diff --git a/analysis/current/analysis-report-hybrid.json b/analysis/current/analysis-report-hybrid.json index 8f311fbbb473bde3eeb12fc426402800529c9ac2..aa53524ac62c8adb7bc3f5a8cd7a04818556691e 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": "20260619T060024Z", - "generated_at": "2026-06-19T06:10:37Z", + "snapshot_id": "20260619T120025Z", + "generated_at": "2026-06-19T12:09:55Z", "evidence_quality": "full", "llm_enrichment": true, "meta_bugs": [ @@ -1810,53 +1810,6 @@ } ] }, - { - "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": 46716, - "issue_numbers": [], - "pr_numbers": [ - 46554, - 46716 - ], - "evidence_types": [ - "soft_similarity" - ], - "pr_comparisons": [ - { - "left_pr_number": 46554, - "right_pr_number": 46716, - "code_similarity": 0.408, - "size_similarity": 0.429, - "file_overlap": 1.0, - "area_overlap": 0.062, - "patch_similarity": 0.0, - "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": [ - [ - 14, - 17 - ] - ] - } - ] - } - ] - }, { "cluster_id": "cluster-44018-2", "summary": "Cluster of 2 related pull requests linked by soft_similarity.", @@ -2242,15 +2195,6 @@ "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": 46716, - "duplicate_pr_numbers": [ - 46554 - ], - "target_issue_number": null, - "reason": "PRs in cluster-46554-2 are treated as duplicates because they share soft_similarity evidence." - }, { "cluster_id": "cluster-46562-2", "canonical_pr_number": 46562, diff --git a/analysis/current/analysis-report-hybrid.llm-reviews.json b/analysis/current/analysis-report-hybrid.llm-reviews.json index ddd06c948c557d9a315d80cca59d6e8fc0a900b8..5e866d53fa9c93b68bd20275fd8b6c9b9693b0c5 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": "20260619T060024Z", - "generated_at": "2026-06-19T06:10:37Z", + "snapshot_id": "20260619T120025Z", + "generated_at": "2026-06-19T12:09:55Z", "model": "gpt-5.4-mini?service_tier=flex", "reviews": [ { @@ -130,412 +130,400 @@ "issue:45237|issue:46015", "issue:43039|issue:44805" ], - "prepared_review_unit_hash": "79408c6e736399fe941148fcd9669c0abe8a9e2358b312e18cfd1c09d49d1cb2", + "prepared_review_unit_hash": "ca75d09a52212709ce0521add24d0d42711a16545bfe67df427db791afd42007", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 62, - "serialized_chars": 45726, - "estimated_input_tokens": 11432, - "estimated_eval_tokens": 23120 + "serialized_chars": 45723, + "estimated_input_tokens": 11431, + "estimated_eval_tokens": 23118 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous; only a few tight duplicates are present around DeepSpeed ZeRO-3 index errors, torch 2.4.1 incompatibility, and Qwen3.5 rotary-embedding/offload behavior. Most other pairs are same subsystem or same model family but different concrete bugs.", - "confidence": 0.36, - "canonical_issue_reason": "issue:44805 is the broadest, most central runtime tensor-shape failure in the set, and it sits closest to several nearby index/mask/device-error reports.", + "summary": "Mostly a grab-bag of unrelated bug reports. The clearest duplicate cluster is the DeepSpeed ZeRO-3 Bert index-out-of-bounds reports (43596/43638); a second likely duplicate pair is the Qwen3.5 MoE rotary-embedding/device-materialization reports (45859/45902). The torch 2.4.1 incompatibility reports (45800/46515) also look duplicate-like, but most other pairs are just broad topical similarity.", + "confidence": 0.82, + "canonical_issue_reason": "issue:43596 is the strongest representative of the cluster: it states the concrete ZeRO-3 \"index 0 out of bounds\" failure clearly, and it has the highest-confidence soft duplicate match to 43638 with essentially the same error path and initialization context.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44805 is the best single representative for the cluster because its title/body family is the most generic among the runtime failures and it connects to the widest set of soft-similarity neighbors.", + "best_issue_reason": "issue:43596 is the best issue candidate overall because it is specific, well-scoped, and sits at the center of the only clearly duplicated bug pair in the set.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44322", "right": "issue:45464", "accept": false, - "reason": "Same model family, but one is a missing config attribute and the other is a streaming chat/completions failure; different code paths." + "reason": "Different failures: Qwen3_5Config missing num_attention_heads vs chat/completions streaming error. Same family, not the same bug." }, { "left": "issue:45237", "right": "issue:45685", "accept": false, - "reason": "Different hardware/runtime bugs: AMD GPU failure versus an MPS histogram op issue in MoE." + "reason": "Both are MoE/GPU-related, but one is AMD GPU execution and the other is an MPS histogram op issue. Different code paths." }, { "left": "issue:43716", "right": "issue:45237", "accept": false, - "reason": "Both are model/runtime failures, but one is a dtype mismatch in Mistral-3 preprocessing and the other is a GPT-OSS AMD GPU issue." + "reason": "Mistral-3 image dtype mismatch and GPT-OSS AMD GPU failure are unrelated problems." }, { "left": "issue:43828", "right": "issue:45237", "accept": false, - "reason": "Both mention execution/runtime problems, but the concrete bug is different: autocast dtype mismatch versus AMD GPU failure." + "reason": "Autocast dtype mismatch in Phi-tiny-MoE is not the same issue as GPT-OSS on AMD GPUs." }, { "left": "issue:45237", "right": "issue:46468", "accept": false, - "reason": "Different model/example failures; no evidence they share the same underlying bug or fix." + "reason": "GPT-OSS AMD GPU failure and a broken Nemotron example are too different to be duplicates." }, { "left": "issue:45237", "right": "issue:45507", "accept": false, - "reason": "Different model-specific failures; one is AMD GPU compatibility, the other is an invalid-method call in GraniteMoEHybrid." + "reason": "AMD GPU failure in GPT-OSS and GraniteMoEHybrid invalid method are different model/code-path issues." }, { "left": "issue:30277", "right": "issue:35545", "accept": false, - "reason": "Unrelated failures: DeepSpeed collective mismatch versus ONNX export error." + "reason": "DeepSpeed ZeRO-3 collective mismatch and ModernBERT ONNX export error are unrelated." }, { "left": "issue:35545", "right": "issue:42915", "accept": false, - "reason": "Different model and error surface; ONNX export failure is not the same as FineGrainedFP8Config execution failure." + "reason": "Both are model-export/config issues, but ModernBERT ONNX export and Qwen3Moe FP8 failure are different bugs." }, { "left": "issue:35545", "right": "issue:41720", "accept": false, - "reason": "Both are runtime errors, but the concrete failures differ: export error versus device-mapped cudaErrorAssert." + "reason": "ModernBERT ONNX export and Qwen3 auto device mapping cudaErrorAssert are unrelated." }, { "left": "issue:45901", "right": "issue:46015", "accept": false, - "reason": "A task crash and outdated RAG examples are both docs/product issues, but they are different changes." + "reason": "Table-question-answering crash and outdated RAG examples are different issues; one is a runtime bug, the other docs." }, { "left": "issue:44509", "right": "issue:44512", "accept": false, - "reason": "Both are docs cleanup for v5 removals, but they refer to different removed content and should not be merged as the same change." + "reason": "Both are docs regressions after v5, but they cover different removed topics and should not be merged as the same change." }, { "left": "issue:35545", "right": "issue:36010", "accept": false, - "reason": "Different regressions: ONNX export versus an import error for GenerationMixin." + "reason": "ModernBERT ONNX export and GenerationMixin import error are different failure modes." }, { "left": "issue:44918", "right": "issue:46165", "accept": false, - "reason": "Different training workflows and failures; unpacking embeddings in SFT is not the same as resuming from a local model with trust_remote_code." + "reason": "TRL SFT embedding unpacking and resume-training with trust_remote_code are different training issues." }, { "left": "issue:44805", "right": "issue:45823", "accept": false, - "reason": "Both are placement/shape-style forward failures, but one is a mask/tensor shape mismatch and the other is a Gemma4 PLE device mismatch." + "reason": "Mask-shape IndexError and Gemma4 PLE device mismatch are different symptoms and code paths." }, { "left": "issue:43638", "right": "issue:44805", "accept": false, - "reason": "Both are IndexError/shape-related, but the concrete bug locations and failure modes differ (zero-sized index during ZeRO-3 init versus mask shape mismatch)." + "reason": "Both involve tensor/index errors, but 43638 is ZeRO-3/Bert init while 44805 is a mask/tensor shape mismatch." }, { "left": "issue:43638", "right": "issue:46165", "accept": false, - "reason": "Different training/loading problems with no clear shared code-path." + "reason": "ZeRO-3 Bert init error and resume-training/remote-code problem are unrelated." }, { "left": "issue:43596", "right": "issue:44805", "accept": false, - "reason": "Both involve index/shape errors, but they are not the same concrete bug or model path." + "reason": "Different IndexError classes: 43596 is ZeRO-3 empty-dimension init, 44805 is mask-vs-tensor shape mismatch." }, { "left": "issue:44805", "right": "issue:46294", "accept": false, - "reason": "Different multimodal input bug: mask/tensor mismatch versus video token/feature mismatch." + "reason": "Mask/tensor shape mismatch and Qwen3.5 video token/feature mismatch are different multimodal bugs." }, { "left": "issue:44805", "right": "issue:45161", "accept": false, - "reason": "Different subsystems: generic tensor-shape error versus GPT-OSS tensor-parallel failure." + "reason": "Index/shape mismatch in one issue and GPT-OSS only-TP failure in the other are unrelated." }, { "left": "issue:46165", "right": "issue:46468", "accept": false, - "reason": "Resume-training with local remote-code models is unrelated to an official example being broken." + "reason": "Resume-training with remote code and broken Nemotron example are not the same bug." }, { "left": "issue:44560", "right": "issue:46165", "accept": false, - "reason": "Different model/runtime issues with no shared concrete failure." + "reason": "Qwen3-vl video StopIteration and resume-training remote-code failure are unrelated." }, { "left": "issue:43596", "right": "issue:46165", "accept": false, - "reason": "ZeRO-3 index error and resume-training failure are unrelated." + "reason": "ZeRO-3 Bert init error and remote-code resume failure are unrelated." }, { "left": "issue:34634", "right": "issue:35545", "accept": false, - "reason": "Different model families and errors: BarkProcessor preset handling versus ModernBERT ONNX export." + "reason": "BarkProcessor voice_preset and ModernBERT ONNX export are unrelated." }, { "left": "issue:30277", "right": "issue:43638", "accept": false, - "reason": "Both touch DeepSpeed, but one is a collective mismatch and the other is a zero-sized index error; different bugs." + "reason": "Both involve DeepSpeed, but collective mismatch and Bert ZeRO-3 index error are different failures." }, { "left": "issue:43596", "right": "issue:43638", "accept": true, - "reason": "Same underlying DeepSpeed ZeRO-3 index-out-of-bounds failure during Bert model initialization/loading." + "reason": "Same underlying ZeRO-3 initialization bug: both report the same out-of-bounds index error on Bert models during deepspeed zero3 init." }, { "left": "issue:43039", "right": "issue:45161", "accept": false, - "reason": "Different bugs in different GPT-OSS paths; one is Liger kernel cross-entropy invocation, the other is tensor-parallel support." + "reason": "Liger Kernel cross-entropy being called and GPT-OSS TP failure are not the same issue." }, { "left": "issue:43975", "right": "issue:44457", "accept": false, - "reason": "Different model behaviors: detokenization bug versus LoRA merge/save/reload mismatch." + "reason": "Incorrect detokenization and LoRA merge/reload output mismatch are different problems." }, { "left": "issue:44292", "right": "issue:44457", "accept": false, - "reason": "Different model/runtime issues; NVFP4 execution error is not the same as LoRA weight merge inconsistency." + "reason": "Qwen-3-8B NVFP4 runtime error and LoRA merge/save inconsistency are unrelated." }, { "left": "issue:44841", "right": "issue:46661", "accept": false, - "reason": "Processor failure and remote-code config_class import failure are different code paths." + "reason": "Processor failure for Voxtral and remote-code model_class config_class error are different loading paths." }, { "left": "issue:24643", "right": "issue:30277", "accept": false, - "reason": "Both are DeepSpeed-related, but the concrete failures are different and not mergeable as one fix." + "reason": "Both mention DeepSpeed training, but weight dimensionality and collective mismatch are distinct bugs." }, { "left": "issue:34928", "right": "issue:44805", "accept": false, - "reason": "Activation-checkpointing recomputation mismatch is a different bug from the mask/tensor shape mismatch in 44805." + "reason": "FSDP activation-checkpointing size mismatch and mask-shape IndexError are not the same failure." }, { "left": "issue:45859", "right": "issue:45902", "accept": true, - "reason": "Same Qwen3_5MoeTextRotaryEmbedding bug family around inv_freq/device materialization and offload support." + "reason": "Same underlying Qwen3.5 MoE rotary-embedding/device handling bug: both center on Qwen3_5MoeTextRotaryEmbedding and its inv_freq behavior under device materialization/offload." }, { "left": "issue:43638", "right": "issue:45161", "accept": false, - "reason": "Zero-3 Bert index error and GPT-OSS tensor-parallel failure are unrelated." + "reason": "ZeRO-3 Bert index error and GPT-OSS TP failure are unrelated." }, { "left": "issue:34928", "right": "issue:36331", "accept": false, - "reason": "FSDP/activation-checkpointing tensor recomputation mismatch versus custom trainer signature change." + "reason": "FSDP checkpointing shape mismatch and CustomTrainer.compute_loss signature mismatch are different bugs." }, { "left": "issue:43596", "right": "issue:45161", "accept": false, - "reason": "Different model and failure mode; no shared underlying bug." + "reason": "ZeRO-3 Bert init error and GPT-OSS TP issue are unrelated." }, { "left": "issue:43716", "right": "issue:45507", "accept": false, - "reason": "Different model-specific runtime bugs: dtype mismatch versus invalid method call." + "reason": "Mistral-3 image dtype mismatch and GraniteMoEHybrid invalid method are different issues." }, { "left": "issue:44292", "right": "issue:45237", "accept": false, - "reason": "Different model/runtime failures; NVFP4 execution error is not the same as GPT-OSS AMD GPU support." + "reason": "Qwen-3-8B NVFP4 error and GPT-OSS AMD GPU failure are different model/runtime problems." }, { "left": "issue:45874", "right": "issue:46294", "accept": false, - "reason": "Both are Gemma/Qwen multimodal issues, but one is reverse-embedding cost and the other is video feature/token mismatch." + "reason": "Gemma4 reverse embedding lookup and Qwen3.5 video token/feature mismatch are unrelated." }, { "left": "issue:39290", "right": "issue:45081", "accept": false, - "reason": "Different model families and failures: Gemma3 vLLM attribute error versus Mistral tokenizer regex patch crash." + "reason": "Gemma3 sliding_window_pattern/vLLM error and Mistral tokenizer regex crash are different code paths." }, { "left": "issue:45823", "right": "issue:45874", "accept": false, - "reason": "Different Gemma4 bugs: device mismatch in forward versus expensive reverse embedding lookup." + "reason": "Both are Gemma4-related, but device mismatch during forward and expensive reverse embedding lookup are different bugs." }, { "left": "issue:45800", "right": "issue:46515", "accept": true, - "reason": "Same torch 2.4.1 incompatibility report, just phrased with slightly different specificity." + "reason": "Both report the same torch 2.4.1 incompatibility; 46515 is a broader restatement of the version-compatibility problem in 45800." }, { "left": "issue:43425", "right": "issue:44292", "accept": false, - "reason": "General torch-version incompatibility and a Qwen NVFP4 runtime error are different issues." - }, - { - "left": "issue:34928", - "right": "issue:41720", - "accept": false, - "reason": "Different failures in different stacks: FSDP checkpointing mismatch versus Qwen3 device-mapping cudaErrorAssert." + "reason": "Torch 2.10 incompatibility and Qwen-3-8B NVFP4 runtime error are unrelated." }, { - "left": "issue:43828", - "right": "issue:45161", + "left": "issue:43827", + "right": "issue:44512", "accept": false, - "reason": "Different model/runtime bugs; autocast dtype mismatch is not tensor-parallel support." + "reason": "Both are docs regressions, but one is pipeline-task references and the other is the removed transformers run command." }, { "left": "issue:43828", "right": "issue:44292", "accept": false, - "reason": "Both are inference-time errors, but they are different concrete bugs with different fixes." + "reason": "Autocast dtype mismatch and Qwen-3-8B NVFP4 runtime error are not the same issue." }, { "left": "issue:43541", "right": "issue:43716", "accept": false, - "reason": "Different models and code paths: torch dynamo tracing grouped_mm error versus image preprocessor dtype mismatch." + "reason": "Mixtral grouped_mm torch dynamo tracing error and Mistral-3 image dtype mismatch are different failures." }, { "left": "issue:43716", "right": "issue:46468", "accept": false, - "reason": "Different model/example failures; no shared underlying bug." + "reason": "Mistral-3 image dtype mismatch and broken Nemotron example are unrelated." }, { "left": "issue:43716", "right": "issue:45685", "accept": false, - "reason": "Different failures and platforms: Mistral-3 dtype mismatch versus MoE MPS histogram op limitation." + "reason": "Mistral-3 image preprocessing dtype mismatch and MPS histogram op error are different problems." }, { "left": "issue:44292", "right": "issue:44918", "accept": false, - "reason": "Different Qwen-related failures with different trigger points and fixes." + "reason": "Qwen-3-8B NVFP4 runtime error and TRL SFT embedding unpacking failure are unrelated." }, { "left": "issue:43716", "right": "issue:44292", "accept": false, - "reason": "Different concrete bugs despite both involving model inference errors." + "reason": "Mistral-3 image dtype mismatch and Qwen-3-8B NVFP4 runtime error are different bugs." }, { "left": "issue:45507", "right": "issue:45685", "accept": false, - "reason": "Different model/platform bugs; invalid method call is not the same as the MPS histogram error." + "reason": "GraniteMoEHybrid invalid method and MPS histogram_mps not implemented are unrelated." }, { "left": "issue:44918", "right": "issue:45161", "accept": false, - "reason": "Different training/runtime failures in different models and subsystems." + "reason": "TRL SFT embedding unpacking and GPT-OSS only-TP failure are different issues." }, { "left": "issue:43039", "right": "issue:44918", "accept": false, - "reason": "Different bugs: Liger kernel cross-entropy call versus Qwen3.5 SFT embedding unpacking." + "reason": "Liger Kernel cross_entropy being called and TRL SFT embedding unpacking are unrelated." }, { "left": "issue:41762", "right": "issue:43638", - "accept": true, - "reason": "Same DeepSpeed ZeRO-3 index-0-out-of-bounds initialization/loading failure, just on a different model." - }, - { - "left": "issue:43827", - "right": "issue:44512", "accept": false, - "reason": "Both are docs issues caused by v5 removals, but they document different removals and are separate changes." + "reason": "Both are ZeRO-3 index errors, but Gemma3 conditional generation and Bert init are not the same concrete bug." }, { "left": "issue:43828", "right": "issue:45685", "accept": false, - "reason": "Different bugs; autocast dtype mismatch is not the same as the MPS int histogram limitation." + "reason": "Autocast dtype mismatch and MPS histogram op error are different runtime issues." }, { "left": "issue:44918", "right": "issue:45685", "accept": false, - "reason": "Different model/runtime problems with no shared fix." + "reason": "TRL SFT embedding unpacking and MPS histogram op error are unrelated." }, { "left": "issue:43039", "right": "issue:43828", "accept": false, - "reason": "Both involve runtime behavior, but one is a kernel-selection issue and the other is a dtype mismatch." + "reason": "Liger Kernel cross_entropy usage and autocast dtype mismatch are different problems." }, { "left": "issue:30277", "right": "issue:36331", "accept": false, - "reason": "DeepSpeed collective mismatch and custom trainer signature mismatch are unrelated." + "reason": "DeepSpeed collective mismatch and compute_loss signature mismatch are unrelated." }, { "left": "issue:34928", "right": "issue:36010", "accept": false, - "reason": "Different bugs in different layers: checkpointing/FSDP mismatch versus import error." + "reason": "FSDP activation-checkpointing mismatch and GenerationMixin import error are different issues." }, { "left": "issue:45237", "right": "issue:46015", "accept": false, - "reason": "Model runtime failure versus outdated RAG examples; not the same issue." + "reason": "GPT-OSS AMD GPU failure and outdated RAG examples are unrelated." }, { "left": "issue:43039", "right": "issue:44805", "accept": false, - "reason": "Different bugs: Liger kernel behavior versus generic mask/tensor shape mismatch." + "reason": "Liger Kernel cross_entropy issue and mask-shape IndexError are different bugs." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative overall. The accepted soft edges are supported by the titles/families in the data (DeepSpeed ZeRO-3 index errors, torch 2.4.1 incompatibility, and Qwen3.5 rotary-embedding/offload issues), and the rejected pairs do not overclaim shared fixes. The canonical issue choice is defensible given the broad mask/shape similarity links." + "feedback": "The analyst summary is well grounded in the packet and stays conservative. The accepted soft edges are supported by closely matching titles and issue themes: 43596/43638 are near-duplicate ZeRO-3 Bert index errors, 45859/45902 are clearly the same Qwen3.5 MoE rotary-embedding/materialization bug family, and 45800/46515 are plausibly the same torch 2.4.1 incompatibility report. The remaining rejections are appropriately cautious and do not overstate evidence." }, "evaluator_used": true, "retried": false, @@ -648,268 +636,286 @@ "issue:43296|issue:43366", "issue:44387|issue:45005" ], - "prepared_review_unit_hash": "1c412572b2423e3934a11c0a6b7cd978a4c910865d5c873f0ad96b6ef0ee7b58", + "prepared_review_unit_hash": "43c5730fe49c0898215644f682f2f8eedc4a8a40f20628fdc61d6ecbe4904abc", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 41, - "serialized_chars": 39791, - "estimated_input_tokens": 9948, - "estimated_eval_tokens": 20152 + "serialized_chars": 39793, + "estimated_input_tokens": 9949, + "estimated_eval_tokens": 20154 }, "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 clear duplicate-like pair is the v5 AutoTokenizer wrong-token-ID regression across model families.", - "confidence": 0.61, - "canonical_issue_reason": "issue:45920 is the broadest and most representative report: it describes the v5 AutoTokenizer regression affecting multiple model families, subsuming the narrower Granite-only report.", + "summary": "The cluster is mostly heterogeneous, with only a couple of true duplicate-looking subthreads. The clearest match is the AutoTokenizer wrong-token-ID regression, and there is also a plausible duplicate pair around the DeepSpeed ZeRO-3 index-out-of-bounds load failure.", + "confidence": 0.77, + "canonical_issue_reason": "Issue 45920 is the broadest and most representative of the tokenizer regression family: it describes the wrong-token-ID bug across multiple model families, subsuming the narrower Granite-only report.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45920 is the best issue to anchor the cluster because it captures the shared tokenizer regression theme most generally and is still open/current.", + "best_issue_reason": "Issue 45920 is the best single issue to anchor the cluster because it is the most general, clearly regression-focused, and spans the widest set of affected models.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43866", "right": "issue:46402", "accept": false, - "reason": "Different models and failure modes: corrupted checkpoint vs missing/unsafely loaded weights." + "reason": "Different model families and failures: corrupted checkpoint vs. missing text-encoder weights." }, { "left": "issue:45800", "right": "issue:46304", "accept": false, - "reason": "Unrelated root causes: torch/Transformers version incompatibility vs USE_DISTRIBUTED=0 support breakage." + "reason": "Both mention compatibility, but they are different incompatibilities and different code paths." }, { "left": "issue:43296", "right": "issue:46304", "accept": false, - "reason": "PaddleOCR-VL loading in vLLM is unrelated to distributed-pytorch support regression." + "reason": "Unrelated symptoms: vLLM load failure for PaddleOCR-VL vs. PyTorch USE_DISTRIBUTED=0 support breakage." }, { "left": "issue:41762", "right": "issue:43596", - "accept": false, - "reason": "Both are zero3 IndexErrors, but the models and loading paths differ too much to call the same bug." + "accept": true, + "reason": "Same ZeRO-3 load-time index-out-of-bounds failure; only the model types differ." }, { "left": "issue:39290", "right": "issue:41720", "accept": false, - "reason": "Gemma3 config attribute error and Qwen3 auto-device-map CUDA assert are different issues." + "reason": "Different problems: missing config attribute in vLLM vs. CUDA assert with auto device mapping." }, { "left": "issue:43901", "right": "issue:44512", "accept": false, - "reason": "Both are docs problems, but they refer to different commands/behaviors and are not the same change." + "reason": "Both are docs-related, but they reference different outdated behaviors and do not share the same bug." }, { "left": "issue:44589", "right": "issue:46304", "accept": false, - "reason": "Float8 storage lookup failure is unrelated to the USE_DISTRIBUTED=0 compatibility break." + "reason": "Completely different failures: missing Float8 storage vs. distributed build/runtime incompatibility." }, { "left": "issue:45237", "right": "issue:45901", "accept": false, - "reason": "Different subsystems and symptoms; no shared bug path." + "reason": "Different subsystems and unrelated crash modes." }, { "left": "issue:43039", "right": "issue:44560", "accept": false, - "reason": "Liger Kernel cross-entropy call and Qwen3-vl StopIteration are unrelated." + "reason": "Unrelated: Liger Kernel cross-entropy path vs. video chat-template StopIteration." }, { "left": "issue:45823", "right": "issue:46294", "accept": false, - "reason": "Device-mapping forward mismatch vs video chat-template token/feature mismatch are different bugs." + "reason": "Different model families and different device/chat-template failures." }, { "left": "issue:43975", "right": "issue:46015", "accept": false, - "reason": "Detokenization regression and outdated RAG docs are unrelated." + "reason": "Different scope: detokenization regression vs. outdated RAG examples." }, { "left": "issue:44841", "right": "issue:45823", "accept": false, - "reason": "Processor failure for Voxtral and Gemma4 device mismatch are different code paths." + "reason": "Different concrete failures: processor load bug vs. device-mismatch bug." }, { "left": "issue:44863", "right": "issue:46015", "accept": false, - "reason": "NemotronH checkpoint loading and RAG docs are unrelated." + "reason": "NemotronH checkpoint loading is unrelated to RAG docs." }, { "left": "issue:45020", "right": "issue:46304", "accept": false, - "reason": "Remote-code model breakage is broader but not the same as the USE_DISTRIBUTED=0 specific incompatibility." + "reason": "Remote-code regression and distributed build support are different bugs." }, { "left": "issue:45812", "right": "issue:45920", "accept": true, - "reason": "Same underlying v5 AutoTokenizer regression: wrong token IDs across model families; 45812 is a narrower instance of the broader 45920 report." + "reason": "Same underlying AutoTokenizer wrong-token-ID regression; 45920 is the broader report covering more models." }, { "left": "issue:43866", "right": "issue:44863", "accept": false, - "reason": "Two different model-loading problems with different likely causes; both are generic checkpoint issues only at a high level." + "reason": "Corrupted checkpoint vs. model-loading support issue; not the same bug." }, { "left": "issue:43872", "right": "issue:46304", "accept": false, - "reason": "bitsandbytes Int8Params kwarg mismatch is unrelated to distributed-build support." + "reason": "bitsandbytes constructor breakage is unrelated to USE_DISTRIBUTED=0 support." }, { "left": "issue:44292", "right": "issue:46015", "accept": false, - "reason": "Qwen-3 NVFP4 runtime error is unrelated to docs/examples." + "reason": "Different topics and failure modes." }, { "left": "issue:39290", "right": "issue:43531", "accept": false, - "reason": "Both mention sliding_window, but one is a missing config attribute and the other is a Qwen3-MoE sliding_window behavior issue." + "reason": "Both mention sliding-window, but the config-attribute error and the Qwen3-MoE sliding-window issue are not clearly the same bug." }, { "left": "issue:41093", "right": "issue:43638", "accept": false, - "reason": "Similar zero3 IndexError symptom, but different models and context; too generic to merge." + "reason": "Both are IndexError reports, but the mask-shape mismatch is a different failure from the ZeRO-3 empty-tensor issue." }, { "left": "issue:43792", "right": "issue:45901", "accept": false, - "reason": "Whisper loading failure and table-question-answering crash are unrelated." + "reason": "Different tasks and unrelated crash causes." }, { "left": "issue:43854", "right": "issue:43866", "accept": false, - "reason": "GLM-4.7-Flash test loading failure is not the same as Ovis2 checkpoint corruption." + "reason": "Different model load failures with no shared root cause." }, { "left": "issue:43425", "right": "issue:45070", "accept": false, - "reason": "Torch-version incompatibility and pydantic-field regression are unrelated." + "reason": "Version incompatibility vs. pydantic/config-field regression; unrelated." }, { "left": "issue:43975", "right": "issue:45901", "accept": false, - "reason": "Detokenization regression and table-question-answering crash are different bugs." + "reason": "Detokenization regression and table QA crash are unrelated." }, { "left": "issue:43596", "right": "issue:45800", "accept": false, - "reason": "Zero3/BertModel IndexError and torch/Transformers version incompatibility are different problems." + "reason": "ZeRO-3 load-time IndexError and torch/transformers version incompatibility are different issues." }, { "left": "issue:43792", "right": "issue:45070", "accept": false, - "reason": "Whisper loading and PretrainedConfig pydantic breakage are unrelated." + "reason": "Whisper runtime failure is unrelated to the pydantic PretrainedConfig regression." }, { "left": "issue:45362", "right": "issue:45464", "accept": false, - "reason": "Both involve Qwen3.5 and chat, but the concrete failures and code paths differ." + "reason": "Both involve Qwen3.5, but the chat crash and streaming API failure are not clearly the same code-path bug." + }, + { + "left": "issue:44292", + "right": "issue:45901", + "accept": false, + "reason": "Unrelated model-specific error vs. table QA task crash." + }, + { + "left": "issue:39290", + "right": "issue:45823", + "accept": false, + "reason": "Different model/config problems; no shared underlying bug is evident." }, { "left": "issue:45464", "right": "issue:45685", "accept": false, - "reason": "Streaming chat API failure and MPS histogram int error are unrelated." + "reason": "Chat/streaming failure and MPS MoE histogram error are unrelated." + }, + { + "left": "issue:43541", + "right": "issue:45685", + "accept": false, + "reason": "Both are numeric/runtime issues, but in different kernels and backends." }, { "left": "issue:36010", "right": "issue:39290", "accept": false, - "reason": "ImportError for GenerationMixin and Gemma3TextConfig sliding-window attr error are unrelated." + "reason": "ImportError for GenerationMixin is unrelated to the Gemma3 sliding_window_pattern error." }, { "left": "issue:43541", - "right": "issue:45685", + "right": "issue:43828", "accept": false, - "reason": "Torch Dynamo grouped_mm failure and MPS histogram int error are different bugs." + "reason": "Both are dtype-related, but they affect different models and ops." }, { "left": "issue:44841", "right": "issue:45084", "accept": false, - "reason": "Processor loading failure and compile-non-template-nodes error do not share the same code path." + "reason": "Processor failure and template-compilation TypeError are different bugs." }, { "left": "issue:43404", "right": "issue:43866", "accept": false, - "reason": "Mistral3 tied-weights bug and Ovis2 checkpoint corruption are unrelated." + "reason": "Untied lm_head weights and corrupted checkpoint are unrelated." }, { "left": "issue:41093", "right": "issue:41762", "accept": false, - "reason": "Both are IndexErrors, but the model families and circumstances differ enough that they are not the same underlying bug." + "reason": "Same general IndexError wording, but the mask-size mismatch is not the ZeRO-3 empty-tensor loading bug." }, { "left": "issue:45399", "right": "issue:46661", "accept": false, - "reason": "Flash-attn fallback gating and remote-code config_class errors are unrelated." + "reason": "Both mention remote-code-ish flows, but the fallback gating bug and missing config_class AttributeError are different code paths." }, { "left": "issue:43572", "right": "issue:45070", "accept": false, - "reason": "StableLmConfig pad_token_idx regression and pydantic PretrainedConfig breakage are unrelated." + "reason": "Missing pad_token_idx and pydantic PretrainedConfig field regression are unrelated." }, { "left": "issue:45800", "right": "issue:45823", "accept": false, - "reason": "Torch version incompatibility and Gemma4 device mismatch are different issues." + "reason": "Torch/version compatibility and Gemma4 device mismatch are separate problems." }, { "left": "issue:43296", "right": "issue:43366", "accept": false, - "reason": "PaddleOCR-VL/vLLM loading and GGUF gpt-oss support are different feature requests/issues." + "reason": "Different model support gaps: PaddleOCR-VL in vLLM vs. GGUF gpt-oss support." }, { "left": "issue:44387", "right": "issue:45005", "accept": false, - "reason": "Int4 OOM regression and tied-weights translation-model issue are unrelated." + "reason": "Int4 OOM regression and tied-weights translation-model issues are unrelated." } ] }, "evaluator_result": { "accept": true, - "feedback": "The cluster summary is grounded in the packet, and the only accepted soft-edge pair (issue:45812 / issue:45920) is well-supported as the same v5 AutoTokenizer wrong-token-ID regression across model families. The remaining soft-edge rejections are conservative and appropriately avoid overclaiming duplicates from only superficial symptom overlap. No pull-request mergeability issues apply here." + "feedback": "The summary is grounded in the packet and stays conservative. The two accepted soft pairs are reasonably supported by the issue titles, and the rest of the pairwise rejections avoid overstating similarity." }, "evaluator_used": true, "retried": false, @@ -1011,20 +1017,20 @@ "issue:43975|issue:44292", "issue:44292|issue:45507" ], - "prepared_review_unit_hash": "104cabdcaa62766739393354ad5b3a6097a29a4d03d82db4acbca5a64556137b", + "prepared_review_unit_hash": "423d94ffa025f77241acf84406bd4459c157fd746124a4eada6eb16e84772efa", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 31, - "serialized_chars": 36260, + "serialized_chars": 36258, "estimated_input_tokens": 9065, "estimated_eval_tokens": 18386 }, @@ -1033,207 +1039,207 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a heterogeneous issue cluster with one clear duplicate-style pair: the broad input_ids/inputs_embeds validation defect and its T5-family error-report variant. The remote-code/custom-module pair is too speculative to merge as a duplicate without stronger evidence of one shared code path.", - "confidence": 0.87, - "canonical_issue_reason": "Issue 45183 is the best canonical issue because it captures a reusable, cross-model validation bug rather than a model-specific symptom, and 46463 reads like a narrower T5-family manifestation of that same input validation problem.", + "summary": "Mostly a heterogeneous issue set with a few near-duplicate themes around Qwen video validation and input-validation error messaging. I only accept soft edges where the titles point to the same concrete code path; most pairs are just the same broad subsystem or adjacent version regressions, not true duplicates.", + "confidence": 0.72, + "canonical_issue_reason": "No single issue truly represents the whole cluster; if one must be chosen, issue:45020 is the broadest version-regression umbrella among the set.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45183 is the strongest representative issue in the set: it describes a general validation defect with broader applicability than the more specific downstream error reports.", + "best_issue_reason": "issue:45020 is the most generally representative issue because it spans recent transformers-version breakage rather than one narrow model symptom, though it is still not a duplicate of most others.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44805", "right": "issue:44918", "accept": false, - "reason": "Both are IndexError failures, but the titles point to different paths: generic mask-shape mismatch vs TRL SFT unpacking for Qwen3.5 embeddings." + "reason": "Both are shape/runtime failures, but they affect different code paths: generic mask indexing vs TRL SFT embedding unpacking." }, { "left": "issue:43541", "right": "issue:45161", "accept": false, - "reason": "Different model families and different failure modes: grouped_mm tracing in Mixtral versus TP behavior in GPT-OSS MoE." + "reason": "Different problems: grouped_mm tracing failure in Mixtral vs GPT-OSS MoE tensor-parallel behavior." }, { "left": "issue:43828", "right": "issue:45464", "accept": false, - "reason": "Both mention dtype issues, but one is autocast on Phi-tiny-MoE and the other is a streaming chat/completions regression on Qwen3.5-0.8B." + "reason": "Unrelated symptoms and paths: autocast dtype mismatch in Phi-tiny-MoE vs chat/completions streaming failure in Qwen3.5." }, { "left": "issue:43366", "right": "issue:45464", "accept": false, - "reason": "GPT-OSS GGUF architecture support and Qwen3.5 streaming inference are separate loader/runtime issues, not one shared bug." + "reason": "GGUF gpt-oss architecture support and Qwen3.5 streaming inference are different issues." }, { "left": "issue:44291", "right": "issue:45070", "accept": false, - "reason": "Different bug classes entirely: init_empty_weights/_is_hf_initialized handling versus a pydantic PretrainedConfig field regression." + "reason": "Both are v5 compatibility regressions, but one is init_empty_weights argument handling and the other is PretrainedConfig/pydantic field behavior." }, { "left": "issue:43531", "right": "issue:45070", "accept": false, - "reason": "Sliding-window behavior for Qwen3-MoE is unrelated to the pydantic/PretrainedConfig regression." + "reason": "Sliding-window handling for Qwen3-MoE is unrelated to the pydantic config break in v5.4.0." }, { "left": "issue:43541", "right": "issue:44589", "accept": false, - "reason": "Both are runtime errors, but they affect different code paths: grouped_mm tracing versus missing Float8 storage resolution." + "reason": "Different low-level runtime failures: dynamo tracing/grouped_mm vs Float8 storage lookup." }, { "left": "issue:43716", "right": "issue:46294", "accept": false, - "reason": "Both involve multimodal dtype/feature mismatches, but the affected models and concrete failures differ too much to treat as the same bug." + "reason": "Both are multimodal, but one is dtype mismatch in image preprocessing and the other is Qwen3.5 video-token alignment." }, { "left": "issue:45020", "right": "issue:45800", "accept": false, - "reason": "Remote-code breakage and a torch/transformers version incompatibility are broad but distinct problems, not the same underlying defect." + "reason": "Both mention broad compatibility regressions, but the concrete failures are different: remote_code loading vs torch/transformers version incompatibility." }, { "left": "issue:41093", "right": "issue:43596", "accept": false, - "reason": "Both are index errors, but one is a mask/tensor length mismatch and the other is a zero-size tensor access during deepspeed zero3 init." + "reason": "Both are IndexErrors, but the causes are different: mask-length mismatch vs DeepSpeed ZeRO3 init with empty tensors." }, { "left": "issue:44479", "right": "issue:46294", - "accept": false, - "reason": "Both concern Qwen video input regressions, but one is a v5.3.0 regression across several models while the other is a chat-template token/feature mismatch for Qwen3.5-9B." + "accept": true, + "reason": "Same underlying Qwen video-input regression: feature/token mismatch across Qwen3.5 video flows." }, { "left": "issue:43054", "right": "issue:43994", "accept": false, - "reason": "These are separate SigLIP2 complaints; one reports worse embeddings, the other nonsensical outputs with AutoModel/pipeline." + "reason": "Different model families and symptoms: siglip2 embedding quality vs SigLIP auto-model/pipeline results." }, { "left": "issue:45759", "right": "issue:45902", "accept": false, - "reason": "One is about unwrapping text_config in composite multimodal configs, the other about uninitialized memory in Qwen3.5 MoE rotary embeddings." + "reason": "Different composite-model bugs: config unwrapping in from_config vs uninitialized inv_freq after meta\u2192to_empty materialization." }, { "left": "issue:46640", "right": "issue:46661", "accept": false, - "reason": "Flash Attention 2 integration and remote-code config_class lookup are unrelated loader/runtime issues." + "reason": "Flash Attention 2 issue and remote-code model_class/config_class failure are separate loading paths." }, { "left": "issue:43716", "right": "issue:44918", "accept": false, - "reason": "Different multimodal/training failure modes on different models; no shared code-path bug is evident." + "reason": "Different failure modes: image preprocessor dtype mismatch vs TRL SFT unpacking of embeddings." }, { "left": "issue:39186", "right": "issue:44918", "accept": false, - "reason": "FSDP weight shape errors and TRL embedding unpacking failures are different problems." + "reason": "FSDP 2-D weight assertion and Qwen3.5 SFT embedding unpacking are unrelated." }, { "left": "issue:40990", "right": "issue:44960", "accept": false, - "reason": "High perplexity on GPT-OSS and a GLM5 issue are not the same underlying bug." + "reason": "Perplexity anomaly on gpt-oss and GLM5 model issue do not share the same bug." }, { "left": "issue:45632", "right": "issue:45698", "accept": false, - "reason": "Too broad and speculative: both mention remote-code/custom-module loading, but the packet does not show a concrete shared bugfix or clearly mergeable code path." + "reason": "Both involve remote-code loading, but one is cache-path collision and the other is loading the wrong custom module after save_pretrained." }, { "left": "issue:43645", "right": "issue:45399", "accept": false, - "reason": "Notebook custom-model initialization and flash-attn fallback logic are unrelated." + "reason": "Jupyter custom-model initialization and flash-attn fallback gating are different breakages." }, { "left": "issue:43404", "right": "issue:45005", "accept": false, - "reason": "Both mention tied weights, but one is a specific Mistral3 lm_head tying bug and the other is a broader translation-model v5 issue." + "reason": "Tied lm_head weights in Mistral3 and translation-model tied weights are not the same concrete bug." }, { "left": "issue:42915", "right": "issue:45399", "accept": false, - "reason": "Qwen3 MoE FP8 failures and flash-attn fallback gating are different subsystems." + "reason": "FineGrainedFP8Config failure in Qwen3Moe is unrelated to flash-attn fallback blocking." }, { "left": "issue:44530", "right": "issue:45005", "accept": false, - "reason": "PagedAttentionCache linear_attention failure and tied-weights translation regression do not share a concrete code path." + "reason": "PagedAttentionCache linear_attention crash and translation tied-weights regression are unrelated." }, { "left": "issue:43873", "right": "issue:45005", "accept": false, - "reason": "Quantization/offloading behavior and tied-weight handling are distinct bug categories." + "reason": "Quantization/offloading behavior and tied-weight handling are different code paths." }, { "left": "issue:44360", "right": "issue:44512", "accept": false, - "reason": "The DSA indexer discussion and the v5 docs command removal are unrelated." + "reason": "DSA indexer/ReLU discussion is unrelated to a docs mention of the removed transformers run command." }, { "left": "issue:44863", "right": "issue:45005", "accept": false, - "reason": "NemotronH checkpoint loading is unrelated to translation-model tied weights." + "reason": "NemotronH checkpoint loading and translation tied-weights are different model-loading bugs." }, { "left": "issue:44492", "right": "issue:44512", "accept": false, - "reason": "A cache-strategy typo and outdated docs are not the same issue." + "reason": "A typo in cache strategies is unrelated to documentation about a removed command." }, { "left": "issue:44918", "right": "issue:45507", "accept": false, - "reason": "TRL SFT embedding unpacking and GraniteMoEHybrid invalid-method calls are different code paths." + "reason": "Qwen3.5 SFT embedding unpacking and GraniteMoEHybrid invalid method calls are different failures." }, { "left": "issue:45183", "right": "issue:46463", "accept": true, - "reason": "46463 looks like a specific T5-family instance of the broader input_ids/inputs_embeds validation problem in 45183, so one PR could plausibly cover both." + "reason": "Same validation/error-message family: misleading input validation for decoder/input selection, with T5 as a specific instance." }, { "left": "issue:46489", "right": "issue:46491", "accept": false, - "reason": "Both are tokenizer regressions, but they affect different tokenizers and symptoms, so they should not be merged." + "reason": "Both are tokenizer regressions, but they affect different model tokenizers and different symptoms." }, { "left": "issue:43975", "right": "issue:44292", "accept": false, - "reason": "DeepSeek-Coder v1 detokenization and Qwen-3 NVFP4 runtime errors are unrelated." + "reason": "DeepSeek tokenizer detokenization and init_empty_weights argument handling are unrelated." }, { "left": "issue:44292", "right": "issue:45507", "accept": false, - "reason": "These involve different model/runtime failures with no shared underlying code path." + "reason": "init_empty_weights loading error and GraniteMoEHybrid invalid method calls are different issues." } ] }, "evaluator_result": { - "accept": false, - "feedback": "The summary is mostly grounded, but one soft-edge acceptance is too speculative: issue:45632 vs issue:45698 is only a broad remote-code/custom-module similarity, and the packet does not show a concrete shared bugfix or clearly mergeable PR path. The 45183/46463 acceptance looks solid, but the remote-code pair should stay conservative unless there is stronger evidence of the same underlying defect." + "accept": true, + "feedback": "The summary is grounded in the packet: it correctly emphasizes that the set is mostly heterogeneous, and the two accepted soft edges are conservative and supported by the issue titles (Qwen video regression; input-validation/error-message regression). The rejected pairs are generally framed as distinct code paths or symptoms rather than true duplicates, which is appropriately cautious." }, "evaluator_used": true, - "retried": true, + "retried": false, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -1336,241 +1342,247 @@ "issue:42491|issue:44368", "issue:43828|issue:45823" ], - "prepared_review_unit_hash": "f784fd092593a4e3311eb6637a26d354772f5a3ed0a5821ad52312d51b9b95e8", + "prepared_review_unit_hash": "976611d02977e5ef075ab0c324480841c5efd43647f84724d4f268240f2063ab", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 34, - "serialized_chars": 38112, - "estimated_input_tokens": 9528, - "estimated_eval_tokens": 19312 + "serialized_chars": 38116, + "estimated_input_tokens": 9529, + "estimated_eval_tokens": 19314 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "There are two main duplicate clusters here: a non-persistent-buffer regression in Transformers 5.x (43950/43644/44534/45902) and a SigLIP2 implementation discrepancy (43054/43493). Most other pairs are different bugs in the same subsystem but not the same underlying issue.", - "confidence": 0.83, - "canonical_issue_reason": "Issue 43950 is the best umbrella report for the non-persistent-buffer regression: it states the root symptom clearly, scopes it to Transformers 5.x, and matches the duplicate reports that describe the same corruption/fill-with-junk behavior.", + "summary": "This cluster is mostly a grab-bag of unrelated Transformers issues. The strongest overlap is the non-persistent-buffer regression pair (43950/44534), which looks like the same underlying bug or a very close re-report, but the packet does not prove duplicate status; the rest should remain separate.", + "confidence": 0.78, + "canonical_issue_reason": "issue:43950 is the clearest representative for the non-persistent-buffer regression: it is more specific and detailed than 44534, which appears to be a shorter re-report of the same behavior.", "canonical_pr_reason": null, - "best_issue_reason": "43950 is the clearest and most central issue in the largest duplicate cluster, with a title that captures the underlying bug rather than one narrow manifestation.", + "best_issue_reason": "issue:43950 is the best representative issue for this cluster because it most completely describes the strongest near-duplicate bug report while the other items are clearly different problems.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43950", "right": "issue:44534", "accept": true, - "reason": "Same bug: both report Transformers 5.x corrupting/filling non-persistent buffers with junk." + "reason": "Same non-persistent-buffer from_pretrained regression; 44534 reads like a shorter re-report of the same underlying bug." }, { "left": "issue:39692", "right": "issue:43054", "accept": false, - "reason": "Both mention SigLIP2, but one is a broken docs example and the other is a model quality/embedding regression; not the same concrete bug." + "reason": "Both mention SigLIP2, but one is a doc/example mismatch plus quantization failure and the other is a text-embedding quality complaint; not the same bug." }, { "left": "issue:43644", "right": "issue:43950", - "accept": true, - "reason": "Same regression about non-persistent buffers being filled/corrupted in Transformers 5.x." + "accept": false, + "reason": "Different failures: missing fields with trust_remote_code versus non-persistent-buffer corruption." }, { "left": "issue:36331", "right": "issue:41762", "accept": false, - "reason": "Different failures: a custom loss API change vs a DeepSpeed ZeRO-3 loading IndexError." + "reason": "Unrelated code paths: custom trainer API break versus DeepSpeed/Gemma loading IndexError." }, { "left": "issue:43872", "right": "issue:44589", "accept": false, - "reason": "Both are quantization/storage-related, but they are distinct failures with different offending types and code paths." + "reason": "Different incompatibilities: bitsandbytes Int8Params kwargs versus missing Float8 storage type." }, { "left": "issue:43828", "right": "issue:44928", "accept": false, - "reason": "Different models and different failure modes; one is autocast dtype mismatch, the other is a FlashAttention/position_ids crash." + "reason": "Different models and symptoms: autocast dtype mismatch versus Qwen3.5 position_ids / SDPA fallback instability." }, { "left": "issue:45920", "right": "issue:46489", "accept": false, - "reason": "Both concern tokenizer regressions, but 46489 is a specific DeepSeek-Coder gap while 45920 is a broader multi-model wrong-token-ID regression." + "reason": "Both involve tokenizer regressions, but 46489 is a model-specific gap in a prior fix, not the same concrete bug as the broader multi-model token-ID regression." }, { "left": "issue:43866", "right": "issue:44403", "accept": false, - "reason": "Unrelated issues: checkpoint corruption vs noisy load-time messages." + "reason": "Checkpoint corruption versus generic loading noise; not the same underlying issue." }, { "left": "issue:43054", "right": "issue:45701", "accept": false, - "reason": "SigLIP2 text-embedding quality issue vs a general tokenization-version regression; not the same bug." + "reason": "SigLIP2 embedding quality issues are unrelated to generic tokenizer version-dependent changes." }, { "left": "issue:38175", "right": "issue:43054", "accept": false, - "reason": "Related SigLIP2 symptoms, but one is zero probabilities and the other is degraded text embeddings; too vague to merge as the same bug." + "reason": "Same model family, but the titles describe different symptoms and no shared concrete failure mode is evident." }, { "left": "issue:43950", "right": "issue:45902", - "accept": true, - "reason": "Same underlying non-persistent-buffer initialization bug; 45902 is a specific manifestation after meta->to_empty materialization." + "accept": false, + "reason": "Both involve non-persistent buffers, but 45902 is about uninitialized memory after meta/to_empty materialization, which is a distinct bug." }, { "left": "issue:41553", "right": "issue:43883", "accept": false, - "reason": "Different model-loading problems with different models and different errors." + "reason": "Different models and failure types: bad AutoTokenizer error for Voxtral versus missing all_tied_weights_keys on Molmo." }, { "left": "issue:44910", "right": "issue:45859", "accept": false, - "reason": "Both involve Qwen3.5/MoE, but one is FlashAttention illegal access from 3D position_ids and the other is CPU-offload incompatibility in rotary embedding." + "reason": "Flash-Attention crash from misread position_ids is unrelated to CPU-offload incompatibility in Qwen3_5Moe rotary embedding." }, { "left": "issue:43054", "right": "issue:43493", - "accept": true, - "reason": "Same SigLIP2 implementation discrepancy: one reports worse text embeddings, the other cites divergence from the original JAX implementation." + "accept": false, + "reason": "Related SigLIP2 area, but one is a performance complaint and the other is an implementation-vs-JAX discrepancy; not clearly the same bug from the titles alone." }, { - "left": "issue:43856", - "right": "issue:43493", + "left": "issue:42947", + "right": "issue:44387", + "accept": false, + "reason": "Different training/memory issues: gradient checkpointing with LoRA versus higher reserved memory under int4 quantization." + }, + { + "left": "issue:45663", + "right": "issue:45902", "accept": false, - "reason": "Unrelated: Qwen3 MoE memory usage vs SigLIP2 implementation mismatch." + "reason": "Gemma-4 FSDP2 shared_kv_states bug is unrelated to Qwen3_5Moe rotary embedding initialization." }, { "left": "issue:41553", "right": "issue:43881", "accept": false, - "reason": "Different models and different load failures." + "reason": "Different models and failure types; only a broad loading-context overlap." }, { "left": "issue:44403", "right": "issue:44451", "accept": false, - "reason": "General loading-noise report vs a specific model load failure; not the same concrete issue." + "reason": "Generic loading noise versus inability to load a specific model are too vague to treat as the same bug." }, { "left": "issue:43975", "right": "issue:45520", "accept": false, - "reason": "Tokenizer detokenization bug vs Python 3.13 import_utils KeyError; unrelated." + "reason": "Tokenizer detokenization bug for DeepSeek-Coder v1 is different from the flash_attn import issue on Python 3.13." }, { "left": "issue:44403", "right": "issue:44488", "accept": false, - "reason": "General load-time noise vs a specific inability to load a model." + "reason": "Same broad loading theme, but the specific model failure is not the same as the generic noise report." }, { "left": "issue:24643", "right": "issue:28530", "accept": false, - "reason": "Different training behaviors: a DeepSpeed weight shape runtime error vs early-stopping metric resolution." + "reason": "Completely different training errors: DeepSpeed weight shape versus early-stopping metric selection." }, { "left": "issue:42947", "right": "issue:43856", "accept": false, - "reason": "Gradient checkpointing/LoRA issue vs Qwen3 MoE memory usage; different problems." + "reason": "Both are training/memory-related, but they describe different problems and code paths." }, { "left": "issue:43643", "right": "issue:45020", "accept": false, - "reason": "Both involve remote code, but one is missing fields in AutoConfig output while the other is a broader breakage across remote-code models." + "reason": "A missing-fields object returned by trust_remote_code is not the same as a broader remote_code breakage report." }, { "left": "issue:41762", "right": "issue:42915", "accept": false, - "reason": "Different model/config failures: Gemma3 ZeRO-3 loading IndexError vs Qwen3 MoE FP8 failure." + "reason": "Different model stacks and error types: Gemma3 loading IndexError versus Qwen3Moe FP8 failure." }, { "left": "issue:44991", "right": "issue:45399", "accept": false, - "reason": "Tokenizer loading failure vs flash-attn2 fallback logic; unrelated." + "reason": "Tokenizer loading failure and flash-attn2 fallback gating are unrelated." }, { "left": "issue:43824", "right": "issue:43872", "accept": false, - "reason": "ImportError for a model class vs a bitsandbytes Int8Params constructor incompatibility." + "reason": "Importing a missing Qwen model class is unrelated to bitsandbytes parameter initialization incompatibility." }, { "left": "issue:45685", "right": "issue:45823", "accept": false, - "reason": "Different model/runtime issues; MPS histogram support is unrelated to Gemma4 device placement." + "reason": "MPS histogram support for Int and Gemma4 device mismatch are different bugs." }, { "left": "issue:43782", "right": "issue:44960", "accept": false, - "reason": "Different models and different errors." + "reason": "Different models and failure modes: Qwen3VL weight_only loading versus GLM5." }, { "left": "issue:43881", "right": "issue:43883", "accept": false, - "reason": "Different load failures affecting different models." + "reason": "Different models and errors; only both are load-time failures." }, { "left": "issue:45020", "right": "issue:45823", "accept": false, - "reason": "Remote-code regression vs Gemma4 device mismatch; not the same bug." + "reason": "Broad remote_code breakage is not the same concrete issue as a Gemma4 device_map forward mismatch." }, { "left": "issue:45081", "right": "issue:45399", "accept": false, - "reason": "Tokenizer regex crash vs flash-attn2 fallback gating; unrelated." + "reason": "Tokenizer regex patch crash is unrelated to flash-attn2 fallback checks." }, { "left": "issue:44661", "right": "issue:45020", "accept": false, - "reason": "Tokenizer mapping edge case vs remote-code model breakage; different code paths." + "reason": "Tokenizer mapping-name edge case is a narrower implementation bug than the broad remote_code regression report." }, { "left": "issue:42491", "right": "issue:44368", "accept": false, - "reason": "LoRA compatibility across versions vs a tie_word_embeddings warning; not the same underlying problem." + "reason": "LoRA model compatibility across HF versions is unrelated to the tie_word_embeddings warning in Qwen3.5 LoRA training." }, { "left": "issue:43828", "right": "issue:45823", "accept": false, - "reason": "Different model/runtime mismatch failures." + "reason": "Different models and distinct runtime failures: dtype mismatch versus device mismatch." } ] }, "evaluator_result": { - "accept": true, - "feedback": "The summary is well grounded in the packet, and the accepted soft edges are conservative and supported by the titles. The two cluster calls (non-persistent buffers and SigLIP2 discrepancy) look plausible, and the chosen canonical issue is a reasonable umbrella report." + "accept": false, + "feedback": "The cluster summary is mostly grounded, but it overstates the evidence by calling 43950/44534 the \"only confirmed duplicate bug\" / \"one true duplicate subgroup.\" The packet only supports that these two issues are very similar and likely duplicates; it does not confirm duplicate status. Please soften that language to reflect uncertainty." }, "evaluator_used": true, - "retried": false, + "retried": true, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -1667,20 +1679,20 @@ "issue:45325|issue:45902", "issue:45663|issue:45823" ], - "prepared_review_unit_hash": "4d07ac6f560c97e856b7e34853fe58057708b182853169b3067d8cf8bbe45039", + "prepared_review_unit_hash": "2043b62a7a42937acdbd3aab0729e02ab3ea30d90fd4562f8e032fc1ae7a24fc", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 36629, + "serialized_chars": 36630, "estimated_input_tokens": 9158, "estimated_eval_tokens": 18572 }, @@ -1689,180 +1701,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is a heterogeneous cluster containing several unrelated issue families. The only strong duplicate pairs are: docs about removed v5 pipeline tasks, the `_is_hf_initialized`/Int8Params constructor-arg incompatibility, the generic mask-shape IndexError, a Qwen3 DeepSpeed ZeRO-3 weight-loading failure, and a Qwen3-VL video-input regression.", - "confidence": 0.78, - "canonical_issue_reason": "Issue 43872 is the clearest concrete bug report in the cluster and has a near-identical duplicate in 44291 about the same `_is_hf_initialized` constructor-arg failure path.", + "summary": "Mostly a heterogeneous cluster of transformer bugs across model loading, tokenizers, DeepSpeed, and docs. I only accept a few pairs that look like the same underlying regression, and reject the rest as merely similar subsystems or generic error patterns.", + "confidence": 0.69, + "canonical_issue_reason": "issue:44291 is the clearest representative of the duplicated `_is_hf_initialized` / `init_empty_weights` loading regression, and it is broader than the bitsandbytes-specific report.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 43872 is the best representative issue because it captures a specific, reproducible runtime incompatibility and directly matches one accepted duplicate.", + "best_issue_reason": "issue:44291 has the most actionable, concrete failure description among the accepted issue duplicates and maps to a specific code-path regression.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45020", "right": "issue:45356", "accept": false, - "reason": "Both are version-regression/tokenizer complaints, but they involve different models and different failure modes (remote_code loading vs tokenizer codec handling)." + "reason": "Both are tokenizer/model-loading regressions, but they target different concrete failures and model families; too broad to be the same bug." }, { "left": "issue:45715", "right": "issue:46315", "accept": false, - "reason": "Same tokenizer subsystem, but different APIs and symptoms: convert_ids_to_tokens vs convert_tokens_to_ids performance regressions." + "reason": "One is about `convert_ids_to_tokens`, the other `convert_tokens_to_ids`; similar performance theme, different code paths and symptom." }, { "left": "issue:43827", "right": "issue:44509", "accept": true, - "reason": "Both report stale docs still referencing removed v5 pipeline tasks for summarization/translation; same underlying docs cleanup." + "reason": "Both report stale docs still referencing pipeline tasks removed in v5; same documentation cleanup issue." }, { "left": "issue:45488", "right": "issue:46489", "accept": false, - "reason": "Both concern DeepSeek tokenization regressions, but one is ByteLevel-to-Metaspace overriding and the other is a DeepSeek-Coder v1 tokenizer correctness gap; not the same concrete bug." + "reason": "Both involve tokenizer correctness, but the model families and failure mechanisms differ; not enough evidence of one underlying bug." }, { "left": "issue:43404", "right": "issue:43950", "accept": false, - "reason": "Different problems: lm_head weight tying in a model class vs corruption of non-persistent buffers during from_pretrained()." + "reason": "Distinct regressions: tied weights in a multimodal model vs non-persistent buffer corruption during `from_pretrained()`." }, { "left": "issue:45759", "right": "issue:46661", "accept": false, - "reason": "Both touch model loading, but one is from_config text_config unwrapping and the other is remote-code config_class lookup failure; distinct code paths." + "reason": "Different loading bugs: composite config unwrapping vs remote-code `config_class` lookup failure." }, { "left": "issue:43872", "right": "issue:44291", "accept": true, - "reason": "Same underlying `_is_hf_initialized` argument incompatibility surfacing in Int8Params/init_empty_weights loading paths." + "reason": "Both describe the same `_is_hf_initialized` argument regression during model init/loading, with 44291 giving the broader root-cause report." }, { "left": "issue:42915", "right": "issue:43872", "accept": false, - "reason": "Different issues: FineGrainedFP8Config failure vs bitsandbytes/_is_hf_initialized incompatibility." + "reason": "Different issues: FineGrainedFP8/Qwen3Moe behavior vs unexpected `_is_hf_initialized` argument handling." + }, + { + "left": "issue:45020", + "right": "issue:46661", + "accept": false, + "reason": "Both are generic loading regressions, but they concern different APIs and failure modes." }, { "left": "issue:43792", "right": "issue:44220", - "accept": false, - "reason": "Whisper model-runtime failure is too broad to match a specific _torch_extract_fbank_features() bug with confidence." + "accept": true, + "reason": "Both point to the same Whisper feature-extraction / fbank pipeline failure and likely duplicate the same underlying bug." }, { "left": "issue:43257", "right": "issue:45305", "accept": false, - "reason": "Both mention DeepSpeed, but one is weight conversion/loading and the other is gradient averaging with GAS; different bugs." + "reason": "Both involve DeepSpeed, but one is weight loading for Qwen3 MoE and the other is gradient averaging with GAS; different bugs." }, { "left": "issue:41553", "right": "issue:44991", "accept": false, - "reason": "Both are tokenizer-loading issues, but on different models and with different root causes; not the same defect." + "reason": "Both are tokenizer-loading problems, but they affect different models and appear to be separate regressions." }, { "left": "issue:42947", "right": "issue:45923", "accept": false, - "reason": "Both involve gradient checkpointing, but one is PEFT LoRA behavior and the other is a missing supports_gradient_checkpointing flag on a remote-code model." + "reason": "Checkpointing-related, but one is PEFT LoRA ineffectiveness and the other is a missing capability flag on a remote-code model." }, { "left": "issue:36331", "right": "issue:41093", "accept": false, - "reason": "Unexpected compute_loss kwarg and mask-shape IndexError are unrelated failures." + "reason": "Different training API and tensor-shape errors; not the same underlying problem." }, { "left": "issue:43940", "right": "issue:45313", - "accept": true, - "reason": "Both describe Qwen3-family DeepSpeed ZeRO-3 weight-loading failures with missing parameters; likely the same loader bug across close variants." + "accept": false, + "reason": "Both are ZeRO-3 load failures for Qwen variants, but the model targets and missing-weight patterns differ enough to treat separately." }, { "left": "issue:43854", "right": "issue:44863", "accept": false, - "reason": "Different model families and different load failures; no shared code-path evidence." + "reason": "Unrelated loading failures for different architectures." }, { "left": "issue:44479", "right": "issue:44560", - "accept": true, - "reason": "Both point to the same Qwen video-input regression in v5.3.0, with 44560 looking like a concrete manifestation of the broader video bug." - }, - { - "left": "issue:41553", - "right": "issue:44488", "accept": false, - "reason": "Different tokenizer-loading problems on different models; one is an error-message issue, the other is a load failure." + "reason": "Both involve Qwen video paths, but one is a broad regression across multiple models and the other is a specific StopIteration case; insufficiently same bug." }, { "left": "issue:44928", "right": "issue:45588", "accept": false, - "reason": "Gradient explosion from Qwen3.5 position_ids/SDPA fallback is unrelated to flash_attention.py crashing on s_aux=None." + "reason": "Different symptoms and code paths: RLHF NaNs from position IDs vs flash-attention attribute crash on `s_aux=None`." }, { "left": "issue:41093", "right": "issue:44805", "accept": true, - "reason": "Same generic mask-vs-tensor shape mismatch IndexError; these look like duplicate reports of the same failure." + "reason": "These are the same generic indexed-tensor vs mask-shape mismatch error pattern and look like duplicate reports of the same bug class." }, { "left": "issue:24643", "right": "issue:30064", "accept": false, - "reason": "Training DeepSpeed weight-dimension error and segmentation-map preprocessing failure are unrelated." + "reason": "Training with DeepSpeed vs image processor void-segmentation-map handling are unrelated." }, { "left": "issue:44530", "right": "issue:44863", "accept": false, - "reason": "PagedAttentionCache linear_attention crash is unrelated to NemotronH checkpoint loading." + "reason": "Different subsystems and failures: PagedAttentionCache linear_attention crash vs checkpoint loading on NemotronH." }, { "left": "issue:44863", "right": "issue:46402", "accept": false, - "reason": "Different architectures and different loading bugs." + "reason": "Both are model-loading issues, but for different models and different failure modes." }, { "left": "issue:43366", "right": "issue:43828", "accept": false, - "reason": "GGUF gpt-oss support and Phi-tiny-MoE autocast dtype mismatch are unrelated." + "reason": "Unrelated: GGUF/gpt-oss architecture support vs autocast dtype mismatch on Phi-tiny-MoE." }, { "left": "issue:43596", "right": "issue:46515", "accept": false, - "reason": "Deepspeed zero3/BertModel index error and torch 2.4.1 incompatibility are not the same bug." + "reason": "Index error with ZeRO-3 init and BertModel is not the same as torch 2.4.1 incompatibility." }, { "left": "issue:43366", "right": "issue:43531", "accept": false, - "reason": "Different Qwen/GGUF support issues with no shared concrete code path." + "reason": "Different Qwen/GGUF concerns; one is architecture support, the other a sliding_window bug." }, { "left": "issue:45325", "right": "issue:45902", "accept": false, - "reason": "Both concern rotary/position-related internals, but one is incorrect scaling of still-image position_ids and the other is uninitialized inv_freq after materialization." + "reason": "Both concern Qwen rotary/position handling, but one is video rope scaling and the other is uninitialized memory after materialization." }, { "left": "issue:45663", "right": "issue:45823", "accept": false, - "reason": "Same model family, but FSDP2 shared_kv_states logic and device_map auto device mismatch are different forward-pass bugs." + "reason": "Same model family, but different bugs: per-layer cache rebuild KeyError vs device mismatch in forward." } ] }, "evaluator_result": { "accept": true, - "feedback": "The cluster summary is well-grounded in the packet and stays conservative. The accepted soft-edge pairs are supported by closely matching titles/symptoms, while the rejected pairs avoid over-claiming duplicates across different models or code paths. No pull-request mergeability issues apply here." + "feedback": "The summary is grounded in the packet and stays mostly conservative. The accepted edges are reasonably supported by near-identical symptoms or clearly shared regressions, and the rejects generally avoid collapsing merely similar subsystems into one bug. The only slightly soft spot is the 41093/44805 pairing, which is a bit broad on title similarity alone, but still not an overreach given the identical error pattern." }, "evaluator_used": true, "retried": false, @@ -1968,14 +1980,14 @@ "issue:44960|issue:44977", "issue:46459|issue:46512" ], - "prepared_review_unit_hash": "b9b6ffbc54c6d987115a030ba70b71031e094fbb538f5ff5d62ede045edb8ac8", + "prepared_review_unit_hash": "57b268b45842e58c1294cacd8687c27f0b721f306436b8ad82e2644b3bd43fa8", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, @@ -1990,126 +2002,126 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is a broad cluster of mostly unrelated Transformers regressions. Only a couple of pairs look like true duplicates; the clearest duplicate is the non-persistent-buffer regression, and the SigLIP2 output-corruption pair also looks like the same underlying bug.", - "confidence": 0.81, - "canonical_issue_reason": "issue:43644 is the clearest exact duplicate target: it has the same non-persistent-buffer regression as 44534 and is specific enough to serve as the anchor issue.", + "summary": "Mostly a noisy similarity cluster: a few true duplicates around Transformers v5 loading/init regressions and repeated model-loading failures, but most pairs are unrelated model-specific bugs that only share broad subsystem vocabulary.", + "confidence": 0.82, + "canonical_issue_reason": "43644 is the clearest anchor for the core framework regression: it states the non-persistent-buffer corruption bug in general terms, and 44534 is a near-exact duplicate.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43644 is the best overall issue to anchor this cluster because it is the most precise, clearly duplicated bug report among the candidates.", + "best_issue_reason": "Best overall representative because it captures the underlying Transformers v5 regression at the framework level rather than a single model symptom.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43638", "right": "issue:46515", "accept": false, - "reason": "Different failures: Zero3/BertModel index error vs a torch version incompatibility report." + "reason": "Different failures: Deepspeed ZeRO3/BERT index error vs a general torch 2.4.1 incompatibility claim." }, { "left": "issue:43366", "right": "issue:44589", "accept": false, - "reason": "Unrelated problems: GGUF gpt-oss support vs missing Float8 storage type during loading." + "reason": "Unrelated problem areas: GGUF/gpt-oss support vs missing Float8 storage object during loading." }, { "left": "issue:43716", "right": "issue:44560", "accept": false, - "reason": "Different models and code paths: image preprocessor dtype mismatch vs Qwen3-vl video StopIteration." + "reason": "Different model paths and symptoms: image preprocessor dtype mismatch vs Qwen3-vl video StopIteration." }, { "left": "issue:44155", "right": "issue:45584", "accept": false, - "reason": "Both involve audio/sequence behavior, but the symptoms and model paths are different enough to treat as separate bugs." + "reason": "Different modalities and bugs: AudioFlamingo embedding/token leakage vs Whisper empty-transcription handling." }, { "left": "issue:43824", "right": "issue:43957", "accept": false, - "reason": "Import error for a missing class is not the same bug as meta-device loading failures." + "reason": "Import error for a specific class vs meta-device loading breakage; not the same underlying bug." }, { "left": "issue:43950", "right": "issue:45834", "accept": false, - "reason": "Non-persistent buffer corruption and long OCR index error are unrelated." + "reason": "Both are loading-related, but one is about non-persistent buffers and the other is a long-OCR index error in Kosmos2.5." }, { "left": "issue:42915", "right": "issue:43957", "accept": false, - "reason": "Different regressions: Qwen3Moe FP8 failure vs meta-device loading breakage." + "reason": "Qwen3Moe FP8 failure is unrelated to meta-device loading of some models." }, { "left": "issue:44560", "right": "issue:45685", "accept": false, - "reason": "Completely different issues: video StopIteration vs an MPS histogram kernel error." + "reason": "Different subsystems: Qwen3-vl video inference vs MPS histogram support in MoE." }, { "left": "issue:43645", "right": "issue:43646", - "accept": false, - "reason": "Related theme, but not clearly the same underlying bug; one is notebook-specific while the other is a broader custom-model init report." + "accept": true, + "reason": "Same custom-model initialization regression in Transformers 5; the second issue is just a shorter rewording of the first." }, { "left": "issue:44560", "right": "issue:44918", "accept": false, - "reason": "Different failures in Qwen3.5/Qwen3-vl workflows; no clear shared code-path bug." + "reason": "Both involve Qwen3-ish workflows, but one is video StopIteration and the other is TRL embedding unpacking." }, { "left": "issue:43550", "right": "issue:44910", "accept": false, - "reason": "Different attention/compile crashes on different models and mechanisms." + "reason": "Different model families and code paths: SDPA/torch.compile on Bamba vs flash-attention position_id handling on Qwen3.5." }, { "left": "issue:44683", "right": "issue:46515", "accept": false, - "reason": "Torch-version incompatibility reports, but on opposite version ranges and different failure modes." + "reason": "torch>=2.9 flex_attention compile failure is unrelated to a torch 2.4.1 incompatibility report." }, { "left": "issue:42617", "right": "issue:43366", "accept": false, - "reason": "Different tasks: running 3d_parallel.py vs GGUF gpt-oss support." + "reason": "3d_parallel.py runtime problem is unrelated to GGUF gpt-oss support." }, { "left": "issue:36010", "right": "issue:42915", "accept": false, - "reason": "ImportError for GenerationMixin is unrelated to Qwen3Moe FP8 loading failure." + "reason": "Importing GenerationMixin from transformers.generation is unrelated to Qwen3Moe FP8 loading." }, { "left": "issue:45812", "right": "issue:46489", "accept": false, - "reason": "Both are tokenizer regressions, but they affect different model families and do not clearly share the same root cause." + "reason": "Both are tokenizer regressions, but they target different model families and likely different code paths." }, { "left": "issue:45823", "right": "issue:46515", "accept": false, - "reason": "Device-mismatch during Gemma4 forward is unrelated to a generic torch version incompatibility report." + "reason": "Gemma4 device-mismatch with device_map=auto is unrelated to a torch 2.4.1 incompatibility." }, { "left": "issue:44560", "right": "issue:44805", "accept": false, - "reason": "Different symptom and likely different bug: video StopIteration vs mask shape mismatch." + "reason": "Different concrete bugs: Qwen3-vl video handling vs generic mask/tensor shape mismatch." }, { "left": "issue:44661", "right": "issue:45823", "accept": false, - "reason": "Different areas: tokenizer mapping expansion failure vs Gemma4 PLE device mismatch." + "reason": "Model-registration logic for tokenizer mappings is unrelated to Gemma4 PLE device placement." }, { "left": "issue:38175", "right": "issue:43994", - "accept": true, - "reason": "Same model family and same broken inference behavior on siglip2-base-patch16-224; these look like the same underlying output-corruption bug." + "accept": false, + "reason": "Same model family, but one reports zero probabilities and the other nonsensical outputs; not enough to treat as the same bug." }, { "left": "issue:43632", @@ -2121,91 +2133,91 @@ "left": "issue:44877", "right": "issue:45030", "accept": false, - "reason": "Both are config-validation issues, but they target different model configs and likely different fixes." + "reason": "Both mention config validation, but they are different model/config failures with no clear shared bug." }, { "left": "issue:43638", "right": "issue:44661", "accept": false, - "reason": "Zero3/BertModel init failure is unrelated to add-new-model-like tokenizer mapping behavior." + "reason": "Deepspeed ZeRO3/BERT initialization error is unrelated to tokenizer-mapping registration." }, { "left": "issue:43596", "right": "issue:44661", "accept": false, - "reason": "Different bugs: Zero3 with BertModel vs tokenizer-mapping failure." + "reason": "Different failure modes: empty-shape index error during ZeRO3 init vs add-new-model-like tokenizer mapping issue." }, { "left": "issue:43317", "right": "issue:44756", "accept": false, - "reason": "Device-map/offload loading bug is unrelated to disabling mmap for Strix Halo OOM." + "reason": "Device-map/offload loading bug is unrelated to mmap OOM on Strix Halo." }, { "left": "issue:44661", "right": "issue:44805", "accept": false, - "reason": "Tokenizer-mapping failure and mask shape mismatch are different bugs." + "reason": "Model-registration bug and tensor mask shape error are unrelated." }, { "left": "issue:42491", "right": "issue:43720", "accept": false, - "reason": "Different subsystems and failure modes: Qwen3_moe LoRA compatibility vs BitNet packed-weight loading." + "reason": "Qwen3_moe LoRA compatibility across branches is unrelated to BitNet packed-weight loading." }, { "left": "issue:43644", "right": "issue:44534", "accept": true, - "reason": "Exact same regression: non-persistent buffers are being filled with junk in Transformers 5.0.0." + "reason": "Near-exact duplicate reports of the same non-persistent-buffer junk regression in Transformers v5." }, { "left": "issue:38175", "right": "issue:39692", "accept": false, - "reason": "The docs-example issue is broader and includes multiple errors; it is not clearly the same root bug as the zero-probabilities report." + "reason": "Both concern SigLIP2, but one is an inference-output anomaly and the other is a docs/example mismatch plus quantization error." }, { "left": "issue:41720", "right": "issue:42915", "accept": false, - "reason": "Qwen3 device-map CUDA assert and Qwen3Moe FP8 failure are different problems." + "reason": "CUDA assert under auto device mapping is unrelated to Qwen3Moe FP8 configuration failures." }, { "left": "issue:42915", "right": "issue:44756", "accept": false, - "reason": "FP8 configuration failure is unrelated to mmap/OOM behavior." + "reason": "FP8 config failure and mmap/OOM on Strix Halo are different issues." }, { "left": "issue:42491", "right": "issue:45276", "accept": false, - "reason": "LoRA cross-version incompatibility is unrelated to resize_token_embeddings not updating Gemma4 embeddings." + "reason": "LoRA branch compatibility for qwen3_moe is unrelated to Gemma4 token-embedding resize behavior." }, { "left": "issue:44451", "right": "issue:44488", - "accept": false, - "reason": "Very similar loading failures, but the reports are for different model checkpoints and the root cause is not clear enough to merge confidently." + "accept": true, + "reason": "Both report the current/latest version failing to load closely related BERT checkpoints, with the same loading symptom." }, { "left": "issue:44960", "right": "issue:44977", "accept": false, - "reason": "GLM5 is a broad model report, while the other is a Qwen3.5 flash-attention generation bug; different issues." + "reason": "GLM5 issue and Qwen3.5 flash-attention generation bug are separate model-specific regressions." }, { "left": "issue:46459", "right": "issue:46512", "accept": false, - "reason": "Both concern dtype handling, but one is silent dtype loss when loading checkpoints and the other is stale config.dtype after from_config; different bugs." + "reason": "One is about dtype being ignored for composite checkpoint loading; the other is stale config.dtype after from_config." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays appropriately cautious. The non-persistent-buffer pair is clearly supported, and the SigLIP2 pair is a reasonable same-bug call from the titles. The rejected pairs are generally conservative and do not overclaim duplication from the available evidence." + "feedback": "The clustering judgment is grounded in the packet. The two accepted pairs are well supported by near-identical titles and clearly overlapping regression descriptions, and the rejected pairs are generally conservative rather than overclaiming duplicates. The summary also stays appropriately cautious about most pairs being broad-similarity noise rather than true duplicates." }, "evaluator_used": true, "retried": false, @@ -2305,20 +2317,20 @@ "issue:36010|issue:39404", "issue:45663|issue:45859" ], - "prepared_review_unit_hash": "40e290c426cc0689d642e5ad1283d5dd943a1fbdf2fbcca43bfc7bfa3bc43728", + "prepared_review_unit_hash": "07e9941bd651b84fcf530a0b43fb5229349dfcbe6c476d6df249226970135739", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 29, - "serialized_chars": 35910, + "serialized_chars": 35909, "estimated_input_tokens": 8978, "estimated_eval_tokens": 18212 }, @@ -2327,196 +2339,187 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly a heterogeneous set of unrelated Transformers bugs, but there is one clear duplicate pair around `resize_token_embeddings` not propagating to model-specific embeddings. I accepted only that pair; the rest look like separate issues.", - "confidence": 0.87, - "canonical_issue_reason": "issue:45276 is the broader and more representative thread in the accepted duplicate pair: it describes the same `resize_token_embeddings` propagation failure and covers the general Gemma4 case, while 45335 is a narrower T5Gemma manifestation.", + "summary": "This set is mostly a grab bag of unrelated Transformers issues. The soft pairs are generally same-subsystem but not the same underlying bug; none look safe to merge as duplicates.", + "confidence": 0.88, + "canonical_issue_reason": "issue:45276 is the closest representative because it matches the strongest near-duplicate pair and captures a concrete, recurring API regression around `resize_token_embeddings` failing to update nested embeddings.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45276 is the best representative issue in this cluster because it captures the shared embedding-resize bug more generally and is the natural target for deduping the closely related T5Gemma report.", + "best_issue_reason": "issue:45276 is the best stand-in for the cluster: it is specific, clearly stated, and sits in the tightest soft-similarity pair among the issues listed.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45313", "right": "issue:45357", "accept": false, - "reason": "Same model family, but different concrete bugs: one is DeepSpeed weight loading for `language_model`, the other is incorrect visual-encoder keys in `save_pretrained`." + "reason": "Both are Qwen3.5 multimodal regressions, but one is ZeRO-3 language_model loading and the other is incorrect visual encoder keys on save; different code paths." }, { "left": "issue:45507", "right": "issue:46468", "accept": false, - "reason": "Different models and failure modes; no evidence they are the same underlying bug or fix." + "reason": "Different models and failures: GraniteMoEHybrid invalid method call vs Nemotron example breakage; no clear shared underlying bug." }, { "left": "issue:43644", "right": "issue:43645", "accept": false, - "reason": "Both are Transformers 5.0 regressions, but one is junk buffers and the other is Jupyter/custom-model initialization; distinct code paths." + "reason": "Both mention Transformers 5.0 breakage, but one is non-persistent buffer initialization and the other is custom-model initialization in Jupyter; distinct bugs." }, { "left": "issue:44488", "right": "issue:44991", "accept": false, - "reason": "Both concern loading specific tokenizer/model repos, but they point to different models and likely different root causes." + "reason": "Both are tokenizer/model-loading failures, but for different models and likely different causes; not the same concrete issue." }, { "left": "issue:43856", "right": "issue:43873", "accept": false, - "reason": "One is Qwen3 MoE training memory usage, the other is quantization/offloading behavior; too different to treat as duplicates." + "reason": "One is Qwen3 MoE training memory inefficiency, the other is quantization offloading behavior; related area but not the same bug." }, { "left": "issue:43856", "right": "issue:44387", "accept": false, - "reason": "Both mention memory/OOM, but one is MoE training efficiency and the other is int4 reserved-memory regression; not the same bug." + "reason": "Both involve memory, but one is MoE training inefficiency and the other is int4 quantization reserving extra CUDA memory; different failure modes." }, { "left": "issue:43257", "right": "issue:43854", "accept": false, - "reason": "Different models and loading scenarios; no shared concrete failure path." + "reason": "Different models and load paths: Qwen3 MoE accelerate/deepspeed conversion vs GLM-4.7-Flash unit test loading." }, { "left": "issue:39692", "right": "issue:43994", "accept": false, - "reason": "Both involve SigLIP2, but one is a documentation example with model/processor mismatch and quantization failure while the other is incorrect outputs in AutoModel/pipeline." + "reason": "Both are SigLIP2-related, but one is a broken documentation example and the other is incorrect model/pipeline outputs; not clearly the same bug." }, { "left": "issue:43854", "right": "issue:43931", "accept": false, - "reason": "Different models and different loading errors; not the same underlying issue." + "reason": "Unrelated model-loading failures for different models; no shared code-path evidence." }, { "left": "issue:42491", "right": "issue:45292", "accept": false, - "reason": "Qwen3 MoE LoRA compatibility on hf5.x is unrelated to `resize_token_embeddings` not updating output embeddings." + "reason": "Qwen3 MoE hf4.x\u2192hf5.x LoRA compatibility is unrelated to `resize_token_embeddings` not updating output embeddings." }, { "left": "issue:43708", "right": "issue:44075", "accept": false, - "reason": "Trainer checkpoint resume/max_steps logic is unrelated to optimizer arguments being ignored." + "reason": "Trainer checkpoint-step math and SGD argument handling are separate training bugs." }, { "left": "issue:42491", "right": "issue:43931", "accept": false, - "reason": "Different model families and distinct bugs." + "reason": "Different models and failure mechanisms; not a duplicate." }, { "left": "issue:44960", "right": "issue:46050", "accept": false, - "reason": "GLM5 loading concerns are unrelated to quantized save/load corruption in sinq." + "reason": "GLM5 issue vs sinq quantized save/load corruption; different model families and bug types." }, { "left": "issue:44164", "right": "issue:44291", "accept": false, - "reason": "Both touch model serialization/loading, but one is `extra_state` handling and the other is `init_empty_weights` with an unexpected argument; different fixes." + "reason": "Both involve loading/saving internals, but extra_state handling and `init_empty_weights`/_is_hf_initialized are distinct bugs." }, { "left": "issue:45488", "right": "issue:45812", "accept": false, - "reason": "Both are tokenizer regressions, but they affect different model families and different mechanisms." + "reason": "Both are tokenizer regressions in v5, but one is LlamaTokenizer pre-tokenizer override for DeepSeek and the other is AutoTokenizer ID corruption for Granite; different root causes." }, { "left": "issue:45362", "right": "issue:45406", "accept": false, - "reason": "Different model families and distinct crashes; not the same code-path problem." + "reason": "Different entrypoints and models: Qwen3.5 chat crash vs Gemma4Processor missing `_tokenizer` in transformers serve." }, { "left": "issue:45276", "right": "issue:45335", - "accept": true, - "reason": "Same underlying `resize_token_embeddings` propagation bug: both report that resizing does not update downstream embedding/output modules in related Gemma/T5Gemma models." + "accept": false, + "reason": "Very similar symptom, but model families and affected submodules differ; insufficient evidence they are the same underlying bug." }, { "left": "issue:44910", "right": "issue:45588", "accept": false, - "reason": "Both involve flash attention, but one is a CUDA illegal-memory-access bug and the other is an AttributeError on `s_aux=None`; separate failures." + "reason": "Both touch flash attention, but one is Qwen3.5 packed-sequence misinterpretation and the other is `s_aux=None` AttributeError for sink-less models; distinct issues." }, { "left": "issue:44402", "right": "issue:44488", "accept": false, - "reason": "Tokenizer/vocab issues, but on different models and with different symptoms." + "reason": "Tokenizer vocab-size mismatch vs inability to load a different model; not the same bug." }, { "left": "issue:43425", "right": "issue:43957", "accept": false, - "reason": "Torch compatibility and `torch.device(\"meta\")` loading are unrelated issues." + "reason": "Torch 2.10 compatibility and `torch.device('meta')` loading breakage are separate compatibility problems." }, { "left": "issue:43404", "right": "issue:43454", "accept": false, - "reason": "Both mention untied `lm_head`, but they are different model implementations and there is no evidence of one shared duplicate bug." + "reason": "Both mention untied `lm_head`, but they affect different model classes and there is no evidence of one shared defect across them." }, { "left": "issue:43873", "right": "issue:44387", "accept": false, - "reason": "Quantization is common to both, but one is offloading behavior and the other is reserved-memory/OOM regression; not the same defect." + "reason": "Both are memory/quantization related, but one is offloading behavior and the other is increased reserved memory under int4 quantization; not the same bug." }, { "left": "issue:39692", "right": "issue:44960", "accept": false, - "reason": "Completely different model families and bug types." + "reason": "SigLIP2 documentation errors are unrelated to a GLM5 model-loading issue." }, { "left": "issue:41720", "right": "issue:44155", "accept": false, - "reason": "Unrelated model/runtime bugs: Qwen3 device-mapping assert vs AudioFlamingo3 track leakage." + "reason": "Qwen3 auto device mapping CUDA assert and AudioFlamingo3 token leakage are unrelated." }, { "left": "issue:42947", "right": "issue:45704", "accept": false, - "reason": "Gradient checkpointing/LoRA is unrelated to a T5 apex RMSNorm memory leak." + "reason": "Gradient checkpointing with PEFT LoRA and T5 Apex RMSNorm memory leak are different problems." }, { "left": "issue:45663", "right": "issue:45759", "accept": false, - "reason": "Different composite-model issues: per-layer KV state rebuild in Gemma-4 training vs `text_config` unwrapping for Qwen multimodal configs." + "reason": "Gemma-4 FSDP2 shared_kv_states KeyError and Qwen 3.5/3.6 text_config unwrapping are different model/config bugs." }, { "left": "issue:44291", "right": "issue:44756", "accept": false, - "reason": "Different loading/runtime problems with no shared underlying code path." + "reason": "Unexpected `_is_hf_initialized` handling and Strix Halo mmap OOM are unrelated." }, { "left": "issue:36010", "right": "issue:39404", "accept": false, - "reason": "Importing `GenerationMixin` and Whisper pipeline `return_language` are unrelated." - }, - { - "left": "issue:45663", - "right": "issue:45859", - "accept": false, - "reason": "Different architectures and failure modes; no duplicate evidence." + "reason": "ImportError for `GenerationMixin` and Whisper pipeline `return_language` regression are unrelated." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The cluster summary is grounded in the packet and stays conservative. The accepted 45276/45335 pair is well-supported by near-identical `resize_token_embeddings` propagation symptoms and looks mergeable into one fix. The other soft-edge rejections do not overclaim duplication and are appropriately cautious." - }, - "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 }, @@ -2618,245 +2621,248 @@ "issue:44589|issue:45084", "issue:44479|issue:44805" ], - "prepared_review_unit_hash": "21918bc8ab6d95ab55eaa8671fefcb5b01d25027fed0ff18a6989e2e54cf0a6f", + "prepared_review_unit_hash": "f31190a2c05b2fd86aaeca6f173d54fc9ae5f52848ac825bab59af6f1a014ea8", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 34, - "serialized_chars": 37890, - "estimated_input_tokens": 9473, - "estimated_eval_tokens": 19202 + "serialized_chars": 37886, + "estimated_input_tokens": 9472, + "estimated_eval_tokens": 19200 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly a set of unrelated Transformers regressions and compatibility reports. The soft-similarity links are largely false positives: they share broad themes like model loading, tokenization, or dtype/runtime issues, but the concrete failures differ. No PRs are present.", - "confidence": 0.91, - "canonical_issue_reason": "issue:43611 is the broadest representative of the custom-model loading regression family in this cluster; several nearby reports touch model initialization/loading, but they describe different breakages rather than the same defect.", + "summary": "Mostly a heterogeneous cluster of Transformers issues with a few loose thematic overlaps. The only clear duplicate family is the resize_token_embeddings / output_embeddings regression across model variants (Gemma4, generic output_embeddings, T5Gemma); the rest are different bugs or different code paths.", + "confidence": 0.81, + "canonical_issue_reason": "issue:45292 is the broadest, most generic formulation of the resize_token_embeddings/output_embeddings regression, making it the best anchor for the one clear duplicate family in this set.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43611 is the best single issue to anchor the cluster because it captures the most general and recurring failure mode here (Transformers 5.x breaking custom model loading), even though the cluster as a whole should not be merged into one duplicate set.", + "best_issue_reason": "issue:45292 is the best representative issue overall because it states the underlying bug cleanly and subsumes the more model-specific variants.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44292", "right": "issue:44912", "accept": false, - "reason": "Both are quantization/loading regressions, but one is Qwen-3-8B-NVFP4 failure and the other is MXFP4 fallback behavior on git-oss-20b; different models and failure modes." + "reason": "Both are quantization/runtime loading failures, but they involve different models, quantization modes, and likely different code paths." }, { "left": "issue:44387", "right": "issue:44756", "accept": false, - "reason": "One is int4 CUDA reserved-memory/OOM under Transformers 5.x; the other is mmap OOM on Strix Halo. Same symptom class, different root cause." + "reason": "One is int4 memory growth/OOM, the other is mmap-related OOM on Strix Halo; same symptom, different cause." }, { "left": "issue:43957", "right": "issue:44756", "accept": false, - "reason": "Meta-device model loading incompatibility is distinct from Strix Halo mmap-related OOM." + "reason": "These are separate OOM/loading regressions: torch.device('meta') model loading vs disabling mmap on Strix Halo." }, { "left": "issue:43335", "right": "issue:45663", "accept": false, - "reason": "Single-layer SwitchTransformers config bug is unrelated to Gemma4 FSDP2 shared_kv_states KeyError." + "reason": "Sparse-layer construction in SwitchTransformers and Gemma4 FSDP2 shared_kv_states are unrelated bugs." }, { "left": "issue:44155", "right": "issue:45823", "accept": false, - "reason": "AudioFlamingo3 batched inference token leak is unrelated to Gemma4 PLE device placement mismatch." + "reason": "AudioFlamingo3 batched inference leakage and Gemma4 PLE device mismatch are different model-specific forward-path issues." }, { "left": "issue:42947", "right": "issue:43381", "accept": false, - "reason": "PEFT LoRA gradient checkpointing ineffectiveness is not the same as gradient checkpointing being disallowed in eval mode." + "reason": "Both mention gradient checkpointing, but one is LoRA ineffectiveness and the other is eval-mode incompatibility." }, { "left": "issue:44863", "right": "issue:46468", "accept": false, - "reason": "NemotronH checkpoint loading failure is unrelated to the broken official Nemotron example." + "reason": "NemotronH checkpoint loading and a broken official Nemotron example are not the same concrete bug." }, { "left": "issue:45276", "right": "issue:45292", - "accept": false, - "reason": "These are closely related resize_token_embeddings regressions, but one is Gemma4-specific and the other is a broader output_embeddings issue; not clearly the same bug." + "accept": true, + "reason": "Same underlying regression: resize_token_embeddings fails to update output-related embeddings; 45276 is just a Gemma4-specific manifestation." }, { "left": "issue:45685", "right": "issue:46468", "accept": false, - "reason": "MPS histogram Int support in MoE is unrelated to the Nemotron example breakage." + "reason": "An MPS histogram kernel limitation is unrelated to the Nemotron example failure." }, { "left": "issue:44479", "right": "issue:44877", "accept": false, - "reason": "Video input regression for Qwen VL models is unrelated to strict-config loading of granite_speech." + "reason": "Video-input regression in Qwen VL models vs strict config blocking granite_speech loading are different issues." }, { "left": "issue:44164", "right": "issue:44756", "accept": false, - "reason": "extra_state save/from_pretrained handling is unrelated to Strix Halo mmap OOM." + "reason": "save/from_pretrained extra_state handling is unrelated to the Strix Halo mmap OOM." }, { "left": "issue:43381", "right": "issue:44928", "accept": false, - "reason": "Eval-mode checkpointing restriction is unrelated to the RLHF NaN/SDPA fallback issue." + "reason": "Eval-mode gradient checkpointing incompatibility and Qwen3.5 RLHF NaNs from 3D position_ids are different failures." }, { "left": "issue:43854", "right": "issue:45216", "accept": false, - "reason": "Broken unit-test model loading for GLM-4.7-Flash is unrelated to Qwen3.5 save_pretrained regression." + "reason": "GLM-4.7-Flash test loading failure and Qwen3.5 save_pretrained regression are separate model/save-path bugs." }, { "left": "issue:45588", "right": "issue:45663", "accept": false, - "reason": "Flash attention crash on s_aux=None is a different code path than Gemma4 FSDP2 shared_kv_states rebuilding." + "reason": "flash_attention.py crashing on s_aux=None is unrelated to Gemma4 FSDP2 shared_kv_states rebuilds." }, { "left": "issue:45684", "right": "issue:45698", "accept": false, - "reason": "Read-only permission propagation during save_pretrained is unrelated to loading the wrong custom module after save_pretrained." + "reason": "Both involve custom model save/load flows, but one is read-only permission propagation and the other is loading the wrong custom module." }, { "left": "issue:42796", "right": "issue:46326", "accept": false, - "reason": "Tokenizer BPE marker handling bug is unrelated to continue_final_message behavior on mistral_common tokenizers." + "reason": "Tokenizer BPE marker handling and continue_final_message behavior are different tokenizer regressions." }, { "left": "issue:43828", "right": "issue:43957", "accept": false, - "reason": "Both are dtype mismatch reports, but one is torch.autocast on Phi-tiny-MoE and the other is meta-device loading on different models; not the same bug." + "reason": "Different root causes: autocast dtype mismatch for Phi-tiny-MoE vs meta-device loading breakage." }, { "left": "issue:45292", "right": "issue:45335", - "accept": false, - "reason": "Both concern resize_token_embeddings propagation, but one targets output_embeddings generally and the other is decoder.embed_tokens in t5gemma; not enough evidence they are the same defect." + "accept": true, + "reason": "Same resize_token_embeddings regression; 45335 is a T5Gemma-specific case where decoder.embed_tokens is not updated." }, { "left": "issue:43646", "right": "issue:43708", "accept": false, - "reason": "Custom model initialization in notebooks is unrelated to Trainer resume/max_steps miscalculation." + "reason": "Custom model initialization under Transformers 5.0.0 and Trainer resume max_steps calculation are unrelated." }, { "left": "issue:44402", "right": "issue:44451", "accept": false, - "reason": "Token IDs outside vocab_size for rmihaylov/bert-base-bg is unrelated to loading vesteinn/ScandiBERT." + "reason": "Tokenizer vocab-size mismatch for one model and inability to load another model are not the same bug." }, { "left": "issue:43645", "right": "issue:44164", "accept": false, - "reason": "Jupyter custom-model initialization breakage is separate from save/from_pretrained extra_state handling." + "reason": "Notebook-defined custom model initialization and extra_state save/load handling are different code paths." }, { "left": "issue:44589", "right": "issue:44756", "accept": false, - "reason": "Float8 storage lookup failure is unrelated to Strix Halo mmap OOM." + "reason": "Float8 storage lookup failure is unrelated to mmap-induced OOM." }, { "left": "issue:43425", "right": "issue:43828", "accept": false, - "reason": "Torch 2.10 compatibility is a general version issue, while the other is a specific autocast dtype mismatch on Phi-tiny-MoE." + "reason": "Torch 2.10 incompatibility is broader environment breakage, not the same as a specific autocast dtype mismatch." }, { "left": "issue:45812", "right": "issue:46315", "accept": false, - "reason": "Granite tokenizer ID regression is unrelated to slow convert_tokens_to_ids performance regression." + "reason": "Wrong Granite token IDs and slow convert_tokens_to_ids are distinct tokenizer issues, one correctness and one performance." }, { "left": "issue:46315", "right": "issue:46491", "accept": false, - "reason": "Performance regression in convert_tokens_to_ids is unrelated to CodeLlama leading-space round-trip regression." + "reason": "Performance regression in token conversion is unrelated to CodeLlama round-trip spacing behavior." }, { "left": "issue:39404", "right": "issue:45701", "accept": false, - "reason": "Whisper return_language pipeline regression is unrelated to tokenizer version-dependent tokenization changes." + "reason": "Whisper return_language pipeline breakage and tokenization changes in version upgrades are not the same bug." }, { "left": "issue:43611", "right": "issue:43645", "accept": false, - "reason": "Both involve custom-model loading, but one is base_model_prefix loading regression and the other is notebook-defined model initialization; different failure modes." + "reason": "Both concern Transformers 5.0.0 custom-model behavior, but one is base_model_prefix loading and the other is notebook-defined model initialization." }, { "left": "issue:39404", "right": "issue:46665", "accept": false, - "reason": "Whisper pipeline return_language bug is unrelated to inconsistent from_pretrained behavior inside Python classes." + "reason": "Whisper pipeline return_language and from_pretrained inside Python classes are unrelated." }, { "left": "issue:44530", "right": "issue:45084", "accept": false, - "reason": "PagedAttentionCache invalid group type on Qwen3.5 is unrelated to compile non template nodes." + "reason": "PagedAttentionCache linear_attention group handling and compile-non-template-node errors are different failures." }, { "left": "issue:44683", "right": "issue:45823", "accept": false, - "reason": "Compiled flex_attention on torch >= 2.9 is unrelated to Gemma4 PLE device mismatch." + "reason": "Compiled flex_attention on torch>=2.9 and Gemma4 device mismatch are unrelated." }, { "left": "issue:43425", "right": "issue:45823", "accept": false, - "reason": "Torch 2.10 compatibility and Gemma4 device_map mismatch are different regressions." + "reason": "Torch version incompatibility is not the same concrete bug as Gemma4 PLE device mismatch." }, { "left": "issue:43425", "right": "issue:43716", "accept": false, - "reason": "Torch version compatibility is unrelated to Mistral-3 image preprocessor dtype mismatch." + "reason": "General torch incompatibility and Mistral-3 image-preprocessor dtype mismatch are different problems." }, { "left": "issue:44589", "right": "issue:45084", "accept": false, - "reason": "Float8 storage lookup failure is unrelated to template-node compilation error." + "reason": "Float8 storage object lookup and non-template node compilation errors are unrelated." }, { "left": "issue:44479", "right": "issue:44805", "accept": false, - "reason": "Qwen VL video-input regression is unrelated to an index/mask shape mismatch error." + "reason": "Qwen VL video regression and mask-shape IndexError do not look like the same underlying issue." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet, and the two accepted links are well supported by the titles and soft-similarity signal: issue:45276/45292 and issue:45292/45335 are clearly the same resize_token_embeddings/output_embeddings family. The other soft-edge rejections stay conservative and do not overclaim identity across clearly different model- or code-path-specific failures." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -2955,220 +2961,220 @@ "issue:42915|issue:44530", "issue:44451|issue:44987" ], - "prepared_review_unit_hash": "796aec3d78326e847de37da7ab4043d961a92dba44e653d8b88e407947e86719", + "prepared_review_unit_hash": "cdfe49bc87154dfd7cc9dc23f402d608644b81ed43166a0af4ac267b8a6f8793", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 37108, - "estimated_input_tokens": 9277, - "estimated_eval_tokens": 18810 + "serialized_chars": 37109, + "estimated_input_tokens": 9278, + "estimated_eval_tokens": 18812 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These items are mostly separate regression reports spanning tokenizer loading, model loading, quantization, compile/export, and doc issues. I don\u2019t see any soft pair that clearly shares the same underlying bug/change, so I rejected all candidate duplicates. No pull requests are present in this cluster.", - "confidence": 0.87, - "canonical_issue_reason": "issue:44462 is the broadest tokenizer-loading regression here and the most plausible umbrella representative for the small tokenizer-related subset.", + "summary": "This is a heterogeneous cluster of mostly unrelated Transformers issues (tokenizer/load regressions, multimodal model loading, quantization, compile/runtime errors, and docs). None of the soft pairs look like the same concrete underlying bug/change, so all pairings should stay separate.", + "confidence": 0.93, + "canonical_issue_reason": "Issue #43611 is the broadest representative of the cluster\u2019s recurring theme of Transformers v5 loading regressions, but the overall set is still too mixed to treat as a single duplicate target.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44462 is the best single representative because it is generic, user-facing, and closest to the common loading-regression theme across the set.", + "best_issue_reason": "#43611 is the most general issue here; it\u2019s a clear v5 model-loading regression and is closer to the dominant pattern than the more model-specific reports. Still, this cluster does not form a strong duplicate group.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43906", "right": "issue:44530", "accept": false, - "reason": "Different failures: isolated repro of an older issue vs a Qwen3.5 cache/group-type crash." + "reason": "One is an isolated reproduction of a previously reported issue; the other is a Qwen3.5 PagedAttentionCache crash. Different code paths and symptoms." }, { "left": "issue:43901", "right": "issue:44509", "accept": false, - "reason": "Both are docs-related, but they cover different pipeline-task deprecations and should stay separate." + "reason": "Both are docs-related, but they cover different pipeline API mismatches: return_all_scores vs removed task names in v5." }, { "left": "issue:43425", "right": "issue:44918", "accept": false, - "reason": "Torch 2.10 compatibility is unrelated to the TRL/Qwen3.5 embedding unpacking failure." + "reason": "Torch 2.10 compatibility is unrelated to the Qwen3.5 TRL SFT embedding-unpacking failure." }, { "left": "issue:43994", "right": "issue:44079", "accept": false, - "reason": "Model/pipeline output quality bug vs ModelOutput key assignment bug; different code paths." + "reason": "SigLIP2 incorrect outputs and ModelOutput key assignment bugs are different problems in different layers." }, { "left": "issue:44265", "right": "issue:44898", "accept": false, - "reason": "torch.export/torch_compilable_check issue is distinct from Perceiver interpolation at non-default resolution." + "reason": "torch.export failures with torch_compilable_check are unrelated to Perceiver non-default-resolution interpolation failures." }, { "left": "issue:44671", "right": "issue:44991", "accept": false, - "reason": "CamemBERT masked-LM behavior regression and tokenizer-loading failure for a different model are not the same bug." + "reason": "CamemBERT masked-LM prediction regression and EMBEDDIA tokenizer loading failure are distinct tokenizer/model issues." }, { "left": "issue:36032", "right": "issue:43452", "accept": false, - "reason": "Different tokenizer-loading problems affecting different models and error surfaces." + "reason": "T5Tokenizer special-token method conflict and gguf_file loading breakage are unrelated tokenizer/load paths." }, { "left": "issue:39404", "right": "issue:44912", "accept": false, - "reason": "Whisper pipeline return_language regression is unrelated to MXFP4 quantization fallback." + "reason": "Whisper pipeline return_language regression is unrelated to MXFP4 quantization fallback behavior." }, { "left": "issue:44164", "right": "issue:45759", "accept": false, - "reason": "save/from_pretrained extra_state handling and composite config unwrapping are separate loading issues." + "reason": "extra_state save/load handling and AutoModelForCausalLM config unwrapping are different serialization/config bugs." }, { "left": "issue:46050", "right": "issue:46326", "accept": false, - "reason": "Quantized model persistence bug vs mistral_common continue_final_message behavior; not the same underlying defect." + "reason": "sinq quantization save/load corruption and mistral_common continue_final_message behavior are unrelated." }, { "left": "issue:43299", "right": "issue:43931", "accept": false, - "reason": "Same model family, but one is a broad v5 loading regression and the other is a specific checkpoint shape mismatch." + "reason": "A generic Qwen3VL MoE loading break in v5.0.0.dev0 is not the same as a specific weight-shape mismatch on a Qwen3-VL model." }, { "left": "issue:43906", "right": "issue:46326", "accept": false, - "reason": "Unrelated tokenizer/pipeline reproducer versus mistral_common chat-template behavior." + "reason": "An isolated reproduction report is not the same bug as mistral_common tokenizer continue_final_message handling." }, { "left": "issue:44991", "right": "issue:45081", "accept": false, - "reason": "Both are tokenizer loading failures, but they hit different tokenizer stacks and model families." + "reason": "Different tokenizer-loading failures affecting different models and different internals." }, { "left": "issue:43856", "right": "issue:44756", "accept": false, - "reason": "Training memory inefficiency and Strix Halo mmap OOM are different root causes." + "reason": "Qwen3 MoE training memory inefficiency and Strix Halo mmap OOM avoidance are unrelated performance issues." }, { "left": "issue:43596", "right": "issue:44683", "accept": false, - "reason": "Deepspeed zero3/BertModel index error is unrelated to flex_attention compilation on torch >=2.9." + "reason": "DeepSpeed ZeRO-3/BertModel index error is unrelated to compiled flex_attention on torch >=2.9." }, { "left": "issue:44164", "right": "issue:44479", "accept": false, - "reason": "Extra_state serialization and video-input regression are separate bugs." + "reason": "extra_state serialization and Qwen video-input regression are separate bugs." }, { "left": "issue:44488", "right": "issue:45081", "accept": false, - "reason": "Different model/tokenizer loading failures with no shared concrete code-path." + "reason": "Loading sleng-bert and patch_mistral_regex crashes are different tokenizer-loading failures." }, { "left": "issue:44488", "right": "issue:44987", "accept": false, - "reason": "Both are load failures, but for different repositories and different failure modes." + "reason": "These are both model-loading issues, but they affect different repositories and likely different tokenizer/model code paths." }, { "left": "issue:43611", "right": "issue:43644", "accept": false, - "reason": "base_model_prefix loading breakage and junk-filled non-persistent buffers are unrelated regressions." + "reason": "Both mention Transformers v5, but one is base_model_prefix loading and the other is non-persistent buffer initialization\u2014different defects." }, { "left": "issue:44462", "right": "issue:46489", "accept": false, - "reason": "Both concern tokenizer behavior, but one is generic tokenizer.json selection and the other is a specific DeepSeek-Coder output regression." + "reason": "Ignoring tokenizer.json on load and DeepSeek-Coder tokenizer output regressions are not the same underlying problem." }, { "left": "issue:42796", "right": "issue:43404", "accept": false, - "reason": "Tokenizer BPE-marker handling and Mistral3 lm_head weight tying are different bugs." + "reason": "Tokenizer BPE-marker handling and Mistral3 lm_head weight tying are unrelated." }, { "left": "issue:43716", "right": "issue:45084", "accept": false, - "reason": "Image preprocessor dtype mismatch is unrelated to template-node compilation failure." + "reason": "Image preprocessor/model dtype mismatch is unrelated to a template compilation TypeError." }, { "left": "issue:36032", "right": "issue:42915", "accept": false, - "reason": "Tokenizer method conflict and Qwen3MoE FineGrainedFP8Config failure are not the same issue." + "reason": "A T5 tokenizer method-name conflict is unrelated to Qwen3Moe FineGrainedFP8Config failure." }, { "left": "issue:46285", "right": "issue:46402", "accept": false, - "reason": "CLIP model structure inconsistency and Sam3Model text-encoder loading are different model-specific regressions." + "reason": "CLIP structure inconsistency and Sam3 text-encoder weight loading are different model-specific regressions." }, { "left": "issue:44737", "right": "issue:45859", "accept": false, - "reason": "Both involve device/offload behavior, but they affect different architectures and code paths." + "reason": "Both are device/offload related, but XLNet CPU arange placement and Qwen3_5Moe rotary embedding CPU-offload compatibility are different code paths." }, { "left": "issue:32090", "right": "issue:39404", "accept": false, - "reason": "Trainer broadcast TypeError is unrelated to Whisper pipeline return_language behavior." + "reason": "Trainer gpu_broadcast_one NoneType error and Whisper return_language regression are unrelated." }, { "left": "issue:44661", "right": "issue:44683", "accept": false, - "reason": "TOKENIZER_MAPPING_NAMES generator failure is unrelated to flex_attention compile support." + "reason": "TOKENIZER_MAPPING_NAMES add-new-model-like behavior is unrelated to compiled flex_attention failures." }, { "left": "issue:43643", "right": "issue:44530", "accept": false, - "reason": "trust_remote_code config field loss and Qwen3.5 linear_attention cache crash are separate problems." + "reason": "trust_remote_code config field loss and Qwen3.5 PagedAttentionCache crashes are separate issues." }, { "left": "issue:43541", "right": "issue:45084", "accept": false, - "reason": "Both are compile/tracing-related, but the concrete failure modes and affected code differ." + "reason": "Torch Dynamo grouped_mm tracing failure and template-node compilation TypeError are different compile-time bugs." }, { "left": "issue:42915", "right": "issue:44530", "accept": false, - "reason": "Different Qwen model families and different failures: FP8 config handling vs cache group-type crash." + "reason": "Qwen3 MoE FP8 config loading and Qwen3.5 cache group-type crash are distinct model/runtime failures." }, { "left": "issue:44451", "right": "issue:44987", "accept": false, - "reason": "Different model repositories with distinct loading failures." + "reason": "Loading a specific BERT checkpoint and loading physical-intelligence/fast are unrelated model-load regressions." } ] }, @@ -3276,20 +3282,20 @@ "issue:42915|issue:44589", "issue:43089|issue:45987" ], - "prepared_review_unit_hash": "d2b9e390d68439332b60ddd0982b3a40f24a26ce84ee966a79a6c01f947713df", + "prepared_review_unit_hash": "4487abee1eba01967a7bb59a08ca760f09ccd379a155cbbb84ec54016596af3d", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 33, - "serialized_chars": 37691, + "serialized_chars": 37692, "estimated_input_tokens": 9423, "estimated_eval_tokens": 19102 }, @@ -3298,204 +3304,204 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a set of separate Transformers bug reports across different subsystems and models. The one tempting v5 pair, 43644/43646, is better treated as a related regression family than a confirmed duplicate. No soft pairs are accepted.", - "confidence": 0.78, - "canonical_issue_reason": "issue:43646 is the broadest representative of the custom-model-initialization v5 regression subgroup, but it should not be treated as the confirmed duplicate of issue:43644.", + "summary": "This cluster is mostly a loose set of Transformers v5 regression reports across unrelated subsystems. I don\u2019t see any soft pair that clearly describes the same concrete bug, so I would not merge any of them as duplicates.", + "confidence": 0.24, + "canonical_issue_reason": "Issue 43646 is the broadest and most representative v5 regression report in the set, but the cluster is heterogeneous and not a strong duplicate group.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43646 is the best representative of the custom-model-initialization regression theme, though the evidence is not strong enough to merge it with issue:43644.", + "best_issue_reason": "Issue 43646 is the best single representative for this cluster because it is general, version-scoped, and closer to the shared v5 regression theme than the more model-specific reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43582", "right": "issue:43632", "accept": false, - "reason": "Different failures in different areas: allocator warmup TypeError vs `_is_hf_initialized` flag break." + "reason": "Different failures: AppleSilicon allocator warmup TypeError vs _is_hf_initialized flag regression." }, { "left": "issue:40990", "right": "issue:43421", "accept": false, - "reason": "High perplexity on a model vs tokenizer backend/post-processor update bug; not the same defect." + "reason": "Unrelated bugs: perplexity quality issue vs tokenizer post-processor update behavior." }, { "left": "issue:43644", "right": "issue:43646", "accept": false, - "reason": "Both are v5 regressions, but the packet does not show the same failing code path; 43644 is about non-persistent buffers and 43646 about custom model initialization, so this is only a related regression, not a duplicate." + "reason": "Both are v5 regressions, but one is buffer initialization corruption and the other is custom model init breakage; not clearly the same concrete bug." }, { "left": "issue:34567", "right": "issue:40444", "accept": false, - "reason": "Unrelated issues: token-count bookkeeping vs multi-image iterable dataset failure." + "reason": "Trainer token-count tracking vs multimodal iterable dataset batching; different code paths." }, { "left": "issue:44206", "right": "issue:44220", "accept": false, - "reason": "Both are audio-related, but one is a specific unsupported-arg regression and the other is a generic helper issue without evidence of the same root cause." + "reason": "Both touch audio features, but one is a center-arg crash and the other is a different fbank helper issue; not enough to treat as one bug." }, { "left": "issue:44811", "right": "issue:45584", "accept": false, - "reason": "Both involve Whisper and special tokens, but the symptoms and code paths differ." + "reason": "Whisper decode skip_special_tokens vs empty-transcription generation after align_special_tokens; separate symptoms." }, { "left": "issue:42175", "right": "issue:46178", "accept": false, - "reason": "Packaging/backend-mapping problems with different backends; not the same defect." + "reason": "Different backend-installation problems: missing tensorflow in extras vs unknown tensorflow_text backend mapping." }, { "left": "issue:43883", "right": "issue:44488", "accept": false, - "reason": "Missing `all_tied_weights_keys` vs a model-loading failure for a specific model; different root causes." + "reason": "Different model-loading failures for different models and likely different root causes." }, { "left": "issue:40444", "right": "issue:46326", "accept": false, - "reason": "IterableDataset multi-image failure is unrelated to `continue_final_message` handling." + "reason": "Different audio-multimodal failures: iterable dataset with multiple images vs mistral_common continue_final_message handling." }, { "left": "issue:36032", "right": "issue:43645", "accept": false, - "reason": "Tokenizer method-name conflict vs custom model initialization break in notebooks; different bugs." + "reason": "Tokenizer method-name conflict vs Jupyter custom model initialization regression; unrelated." }, { "left": "issue:43452", "right": "issue:43482", "accept": false, - "reason": "Both are GGUF-loading related, but one is a broad `from_pretrained` break and the other is a model-specific loading failure; not enough to merge." + "reason": "Both are GGUF-related, but one is a general from_pretrained breakage and the other is a specific Qwen2.5-GGUF v5 issue." }, { "left": "issue:45084", "right": "issue:45507", "accept": false, - "reason": "Template compilation error vs invalid method call in GraniteMoEHybrid; different code paths." + "reason": "Template compilation error vs GraniteMoEHybrid invalid method call; different failures." }, { "left": "issue:43317", "right": "issue:43482", "accept": false, - "reason": "Device-map/offload dequantization failure vs GGUF model loading failure; not the same bug." + "reason": "Dequantized model offload failure vs Qwen2.5-GGUF loading failure; not the same bug." }, { "left": "issue:44987", "right": "issue:44991", "accept": false, - "reason": "Both are loading regressions, but for different models and likely different tokenizer/model paths." + "reason": "Both are load failures on different models; too broad to be duplicates." }, { "left": "issue:43262", "right": "issue:43377", "accept": false, - "reason": "Audio processor sampling-rate default bug vs MIMI encoder padding-mask mismatch; unrelated." + "reason": "Audio chat-template sampling-rate default vs MIMI batching/padding-mask inconsistency; unrelated." }, { "left": "issue:43746", "right": "issue:46326", "accept": false, - "reason": "Local checkpoint loading with PEFT adapters vs tokenizer `continue_final_message` behavior; different issues." + "reason": "PEFT adapter local checkpoint loading vs mistral_common message handling; distinct issues." }, { "left": "issue:42915", "right": "issue:45084", "accept": false, - "reason": "FineGrainedFP8Config failure vs template-node compilation failure; unrelated regressions." + "reason": "FineGrainedFP8Config failure and template compilation failure are unrelated." }, { "left": "issue:43883", "right": "issue:44991", "accept": false, - "reason": "Missing tied-weights attribute vs tokenizer loading failure for a different model." + "reason": "Different models and different loading paths; no shared concrete bug." }, { "left": "issue:44403", "right": "issue:45081", "accept": false, - "reason": "Unnecessary loading noise vs Mistral regex patch crash; not the same underlying bug." + "reason": "General loading noise vs Mistral regex patch crash; not the same problem." }, { "left": "issue:44464", "right": "issue:45987", "accept": false, - "reason": "Chunked generation inconsistency with compiled forward is unrelated to StaticCache seq-length typing." + "reason": "Chunked generation inconsistency vs StaticCache return-type contract bug; different code paths." }, { "left": "issue:43646", "right": "issue:44534", "accept": false, - "reason": "Both are v5 regressions, but one is custom-model initialization and the other is non-persistent buffer serialization; different defects." + "reason": "Both mention v5 and initialization-like behavior, but buffer corruption is not the same concrete bug as custom model initialization breakage." }, { "left": "issue:43645", "right": "issue:44534", "accept": false, - "reason": "Notebook custom-model init break vs non-persistent buffer junk; not the same code-path problem." + "reason": "Jupyter notebook custom-model init breakage vs non-persistent buffer junk; separate regressions." }, { "left": "issue:43404", "right": "issue:44466", "accept": false, - "reason": "Un-tied lm_head weights in one model vs device-dependent serialization inconsistency in another." + "reason": "Different lm_head tying/serialization symptoms on different models and scenarios." }, { "left": "issue:44849", "right": "issue:45325", "accept": false, - "reason": "Different models and different bugs: hidden-states output issue vs RoPE index scaling issue." + "reason": "Different model-specific output/position-id bugs with no clear shared code path." }, { "left": "issue:43377", "right": "issue:43819", "accept": false, - "reason": "Batched-vs-single mismatch from missing padding-mask support vs DAC forward/from_latents mismatch; unrelated." + "reason": "Both are audio model issues, but batched-vs-single padding mask behavior is not the same as DAC latent/STE mismatch." }, { "left": "issue:42915", "right": "issue:44291", "accept": false, - "reason": "FineGrainedFP8Config failure vs unexpected `_is_hf_initialized` init argument; different regressions." + "reason": "Different initialization/quantization failures; only loosely related by v5-era API changes." }, { "left": "issue:43994", "right": "issue:44610", "accept": false, - "reason": "Different vision-model issues: nonsensical outputs vs wrong processor input size." + "reason": "Different models and distinct processor/model input-size problems." }, { "left": "issue:44451", "right": "issue:45081", "accept": false, - "reason": "Model loading failure for ScandiBERT vs Mistral regex patch crash; no shared concrete bug." + "reason": "Different tokenizer/model loading failures with different models and mechanics." }, { "left": "issue:44987", "right": "issue:45081", "accept": false, - "reason": "Model-loading regression vs tokenizer regex patch crash; not the same issue." + "reason": "Different targets: physical-intelligence/fast load regression vs Mistral tokenizer regex crash." }, { "left": "issue:28530", "right": "issue:34567", "accept": false, - "reason": "Early stopping/eval-metric behavior is unrelated to token-count tracking." + "reason": "Early stopping metric selection vs trainer token counter update; unrelated." }, { "left": "issue:36032", "right": "issue:43742", "accept": false, - "reason": "Tokenizer special-token method conflict vs MobileLLM loading key error; different root causes." + "reason": "Tokenizer class-name conflict vs model loading KeyError; different failure modes." }, { "left": "issue:42915", "right": "issue:44589", "accept": false, - "reason": "FineGrainedFP8Config failure vs float8 storage lookup error; distinct failures." + "reason": "FineGrainedFP8Config failure vs missing Float8 storage type; only loosely related by quantization support." }, { "left": "issue:43089", @@ -3505,12 +3511,9 @@ } ] }, - "evaluator_result": { - "accept": false, - "feedback": "The summary is mostly grounded, but it overstates the evidence for the one accepted soft pair. In the packet, issue:43644 (\u201cnon-persistent buffers with junk\u201d) and issue:43646 (\u201ccustom model initialization\u201d) are both v5 regressions, but there isn\u2019t enough direct evidence here to call 43644 just a symptom of 43646 or to make 43646 the canonical duplicate representative. The conservative call would be to keep that pair unaccepted or phrase it as a likely related regression, not a confirmed same-bug match." - }, - "evaluator_used": true, - "retried": true, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null @@ -3608,20 +3611,20 @@ "issue:41553|issue:43582", "issue:43531|issue:44291" ], - "prepared_review_unit_hash": "91d2dfd2a19d2cd525c62a4c866908250817c62ff5684fe9886d47eb67247305", + "prepared_review_unit_hash": "60f813d8bb283f311e0d6dfde92db059a99359924b45602ede2912669994641c", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 30, - "serialized_chars": 35938, + "serialized_chars": 35940, "estimated_input_tokens": 8985, "estimated_eval_tokens": 18226 }, @@ -3630,199 +3633,202 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These soft pairs mostly span unrelated bugs across tokenization, model loading, multimodal preprocessing, and runtime compatibility; none look like the same underlying defect, so all soft edges should be rejected.", - "confidence": 0.93, - "canonical_issue_reason": null, + "summary": "This cluster is mostly heterogeneous transformer/model issues, not one tight duplicate set. The only plausible duplicate-like pair is the Qwen3.5 save_pretrained regression thread; the rest are different bugs in different code paths.", + "confidence": 0.79, + "canonical_issue_reason": "issue:45216 is the clearest anchor: it describes a concrete Qwen3.5 save_pretrained regression and already has downstream references, making it the best representative of the only near-duplicate thread.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44987 is the broadest generic v5 loading-regression report in the set, but the overall cluster is still too mixed to treat it as a true canonical.", + "best_issue_reason": "issue:45812 is the strongest standalone triage candidate overall because it is broad ('all Granite models'), clearly a silent v4\u2192v5 regression, and has the highest apparent external impact.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44062", "right": "issue:44987", "accept": false, - "reason": "Different failures: AddedToken keyword-argument conflict vs a transformers>=5.1.0 model-loading regression." + "reason": "Both involve tokenizer loading failures, but the symptoms and likely code paths differ: one is an AddedToken API conflict, the other is a model-specific load failure on v5.1." }, { "left": "issue:44265", "right": "issue:44610", "accept": false, - "reason": "torch.export + torch_compilable_check is a graph-export issue; OmDet-Turbo is a processor/model input-size mismatch." + "reason": "Different subsystems and bugs: torch.export/torch_compilable_check versus an OmDet-Turbo image-size mismatch in the processor." }, { "left": "issue:43322", "right": "issue:43540", "accept": false, - "reason": "Both are vision-model load/processing problems, but one is a Llava Next segfault and the other is a Qwen3OmniMoe video ValueError." + "reason": "Both are multimodal loading/processing problems, but one is a Llava Next segfault and the other is a Qwen3OmniMoe video ValueError; not the same bug." }, { "left": "issue:44479", "right": "issue:45325", "accept": false, - "reason": "Video-input regression in several Qwen models vs a rope-index scaling bug for still-image temporal positions." + "reason": "Related Qwen VL/video/rope area, but one is a broad v5.3 video regression across models and the other is a specific still-image temporal position_ids scaling bug." }, { "left": "issue:45216", "right": "issue:45357", - "accept": false, - "reason": "Both are Qwen3.5 save_pretrained regressions, but one is incorrect checkpoint content and the other is wrong visual-encoder key saving." + "accept": true, + "reason": "Same model family and same code path: Qwen3.5 save_pretrained producing incorrect checkpoints/visual encoder keys across versions." }, { "left": "issue:43531", "right": "issue:44877", "accept": false, - "reason": "sliding_window behavior in Qwen3-MoE is not the same as strict config rejection for granite_speech." + "reason": "Different failures: Qwen3-MoE sliding_window handling versus strict config blocking granite_speech loading." }, { "left": "issue:36032", "right": "issue:41720", "accept": false, - "reason": "Tokenizer attribute-name conflict is unrelated to Qwen3 auto device mapping triggering cudaErrorAssert." + "reason": "Unrelated issues: a T5 tokenizer method conflict versus Qwen3 auto device mapping causing CUDA assert." }, { "left": "issue:44991", "right": "issue:46315", "accept": false, - "reason": "Both concern tokenizer behavior, but one is model-specific loading failure and the other is a performance regression in convert_tokens_to_ids." + "reason": "Both are tokenizer regressions, but one is a load-time failure for a specific tokenizer and the other is a performance regression in convert_tokens_to_ids; different bug." }, { "left": "issue:43425", "right": "issue:44403", "accept": false, - "reason": "Torch 2.10 incompatibility is a version support issue; unnecessary loading noise is a separate UX/logging problem." + "reason": "No shared underlying change: Torch 2.10 compatibility versus extra loading noise." }, { "left": "issue:43296", "right": "issue:43531", "accept": false, - "reason": "PaddleOCR-VL loading failure is unrelated to Qwen3-MoE sliding_window semantics." + "reason": "Different models and different errors: PaddleOCR-VL/vLLM loading versus Qwen3-MoE sliding_window behavior." }, { "left": "issue:44220", "right": "issue:44683", "accept": false, - "reason": "Audio feature extraction bug vs compiled flex_attention failure on newer torch versions." + "reason": "Different code paths: _torch_extract_fbank_features versus compiled flex_attention on newer torch." }, { "left": "issue:43611", "right": "issue:44534", "accept": false, - "reason": "base_model_prefix loading breakage is a model-init incompatibility; non-persistent buffer junk is a different state-loading bug." + "reason": "One is a model-loading break with base_model_prefix; the other is a v5 buffer-initialization regression." }, { "left": "issue:40444", "right": "issue:45325", "accept": false, - "reason": "IterableDataset multi-image finetuning failure is not the same as still-image temporal-position scaling." + "reason": "Both are Qwen VL-related, but one is iterable-dataset/multi-image finetuning failure and the other is a rope/position_ids bug; not the same defect." }, { "left": "issue:45812", "right": "issue:46491", "accept": false, - "reason": "Granite tokenizer wrong IDs and CodeLlama space-stripping are both tokenizer regressions, but they affect different code paths and models." + "reason": "Both are tokenizer regressions, but they affect different model families and different semantics: Granite token IDs versus CodeLlama space preservation." }, { "left": "issue:43632", "right": "issue:44756", "accept": false, - "reason": "_is_hf_initialized flag handling is a transformers v5 init issue; mmap OOM on Strix Halo is a platform-specific memory policy bug." + "reason": "No meaningful overlap: _is_hf_initialized flag handling versus mmap/OOM behavior on Strix Halo." }, { "left": "issue:44403", "right": "issue:44991", "accept": false, - "reason": "Loading noise/logging regression is unrelated to tokenizer loading failure for EMBEDDIA/est-roberta." + "reason": "Different symptoms and targets: generic loading noise versus a specific tokenizer load failure." }, { "left": "issue:43792", "right": "issue:44589", "accept": false, - "reason": "Whisper model runtime failure is unrelated to missing Float8 storage support." + "reason": "Completely different failure modes: Whisper model run failure versus missing Float8 storage object." }, { "left": "issue:43381", "right": "issue:45923", "accept": false, - "reason": "Gradient-checkpointing eval-mode incompatibility is unrelated to a missing supports_gradient_checkpointing flag in a trust_remote_code model." + "reason": "Related to gradient checkpointing, but one is eval-mode usage semantics and the other is a missing supports_gradient_checkpointing flag on a trust_remote_code model." }, { "left": "issue:42796", "right": "issue:43746", "accept": false, - "reason": "Tokenizer BPE-marker handling and GraniteSpeech PEFT checkpoint loading are different subsystems and failures." + "reason": "Tokenizer BPE marker handling and GraniteSpeech PEFT local checkpoint loading are unrelated." }, { "left": "issue:44291", "right": "issue:44589", "accept": false, - "reason": "Unexpected _is_hf_initialized arg during init_empty_weights is unrelated to Float8 storage lookup." + "reason": "Different backend issues: init_empty_weights/_is_hf_initialized argument handling versus Float8 storage lookup." }, { "left": "issue:33666", "right": "issue:40444", "accept": false, - "reason": "Qwen2-VL multi-GPU training and Qwen2.5-VL iterable-dataset multi-image finetuning are distinct training scenarios and bugs." + "reason": "Both are Qwen VL training/fine-tuning related, but the concrete failures are different and not the same underlying bug." }, { "left": "issue:44079", "right": "issue:45663", "accept": false, - "reason": "ModelOutput key assignment bug is unrelated to Gemma-4 FSDP2 shared_kv_states reconstruction." + "reason": "Different mechanisms: ModelOutput key assignment bug versus Gemma4 FSDP2 shared_kv_states rebuilding error." }, { "left": "issue:40990", "right": "issue:44811", "accept": false, - "reason": "High perplexity on gpt-oss is a model/eval issue, while Whisper batch_decode ignores skip_special_tokens in preprocessing output." + "reason": "No duplicate relationship: perplexity regression on GPT-OSS versus Whisper batch_decode special-token handling." }, { "left": "issue:42796", "right": "issue:46710", "accept": false, - "reason": "Tokenizer BPE-marker regression is unrelated to a DeepSeek-R1-Distill-Llama accuracy regression." + "reason": "Tokenizer BPE-marker bug and DeepSeek accuracy regression are unrelated." }, { "left": "issue:43482", "right": "issue:43525", "accept": false, - "reason": "Qwen2.5-GGUF loading under transformers v5 and Llama4Config missing pad_token_id are separate model/config compatibility issues." + "reason": "Different model/config problems: GGUF loading under transformers v5 versus missing pad_token_id on Llama4Config." }, { "left": "issue:35707", "right": "issue:40444", "accept": false, - "reason": "Progressive generation with inputs_embeds/past_key_values is not the same bug as multi-image IterableDataset finetuning failure." + "reason": "Both mention generation/fine-tuning workflows, but one is inputs_embeds/past_key_values progressive generation and the other is multimodal iterable dataset loading." }, { "left": "issue:43632", "right": "issue:44534", "accept": false, - "reason": "Both involve transformers v5 internals, but one is initialization-flag handling and the other is buffer contents on load." + "reason": "Unrelated regressions: _is_hf_initialized flag versus non-persistent buffer initialization." }, { "left": "issue:43482", "right": "issue:45538", "accept": false, - "reason": "Qwen2.5-GGUF loading failure and CLIPTokenizer max-length handling are unrelated tokenization/loading behaviors." + "reason": "Different tokenizer/model-loading issues: GGUF loading failure versus CLIPTokenizer max length behavior." }, { "left": "issue:41553", "right": "issue:43582", "accept": false, - "reason": "Bad AutoTokenizer error for Voxtral is unrelated to AppleSilicon TypeError in caching_allocator_warmup." + "reason": "Completely different areas: AutoTokenizer error messaging for Voxtral versus Apple Silicon caching allocator TypeError." }, { "left": "issue:43531", "right": "issue:44291", "accept": false, - "reason": "Qwen3-MoE sliding_window behavior and unexpected _is_hf_initialized init argument are different regression types." + "reason": "Different bugs in different layers: Qwen3-MoE sliding_window behavior versus init_empty_weights argument handling." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "Grounded and conservative overall. The broad Granite tokenizer regression is a reasonable best standalone issue, the Qwen3.5 save_pretrained thread is the clearest canonical anchor, and the accepted pair is well-supported by the packet. The rejected soft-edge pairs stay appropriately cautious and do not overclaim duplication." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -3919,218 +3925,215 @@ "issue:43296|issue:45800", "issue:44479|issue:45823" ], - "prepared_review_unit_hash": "d18a4ed78c143add657c21694df90bbe66863b692e8f5ffa9fe6160627ff2236", + "prepared_review_unit_hash": "8774ee2b0987a2361dc348146d6d8924a8a01ac6358cbeb27294e458b4827311", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36626, - "estimated_input_tokens": 9157, - "estimated_eval_tokens": 18570 + "serialized_chars": 36624, + "estimated_input_tokens": 9156, + "estimated_eval_tokens": 18568 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous, but it loosely centers on Transformers v5 regressions. Most pairs are only thematically similar, not true duplicates. The only strong duplicate-like pair is the SigLIP2 regression pair.", - "confidence": 0.66, - "canonical_issue_reason": "Issue 43646 is the broadest representative of the shared v5 model-initialization/loading regression theme; it covers the most general symptom, while nearby reports are narrower or about different subsystems.", + "summary": "This cluster is mostly a loose collection of unrelated or only loosely related Transformers regressions, many around v5 loading/initialization/tokenizer behavior. I did not find any soft pair that looks like the same underlying bug/change, so this should not be merged as duplicate triage.", + "confidence": 0.93, + "canonical_issue_reason": "If a single issue must represent the cluster, issue 43611 is the broadest v5 loading-regression umbrella; however, the cluster is not a true duplicate set and the items are still distinct bugs.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 43646 is the best anchor for the dominant subcluster because its title is broad enough to subsume several v5 init/load breakages, whereas the others are more specific manifestations.", + "best_issue_reason": "43611 is the most representative of the repeated Transformers v5 model-loading breakage theme, but it is still only an umbrella candidate rather than an exact duplicate of the others.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43317", "right": "issue:43856", "accept": false, - "reason": "Both mention model efficiency/loading, but one is quantized offload/device-map failure and the other is Qwen3 MoE training memory usage." + "reason": "Both concern offloading/quantization performance or failure modes, but they describe different symptoms and code paths." }, { "left": "issue:42175", "right": "issue:45542", "accept": false, - "reason": "Both concern TensorFlow-related packaging/backend setup, but one is a missing TF backend in the torch extra and the other is a tensorboard-without-TF runtime error." + "reason": "Both are backend-installation packaging issues, but one is about pip extras missing TensorFlow while the other is about tensorboard/TF backend detection." }, { "left": "issue:43421", "right": "issue:44568", "accept": false, - "reason": "One is about runtime updates to special tokens/post-processing; the other is a specific BOS/EOS insertion regression in a particular tokenizer." + "reason": "Runtime special-token/post-processor update behavior is unrelated to add_special_tokens BOS/EOS insertion regression." }, { "left": "issue:44079", "right": "issue:44534", "accept": false, - "reason": "Different bugs: ModelOutput key assignment vs non-persistent buffer initialization in v5." + "reason": "Both mention ModelOutput/non-persistent buffer handling, but they affect different APIs and failure modes." }, { "left": "issue:44829", "right": "issue:45479", "accept": false, - "reason": "Both affect classification training, but the concrete failures differ: flash_attention_3 instability vs zero-loss behavior for num_labels=1." + "reason": "These are different training regressions: flash_attention_3 classification degeneration vs single-label num_labels=1 zero-loss behavior." }, { "left": "issue:41720", "right": "issue:45081", "accept": false, - "reason": "Unrelated failures: Qwen3 device-map cuda assert vs Mistral tokenizer regex patch crash." + "reason": "One is a CUDA assert in Qwen3 device mapping; the other is a tokenizer regex patch crash." }, { "left": "issue:44462", "right": "issue:45812", "accept": false, - "reason": "Both are tokenizer regressions, but one is ignoring tokenizer.json in AutoTokenizer and the other is wrong Granite token IDs; not enough evidence of the same bug." + "reason": "Both involve tokenizer loading/serialization, but one ignores tokenizer.json and the other returns wrong token IDs for Granite models." }, { "left": "issue:43818", "right": "issue:46402", "accept": false, - "reason": "Different models and failures: Video-LLaVA tower/weights issue vs Sam3 text-encoder weights not loading." + "reason": "Video-LLaVA tower architecture corruption is unrelated to Sam3 text-encoder weights not loading." }, { "left": "issue:43818", "right": "issue:45357", "accept": false, - "reason": "Different code paths: Video-LLaVA architecture problem vs Qwen3.5 save_pretrained key serialization bug." + "reason": "Both are vision/model-loading issues, but one is Video-LLaVA architecture state and the other is Qwen3.5 save_pretrained key serialization." }, { "left": "issue:43611", "right": "issue:43646", "accept": false, - "reason": "Both are v5 loading/init regressions, but one is about base_model_prefix handling and the other is custom model initialization; not clearly the same bug." + "reason": "Both are v5 loading regressions, but one is base_model_prefix loading and the other is custom model initialization." }, { "left": "issue:43377", "right": "issue:45491", "accept": false, - "reason": "Both are batching/padding-related, but they affect different models and different mechanisms (missing padding mask vs sliding-window all-padding NaNs)." + "reason": "MIMI encoder padding-mask batching bug is not the same as Gemma3 sliding-window NaN embeddings." }, { "left": "issue:43644", "right": "issue:44079", "accept": false, - "reason": "Different regressions: non-persistent buffer junk vs ModelOutput key assignment." + "reason": "Non-persistent buffer junk filling is a different defect from ModelOutput key assignment when previous value was None." }, { "left": "issue:42175", "right": "issue:43296", "accept": false, - "reason": "One is a packaging/backend dependency issue; the other is a model load failure in vLLM with Transformers." + "reason": "Package-extras TensorFlow omission is unrelated to PaddleOCR-VL loading in vLLM/transformers." }, { "left": "issue:43749", "right": "issue:44079", "accept": false, - "reason": "Different areas: FSDP CPU RAM efficient loading vs ModelOutput key bookkeeping." + "reason": "FSDP CPU RAM efficient loading is a distributed-loading issue, not a ModelOutput bookkeeping bug." }, { "left": "issue:44265", "right": "issue:45072", "accept": false, - "reason": "Unrelated: torch.export/torch_compilable_check failure vs bfloat16 dtype mismatch in inference." + "reason": "torch.export.export with torch_compilable_check is unrelated to dtype mismatches in bfloat16 inference." }, { "left": "issue:43792", "right": "issue:43957", "accept": false, - "reason": "Different failure modes and models: Whisper runtime failure vs meta-device loading breakage in v5." + "reason": "Whisper model runtime failure and meta-device loading breakage are different model-loading problems." }, { "left": "issue:45325", "right": "issue:45381", "accept": false, - "reason": "Both are Qwen2.5-VL position-id regressions, but one is still-image temporal scaling and the other is video vision_position_ids; related but not clearly the same underlying bug." + "reason": "Both are Qwen2.5-VL video-position issues, but one is still-image temporal scaling while the other is video input vision_position_ids regression." }, { "left": "issue:43454", "right": "issue:44466", "accept": false, - "reason": "One is a missing lm_head weight tie causing bad generation; the other is inconsistent lm_head serialization depending on device." + "reason": "AyaVision lm_head tying failure and inconsistent lm_head serialization are related only at a broad weights level, not the same bug." }, { "left": "issue:43493", "right": "issue:43994", - "accept": true, - "reason": "Both point to the same SigLIP2 implementation mismatch: one compares HF vs JAX behavior, the other reports the downstream bad outputs from that mismatch." + "accept": false, + "reason": "SigLIP2 HF-vs-JAX discrepancy is not the same as wrong AutoModel/pipeline outputs for siglip2-base." }, { "left": "issue:40444", "right": "issue:45381", "accept": false, - "reason": "Both involve Qwen2.5-VL vision inputs, but one is finetuning with multiple images per prompt and the other is a video position-id regression." + "reason": "Multimodal finetuning with multiple images per prompt is unrelated to Qwen2.5-VL video position-id regression." }, { "left": "issue:45325", "right": "issue:45834", "accept": false, - "reason": "Different subsystems: Qwen2.5-VL positional indexing vs Kosmos2.5 OCR long-input index error." + "reason": "Qwen2.5-VL temporal position scaling and Kosmos2.5 OCR index errors are different code paths and symptoms." }, { "left": "issue:42915", "right": "issue:43482", "accept": false, - "reason": "Different models and causes: Qwen3MoE FP8 config failure vs GGUF loading failure in Transformers v5." + "reason": "Qwen3Moe FP8 failure and Qwen2.5-GGUF loading under v5 are distinct model/config incompatibilities." }, { "left": "issue:44291", "right": "issue:44877", "accept": false, - "reason": "Different issues: unexpected _is_hf_initialized arg in init_empty_weights vs strict config rejection for granite_speech." + "reason": "Unexpected _is_hf_initialized argument in init_empty_weights is unrelated to strict config blocking granite_speech." }, { "left": "issue:43632", "right": "issue:45072", "accept": false, - "reason": "Different regressions: _is_hf_initialized flag handling vs dtype mismatches in bfloat16 inference." + "reason": "_is_hf_initialized flag breakage is a loading/init regression, not a dtype mismatch in inference." }, { "left": "issue:43873", "right": "issue:45701", "accept": false, - "reason": "Different categories: quantization/offloading behavior vs tokenizer behavior changing across versions." + "reason": "Quantization offloading behavior and version-dependent tokenization are different classes of bugs." }, { "left": "issue:43701", "right": "issue:45325", "accept": false, - "reason": "Resume-from-checkpoint key mismatch is unrelated to Qwen2.5-VL position-id handling." + "reason": "Resume-from-checkpoint key mismatch is unrelated to Qwen2.5-VL rope index scaling." }, { "left": "issue:42796", "right": "issue:45381", "accept": false, - "reason": "Tokenizer BPE-marker handling and Qwen2.5-VL video position ids are separate bugs." + "reason": "Tokenizer BPE marker handling and Qwen2.5-VL video position IDs are separate tokenizer vs multimodal-vision bugs." }, { "left": "issue:43296", "right": "issue:45800", "accept": false, - "reason": "Different compatibility problems: PaddleOCR-VL load failure in vLLM vs torch 2.4.1/Transformers 5.8 incompatibility." + "reason": "PaddleOCR-VL load failure in vLLM/transformers is not the same as torch 2.4.1 versus transformers 5.8.0 incompatibility." }, { "left": "issue:44479", "right": "issue:45823", "accept": false, - "reason": "Different failures: Qwen video-input regression vs Gemma4 device-map forward mismatch." + "reason": "Video-input regression in Qwen-family models and Gemma4 PLE device mismatch are unrelated model families and failure mechanisms." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet: it accurately characterizes the cluster as heterogeneous with a broad Transformers v5 init/loading subtheme, and it keeps the duplicate claims conservative. The one accepted soft-edge pair (SigLIP2) is plausible from the titles, while the rejected pairs are cautiously distinguished." - }, - "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 }, @@ -4228,214 +4231,214 @@ "issue:43881|issue:44991", "issue:44387|issue:45704" ], - "prepared_review_unit_hash": "581021475c7877b002934e64d07c81f40489e2f40ae6a3d4eb0a5956fddaf6a3", + "prepared_review_unit_hash": "124facf1541ebf7269f694ffeb7d8b8b9095f013c593c5a818992584256ad044", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36834, - "estimated_input_tokens": 9209, - "estimated_eval_tokens": 18674 + "serialized_chars": 36830, + "estimated_input_tokens": 9208, + "estimated_eval_tokens": 18672 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No soft edge looks like a true duplicate: these are mostly unrelated bugs across audio, tokenization, quantization, loading, training, and multimodal processors. I would not merge any of the suggested pairs.", - "confidence": 0.98, - "canonical_issue_reason": "issue:45920 is the strongest representative issue here: it is an open, cross-model tokenizer regression with multiple inbound references and clear user impact.", + "summary": "These items are mostly distinct bug reports across different models, tokenizers, loading paths, and training/inference features. The only loose thematic overlap is around offloading/quantization, but none of the soft pairs are clearly the same underlying bug or mergeable as one PR.", + "confidence": 0.82, + "canonical_issue_reason": "issue:43873 is the broadest representative of the offloading/quantization subgroup and has enough discussion/activity to serve as the best umbrella issue.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45920 is the best standalone issue to keep as the cluster representative because it is broad, active, and well-scoped compared with the other unrelated reports.", + "best_issue_reason": "issue:43873 best captures the common loader/offload failure theme present in the nearest related reports, while still being broader than any single model-specific incident.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43606", "right": "issue:44265", "accept": false, - "reason": "Different failures: Bark CPU offload device mismatch vs torch.export/torch_compilable_check export errors." + "reason": "Different failure classes: CPU-offload device mismatch for Bark vs export/tracing failure with torch_compilable_check." }, { "left": "issue:43262", "right": "issue:43746", "accept": false, - "reason": "Audio chat-template sampling-rate default bug vs GraniteSpeech PEFT local checkpoint loading." + "reason": "Audio chat-template sampling-rate default bug vs a PEFT/local-checkpoint loading problem; no shared code path." }, { "left": "issue:40990", "right": "issue:45584", "accept": false, - "reason": "Perplexity regression on gpt-oss vs Whisper empty-transcription generation failure; unrelated models/code paths." + "reason": "Unrelated symptoms and components: perplexity regression on gpt-oss vs Whisper empty-transcription generation failure." }, { "left": "issue:35707", "right": "issue:41720", "accept": false, - "reason": "inputs_embeds/past_key_values generation bug vs Qwen3 auto device-map CUDA assert; not the same underlying issue." + "reason": "Progressive generation with inputs_embeds/past_key_values is a different path from auto device-mapping cuda assert failures." }, { "left": "issue:43881", "right": "issue:44488", "accept": false, - "reason": "glm-4v load failure vs sleng-bert tokenizer load failure; different model families and causes." + "reason": "Both are loading failures, but for different models and likely different root causes; not the same bug." }, { "left": "issue:34634", "right": "issue:35707", "accept": false, - "reason": "Bark voice_preset issue is unrelated to progressive generation with inputs_embeds and past_key_values." + "reason": "BarkProcessor voice_preset issue is unrelated to generation with inputs_embeds and past_key_values." }, { "left": "issue:43454", "right": "issue:43550", "accept": false, - "reason": "Weight-tying bug in AyaVision is unrelated to Bamba torch.compile/SDPA failure." + "reason": "Weight-tying/generation corruption in AyaVision is unrelated to Bamba torch.compile+SDPA failures." }, { "left": "issue:43606", "right": "issue:43819", "accept": false, - "reason": "Bark offload device mismatch is unrelated to DAC.from_latents missing STE behavior." + "reason": "Both mention latent/model loading behavior, but one is CPU offload mismatch and the other is DAC latent conversion mismatch." }, { "left": "issue:39692", "right": "issue:40444", "accept": false, - "reason": "Documentation example errors for SigLIP2 vs Qwen2.5-VL training with iterable datasets; different problem classes." + "reason": "Doc-example model/processor mismatch and quantization failure is unrelated to iterable-dataset multi-image finetuning." }, { "left": "issue:41720", "right": "issue:45823", "accept": false, - "reason": "Both mention device issues, but they are different models and different code paths (Qwen3 auto-map assert vs Gemma4 PLE forward mismatch)." + "reason": "Both involve device_map='auto', but one is a Qwen3 cuda assert and the other is a Gemma4 PLE forward device mismatch." }, { "left": "issue:44186", "right": "issue:44265", "accept": false, - "reason": "LayoutLMv2Tokenizer NER/padding crash is unrelated to torch.export failures on torch_compilable_check models." + "reason": "Tokenizer crashes on NER/padding are unrelated to torch.export exportability issues." }, { "left": "issue:43377", "right": "issue:44610", "accept": false, - "reason": "MIMI batching/padding-mask discrepancy is unrelated to OmDet-Turbo processor image-size mismatch." + "reason": "MIMI batching/padding-mask correctness is unrelated to OmDet-Turbo processor/model input-size mismatch." }, { "left": "issue:46305", "right": "issue:46315", "accept": false, - "reason": "MoonshineStreaming sliding-window off-by-one is unrelated to tokenizer convert_tokens_to_ids slowness." + "reason": "Streaming mask off-by-one and tokenizer performance regression are different bugs in different code paths." }, { "left": "issue:43257", "right": "issue:45923", "accept": false, - "reason": "Qwen3 MOE weight conversion/loading bug is unrelated to a missing supports_gradient_checkpointing flag in Nemotron trust_remote_code." + "reason": "Accelerate+deepspeed MOE weight conversion failure is unrelated to a missing gradient-checkpointing flag." }, { "left": "issue:44568", "right": "issue:45920", "accept": false, - "reason": "Specific add_special_tokens BOS/EOS regression is not the same bug as multi-model wrong token IDs." + "reason": "Both are tokenizer regressions, but they affect different APIs and models with different symptoms." }, { "left": "issue:43454", "right": "issue:45072", "accept": false, - "reason": "AyaVision lm_head tying problem is unrelated to dtype mismatches in SwitchTransformers/TimmWrapperModel." + "reason": "Weight tying causing garbage generation is not the same as bf16 dtype mismatches in inference wrappers." }, { "left": "issue:42796", "right": "issue:46209", "accept": false, - "reason": "Tokenizer BPE-marker handling bug is unrelated to FP8 inference failure on B200." + "reason": "Tokenizer BPE-marker handling and FP8 inference on B200 are unrelated." }, { "left": "issue:43317", "right": "issue:43873", "accept": false, - "reason": "These are both about offloading/quantization in a broad sense, but the concrete failures differ: dequantized model loading vs quantized offloading behavior." + "reason": "Both touch offloading/quantization, but one is a specific dequantized model load failure and the other is a broad offloading complaint without enough evidence of the same code-path." }, { "left": "issue:43454", "right": "issue:43720", "accept": false, - "reason": "Weight tying bug in AyaVision is unrelated to BitNet packed-weight unpacking during accelerate loading." + "reason": "Missing lm_head weight tying and packed-weight unpacking during accelerate loading are different model-loading bugs." }, { "left": "issue:43377", "right": "issue:44898", "accept": false, - "reason": "MIMI encoder padding-mask bug is unrelated to Perceiver non-default-resolution image classification failure." + "reason": "Padding-mask correctness in MIMI is unrelated to Perceiver resolution/interpolation failures." }, { "left": "issue:40990", "right": "issue:42947", "accept": false, - "reason": "Perplexity regression on gpt-oss is unrelated to gradient checkpointing being ineffective with PEFT LoRA." + "reason": "Perplexity regression on a model is unrelated to gradient-checkpointing behavior with PEFT LoRA." }, { "left": "issue:42617", "right": "issue:42915", "accept": false, - "reason": "3d_parallel.py runtime failure and Qwen3Moe FineGrainedFP8Config failure are different issues." + "reason": "3d_parallel runtime failure and Qwen3Moe FineGrainedFP8Config failure are different execution paths." }, { "left": "issue:43749", "right": "issue:44534", "accept": false, - "reason": "Broken FSDP CPU RAM-efficient loading is unrelated to non-persistent buffers being filled with junk." + "reason": "CPU RAM efficient loading breakage and non-persistent buffer junk are distinct v5 regressions." }, { "left": "issue:43572", "right": "issue:45759", "accept": false, - "reason": "Missing pad_token_idx in StableLmConfig is unrelated to failing to unwrap text_config in composite Qwen multimodal configs." + "reason": "Missing pad_token_idx in StableLmConfig is unrelated to unwrapping text_config in composite Qwen multimodal configs." }, { "left": "issue:43322", "right": "issue:46050", "accept": false, - "reason": "Llava Next loading segfault is unrelated to sinq quantized save/load corruption." + "reason": "Llava Next loading segfault and sinq save/load breakage are different failure modes." }, { "left": "issue:43526", "right": "issue:43883", "accept": false, - "reason": "BeitImageProcessorFast reduce_labels bug is unrelated to Molmo missing all_tied_weights_keys." + "reason": "BeitImageProcessorFast label reduction bug is unrelated to missing all_tied_weights_keys on MolmoForCausalLM." }, { "left": "issue:41720", "right": "issue:43322", "accept": false, - "reason": "Different models and failures: Qwen3 CUDA assert vs Llava Next segmentation fault." + "reason": "Qwen3 device-map assert and Llava Next segfault are both loading-related but not the same concrete bug." }, { "left": "issue:36331", "right": "issue:42617", "accept": false, - "reason": "CustomTrainer.compute_loss signature breakage is unrelated to 3d_parallel.py execution failure." + "reason": "CustomTrainer compute_loss signature breakage is unrelated to 3d_parallel execution failure." }, { "left": "issue:43881", "right": "issue:44991", "accept": false, - "reason": "glm-4v loading failure is unrelated to tokenizer loading regression for EMBEDDIA/est-roberta." + "reason": "Two tokenizer/model-loading regressions, but for different models and likely different fixes." }, { "left": "issue:44387", "right": "issue:45704", "accept": false, - "reason": "Int4 quantization OOM from CUDA reserved memory is unrelated to T5 apex FusedRMSNorm memory leak." + "reason": "Quantization OOM from reserved memory growth is unrelated to an apex RMSNorm memory leak." } ] }, @@ -4540,214 +4543,214 @@ "issue:36331|issue:41720", "issue:44186|issue:44361" ], - "prepared_review_unit_hash": "8998ee4e619138e58a63de166dafd78b99d3ed5c4a5f883d27bf4587bcbd07f2", + "prepared_review_unit_hash": "968c754f131bbd8ba603fc4c03b14911a97b15be46b44aa8641eb095ab697f97", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 31, - "serialized_chars": 36498, - "estimated_input_tokens": 9125, - "estimated_eval_tokens": 18506 + "serialized_chars": 36504, + "estimated_input_tokens": 9126, + "estimated_eval_tokens": 18508 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is a grab-bag of unrelated Transformers issues (trainer regressions, model-loading bugs, tokenizer/processor problems, multimodal serving crashes, quantization/offload issues). None of the soft-edge pairs look like the same underlying bug or change, so no duplicates should be merged.", + "summary": "These are broad semantic-similarity matches, but the paired items mostly describe different concrete bugs or model-specific regressions. I reject all soft edges; there are no PRs in this set. The most representative issue is the detailed Trainer regression in #43708.", "confidence": 0.94, - "canonical_issue_reason": "issue:43708 is the strongest single issue in the set to anchor on: it has a precise regression description, a concrete affected path (`resume_from_checkpoint` / `max_steps`), and a clear reproduction scenario. That said, the overall cluster is not a true duplicate set.", + "canonical_issue_reason": "issue:43708 is the most concrete, actionable regression report here, with a clearly identified code path (`resume_from_checkpoint` / `max_steps`) and a specific reproduction condition.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43708 is the most actionable and well-scoped report among these unrelated issues, with the clearest failure mode and likely fix surface.", + "best_issue_reason": "issue:43708 best represents the cluster because it is well-scoped, reproducible, and more clearly described than the other candidates.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:41720", "right": "issue:44977", "accept": false, - "reason": "Different models and different failures: auto device-map CUDA assert vs flash-attention generation bug." + "reason": "Different failures: one is a CUDA assert during auto device mapping, the other is flash-attention generation behavior for Qwen3.5." }, { "left": "issue:36331", "right": "issue:42915", "accept": false, - "reason": "Different code paths: trainer `compute_loss` API mismatch vs Qwen3-MoE fine-grained FP8 failure." + "reason": "Unrelated bugs: a Trainer API signature change versus a Qwen3MoE FP8 model failure." }, { "left": "issue:44466", "right": "issue:45357", "accept": false, - "reason": "Both touch serialization, but one is lm_head tying/serialization and the other is wrong visual encoder keys; not the same bug." + "reason": "Both involve serialization in v5, but one is tied `lm_head.weight` behavior and the other is wrong visual encoder keys; not the same bug." }, { "left": "issue:44877", "right": "issue:44991", "accept": false, - "reason": "Unrelated loading regressions: strict config handling vs tokenizer loading for a specific model." + "reason": "Different loading failures affecting different model assets and code paths: strict config handling vs tokenizer loading." }, { "left": "issue:43450", "right": "issue:45701", "accept": false, - "reason": "Video processor shape bug is unrelated to a tokenizer/version-dependent tokenization change." + "reason": "Video batch-shape bug is unrelated to a version-dependent tokenization change." }, { "left": "issue:40990", "right": "issue:45701", "accept": false, - "reason": "High perplexity on a model/dataset is not the same as version-dependent tokenization behavior." + "reason": "High perplexity on a specific language model is not the same underlying issue as version-dependent tokenization differences." }, { "left": "issue:43065", "right": "issue:46402", "accept": false, - "reason": "Different models and different issues: dummy Conv2d vs missing text encoder weight loading." + "reason": "Different model components and failures: a dummy Conv2d in Sam3PixelDecoder versus missing text encoder weights in Sam3Model." }, { "left": "issue:41720", "right": "issue:42617", "accept": false, - "reason": "Different runtime failures in different training setups; no shared concrete bug." + "reason": "Qwen3 auto device-mapping CUDA assertion and inability to run `3d_parallel.py` are distinct runtime problems." }, { "left": "issue:43873", "right": "issue:44756", "accept": false, - "reason": "Quantization/offloading behavior and mmap OOM avoidance are different problems." + "reason": "Quantization/offload behavior and Strix Halo mmap OOM are related only broadly by memory usage, not the same defect." }, { "left": "issue:44186", "right": "issue:45072", "accept": false, - "reason": "Tokenizer crash and dtype mismatch are unrelated failure modes." + "reason": "Tokenizer crash on NER/batched padding is unrelated to bfloat16 dtype mismatches in inference." }, { "left": "issue:26413", "right": "issue:28530", "accept": false, - "reason": "Resume-from-checkpoint epoch iterator failure is separate from early-stopping metric detection." + "reason": "Checkpoint resume epoch-iterator failure and early-stopping metric lookup are separate Trainer features with different failure modes." }, { "left": "issue:43550", "right": "issue:43720", "accept": false, - "reason": "One is a torch.compile SDPA issue; the other is packed-weight loading during accelerate\u2014different code paths." + "reason": "One is a torch.compile/SDPA failure in Bamba; the other is an Accelerate loading issue for packed BitNet weights." }, { "left": "issue:43577", "right": "issue:43708", "accept": false, - "reason": "BLIP2 dtype propagation and trainer max_steps computation are unrelated." + "reason": "BLIP2 dtype propagation and Trainer checkpoint-resume max-step calculation are unrelated." }, { "left": "issue:43404", "right": "issue:45357", "accept": false, - "reason": "Weight tying in Mistral3 is unrelated to incorrect visual encoder key serialization." + "reason": "Weight tying in Mistral3 and incorrect saved visual encoder keys are different serialization/model-structure bugs." }, { "left": "issue:43550", "right": "issue:43606", "accept": false, - "reason": "Different models and different runtime symptoms; no evidence of same root cause." + "reason": "Both are model/runtime failures, but the code paths differ: SDPA compilation versus CPU-offload device mismatch." }, { "left": "issue:43646", "right": "issue:46665", "accept": false, - "reason": "Both concern custom-model usage, but one is init breakage in v5 and the other is inconsistent `from_pretrained` behavior inside Python classes." + "reason": "Both concern custom model usage, but one is a Transformers 5.0 initialization breakage and the other is inconsistent `from_pretrained` behavior inside Python classes." }, { "left": "issue:43295", "right": "issue:45701", "accept": false, - "reason": "Processor/tokenizer regression is not the same as broader tokenization changes." + "reason": "Processor/tokenizer regression in v4.57.5 is not the same bug as general tokenization changes across versions." }, { "left": "issue:44186", "right": "issue:44610", "accept": false, - "reason": "Both involve processors, but one is tokenizer crashes on NER/batching and the other is image size mismatch." + "reason": "Different preprocessing bugs: LayoutLMv2 tokenizer crashes versus OmDet processor output-size mismatch." }, { "left": "issue:42915", "right": "issue:43317", "accept": false, - "reason": "Different quantization/offload failures with different load paths and models." + "reason": "Qwen3Moe FP8 failure and dequantized model offload loading failure are different quantization/offload issues." }, { "left": "issue:44008", "right": "issue:45823", "accept": false, - "reason": "Variable-name collision in Gemma3n forward is unrelated to Gemma4 device mismatch." + "reason": "Gemma3n variable-name collision in forward is unrelated to Gemma4 PLE device mismatch under `device_map=auto`." }, { "left": "issue:44610", "right": "issue:44898", "accept": false, - "reason": "Processor output size mismatch and Perceiver non-default resolution failure are different bugs." + "reason": "Both are vision-input shape problems, but they affect different models and the symptoms are not the same underlying bug." }, { "left": "issue:36010", "right": "issue:42617", "accept": false, - "reason": "ImportError on GenerationMixin is unrelated to a 3d_parallel launch problem." + "reason": "ImportError for `GenerationMixin` and failure to run `3d_parallel.py` are unrelated." }, { "left": "issue:44423", "right": "issue:45406", "accept": false, - "reason": "Both are serving crashes, but the failing objects and root causes differ (multimodal batching vs missing processor tokenizer attr)." + "reason": "Both hit `transformers serve`, but one is a `'str' object has no attribute 'to'` crash and the other is a missing `_tokenizer` attribute; different failures." }, { "left": "issue:45632", "right": "issue:45684", "accept": false, - "reason": "Cache-path collision and read-only permission propagation are unrelated save/load issues." + "reason": "Cache-path collision for local trust-remote-code models and read-only permission propagation during save_pretrained are unrelated." }, { "left": "issue:43531", "right": "issue:43572", "accept": false, - "reason": "Sliding-window behavior in Qwen3-MoE is unrelated to missing `pad_token_idx` in StableLmConfig." + "reason": "Qwen3-MoE sliding-window behavior and missing `pad_token_idx` in StableLmConfig are different configuration/model bugs." }, { "left": "issue:28530", "right": "issue:33666", "accept": false, - "reason": "Early stopping metric handling and Qwen2-VL multi-GPU training are not the same problem." + "reason": "Early stopping metric resolution and Qwen2-VL multi-GPU training are not the same issue." }, { "left": "issue:44671", "right": "issue:44743", "accept": false, - "reason": "CamemBERT masked-LM prediction regression and Qwen3.5 recurrent-state reset are different model bugs." + "reason": "CamemBERT masked-LM prediction regression and Qwen3.5 recurrent-state reset are unrelated model bugs." }, { "left": "issue:42915", "right": "issue:46291", "accept": false, - "reason": "FP8 model failure is unrelated to a pipelines import dependency/version error." + "reason": "Qwen3Moe FP8 failure and kernels import requiring version/revision are different dependency/runtime problems." }, { "left": "issue:36331", "right": "issue:41720", "accept": false, - "reason": "Trainer API mismatch and Qwen3 auto device-map assert are different failures." + "reason": "Trainer compute_loss signature mismatch is unrelated to a Qwen3 auto-device CUDA assertion." }, { "left": "issue:44186", "right": "issue:44361", "accept": false, - "reason": "Tokenizer crashes, but on different tokenizers/models and different triggers; too broad to merge." + "reason": "Different tokenizer failures affecting different models and tasks; not the same underlying bug." } ] }, @@ -4851,208 +4854,214 @@ "issue:44077|issue:44625", "issue:43644|issue:46665" ], - "prepared_review_unit_hash": "f02c73930ba064a3b8e66e980a080c6805dd095b728b9c33d08778ea89b4d5b3", + "prepared_review_unit_hash": "6fbd2ab79d1d091db3747769dd09470e9f48f6f179a6601d83be68017e303394", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36374, - "estimated_input_tokens": 9094, - "estimated_eval_tokens": 18444 + "serialized_chars": 36369, + "estimated_input_tokens": 9093, + "estimated_eval_tokens": 18442 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These issues are mostly broad but unrelated regression reports across tokenizers, multimodal processors, config serialization, offloading/quantization, and serving. The soft pairs are title-similar only and do not look like the same underlying bug or change.", - "confidence": 0.91, - "canonical_issue_reason": "No true duplicate stands out in this cluster; issue 44534 is the broadest cross-cutting regression and the best anchor because it describes a general Transformers v5 buffer-state bug rather than a model-specific symptom.", + "summary": "This cluster is over-broad and mostly contains unrelated issue reports. The only clear duplicate-looking pair is the two Transformers v5 non-persistent buffer junk reports; the rest differ by model, subsystem, or bug class.", + "confidence": 0.88, + "canonical_issue_reason": "issue:44534 is the clearest representative of the one true duplicate pair here: it uses the broader, version-agnostic wording and matches issue:43644\u2019s buffer-junk report. The rest of the cluster is not one cohesive issue.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 44534 is the most suitable global anchor: concrete, broadly phrased, and representative of a core runtime regression, whereas the others are narrower or clearly different subsystems.", + "best_issue_reason": "issue:44534 is the best single issue to anchor the duplicate pair because it is the more general and later wording of the same underlying bug.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43526", "right": "issue:43881", "accept": false, - "reason": "Different failures: BEiT fast image label reduction vs GLM-4V model loading; no shared code-path." + "reason": "Different components and failures: BeitImageProcessorFast label reduction vs glm-4v-9b loading failure." }, { "left": "issue:43316", "right": "issue:44913", "accept": false, - "reason": "One is a Gemma config API mismatch, the other a GPTNeoX rotary_pct reload bug; related to configs but not the same issue." + "reason": "Both are config-related, but one is Gemma3TextConfig API mismatch and the other is GPTNeoX rotary_pct persistence." }, { "left": "issue:42915", "right": "issue:43322", "accept": false, - "reason": "Both are model-loading problems, but for different models and symptoms (FineGrainedFP8Config vs Llava Next segfault)." + "reason": "Both are loading failures, but for different models and likely different root causes (Qwen3Moe FP8 vs Llava Next segfault)." }, { "left": "issue:28530", "right": "issue:30064", "accept": false, - "reason": "Early stopping metric selection and void segmentation map processing are unrelated." + "reason": "Early stopping metric lookup bug vs image processor void segmentation map handling; unrelated subsystems." }, { "left": "issue:33357", "right": "issue:35707", "accept": false, - "reason": "CLIP MacOS bus error is unrelated to progressive generation with inputs_embeds/past_key_values." + "reason": "MacOS CLIP bus error vs generation with inputs_embeds/past_key_values; not the same bug." }, { "left": "issue:43540", "right": "issue:43782", "accept": false, - "reason": "Different Qwen model families and different failure modes; not the same bug." + "reason": "Both involve Qwen multimodal models, but one is video input processing and the other is from_pretrained weight_only loading." }, { "left": "issue:30064", "right": "issue:34634", "accept": false, - "reason": "Image segmentation map handling vs BarkProcessor voice_preset behavior; separate components." + "reason": "Different image processor/tokenizer bugs; one is segmentation map void handling, the other BarkProcessor voice_preset behavior." }, { "left": "issue:43873", "right": "issue:45704", "accept": false, - "reason": "Quantization/offloading bug vs T5 apex RMSNorm memory leak; distinct code paths." + "reason": "Both mention memory/offloading-related problems, but one is quantization offloading and the other is apex FusedRMSNorm leak." }, { "left": "issue:43550", "right": "issue:43653", "accept": false, - "reason": "Bamba SDPA/torch.compile issue is unrelated to BigBirdTokenizer special-token registration." + "reason": "Different models and failure modes: Bamba torch.compile SDPA vs BigBirdTokenizer special-token decode issue." }, { "left": "issue:44387", "right": "issue:45701", "accept": false, - "reason": "CUDA memory growth under int4 quantization is not the same as tokenization changes across versions." + "reason": "Int4 quantization OOM vs tokenization changes across versions; not the same underlying defect." }, { "left": "issue:44492", "right": "issue:44509", "accept": false, - "reason": "A typo in cache strategies is unrelated to docs mentioning removed pipeline tasks." + "reason": "A typo in cache strategies is unrelated to removed pipeline docs." }, { "left": "issue:29159", "right": "issue:36032", "accept": false, - "reason": "Slow/fast tokenizer inconsistency is different from the T5 add_special_tokens load-time conflict." + "reason": "Tokenizer inconsistency vs T5Tokenizer add_special_tokens name conflict; different API problems." }, { "left": "issue:43606", "right": "issue:43720", "accept": false, - "reason": "Both involve loading/offload, but one is a Bark CPU-offload device mismatch and the other is packed-weight unpacking for BitNet." + "reason": "Both CI/loading issues, but Bark CPU offload device mismatch is not the same as BitNet packed-weight unpacking." }, { "left": "issue:44534", "right": "issue:46665", "accept": false, - "reason": "Non-persistent buffer junk in v5 is a concrete runtime bug; from_pretrained-in-class inconsistency is a different API behavior issue." + "reason": "One is a specific v5 buffer initialization bug; the other is a generic from_pretrained behavior inconsistency." }, { "left": "issue:43531", "right": "issue:45759", "accept": false, - "reason": "Qwen3-MoE sliding_window behavior and multimodal text_config unwrapping are separate issues." + "reason": "Qwen3-MoE sliding_window behavior vs from_config not unwrapping text_config; different code paths." }, { "left": "issue:41720", "right": "issue:43746", "accept": false, - "reason": "Qwen3 auto device mapping CUDA assert is unrelated to GraniteSpeech PEFT checkpoint loading." + "reason": "Qwen3 auto device mapping CUDA assert vs GraniteSpeech PEFT checkpoint loading; not the same issue." }, { "left": "issue:42490", "right": "issue:43329", "accept": false, - "reason": "Trailing-slash local path import warning is unrelated to multimodal token counting in the video branch." + "reason": "Trailing-slash import warning vs multimodal token counting video-branch bug; unrelated." }, { "left": "issue:43388", "right": "issue:45059", "accept": false, - "reason": "Mask2Former last-batch label dropping and SAM3 text/box behavior are different problems." + "reason": "Metric gathering label truncation vs SAM3 text/bbox behavior; different abstractions." }, { "left": "issue:43606", "right": "issue:43653", "accept": false, - "reason": "CPU offload device mismatch is not the same as BigBirdTokenizer mask-token registration." + "reason": "Both are CI failures, but one is offload/device mismatch and the other is tokenizer special-token registration." }, { "left": "issue:45406", "right": "issue:45464", "accept": false, - "reason": "Gemma4 serve crash from missing _tokenizer is unrelated to Qwen3.5 streaming chat/completions failure." + "reason": "Gemma4Processor missing _tokenizer in serve vs Qwen3.5 streaming chat/completions failure; different failures." }, { "left": "issue:34567", "right": "issue:35707", "accept": false, - "reason": "Trainer token counting bug and progressive generation with past_key_values are unrelated." + "reason": "TrainerState token counting bug vs generation with inputs_embeds/past_key_values; no shared root cause." }, { "left": "issue:34634", "right": "issue:41720", "accept": false, - "reason": "Bark voice_preset handling does not match Qwen3 auto device-mapping CUDA failure." + "reason": "Bark voice_preset issue and Qwen3 device-mapping CUDA assert are unrelated." }, { "left": "issue:45357", "right": "issue:46402", "accept": false, - "reason": "Saving wrong visual encoder keys and not loading Sam3 text encoder weights are different load/save regressions." + "reason": "Qwen3.5 save_pretrained visual encoder key regression vs Sam3Model text encoder weights not loaded; different save/load paths." }, { "left": "issue:43531", "right": "issue:44589", "accept": false, - "reason": "Sliding-window behavior in Qwen3-MoE is not the same as missing Float8 storage objects." + "reason": "Sliding-window Qwen3-MoE bug vs float8 storage lookup error; different model/runtime issues." }, { "left": "issue:44451", "right": "issue:44991", "accept": false, - "reason": "A specific model load failure for ScandiBERT is different from tokenizer loading failure for est-roberta." + "reason": "Both are model/tokenizer load failures, but for different checkpoints and likely different tokenizer incompatibilities." }, { "left": "issue:45759", "right": "issue:45859", "accept": false, - "reason": "from_config text_config unwrapping and CPU-offload rotary embedding compatibility are different Qwen3.5 issues." + "reason": "One is AutoModelForCausalLM config unwrapping; the other is CPU offload compatibility in Qwen3_5MoeTextRotaryEmbedding." }, { "left": "issue:44625", "right": "issue:45479", "accept": false, - "reason": "num_labels propagation in Qwen3.5 configs is not the same as degenerate zero-loss classification behavior." + "reason": "num_labels propagation in Qwen3.5 config is distinct from degenerate zero-loss behavior in sequence classification." }, { "left": "issue:44077", "right": "issue:44625", "accept": false, - "reason": "patchtsmixer post_init policy and Qwen3.5 num_labels propagation are unrelated." + "reason": "PatchTSMixer post_init allowance and Qwen3.5 num_labels propagation are unrelated config issues." + }, + { + "left": "issue:43644", + "right": "issue:44534", + "accept": true, + "reason": "Same underlying Transformers v5 bug: non-persistent buffers being filled with junk; titles and symptoms match closely." }, { "left": "issue:43644", "right": "issue:46665", "accept": false, - "reason": "Filled non-persistent buffers and from_pretrained inside Python classes are different bug classes." + "reason": "Buffer initialization bug is unrelated to from_pretrained behavior inside Python classes." } ] }, @@ -5157,217 +5166,223 @@ "issue:43653|issue:43720", "issue:43596|issue:44220" ], - "prepared_review_unit_hash": "2147ff213331948b7b8830a2f1bcbee6c608d5f3d2e24292d1bb549bdf9268e5", + "prepared_review_unit_hash": "76926e5e3b0c6b04db2736d16353650835d81344e2786c5d819f8da8ce84ee48", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 37065, - "estimated_input_tokens": 9267, - "estimated_eval_tokens": 18790 + "serialized_chars": 37070, + "estimated_input_tokens": 9268, + "estimated_eval_tokens": 18792 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous; nearly all pairs are only loosely related by broad themes like tokenizers, loading, or multimodal processors. The only clear duplicate is the two identical Qwen2_5_VLProcessor padding=False reports.", - "confidence": 0.86, - "canonical_issue_reason": "issue:44545 is the same bug as issue:44514 and has slightly more discussion/inbound signal, so it is the better canonical anchor.", + "summary": "This set is mostly heterogeneous across unrelated tokenizer, vision, training, quantization, and loading bugs. Only one clear duplicate pair stands out: the two Qwen2_5_VLProcessor batched-input padding=False crash reports.", + "confidence": 0.93, + "canonical_issue_reason": "issue:44545 is the best canonical issue because it is an exact duplicate of issue:44514 (same title and same batched-input crash with padding=False).", "canonical_pr_reason": null, - "best_issue_reason": "issue:44545 is the best issue anchor because it is an exact duplicate of issue:44514 and has slightly more discussion/inbound signal than its twin.", + "best_issue_reason": "issue:44545 is the strongest issue representative for the only true duplicate pair in this cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44062", "right": "issue:44451", "accept": false, - "reason": "Both are tokenizer/load-related, but the concrete failures differ: AddedToken keyword handling vs loading a specific model." + "reason": "Different problems: AddedToken constructor special-arg error vs model loading failure for ScandiBERT." }, { "left": "issue:44898", "right": "issue:45072", "accept": false, - "reason": "Different subsystems and symptoms: Perceiver resolution handling versus bfloat16 dtype mismatches." + "reason": "Different vision regressions: Perceiver input-resolution handling vs bfloat16 dtype mismatch in SwitchTransformers/TimmWrapperModel." }, { "left": "issue:44987", "right": "issue:45042", "accept": false, - "reason": "One is a model loading failure, the other is an image-processor dependency regression; not the same bug." + "reason": "One is a model-loading regression; the other is a PIL backend processor dependency issue. Not the same bug." }, { "left": "issue:44368", "right": "issue:45357", "accept": false, - "reason": "A LoRA warning about tie_word_embeddings is unrelated to incorrect visual encoder keys in save_pretrained." + "reason": "Different code paths: tie_word_embeddings warning vs incorrect visual encoder keys saved by Qwen3.5." }, { "left": "issue:43381", "right": "issue:43844", "accept": false, - "reason": "Eval-mode gradient checkpointing prohibition and ZeRO-3 gradient blowup are different issues." + "reason": "Gradient-checkpointing eval-mode guard and ZeRO-3 gradient blow-up are distinct failures." }, { "left": "issue:39692", "right": "issue:44610", "accept": false, - "reason": "Both involve vision models, but they are different model/processor mismatches with different expected sizes and failure modes." + "reason": "Both involve processor/model mismatch, but for different models and different defects." }, { "left": "issue:41720", "right": "issue:45704", "accept": false, - "reason": "Qwen3 auto device-map CUDA asserts and T5 RMSNorm memory leaks are unrelated code paths." + "reason": "Qwen3 auto device mapping CUDA assert is unrelated to the T5 apex RMSNorm memory leak." }, { "left": "issue:45704", "right": "issue:45923", "accept": false, - "reason": "A T5 memory leak is unrelated to a missing supports_gradient_checkpointing flag on a trust_remote_code model." + "reason": "Memory leak in T5 RMSNorm is unrelated to a missing supports_gradient_checkpointing flag." }, { "left": "issue:43531", "right": "issue:43645", "accept": false, - "reason": "Sliding-window behavior in Qwen3-MoE and Jupyter custom-model initialization are unrelated." + "reason": "Sliding-window behavior in Qwen3-MoE is unrelated to custom-model initialization in Jupyter notebooks." }, { "left": "issue:46655", "right": "issue:46693", "accept": false, - "reason": "SmolVLM video padding and FA2 long-context performance regression are different problems." + "reason": "Video padding calculation bug vs long-context FA2 performance regression; different underlying issues." }, { "left": "issue:43746", "right": "issue:46050", "accept": false, - "reason": "PEFT local checkpoint loading and sinq save/load quantization corruption do not look like the same bug." + "reason": "Local checkpoint loading with PEFT adapters is not the same as sinq quantized save/load corruption." }, { "left": "issue:43381", "right": "issue:45704", "accept": false, - "reason": "Gradient checkpointing in eval mode is unrelated to the T5 RMSNorm leak." + "reason": "Different training/runtime bugs: eval-mode checkpointing restriction vs T5 memory leak." }, { "left": "issue:43421", "right": "issue:43927", "accept": false, - "reason": "Runtime special-token post-processing updates and DiaConfig token-ID persistence are different issues." + "reason": "Both touch token handling, but one is runtime post-processor refresh and the other is config persistence of custom token IDs." }, { "left": "issue:44918", "right": "issue:44936", "accept": false, - "reason": "Unpacking Qwen3.5 embeddings in TRL SFT and trainer.evaluate() failing after train() are not the same failure." + "reason": "Different trainer failures: SFT embedding unpacking vs evaluate() failing after train()." }, { "left": "issue:42886", "right": "issue:44843", "accept": false, - "reason": "Both mention offline/tokenizer loading, but 44843 is a specific Mistral-regex root cause while 42886 is broader and not clearly the same bug." + "reason": "Both are offline-related, but one is generic cache loading and the other is a specific mistral-regex model_info call path." }, { "left": "issue:45759", "right": "issue:45923", "accept": false, - "reason": "Composite multimodal config unwrapping and a missing gradient-checkpointing flag are unrelated." + "reason": "Composite-config unwrapping in from_config is unrelated to a missing gradient-checkpointing flag." }, { "left": "issue:44514", "right": "issue:44545", "accept": true, - "reason": "Exact same title and same crash in Qwen2_5_VLProcessor.apply_chat_template with padding=False." + "reason": "Exact duplicate: same title and same batched-input crash when padding=False." }, { "left": "issue:46032", "right": "issue:46463", "accept": false, - "reason": "Mamba2Mixer cache correctness and a T5 error-message issue are unrelated." + "reason": "Mamba2 use_cache correctness bug is unrelated to a T5 error-message validation issue." }, { "left": "issue:39692", "right": "issue:43873", "accept": false, - "reason": "SigLIP2 docs example issues and quantization/offloading behavior are different problems." + "reason": "Different topics: SigLIP2 doc example errors vs quantization offloading behavior." }, { "left": "issue:43746", "right": "issue:43873", "accept": false, - "reason": "Local PEFT checkpoint loading and quantization offloading behavior do not match." + "reason": "PEFT checkpoint loading and quantization offloading are different code paths and different failures." }, { "left": "issue:43994", "right": "issue:44671", "accept": false, - "reason": "Both are model-output correctness reports, but for different models and likely different code paths." + "reason": "Different models and symptoms: SigLIP2 nonsensical outputs vs CamemBERT masked-LM regression." }, { "left": "issue:43421", "right": "issue:43653", "accept": false, - "reason": "Special-token post-processor updates and BigBirdTokenizer mask-token registration are related thematically but not the same concrete bug." + "reason": "Both involve special tokens, but one is backend post-processor updates and the other is BigBird mask-token registration/decode output." }, { "left": "issue:30064", "right": "issue:36010", "accept": false, - "reason": "ImageProcessor void segmentation handling and a GenerationMixin import error are unrelated." + "reason": "Image segmentation-map processing and GenerationMixin import failure are unrelated." }, { "left": "issue:43782", "right": "issue:45381", "accept": false, - "reason": "Different Qwen VL failures: weight_only loading vs incorrect vision_position_ids in video input." + "reason": "Both are Qwen VL issues, but one is weight_only loading and the other is incorrect video vision_position_ids." + }, + { + "left": "issue:43596", + "right": "issue:44220", + "accept": false, + "reason": "Deepspeed ZeRO-3/BertModel init error is unrelated to _torch_extract_fbank_features()." }, { "left": "issue:45081", "right": "issue:45356", "accept": false, - "reason": "Both touch fix_mistral_regex, but one is a tokenizer crash and the other is a codec/regression warning issue; not clearly the same bug." + "reason": "Both mention fix_mistral_regex, but one crashes on backend_tokenizer access while the other is a codec-handling regression with a warning issue." }, { "left": "issue:45356", "right": "issue:45907", "accept": false, - "reason": "A tokenizer regression is unrelated to list_repo_files revision forwarding." + "reason": "Tokenizer regression in Kimi-K2.5 is unrelated to list_repo_files not forwarding revision." }, { "left": "issue:43720", "right": "issue:43819", "accept": false, - "reason": "BitNet packed-weight loading and DAC latent conversion mismatch are unrelated." + "reason": "BitNet packed-weight loading and DAC latent reconstruction mismatch are different bugs." }, { "left": "issue:43653", "right": "issue:43720", "accept": false, - "reason": "BigBirdTokenizer special-token registration and BitNet accelerate loading are unrelated." + "reason": "BigBird special-token decode bug is unrelated to BitNet accelerate-loading unpack behavior." }, { "left": "issue:43596", "right": "issue:44220", "accept": false, - "reason": "DeepSpeed ZeRO-3/BertModel index error and _torch_extract_fbank_features are different issues." + "reason": "Different subsystems and failure modes; not a duplicate pair." } ] }, "evaluator_result": { - "accept": false, - "feedback": "The duplicate judgment for issue:44514/issue:44545 is well-grounded, and choosing issue:44545 as the canonical anchor is reasonable because it has slightly more discussion/inbound signal. However, the best_issue_reason overstates the evidence by calling 44545 \"the richest context in the cluster\"; several other issues in the packet have clearly higher discussion/inbound activity. The rest of the soft-edge rejections look conservative." + "accept": true, + "feedback": "The cluster summary is well grounded in the packet: it identifies one genuinely duplicate pair (44514/44545) and the remaining soft-edge rejections are cautious and supported by the issue titles. No overstatement beyond the evidence is apparent." }, "evaluator_used": true, - "retried": true, + "retried": false, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -5464,20 +5479,20 @@ "issue:34634|issue:36010", "issue:45584|issue:45834" ], - "prepared_review_unit_hash": "069859e623af41b7b18689e1693e3ef21c25c3b5fe682d56634439f9eeabffbb", + "prepared_review_unit_hash": "14175961c4a33ba78afe31bbfc1a23ef4836be96f17040943f466629a68e4cf8", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 36081, + "serialized_chars": 36082, "estimated_input_tokens": 9021, "estimated_eval_tokens": 18298 }, @@ -5486,180 +5501,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: most items are unrelated bug reports across different models, tokenizers, loading paths, and runtime regressions. The soft pairs mostly share only broad subsystem vocabulary (e.g. v5 regressions, tokenizer/loading, Qwen, vision) rather than the same concrete underlying defect.", - "confidence": 0.89, - "canonical_issue_reason": "Issue 43611 is the broadest representative of the cluster\u2019s main recurring theme: v5 compatibility/regression breakage around model loading. It is still only a loose fit, but it is the closest \u201cumbrella\u201d issue among otherwise unrelated reports.", + "summary": "This cluster is mostly heterogeneous; the soft pairs look like different bugs across unrelated models/features rather than true duplicates. No soft edge should be merged.", + "confidence": 0.26, + "canonical_issue_reason": "If a single issue must anchor the cluster, issue 43611 is the broadest user-facing regression report: a Transformers 5.0.0 loading break that can generalize across multiple model classes.", "canonical_pr_reason": null, - "best_issue_reason": "43611 is the best single issue to anchor the cluster because it is generic, central, and captures the widest compatibility failure pattern without being overly model-specific.", + "best_issue_reason": "Issue 43611 is the best representative because it is a broad core regression report rather than a narrow model-specific symptom.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43550", "right": "issue:45072", "accept": false, - "reason": "Different failures: torch.compile+SDPA on Bamba vs dtype mismatches in bfloat16 on SwitchTransformers/TimmWrapperModel." + "reason": "Different failure modes: torch.compile+SDPA on Bamba vs bfloat16 dtype mismatch in SwitchTransformers/TimmWrapperModel." }, { "left": "issue:43295", "right": "issue:45834", "accept": false, - "reason": "One is a processor/tokenizer regression; the other is a Kosmos2.5 long-OCR index error." + "reason": "Unrelated domains: processor.tokenizer regression vs Kosmos2.5 OCR length/indexing error." }, { "left": "issue:43550", "right": "issue:43927", "accept": false, - "reason": "Unrelated model/runtime compile issue vs DiaConfig token ID persistence bug." + "reason": "Different code paths and symptoms: compiled SDPA failure vs config save/load losing token IDs." }, { "left": "issue:43540", "right": "issue:46294", "accept": false, - "reason": "Both involve Qwen video-related behavior, but one is sliding_window/MoE and the other is chat-template video token alignment; not the same bug." + "reason": "Both involve Qwen variants, but one is sliding_window/MoE and the other is video chat-template token/feature mismatch." }, { "left": "issue:44062", "right": "issue:44488", "accept": false, - "reason": "Tokenizer AddedToken argument conflict is unrelated to loading a specific tokenizer model." + "reason": "Token construction bug vs tokenizer loading failure for a specific model; not the same underlying defect." }, { "left": "issue:45381", "right": "issue:46693", "accept": false, - "reason": "Qwen2.5-VL video position IDs bug is distinct from FA2 long-context performance regression." + "reason": "Different regressions: Qwen2.5-VL video position IDs vs FA2 long-context performance degradation." }, { "left": "issue:43618", "right": "issue:43697", "accept": false, - "reason": "Both are v5 output-related, but on different models and different code paths; not a single duplicate defect." + "reason": "Both are v5 model regressions, but on different architectures and different observable failures." }, { "left": "issue:43531", "right": "issue:43643", "accept": false, - "reason": "Sliding-window behavior in Qwen3-MoE is unrelated to AutoConfig trust_remote_code field loss." + "reason": "Qwen3-MoE sliding_window bug is unrelated to trust_remote_code config field loss." }, { "left": "issue:43611", "right": "issue:46665", "accept": false, - "reason": "Both touch loading behavior, but one is a base_model_prefix regression and the other is from_pretrained behavior inside Python classes." + "reason": "Both touch loading, but one is a base_model_prefix regression in v5 and the other is inconsistent from_pretrained behavior inside Python classes." }, { "left": "issue:44220", "right": "issue:44661", "accept": false, - "reason": "Audio feature extraction issue vs add-new-model-like failure inside tokenizer mapping; different subsystems and failures." + "reason": "Audio feature extraction bug vs add-new-model script failure in tokenizer mapping; different subsystems." }, { "left": "issue:43653", "right": "issue:43819", "accept": false, - "reason": "BigBirdTokenizer special-token decode bug is unrelated to DAC latent decoding mismatch." + "reason": "Tokenizer special-token registration bug vs DAC latent reconstruction mismatch; no shared code path." }, { "left": "issue:44186", "right": "issue:44898", "accept": false, - "reason": "Tokenizer NER/padding crash and Perceiver image-classification resolution failure are different model families and bugs." + "reason": "Tokenizer/NER padding crash vs Perceiver image classification resize/interpolation failure." }, { "left": "issue:30064", "right": "issue:36331", "accept": false, - "reason": "ImageProcessor void segmentation map handling is unrelated to CustomTrainer.compute_loss kwargs regression." + "reason": "Image segmentation-map processing failure is unrelated to Trainer.compute_loss signature changes." }, { "left": "issue:43643", "right": "issue:44877", "accept": false, - "reason": "Both involve config loading, but one is missing fields with trust_remote_code and the other is strict config rejection for granite_speech." + "reason": "Both involve config loading, but one loses custom fields under trust_remote_code and the other is strict config rejection of granite_speech." }, { "left": "issue:43335", "right": "issue:43756", "accept": false, - "reason": "SwitchTransformers sparse-layer creation bug is unrelated to Smollm3 RoPE layer dropping." + "reason": "Different model configuration bugs: sparse layer creation vs RoPE layer dropping." }, { "left": "issue:41720", "right": "issue:44991", "accept": false, - "reason": "CUDA assert during Qwen3 auto device mapping is unrelated to tokenizer loading for EMBEDDIA/est-roberta." + "reason": "CUDA assert during Qwen3 device mapping is unrelated to tokenizer loading for EMBEDDIA/est-roberta." }, { "left": "issue:43550", "right": "issue:43819", "accept": false, - "reason": "Different model/runtime bug vs DAC forward/from_latents mismatch; no shared code-path." + "reason": "SDPA/compile runtime failure and DAC.from_latents mismatch are distinct bugs." }, { "left": "issue:44821", "right": "issue:45084", "accept": false, - "reason": "AutoImageProcessor URL loading and template compilation failure are unrelated." + "reason": "AutoImageProcessor URL loading and template compilation errors are unrelated." }, { "left": "issue:43540", "right": "issue:44008", "accept": false, - "reason": "Qwen3OmniMoe video-input processing differs from Gemma3n audio_mel_mask variable collision." + "reason": "Video-input validation error in Qwen3OmniMoe vs variable-name collision in Gemma3n forward." }, { "left": "issue:44843", "right": "issue:45081", "accept": false, - "reason": "Same broad _patch_mistral_regex area, but one is offline model_info access and the other is missing backend_tokenizer; not the same concrete bug." + "reason": "Both mention _patch_mistral_regex, but one is offline model_info access and the other is backend_tokenizer attribute access; separate defects." }, { "left": "issue:39692", "right": "issue:45834", "accept": false, - "reason": "SigLIP2 documentation mistakes are unrelated to Kosmos2.5 OCR long-input indexing." + "reason": "Doc/example errors for SigLIP2 vs long OCR input failure in Kosmos2.5." }, { "left": "issue:44485", "right": "issue:44509", "accept": false, - "reason": "GLM-5 RoPE implementation discussion is unrelated to docs mentioning removed pipeline tasks." + "reason": "GLM-5 RoPE implementation discussion is unrelated to docs still listing removed pipeline tasks." }, { "left": "issue:44448", "right": "issue:44671", "accept": false, - "reason": "Both are v5 prediction differences, but on different models and likely different regressions." + "reason": "Different model-specific output regressions in v5, not the same underlying change." }, { "left": "issue:44186", "right": "issue:44442", "accept": false, - "reason": "LayoutLMv2 tokenizer crash and AutoTokenizer failing to load FastSpeech2ConformerTokenizer are distinct tokenizer issues." + "reason": "LayoutLMv2 tokenizer padding/NER crash and FastSpeech2Conformer AutoTokenizer loading failure are different tokenizer issues." }, { "left": "issue:43644", "right": "issue:43749", "accept": false, - "reason": "Non-persistent buffer junk and FSDP CPU RAM efficient loading are different v5 regressions without the same defect." + "reason": "Non-persistent buffer junk values and FSDP CPU RAM efficient loading failure are separate loading/runtime problems." }, { "left": "issue:43322", "right": "issue:45685", "accept": false, - "reason": "Llava Next segmentation fault and MPS histogram error in moe are unrelated runtime failures." + "reason": "Llava Next loading segfault is unrelated to MPS histogram support for Int in MoE." }, { "left": "issue:34634", "right": "issue:36010", "accept": false, - "reason": "BarkProcessor voice_preset and GenerationMixin import error are unrelated." + "reason": "BarkProcessor voice_preset behavior and GenerationMixin import error are unrelated." }, { "left": "issue:45584", "right": "issue:45834", "accept": false, - "reason": "Whisper empty transcription handling is unrelated to Kosmos2.5 OCR indexing on long inputs." + "reason": "Whisper empty-transcription generation bug is unrelated to Kosmos2.5 long OCR input indexing." } ] }, @@ -5761,188 +5776,203 @@ "issue:42175|issue:45081", "issue:34634|issue:36331" ], - "prepared_review_unit_hash": "c476fd1b2833047cdcebbf9a3f40d9a8d8e5b3f4d453ab2dee95f4ba2deb4d7b", + "prepared_review_unit_hash": "cde82b580fd372ade7235eabb839ae372b4c1ebfd4d5785888d85c8815c2bf2b", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 27, - "serialized_chars": 36336, - "estimated_input_tokens": 9084, - "estimated_eval_tokens": 18424 + "serialized_chars": 36340, + "estimated_input_tokens": 9085, + "estimated_eval_tokens": 18426 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly heterogeneous bug reports with no PRs; nearly all soft pairs are false positives across unrelated subsystems. The only plausible duplicate is the tokenizer added-token spacing/inconsistency pair.", + "summary": "This set is a loose mix of unrelated Transformers issues; none of the soft pairs look like the same underlying bug/change, so no merges are justified. The best single anchor is the highly discussed FSDP loading regression issue, but even that is only a representative issue, not a duplicate hub.", "confidence": 0.89, - "canonical_issue_reason": "issue:29159 is the broader tokenizer inconsistency report and best captures the shared added-token spacing symptom from issue:28218.", + "canonical_issue_reason": "issue:43749 is the strongest central anchor by discussion activity and inbound references, and it is a concrete regression report with substantial debugging context.", "canonical_pr_reason": null, - "best_issue_reason": "issue:29159 is the best issue anchor because it is the most general and likely to subsume the duplicate tokenizer behavior report.", + "best_issue_reason": "issue:43749 is the most suitable global representative because it has the most engagement and a clearly scoped, actionable failure mode.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43606", "right": "issue:45072", "accept": false, - "reason": "Different failures: device mismatch during CPU offload vs dtype mismatch in bfloat16 inference." + "reason": "Both mention mismatch errors, but they affect different models and different paths: CPU offload/device placement vs bfloat16 dtype handling." }, { "left": "issue:45183", "right": "issue:45479", "accept": false, - "reason": "Unrelated bugs: input validation error messaging vs degenerate classification loss." + "reason": "One is about input_ids/inputs_embeds validation messaging; the other is a loss/computation bug in sequence classification. Different problems." }, { "left": "issue:43716", "right": "issue:44008", "accept": false, - "reason": "Different multimodal model issues and different failure modes." + "reason": "Different model families and different root causes: image-preprocessor dtype mismatch vs an audio variable-name collision." }, { "left": "issue:43388", "right": "issue:46032", "accept": false, - "reason": "Different subsystems: metrics gathering drops labels vs Mamba cache correctness." + "reason": "Different subsystems and failure modes: metrics gathering on the last batch vs incorrect Mamba2 cache results." }, { "left": "issue:44479", "right": "issue:45834", "accept": false, - "reason": "Both are video/vision regressions, but the models and likely code paths are different." + "reason": "Both are vision/video related, but they describe different models and distinct regressions." }, { "left": "issue:43950", "right": "issue:45987", "accept": false, - "reason": "Different areas: buffer loading corruption vs cache length return type." + "reason": "from_pretrained buffer corruption and StaticCache return-type issues are unrelated bugs." }, { "left": "issue:43582", "right": "issue:46512", "accept": false, - "reason": "Apple Silicon allocator TypeError is unrelated to stale config dtype after from_config." + "reason": "Apple Silicon warmup TypeError is unrelated to from_config leaving config.dtype stale." }, { "left": "issue:43825", "right": "issue:43994", "accept": false, - "reason": "One is an error-message/doc issue, the other is a model inference regression." + "reason": "An error-message regression is not the same as a model output correctness bug." }, { "left": "issue:44991", "right": "issue:46291", "accept": false, - "reason": "Tokenizer loading failure and kernels import/version requirement are unrelated." + "reason": "Tokenizer-loading failure for a specific model and kernels package import failure are different breakages." }, { "left": "issue:43577", "right": "issue:43818", "accept": false, - "reason": "Different model families and different dtype/architecture problems." + "reason": "Blip2 dtype-loading regression and Video-LLaVA architecture/weight-sharing issue do not share a concrete code path." }, { "left": "issue:46294", "right": "issue:46463", "accept": false, - "reason": "Video chat-template mismatch vs T5 decoder-input validation error." + "reason": "One concerns Qwen video chat template feature alignment; the other is T5 decoder-input validation messaging." }, { "left": "issue:43782", "right": "issue:44977", "accept": false, - "reason": "Different regressions: weight_only loading error vs flash-attention generation problem." + "reason": "Qwen3VL weight-only loading and Qwen3.5 flash-attention generation failures are different issues." }, { "left": "issue:44561", "right": "issue:45020", "accept": false, - "reason": "Both mention remote_code/v5 breakage, but the titles do not point to the same concrete underlying bug." + "reason": "Both involve trust_remote_code/remote_code breakage, but the specific removed API and the broader compatibility regression are not the same bug." + }, + { + "left": "issue:43421", + "right": "issue:44811", + "accept": false, + "reason": "Runtime special-token post-processing update and Whisper skip_special_tokens handling are separate tokenizer behaviors." + }, + { + "left": "issue:44805", + "right": "issue:44936", + "accept": false, + "reason": "Mask shape IndexError and trainer.evaluate() failing after train are unrelated training/runtime issues." + }, + { + "left": "issue:43540", + "right": "issue:45381", + "accept": false, + "reason": "Different Qwen video regressions with different symptoms and likely different fixes." }, { "left": "issue:44610", "right": "issue:45072", "accept": false, - "reason": "Processor image-size mismatch is unrelated to dtype mismatches." + "reason": "Processor image-size mismatch and dtype mismatch in inference are unrelated." }, { "left": "issue:43720", "right": "issue:45072", "accept": false, - "reason": "Packed-weight loading failure is unrelated to dtype inference issues." + "reason": "Packed-weight unpacking during accelerate loading is a different failure mode from dtype mismatches." }, { "left": "issue:43825", "right": "issue:43827", "accept": false, - "reason": "Both are migration fallout around pipeline removals, but one is runtime messaging and the other is docs cleanup." + "reason": "One is an error message bug; the other is stale documentation after pipeline removals." }, { "left": "issue:43749", "right": "issue:44291", "accept": false, - "reason": "Different loading paths and different exceptions." + "reason": "These are both loading-related, but one is FSDP CPU RAM-efficient loading and the other is init_empty_weights argument handling." }, { "left": "issue:28218", "right": "issue:29159", - "accept": true, - "reason": "These describe the same tokenizer behavior regression: extra spacing after an added token / slow-fast tokenizer inconsistency." + "accept": false, + "reason": "Tokenizer spacing after added tokens and slow/fast tokenizer inconsistency are too broad and not clearly the same underlying bug." }, { "left": "issue:43645", "right": "issue:46665", "accept": false, - "reason": "Notebook custom-model initialization and from_pretrained-in-classes are different behaviors." + "reason": "Jupyter custom-model initialization and from_pretrained-in-classes inconsistencies are different usage scenarios." }, { "left": "issue:43577", "right": "issue:43582", "accept": false, - "reason": "Unrelated dtype-loading bug vs allocator warmup TypeError." + "reason": "Blip2 dtype persistence and Apple Silicon allocator warmup TypeError are unrelated." }, { "left": "issue:45987", "right": "issue:46192", "accept": false, - "reason": "Cache length return-type bug is unrelated to return_tensors being ignored." + "reason": "A cache-length contract bug and text_kwargs/return_tensors handling are different API issues." }, { "left": "issue:44671", "right": "issue:45701", "accept": false, - "reason": "Tokenization changes may be involved, but one is a CamemBERT prediction regression and the other is a broad version-change complaint." + "reason": "CamemBERT masked-LM regression and version-dependent tokenization changes are not the same code-path problem." }, { "left": "issue:42175", "right": "issue:45081", "accept": false, - "reason": "Backend packaging and tokenizer regex patching are unrelated." + "reason": "Backend dependency packaging and Mistral regex patch/tokenizer attribute access are unrelated." }, { "left": "issue:34634", "right": "issue:36331", "accept": false, - "reason": "BarkProcessor voice_preset failure is unrelated to CustomTrainer compute_loss signature breakage." + "reason": "BarkProcessor voice_preset and CustomTrainer.compute_loss signature errors are entirely separate issues." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary stays within the packet evidence: the tokenizer pair is the only clearly plausible duplicate, and the other soft-edge rejections are phrased conservatively based on titles/subsystems without overstating linkage. No PR-pair mergeability issues apply here." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -6038,212 +6068,209 @@ "issue:42490|issue:43257", "issue:39692|issue:44912" ], - "prepared_review_unit_hash": "f67792f8a5d9b1e73b50e3f8eb82e8f218db66bb2884601acaaa44c9a800a06f", + "prepared_review_unit_hash": "3171dd26f9576afb0a038c7ac239502b504594adbde96a2297f821da8ebb4d8e", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 36589, - "estimated_input_tokens": 9148, - "estimated_eval_tokens": 18552 + "serialized_chars": 36578, + "estimated_input_tokens": 9145, + "estimated_eval_tokens": 18546 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most proposed pairs are false positives across distinct models/code paths. The only clear duplicate-like pair is the Qwen3 tied-weights warning reports; the rest are too different to merge safely.", - "confidence": 0.91, - "canonical_issue_reason": "Issue 44060 is the most concrete description of the only true duplicate pair: it identifies the specific incorrect tie between `embed_tokens.weight` and `linear_attn.dt_bias`.", + "summary": "This cluster is a grab bag of mostly unrelated Transformers issues; the soft pairs are only superficially similar and do not appear to describe the same underlying bug or change.", + "confidence": 0.89, + "canonical_issue_reason": "issue:45020 is the best canonical issue because it is broad, user-facing, and describes a core regression class ('recent versions break models using remote_code') that can subsume multiple model-loading failures.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 44060 is the best representative overall because it pinpoints the root cause, while 44368 is mainly a user-facing symptom of the same warning.", + "best_issue_reason": "issue:45020 is the strongest representative issue in the set due to its broad scope and centrality to model-loading regressions.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45003", "right": "issue:46285", "accept": false, - "reason": "Different bugs: unsafe `sys.modules` access in `modeling_utils` vs CLIP text-model structure inconsistency." + "reason": "Both touch model internals, but one is about unsafe sys.modules access and the other about CLIPTextModel structure changes; different bugs and code paths." }, { "left": "issue:43653", "right": "issue:44912", "accept": false, - "reason": "Unrelated failures: BigBird tokenizer special-token registration vs MXFP4 quantization fallback/loading." + "reason": "BigBirdTokenizer special-token decoding and MXFP4 quantization/loading fallback are unrelated." }, { "left": "issue:46697", "right": "issue:46728", "accept": false, - "reason": "Both are image-processor geometry bugs, but they affect different models/functions and do not look like the same underlying defect." + "reason": "Both are image-processor sizing issues, but they affect different models and different calculations (over-padding vs patch-count math)." }, { "left": "issue:43582", "right": "issue:44857", "accept": false, - "reason": "Different runtime errors in unrelated paths: Apple Silicon warmup TypeError vs CUDA AMP loss crash." + "reason": "Apple Silicon allocator warmup TypeError and CUDA float16 AMP crash are different runtime/platform failures." }, { "left": "issue:43262", "right": "issue:43819", "accept": false, - "reason": "Audio chat-template sampling-rate default bug is unrelated to DAC latent reconstruction / missing STE behavior." + "reason": "Audio chat-template sampling-rate defaulting and DAC latent/forward mismatch are unrelated audio bugs." }, { "left": "issue:44589", "right": "issue:45042", "accept": false, - "reason": "Float8 storage lookup failure and PIL/torchvision backend requirement are separate loading/runtime issues." + "reason": "A missing Float8 storage type and a PIL backend torchvision dependency regression are not the same problem." }, { "left": "issue:43742", "right": "issue:45020", "accept": false, - "reason": "`MobileLLM-125M` key error may be a loading regression, but it is not clearly the same concrete remote-code breakage as the broader umbrella issue." + "reason": "A specific key error while loading one model is not the same as the broad remote_code regression." }, { "left": "issue:43782", "right": "issue:45325", "accept": false, - "reason": "Different Qwen-VL bugs: weight_only loading failure vs rope-index temporal position scaling." + "reason": "One is a Qwen3VL weight_only load error; the other is a Qwen2.5-VL rope-index scaling bug. Different model families and failures." }, { "left": "issue:45081", "right": "issue:45406", "accept": false, - "reason": "Both involve processor/tokenizer loading, but they are different attribute failures in different models/code paths." + "reason": "Both involve tokenizer/processor internals, but one is a Mistral regex patch crash and the other is a Gemma4 processor missing _tokenizer; not the same bug." }, { "left": "issue:45372", "right": "issue:46326", "accept": false, - "reason": "Both mention `mistral_common`, but one is an import failure and the other is `continue_final_message` behavior; not the same bug." + "reason": "Both mention mistral_common, but one is an import failure for ReasoningEffort and the other is a continue_final_message behavior bug." }, { "left": "issue:43550", "right": "issue:44912", "accept": false, - "reason": "Different model/runtime problems: Bamba torch.compile+SDPA vs git-oss MXFP4 quantization fallback." + "reason": "torch.compile+SDPA failure and MXFP4 quantization fallback are unrelated." }, { "left": "issue:45084", "right": "issue:45538", "accept": false, - "reason": "Template compilation error is unrelated to CLIP tokenizer `model_max_length` handling." + "reason": "Template compilation TypeError and CLIPTokenizer max_length semantics are different issues." }, { "left": "issue:44060", "right": "issue:44368", - "accept": true, - "reason": "Same underlying tied-weights warning bug: 44368 reports the warning seen in practice, while 44060 identifies the incorrect cross-layer tie causing it." + "accept": false, + "reason": "They are both about Qwen weight-tying warnings, but one reports an incorrect tie to dt_bias while the other is a repeated warning during fine-tuning; not clearly the same underlying defect." }, { "left": "issue:43526", "right": "issue:43697", "accept": false, - "reason": "BEiT Fast label reduction bug and RTDetrV2 output mismatch are separate vision-model issues." + "reason": "BeitImageProcessorFast label reduction and RTDetrV2 output drift in Transformers v5 are unrelated." }, { "left": "issue:43329", "right": "issue:45600", "accept": false, - "reason": "Video-token counting bug in multimodal processors is unrelated to removed auto-mapping config references." + "reason": "A multimodal-token counting bug and stale auto_mappings config references are different code paths." }, { "left": "issue:43526", "right": "issue:44554", "accept": false, - "reason": "Different multimodal/vision bugs with no shared code path." + "reason": "Different components and failure modes: image-label reduction vs MPS attention correctness." }, { "left": "issue:43452", "right": "issue:44991", "accept": false, - "reason": "Both are tokenizer/loading regressions, but one is gguf_file support and the other is a specific model tokenizer failure; not the same defect." + "reason": "Both are loading-related, but gguf_file incompatibility and a tokenizer failure for a specific model are not the same issue." }, { "left": "issue:43257", "right": "issue:43329", "accept": false, - "reason": "Qwen3 MoE loading/weight conversion issue is unrelated to multimodal video token counting." + "reason": "Qwen3 MoE weight conversion under accelerate/deepspeed and multimodal video token counting are unrelated." }, { "left": "issue:36010", "right": "issue:36331", "accept": false, - "reason": "Import error for `GenerationMixin` and `compute_loss` signature mismatch are unrelated." + "reason": "Importing GenerationMixin and CustomTrainer.compute_loss signature changes are separate API breakages." }, { "left": "issue:36010", "right": "issue:43645", "accept": false, - "reason": "Different regressions: import path breakage vs Jupyter custom-model initialization failure." + "reason": "GenerationMixin import failure and Jupyter custom-model initialization regression are distinct issues." }, { "left": "issue:43906", "right": "issue:46209", "accept": false, - "reason": "A reproduction of an old issue and an FP8/B200 inference problem are not the same bug." + "reason": "An isolated reproduction of an older issue and an FP8 inference problem on B200 are not the same bug." }, { "left": "issue:43819", "right": "issue:43927", "accept": false, - "reason": "DAC latent conversion mismatch and DiaConfig token-ID persistence loss affect different components and symptoms." + "reason": "DAC latent reconstruction mismatch and DiaConfig token-ID persistence are unrelated." }, { "left": "issue:43531", "right": "issue:45042", "accept": false, - "reason": "Qwen3-MoE sliding-window behavior is unrelated to PIL backend torchvision dependency checks." + "reason": "Qwen3-MoE sliding-window behavior and PIL backend torchvision dependency regression have no common underlying cause." }, { "left": "issue:43638", "right": "issue:44936", "accept": false, - "reason": "Deepspeed ZeRO3/BERT index error and post-train evaluate failure are not the same code-path problem." + "reason": "Zero3/Bert training crash and evaluate-after-train failure are different trainer failures." }, { "left": "issue:42175", "right": "issue:46291", "accept": false, - "reason": "Packaging/dependency issues in different backends and import paths; not the same underlying bug." + "reason": "Packaging TensorFlow with '.[torch]' and kernels import/version requirements are separate dependency issues." }, { "left": "issue:43720", "right": "issue:44610", "accept": false, - "reason": "Packed-weight loading failure in BitNet is unrelated to OmDet processor input-size mismatch." + "reason": "BitNet packed-weight loading and OmDet-Turbo processor image-size mismatch are unrelated." }, { "left": "issue:42490", "right": "issue:43257", "accept": false, - "reason": "Trailing-slash local model path warning and Qwen3 MoE weight conversion are different loading problems." + "reason": "A trailing-slash local-path import warning is unrelated to Qwen3 MoE accelerate/deepspeed conversion." }, { "left": "issue:39692", "right": "issue:44912", "accept": false, - "reason": "Documentation example errors for SigLIP2 are unrelated to MXFP4 quantization fallback." + "reason": "SigLIP2 documentation errors and git-oss-20b MXFP4 quantization fallback are different problems despite both mentioning quantization." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The assessment is grounded in the packet and stays conservative overall. The accepted 44060\u219444368 pair is plausibly the same tied-weights warning/root-cause issue, and the other rejections are appropriately cautious given only title-level evidence. No mergeability concerns for PR pairs were present." - }, - "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 }, @@ -6336,20 +6363,20 @@ "issue:29159|issue:34567", "issue:44987|issue:45538" ], - "prepared_review_unit_hash": "011e18cfd1c6b578278957e13062835da04094726843519fbd3e5c0b3959b16d", + "prepared_review_unit_hash": "d6c4cfae12711efa71fb74a7c786713d7694989a536ef722234b668e31e6acfc", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 26, - "serialized_chars": 34991, + "serialized_chars": 34989, "estimated_input_tokens": 8748, "estimated_eval_tokens": 17752 }, @@ -6358,168 +6385,162 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The candidate pairs are mostly unrelated transformer bugs across different models, loaders, processors, and runtime features. No soft edges should be merged; this cluster looks heterogeneous rather than duplicate-heavy.", - "confidence": 0.95, - "canonical_issue_reason": "issue:43749 is the strongest anchor if one must be chosen: it has the highest discussion/inbound activity and describes a broad, well-scoped runtime-loading failure with clear user impact. That said, the set is not a true duplicate cluster.", + "summary": "The set is mostly heterogeneous: loading, tokenizer, quantization, multimodal/video, cache, and memory issues are all mixed together. No PRs are present. The only near-looking pair is the video-path issues, but they still read as separate regressions rather than one duplicate fix.", + "confidence": 0.93, + "canonical_issue_reason": "Issue 43329 is the most concrete, code-path-specific report in the set: it names undefined helpers in the video branch of `_get_num_multimodal_tokens`. It is the best single anchor for the small video-regression subset, though the overall cluster is too mixed for a true universal canonical.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43749 is the best issue candidate overall because it is prominent, actionable, and likely to attract related reports about loading/runtime failures. It is not a perfect canonical for the whole set since the other issues are about different subsystems.", + "best_issue_reason": "Issue 43329 is the strongest representative issue because it has the clearest failing helper/function and the most actionable root cause among the items, but it only fits a narrow subset of the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:38175", "right": "issue:44610", "accept": false, - "reason": "Different bugs: SigLIP zero probabilities vs OmDet-Turbo image size mismatch." + "reason": "Different bugs: one is zero-probability behavior in SigLIP2, the other is an OmDet-Turbo processor/model input-size mismatch." }, { "left": "issue:43746", "right": "issue:44811", "accept": false, - "reason": "Different subsystems: PEFT/local checkpoint loading vs Whisper batch_decode skip_special_tokens behavior." + "reason": "Separate code paths: GraniteSpeech PEFT/local-checkpoint loading vs Whisper `batch_decode()` skipping special-token handling." }, { "left": "issue:43643", "right": "issue:45759", "accept": false, - "reason": "Different composite-config issues: missing fields from trust_remote_code AutoConfig vs AutoModelForCausalLM not unwrapping text_config." + "reason": "Both concern config/model loading, but one is `trust_remote_code` dropping fields in `AutoConfig`, while the other is `AutoModelForCausalLM.from_config` not unwrapping `text_config` for composite Qwen configs." }, { "left": "issue:44387", "right": "issue:44617", "accept": false, - "reason": "Both are memory-related but distinct: int4 quantization reserved memory regression vs Sam3Video CUDA OOM." + "reason": "Both mention OOM, but the underlying problems are different: int4 quantization reserved-memory growth vs a Sam3Video CUDA-memory failure." }, { "left": "issue:43854", "right": "issue:44977", "accept": false, - "reason": "Different failure modes and models: GLM-4.7-Flash unit test load issue vs Qwen3.5 flash-attention generation bug." + "reason": "Different model-specific failures: GLM-4.7-Flash loading in tests vs Qwen3.5 generation with flash-attention." }, { "left": "issue:43257", "right": "issue:43696", "accept": false, - "reason": "Different causes: Qwen3 MoE weight conversion with accelerate/deepspeed vs GPT-oss-20b CUDA OOM." + "reason": "Qwen3 MOE weight-conversion/loading under accelerate+deepspeed is unrelated to a generic GPT-oss-20b CUDA OOM." }, { "left": "issue:43257", "right": "issue:43931", "accept": false, - "reason": "Both involve Qwen models, but one is MoE weight conversion/loading and the other is a Qwen3-VL shape mismatch." + "reason": "Both involve model loading, but the symptoms and models differ: MOE weight conversion vs Qwen3-VL shape mismatch." }, { "left": "issue:43262", "right": "issue:44811", "accept": false, - "reason": "Audio apply_chat_template sampling-rate default bug is unrelated to Whisper batch_decode special-token handling." + "reason": "Different audio-processing bugs: chat-template sampling-rate defaulting vs Whisper processor decode parameters." }, { "left": "issue:46421", "right": "issue:46424", "accept": false, - "reason": "Related cache area, but one is lazy initialization causing recompilation and the other is max_cache_length config handling." + "reason": "Same broad cache area, but distinct defects: spurious recompilation from lazy cache init vs `max_cache_length` only applying to Quantized Cache." }, { "left": "issue:42915", "right": "issue:43742", "accept": false, - "reason": "Qwen3Moe FineGrainedFP8Config failure is unrelated to MobileLLM key error on load." + "reason": "Unrelated failures: Qwen3 MOE + FineGrainedFP8Config vs a key error loading MobileLLM-125M." }, { "left": "issue:43329", "right": "issue:44479", "accept": false, - "reason": "Video-token counting undefined variables vs a broader v5.3.0 video input regression across Qwen VL/35 models; not the same bug." + "reason": "Both are video/multimodal regressions, but 43329 is a specific undefined-variable/helper bug in `_get_num_multimodal_tokens` while 44479 is a broader v5.3.0 video-input regression across several models; not enough evidence they are the same fix." }, { "left": "issue:45459", "right": "issue:45803", "accept": false, - "reason": "Both involve error handling, but tokenizer protobuf decode masking and Fuyu bare except are different code paths and bug classes." + "reason": "Both involve swallowed errors, but the functions and failure modes differ: protobuf decode fallback masking tokenizer errors vs bare `except:` in `FuyuBatchFeature.convert_to_tensors()`." }, { "left": "issue:43329", "right": "issue:43450", "accept": false, - "reason": "Undefined multimodal token helper variables vs batched video processor shape bug; different layers of the VLM stack." + "reason": "Different video bugs: token counting in the multimodal helper vs incorrect batched video processor output shape." }, { "left": "issue:43749", "right": "issue:44756", "accept": false, - "reason": "Both mention loading/memory, but FSDP CPU RAM efficient loading and Strix Halo mmap OOM are distinct issues." + "reason": "Separate memory/loading problems: FSDP CPU RAM efficient loading is broken, while Strix Halo mmap is being disabled to avoid OOM." }, { "left": "issue:44291", "right": "issue:45042", "accept": false, - "reason": "init_empty_weights/_is_hf_initialized regression is unrelated to PIL image processors incorrectly requiring torchvision." - }, - { - "left": "issue:42915", - "right": "issue:44821", - "accept": false, - "reason": "Qwen3 MoE FP8 loading failure is unrelated to AutoImageProcessor URL loading." + "reason": "Unrelated issues: `init_empty_weights` passes an unexpected `_is_hf_initialized` argument, whereas PIL image processors incorrectly require torchvision." }, { "left": "issue:43606", "right": "issue:45859", "accept": false, - "reason": "CPU offload device mismatch for bark-small is a different problem from Qwen3_5Moe rotary embedding CPU-offload incompatibility." + "reason": "Both mention CPU offload, but the model/code path is different: suno/bark-small device mismatch vs Qwen3_5MoeTextRotaryEmbedding incompatibility." }, { "left": "issue:34567", "right": "issue:40990", "accept": false, - "reason": "TrainerState token counter bug is unrelated to GPT-oss-20b perplexity on WikiText-2." + "reason": "No relation between TrainerState token-count tracking and GPT-oss-20b perplexity on WikiText-2." }, { "left": "issue:43650", "right": "issue:43906", "accept": false, - "reason": "The generic 'ADD THE DATA' issue is not a concrete bug and is unrelated to the isolated reproduction report." + "reason": "`ADD THE DATA` is not a meaningful duplicate target, and the other item is an isolated reproduction of a separate issue." }, { "left": "issue:43540", "right": "issue:46701", "accept": false, - "reason": "Qwen3OmniMoe video processing ValueError and DiffusionGemma image-text-to-text pipeline failure are different model/pipeline bugs." + "reason": "Different multimodal failures: Qwen3OmniMoe video processing ValueError vs DiffusionGemma image-text-to-text pipeline malfunction." }, { "left": "issue:43927", "right": "issue:44912", "accept": false, - "reason": "DiaConfig custom token IDs not saved/loaded is unrelated to MXFP4 quantization fallback to bf16." + "reason": "Separate concerns: DiaConfig losing custom token IDs on save/load vs MXFP4 quantization falling back to bf16." }, { "left": "issue:43454", "right": "issue:43577", "accept": false, - "reason": "AyaVision lm_head tying bug and Blip2 dtype propagation bug are separate generation vs dtype-loading issues." + "reason": "Different model-loading bugs: AyaVision weight tying/generation corruption vs BLIP-2 dtype propagation remaining float32." }, { "left": "issue:43525", "right": "issue:46291", "accept": false, - "reason": "Llama4Config missing pad_token_id and importing pipelines with kernels v0.15 are unrelated configuration/import failures." + "reason": "No overlap: missing `pad_token_id` in Llama4Config vs pipeline import failing because kernels v0.15 requires version/revision." }, { "left": "issue:28282", "right": "issue:29870", "accept": false, - "reason": "PyTorch AutoModel import error is unrelated to Bart generate() decoder_attention_mask assignment bug." + "reason": "Unrelated: missing PyTorch for AutoModel import vs a Bart generation/evaluation `UnboundLocalError`." }, { "left": "issue:29159", "right": "issue:34567", "accept": false, - "reason": "Tokenizer slow/fast inconsistency does not match TrainerState num_input_tokens_seen not updating." + "reason": "Tokenizer slow/fast inconsistency is unrelated to TrainerState's `num_input_tokens_seen` not updating." }, { "left": "issue:44987", "right": "issue:45538", "accept": false, - "reason": "Physical-intelligence/fast loading failure and CLIPTokenizer model_max_length bug are different loading/tokenizer issues." + "reason": "Different bugs in different areas: loading `physical-intelligence/fast` under transformers>=5.1.0 vs CLIPTokenizer's `model_max_length` constant." } ] }, @@ -6622,227 +6643,218 @@ "issue:38175|issue:44898", "issue:44568|issue:45488" ], - "prepared_review_unit_hash": "7b30cac1445679002cb013c940ba6ce0d85b956298fcc0f2828155967bbdcc88", + "prepared_review_unit_hash": "971ffbef58123bbe74ac086d2b321e3411bbe628c6e5be54252cd9e267355bd8", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 29, - "serialized_chars": 35867, - "estimated_input_tokens": 8967, - "estimated_eval_tokens": 18190 + "serialized_chars": 35858, + "estimated_input_tokens": 8965, + "estimated_eval_tokens": 18186 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: tokenizer regressions, model-loading failures, video-input bugs, dtype/config issues, DeepSpeed problems, and packaging/import errors. None of the soft-edge pairs look like the same underlying bug or change.", - "confidence": 0.93, - "canonical_issue_reason": "Issue 45701 is the broadest umbrella-style tokenization regression report, so it is the closest thing to a cluster anchor, though the set is too mixed for a strong canonical duplicate.", + "summary": "Mostly a loose similarity cluster of unrelated Transformers regressions; only the Qwen video-input reports look like the same underlying bug.", + "confidence": 0.74, + "canonical_issue_reason": "issue:44479 is the broadest and most representative report in the one clearly matching subgroup: it covers the Qwen video-input regression across multiple variants, while 45381 narrows to qwen2.5-vl and its vision_position_ids symptom.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45701 is the most general and representative report in the set, with broader tokenization scope than the model-specific or environment-specific failures.", + "best_issue_reason": "issue:44479 is the strongest global representative: it describes the broadest concrete failure mode among the items and best captures the shared Qwen video regression that appears to recur in the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44568", "right": "issue:45356", "accept": false, - "reason": "Both are tokenizer regressions, but they affect different models and different failure modes (special-token insertion vs codec/warning handling)." + "reason": "Different tokenizer bugs: mdeberta special-token insertion vs Kimi-K2.5 codec/regex regression." }, { "left": "issue:28282", "right": "issue:34689", "accept": false, - "reason": "One is missing-PyTorch import detection; the other is a model-loading breakage for a specific vision model. Different issues." + "reason": "Generic PyTorch-missing import error vs a model-specific Llama 3.2 vision loading regression." }, { "left": "issue:43404", "right": "issue:45127", "accept": false, - "reason": "Both involve tied weights, but one is a missing lm_head tie in model init and the other is LoRA merge collapse after vocabulary extension. Not the same bug." + "reason": "Different model-path bugs: untied lm_head in Mistral3 vs LoRA merge collapse after vocab extension." }, { "left": "issue:43425", "right": "issue:46291", "accept": false, - "reason": "Torch version incompatibility is unrelated to the kernels import requirement/revision API change." + "reason": "Torch version incompatibility is unrelated to the kernels v0.15 revision/version import failure." }, { "left": "issue:42915", "right": "issue:45538", "accept": false, - "reason": "Qwen3Moe FP8 config failure and CLIPTokenizer max_length are unrelated code paths." + "reason": "Qwen3Moe FP8 config failure and CLIPTokenizer max-length bug are unrelated." }, { "left": "issue:43697", "right": "issue:45357", "accept": false, - "reason": "RTDetrV2 output regression and Qwen3.5 save_pretrained key corruption are different model-path bugs." + "reason": "RTDetrV2 output drift in v5 is unrelated to Qwen3.5 save_pretrained visual-key corruption." }, { "left": "issue:45137", "right": "issue:45405", "accept": false, - "reason": "DeepSpeed ZeRO3 deque failure is unrelated to the PEFT minimum-version packaging issue." + "reason": "DeepSpeed ZeRO3 deque error is unrelated to the PEFT minimum-version bump." }, { "left": "issue:33357", "right": "issue:34567", "accept": false, - "reason": "MacOS bus error with a CLIP model and TrainerState token counter not updating are unrelated." + "reason": "MacOS CLIP bus error and TrainerState token-count bookkeeping are different subsystems and failures." }, { "left": "issue:42175", "right": "issue:43824", "accept": false, - "reason": "Backend extra-install packaging and missing Qwen2_5_VL import are different problems." + "reason": "Packaging/backend install issue vs missing Qwen2_5_VL import symbol; not the same bug." }, { "left": "issue:43742", "right": "issue:44291", "accept": false, - "reason": "KeyError while loading a model and unexpected _is_hf_initialized during init_empty_weights are distinct loading failures." + "reason": "Different load-time problems: MobileLLM key error vs unexpected _is_hf_initialized arg in init_empty_weights." }, { "left": "issue:28282", "right": "issue:29127", "accept": false, - "reason": "Missing PyTorch import and LayoutLMv3 box-content validation are unrelated." + "reason": "Missing-PyTorch ImportError is unrelated to LayoutLMv3 box-validation messaging." }, { "left": "issue:44479", "right": "issue:46693", "accept": false, - "reason": "Both mention regressions, but one is a video-input functional break and the other is a long-context performance regression." + "reason": "Video-input regression is unrelated to the FA2 long-context performance regression." }, { "left": "issue:43317", "right": "issue:45538", "accept": false, - "reason": "Dequantized model offload failure and CLIPTokenizer max_length are unrelated." + "reason": "device_map/offload dequantization failure is unrelated to CLIPTokenizer max_length." }, { "left": "issue:44062", "right": "issue:44991", "accept": false, - "reason": "The tokenizer AddedToken TypeError may underlie a load failure, but the reports are not clearly the same concrete bug." + "reason": "Both are tokenizer-related, but one is AddedToken argument duplication and the other is a model-loading regression for est-roberta." }, { "left": "issue:43577", "right": "issue:46512", "accept": false, - "reason": "One is Blip2 dtype propagation in loaded models; the other is from_config leaving config.dtype stale. Related theme, different bug." + "reason": "Both mention dtype, but one is BLIP2 module casting and the other is stale config.dtype from from_config." }, { "left": "issue:42915", "right": "issue:43749", "accept": false, - "reason": "Qwen3Moe FP8 config and FSDP CPU RAM efficient loading are different training/loading paths." + "reason": "Qwen3Moe FP8 config failure and FSDP CPU-RAM-efficient-loading break are different code paths." }, { "left": "issue:43452", "right": "issue:43582", "accept": false, - "reason": "GGUF loading breakage and Apple Silicon allocator TypeError are unrelated." + "reason": "gguf loading failure and Apple Silicon caching_allocator_warmup TypeError are unrelated." }, { "left": "issue:44849", "right": "issue:45701", "accept": false, - "reason": "Qwen3.5 hidden-state output bug and a broad tokenization-regression report are different issues." + "reason": "Qwen3.5 hidden-states bug is unrelated to a general tokenization-version change report." }, { "left": "issue:41720", "right": "issue:43540", "accept": false, - "reason": "Auto device mapping CUDA assert and video-input processing error in Qwen3OmniMoe are unrelated." + "reason": "Auto device-mapping CUDA assert and Qwen3OmniMoe video ValueError are different failures." }, { "left": "issue:27957", "right": "issue:28282", "accept": false, - "reason": "Flash Attention 2 behavior for XLMRoberta and missing PyTorch import detection are not the same bug." + "reason": "FlashAttention2 support for XLMRoberta is unrelated to missing PyTorch installation." }, { "left": "issue:43540", "right": "issue:44560", "accept": false, - "reason": "Both are video-related, but they affect different models and symptoms (ValueError vs StopIteration)." + "reason": "Both are video-input failures, but different Qwen model families and error modes with no clear shared code path." }, { "left": "issue:44291", "right": "issue:44821", "accept": false, - "reason": "init_empty_weights argument handling and AutoImageProcessor URL loading are unrelated." + "reason": "init_empty_weights TypeError and AutoImageProcessor URL loading are unrelated load paths." }, { "left": "issue:24643", "right": "issue:29159", "accept": false, - "reason": "DeepSpeed weight-dimension RuntimeError and slow/fast tokenizer inconsistency are different problems." + "reason": "DeepSpeed weight-shape RuntimeError and tokenizer slow/fast inconsistency are unrelated." }, { "left": "issue:42175", "right": "issue:44488", "accept": false, - "reason": "Packaging backend selection and a specific model-loading failure are not the same underlying issue." + "reason": "TensorFlow not included in a torch install extra is unrelated to loading a specific model." }, { - "left": "issue:43452", - "right": "issue:43582", + "left": "issue:43540", + "right": "issue:44479", "accept": false, - "reason": "GGUF loading and caching_allocator_warmup on Apple Silicon are unrelated." + "reason": "Both are video-related, but different Qwen model stacks and distinct failures; not enough to call them the same bug." }, { - "left": "issue:44849", - "right": "issue:45701", + "left": "issue:44987", + "right": "issue:45907", "accept": false, - "reason": "Different concrete bugs: hidden-states handling versus broad version-dependent tokenization changes." + "reason": "Loading a specific model in transformers>=5.1.0 is unrelated to list_repo_files needing revision forwarding." + }, + { + "left": "issue:44479", + "right": "issue:45381", + "accept": true, + "reason": "Both point to the same Qwen video-input regression; 45381\u2019s wrong vision_position_ids symptom fits the broader 44479 failure mode." }, { "left": "issue:38175", "right": "issue:44898", "accept": false, - "reason": "Unexpected zero probabilities in SigLIP2 and Perceiver non-default-resolution failures are different model behaviors." + "reason": "Siglip2 zero probabilities and Perceiver non-default-resolution failure are different vision-model bugs." }, { "left": "issue:44568", "right": "issue:45488", "accept": false, - "reason": "Both tokenizer-related, but one is special-token insertion and the other is a LlamaTokenizer pre-tokenizer override regression." - }, - { - "left": "issue:41720", - "right": "issue:43540", - "accept": false, - "reason": "Qwen3 auto-device-map CUDA assert and Qwen3OmniMoe video processing failure are different code paths." - }, - { - "left": "issue:44568", - "right": "issue:45356", - "accept": false, - "reason": "Reiterated: separate tokenizer regressions with different affected models and root causes." - }, - { - "left": "issue:42175", - "right": "issue:43824", - "accept": false, - "reason": "Different packaging/import failures; no shared concrete code path." + "reason": "mdeberta special-token insertion is unrelated to the LlamaTokenizer ByteLevel/Metaspace override regression." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The only accepted soft pair, issue:44479 vs issue:45381, is plausibly the same Qwen video-input regression and looks mergeable as a single bugfix. The other verdicts are appropriately cautious and do not overstate relatedness." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -6936,196 +6948,196 @@ "issue:44913|issue:46121", "issue:28282|issue:32090" ], - "prepared_review_unit_hash": "38e3db4c0b3c525de8b292d10d52756083f253fd7aef2fa2bc7903918070f9b0", + "prepared_review_unit_hash": "c13e9d1690635973771fce1eb4d245e7d07c1c278ce88118aa5c5a70635f4676", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 27, - "serialized_chars": 35588, - "estimated_input_tokens": 8897, - "estimated_eval_tokens": 18050 + "serialized_chars": 35593, + "estimated_input_tokens": 8899, + "estimated_eval_tokens": 18054 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These items are mostly unrelated model/tokenizer/config bugs, so there is no strong duplicate cluster here. If a single issue must represent the group, the broad v5 tokenizer regression is the closest fit.", - "confidence": 0.78, - "canonical_issue_reason": "issue:45920 is the broadest and most central report in the set: it describes a multi-model tokenizer regression affecting several families and has active discussion, making it the best stand-in for this otherwise loosely related cluster.", + "summary": "The cluster is heterogeneous: most pairs share only superficial subsystem overlap (tokenizers, multimodal models, configs, dtype, loading) but describe different concrete failures and fixes. I found no true duplicate pair among the soft edges.", + "confidence": 0.96, + "canonical_issue_reason": "issue:45920 is the strongest representative issue: it is a broad, high-impact v5 regression affecting multiple model families and clearly describes a concrete user-facing tokenizer failure.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45920 is the most representative issue because it is broad, regression-oriented, and impacts multiple downstream models rather than a single narrow code path.", + "best_issue_reason": "issue:45920 is the best global issue candidate because it has the broadest scope and clearest regression framing, making it the most representative triage anchor for this set.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43257", "right": "issue:43756", "accept": false, - "reason": "Different bugs: one is Qwen3 MoE weight conversion/loading with accelerate+deepspeed; the other is Smollm3 RoPE-layer counting. No shared code path." + "reason": "Both mention model-specific loading/regression problems, but one is about Qwen3 MoE weight conversion under accelerate+deepspeed and the other about Smollm3 RoPE layer dropping; different components and fixes." }, { "left": "issue:34567", "right": "issue:36331", "accept": false, - "reason": "Different Trainer APIs: one is token-count tracking not updating, the other is a custom compute_loss signature change. Related to Trainer but not the same defect." + "reason": "Both involve Trainer behavior, but num_input_tokens_seen not updating is a state/metrics bug while compute_loss getting num_items_in_batch is an API signature incompatibility." }, { "left": "issue:44792", "right": "issue:44964", "accept": false, - "reason": "Both are model-loading failures, but one is a janus generate test failure and the other is a Phi-4 multimodal load regression. Too different to merge." + "reason": "Both are multimodal model failures, but one is a failed janus image generation test and the other is a Phi-4-multimodal model loading error; not the same code path." }, { "left": "issue:44062", "right": "issue:45081", "accept": false, - "reason": "Tokenizer-related crashes, but the underlying failures differ: AddedToken duplicate kwargs versus Mistral regex patch expecting backend_tokenizer." + "reason": "Both are tokenizer-related crashes, but one is AddedToken duplication with special tokens and the other is a Mistral regex patch/backend_tokenizer attribute crash; distinct bugs." }, { "left": "issue:43742", "right": "issue:44589", "accept": false, - "reason": "Both are loading/runtime errors, but one is a MobileLLM key error and the other is a float8 storage lookup issue. No concrete overlap." + "reason": "These are separate loading failures: MobileLLM key lookup vs float8 storage deserialization; no shared root cause is evident." }, { "left": "issue:44737", "right": "issue:46463", "accept": false, - "reason": "Both mention validation/decoder behavior, but one is XLNet rotary position CPU placement and the other is T5 decoder-input error messaging. Not the same bug." + "reason": "XLNet CPU allocation in rotary positional encoding and T5 decoder-input validation error messaging are unrelated." }, { "left": "issue:45042", "right": "issue:45081", "accept": false, - "reason": "One is PIL backend image processors incorrectly requiring torchvision; the other is a Mistral tokenizer regex crash. Different subsystems and fixes." + "reason": "One is an image-processor dependency issue requiring torchvision, the other is a tokenizer regex patch crash; different subsystems and fixes." }, { "left": "issue:43716", "right": "issue:46701", "accept": false, - "reason": "Both involve multimodal/image-text models, but one is a Mistral-3 dtype mismatch and the other is a DiffusionGemmaGenerationOutput pipeline failure. Not the same code-path problem." + "reason": "Both touch multimodal generation, but one is a dtype mismatch in Mistral-3 image preprocessing and the other is a DiffusionGemma image-text-to-text pipeline failure." }, { "left": "issue:43756", "right": "issue:46399", "accept": false, - "reason": "Both concern Qwen3-related internals, but one is RoPE layer selection and the other is missing registration for apply_rotary_pos_emb when use_kernels=True. Different defects." + "reason": "Both reference RoPE/kernel behavior, but Smollm3 drops RoPE layers while Qwen3.5 fails to register apply_rotary_pos_emb under use_kernels=True; not the same defect." }, { "left": "issue:43531", "right": "issue:44821", "accept": false, - "reason": "Both are generate/loading issues, but one is Qwen3-MoE sliding_window behavior and the other is AutoImageProcessor URL loading. No duplicate evidence." + "reason": "Qwen3-MoE sliding-window behavior and AutoImageProcessor URL loading are unrelated." }, { "left": "issue:43756", "right": "issue:45325", "accept": false, - "reason": "Both touch Qwen3/Qwen3.5-ish rotary behavior, but one is RoPE layer dropping and the other is Qwen2.5-VL rope index scaling for images. Different models and logic." + "reason": "Both involve Qwen3.5/Qwen2.5-VL and positional handling, but one is about RoPE layer dropping and the other about temporal position_ids scaling for still images." }, { "left": "issue:44568", "right": "issue:46491", "accept": false, - "reason": "Both are tokenizer regressions, but one is missing BOS/EOS insertion in mdeberta-v3-base and the other is CodeLlama leading-space round-trip loss. Distinct symptoms and fixes." + "reason": "Tokenizer regressions in different models: mdeberta special-token addition vs CodeLlama leading-space round-trip loss; separate tokenization logic." }, { "left": "issue:43577", "right": "issue:45072", "accept": false, - "reason": "Both are dtype-related, but one is BLIP2 model/qformer dtype staying float32 and the other is CI inference mismatches in SwitchTransformers/TimmWrapperModel. Not the same underlying bug." + "reason": "Both are dtype-mismatch reports, but they affect different architectures and code paths (BLIP2 qformer vs SwitchTransformers/TimmWrapperModel)." }, { "left": "issue:45405", "right": "issue:45600", "accept": false, - "reason": "Both are repository maintenance/config issues, but one is a PEFT version bump and the other is stale auto_mappings references breaking CI. Different concrete problems." + "reason": "One is an external dependency version pin mismatch, the other is stale auto_mappings entries causing CI breakage; different maintenance issues." }, { "left": "issue:43906", "right": "issue:46468", "accept": false, - "reason": "One is an isolated reproduction of an earlier issue; the other is a broken official Nemotron example. No evidence they share the same bug." + "reason": "An isolated reproduction of one prior issue and an official Nemotron example breakage are not the same bug without stronger evidence." }, { "left": "issue:45823", "right": "issue:46463", "accept": false, - "reason": "Both are model usage failures, but one is Gemma4 PLE device mismatch on forward and the other is a T5 decoder-input validation message bug. Different path and symptom." + "reason": "Gemma4 device-map mismatch during forward and T5 decoder input validation messaging are unrelated." }, { "left": "issue:24643", "right": "issue:29870", "accept": false, - "reason": "Both are generation/training errors, but one is a DeepSpeed weight-dimension RuntimeError and the other is a Bart generate() UnboundLocalError. Not duplicates." + "reason": "DeepSpeed training weight-shape runtime error and Bart generate decoder_attention_mask unbound variable are different failures in different pipelines." }, { "left": "issue:43404", "right": "issue:45987", "accept": false, - "reason": "Both involve model state/config tracking, but one is Mistral3 lm_head weight tying and the other is TrainerState token counting. Different layers of the stack." + "reason": "Mistral3 lm_head weight tying and StaticCache sequence-length return type are unrelated model/utility bugs." }, { "left": "issue:44737", "right": "issue:45588", "accept": false, - "reason": "Both involve runtime errors in model internals, but one is XLNet CPU arange placement and the other is flash_attention sink-less s_aux handling. No same code path." + "reason": "XLNet positional encoding CPU placement and flash_attention s_aux=None crash are different code paths." }, { "left": "issue:43452", "right": "issue:44488", "accept": false, - "reason": "Both mention loading tokenizers/models, but one is gguf_file support and the other is loading cjvt/sleng-bert. Different failure modes and likely different fixes." + "reason": "Both involve model/tokenizer loading, but gguf_file regression in from_pretrained is not the same as a specific model no longer loading." }, { "left": "issue:43906", "right": "issue:46693", "accept": false, - "reason": "One is a specific isolated reproduction issue; the other is a performance regression with FA2 long-context generation. Different class of problem." + "reason": "An issue reproduction and a performance regression in long-context FA2 generation are not the same underlying bug." }, { "left": "issue:44625", "right": "issue:44877", "accept": false, - "reason": "Both are config propagation/loading bugs, but one is num_labels not copied into Qwen3.5 text config and the other is strict config blocking granite_speech. Not the same defect." + "reason": "Both are config-related, but one is num_labels propagation in Qwen3.5 and the other is strict config blocking granite_speech loading." }, { "left": "issue:43526", "right": "issue:45479", "accept": false, - "reason": "Both concern label handling, but one is BeitImageProcessorFast reduce_labels returning only one label and the other is zero-loss classification when num_labels=1. Different behavior and layers." + "reason": "BeitImageProcessorFast reduce_labels behavior and sequence-classification zero-loss with num_labels=1 are unrelated label-handling bugs." }, { "left": "issue:45357", "right": "issue:45920", "accept": false, - "reason": "Both are Qwen3.5-related regressions, but one is incorrect visual encoder keys in save_pretrained and the other is wrong token IDs from AutoTokenizer across many models. Not mergeable." + "reason": "One is incorrect visual encoder keys saved for Qwen3.5, the other is wrong token IDs from AutoTokenizer across several models; different failure modes." }, { "left": "issue:43540", "right": "issue:46693", "accept": false, - "reason": "Both mention generation, but one is a video-input ValueError in Qwen3OmniMoe and the other is a performance regression with FA2 long-context generation. Different issues." + "reason": "Qwen3OmniMoe video input processing and FA2 long-context performance regression do not share a concrete code-path failure." }, { "left": "issue:44913", "right": "issue:46121", "accept": false, - "reason": "Both touch config handling, but one is GPTNeoX rotary_pct not persisting and the other is JSON-loaded ignore_keys_at_rope_validation causing TypeError. Different specific bugs." + "reason": "GPTNeoX rotary_pct persistence and JSON-loaded ignore_keys_at_rope_validation type handling are distinct configuration bugs." }, { "left": "issue:28282", "right": "issue:32090", "accept": false, - "reason": "Both are import/runtime environment errors around model use, but one is missing PyTorch import support and the other is Trainer passing NoneType into _gpu_broadcast_one. Not the same underlying bug." + "reason": "PyTorch missing ImportError for AutoModel and Trainer _gpu_broadcast_one NoneType type error are completely different problems." } ] }, @@ -7229,14 +7241,14 @@ "issue:43756|issue:46402", "issue:43688|issue:45704" ], - "prepared_review_unit_hash": "c44d2461a72034c009c8e9ba5f4eebb2136d1f52549dbb54eeb9c8ec7df079d0", + "prepared_review_unit_hash": "22a96c90be7313ede2b0f825a040f4c5886a85f80a7106199ddf636a678c4bc1", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, @@ -7251,193 +7263,196 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is broadly heterogeneous: mostly distinct Transformers regressions spanning tokenizers, model loading, generation, and backend handling. No soft pair looks like a clear duplicate on the evidence given, so I would not merge any of them into the same issue cluster.", - "confidence": 0.34, - "canonical_issue_reason": "issue:43646 is the broadest and most representative v5 regression report in the set, but overall the collection is too diverse for a strong single canonical anchor.", + "summary": "Mostly heterogeneous transformer issues with two tokenizer-related themes. I\u2019m keeping issue:42898 as the broad tokenizer-regression representative, but I\u2019m rejecting the 42898\u219446491 duplicate claim because the evidence is too title-level and the concrete failures differ.", + "confidence": 0.71, + "canonical_issue_reason": "Issue:42898 is the broadest tokenizer-regression report in the set and the cleanest cluster representative for the v5 tokenization-behavior theme.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43646 is the most umbrella-like issue title and best fits a cluster-level representative, though it is still not a true duplicate magnet for the rest of the items.", + "best_issue_reason": "Issue:42898 is the best general representative because it describes the upstream tokenizer cleanup-behavior change directly, even though the cluster is otherwise mostly unrelated.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43720", "right": "issue:44898", "accept": false, - "reason": "Different models and failures: BitNet packed-weight loading vs Perceiver interpolation at non-default resolution." + "reason": "Different subsystems and failure modes: BitNet packed-weight loading vs Perceiver resolution handling." }, { "left": "issue:43697", "right": "issue:45216", "accept": false, - "reason": "Different bugs: RTDetrV2 output divergence vs Qwen3.5 save_pretrained checkpoint corruption." + "reason": "RTDetrV2 output drift in v5 and a Qwen3.5 save_pretrained regression are different bugs." }, { "left": "issue:43643", "right": "issue:45375", "accept": false, - "reason": "Both involve config fields, but one is missing fields from trust_remote_code AutoConfig and the other is a model-specific strict-drop field bug." + "reason": "Both mention missing fields, but the specific loading paths and root causes differ." }, { "left": "issue:44466", "right": "issue:45005", - "accept": false, - "reason": "Both mention tied weights, but the titles point to different symptoms and scopes; not enough to treat as the same concrete bug." + "accept": true, + "reason": "Both describe the same tied-weights/save_pretrained serialization regression in v5." }, { "left": "issue:45200", "right": "issue:46463", "accept": false, - "reason": "Unrelated model-specific issues: Gemma 4 token type defaults vs T5 decoder input validation messaging." + "reason": "Gemma4 mm_token_type_ids handling is unrelated to T5 decoder-input validation messaging." }, { "left": "issue:41762", "right": "issue:44936", "accept": false, - "reason": "Gemma3 DeepSpeed loading crash and trainer.evaluate after train are unrelated code paths." + "reason": "DeepSpeed ZeRO-3 loading IndexError is a different path from trainer.evaluate() failing after train()." }, { "left": "issue:44361", "right": "issue:44442", "accept": false, - "reason": "Different tokenizers and different failures; both are loading issues but not the same bug." + "reason": "Different tokenizer problems: MLukeTokenizer task AttributeError vs AutoTokenizer loading failure." }, { "left": "issue:44493", "right": "issue:46050", "accept": false, - "reason": "Position-id key warnings across models vs sinq quantized save/load breakage are not the same defect." + "reason": "Unexpected position-id keys and sinq quantization save/load breakage are unrelated issues." }, { "left": "issue:43688", "right": "issue:43749", "accept": false, - "reason": "Auxiliary loss normalization and FSDP RAM-efficient loading are unrelated." + "reason": "Aux-loss normalization in MoE models is unrelated to FSDP CPU RAM efficient loading." }, { "left": "issue:43653", "right": "issue:44361", "accept": false, - "reason": "Both are tokenizer-related, but one is special-token registration and the other is an AttributeError on tasks." + "reason": "BigBirdTokenizer special-token registration and MLukeTokenizer task AttributeError are different bugs." }, { "left": "issue:43525", "right": "issue:46701", "accept": false, - "reason": "Llama4 config padding-field error and DiffusionGemma pipeline failure are unrelated." + "reason": "Llama4Config pad_token_id and DiffusionGemma image-text pipeline failure are unrelated." }, { "left": "issue:43819", "right": "issue:44811", "accept": false, - "reason": "DAC latent reconstruction mismatch and Whisper batch_decode special-token handling are different code paths." + "reason": "DAC latent reconstruction mismatch and Whisper batch_decode skip_special_tokens handling are different issues." }, { "left": "issue:43994", "right": "issue:44898", "accept": false, - "reason": "Different vision-model regressions with different triggers and symptoms." + "reason": "SigLIP2 nonsensical outputs are unrelated to Perceiver non-default-resolution failure." }, { "left": "issue:43452", "right": "issue:44843", "accept": false, - "reason": "AutoTokenizer/AutoModel gguf_file breakage and mistral_regex offline model_info calls are separate loader regressions." + "reason": "gguf_file loading breakage is a different code path from the offline-mode model_info() call." }, { "left": "issue:38175", "right": "issue:43295", "accept": false, - "reason": "SigLIP2 zero-probability outputs and processor.tokenizer regression are unrelated." + "reason": "SigLIP2 zero probabilities and processor.tokenizer regression are different model/pipeline bugs." }, { "left": "issue:44936", "right": "issue:45823", "accept": false, - "reason": "Trainer lifecycle bug and Gemma4 device mismatch during forward are different issues." + "reason": "Trainer lifecycle failure is unrelated to Gemma4 device-mismatch during forward." }, { "left": "issue:45372", "right": "issue:46170", "accept": false, - "reason": "Gemma 4 processor import dependency failure and Windows ROCm torch.distributed failure are unrelated." + "reason": "Gemma4 processor import dependency breakage is unrelated to the Windows ROCm torch.distributed issue." }, { "left": "issue:43531", "right": "issue:43742", "accept": false, - "reason": "Qwen3-MoE sliding_window behavior and MobileLLM loading key error do not share the same underlying bug." + "reason": "Qwen3-MoE sliding_window behavior and MobileLLM loading KeyError are distinct problems." }, { "left": "issue:45542", "right": "issue:46178", "accept": false, - "reason": "Both concern backend dependencies, but they are distinct missing-backend/reporting problems." + "reason": "TensorBoard-without-TF backend setup and tensorflow_text backend mapping are separate dependency/backend issues." }, { "left": "issue:42898", "right": "issue:46491", "accept": false, - "reason": "Generic v5 tokenization-space behavior change vs CodeLlama leading-space round-trip regression are not the same concrete bug." + "reason": "The titles point to a related tokenizer-regression theme, but not enough to treat them as the same concrete bug from title-level evidence alone: cleanup-behavior change vs. CodeLlama-specific leading-space stripping." }, { "left": "issue:43646", "right": "issue:43749", "accept": false, - "reason": "Custom model initialization regression and FSDP CPU RAM-efficient loading are unrelated." + "reason": "Custom model initialization breakage is unrelated to FSDP CPU RAM efficient loading." }, { "left": "issue:43653", "right": "issue:44568", "accept": false, - "reason": "Different tokenizer special-token problems with different models and APIs." + "reason": "Both involve tokenizer special tokens, but one is mask-token registration and the other is BOS/EOS insertion; not the same bug." }, { "left": "issue:43577", - "right": "issue:43697", + "right": "issue:46402", "accept": false, - "reason": "Blip2 dtype propagation and RTDetrV2 output divergence are unrelated." + "reason": "Blip2 dtype propagation and Sam3Model text-encoder weight loading are different loading bugs." }, { "left": "issue:43452", "right": "issue:43577", "accept": false, - "reason": "Model loading via gguf_file and Blip2 dtype handling are different bugs." + "reason": "gguf_file loading failures are unrelated to Blip2 dtype remaining float32." }, { "left": "issue:43295", "right": "issue:45356", "accept": false, - "reason": "Both are regressions around tokenizers/processors, but the affected models and failure modes differ." + "reason": "processor.tokenizer/images regression is unrelated to the Kimi-K2.5 codec-handling issue." }, { "left": "issue:44464", "right": "issue:44743", "accept": false, - "reason": "Chunked generation with compiled forward and Qwen3.5 recurrent-state reset are different generation bugs." + "reason": "Chunked generation with compiled forward and Qwen3 recurrent-state reset are different generation bugs." }, { "left": "issue:29159", "right": "issue:33357", "accept": false, - "reason": "Tokenizer inconsistency and MacOS bus error on CLIP are unrelated." + "reason": "Tokenizer slow/fast inconsistency is unrelated to the MacOS bus error on a CLIP community model." }, { "left": "issue:43756", "right": "issue:46402", "accept": false, - "reason": "Smollm3 RoPE-layer issue and Sam3 text-encoder weight loading are different model-specific defects." + "reason": "Smollm3 RoPE-layer dropping and Sam3Model text-encoder loading are unrelated model-definition bugs." }, { "left": "issue:43688", "right": "issue:45704", "accept": false, - "reason": "Auxiliary-loss normalization and T5 apex FusedRMSNorm memory leak are unrelated." + "reason": "MoE auxiliary-loss normalization is unrelated to T5's apex.FusedRMSNorm memory leak." } ] }, - "evaluator_result": null, - "evaluator_used": false, - "retried": false, - "accepted_nontrivial_soft_edge": false, + "evaluator_result": { + "accept": false, + "feedback": "The cluster summary is broadly grounded, but one soft-edge verdict is too assertive for the packet: `issue:42898` vs `issue:46491`. The titles support a related tokenizer regression theme, but not enough to confidently call them the same bug from title-level evidence alone (`clean_up_tokenization_spaces` behavior change vs. a CodeLlama-specific encode\u2192decode leading-space strip). That makes the duplicate claim less conservative than required. The other judgments look more defensible." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -7537,20 +7552,20 @@ "issue:43493|issue:43825", "issue:43525|issue:43572" ], - "prepared_review_unit_hash": "3ae5f08188cb761aa9eb5d31245cbd9c552916c4d9834d9895d92339fed29867", + "prepared_review_unit_hash": "19caa81b5de2a0e1e1bb07aa6fd6925db72e662ca288d2ce267897b911be8e6e", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 32, - "serialized_chars": 37418, + "serialized_chars": 37417, "estimated_input_tokens": 9355, "estimated_eval_tokens": 18966 }, @@ -7559,204 +7574,204 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These items are mostly unrelated Transformers issues across different model families and failure modes. No soft pair looks like the same underlying bug or change, so all candidate links are rejected.", - "confidence": 0.19, - "canonical_issue_reason": "If one issue must stand in for the cluster, `issue:44561` is the broadest umbrella: a v5.0 regression breaking `trust_remote_code` models. The rest are mostly model-specific loading, generation, dtype, or docs problems rather than one shared root cause.", + "summary": "This cluster is heterogeneous: the soft pairs span unrelated bugs across tokenizers, configs, generation, multimodal inputs, loading, and training. I would reject every soft edge; none look like the same underlying fix. If one issue has to serve as a loose representative, 45701 is the broadest and most referenced item.", + "confidence": 0.89, + "canonical_issue_reason": "No clear canonical duplicate target; the set mixes unrelated issues, so there is no single issue that obviously anchors the cluster.", "canonical_pr_reason": null, - "best_issue_reason": "`issue:44561` is the best representative because it describes a general version-wide compatibility break affecting many remote-code models, which is closer to a cluster-level theme than the other highly specific issues.", + "best_issue_reason": "issue:45701 is the broadest and most referenced report in the set, so it is the best loose representative if one must be chosen.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43653", "right": "issue:43927", "accept": false, - "reason": "Both mention token IDs/tokens, but one is a tokenizer special-token registration bug and the other is a config persistence bug; different components and root causes." + "reason": "Different bugs: BigBird special-token registration vs DiaConfig losing custom token IDs after save/load." }, { "left": "issue:43688", "right": "issue:45701", "accept": false, - "reason": "Aux-loss normalization and tokenization changes are unrelated subsystems and failure modes." + "reason": "Aux-loss normalization bug is unrelated to tokenization changes." }, { "left": "issue:44843", "right": "issue:44991", "accept": false, - "reason": "Both are tokenizer-loading problems, but one is an offline/model_info patching bug and the other is a specific model tokenizer load failure." + "reason": "Both affect tokenizer loading, but one is offline model_info patching and the other is an est-roberta tokenizer regression." }, { "left": "issue:44479", "right": "issue:44625", "accept": false, - "reason": "Video-input regression for Qwen VL models is not the same as missing `num_labels` propagation in Qwen3.5 config." + "reason": "Video-input regression across Qwen VL variants vs Qwen3.5 num_labels propagation; different code paths." }, { "left": "issue:45491", "right": "issue:46002", "accept": false, - "reason": "Both involve NaNs, but one is a sliding-window attention embedding bug and the other is a classification-head init/checkout issue." + "reason": "Gemma3 sliding-window NaNs are unrelated to classification-head NaNs after checkout/init." }, { "left": "issue:43550", "right": "issue:44361", "accept": false, - "reason": "Torch.compile/SDPA behavior for Bamba is unrelated to an MLukeTokenizer AttributeError." + "reason": "Bamba torch.compile/SDPA failure and MLukeTokenizer AttributeError are distinct problems." }, { "left": "issue:44484", "right": "issue:44704", "accept": false, - "reason": "Default shard size and AutoProcessor kwarg forwarding are different save/load API issues." + "reason": "max_shard_size behavior is unrelated to AutoProcessor kwargs forwarding." }, { "left": "issue:33666", "right": "issue:36010", "accept": false, - "reason": "Multi-GPU Qwen2-VL training and a GenerationMixin import error are unrelated." + "reason": "Multi-GPU Qwen2-VL training and GenerationMixin import failure are unrelated." }, { "left": "issue:43688", "right": "issue:45440", "accept": false, - "reason": "Both are model-quality regressions, but they affect different architectures and different code paths; not the same bug." + "reason": "Aux-loss normalization and DeepseekV3MoE implementation divergence are different issues." }, { "left": "issue:43742", "right": "issue:44821", "accept": false, - "reason": "Loading a text model and loading an image processor from URL are different failures." + "reason": "MobileLLM load key error is unrelated to AutoImageProcessor URL loading." }, { "left": "issue:44734", "right": "issue:46612", "accept": false, - "reason": "Both touch generation cache logic, but one is server KV-cache continuation indexing and the other is beam-search cache reorder omission; not the same concrete bug." + "reason": "Both involve generation/cache, but one is KV-cache continuation indexing and the other is beam-search cache reorder skipping for specific models." }, { "left": "issue:32090", "right": "issue:33666", "accept": false, - "reason": "A generic Trainer broadcast TypeError is not the same issue as Qwen2-VL multi-GPU training." + "reason": "Trainer GPU broadcast NoneType error is unrelated to Qwen2-VL multi-GPU training." }, { "left": "issue:44466", "right": "issue:46050", "accept": false, - "reason": "Inconsistent lm_head serialization and broken sinq quantized save/load are both serialization-related, but they are distinct mechanisms." + "reason": "lm_head serialization/tied-weights behavior is unrelated to sinq quantized save/load breakage." }, { "left": "issue:41720", "right": "issue:44464", "accept": false, - "reason": "Auto device mapping causing cudaErrorAssert is unrelated to chunked generation mismatches under compiled forward." + "reason": "Qwen3 auto device-map cuda assert and compiled chunked-generation inconsistency are different generation bugs." }, { "left": "issue:44464", "right": "issue:44977", "accept": false, - "reason": "Both affect generation correctness, but one is compiled-forward chunking and the other is flash-attention behavior in Qwen3.5; different bugs." + "reason": "Chunked generation with compiled forward is not the same as Qwen3.5 flash-attention generation failure." }, { "left": "issue:43504", "right": "issue:43550", "accept": false, - "reason": "A legacy-field load issue in BEiT segmentation is unrelated to Bamba compile/SDPA behavior." + "reason": "Legacy-field loading for Beit segmentation and Bamba SDPA torch.compile failure are unrelated." }, { "left": "issue:43504", "right": "issue:43720", "accept": false, - "reason": "Legacy-field model loading and BitNet packed-weight unpacking are different load-path problems." + "reason": "Legacy-field model loading and BitNet packed-weight unpacking are different loading issues." }, { "left": "issue:44977", "right": "issue:46693", "accept": false, - "reason": "One is incorrect generation output with flash-attention; the other is a performance regression with FA2 and long context." + "reason": "Generation correctness with flash-attn is not the same as a performance regression with FA2 long-context generation." }, { "left": "issue:43504", "right": "issue:44610", "accept": false, - "reason": "BEiT pretrained loading and OmDet-Turbo processor size mismatch are unrelated." + "reason": "Beit pretrained preset loading and OmDet processor/model size mismatch are unrelated." }, { "left": "issue:43577", "right": "issue:44554", "accept": false, - "reason": "A dtype propagation bug in BLIP2 is not the same as an MPS attention correctness issue." + "reason": "Blip2 dtype loading drift and MPS attention correctness are separate bugs." }, { "left": "issue:43720", "right": "issue:43927", "accept": false, - "reason": "BitNet accelerate-loading behavior and DiaConfig token ID persistence are unrelated." + "reason": "BitNet packed-weight loading and DiaConfig token-id persistence have no shared code path." }, { "left": "issue:33666", "right": "issue:36331", "accept": false, - "reason": "A multi-GPU training report and a `compute_loss` signature change are different trainer issues." + "reason": "Multi-GPU training and CustomTrainer compute_loss signature mismatch are distinct trainer issues." }, { "left": "issue:43643", "right": "issue:44561", "accept": false, - "reason": "Both involve trust_remote_code regressions, but one is missing config fields from `AutoConfig` and the other is an `is_torch_fx_available` removal breakage; not the same underlying bug." + "reason": "Both touch trust_remote_code, but one is missing config fields and the other is removal of is_torch_fx_available." }, { "left": "issue:43278", "right": "issue:43688", "accept": false, - "reason": "Embedding dtype drift and auxiliary-loss normalization are unrelated." + "reason": "Embedding dtype drift during evaluate is unrelated to auxiliary-loss normalization." }, { "left": "issue:43606", "right": "issue:43927", "accept": false, - "reason": "CPU offload device mismatch is unrelated to lost config token IDs." + "reason": "CPU offload device mismatch and custom token-ID loss after save/load are unrelated." }, { "left": "issue:43782", "right": "issue:46693", "accept": false, - "reason": "Qwen3VL weight_only loading failure and FA2 long-context performance regression are different problems." + "reason": "Qwen3VL weight_only loading error is unrelated to FA2 long-context performance regression." }, { "left": "issue:43262", "right": "issue:43295", "accept": false, - "reason": "Audio chat-template sampling-rate defaults and processor/tokenizer API regression are distinct processor issues." + "reason": "Audio chat-template sample-rate default and processor.tokenizer regression are different processor bugs." }, { "left": "issue:39692", "right": "issue:43493", "accept": false, - "reason": "Documentation example errors are not the same as SigLIP2 implementation divergence." + "reason": "SigLIP2 docs/example errors and HF-vs-JAX implementation divergence are related to the same model family but not the same bug." }, { "left": "issue:44964", "right": "issue:45405", "accept": false, - "reason": "A model-loading regression and an unreleased PEFT version bump are unrelated." + "reason": "Phi-4 multimodal load failure and PEFT version pinning are unrelated." }, { "left": "issue:43720", "right": "issue:44361", "accept": false, - "reason": "BitNet loading and MLukeTokenizer attribute errors do not share a code path." + "reason": "BitNet weight loading and MLukeTokenizer task errors are unrelated." }, { "left": "issue:43493", "right": "issue:43825", "accept": false, - "reason": "SigLIP2 implementation mismatch and a pipeline error-message mismatch are unrelated." + "reason": "SigLIP2 implementation mismatch and pipeline translation error messaging are unrelated." }, { "left": "issue:43525", "right": "issue:43572", "accept": false, - "reason": "Both are missing pad-token config fields, but they affect different models/config keys and are not the same bug." + "reason": "Missing pad_token_id and missing pad_token_idx are similar-shaped but distinct model/config regressions." } ] }, @@ -7860,208 +7875,208 @@ "issue:43504|issue:43606", "issue:43334|issue:43572" ], - "prepared_review_unit_hash": "bfc114bd0b4c46b1c8b71795882105ac59a8e1e48cf937a284cb7a88c749407f", + "prepared_review_unit_hash": "4f1614f9717c6b0e4db502c6d59193883e2b5d3f5638cdf80ae3f2a6069528cf", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36451, - "estimated_input_tokens": 9113, - "estimated_eval_tokens": 18482 + "serialized_chars": 36441, + "estimated_input_tokens": 9111, + "estimated_eval_tokens": 18478 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is very heterogeneous: most items are unrelated tokenizer, model-loading, training, and config regressions. No soft pair clearly shares the same underlying bug/change, so I rejected them all. If forced to pick a weak canonical issue, 45812 is the broadest tokenizer-regression anchor.", - "confidence": 0.36, - "canonical_issue_reason": "issue:45812 is the broadest and most representative tokenization regression here: it is a silent v4\u2192v5 AutoTokenizer bug, affects all Granite models, and has the strongest cross-cutting impact/inbound references among the tokenizer-related items.", + "summary": "This cluster is a loose collection of mostly unrelated Transformers bug reports, with many tokenizer/config/model-loading regressions but no pair that clearly represents the same underlying issue. The closest matches still differ in trigger, affected model, or code path, so I would not merge any soft edge.", + "confidence": 0.88, + "canonical_issue_reason": "issue:45812 is the broadest, most representative report in the set: a silent tokenizer regression affecting an entire model family with clear v4\u2192v5 impact and multiple references.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45812 is the best global representative because it captures a general tokenizer regression rather than a model-specific one, and it is the most broadly impactful issue in the set.", + "best_issue_reason": "issue:45812 best fits the cluster because it is concrete, widely scoped, and regression-focused, making it a reasonable anchor for this group of mostly tokenizer/model-compatibility reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44554", "right": "issue:46365", "accept": false, - "reason": "Different subsystems and failure modes: MPS attention correctness vs Trainer import broken by datasets." + "reason": "Different subsystems and failures: MPS attention correctness vs. Trainer import resolution." }, { "left": "issue:44514", "right": "issue:45290", "accept": false, - "reason": "Both involve apply_chat_template, but the triggers are different enough (batched padding=False vs assistant tool-call/no-content) that they do not look like the same bug." + "reason": "Both involve apply_chat_template, but the crash conditions and code paths differ materially." }, { "left": "issue:44625", "right": "issue:45584", "accept": false, - "reason": "Unrelated regressions: Qwen3.5 num_labels propagation vs Whisper empty transcription generation after token alignment." + "reason": "Unrelated models and bugs: Qwen3.5 config propagation vs Whisper empty-transcription handling." }, { "left": "issue:44977", "right": "issue:46209", "accept": false, - "reason": "Different concrete problems: Qwen3.5 flash-attention generation bug vs FP8 inference failure on B200." + "reason": "Different model/runtime issues: flash-attention behavior vs FP8 inference on B200." }, { "left": "issue:43452", "right": "issue:44451", "accept": false, - "reason": "Both are loading-related, but one is gguf_file path handling for AutoTokenizer/AutoModel and the other is a specific model load failure for ScandiBERT; not the same bug." + "reason": "Both are load failures, but for different models and seemingly different causes." }, { "left": "issue:45823", "right": "issue:46050", "accept": false, - "reason": "Device-mismatch during forward is not the same as quantized model save/load breakage." + "reason": "Device-map forward mismatch is not the same as quantized save/load corruption." }, { "left": "issue:28218", "right": "issue:33357", "accept": false, - "reason": "Tokenizer spacing bug and MacOS CLIP bus error are unrelated." + "reason": "Tokenizer spacing bug vs MacOS bus error; no shared underlying defect." }, { "left": "issue:43994", "right": "issue:46192", "accept": false, - "reason": "Wrong SigLIP2 outputs and silently ignored return_tensors are different tokenizer/processor issues with no shared code-path evidence." + "reason": "SigLIP2 output quality issue is unrelated to kwargs precedence/return_tensors behavior." }, { "left": "issue:42796", "right": "issue:44977", "accept": false, - "reason": "Tokenizer BPE-marker handling for Ministral is unrelated to Qwen3.5 flash-attention generation." + "reason": "Tokenizer BPE marker handling and Qwen3.5 flash-attention generation failure are unrelated." }, { "left": "issue:43278", "right": "issue:43844", "accept": false, - "reason": "Embedding dtype drift in eval is not the same bug as gradient explosion under ZeRO-3." + "reason": "Training/eval dtype mismatch differs from ZeRO-3 gradient inflation behavior." }, { "left": "issue:43493", "right": "issue:43844", "accept": false, - "reason": "SigLIP2 implementation discrepancy and ZeRO-3 gradient instability are distinct problems." + "reason": "SigLIP2 implementation mismatch is unrelated to DeepSpeed gradient regression." }, { "left": "issue:39692", "right": "issue:43295", "accept": false, - "reason": "Both touch multimodal processor/model usage, but one is a SigLIP2 doc example error set and the other is a regression in processor.tokenizer/images handling." + "reason": "Doc example mistakes are not the same as a processor/tokenizer regression in custom code." }, { "left": "issue:44849", "right": "issue:46209", "accept": false, - "reason": "Qwen3.5 hidden_states bug and FP8 inference-on-B200 bug are unrelated." + "reason": "Different failure modes: hidden_states output bug vs FP8 inference support issue." }, { "left": "issue:33357", "right": "issue:33666", "accept": false, - "reason": "MacOS bus error on a CLIP model is unrelated to Qwen2-VL multi-GPU training." + "reason": "MacOS bus error on a CLIP model and Qwen2-VL multi-GPU training are unrelated." }, { "left": "issue:43646", "right": "issue:44466", "accept": false, - "reason": "Custom model initialization breakage is unrelated to lm_head.weight serialization differences." + "reason": "Generic custom-model init breakage is not the same as device-dependent serialization of tied weights." }, { "left": "issue:44568", "right": "issue:46489", "accept": false, - "reason": "Both are tokenizer regressions in v5+, but they affect different models and different tokenizer behaviors; not enough evidence they are the same underlying bug." + "reason": "Different tokenizer regressions affecting different model families and symptoms." }, { "left": "issue:46209", "right": "issue:46710", "accept": false, - "reason": "FP8 inference failure on B200 is unrelated to DeepSeek-R1 accuracy regression after upgrade." + "reason": "Both mention performance/regression, but one is FP8 inference and the other is model accuracy after upgrade." }, { "left": "issue:43278", "right": "issue:43701", "accept": false, - "reason": "Embedding dtype mismatch and resume_from_checkpoint key mismatch are unrelated training issues." + "reason": "Checkpoint resume key mismatch is unrelated to training/eval dtype changes." }, { "left": "issue:44792", "right": "issue:45405", "accept": false, - "reason": "Janus image generation test failure and MIN_PEFT_VERSION bump are unrelated." + "reason": "A failing Janus image-generation test is unrelated to an unavailable PEFT version bump." }, { "left": "issue:43575", "right": "issue:45405", "accept": false, - "reason": "Tensor-parallel OOM for Qwen2-57B is unrelated to the PEFT version bump." + "reason": "Tensor-parallel OOM and dependency version pinning are entirely different issues." }, { "left": "issue:43927", "right": "issue:44361", "accept": false, - "reason": "Custom token ID loss in DiaConfig and MLukeTokenizer AttributeError are different code paths." + "reason": "Different model/config bugs: custom token IDs lost on save/load vs tokenizer AttributeError on tasks." }, { "left": "issue:44849", "right": "issue:45357", "accept": false, - "reason": "Qwen3.5 output_hidden_states bug and save_pretrained visual-key regression are distinct issues." + "reason": "Hidden-state output bug and incorrect visual encoder keys are different Qwen3.5 regressions." }, { "left": "issue:43257", "right": "issue:43299", "accept": false, - "reason": "Both concern Qwen3 MOE/Qwen3VL loading, but the symptoms and code paths differ enough that they do not look like one bug." + "reason": "Both involve Qwen MoE loading, but one is accelerate/deepspeed conversion and the other is a dev-version load failure; not the same bug." }, { "left": "issue:44589", "right": "issue:45520", "accept": false, - "reason": "Float8 storage lookup failure and flash_attn import KeyError are unrelated runtime/import problems." + "reason": "Float8 storage lookup and flash_attn import_utils KeyError are unrelated runtime/environment issues." }, { "left": "issue:43824", "right": "issue:44589", "accept": false, - "reason": "Qwen2_5_VL import error and Float8 storage lookup failure do not share the same bug." + "reason": "Import error for a VL class and float8 storage resolution failure do not share a code path." }, { "left": "issue:44488", "right": "issue:44843", "accept": false, - "reason": "Model load failure for cjvt/sleng-bert and HF_HUB_OFFLINE breakage in _patch_mistral_regex are unrelated." + "reason": "Model loading failure for a specific BERT model is not the same as offline regex patch behavior in AutoTokenizer." }, { "left": "issue:44568", "right": "issue:45812", "accept": false, - "reason": "Both are tokenizer regressions, but 44568 is about add_special_tokens/BOS-EOS on one tokenizer while 45812 is a broad AutoTokenizer wrong-ID regression across Granite models; different underlying bugs." + "reason": "Different tokenizer regressions: missing BOS/EOS insertion vs wrong IDs for Granite models." }, { "left": "issue:43504", "right": "issue:43606", "accept": false, - "reason": "BEiT pretrained preset loading and bark-small CPU-offload device mismatch are unrelated." + "reason": "Legacy-field model loading and CPU-offload device mismatch are unrelated." }, { "left": "issue:43334", "right": "issue:43572", "accept": false, - "reason": "Qwen3-VL pad_token_id missing and StableLm pad_token_idx missing are similar-looking config issues, but they affect different model families and are not clearly the same underlying defect." + "reason": "Similar symptom of missing config attributes, but different fields, models, and failure contexts." } ] }, @@ -8161,20 +8176,20 @@ "issue:15354|issue:38175", "issue:44423|issue:44734" ], - "prepared_review_unit_hash": "b28358a882847f17ddec635922746b97fdbe072d555e33dd465bc7a7da45090b", + "prepared_review_unit_hash": "a1e0111eb58313cf27baadfc64462ac8b01c68736acca28e1e424e88c1336f5f", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 26, - "serialized_chars": 34995, + "serialized_chars": 34993, "estimated_input_tokens": 8749, "estimated_eval_tokens": 17754 }, @@ -8183,168 +8198,168 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The soft pairs are mostly superficial overlaps by subsystem or model family, not true duplicate bugs. I would reject every proposed merge; the set is heterogeneous and does not form one coherent issue cluster.", - "confidence": 0.95, - "canonical_issue_reason": "issue:45759 is the closest candidate because it is the broadest multimodal-config-loading report in the set, but the overall collection is too heterogeneous to treat it as a true canonical duplicate.", + "summary": "The cluster is a grab-bag of unrelated transformer bugs across tokenizers, model loading, generation, quantization, serving, and CI. None of the soft pairs appear to be the same underlying issue, so I reject all of them.", + "confidence": 0.98, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:45759 is the least narrow issue among the items, but it still only fits a small subset of the set and is not a good global representative for the whole cluster.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43334", "right": "issue:45759", "accept": false, - "reason": "Same general Qwen multimodal loading area, but one is a missing pad_token_id on Qwen3-VL and the other is from_config failing to unwrap text_config; different root causes." + "reason": "Both involve Qwen multimodal config/loading, but one is a missing `pad_token_id` attribute and the other is `from_config` not unwrapping `text_config`; different bugs and fixes." }, { "left": "issue:43606", "right": "issue:44112", "accept": false, - "reason": "Both mention CI/device behavior, but one is a CPU offload mismatch in suno/bark-small and the other is a stale GraniteSpeech test on CI; not the same bug." + "reason": "Different models and failure modes: Bark CPU offload device mismatch vs a stale GraniteSpeech CI test." }, { "left": "issue:43572", "right": "issue:44291", "accept": false, - "reason": "Both are transformer-init regression reports, but one is missing pad_token_idx in StableLmConfig and the other is an unexpected _is_hf_initialized argument; distinct fixes." + "reason": "Both are load-time regressions after 5.0, but one is a missing config field and the other is an unexpected `_is_hf_initialized` argument; not the same code path." }, { "left": "issue:45245", "right": "issue:46710", "accept": false, - "reason": "Completely different failures: category-count limit error versus an accuracy regression after upgrade." + "reason": "Category-count runtime error versus an accuracy regression in a specific model; unrelated symptoms and causes." }, { "left": "issue:44493", "right": "issue:46285", "accept": false, - "reason": "Different code paths and symptoms: unexpected position-id key versus CLIP text-model structure inconsistency." + "reason": "Both mention structural/model-output changes, but one is an unexpected key warning and the other is a CLIP text model API inconsistency." }, { "left": "issue:43504", "right": "issue:43653", "accept": false, - "reason": "One is a legacy-field model-loading bug; the other is tokenizer special-token registration/empty decode output." + "reason": "Different components and bugs: BEiT legacy-field loading failure vs BigBirdTokenizer special-token registration/empty decode issue." }, { "left": "issue:44464", "right": "issue:46424", "accept": false, - "reason": "Chunked generation with compiled forward is unrelated to cache max length being honored only for quantized cache." + "reason": "Both touch generation/cache behavior, but one is compiled-forward inconsistency and the other is `max_cache_length` handling for quantized cache; not the same defect." }, { "left": "issue:43937", "right": "issue:46701", "accept": false, - "reason": "GenerationConfig validation failure and image-text pipeline breakage are different bugs with different surfaces." + "reason": "Different models and failure modes: invalid `GenerationConfig` vs a pipeline failure in DiffusionGemmaGenerationOutput." }, { "left": "issue:44464", "right": "issue:45584", "accept": false, - "reason": "Compiled-forward chunked generation inconsistency is unrelated to Whisper empty-transcription behavior after align_special_tokens." + "reason": "Chunked generation with compiled forward is unrelated to Whisper failing on empty transcription after `align_special_tokens`." }, { "left": "issue:45200", "right": "issue:45229", "accept": false, - "reason": "mm_token_type_ids defaulting to zeros and multi-GPU CUDA OOM are not the same underlying issue." + "reason": "One is a missing default for `mm_token_type_ids`; the other is a multi-GPU CUDA OOM. Different problems entirely." }, { "left": "issue:45229", "right": "issue:45823", "accept": false, - "reason": "Both are Gemma4 inference reports, but one is OOM and the other is a PLE device-mismatch during forward; different bugs." + "reason": "Both are Gemma4 inference issues, but OOM from multi-GPU versus a device mismatch in forward are distinct bugs." }, { "left": "issue:44869", "right": "issue:46752", "accept": false, - "reason": "Whisper timestamp decode crash and apply_chat_template empty-list IndexError are unrelated failures." + "reason": "Both are edge-case crashes, but one is Whisper timestamp decoding on a trailing replacement character and the other is `apply_chat_template` on an empty list." }, { "left": "issue:43844", "right": "issue:44964", "accept": false, - "reason": "ZeRO-3 gradient explosion is unrelated to a model-loading failure for Phi-4 multimodal." + "reason": "Different model/training stack issue versus a model-loading failure; no shared underlying bug." }, { "left": "issue:43873", "right": "issue:46050", "accept": false, - "reason": "Both involve quantization/offloading broadly, but one is offloading not working as expected and the other is broken save_pretrained/from_pretrained behavior." + "reason": "Both involve quantization/offloading, but one is offloading not working as expected and the other is save/load corruption after `sinq`; not mergeable as one fix." }, { "left": "issue:43994", "right": "issue:44493", "accept": false, - "reason": "SigLIP2 nonsensical outputs and unexpected position-id keys are different symptoms with different likely fixes." + "reason": "SIGLIP2 bad outputs and unexpected position-id key warnings are different regressions in different paths." }, { "left": "issue:29942", "right": "issue:30333", "accept": false, - "reason": "Flash Attention 2 test failures and MLflow job-status reporting are unrelated subsystems." + "reason": "Flash Attention 2 test failures are unrelated to MLFlow not reporting failed training jobs." }, { "left": "issue:45588", "right": "issue:46050", "accept": false, - "reason": "flash_attention.py crashing on s_aux=None is not the same bug as quantized model save/load breakage." + "reason": "A flash-attention integration crash with `s_aux=None` is unrelated to quantized save/load breakage." }, { "left": "issue:29159", "right": "issue:32090", "accept": false, - "reason": "Tokenizer slow/fast inconsistency and _gpu_broadcast_one rejecting NoneType are unrelated." + "reason": "Tokenizer slow/fast inconsistency and a trainer `_gpu_broadcast_one` TypeError are unrelated." }, { "left": "issue:42490", "right": "issue:43696", "accept": false, - "reason": "Local-path import warning with trailing slash and GPT-oss CUDA OOM are unrelated." + "reason": "Import-path warning from trailing slash and a CUDA OOM are not the same issue." }, { "left": "issue:45084", "right": "issue:45520", "accept": false, - "reason": "Non-template-node compilation failure and flash_attn import KeyError on Python 3.13 are different issues." + "reason": "Template-node compilation error and a Python 3.13 `flash_attn` import KeyError are unrelated." }, { "left": "issue:44977", "right": "issue:46701", "accept": false, - "reason": "Qwen3.5 flash-attention generation problems and DiffusionGemma pipeline failure do not share the same code path." + "reason": "Qwen3.5 flash-attention generation problems do not match a DiffusionGemma pipeline failure." }, { "left": "issue:43065", "right": "issue:46369", "accept": false, - "reason": "Same SAM3 family, but a dummy Conv2d in the decoder and a memory overflow in sam3-litetext are not clearly the same bug." + "reason": "Same broad SAM3 area, but dummy `nn.Conv2d` in one component is not the same as memory overflow in `sam3-litetext`." }, { "left": "issue:28218", "right": "issue:34567", "accept": false, - "reason": "Tokenizer added-token spacing and TrainerState token-count updates are unrelated." + "reason": "Tokenizer spacing behavior and `TrainerState.num_input_tokens_seen` not updating are different features with different symptoms." }, { "left": "issue:44554", "right": "issue:45446", "accept": false, - "reason": "MPS attention correctness with mismatched head dims and a flex_attention import version check are separate problems." + "reason": "MPS attention correctness with mismatched head dims is unrelated to an AuxRequest PyTorch version check in flex_attention." }, { "left": "issue:15354", "right": "issue:38175", "accept": false, - "reason": "torch.jit.script GeneratorExp support and SigLIP2 zero probabilities have no concrete overlap." + "reason": "TorchScript GeneratorExp export failure and SIGLIP2 zero-probability outputs are unrelated bugs." }, { "left": "issue:44423", "right": "issue:44734", "accept": false, - "reason": "Both are serve/runtime crashes, but continuous-batching multimodal failure and KV-cache continuation indexing are different code paths." + "reason": "Both are `transformers serve` crashes, but one is multimodal continuous batching and the other is KV-cache continuation tensor indexing; different code paths and fixes." } ] }, @@ -8448,214 +8463,214 @@ "issue:43278|issue:43856", "issue:45381|issue:46209" ], - "prepared_review_unit_hash": "019e2afa0349e6b01c9be8385ce0e2e7f97bc4f2df796087a0b4e9b5b7c11f69", + "prepared_review_unit_hash": "b5cdc853d90e2bac5ee276d272edd0fcd4a90f2d90d05a2200ef25be3e4f6b11", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 30, - "serialized_chars": 36391, - "estimated_input_tokens": 9098, - "estimated_eval_tokens": 18452 + "serialized_chars": 36397, + "estimated_input_tokens": 9100, + "estimated_eval_tokens": 18456 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: the issues span generation bugs, tokenizer/processor regressions, model-loading/import failures, vision/video preprocessing, device-map/offloading, and training/runtime correctness. None of the soft pairs look like the same underlying bug or change, so I\u2019d keep them separate. I\u2019d use the offloading/quantization report as the representative issue only because it has the most activity.", - "confidence": 0.84, - "canonical_issue_reason": "issue:43873 has the strongest engagement signal in the set (many comments and inbound references) and is a concrete runtime bug, but the cluster is not actually a coherent duplicate group.", + "summary": "This cluster is mostly heterogeneous: the soft-similarity links span unrelated bugs across generation, tokenizers, vision/video processors, quantization, and config/loading issues. I would keep 43295 as the best anchor only because it is the most connected and representative-looking issue in the set, but there is no true coherent duplicate cluster here.", + "confidence": 0.64, + "canonical_issue_reason": "No strong true canonical emerges from the titles alone; if one anchor is needed, issue 43295 is the best hub because it has the most soft neighbors and a clear, concrete regression report.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43873 is the best standalone representative because it is well-discussed, has inbound references, and describes a clear actionable bug around quantization/offloading behavior.", + "best_issue_reason": "Issue 43295 is the best representative by connectivity and specificity: it is a concrete regression in processor/tokenizer behavior and is the most central item among the soft-similarity candidates.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43493", "right": "issue:43701", "accept": false, - "reason": "Both are bug reports, but one is SigLIP2 implementation discrepancy and the other is a resume_from_checkpoint key mismatch; different code paths and symptoms." + "reason": "Different problems: SigLIP2 model-output discrepancy vs a resume_from_checkpoint key mismatch." }, { "left": "issue:41720", "right": "issue:46710", "accept": false, - "reason": "Qwen3 auto device-mapping CUDA assert and DeepSeek accuracy regression after upgrade are unrelated failures." + "reason": "Different failures: device-map CUDA assert on A800 vs accuracy regression after version upgrade." }, { "left": "issue:45406", "right": "issue:46121", "accept": false, - "reason": "Gemma4Processor _tokenizer attribute crash and rope-validation list TypeError affect different layers of the stack." + "reason": "Gemma4 processor attribute regression and rope-validation TypeError are unrelated code paths." }, { "left": "issue:43404", "right": "issue:46710", "accept": false, - "reason": "lm_head weight tying in Mistral3 is a model-weighting bug, while the DeepSeek issue is an accuracy regression; not the same bug." + "reason": "Different model families and symptoms; tied lm_head bug vs DeepSeek accuracy regression." }, { "left": "issue:45072", "right": "issue:45198", "accept": false, - "reason": "dtype mismatches in SwitchTransformers/TimmWrapperModel do not match Wav2Vec2 save/tokenization failures." + "reason": "Different areas: dtype mismatches in inference vs Wav2Vec2 save/tokenization failure." }, { "left": "issue:37428", "right": "issue:43957", "accept": false, - "reason": "Both involve model loading, but one is a missing flash-attention helper import and the other is meta-device loading breakage; different root causes." + "reason": "One is a flash-attention import symbol issue; the other is a meta-device model-loading regression." }, { "left": "issue:45588", "right": "issue:45759", "accept": false, - "reason": "flash_attention.py crashing on s_aux=None is unrelated to AutoModelForCausalLM.from_config not unwrapping text_config." + "reason": "Flash-attention sink handling crash and composite-config text_config unwrapping are unrelated." }, { "left": "issue:43295", "right": "issue:43450", "accept": false, - "reason": "Processor.tokenizer regression and batched video shape bug are both processor issues, but they are separate code paths." + "reason": "Both involve processors, but one is tokenizer access/images handling and the other is batched video shape; distinct bugs." }, { "left": "issue:43540", "right": "issue:45200", "accept": false, - "reason": "Qwen3OmniMoe video-input ValueError and Gemma4 mm_token_type_ids defaulting are different multimodal input problems." + "reason": "Video-input processing error in Qwen3OmniMoe vs Gemma4 mm_token_type_ids defaulting; not the same issue." }, { "left": "issue:45200", "right": "issue:45823", "accept": false, - "reason": "Gemma4 mm_token_type_ids defaulting and Gemma4 PLE device mismatch are both Gemma4-related but not the same underlying bug." + "reason": "Both are Gemma4-related, but one is missing mm_token_type_ids and the other is a device_map forward mismatch." }, { "left": "issue:44186", "right": "issue:46192", "accept": false, - "reason": "LayoutLMv2Tokenizer crashing on NER/batched padding is unrelated to return_tensors being ignored when text_kwargs is passed." + "reason": "Tokenizer crash on NER/batched padding is different from silently ignored return_tensors handling." }, { "left": "issue:43295", "right": "issue:43873", "accept": false, - "reason": "A processor.tokenizer regression and quantization/offloading behavior are unrelated subsystems." + "reason": "Processor/tokenizer regression and offloading-with-quantization are separate concerns." }, { "left": "issue:45823", "right": "issue:46192", "accept": false, - "reason": "Device placement in Gemma4 PLE is unrelated to the return_tensors/text_kwargs precedence bug." + "reason": "Gemma4 device mismatch and return_tensors precedence are unrelated bugs." }, { "left": "issue:43295", "right": "issue:44610", "accept": false, - "reason": "Both are processor-facing regressions, but one is about tokenizer access and the other about video input sizing." + "reason": "Custom processor/tokenizer regression vs OmDet-Turbo image-size mismatch are different failures." }, { "left": "issue:29870", "right": "issue:29942", "accept": false, - "reason": "Bart generate UnboundLocalError and failing Flash Attention 2 tests are not the same failure." + "reason": "Bart generate() UnboundLocalError and failing Flash Attention 2 tests do not describe the same bug." }, { "left": "issue:45520", "right": "issue:46178", "accept": false, - "reason": "flash_attn import handling on Python 3.13 and an unsupported tensorflow_text backend mapping are unrelated import/configuration issues." + "reason": "Flash-attn import KeyError and tensorflow_text backend-mapping error are different import-system issues." }, { "left": "issue:43572", "right": "issue:44987", "accept": false, - "reason": "Missing pad_token_idx in StableLmConfig and a load failure for physical-intelligence/fast are different regression reports." + "reason": "StableLmConfig pad_token_idx regression and loading physical-intelligence/fast are unrelated." }, { "left": "issue:43450", "right": "issue:43575", "accept": false, - "reason": "Video processor batch-shape bugs do not match Qwen2-57B-A14B-Instruct tensor-parallel OOM." + "reason": "Batched video processor shape bug and tensor-parallel OOM are not the same underlying problem." }, { "left": "issue:43550", "right": "issue:43844", "accept": false, - "reason": "torch.compile+SDPA failure in Bamba-9B-v2 is unrelated to ZeRO-3 gradient growth in training." + "reason": "torch.compile+SDPA failure in Bamba is distinct from ZeRO-3 gradient growth during training." }, { "left": "issue:43232", "right": "issue:44849", "accept": false, - "reason": "Generation kwargs update after sync_gpus and Qwen3.5 hidden-state output bugs are different generation path problems." + "reason": "Generation kwargs after sync_gpus and output_hidden_states regression are separate code paths." }, { "left": "issue:43572", "right": "issue:43957", "accept": false, - "reason": "StableLmConfig pad_token_idx regression and meta-device loading breakage are not the same defect." + "reason": "Config pad token issue is unrelated to meta-device model loading breakage." }, { "left": "issue:44554", "right": "issue:45356", "accept": false, - "reason": "MPS attention correctness with mismatched head dims and Kimi tokenizer regression are unrelated." + "reason": "MPS attention correctness bug and Kimi tokenizer codec regression are unrelated." }, { "left": "issue:37428", "right": "issue:43824", "accept": false, - "reason": "These are both import errors, but one is a missing flash-attention utility and the other is a missing Qwen2.5-VL model export." + "reason": "Different import errors for different symbols/models; no shared concrete bug." }, { "left": "issue:42490", "right": "issue:43493", "accept": false, - "reason": "Local-path trailing-slash import warning and SigLIP2 implementation discrepancy have no shared underlying bug." + "reason": "Trailing-slash local model path warning and SigLIP2 implementation discrepancy are unrelated." }, { "left": "issue:44466", "right": "issue:45127", "accept": false, - "reason": "Inconsistent lm_head serialization by device and LoRA-merge collapse with extended vocab may both touch tied weights, but the symptoms and triggers differ." + "reason": "Tied-weight serialization inconsistency and LoRA merge collapse are different weight-handling bugs." }, { "left": "issue:38175", "right": "issue:43493", "accept": false, - "reason": "Unexpected zero probabilities in SigLIP2 and broader HF-vs-JAX discrepancy are related by model family only, not clearly the same bug." + "reason": "Close in model family, but zero probabilities on a checkpoint vs HF-vs-JAX discrepancy are not clearly the same concrete bug." }, { "left": "issue:39692", "right": "issue:46192", "accept": false, - "reason": "SigLIP2 doc-example errors and return_tensors being ignored are unrelated." + "reason": "SigLIP2 doc-example/model-processor issues and return_tensors precedence are unrelated." }, { "left": "issue:43653", "right": "issue:44112", "accept": false, - "reason": "BigBirdTokenizer special-token registration and GraniteSpeech stale device override failures are separate CI/runtime issues." + "reason": "BigBirdTokenizer special-token registration and GraniteSpeech CI stale device override are different issues." }, { "left": "issue:43278", "right": "issue:43856", "accept": false, - "reason": "BF16-to-FP32 embedding dtype drift in evaluate is unrelated to Qwen3 MoE memory inefficiency." + "reason": "Embedding dtype drift in eval and Qwen3 MoE memory inefficiency are unrelated." }, { "left": "issue:45381", "right": "issue:46209", "accept": false, - "reason": "Qwen2.5-VL video position IDs and FP8 inference on B200 are different hardware/model issues." + "reason": "Qwen2.5-VL video position IDs and FP8 inference on B200 are different hardware/model problems." } ] }, @@ -8757,20 +8772,20 @@ "issue:43606|issue:45083", "issue:38175|issue:45412" ], - "prepared_review_unit_hash": "650d7cd92f7fab08f78bb579520e0a10c480050b7a6decb774556d8e18c968c8", + "prepared_review_unit_hash": "404dc1decb55664b032d4fb88f00caf98dd3cffff40e9e0a67879f6cc8ed9454", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 27, - "serialized_chars": 36145, + "serialized_chars": 36148, "estimated_input_tokens": 9037, "estimated_eval_tokens": 18330 }, @@ -8779,174 +8794,174 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: the soft pairs are mostly superficial subsystem similarities, not the same underlying bug/change. I would not merge any of the candidate pairs. If a representative issue is needed, the multimodal `apply_chat_template` bug (44521) is the strongest anchor by activity and references, but the cluster is not a true duplicate set.", - "confidence": 0.86, - "canonical_issue_reason": "issue:44521 is the most active and referenced item, with a concrete reproducible regression in a core API (`apply_chat_template`) and an open status, so it is the best loose representative for this noisy cluster.", + "summary": "This cluster is highly heterogeneous: the soft pairs span unrelated tokenizer, multimodal processor, config, loading, docs, and hardware-specific bugs. I do not see any pair that clearly describes the same underlying defect, so all soft edges are rejected. No PRs are present.", + "confidence": 0.94, + "canonical_issue_reason": "issue:44521 is the most discussion-heavy, actively open issue in the set and is a reasonable representative node for the cluster, but it is not a duplicate of the other issues.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44521 is the best global issue candidate because it is open, well-discussed, and clearly actionable; the other issues are narrower and/or unrelated to one another.", + "best_issue_reason": "issue:44521 is the best overall representative issue by activity and recency; it has the strongest signal among otherwise unrelated issues, but it should not be treated as semantically merged with the others.", "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 multimodal assistant masks and the other about tool-call messages with empty content; different failure modes and likely different fixes." + "reason": "Both concern apply_chat_template, but one is about multimodal assistant masks and the other is a tokenize=True crash with tool-call assistant messages; different failure modes and code paths." }, { "left": "issue:28218", "right": "issue:35141", "accept": false, - "reason": "Tokenizer spacing after added tokens is unrelated to embedding resizing / `post_init` reinitialization." + "reason": "Tokenizer spacing after added tokens is unrelated to embedding reinitialization during post_init when tie_word_embedding=False." }, { "left": "issue:42915", "right": "issue:43824", "accept": false, - "reason": "FP8 failure in Qwen3MoE and missing Qwen2.5-VL import are unrelated issues." + "reason": "Different models and failures: Qwen3MoE FP8 inference vs Qwen2.5-VL import error." }, { "left": "issue:43950", "right": "issue:46285", "accept": false, - "reason": "Both touch model loading/structure, but one is a `from_pretrained` buffer corruption regression and the other is a CLIP class API inconsistency; not the same bug." + "reason": "Non-persistent buffer corruption in from_pretrained is unrelated to CLIP text model structural inconsistency." }, { "left": "issue:43452", "right": "issue:46512", "accept": false, - "reason": "GGUF loading failure and stale `config.dtype` in `from_config(dtype=...)` are different code paths and different regressions." + "reason": "gguf_file loading breakage and stale config.dtype after from_config(dtype=...) are separate loading/configuration bugs." }, { "left": "issue:43824", "right": "issue:43975", "accept": false, - "reason": "Missing export for `Qwen2_5_VLForConditionalGeneration` is unrelated to detokenization behavior in DeepSeek coder." + "reason": "Import failure for Qwen2.5-VL is unrelated to incorrect detokenization for deepseek-coder." }, { "left": "issue:43334", "right": "issue:43525", "accept": false, - "reason": "Both are missing `pad_token_id`, but they are distinct model/config regressions for different architectures and are not clearly one bug." + "reason": "Both mention pad_token_id, but they are different model families with likely model-specific config issues, not the same concrete bug." }, { "left": "issue:43262", "right": "issue:43450", "accept": false, - "reason": "Audio chat-template sampling rate and batched video processor output shape are unrelated processor bugs." + "reason": "Audio processor sampling-rate defaulting and video batch shape handling are different processor bugs." }, { "left": "issue:44442", "right": "issue:45198", "accept": false, - "reason": "Tokenizer loading for FastSpeech2Conformer and save/tokenization failure for Wav2Vec2 are separate speech-model issues." + "reason": "FastSpeech2ConformerTokenizer loading failure is unrelated to Wav2Vec2 save_pretrained/tokenization failure." }, { "left": "issue:44792", "right": "issue:45563", "accept": false, - "reason": "A Janus image-generation test failure and a stale paged-generate warning about `num_return_sequences` are not the same underlying problem." + "reason": "Janus image-generation test failure and a stale paged-generate warning are different generation-path issues." }, { "left": "issue:44479", "right": "issue:46192", "accept": false, - "reason": "A video-input regression in Qwen VL models is unrelated to `return_tensors` being ignored when `text_kwargs` is passed." + "reason": "Qwen multimodal video-input regression is unrelated to return_tensors being ignored when text_kwargs is passed." }, { "left": "issue:44898", "right": "issue:45412", "accept": false, - "reason": "Perceiver non-default-resolution failure and RT-DETR memory-release issue are unrelated vision bugs." + "reason": "Perceiver non-default-resolution failure and RT-DETR memory release are unrelated." }, { "left": "issue:43819", "right": "issue:44112", "accept": false, - "reason": "DAC latent/STE mismatch and a GraniteSpeech CI device-override test failure are different problems." + "reason": "DAC.from_latents mismatch and a GraniteSpeech CI test failure are different audio/model behaviors." }, { "left": "issue:43827", "right": "issue:43901", "accept": false, - "reason": "Both are docs issues around pipelines, but they concern different APIs and different deprecations; not one duplicate." + "reason": "Both are docs issues, but they concern different pipeline APIs and different deprecations." }, { "left": "issue:45200", "right": "issue:45325", "accept": false, - "reason": "Gemma 4 text-only `mm_token_type_ids` handling and Qwen2.5-VL temporal position scaling are different multimodal bugs." + "reason": "Gemma 4 mm_token_type_ids handling is unrelated to Qwen2.5-VL rope index scaling." }, { "left": "issue:44610", "right": "issue:44857", "accept": false, - "reason": "OmDet-Turbo preprocessing size mismatch and LwDETR AMP crash are unrelated detection issues." + "reason": "OmDet-Turbo input-size mismatch and LwDetrImageLoss AMP crash are different detection bugs." }, { "left": "issue:43388", "right": "issue:43526", "accept": false, - "reason": "Both are label-handling bugs, but one drops labels in metric gathering and the other reduces labels incorrectly in an image processor; different fixes." + "reason": "Both involve labels, but one is gather_for_metrics truncation and the other is BeitImageProcessorFast reduce_labels returning the wrong shape." }, { "left": "issue:44561", "right": "issue:46234", "accept": false, - "reason": "Both concern `import_utils`, but one is removal of `is_torch_fx_available` breaking remote-code models and the other is a missing Falcon-Mamba availability helper." + "reason": "Different missing utility functions with different downstream breakages; not the same code-path problem." }, { "left": "issue:43335", "right": "issue:43550", "accept": false, - "reason": "Sparse-layer construction in SwitchTransformers and Bamba torch.compile/SDPA failure are unrelated." + "reason": "SwitchTransformers sparse-layer construction and Bamba torch.compile/SDPA failure are unrelated." }, { "left": "issue:43819", "right": "issue:45083", "accept": false, - "reason": "DAC latent matching and `_get_feat_extract_output_lengths` behavior in qwen3_omni_moe are unrelated." + "reason": "DAC latent decoding mismatch is unrelated to qwen3_omni_moe feature-length helper behavior." }, { "left": "issue:44857", "right": "issue:45593", "accept": false, - "reason": "Both are detection-related, but one is an AMP/CUDA crash in a loss and the other is missing auxiliary losses when denoising is off." + "reason": "AMP crash in LwDetrImageLoss is unrelated to D-FINE auxiliary-loss handling." }, { "left": "issue:41720", "right": "issue:46209", "accept": false, - "reason": "Both are hardware/runtime failures, but they affect different model families and likely different root causes." + "reason": "Different model/hardware inference failures: Qwen3 auto device mapping on A800 vs FP8 inference on B200." }, { "left": "issue:44206", "right": "issue:45584", "accept": false, - "reason": "Unsupported `center` in a feature extractor and Whisper empty-transcription failure after token alignment are different audio bugs." + "reason": "Unsupported center arg in LasrFeatureExtractor is unrelated to Whisper empty-transcription failure after align_special_tokens." }, { "left": "issue:45759", "right": "issue:46285", "accept": false, - "reason": "Composite Qwen multimodal config unwrapping and CLIP text-model structure inconsistency are separate API regressions." + "reason": "Qwen multimodal from_config text_config unwrapping and CLIP text-model structure divergence are different model-initialization issues." }, { "left": "issue:45520", "right": "issue:46170", "accept": false, - "reason": "Python 3.13 `flash_attn` import-key error and Windows ROCm `torch.distributed` dependency failure are platform-specific but unrelated." + "reason": "Python 3.13 flash_attn import_utils KeyError and Windows ROCm torch.distributed import/type errors are different platform-specific issues." }, { "left": "issue:43606", "right": "issue:45083", "accept": false, - "reason": "CPU offload device mismatch in Bark and qwen3_omni_moe feature-length helper behavior are unrelated." + "reason": "CPU offload device mismatch in bark-small is unrelated to qwen3_omni_moe feature-length helper behavior." }, { "left": "issue:38175", "right": "issue:45412", "accept": false, - "reason": "Unexpected zero probabilities in SigLIP2 and RT-DETR memory leakage are unrelated issues." + "reason": "Siglip2 zero probabilities and RT-DETR memory release are unrelated bugs." } ] }, @@ -9049,217 +9064,217 @@ "issue:43819|issue:44493", "issue:44448|issue:44779" ], - "prepared_review_unit_hash": "254cb069db7d573305d88a336e53b8fc31fb48b729135a3e6a24e70090ee3184", + "prepared_review_unit_hash": "5995cea2ef3885348e665c17a1bdfc0899317e8965b653c6fa633664fed6056c", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 29, - "serialized_chars": 36058, - "estimated_input_tokens": 9015, - "estimated_eval_tokens": 18286 + "serialized_chars": 36062, + "estimated_input_tokens": 9016, + "estimated_eval_tokens": 18288 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly loose similarity across unrelated Transformers bugs, with one strong duplicate-like candidate: 45310/45478. The packet supports that pair well via identical titles and the same Qwen3.5 MoE from_pretrained failure, but it does not explicitly show a duplicate link, so this should be treated conservatively as a likely duplicate rather than a confirmed exact duplicate set.", - "confidence": 0.89, - "canonical_issue_reason": "issue:45310 is the best canonical issue because it is the earlier report and the clearest anchor for the Qwen3.5 MoE from_pretrained failure; issue:45478 is a highly similar follow-up that appears duplicate-like, though the packet does not explicitly confirm duplication.", + "summary": "The cluster is mostly heterogeneous; the only clear duplicate pair is the Qwen3.5 MoE from_pretrained regression (45310/45478). The other soft links are mostly same-subsystems but distinct bugs and should stay separate.", + "confidence": 0.88, + "canonical_issue_reason": "issue:45310 is the earlier and clearer representative of the only exact duplicate pair here; issue:45478 has the same title and describes the same Qwen3.5 MoE from_pretrained failure on transformers>=5.4.0.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45310 is the strongest issue to anchor the cluster: it matches the closely related 45478 report and describes a concrete Qwen3.5 MoE from_pretrained loading failure.", + "best_issue_reason": "issue:45310 is the best cluster representative because it is the strongest confirmed duplicate and the rest of the set is too diverse to unify cleanly under one issue.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:37428", "right": "issue:43425", "accept": false, - "reason": "Different failures: a missing internal import symbol vs Torch 2.10 compatibility." + "reason": "Different failures: missing internal flash-attn helper import vs a Torch 2.10 compatibility report." }, { "left": "issue:43335", "right": "issue:43653", "accept": false, - "reason": "Different models and bugs: SwitchTransformers config construction vs BigBird tokenizer special-token handling." + "reason": "Different models and bugs: SwitchTransformers sparse-layer config vs BigBirdTokenizer special-token decoding." }, { "left": "issue:43550", "right": "issue:44112", "accept": false, - "reason": "One is a torch.compile/SDPA failure, the other is a CI stale device override test." + "reason": "Bamba SDPA/torch.compile failure and GraniteSpeech CI stale-device test are unrelated code paths." }, { "left": "issue:29942", "right": "issue:32090", "accept": false, - "reason": "Different code paths: Flash Attention 2 test failures vs Trainer broadcast of NoneType." + "reason": "Flash Attention 2 test failures and Trainer _gpu_broadcast_one NoneType errors are distinct issues." }, { "left": "issue:28282", "right": "issue:33357", "accept": false, - "reason": "Missing PyTorch import is unrelated to the MacOS CLIP bus error." + "reason": "PyTorch-missing ImportError vs Mac bus error loading CLIP are different root causes." }, { "left": "issue:43611", "right": "issue:43931", "accept": false, - "reason": "Base-model-prefix loading regression and weight-shape mismatch are distinct model-loading bugs." + "reason": "base_model_prefix loading regression and Qwen3-VL weight-shape mismatch are not the same bug." }, { "left": "issue:44448", "right": "issue:44743", "accept": false, - "reason": "Pegasus output drift is unrelated to Qwen recurrent state reset behavior." + "reason": "Pegasus v4/v5 output drift and Qwen3_5 recurrent-state reset are different model behaviors." }, { "left": "issue:43335", "right": "issue:43504", "accept": false, - "reason": "Different underlying problems in different model families; not the same bug." + "reason": "SwitchTransformers sparse-layer creation and Beit legacy-field loading are unrelated." }, { "left": "issue:45759", "right": "issue:46512", "accept": false, - "reason": "Both involve config handling, but one is text_config unwrapping and the other is stale dtype metadata." + "reason": "Both involve from_config, but one is text_config unwrapping for multimodal configs and the other is stale config.dtype after construction." }, { "left": "issue:44855", "right": "issue:45083", "accept": false, - "reason": "Python 3.13 torch.jit parsing error vs helper length calculation behavior." + "reason": "Python 3.13 TorchScript indentation parsing error and qwen3_omni output-length helper behavior are unrelated." }, { "left": "issue:43278", "right": "issue:43825", "accept": false, - "reason": "Embedding dtype drift and pipeline translation messaging are unrelated." + "reason": "BF16-to-FP32 dtype regression during evaluate and an incorrect pipeline error message are different problems." }, { "left": "issue:43720", "right": "issue:44112", "accept": false, - "reason": "Packed-weight unpacking during accelerate loading is not the same as a stale CI device override." + "reason": "BitNet packed-weight unloading bug and GraniteSpeech stale-device CI failure do not share the same code path." }, { "left": "issue:43335", "right": "issue:45083", "accept": false, - "reason": "SwitchTransformers sparse-layer config bug is unrelated to qwen3_omni_moe feature-length logic." + "reason": "SwitchTransformers sparse-layer config bug is unrelated to qwen3_omni feature-length computation." }, { "left": "issue:29127", "right": "issue:32090", "accept": false, - "reason": "Different error classes: clearer LayoutLMv3 message vs GPU broadcast TypeError." + "reason": "LayoutLMV3 error-message clarity and Trainer broadcast NoneType failure are different issues." }, { "left": "issue:43335", "right": "issue:44112", "accept": false, - "reason": "No shared code path or concrete bug; only broad Transformers similarity." + "reason": "SwitchTransformers config bug and GraniteSpeech CI stale-device test are unrelated." }, { "left": "issue:43335", "right": "issue:45593", "accept": false, - "reason": "Sparse-layer creation and auxiliary-loss omission are distinct issues." + "reason": "SwitchTransformers sparse-layer creation and D-FINE auxiliary-loss handling are different model bugs." }, { "left": "issue:37428", "right": "issue:44351", "accept": false, - "reason": "Both are import errors, but they involve different symbols and modules." + "reason": "Both are import errors, but they concern different missing symbols and likely different fixes." }, { "left": "issue:44514", "right": "issue:44521", "accept": false, - "reason": "Both touch apply_chat_template, but one is a batched padding crash and the other is all-zero assistant masks for multimodal inputs." + "reason": "Both are apply_chat_template issues, but one is a batched padding=False crash and the other is incorrect assistant_masks for multimodal inputs." }, { "left": "issue:45588", "right": "issue:45859", "accept": false, - "reason": "Flash-attention crash on sink-less models is different from CPU-offload incompatibility in Qwen3.5 MoE." + "reason": "flash_attention.py s_aux=None crash and Qwen3_5 rotary embedding CPU-offload incompatibility are different failures." }, { "left": "issue:34689", "right": "issue:35141", "accept": false, - "reason": "Model loading breakage and embedding reinitialization after resize are different bugs." + "reason": "Model loading breakage for Llama 3.2 Vision and token-embedding reinitialization after resize are separate bugs." }, { "left": "issue:46431", "right": "issue:46566", "accept": false, - "reason": "Missing chat template ValueError is unrelated to meta-device weights under device_map=auto." + "reason": "Missing chat template error and DiffusionGemma meta-device loading issue are unrelated." }, { "left": "issue:45290", "right": "issue:46326", "accept": false, - "reason": "Different apply_chat_template failures with different triggers and symptoms." + "reason": "Both involve chat templating, but tool-call/no-content crashes and mistral_common continue_final_message behavior are not the same bug." }, { "left": "issue:45127", "right": "issue:45245", "accept": false, - "reason": "LoRA merge collapse with tied embeddings is unrelated to the category-count limit error." + "reason": "LoRA merge collapse with extended vocab and category-count overflow are unrelated." }, { "left": "issue:42490", "right": "issue:43854", "accept": false, - "reason": "A local-path import warning and a GLM-4.7 loading test failure are not the same bug." + "reason": "Trailing-slash local-path warning and GLM-4.7-Flash loading failure are distinct issues." }, { "left": "issue:45200", "right": "issue:45588", "accept": false, - "reason": "Gemma 4 token-type defaults and flash_attention s_aux handling are unrelated." + "reason": "mm_token_type_ids defaulting and flash_attention s_aux handling are unrelated." }, { "left": "issue:45310", "right": "issue:45478", "accept": true, - "reason": "Same title and same Qwen3.5 MoE from_pretrained error; even without an explicit duplicate link, they look like the same underlying report." + "reason": "Same title and same Qwen3.5 MoE from_pretrained regression on transformers>=5.4.0; clear duplicate." }, { "left": "issue:43606", "right": "issue:44857", "accept": false, - "reason": "CPU offload device mismatch is distinct from LwDetrImageLoss float16 AMP failure." + "reason": "CPU offload device mismatch for bark-small and float16 AMP CUDA crash in LwDetrImageLoss are different bugs." }, { "left": "issue:43819", "right": "issue:44493", "accept": false, - "reason": "DAC latent decoding mismatch and unexpected position-id keys are different model issues." + "reason": "DAC from_latents/STE mismatch and unexpected position-id keys affect different code paths." }, { "left": "issue:44448", "right": "issue:44779", "accept": false, - "reason": "Pegasus v4/v5 output differences are unrelated to Deepseek tokenizer regressions." + "reason": "Pegasus v4/v5 output differences and Deepseek tokenizer v5 regression are separate tokenizer/model issues." } ] }, "evaluator_result": { - "accept": false, - "feedback": "The summary is a bit too strong for the packet. The 45310/45478 pair is well-supported as a likely duplicate, but the packet does not explicitly show a duplicate link, so phrases like \"true duplicate pair\" and \"exactly duplicated\" overstate the evidence. Keep that pairing as a conservative duplicate-like match, and phrase the cluster conclusion as mostly loose similarity with one strong candidate duplicate, not a confirmed duplicate set." + "accept": true, + "feedback": "Grounded overall. The only clearly duplicated pair in the packet is 45310/45478, which have identical titles and matching Qwen3.5 MoE from_pretrained regression descriptions. The remaining soft links are handled conservatively as distinct bugs, with reasons tied to different models/code paths rather than overstating duplication." }, "evaluator_used": true, - "retried": true, + "retried": false, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -9357,20 +9372,20 @@ "issue:44857|issue:45588", "issue:43927|issue:45698" ], - "prepared_review_unit_hash": "f30806ddd26d641c1a63b818bd1cb25cba85dd8229d89345fe740f9957b71e30", + "prepared_review_unit_hash": "b597d41b7b2b7dee9e8624538671f5712989373fed191624e88cd9749ea9cb75", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36480, + "serialized_chars": 36477, "estimated_input_tokens": 9120, "estimated_eval_tokens": 18496 }, @@ -9379,180 +9394,186 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: the soft pairs span unrelated bugs in tokenization, training, multimodal models, dtype/device issues, and import/runtime regressions. I would not merge any of the candidate soft edges; they do not look like the same underlying change or defect.", - "confidence": 0.61, - "canonical_issue_reason": "issue:45920 is the best representative because it is a broad, high-signal regression affecting multiple model families, with inbound references and an active cross-model failure pattern.", + "summary": "This cluster is a grab-bag of unrelated Transformers issues across tokenization, model loading, training, Flash Attention, JIT/export, dtype, and OOM behavior. None of the soft pairs look like the same underlying bug or change, so I reject all edges.", + "confidence": 0.97, + "canonical_issue_reason": "No clear canonical issue: the items span many unrelated subsystems and failure modes rather than one duplicate bug.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45920 is the strongest standalone issue in the set for triage/anchoring because it covers a widespread tokenizer regression rather than a model-specific edge case.", + "best_issue_reason": "Issue 45920 is the best representative only because it is the broadest, highest-signal regression report in the set; it still is not a duplicate of the others.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:39692", "right": "issue:43526", "accept": false, - "reason": "Different subsystems and failures: SigLIP2 docs/model-processor mismatch vs BEiT image processor label reduction bug." + "reason": "SigLIP2 docs/model-processor example errors and BeitImageProcessorFast reduce_labels are different bugs in different components." }, { "left": "issue:26413", "right": "issue:31501", "accept": false, - "reason": "Both involve training startup, but the concrete failures differ: checkpoint resume iterator handling vs FineWeb SLM training not starting." + "reason": "Both involve training startup, but one is resume_from_checkpoint epoch handling while the other is a FineWeb dataset/training launch failure." }, { "left": "issue:45459", "right": "issue:46234", "accept": false, - "reason": "Tokenizer protobuf error hiding decode failures is unrelated to a missing import_utils availability helper." + "reason": "One is a tokenizer exception-masking problem, the other is a missing import_utils feature flag; unrelated fixes." }, { "left": "issue:28218", "right": "issue:32090", "accept": false, - "reason": "Tokenizer spacing after added tokens is unrelated to a Trainer broadcast NoneType crash." + "reason": "Tokenizer added-space behavior and Trainer _gpu_broadcast_one NoneType are separate code paths and symptoms." }, { "left": "issue:39692", "right": "issue:44493", "accept": false, - "reason": "Docs/example mismatch for SigLIP2 does not match the position-id key regression affecting model loading/generation." + "reason": "Doc/example mismatch for SigLIP2 is not the same as the position_id unexpected-key regression." }, { "left": "issue:44617", "right": "issue:46697", "accept": false, - "reason": "One is a GPU OOM in Sam3Video; the other is an image preprocessor over-padding bug." + "reason": "CUDA OOM in Sam3Video and Swin2SR over-padding are distinct model/image-processing issues." }, { "left": "issue:44206", "right": "issue:45823", "accept": false, - "reason": "Unsupported center argument in LASR feature extraction is a different code path from Gemma4 PLE device-map mismatch." + "reason": "Unsupported center arg in LasrFeatureExtractor and Gemma4 device_map mismatch are different regressions." }, { "left": "issue:43701", "right": "issue:45083", "accept": false, - "reason": "Resume checkpoint key mismatch is unrelated to a Qwen3 Omni MoE helper-length calculation bug." + "reason": "resume_from_checkpoint key mismatch is unrelated to qwen3_omni_moe feature-extraction length logic." }, { "left": "issue:44857", "right": "issue:45072", "accept": false, - "reason": "Float16 AMP crash in LwDetrImageLoss is not the same as bfloat16 dtype mismatches in SwitchTransformers/TimmWrapperModel." + "reason": "Both mention precision, but an AMP crash in LwDetrImageLoss is not the same bug as bfloat16 dtype mismatches in inference." }, { "left": "issue:32090", "right": "issue:33357", "accept": false, - "reason": "Trainer NoneType broadcast failure and MacOS CLIP bus error are distinct runtime failures with different symptoms and likely causes." + "reason": "Trainer NoneType broadcast failure and MacOS bus error on CLIP are different failure modes and subsystems." }, { "left": "issue:42490", "right": "issue:43575", "accept": false, - "reason": "Trailing-slash import warning is unrelated to Qwen2-57B-A14B-Instruct tensor-parallel OOM." + "reason": "Trailing-slash import warning and tp-related OOM when loading Qwen2-57B are unrelated." }, { "left": "issue:44360", "right": "issue:44561", "accept": false, - "reason": "Missing ReLU in DSA indexer is not the same as removal of is_torch_fx_available breaking trust_remote_code imports." + "reason": "DSA indexer missing ReLU and removal of is_torch_fx_available breaking trust_remote_code models are different API/code changes." }, { "left": "issue:43550", "right": "issue:45083", "accept": false, - "reason": "torch.compile/SDPA failure in Bamba-9B-v2 is unrelated to the Qwen3 Omni MoE output-length helper bug." + "reason": "Bamba torch.compile/SDPA failure is not the same as qwen3_omni_moe feature-length behavior." }, { "left": "issue:38175", "right": "issue:44493", "accept": false, - "reason": "Unexpected zero probabilities in SigLIP2 is a model-specific inference issue, not the same as unexpected position-id keys." + "reason": "Unexpected zero probabilities in SigLIP2 and unexpected position_id keys are separate model-level regressions." }, { "left": "issue:29942", "right": "issue:35141", "accept": false, - "reason": "Flash Attention 2 test failures are unrelated to output embedding reinitialization during resize/post_init." + "reason": "Flash Attention 2 test failures and post_init reinitializing output embeddings are unrelated issues." }, { "left": "issue:27957", "right": "issue:34689", "accept": false, - "reason": "Both mention model loading/attention, but XLMRoberta FA2 support and Llama 3.2 Vision load regression are different concrete problems." + "reason": "XLMRoberta Flash Attention 2 and Llama 3.2 Vision model loading on 4.46.2 are different bugs." }, { "left": "issue:43278", "right": "issue:43329", "accept": false, - "reason": "Embedding dtype drift in eval is unrelated to undefined video-branch variables in multimodal token counting." + "reason": "Training/evaluate embedding dtype drift and undefined video-branch variables in _get_num_multimodal_tokens are unrelated." }, { "left": "issue:29870", "right": "issue:32090", "accept": false, - "reason": "Bart generate decoder_attention_mask initialization bug is unrelated to Trainer GPU broadcast of NoneType." + "reason": "Bart generate decoder_attention_mask UnboundLocalError and _gpu_broadcast_one NoneType are distinct problems." }, { "left": "issue:44448", "right": "issue:44977", "accept": false, - "reason": "Pegasus v4/v5 output drift is a different issue from Qwen3.5 flash-attention generation failure." + "reason": "Pegasus v4/v5 output difference and Qwen3.5 flash-attention generation failure are not the same underlying bug." }, { "left": "issue:45920", "right": "issue:46192", "accept": false, - "reason": "Both touch tokenization, but one is a wrong-token-ID regression across models while the other is an API precedence bug for return_tensors." + "reason": "Both touch tokenizer APIs, but wrong token IDs and return_tensors being ignored are different behaviors with different fixes." }, { "left": "issue:40444", "right": "issue:43526", "accept": false, - "reason": "Qwen2.5-VL iterable dataset multi-image finetuning failure is unrelated to BEiT label reduction behavior." + "reason": "Qwen2.5-VL multi-image iterable dataset finetuning failure and BeitImageProcessorFast reduce_labels are unrelated." }, { "left": "issue:43278", "right": "issue:43381", "accept": false, - "reason": "Eval/training dtype mismatch is not the same as gradient checkpointing being disallowed in eval mode." + "reason": "Embedding dtype change in eval and gradient checkpointing being disallowed in eval mode are separate eval-mode issues." }, { "left": "issue:10000", "right": "issue:15354", "accept": false, - "reason": "Older model/script errors, but different model families and different root causes: DistilBert German issue vs torch.jit GeneratorExp export failure." + "reason": "German DistilBertModel issue and GeneratorExp unsupported by torch.jit.script are unrelated historical bugs." }, { "left": "issue:44977", "right": "issue:45229", "accept": false, - "reason": "Qwen3.5 flash-attention generation bug and Gemma4 multi-GPU OOM are unrelated failures." + "reason": "Qwen3.5 flash-attention generation and Gemma4 multi-GPU CUDA OOM are different classes of problem." }, { "left": "issue:45290", "right": "issue:45381", "accept": false, - "reason": "Chat-template tool-call handling is a different path from Qwen2.5-VL video vision_position_ids." + "reason": "Chat template tool-call crash and qwen2.5-vl video position_id bug affect different formatting/code paths." }, { "left": "issue:44855", "right": "issue:46399", "accept": false, - "reason": "Python 3.13 IndentationError in DebertaV2Model parsing is unrelated to a missing decorator registration in Qwen3.5 kernels." + "reason": "Both involve decorator/JIT-related errors, but the Python 3.13 IndentationError and missing-decorator registration bug are different failures." + }, + { + "left": "issue:29870", + "right": "issue:33290", + "accept": false, + "reason": "Bart generate crash and Adafactor OOM in deepspeed are unrelated training/inference issues." }, { "left": "issue:44857", "right": "issue:45588", "accept": false, - "reason": "Mixed precision loss crash and flash_attention.py s_aux=None AttributeError are separate code paths and symptoms." + "reason": "LwDetrImageLoss AMP crash and flash_attention.py s_aux=None AttributeError are separate code paths." }, { "left": "issue:43927", "right": "issue:45698", "accept": false, - "reason": "Both concern save/load of custom model metadata, but one loses token IDs and the other loads the wrong custom module; too different to treat as the same bug." + "reason": "DiaConfig losing custom token IDs on save/load and from_pretrained loading the wrong custom module are both persistence bugs but not the same one." } ] }, @@ -9654,202 +9675,202 @@ "issue:43412|issue:45588", "issue:45372|issue:45406" ], - "prepared_review_unit_hash": "86f2e04fa1dcc57158ad5d87d623f6d86e64b4aaedf1b090ed91b5bc0849ec94", + "prepared_review_unit_hash": "adece12eb045bcf399c048a9f1614dbd6683447e2466c8d918e728e6b94015f7", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 28, - "serialized_chars": 35726, - "estimated_input_tokens": 8932, - "estimated_eval_tokens": 18120 + "serialized_chars": 35729, + "estimated_input_tokens": 8933, + "estimated_eval_tokens": 18122 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These soft pairs are overwhelmingly false positives: they share only broad subsystem keywords, not the same underlying bug or change.", - "confidence": 0.98, - "canonical_issue_reason": "No single canonical issue fits well: the items span unrelated tokenizer, generation, device-map, packaging, processor, and training bugs.", + "summary": "This cluster is very heterogeneous: the paired items share only broad Transformers-related wording, not the same underlying bug or fix. No soft pair should be merged; they span unrelated models, tokenizers, generation paths, packaging, and runtime errors.", + "confidence": 0.84, + "canonical_issue_reason": "If a single issue must anchor the cluster, issue 43873 is the most central by activity and inbound references and it covers a broad runtime concern (quantization/offloading). Still, it is not a true representative of the many unrelated bugs in this set.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43873 is the best anchor among the set because it is open, highly discussed, and broadly about quantization/offloading behavior, but it still does not represent the other unrelated issues.", + "best_issue_reason": "Issue 43873 is the best available anchor because it has the most discussion/inbound references and a broad, user-facing runtime impact. It is still only a loose fit, not a duplicate umbrella for the rest.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45290", "right": "issue:46655", "accept": false, - "reason": "Different failure modes in different code paths: chat templating with tool calls vs video processor padding." + "reason": "Assistant chat-template tool-call crash and video processor padding bug are different code paths and different failure modes." }, { "left": "issue:37428", "right": "issue:42175", "accept": false, - "reason": "ImportError in flash-attention utils vs packaging/backend installation issue; unrelated." + "reason": "One is a flash-attention import error; the other is a packaging/backend dependency issue for TensorFlow." }, { "left": "issue:32090", "right": "issue:33290", "accept": false, - "reason": "Trainer broadcast NoneType error and DeepSpeed/Adafactor OOM are distinct bugs." + "reason": "A NoneType broadcast error in Trainer is unrelated to Deepspeed Adafactor OOM." }, { "left": "issue:43873", "right": "issue:45229", "accept": false, - "reason": "Both mention memory/offloading, but one is quantization offloading behavior and the other is Gemma4 multi-GPU CUDA OOM; not the same code path." + "reason": "Quantization/offloading behavior is a different problem from Gemma4 multi-GPU CUDA OOM." }, { "left": "issue:44561", "right": "issue:44779", "accept": false, - "reason": "Removal of a utility API for trust_remote_code models is unrelated to DeepSeek tokenizer correctness." + "reason": "Removal of a utility function breaking trust_remote_code models is unrelated to DeepSeek tokenizer output regressions." }, { "left": "issue:25251", "right": "issue:32090", "accept": false, - "reason": "Pipeline top_k nesting behavior is unrelated to Trainer GPU broadcast failure." + "reason": "Pipeline top_k nesting behavior is unrelated to the GPU broadcast NoneType error." }, { "left": "issue:44448", "right": "issue:44466", "accept": false, - "reason": "One is model output drift between v4 and v5; the other is serialization of tied weights by device." + "reason": "Both involve v5 behavior changes, but one is generation output drift and the other is serialization of tied weights; not the same fix." }, { "left": "issue:43976", "right": "issue:45440", "accept": false, - "reason": "Python version compatibility issue vs divergence in DeepSeekV3MoE implementation." + "reason": "Python version compatibility and DeepSeekV3 implementation divergence are unrelated issues." }, { "left": "issue:42490", "right": "issue:43873", "accept": false, - "reason": "Local-path import warning is unrelated to quantization offloading behavior." + "reason": "Local-path import warning has nothing to do with quantization/offloading behavior." }, { "left": "issue:43278", "right": "issue:43575", "accept": false, - "reason": "Dtype change during evaluate is unrelated to tensor-parallel OOM on loading a specific model." + "reason": "Embedding dtype drift in evaluate and Qwen2 tp OOM are different runtime problems." }, { "left": "issue:45588", "right": "issue:45823", "accept": false, - "reason": "Flash attention crash on sink-less models vs Gemma4 device mismatch in forward; different bugs." + "reason": "Flash-attention sinkless-model crash and Gemma4 PLE device mismatch are distinct forward-pass bugs." }, { "left": "issue:41720", "right": "issue:45229", "accept": false, - "reason": "Both are inference-memory/device issues, but one is auto device mapping assert and the other is multi-GPU OOM; not the same defect." + "reason": "Qwen3 auto device mapping cuda assert and Gemma4 multi-GPU OOM are separate model-specific failures." }, { "left": "issue:44568", "right": "issue:45491", "accept": false, - "reason": "Tokenizer special-token handling is unrelated to Gemma3 NaN embeddings in sliding-window attention." + "reason": "Tokenizer special-token handling and Gemma3 sliding-window NaNs are unrelated." }, { "left": "issue:43381", "right": "issue:43701", "accept": false, - "reason": "Eval-mode gradient checkpointing restriction and resume checkpoint key mismatch are separate training-flow problems." + "reason": "Eval-mode gradient checkpointing restriction and resume checkpoint key mismatch are different training-state issues." }, { "left": "issue:33357", "right": "issue:39692", "accept": false, - "reason": "MacOS bus error with a model load versus documentation example errors; no shared code path." + "reason": "MacOS bus error in CLIP loading and SigLIP2 documentation/example errors do not share the same bug." }, { "left": "issue:44977", "right": "issue:45563", "accept": false, - "reason": "Flash-attention generation bug and stale warning about num_return_sequences are not the same issue." + "reason": "Flash-attention generation failure and a stale warning in paged generate are not the same change." }, { "left": "issue:29127", "right": "issue:34567", "accept": false, - "reason": "LayoutLMv3 error messaging and TrainerState token counter updating are unrelated." + "reason": "LayoutLMv3 error clarity and TrainerState token counter not updating are unrelated." }, { "left": "issue:44871", "right": "issue:44977", "accept": false, - "reason": "EOS token config inconsistency is unrelated to Qwen3.5 flash-attention generation failure." + "reason": "Gemma eos_token_id config mismatch is a different issue from Qwen3.5 flash-attention generation failure." }, { "left": "issue:45198", "right": "issue:45803", "accept": false, - "reason": "Wav2Vec2 save/tokenization failure and a bare-except bug in Fuyu tensor conversion are different components." + "reason": "Wav2Vec2 save/tokenization failure and a bare-except cleanup in FuyuBatchFeature are unrelated." }, { "left": "issue:43653", "right": "issue:44743", "accept": false, - "reason": "BigBirdTokenizer special-token registration and Qwen3 recurrent-state reset are unrelated." + "reason": "BigBirdTokenizer special-token registration and Qwen3_5 recurrent-state reset are different tokenizer/model behaviors." }, { "left": "issue:45372", "right": "issue:45588", "accept": false, - "reason": "Gemma4 processor import failure and flash_attention.py s_aux=None crash are different layers and code paths." + "reason": "Gemma4 processor import failure and flash-attention s_aux=None crash are separate errors in different layers." }, { "left": "issue:44361", "right": "issue:45198", "accept": false, - "reason": "MLukeTokenizer AttributeError is unrelated to Wav2Vec2 save_pretrained/tokenization failure." + "reason": "MLukeTokenizer AttributeError and Wav2Vec2 save/tokenization failure do not indicate the same underlying bug." }, { "left": "issue:44186", "right": "issue:45198", "accept": false, - "reason": "LayoutLMv2 tokenizer crashes on NER/padding are not the same as Wav2Vec2 serialization/tokenization failures." + "reason": "LayoutLMv2 NER/padding crash and Wav2Vec2 save/tokenization fail are unrelated tokenizer issues." }, { "left": "issue:43611", "right": "issue:46121", "accept": false, - "reason": "base_model_prefix loading regression and rope-validation TypeError are unrelated." + "reason": "base_model_prefix loading regression and rope-validation JSON list type error affect different loading/validation paths." }, { "left": "issue:42915", "right": "issue:42994", "accept": false, - "reason": "Both involve quantization, but one is FineGrainedFP8Config load failure and the other is saving quantized models; different problems." + "reason": "Qwen3Moe FP8 config failure and quantized model saving failure are different quantization workflows." }, { "left": "issue:29870", "right": "issue:34689", "accept": false, - "reason": "Bart generate UnboundLocalError and Llama 3.2 Vision model loading breakage are unrelated." + "reason": "BART generate decoder mask bug and Llama 3.2 model loading break are unrelated model-specific regressions." }, { "left": "issue:43412", "right": "issue:45588", "accept": false, - "reason": "Executorch export/training-guard and flash_attention s_aux=None crash are distinct export vs inference bugs." + "reason": "ExecuTorch export problems for Gemma3n and a flash-attention runtime crash are different fixes." }, { "left": "issue:45372", "right": "issue:45406", "accept": false, - "reason": "Processor loading import error and serve-time missing _tokenizer AttributeError are related only by Gemma4, not the same bug." + "reason": "Gemma4 processor import error and transformers serve missing _tokenizer are distinct processor-loading/serving issues." } ] }, @@ -9957,232 +9978,232 @@ "issue:43334|issue:43531", "issue:33290|issue:34689" ], - "prepared_review_unit_hash": "1710490b863d8496560649d156f0d8b350f2a3bf7da7b437ddc1cc5425f083d9", + "prepared_review_unit_hash": "542a67a1ef1a3181ef9a7ce847a71f549446d4720270a522579ec8391ad16085", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 33, - "serialized_chars": 37508, - "estimated_input_tokens": 9377, - "estimated_eval_tokens": 19010 + "serialized_chars": 37514, + "estimated_input_tokens": 9379, + "estimated_eval_tokens": 19014 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is a heterogeneous mix of unrelated Transformer bug reports and regressions. None of the soft pairs look like true duplicates; they mostly share only broad subsystem keywords (tokenizers, vision, memory, FlashAttention, etc.) but describe different code paths and failure modes.", - "confidence": 0.95, - "canonical_issue_reason": "No single true canonical duplicate stands out because the cluster is not cohesive. If one issue must serve as a representative, issue:44462 is the most broadly relevant core-loader regression and has the clearest user-facing impact, but the rest of the set is largely unrelated.", + "summary": "This set is mostly a grab-bag of unrelated Transformers issues; I don\u2019t see any soft pair that looks like the same underlying bug/change. No PRs are present. If one issue must anchor the cluster, the broad and active FA2 regression report is the best fit.", + "confidence": 0.9, + "canonical_issue_reason": "issue:46693 is the most active, broadly scoped open bug report and is the best representative anchor if this cluster must be summarized around one issue.", "canonical_pr_reason": null, - "best_issue_reason": "Issue:44462 is the best representative issue among this set because it targets core AutoTokenizer/model loading behavior with broad impact and a concrete, reproducible bug, unlike the mostly model-specific reports.", + "best_issue_reason": "issue:46693 has the strongest combination of discussion, current relevance, and generalized failure mode compared with the other highly specific reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44206", "right": "issue:44479", "accept": false, - "reason": "Both are regression reports, but they affect different features and model families: LasrFeatureExtractor center-arg handling vs video input regressions in Qwen VL models." + "reason": "Different regressions: unsupported feature-extractor arg vs video input handling for Qwen VL models." }, { "left": "issue:29942", "right": "issue:30064", "accept": false, - "reason": "Flash Attention 2 test failures and void segmentation-map processing are unrelated bugs in different code paths." + "reason": "Unrelated areas: Flash Attention 2 test failures vs segmentation-map processing in image processor." }, { "left": "issue:43756", "right": "issue:45440", "accept": false, - "reason": "Different model implementations and different defects: Smollm3 RoPE-layer count vs DeepseekV3MoE divergence from the remote implementation." + "reason": "Different model families and failure modes; RoPE-layer mismatch is not the same as DeepSeek V3 implementation divergence." }, { "left": "issue:43653", "right": "issue:44442", "accept": false, - "reason": "Tokenizer special-token registration causing empty decode output is unrelated to AutoTokenizer failing to load a specific tokenizer class." + "reason": "Tokenizer special-token registration bug vs AutoTokenizer loading a specific tokenizer file; not the same code path." }, { "left": "issue:45459", "right": "issue:45712", "accept": false, - "reason": "One is tokenizer error masking when protobuf is missing; the other is leftover dummy class cleanup. No shared bug." + "reason": "One is tokenizer error masking without protobuf, the other is leftover dummy classes leaking into module namespace." }, { "left": "issue:45491", "right": "issue:46693", "accept": false, - "reason": "Both mention attention/windowing, but one is a correctness NaN bug in Gemma3 and the other is a performance regression in long-context generation; not the same underlying issue." + "reason": "Both mention FA2/long-context, but one is NaN embeddings in Gemma3 and the other is a performance regression; not the same concrete bug." }, { "left": "issue:43065", "right": "issue:45412", "accept": false, - "reason": "Sam3PixelDecoder dummy Conv2d cleanup and RT-DETR memory not released are separate issues with different symptoms and fixes." + "reason": "Dummy Conv2d in Sam3PixelDecoder is unrelated to RT-DETR memory not being released." }, { "left": "issue:29127", "right": "issue:33357", "accept": false, - "reason": "Improving a LayoutLMv3 error message is unrelated to a MacOS bus error loading a CLIP model." + "reason": "Clarity of an error message in LayoutLMv3 vs a MacOS bus error in CLIP loading are unrelated." }, { "left": "issue:43065", "right": "issue:43994", "accept": false, - "reason": "Dummy layer cleanup in Sam3PixelDecoder is not the same as SigLIP2 producing nonsensical outputs via AutoModel/pipeline." + "reason": "Different models and problems: dummy layer construction vs nonsensical SigLIP2 outputs." }, { "left": "issue:45245", "right": "issue:45563", "accept": false, - "reason": "Category-count overflow and a stale num_return_sequences warning are different bugs with no common root cause." + "reason": "Category-limit runtime error is unrelated to a stale warning in paged generate()." }, { "left": "issue:45372", "right": "issue:45459", "accept": false, - "reason": "A missing ReasoningEffort import in mistral_common is unrelated to hidden tokenizer errors when protobuf is absent." + "reason": "Dependency import breakage for Gemma 4 processor is unrelated to protobuf-based tokenizer error handling." }, { "left": "issue:44265", "right": "issue:45198", "accept": false, - "reason": "torch.export failures from torch_compilable_check and Wav2Vec2 save/tokenization failures are distinct problems in different subsystems." + "reason": "torch.export failure with torch_compilable_check is unrelated to Wav2Vec2 save/tokenization failures." }, { "left": "issue:43575", "right": "issue:43701", "accept": false, - "reason": "TP-induced OOM when loading a Qwen model is unrelated to resume_from_checkpoint key mismatches." + "reason": "TP OOM when loading a Qwen model is unrelated to resume_from_checkpoint key mismatch." }, { "left": "issue:45440", "right": "issue:45593", "accept": false, - "reason": "DeepseekV3MoE parity with remote code and D-FINE auxiliary-loss behavior are separate model-specific changes." + "reason": "DeepSeekV3 implementation divergence is unrelated to auxiliary-loss behavior in D-FINE." }, { "left": "issue:43066", "right": "issue:44462", "accept": false, - "reason": "Tokenizer decoder type mismatch in v5 and ignoring tokenizer.json during AutoTokenizer load are both tokenizer-related, but they are different failure modes and fixes." + "reason": "Wrong tokenizer decoder type in v5 is unrelated to AutoTokenizer ignoring tokenizer.json." }, { "left": "issue:29127", "right": "issue:35141", "accept": false, - "reason": "LayoutLMv3 error clarity and token-embedding resize reinitialization are unrelated issues." + "reason": "Error-message clarity in LayoutLMv3 is unrelated to embedding reinitialization after resize." }, { "left": "issue:42490", "right": "issue:43381", "accept": false, - "reason": "Path-trailing-slash import warning and gradient-checkpointing-in-eval-mode are separate behaviors with different code paths." + "reason": "Local-path import warning and gradient-checkpointing-in-eval are separate behavioral issues." }, { "left": "issue:43994", "right": "issue:45412", "accept": false, - "reason": "SigLIP2 inference correctness and RT-DETR memory release are unrelated model bugs." + "reason": "SigLIP2 inference quality issue is unrelated to RT-DETR memory retention on deletion." }, { "left": "issue:45412", "right": "issue:46369", "accept": false, - "reason": "Both mention memory, but one is a garbage-collection leak in RT-DETR and the other is memory overflow for sam3-litetext; not the same defect." + "reason": "Memory leak on GC and memory overflow when calling sam3-litetext are different failure modes with different triggers." }, { "left": "issue:43421", "right": "issue:45491", "accept": false, - "reason": "Tokenizer post-processor updates at runtime and Gemma3 NaN embeddings are different subsystems and symptoms." + "reason": "Runtime special-token post-processing bug is unrelated to Gemma3 sliding-window NaNs." }, { "left": "issue:43575", "right": "issue:44387", "accept": false, - "reason": "Qwen TP loading OOM and increased CUDA reserved memory under int4 quantization are both memory-related but arise from different mechanisms and likely different fixes." + "reason": "Both are memory-related, but TP OOM and int4 reserved-memory regression are not the same code-path problem." }, { "left": "issue:44442", "right": "issue:45072", "accept": false, - "reason": "AutoTokenizer loading failure and bfloat16 dtype mismatches in SwitchTransformers/TimmWrapperModel are unrelated." + "reason": "Tokenizer loading failure is unrelated to bfloat16 dtype mismatches in inference." }, { "left": "issue:42915", "right": "issue:43931", "accept": false, - "reason": "FineGrainedFP8Config failure in Qwen3Moe and Qwen3-VL weight-shape mismatch are separate model-loading problems." + "reason": "FP8 config failure for Qwen3Moe is unrelated to Qwen3-VL weight-shape mismatch." }, { "left": "issue:43526", "right": "issue:45834", "accept": false, - "reason": "BeitImageProcessorFast label reduction bug and Kosmos2.5 OCR index error are different image-processing failures." + "reason": "Label reduction bug in BeitImageProcessorFast is unrelated to Kosmos2.5 OCR index errors." }, { "left": "issue:43526", "right": "issue:46693", "accept": false, - "reason": "A label-array handling bug in BeitImageProcessorFast is unrelated to FA2 long-context performance regression." + "reason": "Different subsystems and symptoms: image-label reduction vs FA2 long-context performance regression." }, { "left": "issue:43824", "right": "issue:44279", "accept": false, - "reason": "Missing Qwen2_5_VL export and a vague dependency issue are not concrete duplicates of the same underlying bug." + "reason": "Missing Qwen2.5-VL import is not the same as a generic dependency issue." }, { "left": "issue:43526", "right": "issue:44186", "accept": false, - "reason": "Different tokenizer/image-processor bugs: one returns the wrong label shape, the other crashes on NER inputs and padding/truncation." + "reason": "BeitImageProcessorFast label handling and LayoutLMv2Tokenizer NER/padding crashes are unrelated." }, { "left": "issue:27957", "right": "issue:29127", "accept": false, - "reason": "XLMRoberta Flash Attention 2 support and LayoutLMv3 error-message clarity are unrelated." + "reason": "Flash Attention 2 support for XLM-Roberta is unrelated to LayoutLMv3 error-message clarity." }, { "left": "issue:27957", "right": "issue:33290", "accept": false, - "reason": "Flash Attention 2 model support and Adafactor OOM in DeepSpeed are different bugs." + "reason": "FA2 model support and deepspeed adafactor OOM are separate issues." }, { "left": "issue:46121", "right": "issue:46399", "accept": false, - "reason": "JSON-loaded list handling in rope validation and missing decorator registration for apply_rotary_pos_emb are both rope-related, but they are distinct code-path failures." + "reason": "RoPE validation TypeError and Qwen3.5 kernel registration failure are different bugs." }, { "left": "issue:43994", "right": "issue:46369", "accept": false, - "reason": "SigLIP2 incorrect outputs and sam3-litetext memory overflow are unrelated model-specific issues." + "reason": "Different models and failure modes: nonsensical outputs vs memory overflow." }, { "left": "issue:43334", "right": "issue:43531", "accept": false, - "reason": "Qwen3-VL pad_token_id loading failure and Qwen3-MoE sliding_window issues are different configuration/architecture bugs." + "reason": "Qwen3-VL pad_token_id loading error is unrelated to Qwen3-MoE sliding_window behavior." }, { "left": "issue:33290", "right": "issue:34689", "accept": false, - "reason": "Adafactor OOM in DeepSpeed and Llama 3.2 Vision load breakage are separate regression reports with no shared root cause." + "reason": "Optimizer OOM in deepspeed is unrelated to Llama 3.2 Vision model loading regression." } ] }, @@ -10284,20 +10305,20 @@ "issue:43931|issue:45216", "issue:43643|issue:44913" ], - "prepared_review_unit_hash": "d4efb99eed89a0441f77ad95d7dcedc28c1b6af5093452c0abaeb78004729fc2", + "prepared_review_unit_hash": "97d75f214290700cd9651ca03509696005d77115cab15e1eac2783e7edd4ef7d", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 27, - "serialized_chars": 35841, + "serialized_chars": 35844, "estimated_input_tokens": 8961, "estimated_eval_tokens": 18178 }, @@ -10306,184 +10327,181 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a heterogeneous set of Transformer issues with a few plausible related pairs, but the evidence is generally too thin for duplicate merging. The Qwen2.5-VL items look related at the subsystem level, yet `issue:44514` and `issue:45381` do not clearly share the same underlying bug or a single mergeable fix.", - "confidence": 0.61, - "canonical_issue_reason": "issue:44479 is the best representative of the Qwen2.5-VL regression cluster because it describes the broadest concrete video-input regression and anchors the v5.3.0 context most clearly.", + "summary": "The cluster is noisy: the soft pairs only share broad subsystem vocabulary and do not describe the same concrete bug or change, so none should be merged.", + "confidence": 0.93, + "canonical_issue_reason": "issue:43749 is the strongest anchor in this set: it has the most discussion/inbound references and a detailed, high-signal report, making it the best representative issue for the cluster.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44479 is the clearest, most actionable issue in the cluster because it frames the regression at the right level of specificity and covers the broader affected model family.", + "best_issue_reason": "issue:43749 is the most informative and central issue here; if one issue must represent the cluster, it is the clearest choice.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45250", "right": "issue:46640", "accept": false, - "reason": "Both mention Flash Attention 2, but the titles are too generic to establish the same underlying bug." + "reason": "Both mention Flash Attention 2, but one is a generic topic stub and the other is a specific flash-attn bug report; not the same underlying issue." }, { "left": "issue:33290", "right": "issue:35141", "accept": false, - "reason": "Different problems: DeepSpeed/Adafactor OOM versus embedding reinitialization after resizing." + "reason": "OOM with Adafactor/DeepSpeed and embedding reinitialization after resize are unrelated problems." }, { "left": "issue:43761", "right": "issue:45859", "accept": false, - "reason": "Both are model/runtime regressions, but they affect different code paths and model families." + "reason": "CLIPVision hidden_states regression and Qwen3_5MoE CPU offload incompatibility affect different code paths and symptoms." }, { "left": "issue:45588", "right": "issue:45600", "accept": false, - "reason": "Flash-attention sink handling and auto-mapping cleanup are unrelated bugs." + "reason": "A flash_attention AttributeError on s_aux=None is unrelated to auto_mappings referencing removed configs." }, { "left": "issue:43381", "right": "issue:43873", "accept": false, - "reason": "Eval-mode gradient checkpointing and quantization offloading are separate behaviors." + "reason": "Gradient checkpointing in eval mode and quantization offloading are separate behaviors with different failure modes." }, { "left": "issue:44514", "right": "issue:45381", "accept": false, - "reason": "`44514` is about batched `apply_chat_template()` crashing with `padding=False`, while `45381` is about incorrect `vision_position_ids` for Qwen2.5-VL video input; the packet does not establish a shared root cause or a single mergeable fix." + "reason": "Batched apply_chat_template padding behavior and Qwen2.5-VL video position_ids are different bugs." }, { "left": "issue:43646", "right": "issue:44077", "accept": false, - "reason": "Missing fields from `trust_remote_code` config loading is not the same bug as patchtsmixer's post_init restriction." + "reason": "Custom model initialization breakage and a specific patchtsmixer post_init API issue are not the same code-path problem." }, { "left": "issue:43262", "right": "issue:44206", "accept": false, - "reason": "Both touch audio preprocessing, but one is chat-template sampling-rate defaulting and the other is a feature-extractor argument crash." + "reason": "Audio chat template sampling-rate defaults and LaserFeatureExtractor center-arg crashes are unrelated." }, { "left": "issue:44977", "right": "issue:45245", "accept": false, - "reason": "Qwen3.5 flash-attention generation and a categories-limit runtime error are unrelated." + "reason": "Qwen3.5 flash-attention generation and a category-cardinality runtime error are unrelated." }, { "left": "issue:44743", "right": "issue:44811", "accept": false, - "reason": "Recurrent-state reset and Whisper batch_decode `skip_special_tokens` are different model/processor bugs." + "reason": "Recurrent-state reset in modular_qwen3_5 and Whisper batch_decode skip_special_tokens are different components." }, { "left": "issue:44442", "right": "issue:44898", "accept": false, - "reason": "Tokenizer loading and Perceiver image resizing are unrelated." + "reason": "Tokenizer loading failure and Perceiver image-size handling are unrelated model-specific bugs." }, { "left": "issue:42490", "right": "issue:44743", "accept": false, - "reason": "Trailing-slash path warning and recurrent-state caching are unrelated." + "reason": "Local path import warning and Qwen3_5 recurrent-state reset do not share the same bug." }, { "left": "issue:45003", "right": "issue:45440", "accept": false, - "reason": "Unsafe `sys.modules` access and DeepseekV3MoE divergence are different code paths and fixes." + "reason": "Unsafe sys.modules access and DeepseekV3MoE divergence are not the same defect." }, { "left": "issue:43720", "right": "issue:44442", "accept": false, - "reason": "BitNet packed-weight loading and tokenizer loading failure are unrelated." + "reason": "BitNet packed-weight loading and tokenizer loading for FastSpeech2Conformer are unrelated." }, { "left": "issue:42175", "right": "issue:46234", "accept": false, - "reason": "Packaging/backends issue and missing `import_utils` helper are different maintenance bugs." + "reason": "Missing tensorflow in a pip extra and a missing import_utils helper are different packaging/import issues." }, { "left": "issue:27957", "right": "issue:29870", "accept": false, - "reason": "XLMRoberta Flash Attention 2 support and a BART generate `decoder_attention_mask` error are unrelated." + "reason": "XLMRoberta Flash Attention 2 support and BART decoder_attention_mask assignment are unrelated." }, { "left": "issue:43316", "right": "issue:43525", "accept": false, - "reason": "Both are config-related, but they concern different model families and different missing fields." + "reason": "A generic config API discrepancy and Llama4Config missing pad_token_id are not the same underlying bug." }, { "left": "issue:43723", "right": "issue:45479", "accept": false, - "reason": "Tokenizer loading in v5 and single-label classification zero-loss behavior are unrelated." + "reason": "Tokenizer loading in v5 and zero-loss sequence classification are unrelated regressions." }, { "left": "issue:43493", "right": "issue:43827", "accept": false, - "reason": "SigLIP2 implementation parity and docs referencing removed pipeline APIs are unrelated." + "reason": "SigLIP2 JAX/HF divergence and docs still referencing pipeline() are different problem classes." }, { "left": "issue:44077", "right": "issue:44479", "accept": false, - "reason": "Patchtsmixer post_init policy and Qwen video-input regression are unrelated." + "reason": "PatchTSMixer post_init policy and Qwen video-input regression are unrelated." }, { "left": "issue:45325", "right": "issue:46697", "accept": false, - "reason": "Qwen2.5-VL rope/index scaling and Swin2SR padding are different preprocessing bugs." + "reason": "Qwen2.5-VL rope-index scaling and Swin2SR over-padding are different image/video preprocessing bugs." }, { "left": "issue:43873", "right": "issue:44871", "accept": false, - "reason": "Quantization offloading and `eos_token_id` config mismatch are unrelated." + "reason": "Quantization offloading and Gemma-3 eos_token_id inconsistency are unrelated." }, { "left": "issue:44062", "right": "issue:45907", "accept": false, - "reason": "AddedToken argument duplication and `list_repo_files` revision forwarding are unrelated." + "reason": "AddedToken special-arg duplication and list_repo_files revision forwarding are unrelated API bugs." }, { "left": "issue:43761", "right": "issue:44079", - "accept": true, - "reason": "The CLIP hidden_states regression plausibly stems from the `ModelOutput` key-assignment bug when a key was previously `None`." + "accept": false, + "reason": "CLIPVisionModel hidden_states regression and ModelOutput key assignment are different internal behaviors." }, { "left": "issue:43749", "right": "issue:43931", "accept": false, - "reason": "FSDP CPU-efficient loading and Qwen3-VL weight-shape mismatch are distinct loading failures." + "reason": "FSDP CPU RAM efficient loading and Qwen3-VL shape mismatch are unrelated loading failures." }, { "left": "issue:43931", "right": "issue:45216", "accept": false, - "reason": "Shape-mismatch loading of a third-party checkpoint is not the same as a save_pretrained regression in Qwen3.5." + "reason": "Model shape mismatch and incorrect save_pretrained output are different Qwen3-related regressions, not one bug." }, { "left": "issue:43643", "right": "issue:44913", "accept": false, - "reason": "Missing fields from `trust_remote_code` config loading and `rotary_pct` reset on reload are separate config issues." + "reason": "trust_remote_code missing fields and GPTNeoX rotary_pct reload behavior are distinct config-handling issues." } ] }, - "evaluator_result": { - "accept": false, - "feedback": "The cluster-level takeaway is mostly cautious, but one soft-edge acceptance is too speculative for the packet: issue:44514 vs issue:45381. From the provided titles, 44514 is about `apply_chat_template` crashing on batched input with `padding=False`, while 45381 is about incorrect `vision_position_ids` for Qwen2.5-VL video input. The packet does not itself establish a shared root cause or a clearly mergeable single-PR fix, so labeling them as the same bug is stronger than the evidence supports. The summary also overstates this overlap by calling it the \u2018Qwen2.5-VL video-position regression pair\u2019 rather than a more conservative \u2018related Qwen2.5-VL input-handling issues\u2019." - }, - "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 }, @@ -10580,208 +10598,208 @@ "issue:43452|issue:44496", "issue:45803|issue:46365" ], - "prepared_review_unit_hash": "0671690d58808c4765f1db183113aea8c745e0dad6a8c2e7b184ea090ad3b405", + "prepared_review_unit_hash": "f685435ffbbfcc7a0df298f72517b3cea4de3c701ec28139ea307d8e209b712f", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36447, - "estimated_input_tokens": 9112, - "estimated_eval_tokens": 18480 + "serialized_chars": 36449, + "estimated_input_tokens": 9113, + "estimated_eval_tokens": 18482 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is a heterogeneous issue cluster, not a duplicate set. The soft pairs mostly share broad transformer/model-loading vocabulary, but each report points to a different subsystem, model family, or failure mode.", - "confidence": 0.98, - "canonical_issue_reason": null, + "summary": "This cluster is heterogeneous: the paired issues span unrelated loading, generation, multimodal, backend, and distributed-training problems. None of the soft pairs looks like the same underlying bug or change, so I would not merge any of them.", + "confidence": 0.62, + "canonical_issue_reason": "Issue 45698 is the most active open report and the clearest concrete save/load regression in the set, so it is the best single anchor issue here.", "canonical_pr_reason": null, - "best_issue_reason": "No single issue is a good global representative; the items span unrelated bugs in config handling, loading, generation, compilation, training, and backend/import behavior.", + "best_issue_reason": "Issue 45698 has the strongest discussion signal and a well-defined reproduction path, making it the most useful representative issue among these items.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43316", "right": "issue:45759", "accept": false, - "reason": "Both involve model/config APIs, but one is a Gemma3TextConfig discrepancy and the other is unwrapping text_config for Qwen multimodal configs; different code paths and failures." + "reason": "Both are multimodal-config related, but they hit different code paths: Gemma3TextConfig API mismatch vs from_config not unwrapping text_config." }, { "left": "issue:44855", "right": "issue:45137", "accept": false, - "reason": "Completely unrelated: DebertaV2 import parsing on Python 3.13 vs a DeepSpeed ZeRO3 deque error." + "reason": "Unrelated failures: Python 3.13 TorchScript parsing on import vs a DeepSpeed ZeRO3 deque IndexError." }, { "left": "issue:44987", "right": "issue:46178", "accept": false, - "reason": "One is a loading regression for physical-intelligence/fast; the other is a missing tensorflow_text backend mapping." + "reason": "Both are load/import problems, but one is a model-specific v5.1.0 regression and the other is a missing tensorflow_text backend mapping." }, { "left": "issue:43761", "right": "issue:44188", "accept": false, - "reason": "Different bugs: CLIPVisionModel hidden_states regression vs torch.compile attention-kernel branching divergence." + "reason": "CLIP hidden_states regression and torch.compile attention-kernel divergence are different bugs." }, { "left": "issue:44190", "right": "issue:44496", "accept": false, - "reason": "Unrelated failures: local dataset loading in a training script vs unrecognized model/config loading for OLMo." + "reason": "Example script dataset loading failure and unrecognized-model config loading failure are unrelated." }, { "left": "issue:43232", "right": "issue:43388", "accept": false, - "reason": "One is a generation-state issue after sync_gpus; the other is gather_for_metrics dropping tuple label elements." + "reason": "Generation kwargs sync bug vs gather_for_metrics dropping tuple labels; different subsystems and symptoms." }, { "left": "issue:44442", "right": "issue:45803", "accept": false, - "reason": "Different tokenizer/batching issues: FastSpeech2ConformerTokenizer loading vs a bare except swallowing errors in FuyuBatchFeature." + "reason": "Tokenizer class loading failure and bare-except tensor conversion bug are separate issues." }, { "left": "issue:42907", "right": "issue:45698", "accept": false, - "reason": "Both touch save/load flows, but one is about saving dequantized models and the other about loading the wrong custom module after save_pretrained; not the same bug." + "reason": "Both involve save/load behavior, but one is dequantized-model persistence and the other is wrong custom module restoration." }, { "left": "issue:43701", "right": "issue:46506", "accept": false, - "reason": "Key mismatch on resume_from_checkpoint is unrelated to CPU/CUDA output differences in PPDocLayoutV3." + "reason": "Checkpoint resume key mismatch is unrelated to CPU/CUDA output divergence in PPDocLayoutV3." }, { "left": "issue:42617", "right": "issue:43575", "accept": false, - "reason": "3d_parallel.py runtime failure is unrelated to Qwen2-57B tp OOM during model loading." + "reason": "3d_parallel.py failure and Qwen TP OOM are different distributed-loading problems." }, { "left": "issue:43575", "right": "issue:43742", "accept": false, - "reason": "Different model-load failures: Qwen2 tp OOM vs a key error for MobileLLM-125M." + "reason": "OOM on a large TP load and a key error on MobileLLM loading are not the same bug." }, { "left": "issue:27957", "right": "issue:32090", "accept": false, - "reason": "XLMRoberta FlashAttention behavior is unrelated to the Trainer _gpu_broadcast_one NoneType error." + "reason": "Flash Attention 2 support issue and Trainer _gpu_broadcast_one NoneType error are unrelated." }, { "left": "issue:25251", "right": "issue:28282", "accept": false, - "reason": "Pipeline top_k output-shape change is unrelated to missing PyTorch for AutoModel import." + "reason": "Pipeline output-shape bug and missing PyTorch ImportError are entirely different." }, { "left": "issue:43550", "right": "issue:43701", "accept": false, - "reason": "torch.compile/SDPA divergence for Bamba-9B-v2 is unrelated to checkpoint key mismatch." + "reason": "torch.compile SDPA failure and resume_from_checkpoint key mismatch do not share a concrete code path." }, { "left": "issue:44568", "right": "issue:45479", "accept": false, - "reason": "Tokenizer special-token insertion regression is unrelated to degenerate zero-loss classification behavior." + "reason": "Tokenizer special-token regression and degenerate single-label classification loss are different bugs." }, { "left": "issue:46234", "right": "issue:46291", "accept": false, - "reason": "Missing import_utils helper is unrelated to pipelines import failing on kernels version/revision requirements." + "reason": "Missing import_utils helper and kernels v0.15 import requirement are separate optional-dependency issues." }, { "left": "issue:43606", "right": "issue:43701", "accept": false, - "reason": "CPU offload device mismatch for bark-small is unrelated to resume_from_checkpoint key mismatch." + "reason": "CPU offload device mismatch and checkpoint key mismatch are unrelated." }, { "left": "issue:44464", "right": "issue:46421", "accept": false, - "reason": "Both mention chunked generation/compile, but one reports inconsistent outputs and the other spurious recompilation from cache lazy init; different concrete bugs." + "reason": "Both mention chunked generation, but one is output inconsistency under compiled forward while the other is cache-init causing recompilation; not the same failure." }, { "left": "issue:16998", "right": "issue:22355", "accept": false, - "reason": "DeBERTa model_max_length question is unrelated to the missing transformers.onnx module error." + "reason": "Model_max_length question and missing transformers.onnx module are unrelated." }, { "left": "issue:43295", "right": "issue:43825", "accept": false, - "reason": "Different regressions: processor.tokenizer/image passthrough breakage vs pipeline error-message wording about translation tasks." + "reason": "Custom processor/tokenizer regression and pipeline error-message regression are different issues." }, { "left": "issue:43295", "right": "issue:44871", "accept": false, - "reason": "Processor/tokenizer regression is unrelated to Gemma-3 eos_token_id configuration mismatch." + "reason": "Processor/tokenizer image-passing regression and Gemma3 eos_token_id mismatch affect different models and code paths." }, { "left": "issue:44871", "right": "issue:45584", "accept": false, - "reason": "Both involve special tokens, but one is a Gemma-3 eos_token_id config mismatch and the other is Whisper empty-transcription generation failure." + "reason": "Gemma3 eos_token config inconsistency is not the same as Whisper empty-transcription failure after align_special_tokens." }, { "left": "issue:43316", "right": "issue:43716", "accept": false, - "reason": "Gemma3TextConfig API mismatch is unrelated to Mistral-3 image-preprocessor dtype mismatch." + "reason": "Both are multimodal, but one is a config API discrepancy and the other is an image-preprocessor dtype mismatch." }, { "left": "issue:42175", "right": "issue:43316", "accept": false, - "reason": "Packaging/backend installation issue for tensorflow is unrelated to Gemma3TextConfig API behavior." + "reason": "Installation/backend packaging issue vs Gemma3 config API discrepancy." }, { "left": "issue:44077", "right": "issue:44534", "accept": false, - "reason": "Patchtsmixer post_init restriction and non-persistent buffer junk in v5 are different issues." + "reason": "PatchTSMixer post_init policy and non-persistent buffer junk are unrelated v5 issues." }, { "left": "issue:44871", "right": "issue:46050", "accept": false, - "reason": "Gemma-3 eos_token_id inconsistency is unrelated to sinq quantization save/load breakage." + "reason": "EOS-token configuration mismatch is not the same as quantized model save/load corruption." }, { "left": "issue:43646", "right": "issue:44913", "accept": false, - "reason": "Custom model initialization regression and GPTNeoX rotary_pct reload bug are different model-config problems." + "reason": "Broad custom-model initialization breakage and GPTNeoX rotary_pct reload reset are different problems." }, { "left": "issue:43452", "right": "issue:44496", "accept": false, - "reason": "GGUF loading failures for AutoTokenizer/AutoModel are unrelated to the unrecognized-model config.json error." + "reason": "gguf_file loader breakage and missing model_type in config.json are separate model-loading failures." }, { "left": "issue:45803", "right": "issue:46365", "accept": false, - "reason": "Bare except swallowing errors in FuyuBatchFeature is unrelated to Trainer import failure caused by datasets." + "reason": "Bare-except in Fuyu conversion and Trainer import blocked by datasets are unrelated import/runtime issues." } ] }, @@ -10881,184 +10899,184 @@ "issue:44734|issue:45406", "issue:46082|issue:46697" ], - "prepared_review_unit_hash": "7f354d2ae3d31a537c4bfd3d4807c087836e18b1d88f5c94c604e42d1c735d61", + "prepared_review_unit_hash": "b513df637688cf5909dbcd0f558a2a0b1e3e5c8b7efe6b32228be15d2a7b0592", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 25, - "serialized_chars": 35158, - "estimated_input_tokens": 8790, - "estimated_eval_tokens": 17836 + "serialized_chars": 35153, + "estimated_input_tokens": 8789, + "estimated_eval_tokens": 17834 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster looks like a grab-bag of unrelated Transformers issues that only share broad keywords (cache, video, flash-attn, config, processors). None of the proposed soft pairs appear to describe the same underlying bug or change.", - "confidence": 0.9, - "canonical_issue_reason": "issue:45698 is the strongest standalone report in the set: it has the most discussion/inbound references and describes a concrete, reproducible from_pretrained/save_pretrained loading bug.", + "summary": "These soft pairs are mostly unrelated false positives across different models, pipelines, and code paths. I would reject every soft edge; there is no single duplicate bug cluster here.", + "confidence": 0.95, + "canonical_issue_reason": "No single canonical issue: the items span unrelated bugs, models, and subsystems rather than one underlying defect.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45698 is the best representative issue here because it is detailed, actively discussed, and tied to a specific failure mode rather than a vague regression headline.", + "best_issue_reason": "Issue 45698 is the most broadly impactful open bug in the set, but it still does not match the others well enough to serve as a duplicate anchor.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44162", "right": "issue:44792", "accept": false, - "reason": "ESM2 workflow breakage and a Janus image-generation test failure are unrelated model-specific issues." + "reason": "Different bugs: ESM2 breakage vs. Janus image-generation test failure." }, { "left": "issue:45803", "right": "issue:46170", "accept": false, - "reason": "Swallowing KeyboardInterrupt in Fuyu tensor conversion is unrelated to a Windows/ROCm torch.distributed import failure." + "reason": "Unrelated causes: bare except in Fuyu tensor conversion vs Windows ROCm torch.distributed import error." }, { "left": "issue:43404", "right": "issue:43441", "accept": false, - "reason": "Missing lm_head tying in Mistral3 is a different bug from Ministral-3 flash-attention incompatibility." + "reason": "Mistral3 lm_head tying bug and Ministral-3 FlashAttention support failure are different code paths." }, { "left": "issue:45632", "right": "issue:46424", "accept": false, - "reason": "A local-cache path collision for trust_remote_code is unrelated to max_cache_length handling in cache config." + "reason": "Cache-path collision and max_cache_length handling are separate cache issues." }, { "left": "issue:44610", "right": "issue:46369", "accept": false, - "reason": "OmDet-Turbo processor output size mismatch is not the same as Sam3-litetext memory overflow." + "reason": "Processor/model input-size mismatch is unrelated to SAM3-LiteText memory overflow." }, { "left": "issue:43650", "right": "issue:46202", "accept": false, - "reason": "The placeholder text 'ADD THE DATA' is not a duplicate of a Gemma pipeline roundtrip bug." + "reason": "No clear shared bug; one title is non-technical while the other is a Gemma tokenizer roundtrip failure." }, { "left": "issue:43761", "right": "issue:44479", "accept": false, - "reason": "CLIPVisionModel hidden_states regression and Qwen multimodal video regression are different code paths and failures." + "reason": "CLIP hidden_states regression and Qwen video-input regression are different failures." }, { "left": "issue:43572", "right": "issue:46178", "accept": false, - "reason": "Missing pad_token_idx in StableLmConfig is unrelated to the tensorflow_text backend mapping error." + "reason": "Missing pad_token_idx in StableLmConfig is unrelated to tensorflow_text backend mapping." }, { "left": "issue:45542", "right": "issue:46234", "accept": false, - "reason": "A TensorFlow/tensorboard backend import problem is not the same as a missing import_utils helper function." + "reason": "Different import/backend problems: tensorboard-only TF backend vs missing import_utils symbol." }, { "left": "issue:43540", "right": "issue:44514", "accept": false, - "reason": "Qwen3OmniMoe video processing ValueError and Qwen2_5_VL batched chat-template crash are distinct failures." + "reason": "Qwen3OmniMoe video processing error and Qwen2_5_VL batched chat-template crash are distinct." }, { "left": "issue:43065", "right": "issue:44617", "accept": false, - "reason": "A dummy Conv2d concern in Sam3PixelDecoder is not the same as Sam3Video CUDA OOM." + "reason": "Dummy Conv2d in Sam3PixelDecoder is not the same as SAM3Video CUDA OOM." }, { "left": "issue:43504", "right": "issue:45698", "accept": false, - "reason": "Legacy Beit preset loading and wrong custom-module loading after save_pretrained are different persistence/serialization bugs." + "reason": "Legacy field loading in BEiT and wrong custom module loading after save_pretrained are different bugs." }, { "left": "issue:45127", "right": "issue:45200", "accept": false, - "reason": "LoRA merge collapse with extended vocab is unrelated to Gemma4 mm_token_type_ids defaults for text-only fine-tuning." + "reason": "LoRA merge collapse with tied embeddings is unrelated to Gemma4 mm_token_type_ids defaulting." }, { "left": "issue:43388", "right": "issue:43688", "accept": false, - "reason": "Metric-gathering label truncation and auxiliary-loss normalization in MoE models are unrelated." + "reason": "Gather truncation in metric collection and auxiliary-loss normalization are separate issues." }, { "left": "issue:44792", "right": "issue:44977", "accept": false, - "reason": "A Janus image-generation test failure is not the same as Qwen3.5 flash-attention generation failure." + "reason": "Janus test failure and Qwen3.5 FlashAttention generation bug do not share the same root cause." }, { "left": "issue:43352", "right": "issue:46640", "accept": false, - "reason": "Nemotron flash-attn-2 unsupported is not the same as a vague flash-attention issue with no matching concrete code path." + "reason": "Too generic to merge: specific Nemotron FA2 unsupported issue vs a vague Flash Attention 2 report." }, { "left": "issue:44792", "right": "issue:45600", "accept": false, - "reason": "Janus generation test failure and Sam3LiteText auto-mapping CI breakage are unrelated." + "reason": "Janus image-generation failure is unrelated to Sam3LiteText auto-mapping/CI breakage." }, { "left": "issue:43673", "right": "issue:44743", "accept": false, - "reason": "Generic GenerationMixin cache missing in chunked prefill and Qwen3.5 recurrent-state reset are different caching bugs in different implementations." + "reason": "Missing GenerationMixin cache in chunked prefill and recurrent-state reset with cache are different cache bugs." }, { "left": "issue:41950", "right": "issue:43475", "accept": false, - "reason": "Video-classification pipeline processor mismatch is unrelated to Sam3VisionEncoderOutput missing fpn_position_embeddings." + "reason": "Video-classification processor lookup and missing fpn_position_embeddings in SAM3 are unrelated." }, { "left": "issue:43525", "right": "issue:43531", "accept": false, - "reason": "Llama4Config pad_token_id absence and Qwen3-MoE sliding_window behavior are unrelated config issues." + "reason": "Missing pad_token_id in Llama4Config is not the same as the Qwen3-MoE sliding_window issue." }, { "left": "issue:44190", "right": "issue:44964", "accept": false, - "reason": "Local dataset loading in a training script and loading Phi-4-multimodal-instruct are different failure scenarios." + "reason": "Local dataset loading failure and Phi-4 multimodal loading failure are unrelated." }, { "left": "issue:44534", "right": "issue:44913", "accept": false, - "reason": "Non-persistent buffer initialization junk and GPTNeoX rotary_pct reload persistence are unrelated." + "reason": "Non-persistent buffer junk and rotary_pct reload regression are different config/state bugs." }, { "left": "issue:35141", "right": "issue:42490", "accept": false, - "reason": "Token-embedding resize reinitialization is unrelated to import warnings from local paths with trailing slashes." + "reason": "Embedding resize reinitialization and trailing-slash import warning are unrelated loading behaviors." }, { "left": "issue:44734", "right": "issue:45406", "accept": false, - "reason": "KV-cache tensor indexing in serve is a different bug from Gemma4Processor missing _tokenizer." + "reason": "KV-cache continuation tensor indexing and Gemma4Processor missing _tokenizer are different serve failures." }, { "left": "issue:46082", "right": "issue:46697", "accept": false, - "reason": "LlamaConfig head_dim validation and Swin2SR over-padding are unrelated configuration/processor bugs." + "reason": "LlamaConfig head_dim validation is unrelated to Swin2SR over-padding." } ] }, @@ -11160,14 +11178,14 @@ "issue:44617|issue:46369", "issue:44779|issue:46489" ], - "prepared_review_unit_hash": "97ebe4ee0e8dfde78878cf683f1b69d699f5206641e5cac7ffe86363a3a1cf1e", + "prepared_review_unit_hash": "25d6542070ee4375c0e6afd4e70c1237556d205f60a2626edb97a9343d93da60", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 47, @@ -11182,180 +11200,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly a set of unrelated issue reports that only share broad subsystem words (generation, cache, vision, tokenizer, CI). None of the soft pairs are strong enough to merge as duplicates under a strict triage standard.", - "confidence": 0.86, - "canonical_issue_reason": "Issue 46489 is the strongest standalone report: it names a concrete model family, a clear v5 regression boundary, and points to a specific gap in a prior fix, making it the most actionable and diagnostically precise issue in the set.", + "summary": "The cluster is heterogeneous: most items are distinct model-, processor-, config-, or generation-path bugs, and none of the soft-edge pairs look like the same underlying change closely enough to merge.", + "confidence": 0.95, + "canonical_issue_reason": "issue:46489 is the most detailed and actionable report in the set, with a concrete DeepSeek-Coder v1 v5 regression and an explicit link to a gap in a prior fix, making it the strongest single anchor even though the cluster is not truly duplicate-heavy.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 46489 is the best overall issue because it is specific, regression-focused, and directly actionable for maintainers, unlike several broader or more ambiguous reports.", + "best_issue_reason": "issue:46489 is the best standalone issue to keep as a representative because it has the clearest root-cause framing and reproduction context.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45632", "right": "issue:46617", "accept": false, - "reason": "Both are trust_remote_code/cache-related, but one is a leaf-directory cache collision and the other is a FileNotFoundError loading from cache; related area, not the same bug." + "reason": "Both involve trust_remote_code/cache, but one is a leaf-directory collision for local models while the other is a FileNotFoundError during cache load; different failure modes." }, { "left": "issue:43931", "right": "issue:44060", "accept": false, - "reason": "Different Qwen model families and different failures: weight-shape mismatch versus an incorrect tied-weights warning." + "reason": "Different model families and symptoms: load-time weight-shape mismatch versus a tied-weights warning bug in Qwen3-Next." }, { "left": "issue:44855", "right": "issue:45588", "accept": false, - "reason": "Unrelated code paths: Python 3.13 TorchScript parsing error versus flash-attention crash on s_aux=None." + "reason": "Unrelated problems: Python 3.13/JIT parsing import error versus a flash-attention runtime crash on sink-less models." }, { "left": "issue:43827", "right": "issue:43994", "accept": false, - "reason": "One is a docs stale-reference problem; the other is an actual model/pipeline output correctness issue." + "reason": "One is stale docs still mentioning pipeline(), the other is a SigLIP2 model/pipeline correctness issue; not the same bug." }, { "left": "issue:43673", "right": "issue:46421", "accept": false, - "reason": "Both mention chunked prefill/cache, but one is a missing GenerationMixin cache regression and the other is recompilation from lazy cache initialization; not clearly the same defect." + "reason": "Both touch chunked prefill/cache, but one is missing cache in v5.0.0 and the other is lazy-init causing recompilation; distinct code-path issues." }, { "left": "issue:43761", "right": "issue:44242", "accept": false, - "reason": "Different model behaviors and symptoms: CLIP hidden states regression versus missing load-balancing loss computation." + "reason": "Different subsystems: hidden-states propagation in CLIPVisionModel versus omission of load-balancing loss when router logits are disabled." }, { "left": "issue:29870", "right": "issue:35141", "accept": false, - "reason": "Bart generate() UnboundLocalError and token-embedding resize/post_init reinitialization are separate bugs." + "reason": "Both are generation/model-init regressions, but one is an unbound decoder_attention_mask path and the other is embedding reinitialization after resize; not the same fix." }, { "left": "issue:44617", "right": "issue:45584", "accept": false, - "reason": "Sam3Video OOM and Whisper empty-transcription failure are different models, different processors, and different failure modes." + "reason": "CUDA OOM in Sam3Video is unrelated to Whisper empty-transcription handling after align_special_tokens." }, { "left": "issue:45563", "right": "issue:45584", "accept": false, - "reason": "A stale warning in paged generate() is unrelated to Whisper failing after align_special_tokens." + "reason": "A stale warning in paged generate() is a different bug from Whisper failing on empty transcription." }, { "left": "issue:43644", "right": "issue:44913", "accept": false, - "reason": "Non-persistent buffer junk fill and GPTNeoX rotary_pct reload behavior are unrelated state/config issues." + "reason": "A broad non-persistent-buffer initialization regression is not the same as GPTNeoXConfig rotary_pct losing its value on reload." }, { "left": "issue:44933", "right": "issue:45362", "accept": false, - "reason": "Missing image_utils import and Qwen3.5 chat crash are unrelated." + "reason": "Importing a nonexistent image_utils symbol and Qwen3.5 chat crashing are unrelated." }, { "left": "issue:25251", "right": "issue:27957", "accept": false, - "reason": "Pipeline top_k output-shape bug and XLMRoberta Flash Attention 2 issue are unrelated." + "reason": "Pipeline top_k nesting behavior and XLMRoberta Flash Attention 2 support are unrelated issues." }, { "left": "issue:43335", "right": "issue:44855", "accept": false, - "reason": "SwitchTransformers sparse-layer config bug is unrelated to the Python 3.13 TorchScript parsing issue." + "reason": "A sparse-layer config bug and a Python 3.13 decorator/comment parsing failure are different problems." }, { "left": "issue:44479", "right": "issue:44734", "accept": false, - "reason": "Video-input regression in Qwen VL models and KV-cache continuation crash in serve are different code paths." + "reason": "Both are generation-related, but one is a Qwen VL video-input regression and the other is a server KV-cache indexing crash." }, { "left": "issue:45381", "right": "issue:46369", "accept": false, - "reason": "Both involve video/Sam-like model families, but one is Qwen VL vision-position IDs and the other is Sam3-LiteText memory overflow." + "reason": "Different model families and failures: Qwen2.5-VL video position IDs versus Sam3-LiteText memory overflow." }, { "left": "issue:44077", "right": "issue:44877", "accept": false, - "reason": "Optional post_init enforcement and strict config loading for granite_speech are different configuration-policy issues." + "reason": "Both are config-related, but optional post_init in patchtsmixer and strict granite_speech config loading are not the same change." }, { "left": "issue:43550", "right": "issue:44190", "accept": false, - "reason": "torch.compile/SDPA failure in Bamba is unrelated to the local-dataset loading problem in the training script." + "reason": "A torch.compile+SDPA model failure is unrelated to loading a local dataset in a training script." }, { "left": "issue:41720", "right": "issue:45245", "accept": false, - "reason": "Qwen3 auto device mapping CUDA assert and category-cardinality RuntimeError are unrelated." + "reason": "Qwen3 auto device mapping cudaErrorAssert and the 2^24 categories runtime error are unrelated." }, { "left": "issue:43335", "right": "issue:44190", "accept": false, - "reason": "Sparse-layer creation in SwitchTransformers and local dataset loading failure are unrelated." + "reason": "A SwitchTransformers sparse-layer config bug is unrelated to the dataset-loading script failure." }, { "left": "issue:43065", "right": "issue:44610", "accept": false, - "reason": "A dummy Conv2d in Sam3PixelDecoder and an OmDet-Turbo processor/model input-size mismatch are separate problems." + "reason": "Sam3PixelDecoder dummy Conv2d and OmDet-Turbo processor size mismatch are different model/processor issues." }, { "left": "issue:44464", "right": "issue:45563", "accept": false, - "reason": "Compiled forward chunked-generation inconsistency and stale num_return_sequences warning are unrelated generation issues." + "reason": "Compiled-forward chunked-generation inconsistency and a stale warning about num_return_sequences are distinct bugs." }, { "left": "issue:45127", "right": "issue:45357", "accept": false, - "reason": "LoRA merge collapse with extended vocabulary and incorrect visual encoder keys in save_pretrained are different save/weight-tie bugs." + "reason": "LoRA merge collapse with extended vocab and incorrect visual encoder keys in save_pretrained are separate problems." }, { "left": "issue:39692", "right": "issue:44190", "accept": false, - "reason": "SigLIP2 docs/model-processor mismatch and local dataset loading are unrelated." + "reason": "SigLIP2 documentation errors and local dataset loading in a training script are unrelated." }, { "left": "issue:45381", "right": "issue:46655", "accept": false, - "reason": "Both are video-processor related, but the reported bugs are different models and different preprocessing mistakes." + "reason": "Both are video-related, but wrong vision_position_ids and incorrect padding calculation are different processor bugs." }, { "left": "issue:44112", "right": "issue:45600", "accept": false, - "reason": "GraniteSpeech CI stale device override test and Sam3LiteText auto_mappings cleanup are unrelated CI issues." + "reason": "CI failures in GraniteSpeech and Sam3LiteText auto_mappings breakage have different root causes." }, { "left": "issue:43526", "right": "issue:46697", "accept": false, - "reason": "BeitImageProcessorFast label reduction and Swin2SR over-padding are different image-processor bugs." + "reason": "One image processor drops labels, the other over-pads images; same subsystem, different bug." }, { "left": "issue:44617", "right": "issue:46369", "accept": false, - "reason": "Same broader Sam3 family, but one is a video-model OOM and the other is a LiteText memory overflow; not enough to call them the same bug." + "reason": "Same broad SAM3 family, but CUDA OOM in Sam3Video and memory overflow in Sam3-LiteText are not the same underlying issue." }, { "left": "issue:44779", "right": "issue:46489", "accept": false, - "reason": "Both concern DeepSeek tokenizer regressions in v5+, but 44779 is a broad incorrect-output report while 46489 is a specific DeepSeek-Coder v1 gap in PR #44801's fix; related, but not clearly the same underlying bug." + "reason": "Both are DeepSeek tokenizer regressions, but 46489 is a specific DeepSeek-Coder v1 follow-up to an earlier fix gap, not the same report." } ] }, @@ -11458,14 +11476,14 @@ "issue:43479|issue:45375", "issue:41950|issue:44016" ], - "prepared_review_unit_hash": "e50930abb1c3bdde0801d3eab660532b2df3fb5504a8eae9075db329176fb45d", + "prepared_review_unit_hash": "44b0314d34ebac571b68863b70fb068261c56a306d5273c3383aa31192543e3b", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, @@ -11480,180 +11498,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab bag of unrelated Transformers issues. The only loose thematic area is multimodal/chat-template regressions, but the soft pairs are different concrete bugs rather than true duplicates.", - "confidence": 0.89, - "canonical_issue_reason": "issue:44521 is the strongest representative of the only coherent subgroup here: it is an active, well-discussed apply_chat_template multimodal bug with inbound references.", + "summary": "This is a loose cluster of mostly unrelated Transformers issues. The only mild theme is chat-template/multimodal edge cases, but the paired reports still describe different failures, so I rejected all soft merges.", + "confidence": 0.39, + "canonical_issue_reason": "issue:44521 is the strongest representative: it is a clearly described, actively discussed multimodal `apply_chat_template` bug with multiple inbound references.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44521 is the best overall issue candidate because it is open, broadly scoped within a recurring API area, and has the most evidence of user impact/interest in this set.", + "best_issue_reason": "issue:44521 is the best single issue to anchor the set because it has the broadest relevance and clearest reproduction among the nearby chat-template reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45290", "right": "issue:46752", "accept": false, - "reason": "Both involve apply_chat_template, but one is tool-call/no-content handling and the other is empty-list input. Different failure modes and likely separate fixes." + "reason": "Both are `apply_chat_template` crashes, but one is about tool-call assistant messages with no content and the other is an empty-list `IndexError`; different edge cases." }, { "left": "issue:43295", "right": "issue:45276", "accept": false, - "reason": "These are unrelated regressions: one is processor/tokenizer access in multimodal custom code, the other is resize_token_embeddings not propagating to tied output embeddings." + "reason": "These hit different code paths: custom model code accessing `processor.tokenizer`/image passing vs `resize_token_embeddings` not updating Gemma4 output embeddings." }, { "left": "issue:44737", "right": "issue:45803", "accept": false, - "reason": "Different code paths and bugs: XLNet CPU tensor allocation vs a bare except swallowing errors in FuyuBatchFeature." + "reason": "One is an XLNet device placement bug in `torch.arange`; the other is a bare-`except` swallowing errors in Fuyu tensor conversion." }, { "left": "issue:43540", "right": "issue:44734", "accept": false, - "reason": "Both are multimodal/video-related, but the reported crashes are in different subsystems and different code paths." + "reason": "Both mention video/generation areas, but one is a Qwen3OmniMoe video-input validation error and the other is a server KV-cache indexing bug." }, { "left": "issue:44938", "right": "issue:45542", "accept": false, - "reason": "Both are import/backend problems, but one is Python 3.14 import failure and the other is missing TensorFlow backend wiring from a pip extra." + "reason": "Different runtime/import failures: Python 3.14 load failure vs tensorboard installed without TensorFlow causing a backend error." }, { "left": "issue:45276", "right": "issue:45584", "accept": false, - "reason": "Different bugs: embedding resize propagation vs Whisper generation failing on empty transcription after align_special_tokens." + "reason": "One is a Gemma4 embedding-resize regression; the other is Whisper failing on empty transcription after token alignment." }, { "left": "issue:44521", "right": "issue:45381", "accept": false, - "reason": "Both are multimodal, but one is assistant mask generation in apply_chat_template and the other is wrong vision_position_ids in Qwen2.5-VL video inputs." + "reason": "Both involve multimodal chat/video behavior, but one is zeroed assistant masks and the other is wrong `vision_position_ids` for qwen2.5-vl video inputs." }, { "left": "issue:25251", "right": "issue:29870", "accept": false, - "reason": "Unrelated issues: pipeline top_k nesting behavior versus a Bart generate decoder_attention_mask UnboundLocalError." + "reason": "Different bugs in different paths: pipeline `top_k` nesting vs Bart `generate()` using `decoder_attention_mask` before assignment." }, { "left": "issue:44462", "right": "issue:45907", "accept": false, - "reason": "Different concerns: AutoTokenizer ignoring tokenizer.json versus list_repo_files not forwarding revision." + "reason": "One is `AutoTokenizer` ignoring `tokenizer.json`; the other is `list_repo_files` not forwarding `revision`." }, { "left": "issue:45632", "right": "issue:46234", "accept": false, - "reason": "One is a trust_remote_code cache-path collision; the other is a missing import_utils symbol. Not the same bug." + "reason": "A local trust-remote-code cache collision is unrelated to a missing `is_falcon_mamba_ssm_available` import utility." }, { "left": "issue:44246", "right": "issue:45003", "accept": false, - "reason": "Both touch import internals, but one is import slowness and the other is unsafe sys.modules access. Different root causes." + "reason": "Import slowness during `import transformers` is not the same as unsafe `sys.modules[]` access in `modeling_utils`." }, { "left": "issue:43854", "right": "issue:44964", "accept": false, - "reason": "Different model-loading failures for different model families; not the same underlying issue." + "reason": "Distinct model-loading problems: GLM-4.7-Flash test failure vs Phi-4 multimodal load failure on latest transformers." }, { "left": "issue:44964", "right": "issue:45600", "accept": false, - "reason": "Different problems: model loading for Phi-4 multimodal versus stale auto_mappings references breaking CI." + "reason": "One is a model load regression; the other is `auto_mappings.py` referencing removed Sam3LiteText configs and breaking CI." }, { "left": "issue:43643", "right": "issue:44496", "accept": false, - "reason": "One is missing fields when using trust_remote_code=True; the other is an unrecognized model lacking model_type. Related area, different bug." + "reason": "`trust_remote_code=True` dropping fields from `AutoConfig` is different from an unrecognized model missing `model_type`." }, { "left": "issue:44112", "right": "issue:44964", "accept": false, - "reason": "A CI test failure in GraniteSpeech is unrelated to a Phi-4 multimodal loading regression." + "reason": "A stale CI override test failure is unrelated to a model loading error for Phi-4 multimodal." }, { "left": "issue:27957", "right": "issue:35141", "accept": false, - "reason": "XLMRoberta Flash Attention 2 support and embedding reinitialization after resize are separate issues." + "reason": "XLMRoberta Flash Attention support and token-embedding resize/post_init reinitialization are unrelated issues." }, { "left": "issue:46082", "right": "issue:46121", "accept": false, - "reason": "LlamaConfig head_dim validation and rope-parameter JSON list handling are distinct validation bugs." + "reason": "Different validation paths: LlamaConfig head_dim rejection vs rope-parameter conversion failing on a JSON-loaded list." }, { "left": "issue:42175", "right": "issue:44279", "accept": false, - "reason": "A missing TensorFlow backend in a torch extra install is not the same as a generic dependency issue with transformers." + "reason": "One is a packaging/backend extra issue (`.[torch]` missing TensorFlow backend support); the other is a generic dependency complaint." }, { "left": "issue:43404", "right": "issue:45216", "accept": false, - "reason": "Different generation/checkpoint bugs for different models and APIs; no shared concrete code-path problem." + "reason": "Mistral3 tied-weights behavior and Qwen3.5 checkpoint save corruption are separate model-specific regressions." }, { "left": "issue:45245", "right": "issue:45584", "accept": false, - "reason": "Category-cardinality limit error and Whisper empty-transcription failure are unrelated." + "reason": "Category-cardinality limits and Whisper empty-transcription handling are unrelated failures." }, { "left": "issue:42796", "right": "issue:44521", "accept": false, - "reason": "Tokenizer BPE-marker handling and chat-template assistant-mask generation are different features with different bugs." + "reason": "Tokenizer BPE-marker handling is a tokenizer bug, not the multimodal assistant-mask issue in `apply_chat_template`." }, { "left": "issue:36246", "right": "issue:41628", "accept": false, - "reason": "Both are import errors, but they concern different symbols and different export paths." + "reason": "Both are import problems around image processors, but one is a specific missing Qwen2.5-VL symbol and the other is a top-level `AutoImageProcessor` import failure." }, { "left": "issue:44186", "right": "issue:44734", "accept": false, - "reason": "LayoutLMv2Tokenizer crashes and transformers serve KV-cache indexing are unrelated code paths." + "reason": "LayoutLMv2 tokenizer padding/NER crashes are not the same as server KV-cache continuation indexing." }, { "left": "issue:43650", "right": "issue:44220", "accept": false, - "reason": "A vague data issue and a torch_extract_fbank_features bug are not the same underlying change." + "reason": "The titles and repros are unrelated; one is non-technical noise, the other is `_torch_extract_fbank_features()`." }, { "left": "issue:36246", "right": "issue:38617", "accept": false, - "reason": "Different missing-import failures for different symbols; not a single concrete bug." + "reason": "Different missing imports from different modules; no shared code path." }, { "left": "issue:32090", "right": "issue:35141", "accept": false, - "reason": "Trainer GPU broadcast NoneType and embedding reinitialization after resize are unrelated." + "reason": "Trainer GPU broadcast of `None` is unrelated to embedding resize/post-init reinitialization." }, { "left": "issue:43479", "right": "issue:45375", "accept": false, - "reason": "One is default config initialization with None; the other is a missing deepstack_visual_indexes field dropped by strict validation." + "reason": "Different config bugs in different models: default vision/audio config initialization vs missing `deepstack_visual_indexes`." }, { "left": "issue:41950", "right": "issue:44016", "accept": false, - "reason": "Video-classification pipeline behavior and a notebook syntax error are completely unrelated." + "reason": "A video-classification pipeline processor lookup bug is unrelated to a notebook syntax error." } ] }, @@ -11759,220 +11777,226 @@ "issue:43295|issue:44779", "issue:43976|issue:45405" ], - "prepared_review_unit_hash": "3a92cae1e48e88d99b61f21804f7cd6170e554676c473c299c9ae8bc0463cae0", + "prepared_review_unit_hash": "e2ef941e3b21b94cc03e2cd8eb3edd3628c06980a09fa70017f4919f376c9170", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 32, - "serialized_chars": 36636, - "estimated_input_tokens": 9159, - "estimated_eval_tokens": 18574 + "serialized_chars": 36645, + "estimated_input_tokens": 9162, + "estimated_eval_tokens": 18580 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a loose semantic group of unrelated Transformers issues. The items share broad themes like multimodal APIs, model loading, cache/version regressions, and docs, but they do not describe the same concrete bug/change, so none of the soft pairs should be merged as duplicates.", - "confidence": 0.89, - "canonical_issue_reason": "issue:44521 is the strongest representative of the cluster\u2019s multimodal API regressions: it is open, has the most inbound references/activity among the clearly user-facing multimodal issues, and describes a concrete, reproducible failure in `apply_chat_template`.", + "summary": "The set is a heterogeneous mix of unrelated Transformers issues across different models, pipelines, and utilities. None of the soft edges look like true duplicates or the same fixable code-path, so I would not merge any of them into a single canonical artifact.", + "confidence": 0.97, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:44521 is the best triage anchor because it is specific, impactful, and actively discussed; it is a clearer representative than the more generic docs/version issues.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44295", "right": "issue:46506", "accept": false, - "reason": "Different bugs: `position_ids` buffer access vs CPU/CUDA output divergence in PPDocLayoutV3." + "reason": "Different bugs: position_ids buffer access vs CPU/CUDA output divergence in PPDocLayoutV3." }, { "left": "issue:43531", "right": "issue:45600", "accept": false, - "reason": "Different code paths: Qwen3-MoE `sliding_window` behavior vs stale `auto_mappings.py` config references." + "reason": "Qwen3-MoE sliding_window behavior is unrelated to stale auto_mappings references for Sam3LiteText." }, { "left": "issue:43673", "right": "issue:46424", "accept": false, - "reason": "Both mention cache, but one is missing generation cache in chunked prefill while the other is cache-length config handling." + "reason": "Cache API regression during chunked_prefill is distinct from max_cache_length being honored only for Quantized Cache." }, { "left": "issue:44016", "right": "issue:44220", "accept": false, - "reason": "Notebook syntax error and audio feature extraction error are unrelated." + "reason": "Notebook syntax error and _torch_extract_fbank_features are unrelated failures." }, { "left": "issue:43976", "right": "issue:44987", "accept": false, - "reason": "Generic Python/version support issue vs a specific model load failure; not the same bug." + "reason": "Python-version support issue is unrelated to loading physical-intelligence/fast." }, { "left": "issue:44683", "right": "issue:45446", "accept": false, - "reason": "Both involve flex_attention/PyTorch, but they fail for different reasons and at different code paths." + "reason": "Both touch flex_attention, but one is a torch>=2.9 compile failure and the other is a bad version gate for AuxRequest import." }, { "left": "issue:43295", "right": "issue:45335", "accept": false, - "reason": "Multimodal processor/tokenizer regression vs token embedding resize not updating decoder embeddings." + "reason": "Different model families and code paths: processor/tokenizer regression vs t5gemma embedding resize bug." }, { "left": "issue:45083", "right": "issue:46506", "accept": false, - "reason": "Different functionality: feature-length helper in qwen3_omni_moe vs PPDocLayoutV3 CPU/CUDA mismatch." + "reason": "Helper-length logic in qwen3_omni_moe is unrelated to PPDocLayoutV3 CPU/CUDA discrepancies." }, { "left": "issue:43526", "right": "issue:43901", "accept": false, - "reason": "Image processor label reduction bug vs pipeline docs mismatch; only loosely related by being docs/API adjacent." + "reason": "BeitImageProcessorFast label reduction bug is unrelated to TextClassificationPipeline docs mismatch." }, { "left": "issue:43381", "right": "issue:45593", "accept": false, - "reason": "Gradient checkpointing in eval mode is unrelated to D-FINE auxiliary losses." + "reason": "Gradient checkpointing in eval mode and D-FINE auxiliary-loss behavior are different issues." }, { "left": "issue:42907", "right": "issue:43441", "accept": false, - "reason": "Both involve Ministral/Devstral, but one is save/load dequantized weights and the other is FlashAttention failure." + "reason": "Dequantized save failure for Ministral/Devstral is not the same as a flash-attention load/runtime bug." }, { "left": "issue:43901", "right": "issue:44496", "accept": false, - "reason": "Pipeline docs wording issue vs model config/import failure; unrelated." + "reason": "A docs/behavior mismatch in a pipeline is unrelated to an unrecognized model config." }, { "left": "issue:44496", "right": "issue:45042", "accept": false, - "reason": "Unrecognized model config vs unnecessary torchvision requirement in PIL image processors." + "reason": "Unrecognized model loading and PIL image processor torchvision requirements are separate problems." }, { "left": "issue:43976", "right": "issue:44279", "accept": false, - "reason": "Python support/versioning issue vs unspecified dependency issue; too different." + "reason": "General Python support incompatibility is not the same as a dependency issue with transformers." }, { "left": "issue:43673", "right": "issue:44336", "accept": false, - "reason": "Cache behavior in generation vs ANSI output in loading_report; unrelated." + "reason": "Generation cache handling and ANSI output in loading_report are unrelated." }, { "left": "issue:43441", "right": "issue:45588", "accept": false, - "reason": "Both touch FlashAttention, but one is a model-specific failure and the other is an `s_aux=None` crash in sink-less models." + "reason": "Both involve flash_attention, but they affect different failures/models and would not merge into one fix." }, { "left": "issue:44496", "right": "issue:45759", "accept": false, - "reason": "Model not recognized from config vs `from_config` not unwrapping `text_config`; different loading bugs." + "reason": "Unrecognized model config and text_config unwrapping for composite multimodal configs are different loading paths." }, { "left": "issue:44016", "right": "issue:44062", "accept": false, - "reason": "Notebook syntax error vs token insertion constructor error; unrelated." + "reason": "Notebook syntax error is unrelated to AddedToken special-argument duplication." }, { "left": "issue:44792", "right": "issue:46424", "accept": false, - "reason": "Janus image-generation test failure vs cache-length config behavior; unrelated." + "reason": "Janus image-generation test failure is unrelated to cache max length handling." }, { "left": "issue:43824", "right": "issue:44933", "accept": false, - "reason": "Different import failures in different modules/symbols, not the same missing export." + "reason": "Missing Qwen2_5_VLForConditionalGeneration import is unrelated to a nonexistent image_utils import." }, { "left": "issue:43299", "right": "issue:45216", "accept": false, - "reason": "Qwen3VL MoE model loading regression vs Qwen3.5 save_pretrained checkpoint regression." + "reason": "Both are Qwen regressions, but one breaks model loading and the other breaks save_pretrained checkpoint saving." }, { "left": "issue:44327", "right": "issue:46752", "accept": false, - "reason": "QA span decoding top-k bounds error vs empty-list `apply_chat_template` crash." + "reason": "QA decode_spans top-k bounds error is unrelated to apply_chat_template on an empty list." }, { "left": "issue:46123", "right": "issue:46369", "accept": false, - "reason": "Partial-batch pipeline bookkeeping bug vs memory overflow in sam3-litetext." + "reason": "Final partial batch dropping results and Sam3-LiteText memory overflow are different pipeline issues." }, { "left": "issue:44279", "right": "issue:46515", "accept": false, - "reason": "Generic dependency issue vs torch 2.4.1 incompatibility; not the same defect." + "reason": "A vague dependency issue is not the same as torch 2.4.1 incompatibility." }, { "left": "issue:44521", "right": "issue:46693", "accept": false, - "reason": "Both are multimodal/transformer-adjacent, but one is an assistant mask correctness bug and the other is a FA2 performance regression." + "reason": "Multimodal assistant mask generation and FA2 long-context performance regression are unrelated." }, { "left": "issue:45335", "right": "issue:45584", "accept": false, - "reason": "Tokenizer embedding resize bug vs Whisper empty-transcription generation failure." + "reason": "Token embedding resize behavior in t5gemma is unrelated to Whisper empty-transcription generation." }, { "left": "issue:46082", "right": "issue:46399", "accept": false, - "reason": "Llama config validation vs missing decorator on `apply_rotary_pos_emb`; different subsystems." + "reason": "LlamaConfig head_dim validation and Qwen3.5 kernel registration failure are different code paths." }, { "left": "issue:44279", "right": "issue:45405", "accept": false, - "reason": "Dependency problem vs PEFT minimum-version bump; not duplicates." + "reason": "General dependency complaints do not match the specific unreleased MIN_PEFT_VERSION bump." }, { "left": "issue:43577", "right": "issue:45216", "accept": false, - "reason": "Blip2 dtype loading issue vs Qwen3.5 checkpoint save regression." + "reason": "Blip2 dtype propagation and Qwen3.5 checkpoint saving are unrelated regressions." }, { "left": "issue:39692", "right": "issue:43901", "accept": false, - "reason": "SigLIP2 docs example errors vs TextClassificationPipeline docs mismatch; both docs-related but different content." + "reason": "SigLIP2 doc-example errors are unrelated to TextClassificationPipeline return_all_scores docs." }, { "left": "issue:43295", "right": "issue:44779", "accept": false, - "reason": "Processor/tokenizer multimodal regression vs Deepseek tokenizer output change; different model families and failure modes." + "reason": "Processor/tokenizer regression for multimodal inputs is not the same as Deepseek tokenizer output changes." + }, + { + "left": "issue:43976", + "right": "issue:45405", + "accept": false, + "reason": "Python-version compatibility and PEFT minimum-version bump are unrelated dependency issues." } ] }, @@ -12072,190 +12096,190 @@ "issue:44261|issue:44263", "issue:44623|issue:44749" ], - "prepared_review_unit_hash": "1435535645c07be7802b5fd574a24a7d0b6ecc7798a2dca1a82adf93ab85ec60", + "prepared_review_unit_hash": "6c3acb213129c96977e50a5ccf8618e7b4e7b0a3cbff280def924f24bbf8e44c", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 26, - "serialized_chars": 35016, - "estimated_input_tokens": 8754, - "estimated_eval_tokens": 17764 + "serialized_chars": 35021, + "estimated_input_tokens": 8756, + "estimated_eval_tokens": 17768 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is mostly a heterogeneous mix of unrelated Transformer bugs, with no strong duplicate cluster emerging. The soft-similarity pairs share broad themes (tokenizers, loading, multimodal configs, deepspeed, import errors) but not the same concrete failure path.", - "confidence": 0.16, - "canonical_issue_reason": "issue:45812 is the clearest standalone bug report here: a concrete, high-impact regression with a specific symptom (wrong token IDs), a specific subsystem (AutoTokenizer), and strong user-facing impact.", + "summary": "This cluster is heterogeneous and does not resolve into a single duplicate family. The soft pairs are mostly topical overlaps (tokenizers, configs, deepspeed, multimodal, serialization) but each points to a different concrete bug or regression, so none should be merged as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:45812 is the best representative issue to keep as a canonical example because it is specific, reproducible, and well-scoped compared with the more generic or ambiguous reports in the set.", + "best_issue_reason": "issue:44261 is the cleanest concrete bug report with a focused code path and clear symptom, but it should only serve as a loose representative anchor; the cluster as a whole is not a duplicate set.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44075", "right": "issue:44368", "accept": false, - "reason": "Both mention optimizer-related behavior, but one is about SGD args being ignored and the other about a tied-embeddings warning during LoRA fine-tuning; different bugs and code paths." + "reason": "Different problems: unused SGD args in optimizer handling vs a tie_word_embeddings warning during Qwen fine-tuning." }, { "left": "issue:46082", "right": "issue:46512", "accept": false, - "reason": "One is about LlamaConfig head_dim validation; the other is about dtype metadata staying stale after from_config. Different model config issues." + "reason": "Both are config-related, but one is about LlamaConfig head_dim validation and the other about dtype being left stale in from_config." }, { "left": "issue:44351", "right": "issue:45278", "accept": false, - "reason": "Both involve imports, but one is a missing HybridCache symbol and the other is a broad post-upgrade import breakage report; too different to be the same defect." + "reason": "45278 is a broad import-regression report; 44351 is a specific HybridCache import failure. Not the same concrete bug." }, { "left": "issue:43299", "right": "issue:45357", "accept": false, - "reason": "Both are Qwen/Gemma multimodal regressions, but one is loading failure for Qwen3VL MoE models and the other is incorrect visual-encoder keys on save_pretrained; not the same bug." + "reason": "Different Qwen multimodal issues: model loading failure vs incorrect save_pretrained visual encoder keys." }, { "left": "issue:41950", "right": "issue:45020", "accept": false, - "reason": "A pipeline processor lookup bug is not the same as a broader remote_code compatibility regression." + "reason": "Different subsystems and symptoms: pipeline processor lookup vs remote_code breakage in recent transformers versions." }, { "left": "issue:42796", "right": "issue:43650", "accept": false, - "reason": "Tokenizer BPE-marker handling and a generic \"ADD THE DATA\" issue are unrelated." + "reason": "Tokenizer BPE-marker handling is unrelated to the generic 'ADD THE DATA' issue." }, { "left": "issue:44779", "right": "issue:45812", "accept": false, - "reason": "Both are tokenizer regressions, but they affect different model families with different symptoms and likely different fixes." + "reason": "Both are tokenizer regressions, but they affect different model families and appear to be separate bugs." }, { "left": "issue:43299", "right": "issue:43854", "accept": false, - "reason": "Different models and failure modes: Qwen3VL MoE loading versus GLM-4.7-Flash test loading." + "reason": "Different multimodal model failures with different loaders and symptoms; not the same underlying defect." }, { "left": "issue:44242", "right": "issue:45200", "accept": false, - "reason": "One concerns MoE load-balancing loss gating; the other is a Gemma 4 multimodal training default for mm_token_type_ids. Related area, not the same underlying bug." + "reason": "Different model logic: MoE load-balancing loss gating vs mm_token_type_ids defaults for Gemma 4 text-only fine-tuning." }, { "left": "issue:43479", "right": "issue:43525", "accept": false, - "reason": "Different config classes and different attributes; no shared code-path evidence." + "reason": "Different config initialization bugs on different classes and fields (vision/audio defaults vs missing pad_token_id)." }, { "left": "issue:44466", "right": "issue:45276", "accept": false, - "reason": "Both touch Gemma4 weights/embeddings, but one is about serialization of lm_head.weight and the other about resize_token_embeddings not propagating to per-layer/output embeddings; not clearly the same defect." + "reason": "Both touch tied/embedding behavior, but one is about serialization depending on device and the other about resize_token_embeddings propagation." }, { "left": "issue:41950", "right": "issue:45823", "accept": false, - "reason": "Pipeline processor lookup versus Gemma4 PLE device-mismatch during forward are distinct failures." + "reason": "Both mention multimodal models, but one is a processor lookup bug and the other is a device_map forward device mismatch." }, { "left": "issue:43653", "right": "issue:45593", "accept": false, - "reason": "Tokenizer special-token registration and D-FINE auxiliary-loss behavior are unrelated." + "reason": "Tokenizer special-token registration is unrelated to D-FINE auxiliary-loss behavior." }, { "left": "issue:44683", "right": "issue:45542", "accept": false, - "reason": "FlexAttention compilation on torch 2.9 and a TensorBoard/TensorFlow backend error are unrelated." + "reason": "Compiled flex_attention on torch>=2.9 has no relation to the TensorFlow backend import error." }, { "left": "issue:43650", "right": "issue:46209", "accept": false, - "reason": "No meaningful overlap: a placeholder issue versus an FP8 inference failure on B200." + "reason": "Completely different issues: a generic 'ADD THE DATA' ticket vs FP8 inference failure on B200." }, { "left": "issue:44016", "right": "issue:44291", "accept": false, - "reason": "A notebook syntax error is not the same as an init_empty_weights argument regression." + "reason": "Notebook syntax error and init_empty_weights TypeError are unrelated failures." }, { "left": "issue:44462", "right": "issue:45081", "accept": false, - "reason": "Both are tokenizer-related, but one is AutoTokenizer ignoring tokenizer.json while the other is a Mistral regex patch crashing on a backend_tokenizer attribute; different bug paths." + "reason": "Both involve tokenizers, but one is AutoTokenizer ignoring tokenizer.json and the other is a Mistral regex patch crash." }, { "left": "issue:43441", "right": "issue:45834", "accept": false, - "reason": "Different model-specific runtime errors: FlashAttention failure versus OCR long-input index error." + "reason": "Different model-specific runtime failures: FlashAttention on Ministral-3 vs long OCR index error in Kosmos2.5." }, { "left": "issue:44361", "right": "issue:45593", "accept": false, - "reason": "Tokenizer attribute errors and D-FINE auxiliary-loss logic are unrelated." + "reason": "Tokenizer AttributeError and D-FINE loss logic are unrelated." }, { "left": "issue:45103", "right": "issue:45372", "accept": false, - "reason": "auto_docstring annotation handling and a mistral_common import failure are different issues." + "reason": "Docstring helper crash from future annotations is unrelated to Gemma 4 processor loading/import failure." }, { "left": "issue:42994", "right": "issue:43317", "accept": false, - "reason": "Quantized model saving and dequantized model loading with device_map=auto are separate stages and different failure modes." + "reason": "Both are quantization-related, but one is saving a quantized model and the other is loading a dequantized model with device_map offload." }, { "left": "issue:33290", "right": "issue:45656", "accept": false, - "reason": "Both mention deepspeed/optimizer behavior, but OOM with Adafactor and double optimizer.step calls are different concrete bugs." + "reason": "Both mention deepspeed/optimizers, but one is OOM with Adafactor and the other is an optimizer.step being called twice." }, { "left": "issue:45081", "right": "issue:45907", "accept": false, - "reason": "Tokenizer regex patch crash and list_repo_files revision forwarding are unrelated." + "reason": "Tokenizer regex crash and list_repo_files revision forwarding are unrelated code paths." }, { "left": "issue:45600", "right": "issue:45712", "accept": false, - "reason": "Both are repo-maintenance/C I-related, but one is stale auto_mappings references and the other is leftover dummy classes; different fixes." + "reason": "Both are repo-maintenance/CI-ish, but one is stale auto_mappings references and the other is leftover dummy classes leaking into imports." }, { "left": "issue:44261", "right": "issue:44263", "accept": false, - "reason": "One is a precision/config epsilon issue in MLA q_a_layernorm; the other is torch.split return handling in GlmMoeDsaIndexer." + "reason": "Completely different bugs: MLA layernorm epsilon handling vs torch.split return values in a GLM-MoE indexer." }, { "left": "issue:44623", "right": "issue:44749", "accept": false, - "reason": "Processor save_pretrained missing files and a performance regression after upgrading versions are unrelated." + "reason": "save_pretrained missing files is unrelated to the performance regression after upgrading Transformers." } ] }, @@ -12357,20 +12381,20 @@ "issue:42491|issue:43299", "issue:43531|issue:45362" ], - "prepared_review_unit_hash": "64e40ace2206fdc67ce2b5db81f202698d5bca95e4129916bbe31819c2e2bda8", + "prepared_review_unit_hash": "6a64ba4aa55eacc29588d5d7083aeb0415e019c36df03f18f3b75625392e28ed", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 28, - "serialized_chars": 35375, + "serialized_chars": 35373, "estimated_input_tokens": 8844, "estimated_eval_tokens": 17944 }, @@ -12379,180 +12403,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: most items are unrelated v5 regressions, model-loading failures, backend/install issues, or docs/test bugs. None of the soft pairs look like the same underlying bug, so I would not merge any of them as duplicates.", - "confidence": 0.9, - "canonical_issue_reason": "Issue 43976 is the broadest, most release-level regression in the set: it describes a general Transformers v5 compatibility break rather than a model-specific edge case.", + "summary": "This is a loose mix of Transformers v5-era regressions around loading/imports, tokenizers/processors, and model-specific runtime bugs. The suggested soft pairs mostly share broad subsystem language but not the same concrete underlying defect, so none should be merged as duplicates.", + "confidence": 0.27, + "canonical_issue_reason": "Issue 43723 is the broadest user-facing loading regression in the set and best represents the cluster\u2019s repeated v5 import/load failures, even though the cluster is not a tight duplicate set.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45561 is the most actionable standalone issue in the cluster: it is open, has the most discussion/inbound references, and contains a concrete failure mode with a likely reproducible test-race root cause.", + "best_issue_reason": "Issue 43723 is the best single representative issue because it describes a top-level API loading failure that is easy to understand and closest to the dominant theme of the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:42175", "right": "issue:43976", "accept": false, - "reason": "Different failures: backend/dependency packaging vs Python-version compatibility. Same library, but not the same bug." + "reason": "One is a backend-install packaging problem; the other is a Python-version compatibility issue. Different failure modes." }, { "left": "issue:44623", "right": "issue:45397", "accept": false, - "reason": "One is missing files in processor.save_pretrained; the other is a from_pretrained/load failure on Gemma-4 ZeRO3. Different code paths." + "reason": "Missing files from processor.save_pretrained and a zero3 from_pretrained crash on Gemma-4 are unrelated bugs." }, { "left": "issue:44671", "right": "issue:44779", "accept": false, - "reason": "Both affect model behavior in v5, but CamemBERT MLM predictions and Deepseek tokenizer output are separate model-specific regressions." + "reason": "CamemBERT MLM prediction corruption and Deepseek tokenizer corruption affect different code paths and models." }, { "left": "issue:43901", "right": "issue:44246", "accept": false, - "reason": "Docs mismatch for return_all_scores is unrelated to intermittent import latency." + "reason": "Docs mismatch for return_all_scores is not the same as intermittent slow import of transformers." }, { "left": "issue:43441", "right": "issue:44683", "accept": false, - "reason": "Both mention attention/kernel issues, but one is Ministral-3 FlashAttention in v5 RC and the other is compiled flex_attention on torch>=2.9; different bug surfaces." + "reason": "Ministral FlashAttention failure and compiled flex_attention failure on torch>=2.9 are different kernel/backend issues." }, { "left": "issue:43673", "right": "issue:43761", "accept": false, - "reason": "Chunked-prefill cache handling and CLIPVisionModel hidden_states are different failures in different subsystems." + "reason": "Chunked_prefill cache regression and CLIPVisionModel hidden_states regression are distinct model-path bugs." }, { "left": "issue:45542", "right": "issue:45800", "accept": false, - "reason": "One is a missing TensorFlow backend in install extras; the other is a torch/Transformers version incompatibility. Not the same defect." + "reason": "One is a missing TensorFlow backend after an extras install; the other is a version incompatibility between torch and transformers." }, { "left": "issue:43723", "right": "issue:44162", "accept": false, - "reason": "Tokenizer loading in v5 and ESM2 breakage are unrelated model/loading problems." + "reason": "Generic tokenizer loading failure and ESM2 model breakage are different concrete problems." }, { "left": "issue:43824", "right": "issue:43976", "accept": false, - "reason": "ImportError for a specific VL model class is unrelated to the Python 3.9+ support regression." + "reason": "Missing Qwen2.5-VL import symbol and Python-version incompatibility are unrelated." }, { "left": "issue:46291", "right": "issue:46399", "accept": false, - "reason": "Kernel package import/version-revision validation and missing decorator registration are different kernel-integration bugs." + "reason": "pipelines import failure from kernels versioning and a missing decorator in Qwen3.5 rotary-pos-emb registration are different bugs." }, { "left": "issue:43525", "right": "issue:43881", "accept": false, - "reason": "Both are load failures, but for different models and error modes: missing pad_token_id vs a failed glm-4v-9b load." + "reason": "Llama4Config pad_token_id AttributeError and glm-4v loading failure do not point to the same root cause." }, { "left": "issue:44016", "right": "issue:45823", "accept": false, - "reason": "Notebook syntax error is unrelated to Gemma4 device placement during forward." + "reason": "Notebook syntax error and Gemma4 device-map forward mismatch are unrelated." }, { "left": "issue:43525", "right": "issue:43643", "accept": false, - "reason": "Llama4Config missing pad_token_id is not the same as AutoConfig returning missing fields under trust_remote_code." + "reason": "A missing config attribute in Llama4 and trust_remote_code dropping fields are both config-related, but they are not the same bug." }, { "left": "issue:44589", "right": "issue:45362", "accept": false, - "reason": "Float8 storage lookup failure and Qwen3.5 chat crash are distinct runtime errors." + "reason": "Float8 storage lookup failure and Qwen3.5 chat crash are unrelated runtime issues." }, { "left": "issue:44188", "right": "issue:44336", "accept": false, - "reason": "Attention-kernel divergence under torch.compile is unrelated to ANSI color leakage in loading_report." + "reason": "torch.compile attention-kernel divergence and ANSI codes in loading_report are completely different." }, { "left": "issue:46655", "right": "issue:46726", "accept": false, - "reason": "Video padding miscalculation in SmolVLM is different from AutoProcessor failing when a required vision backend is missing." + "reason": "Video padding calculation in SmolVLM and missing vision-backend handling in AutoProcessor are separate processor bugs." }, { "left": "issue:44336", "right": "issue:44373", "accept": false, - "reason": "Terminal ANSI output and docstring text are unrelated." + "reason": "ANSI code emission and wrong docstring content are unrelated documentation/reporting issues." }, { "left": "issue:41628", "right": "issue:43976", "accept": false, - "reason": "Missing AutoImageProcessor import and Python 3.9+ incompatibility are separate import/environment issues." + "reason": "AutoImageProcessor import error and Python3.9/3.10 compatibility are different import/version issues." }, { "left": "issue:43784", "right": "issue:45542", "accept": false, - "reason": "Both mention missing symbols, but sentence-transformers import NameError and undefined tf backend are different dependency issues." + "reason": "sentence-transformers import NameError and undefined tf backend after extras install do not share a root cause." }, { "left": "issue:44246", "right": "issue:44496", "accept": false, - "reason": "Slow import performance is unrelated to config.json missing model_type." + "reason": "Performance on import and unrecognized model/config parsing failure are unrelated." }, { "left": "issue:44297", "right": "issue:45397", "accept": false, - "reason": "Tokenizer config serialization mismatch and Gemma-4 ZeRO3 from_pretrained failure are different bugs." + "reason": "Tokenizer serialization metadata mismatch and Gemma-4 from_pretrained failure are different code paths." }, { "left": "issue:45561", "right": "issue:45563", "accept": false, - "reason": "A pytest-xdist file race and a stale generation warning are unrelated test/runtime issues." + "reason": "pytest-xdist file race and stale num_return_sequences warning are unrelated testing/generation issues." }, { "left": "issue:46614", "right": "issue:46665", "accept": false, - "reason": "Phonemizer language-state retention and from_pretrained behavior inside Python classes are different state-management problems." + "reason": "Stale phonemizer backend state and inconsistent from_pretrained behavior inside Python classes are not the same defect." }, { "left": "issue:45103", "right": "issue:46399", "accept": false, - "reason": "Auto-docstring annotation parsing crash and Qwen3.5 kernel registration error are unrelated." + "reason": "auto_docstring future-annotations crash and missing decorator registration in Qwen3.5 are unrelated." }, { "left": "issue:44242", "right": "issue:45593", "accept": false, - "reason": "Missing load-balancing loss and missing auxiliary losses are similar in wording but refer to different models and training branches." + "reason": "Missing load-balancing loss in one MoE model and missing auxiliary losses in D-FINE are model-specific, not one bug." }, { "left": "issue:43526", "right": "issue:45003", "accept": false, - "reason": "BeitImageProcessorFast label reduction and unsafe sys.modules access are unrelated code defects." + "reason": "BeitImageProcessorFast label reduction bug and unsafe sys.modules access are unrelated." }, { "left": "issue:42491", "right": "issue:43299", "accept": false, - "reason": "Both involve Qwen3 MoE, but one is LoRA compatibility across hf4.x/main and the other is Qwen3VL MoE loading; not clearly the same bug." + "reason": "Both involve Qwen3 MoE, but one is a LoRA compatibility issue and the other is a direct model-loading regression; not the same concrete bug." }, { "left": "issue:43531", "right": "issue:45362", "accept": false, - "reason": "Sliding_window behavior in Qwen3-MoE and Qwen3.5 chat crashes are different issues." + "reason": "Sliding-window behavior in Qwen3-MoE and Qwen3.5 chat crashes are different symptoms and code paths." } ] }, @@ -12659,226 +12683,226 @@ "issue:44561|issue:46399", "issue:43295|issue:45292" ], - "prepared_review_unit_hash": "7a377111234a053a37f560d41db8cd46b00b180c326429db99c8d908fd726588", + "prepared_review_unit_hash": "aa146d244bb03480c92de4f923b112c16e5860d497780618eb96bd14e2fcabe1", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 32, - "serialized_chars": 37240, - "estimated_input_tokens": 9310, - "estimated_eval_tokens": 18876 + "serialized_chars": 37244, + "estimated_input_tokens": 9311, + "estimated_eval_tokens": 18878 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is a grab bag of mostly unrelated Transformers issues spanning multimodal/video regressions, config/import breakages, cache/deepspeed bugs, and docs/UX problems. None of the soft edges look like true duplicates of the same underlying bug.", - "confidence": 0.73, - "canonical_issue_reason": "issue:44479 is the broadest representative regression in the set, with an explicit v5.3.0 video-input failure affecting multiple Qwen multimodal variants.", + "summary": "This cluster is highly heterogeneous: most items are standalone Transformers bug reports across unrelated areas (imports, configs, deepspeed, multimodal/video, caching, docs). I did not find any soft-edge pair that clearly describes the same underlying issue, so all candidate merges should be rejected.", + "confidence": 0.27, + "canonical_issue_reason": "Issue 44479 is the broadest representative of the strongest visible theme in this set: a versioned multimodal/video regression affecting multiple Qwen model families. It is more cluster-like than the many narrow one-off failures, though the overall cluster remains quite mixed.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44479 best captures the largest coherent subtheme here: a shared video-input regression across several related models.", + "best_issue_reason": "44479 is the best single issue to anchor the cluster because it names a wide regression, spans several related model variants, and is more general than the other candidates.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43540", "right": "issue:46752", "accept": false, - "reason": "Different failures: Qwen3OmniMoe video processing error vs empty-list apply_chat_template IndexError." + "reason": "Both mention input processing, but one is a Qwen3OmniMoe video-input ValueError and the other is an empty-list apply_chat_template crash; different code paths and failure modes." }, { "left": "issue:44246", "right": "issue:44871", "accept": false, - "reason": "Unrelated bugs: router-loss handling vs Gemma eos_token_id config mismatch." + "reason": "Import-time slowness and Gemma eos_token_id inconsistency are unrelated bugs." }, { "left": "issue:44295", "right": "issue:45397", "accept": false, - "reason": "Different code paths and symptoms: position_ids buffer access vs Gemma-4 ZeRO3 from_pretrained loading." + "reason": "A position_ids buffer access error and a Gemma-4 ZeRO3 from_pretrained bug are different issues." }, { "left": "issue:44336", "right": "issue:44485", "accept": false, - "reason": "ANSI output formatting bug vs GLM-5 RoPE implementation discussion; not the same bug." + "reason": "ANSI output in loading_report and GLM-5 RoPE implementation are unrelated." }, { "left": "issue:44451", "right": "issue:46752", "accept": false, - "reason": "Different model-loading failure vs empty-chat-template input handling." + "reason": "ScandiBERT loading failure and empty-list chat-template handling are different failures." }, { "left": "issue:44336", "right": "issue:46424", "accept": false, - "reason": "Loading-report terminal detection bug vs cache max length behavior; unrelated." + "reason": "Terminal ANSI formatting and cache max length handling do not share the same underlying bug." }, { "left": "issue:42886", "right": "issue:46132", "accept": false, - "reason": "Offline tokenizer cache loading vs AttentionInterface.register behavior; different subsystems." + "reason": "Offline tokenizer cache loading and AttentionInterface.register behavior are unrelated subsystems." }, { "left": "issue:43550", "right": "issue:46506", "accept": false, - "reason": "torch.compile+SDPA issue in Bamba vs CPU/CUDA divergence in PPDocLayoutV3; not duplicate." + "reason": "torch.compile with SDPA on Bamba and CPU/CUDA divergence in PPDocLayoutV3 are distinct model/runtime problems." }, { "left": "issue:41950", "right": "issue:44220", "accept": false, - "reason": "Video-classification pipeline import/processor lookup issue vs audio feature extraction bug." + "reason": "Video-classification image-processor lookup and audio fbank extraction are different pipeline bugs." }, { "left": "issue:44779", "right": "issue:45292", "accept": false, - "reason": "Tokenizer correctness regression vs resize_token_embeddings/output_embeddings mismatch." + "reason": "DeepSeek tokenizer correctness and resize_token_embeddings not updating output embeddings are different code paths." }, { "left": "issue:36246", "right": "issue:42757", "accept": false, - "reason": "Different missing symbol imports from different packages; only superficially similar." + "reason": "Qwen2_5_VLImageProcessor import failure and missing is_offline_mode import are separate missing-symbol regressions." }, { "left": "issue:36246", "right": "issue:44351", "accept": false, - "reason": "Both are import errors, but for different removed symbols and different fixes." + "reason": "Both are import errors, but for different exported symbols/classes and different root causes." }, { "left": "issue:43901", "right": "issue:45003", "accept": false, - "reason": "Docs mention return_all_scores vs unsafe sys.modules access; unrelated." + "reason": "Pipeline docs mismatch and unsafe sys.modules access are unrelated." }, { "left": "issue:43257", "right": "issue:45137", "accept": false, - "reason": "Different DeepSpeed-related problems: weight conversion vs empty deque during ZeRO3." + "reason": "Both involve DeepSpeed, but one is Qwen3 MOE weight conversion and the other is a ZeRO3 deque/index error; not the same bug." }, { "left": "issue:45003", "right": "issue:45561", "accept": false, - "reason": "sys.modules access bug vs pytest-xdist captured_info.txt race; different issues." + "reason": "Unsafe sys.modules access and xdist captured_info.txt race are different concurrency/robustness issues." }, { "left": "issue:44734", "right": "issue:46424", "accept": false, - "reason": "KV cache continuation crash in serve vs cache max length semantics; different code paths." + "reason": "KV-cache continuation indexing and cache max_length configuration concern different cache logic." }, { "left": "issue:44496", "right": "issue:45003", "accept": false, - "reason": "Model config recognition failure vs sys.modules access bug; not the same underlying defect." + "reason": "Unrecognized model config loading and sys.modules access are unrelated." }, { "left": "issue:43540", "right": "issue:46655", "accept": false, - "reason": "Both involve video, but different models and distinct preprocessing bugs." + "reason": "Both are video-related, but Qwen3OmniMoe video-input handling and SmolVLM padding calculation are not the same defect." }, { "left": "issue:45859", "right": "issue:46399", "accept": false, - "reason": "CPU offload incompatibility vs missing decorator for kernel registration; not mergeable as one fix." + "reason": "CPU-offload incompatibility in rotary embedding and a missing registration decorator for kernels are different Qwen3.5 issues." }, { "left": "issue:43673", "right": "issue:44561", "accept": false, - "reason": "Chunked prefill cache issue vs removal of is_torch_fx_available breaking trust_remote_code; unrelated." + "reason": "Chunked-prefill cache regression and removal of is_torch_fx_available break different behaviors." }, { "left": "issue:44162", "right": "issue:45200", "accept": false, - "reason": "ESM2 model breakage vs Gemma-4 mm_token_type_ids defaulting; different models and failures." + "reason": "ESM2 breakage and Gemma 4 mm_token_type_ids defaulting are unrelated model-specific bugs." }, { "left": "issue:44479", "right": "issue:46655", "accept": false, - "reason": "Both are video-input regressions, but for different model families and different preprocessing defects." + "reason": "Both are video-input regressions, but they affect different model families and point to different preprocessing bugs." }, { "left": "issue:43475", "right": "issue:46369", "accept": false, - "reason": "Missing output attribute in SAM3 video encoder vs memory overflow in sam3-litetext; not duplicate." + "reason": "SAM 3 video encoder output attribute mismatch and SAM3-litetext memory overflow are different issues." }, { "left": "issue:44242", "right": "issue:44857", "accept": false, - "reason": "Mixture-of-experts router loss bug vs LwDetr image loss AMP crash; unrelated." + "reason": "MoE load-balancing loss omission and LwDetrImageLoss AMP crash are unrelated." }, { "left": "issue:45375", "right": "issue:45759", "accept": false, - "reason": "Strict config field dropping vs text_config unwrapping in AutoModelForCausalLM; related area but different bug." + "reason": "Both touch Qwen multimodal configs, but one is strict-field dropping and the other is from_config not unwrapping text_config; different fixes." }, { "left": "issue:43262", "right": "issue:44246", "accept": false, - "reason": "Audio chat-template sampling-rate default vs router-loss omission; unrelated." + "reason": "Audio chat-template sampling-rate default and import slowness are unrelated." }, { "left": "issue:44295", "right": "issue:44749", "accept": false, - "reason": "Buffer access error vs performance regression during filtering; different symptom and fix." + "reason": "A buffer access IndexError and a broad slowdown regression are not the same bug." }, { "left": "issue:45137", "right": "issue:45600", "accept": false, - "reason": "ZeRO3 deque failure vs stale auto_mappings references; different maintenance issues." + "reason": "DeepSpeed ZeRO3 deque failure and Sam3LiteText auto_mappings CI breakage are unrelated." }, { "left": "issue:36246", "right": "issue:43596", "accept": false, - "reason": "Qwen processor import error vs ZeRO3/BertModel index error; unrelated." + "reason": "Qwen2_5_VLImageProcessor import error and ZeRO3/BertModel empty-dimension IndexError are different failures." }, { "left": "issue:44877", "right": "issue:45375", "accept": false, - "reason": "Strict config loading failure vs silent field drop in a different config class; not the same bug." + "reason": "A generic strict-config loading failure and a missing Qwen3_5MoeVisionConfig field are related only at a very high level, not the same underlying bug." }, { "left": "issue:44561", "right": "issue:46399", "accept": false, - "reason": "Removal of is_torch_fx_available vs Qwen3.5 apply_rotary_pos_emb registration crash; different failure modes." + "reason": "Removal of is_torch_fx_available and a missing decorator for apply_rotary_pos_emb are unrelated breakages." }, { "left": "issue:43295", "right": "issue:45292", "accept": false, - "reason": "Processor/tokenizer regression vs embeddings resizing bug; separate code paths." + "reason": "Processor.tokenizer/image-passing regression and resize_token_embeddings output-embedding mismatch are different issues." } ] }, @@ -12979,20 +13003,20 @@ "issue:44483|issue:46202", "issue:44188|issue:46421" ], - "prepared_review_unit_hash": "bc01a64c070bb3f2e157df977744024836dea2f93f70be2720c6f66550fd40b8", + "prepared_review_unit_hash": "4bf917949490516578e0f6b691a22fa99e6a9e578d6e9cd96c68032f0895a648", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 27, - "serialized_chars": 35200, + "serialized_chars": 35198, "estimated_input_tokens": 8800, "estimated_eval_tokens": 17856 }, @@ -13001,174 +13025,174 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is highly heterogeneous: the soft pairs span unrelated tokenizer, multimodal, caching, config, and training/runtime issues. None look like true duplicates or PRs that should be merged together.", - "confidence": 0.96, - "canonical_issue_reason": "No single issue is a clear canonical representative because the cluster does not describe one underlying bug; choosing any item would misrepresent the rest.", + "summary": "All suggested soft pairs look like false positives: they span unrelated failures across tokenization, model loading, generation, cache behavior, DeepSpeed, vision/video models, and config handling. No single issue stands out as a credible duplicate hub for the cluster.", + "confidence": 0.95, + "canonical_issue_reason": "No coherent canonical issue emerges; the items are too heterogeneous and the soft pairs do not share the same underlying bug or code path.", "canonical_pr_reason": null, - "best_issue_reason": "No issue is suitable as a global best representative for deduping since the items are not the same problem family.", + "best_issue_reason": "No issue is a strong global representative here because the cluster is dominated by unrelated or only loosely similar bugs.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45292", "right": "issue:45584", "accept": false, - "reason": "Different bugs: one is embedding resize sync, the other is Whisper empty transcription handling after token alignment." + "reason": "Resize-token-embeddings/output-embeddings bug vs empty-transcription Whisper failure; unrelated code paths." }, { "left": "issue:44466", "right": "issue:45335", "accept": false, - "reason": "Both mention tied/resize embeddings, but one is device-dependent serialization and the other is T5Gemma decoder embedding not updating after resize." + "reason": "Both mention embeddings, but one is tied-weight serialization by device and the other is decoder.embed_tokens not updating on resize; different bugs." }, { "left": "issue:44623", "right": "issue:46726", "accept": false, - "reason": "Different failure points: saving processor files vs AutoProcessor falling back badly when a vision backend is missing." + "reason": "save_pretrained missing files vs load-time AutoProcessor error when a vision backend is missing; different lifecycle stages." }, { "left": "issue:44038", "right": "issue:46506", "accept": false, - "reason": "Unrelated models and symptoms: Qwen3-VL-Moe loading bug vs CPU/CUDA divergence in PPDocLayoutV3." + "reason": "Qwen3-VL-Moe loading bug vs CPU/CUDA output divergence in PPDocLayoutV3; unrelated models and symptoms." }, { "left": "issue:44987", "right": "issue:45459", "accept": false, - "reason": "Different tokenizer/loading issues; one is a model load failure on transformers>=5.1, the other is exception masking when protobuf is absent." + "reason": "Model loading regression vs tokenizer error masking when protobuf is absent; different failure modes." }, { "left": "issue:29870", "right": "issue:30333", "accept": false, - "reason": "Completely different domains: BART generate() error vs MLflow reporting after failed training jobs." + "reason": "Generation UnboundLocalError vs MLflow reporting not updated after failed training; unrelated." }, { "left": "issue:44479", "right": "issue:46170", "accept": false, - "reason": "Different subsystems and failure modes: video-input regression in Qwen models vs Windows/ROCm import dependency problems." + "reason": "Video-input regression in Qwen VL models vs Windows ROCm import/dependency failure; not the same bug." }, { "left": "issue:43352", "right": "issue:45250", "accept": false, - "reason": "Only loosely related by Flash Attention 2.0; one is a specific model incompatibility, the other is a generic feature/issue title." + "reason": "Specific model lacking Flash Attention 2 support vs a generic flash-attention topic; too broad to be the same concrete issue." }, { "left": "issue:44060", "right": "issue:44077", "accept": false, - "reason": "Different concrete bugs: tied-weights warning in Qwen3-Next vs a patchtsmixer post_init constraint issue." + "reason": "Qwen3-Next tied-weights warning vs patchtsmixer post_init policy; unrelated subsystems." }, { "left": "issue:45412", "right": "issue:46123", "accept": false, - "reason": "Unrelated: RT-DETR memory release vs MaskGenerationPipeline partial-batch result dropping." + "reason": "Memory leak on delete vs partial-batch result dropping in a pipeline; different problems." }, { "left": "issue:43653", "right": "issue:44162", "accept": false, - "reason": "Tokenizer special-token registration bug is unrelated to the ESM2 broken-workflow report." + "reason": "BigBirdTokenizer special-token registration bug vs ESM2 breakage; unrelated." }, { "left": "issue:44368", "right": "issue:44977", "accept": false, - "reason": "Different Qwen3.5 problems: tied-embeddings warning during fine-tuning vs flash-attention generation regression." + "reason": "Tied-embeddings warning in Qwen3.5 fine-tuning vs flash-attention generation regression; different bugs." }, { "left": "issue:44995", "right": "issue:45200", "accept": false, - "reason": "Different models and root causes: GlmMoeDsa stale cache crash vs Gemma 4 token-id defaulting for text-only fine-tuning." + "reason": "Stale indexer cache crash vs Gemma 4 token-type-id defaulting; unrelated." }, { "left": "issue:44877", "right": "issue:46170", "accept": false, - "reason": "Different issues: strict Granite config parsing vs Windows ROCm import dependency handling." + "reason": "Strict config loading failure vs torch.distributed import issue on Windows/ROCm; unrelated." }, { "left": "issue:43475", "right": "issue:43540", "accept": false, - "reason": "Both are multimodal/video-related, but the concrete bugs differ: missing vision encoder field vs Qwen3OmniMoe video processing ValueError." + "reason": "SAM3 vision-encoder attribute error vs Qwen3OmniMoe video ValueError; different models and code paths." }, { "left": "issue:42886", "right": "issue:45229", "accept": false, - "reason": "Tokenizer offline cache loading is unrelated to Gemma4 multi-GPU CUDA OOM." + "reason": "Offline tokenizer cache load failure vs Gemma4 multi-GPU OOM; unrelated." }, { "left": "issue:43065", "right": "issue:43408", "accept": false, - "reason": "Related SAM3 area, but one is a dummy Conv2d implementation detail and the other is a model-type mismatch warning." + "reason": "Dummy Conv2d implementation vs model-type mismatch warning; related family, but not the same bug." }, { "left": "issue:45397", "right": "issue:45656", "accept": false, - "reason": "Both involve deepspeed, but one is Gemma-4 zero3 loading and the other is an optimizer step being called twice." + "reason": "DeepSpeed zero3 from_pretrained issue vs optimizer stepping twice; not the same concrete failure." }, { "left": "issue:43931", "right": "issue:44368", "accept": false, - "reason": "Qwen3-VL weight-shape mismatch is not the same as a tied-embeddings warning in Qwen3.5 fine-tuning." + "reason": "Model weight-shape mismatch on load vs tied-weights warning; different symptoms and causes." }, { "left": "issue:44368", "right": "issue:45563", "accept": false, - "reason": "One is a tied-weights warning, the other is a stale paged-generate warning for num_return_sequences." + "reason": "Tied-weights warning vs stale num_return_sequences warning in paged generate; unrelated." }, { "left": "issue:44995", "right": "issue:45588", "accept": false, - "reason": "Different code paths: GlmMoeDsa second forward-pass crash vs flash_attention integration AttributeError on s_aux=None." + "reason": "Stale cache crash vs flash_attention AttributeError on s_aux=None; different code paths." }, { "left": "issue:43723", "right": "issue:46710", "accept": false, - "reason": "Tokenizer loading in v5 is unrelated to a model accuracy regression after upgrading versions." + "reason": "Tokenizer loading issue in v5 vs accuracy regression on a specific model after upgrade; not duplicates." }, { "left": "issue:45987", "right": "issue:46082", "accept": false, - "reason": "StaticCache return type contract bug is unrelated to LlamaConfig head_dim validation." + "reason": "Cache length return-type contract vs LlamaConfig head_dim validation; unrelated." }, { "left": "issue:44188", "right": "issue:46202", "accept": false, - "reason": "Different problems: torch.compile branching changes attention kernels vs Gemma pipeline roundtrip on the \u2581 token." + "reason": "torch.compile attention-kernel divergence vs Gemma pipeline roundtrip failure; different bugs." }, { "left": "issue:43906", "right": "issue:46202", "accept": false, - "reason": "The reproduction issue is not clearly the same bug as Gemma roundtrip failure; no concrete shared failure path is shown." + "reason": "An isolated reproduction of another issue is not enough to show the same underlying bug as the Gemma roundtrip failure." }, { "left": "issue:44483", "right": "issue:46202", "accept": false, - "reason": "Chat/completions request rejection in v5.3 is a different runtime/API bug from Gemma tokenizer roundtrip failure." + "reason": "Chat-completions request rejection vs Gemma pipeline roundtrip issue; unrelated." }, { "left": "issue:44188", "right": "issue:46421", "accept": false, - "reason": "Both mention caching/compile behavior, but one is attention-kernel divergence and the other is spurious recompilation from lazy cache init." + "reason": "Both touch cache/compile behavior, but one is attention-kernel divergence and the other is lazy cache init causing recompilation; different concrete problems." } ] }, @@ -13272,220 +13296,184 @@ "issue:43475|issue:45381", "issue:43723|issue:46326" ], - "prepared_review_unit_hash": "cd1f4540402eefc258ef5fd711be66f432a5c32bfb1ce23f9bcdb5df693593f1", + "prepared_review_unit_hash": "fff10ed3d4884b951d60a06382fa2e89879a2900f362bde0bb903710484d8508", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36164, - "estimated_input_tokens": 9041, - "estimated_eval_tokens": 18338 + "serialized_chars": 36157, + "estimated_input_tokens": 9040, + "estimated_eval_tokens": 18336 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly distinct bug reports with only one likely duplicate-ish pair in the Qwen3.5-Moe v5 loading path. The rest span unrelated failures across processors, tokenizers, generation, quantization, and model-specific regressions.", - "confidence": 0.66, - "canonical_issue_reason": "issue:45310 is the strongest representative of the cluster\u2019s nearest duplicate theme: a concrete Qwen3.5-Moe from_pretrained regression in Transformers v5, likely overlapping with the related config-field-drop report.", + "summary": "This cluster is mostly a grab-bag of Transformers v5 regression reports, with only a couple of plausible duplicate pairs around Qwen3.5-Moe config loading and AutoProcessor backend-missing handling. Most other pairs are different bug paths and should stay separate.", + "confidence": 0.74, + "canonical_issue_reason": "issue:43646 is the broadest, most central regression report here: it describes Transformers 5.0 breaking custom model initialization, which best matches the cluster\u2019s overall v5-loading/regression theme.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45310 is the best single issue to anchor this cluster because it is specific, actionable, and appears closest to another near-match in the set.", + "best_issue_reason": "issue:43646 is the best single issue to anchor this cluster because it is the most umbrella-like report and can represent the shared v5 initialization/loading regression theme better than the narrower model-specific bugs.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45685", "right": "issue:46431", "accept": false, - "reason": "Different subsystems and failure modes: MoE/MPS histogram error vs chat-template validation." + "reason": "Different problems: an MPS histogram backend error in MoE code vs an empty-list apply_chat_template crash." }, { "left": "issue:45584", "right": "issue:46170", "accept": false, - "reason": "Whisper decoding bug and Windows/ROCm torch.distributed import failure are unrelated." + "reason": "Unrelated: Whisper empty-transcription generation failure vs Windows/ROCm torch.distributed import dependency issue." }, { "left": "issue:43065", "right": "issue:46123", "accept": false, - "reason": "Sam3PixelDecoder dummy layer issue and MaskGenerationPipeline batch-finalization bug affect different code paths." + "reason": "Both involve segmentation-related code, but one is a dummy Conv2d in Sam3PixelDecoder and the other is a MaskGenerationPipeline batching bug; different code paths and fixes." }, { "left": "issue:44361", "right": "issue:44749", "accept": false, - "reason": "Tokenizer AttributeError vs a performance regression after upgrade; not the same bug." + "reason": "Unrelated: MLukeTokenizer task attribute error vs a broad performance regression after upgrading Transformers." }, { "left": "issue:44297", "right": "issue:44623", "accept": false, - "reason": "Both are save_pretrained-related, but one is tokenizer metadata mismatch and the other is missing processor files; different fixes." + "reason": "Both are save_pretrained-related, but one is tokenizer metadata mismatch and the other is missing processor files; not the same bug." }, { "left": "issue:43994", "right": "issue:46726", "accept": false, - "reason": "Both involve vision models, but one is bad outputs from AutoModel/pipeline and the other is missing-backend import handling in AutoProcessor." + "reason": "Different symptoms and targets: SigLIP2 bad outputs with AutoModel/pipeline vs AutoProcessor's missing-image-processor error when vision backend is absent." }, { "left": "issue:43646", "right": "issue:44162", "accept": false, - "reason": "Generic custom-model initialization breakage vs a specific ESM2 regression; too broad to be the same issue." + "reason": "Different subsystems: custom model initialization regression vs ESM2 breakage." }, { "left": "issue:44749", "right": "issue:45656", "accept": false, - "reason": "Performance slowdown and double optimizer-step behavior are different problems." + "reason": "Performance slowdown after upgrade vs optimizer step being called twice under deepspeed; not the same failure mode." }, { "left": "issue:44871", "right": "issue:46132", "accept": false, - "reason": "Gemma eos_token_id mismatch is unrelated to AttentionInterface.register behavior change." + "reason": "Gemma eos_token_id configuration inconsistency vs AttentionInterface.register behavior change; unrelated." }, { "left": "issue:43701", "right": "issue:44749", "accept": false, - "reason": "Checkpoint resume key mismatch and data-path slowdown are unrelated." + "reason": "Checkpoint key mismatch vs filtering performance regression; different bugs." }, { "left": "issue:42994", "right": "issue:43482", "accept": false, - "reason": "Quantized-model saving failure and GGUF loading failure are different directions and code paths." + "reason": "Quantized model saving failure vs Qwen2.5-GGUF loading failure; save/load and model-specific issues differ." }, { "left": "issue:44623", "right": "issue:45656", "accept": false, - "reason": "Processor save_pretrained file emission bug is unrelated to deepspeed optimizer-step duplication." + "reason": "Processor.save_pretrained missing files vs optimizer double-step in deepspeed; unrelated." }, { "left": "issue:43441", "right": "issue:46752", "accept": false, - "reason": "FlashAttention/model-load failure and empty-chat-template IndexError are unrelated." + "reason": "FlashAttention failure for Ministral-3 vs apply_chat_template empty-list IndexError; different code paths." }, { "left": "issue:43232", "right": "issue:43673", "accept": false, - "reason": "Both are generation-related, but one is sync_gpus kwargs handling and the other is chunked_prefill cache absence; not the same bug." + "reason": "Both are generation-related, but one is sync_gpus/_update_model_kwargs_for_generation and the other is chunked_prefill cache missing; too different to merge." }, { "left": "issue:43475", "right": "issue:43643", "accept": false, - "reason": "Missing attribute on Sam3 video output is not the same as missing config fields from trust_remote_code AutoConfig." + "reason": "Sam3VisionEncoderOutput missing attribute vs AutoConfig trust_remote_code missing fields; different model families and failure points." }, { "left": "issue:45375", "right": "issue:45478", - "accept": false, - "reason": "Both are Qwen3.5-Moe related, but one is a specific config-field drop and the other is a broader from_pretrained error without enough evidence of the same root cause." + "accept": true, + "reason": "Same underlying Qwen3.5-Moe from_pretrained/config regression: one report identifies the missing strict-dropped field, the other reports the resulting load failure on the same model family." }, { "left": "issue:28282", "right": "issue:30990", "accept": false, - "reason": "PyTorch import missing and Sentence Transformers loading stall are unrelated." + "reason": "ImportError for missing PyTorch vs Sentence Transformers loading stuck; not the same issue." }, { "left": "issue:41628", "right": "issue:44933", "accept": false, - "reason": "Both are import/export complaints around image utilities, but the reported missing symbols differ and the underlying paths are not clearly the same." + "reason": "Both are import-related, but they concern different missing symbols/modules and different API surfaces; not clearly the same bug." }, { "left": "issue:44964", "right": "issue:45137", "accept": false, - "reason": "Phi-4 multimodal loading and DeepSpeed ZeRO3 deque error are unrelated." + "reason": "Phi-4 multimodal model loading failure vs DeepSpeed ZeRO3 deque IndexError; unrelated." }, { "left": "issue:43986", "right": "issue:46726", - "accept": false, - "reason": "Both mention missing vision backends, but one is video-model loading without torchvision and the other is image-processor resolution; too different to merge confidently." - }, - { - "left": "issue:43646", - "right": "issue:44162", - "accept": false, - "reason": "Custom model init regression and ESM2 breakage may both be v5-related, but they are not the same concrete failure." - }, - { - "left": "issue:45310", - "right": "issue:45375", "accept": true, - "reason": "Both are Qwen3.5-Moe v5 from_pretrained/config-loading regressions, and the config-field-drop report plausibly explains the broader loading error." - }, - { - "left": "issue:44869", - "right": "issue:45356", - "accept": false, - "reason": "Whisper timestamp decode crash and Kimi tokenizer codec/regex regression are different tokenization bugs." - }, - { - "left": "issue:42994", - "right": "issue:43525", - "accept": false, - "reason": "Quantized save failure and Llama4Config pad_token_id AttributeError are separate issues." - }, - { - "left": "issue:44038", - "right": "issue:44623", - "accept": false, - "reason": "Qwen3-VL-Moe loading and processor save_pretrained missing files do not share the same bug." - }, - { - "left": "issue:45230", - "right": "issue:45397", - "accept": false, - "reason": "Generic bug report and gemma-4 zero3 from_pretrained regression are not comparable." + "reason": "Both are AutoProcessor failures caused by missing vision-related optional dependencies, with the same underlying class of backend-detection/error-reporting bug." }, { "left": "issue:39692", "right": "issue:46726", "accept": false, - "reason": "SigLIP2 docs/example errors and AutoProcessor missing-backend handling are different problems." + "reason": "SigLIP2 documentation/model/processor mismatch plus quantization issue vs generic missing-vision-backend processor error; different bugs." }, { "left": "issue:42371", "right": "issue:45092", "accept": false, - "reason": "TF32 API migration guidance and InternVL2 meta-init incompatibility are unrelated." + "reason": "TF32 API settings warning vs old InternVL2 remote-code/meta-init incompatibility; unrelated." }, { "left": "issue:43475", "right": "issue:45381", "accept": false, - "reason": "Sam3 video encoder output missing field and Qwen2.5-VL video position-id issue are different model-family regressions." + "reason": "Different SAM3 vs Qwen2.5-VL issues with no shared concrete failure path." }, { "left": "issue:43723", "right": "issue:46326", "accept": false, - "reason": "AutoTokenizer loading in v5 and mistral_common continue_final_message behavior are separate tokenizer issues." + "reason": "Tokenizer loading regression vs mistral_common continue_final_message bug; different areas." } ] }, "evaluator_result": { "accept": true, - "feedback": "The cluster summary is grounded in the packet, and the soft-edge calls are generally conservative. The only accepted pair, issue:45310 and issue:45375, is supported as a plausible Qwen3.5-Moe Transformers v5 loading/config regression, while the rest are reasonably treated as distinct by subsystem and failure mode. No obvious overstatement beyond the evidence in the titles and descriptions." + "feedback": "The summary is well grounded in the packet and stays conservative about the soft edges. The two accepted pairs are plausibly mergeable into single PRs for the same bugfix class (Qwen3.5-Moe config/load regression and AutoProcessor missing-vision-backend handling), and the rejected pairs are appropriately treated as separate bugs rather than overstated duplicates." }, "evaluator_used": true, "retried": false, @@ -13587,220 +13575,220 @@ "issue:43479|issue:44869", "issue:42673|issue:46693" ], - "prepared_review_unit_hash": "ebc5d4008d1f341715298087cd9e652b5be08de503a94bf99ec676e5f06a1d7a", + "prepared_review_unit_hash": "6c265b5263425e093d09e003061d69334649e67339d39f039c8df2c40eadb82a", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36619, - "estimated_input_tokens": 9155, - "estimated_eval_tokens": 18566 + "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 cluster is mostly heterogeneous and does not form a single duplicate set. The only clear duplicate pair is the Qwen3ForCausalLM VRAM leak report; all other soft pairs look related only by broad subsystem or surface similarity, not the same underlying bug.", - "confidence": 0.93, - "canonical_issue_reason": "issue 42673 is the clearest canonical issue because it is the older, more detailed report of the exact duplicate pair 42673/43704.", + "summary": "This cluster is mostly a mix of unrelated Transformers issues; only one soft pair is a clear duplicate. The exact duplicate is the Qwen3ForCausalLM VRAM leak report, where the earlier issue is the better anchor.", + "confidence": 0.92, + "canonical_issue_reason": "issue:42673 is the best canonical issue because it is the earlier, fuller report for the exact duplicate pair and has more discussion/inbound references than issue:43704.", "canonical_pr_reason": null, - "best_issue_reason": "issue 42673 is the best overall representative: it matches an exact duplicate, has the richer original report, and is the most suitable anchor for deduping that pair.", + "best_issue_reason": "issue:42673 is the strongest representative in the set: it is an exact duplicate of issue:43704 and has the richest signal for the underlying bug.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43784", "right": "issue:46399", "accept": false, - "reason": "Both are import/runtime failures in the transformers ecosystem, but they concern different symbols and different code paths (sentence-transformers import vs Qwen3.5 kernel registration)." + "reason": "Both are import/runtime failures, but they concern different symbols and code paths (sentence-transformers import vs Qwen3.5 rotary registration)." }, { "left": "issue:43232", "right": "issue:44188", "accept": false, - "reason": "Both involve generation internals, but one is about sync_gpus/model kwargs handling and the other is about attention-kernel divergence under torch.compile; not the same bug." + "reason": "Different generation-related bugs: one is sync_gpus/_update_model_kwargs_for_generation, the other is attention kernel divergence under torch.compile." }, { "left": "issue:43873", "right": "issue:44945", "accept": false, - "reason": "Quantization/offloading behavior and pipeline-parallel output correctness are different failure modes and would not be fixed by one change." + "reason": "One is a quantization/offloading problem, the other is incorrect output under pipeline parallelism; not the same underlying defect." }, { "left": "issue:45406", "right": "issue:46202", "accept": false, - "reason": "Both mention Gemma-related processing, but one is a missing _tokenizer attribute in serve and the other is a tokenizer roundtrip failure on a specific character." + "reason": "Both involve Gemma, but one is a processor attribute crash and the other is a token roundtrip failure; different bugs." }, { "left": "issue:44945", "right": "issue:45753", "accept": false, - "reason": "Pipeline parallelism output corruption and a Qwen3_5 image-triggered infinite loop are unrelated problems." + "reason": "Different models and symptoms: pipeline parallelism output corruption vs a Qwen3_5 image-specific infinite loop." }, { "left": "issue:41628", "right": "issue:44351", "accept": false, - "reason": "Both are top-level import errors, but they are for different missing exports (AutoImageProcessor vs HybridCache) and different fixes." + "reason": "Both are import errors, but for different exported names and likely different missing-export causes." }, { "left": "issue:46202", "right": "issue:46701", "accept": false, - "reason": "Both are Gemma/pipeline issues, but one is a text token roundtrip bug and the other is an image-text-to-text pipeline failure; too different to merge." + "reason": "Gemma tokenizer roundtrip failure is not the same as the DiffusionGemma image-text-to-text pipeline failure." }, { "left": "issue:42994", "right": "issue:43824", "accept": false, - "reason": "Quantized model saving and Qwen2_5_VL import errors are unrelated." + "reason": "Unrelated failures: quantized model saving vs missing Qwen2_5_VLForConditionalGeneration import." }, { "left": "issue:42673", "right": "issue:43704", "accept": true, - "reason": "These are exact title duplicates describing the same Qwen3ForCausalLM VRAM leak in multiple dataloader threads." + "reason": "Exact same bug and identical title: Qwen3ForCausalLM leaks VRAM in multiple dataloader threads." }, { "left": "issue:45081", "right": "issue:45103", "accept": false, - "reason": "Tokenizer regex patching and auto-docstring annotation processing are separate utilities with different crash conditions." + "reason": "Different crash sites and behaviors: tokenizer regex patching vs auto-docstring kwargs processing." }, { "left": "issue:43278", "right": "issue:45137", "accept": false, - "reason": "Embedding dtype drift in evaluate and a DeepSpeed ZeRO3 deque IndexError are unrelated." + "reason": "Embedding dtype drift and a DeepSpeed ZeRO3 deque IndexError are unrelated." }, { "left": "issue:44038", "right": "issue:45310", "accept": false, - "reason": "Both mention Qwen3.5 / transformers versioning, but one is a general compatibility bug and the other is a specific from_pretrained error for Qwen3.5-MoE." + "reason": "Both mention Qwen3.x model variants, but the reported failures and version contexts differ; not the same concrete bug." }, { "left": "issue:36010", "right": "issue:41628", "accept": false, - "reason": "Both are missing-symbol import issues, but the symbols and affected modules differ and do not indicate the same root cause." + "reason": "Both are import errors from transformers, but they are distinct missing symbols and distinct fixes." }, { "left": "issue:44222", "right": "issue:45310", "accept": false, - "reason": "FP8 save_pretrained for MoE and Qwen3.5 from_pretrained loading are different code paths and symptoms." + "reason": "FP8 save_pretrained behavior is different from a Qwen3.5 from_pretrained import/load failure." }, { "left": "issue:42831", "right": "issue:44222", "accept": false, - "reason": "Accuracy regressions in FineGrainedFP8 and FP8 save_pretrained failures are not the same bug." + "reason": "FineGrainedFP8 accuracy issues and FP8 save_pretrained failures are different problem classes." }, { "left": "issue:43232", "right": "issue:44792", "accept": false, - "reason": "Generation kwargs update logic and a janus image generation test failure are not the same underlying issue." + "reason": "Generation kwargs handling and a janus image generation test failure are not the same bug." }, { "left": "issue:42175", "right": "issue:44933", "accept": false, - "reason": "Backend dependency installation and a missing image_utils import are unrelated packaging/API problems." + "reason": "Packaging/backend dependency resolution and an image_utils import failure are unrelated." }, { "left": "issue:43504", "right": "issue:45561", "accept": false, - "reason": "A legacy field breaking BEiT pretrained loading and a pytest-xdist race on captured_info.txt are unrelated." + "reason": "Pretrained model config loading and a pytest-xdist file race are completely different issues." }, { "left": "issue:44496", "right": "issue:45561", "accept": false, - "reason": "Unrecognized model config and a testing race condition are completely different problems." + "reason": "Model config/model_type resolution and a test utility race condition are unrelated." }, { "left": "issue:30333", "right": "issue:33290", "accept": false, - "reason": "MLflow reporting on failed training jobs and Adafactor/DeepSpeed OOM are unrelated." + "reason": "MLFlow reporting after failed training and a deepspeed OOM are different failure modes." }, { "left": "issue:45561", "right": "issue:45600", "accept": false, - "reason": "A parallel test-file race and stale auto_mappings entries are different CI issues." + "reason": "A parallel test file race and stale auto_mappings references are not the same bug." }, { "left": "issue:43976", "right": "issue:44933", "accept": false, - "reason": "Python version compatibility and missing image_utils import are unrelated." + "reason": "Python version compatibility and a missing image_utils import are unrelated." }, { "left": "issue:30990", "right": "issue:35141", "accept": false, - "reason": "Sentence Transformers loading hangs and token-embedding resizing reinitialization are different bugs." + "reason": "Stuck loading vs token embedding reinitialization are different issues." }, { "left": "issue:42175", "right": "issue:45712", "accept": false, - "reason": "TensorFlow backend installation and leftover dummy classes in dummy_pt_objects.py are not the same change." + "reason": "Both touch packaging/dummy backend behavior, but one is missing tensorflow in extras and the other is leftover dummy classes leaking into dir()." }, { "left": "issue:42898", "right": "issue:44568", "accept": false, - "reason": "Tokenization cleanup behavior changes and mdeberta special-token insertion are separate tokenizer regressions." + "reason": "Both concern tokenizer behavior changes, but they affect different APIs and symptoms." }, { "left": "issue:43377", "right": "issue:44242", "accept": false, - "reason": "Missing padding-mask support in MIMI encoder and load-balancing loss omission in MoE routing are different model behaviors." + "reason": "Different MoE-related bugs: padding-mask batching mismatch vs missing load-balancing loss when router logits are disabled." }, { "left": "issue:46077", "right": "issue:46123", "accept": false, - "reason": "Bidirectional attention mask construction and MaskGenerationPipeline batch-finalization logic are unrelated." + "reason": "Bidirectional attention mask construction and partial-batch pipeline bookkeeping are unrelated." }, { "left": "issue:44623", "right": "issue:45310", "accept": false, - "reason": "Processor save_pretrained file emission and Qwen3.5 MoE loading are different areas and failures." + "reason": "Missing processor save files and Qwen3.5 from_pretrained errors are different code paths." }, { "left": "issue:43479", "right": "issue:44869", "accept": false, - "reason": "Default multimodal config initialization and Whisper timestamp decode crash are unrelated." + "reason": "Config default initialization and Whisper decode crashing on a trailing replacement character are unrelated." }, { "left": "issue:42673", "right": "issue:46693", "accept": false, - "reason": "VRAM leakage in dataloader threads and long-context FA2 performance regression are different performance bugs." + "reason": "VRAM leak in dataloader threads and a long-context FA2 performance regression are different performance problems." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is well grounded in the packet: the only clearly duplicate pair is 42673/43704, and the rest of the soft pairs are rejected conservatively with reasons that stay within the evidence. Canonical-issue choice and confidence look reasonable." + "feedback": "The cluster summary is well grounded in the packet: 42673 and 43704 have identical titles and are a clear duplicate pair, and the canonical choice is conservative and supported by earlier date plus richer discussion/body. The soft-edge rejections are also appropriately cautious and do not overclaim equivalence from title-level similarity alone." }, "evaluator_used": true, "retried": false, @@ -13902,220 +13890,220 @@ "issue:45245|issue:46132", "issue:43479|issue:46665" ], - "prepared_review_unit_hash": "3fcc1144980efcd86c1a33494c9db4f11cd99621e002da284e1cf7937c3e4c50", + "prepared_review_unit_hash": "308a6781a2b6a267258e66ff99972b45a946b4919ce9fabebb7327baa8824d6f", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 31, - "serialized_chars": 36514, - "estimated_input_tokens": 9129, - "estimated_eval_tokens": 18514 + "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 highly heterogeneous: the issues span unrelated tokenizer regressions, multimodal config/loading bugs, training/optimizer behavior, docs/notebook problems, and import errors. The soft-similarity links are largely false positives.", - "confidence": 0.93, - "canonical_issue_reason": "No single issue is canonical: the items cover unrelated bugs across tokenization, multimodal configs, optimizer/training, docs, and import errors.", + "summary": "This cluster is mostly a loose set of unrelated issues that only share high-level transformers/tokenizer/regression language. I did not find any soft pair that looks like the same underlying bug or change, so all pairwise duplicates should be rejected.", + "confidence": 0.46, + "canonical_issue_reason": "Issue 42898 is the broadest representative of the weakest shared theme here: a Transformers v5 tokenizer behavior regression. It is still not a true duplicate of the others, but it is the closest umbrella issue among the set.", "canonical_pr_reason": null, - "best_issue_reason": "issue:42898 is the closest thing to a cluster hub because several candidates are tokenizer/v5 regressions, but it still describes a distinct `clean_up_tokenization_spaces` behavior change.", + "best_issue_reason": "Issue 42898 is the best single anchor because it describes a core tokenizer regression with broad impact, whereas the other issues are mostly model-specific, import/config, docs, or unrelated training/runtime bugs.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44016", "right": "issue:46231", "accept": false, - "reason": "Notebook syntax error vs a documentation parameter-name mismatch; unrelated problems." + "reason": "Notebook syntax error and a docs parameter-name mismatch are unrelated." }, { "left": "issue:42898", "right": "issue:45812", "accept": false, - "reason": "Both involve tokenization, but one is `clean_up_tokenization_spaces` behavior and the other is Granite token IDs; different bugs." + "reason": "Both involve v5 tokenizer regressions, but they are different concrete bugs affecting different behaviors/models." }, { "left": "issue:43723", "right": "issue:44568", "accept": false, - "reason": "Tokenizer loading failure vs missing BOS/EOS on `add_special_tokens=True`; distinct code paths." + "reason": "Generic AutoTokenizer loading issue vs missing BOS/EOS insertion for a specific tokenizer; not the same bug." }, { "left": "issue:42222", "right": "issue:45200", "accept": false, - "reason": "VitPose model breakage vs Gemma 4 multimodal token-type defaults; unrelated model families and fixes." + "reason": "Different model families and different config defaults; no shared code-path bug." }, { "left": "issue:43479", "right": "issue:44079", "accept": false, - "reason": "Multimodal config initialization bug vs generic `ModelOutput` key handling; not the same issue." + "reason": "Multimodal config initialization and ModelOutput key assignment are unrelated internals." }, { "left": "issue:42898", "right": "issue:44779", "accept": false, - "reason": "Both are tokenizer regressions, but they affect different models and have different symptoms/causes." + "reason": "Both are tokenizer regressions in v5, but the symptoms and affected model behavior differ." }, { "left": "issue:43784", "right": "issue:43824", "accept": false, - "reason": "Sentence-transformers import error vs missing Qwen2_5_VL class import; separate import failures." + "reason": "Different import errors from different modules and causes." }, { "left": "issue:44734", "right": "issue:45103", "accept": false, - "reason": "KV-cache tensor indexing crash vs auto-docstring annotation parsing crash; unrelated subsystems." + "reason": "Serving KV-cache indexing bug and auto_docstring annotation crash are unrelated." }, { "left": "issue:43874", "right": "issue:45092", "accept": false, - "reason": "Missing image-processor method vs remote-code meta-init incompatibility; both multimodal, but not the same bug." + "reason": "Two different multimodal model-loading failures with different root causes." }, { "left": "issue:27957", "right": "issue:30333", "accept": false, - "reason": "Flash Attention 2 model support vs MLflow reporting after training failure; unrelated." + "reason": "Flash Attention 2 support and MLFlow reporting on failed jobs are unrelated." }, { "left": "issue:30333", "right": "issue:34689", "accept": false, - "reason": "MLflow job-status bug vs Llama 3.2 Vision model loading regression; different code paths." + "reason": "Training-job reporting and Llama vision model loading are separate problems." }, { "left": "issue:42898", "right": "issue:45356", "accept": false, - "reason": "Both are tokenizer regressions, but one is cleanup-space behavior and the other is codec handling/fix warning for Kimi-K2.5." + "reason": "Both are tokenizer regressions, but they concern different tokenization/codecs and warning behavior." }, { "left": "issue:42898", "right": "issue:43295", "accept": false, - "reason": "Tokenizer cleanup behavior vs processor/tokenizer regression with images; different APIs and failures." + "reason": "General tokenizer cleanup behavior vs processor/tokenizer access regression are different issues." }, { "left": "issue:43116", "right": "issue:44190", "accept": false, - "reason": "Text-classification example bug vs local dataset loading in image-classification example; unrelated examples." + "reason": "Example-script multi-label output bug and local dataset loading are unrelated." }, { "left": "issue:44060", "right": "issue:45125", "accept": false, - "reason": "Incorrect tied-weights warning vs missing tensor-parallel plan; both Qwen, but not the same concrete defect." + "reason": "Different Qwen model bugs: tied-weights warning vs missing tensor-parallel plan." }, { "left": "issue:45405", "right": "issue:45712", "accept": false, - "reason": "Dependency version bump issue vs leftover dummy class cleanup; unrelated." + "reason": "Released-version dependency bump and leftover dummy classes are unrelated maintenance issues." }, { "left": "issue:44295", "right": "issue:45656", "accept": false, - "reason": "Buffer read error vs optimizer stepping twice with DeepSpeed; unrelated runtime issues." + "reason": "Buffer registration/read bug and deepspeed optimizer-step duplication are different runtime problems." }, { "left": "issue:30333", "right": "issue:35141", "accept": false, - "reason": "MLflow reporting bug vs embedding reinitialization on resize; no shared underlying fix." + "reason": "MLFlow failure reporting vs embedding resize/post_init reinitialization are unrelated." }, { "left": "issue:22355", "right": "issue:28282", "accept": false, - "reason": "Import error for `transformers.onnx` vs missing PyTorch for `AutoModel`; both are import errors, but different causes." + "reason": "Two distinct import/dependency errors with different missing modules/libraries." }, { "left": "issue:44933", "right": "issue:45712", "accept": false, - "reason": "Missing import from `image_utils` vs dummy-class cleanup; unrelated import surfaces." + "reason": "Import exposure from image_utils and dummy_pt_objects cleanup are different packaging issues." }, { "left": "issue:44075", "right": "issue:46231", "accept": false, - "reason": "SGD arguments ignored vs documentation parameter-name typo; different severity and code paths." + "reason": "Optimizer behavior bug and documentation typo are unrelated." }, { "left": "issue:44222", "right": "issue:44297", "accept": false, - "reason": "FP8 MoE save-pretrained bug vs tokenizer class mismatch in tokenizer config; separate serialization problems." + "reason": "FP8 save_pretrained issue and tokenizer config mismatch are different save-path bugs." }, { "left": "issue:44360", "right": "issue:46762", "accept": false, - "reason": "DSA indexer discussion vs Minimax attention numerical differences; unrelated." + "reason": "DSA indexer discussion and Minimax attention numerical differences are unrelated." }, { "left": "issue:44448", "right": "issue:46132", "accept": false, - "reason": "Pegasus v4/v5 output regression vs AttentionInterface.register behavior change; different subsystems." + "reason": "Pegasus v4/v5 output regression and AttentionInterface.register behavior change are different code paths." }, { "left": "issue:42831", "right": "issue:44623", "accept": false, - "reason": "FineGrainedFP8 accuracy issue vs processor save-pretrained missing files; unrelated save/serialization behavior." + "reason": "FP8 accuracy issue and processor save_pretrained missing files are not the same bug." }, { "left": "issue:44297", "right": "issue:45656", "accept": false, - "reason": "Tokenizer config mismatch vs DeepSpeed double optimizer step; no common bug." + "reason": "Tokenizer save metadata mismatch has no relation to deepspeed optimizer stepping." }, { "left": "issue:42796", "right": "issue:43723", "accept": false, - "reason": "Ministral tokenizer BPE-marker bug vs tokenizer loading in v5; both tokenizer-related, but not the same defect." + "reason": "Both are tokenizer-related, but one is a model-specific BPE-marker bug and the other is a generic loading issue." }, { "left": "issue:43479", "right": "issue:43644", "accept": false, - "reason": "Phi4 multimodal config defaults vs non-persistent buffer junk; unrelated initialization problems." + "reason": "Config default initialization and non-persistent buffer filling are unrelated." }, { "left": "issue:44222", "right": "issue:44623", "accept": false, - "reason": "FP8 MoE save-pretrained bug vs processor save-pretrained missing files; both save-related, but different objects and failure modes." + "reason": "Different save_pretrained failures affecting different artifact types." }, { "left": "issue:45245", "right": "issue:46132", "accept": false, - "reason": "Category-count runtime limit vs AttentionInterface.register behavior change; unrelated." + "reason": "Category-count runtime error and attention registration behavior change are unrelated." }, { "left": "issue:43479", "right": "issue:46665", "accept": false, - "reason": "Multimodal config defaulting bug vs inconsistent `from_pretrained` usage inside Python classes; different underlying issues." + "reason": "Multimodal config initialization and from_pretrained behavior inside classes are different problems." } ] }, @@ -14220,214 +14208,214 @@ "issue:44261|issue:45092", "issue:22355|issue:27957" ], - "prepared_review_unit_hash": "cfb7b1f544afd1ee7b95648e0e065f9c0c1cffb083d7581835ebe20f1e0b40ed", + "prepared_review_unit_hash": "ad815562c3f38bca37d13e0a9fd156c050c1e3223f10f43634c75f20f3f1bd9d", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36310, - "estimated_input_tokens": 9078, - "estimated_eval_tokens": 18412 + "serialized_chars": 36306, + "estimated_input_tokens": 9077, + "estimated_eval_tokens": 18410 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These are mostly distinct bug reports across different Transformers v5 regressions and model-specific failures; none of the soft pairs look like true duplicates. If forced to pick an umbrella issue, 45071 is the broadest core-regression report, but it is still not a good duplicate match for the others.", - "confidence": 0.78, - "canonical_issue_reason": "45071 is the closest umbrella-style issue in this set because it reports a broad core regression in v5.4.0 (PreTrainedConfig type checking), but the cluster overall spans many different code paths.", + "summary": "The cluster is a grab bag of superficially similar Transformers issues, mostly around v5 regressions, save/load, generation, cache, or model-specific bugs, but the paired artifacts do not describe the same underlying defect. No soft edge is a safe duplicate merge.", + "confidence": 0.34, + "canonical_issue_reason": "issue:44263 is the best anchor for this cluster because it is the most concrete code-path bug and sits near the only small related GlmMoeDsaIndexer/DSA subset; it is more substantive than the nearby docstring-only item.", "canonical_pr_reason": null, - "best_issue_reason": "45071 is the most representative of the set as a general v5 regression report, even though the rest are not duplicates of it.", + "best_issue_reason": "issue:44263 is the strongest representative issue in the set: specific, technical, and the only one with adjacent near-match discussion, while the rest are mostly isolated model- or version-specific reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:41628", "right": "issue:42757", "accept": false, - "reason": "Different import failures from different modules: `transformers.onnx` vs `huggingface_hub.is_offline_mode`." + "reason": "Both are import errors, but for different symbols and different dependency layers; not the same bug." }, { "left": "issue:45071", "right": "issue:46082", "accept": false, - "reason": "One is a config type-checking regression; the other is LlamaConfig rejecting explicit `head_dim` when shapes are incompatible." + "reason": "One is type-checking for PretrainedConfig, the other is LlamaConfig head_dim validation; unrelated failures." }, { "left": "issue:44373", "right": "issue:44485", "accept": false, - "reason": "Docstring typo for `position_ids` is not the same bug as GLM-5 RoPE implementation behavior." + "reason": "Docstring mismatch versus RoPE implementation bug; different problem types and code paths." }, { "left": "issue:46132", "right": "issue:46710", "accept": false, - "reason": "Attention registration API behavior changes are unrelated to the DeepSeek accuracy regression after upgrade." + "reason": "AttentionInterface registration behavior and DeepSeek accuracy regression are not the same defect." }, { "left": "issue:44297", "right": "issue:45310", "accept": false, - "reason": "Both involve Qwen/tokenizer workflows, but one is tokenizer metadata saved incorrectly while the other is a `from_pretrained` load error; not the same code-path bug." + "reason": "Tokenizer save_pretrained metadata mismatch versus Qwen3.5 from_pretrained failure; too different and model-specific." }, { "left": "issue:43723", "right": "issue:44977", "accept": false, - "reason": "Generic tokenizer loading failure is different from Qwen3.5 flash-attention generation corruption." + "reason": "Tokenizer loading regression and flash-attention generation issue are unrelated." }, { "left": "issue:44222", "right": "issue:44484", "accept": false, - "reason": "FP8 `save_pretrained` on MoE models is unrelated to the `max_shard_size` default question." + "reason": "FP8 save_pretrained bug versus max_shard_size default discussion; not a duplicate." }, { "left": "issue:44190", "right": "issue:46082", "accept": false, - "reason": "A local dataset loading script issue is unrelated to LlamaConfig head-dimension validation." + "reason": "Dataset loading in a training script versus LlamaConfig head_dim validation; no shared code bug." }, { "left": "issue:43232", "right": "issue:46424", "accept": false, - "reason": "Generation kwarg update handling and cache-length config semantics are different bugs." + "reason": "Both mention generation/cache, but one is sync_gpus kwargs handling and the other cache length honoring; different paths." }, { "left": "issue:44206", "right": "issue:45103", "accept": false, - "reason": "Feature-extractor argument crash and auto-docstring annotation crash are unrelated." + "reason": "Feature extractor center-arg crash and auto_docstring annotation crash are unrelated." }, { "left": "issue:44263", "right": "issue:45092", "accept": false, - "reason": "A `torch.split`/indexer bug is not the same as remote-code checkpoint incompatibility with v5 meta initialization." + "reason": "GlmMoeDsaIndexer split handling and remote-code/meta-init incompatibility are different issues." }, { "left": "issue:43122", "right": "issue:46209", "accept": false, - "reason": "Tokenizer behavior drift across versions is unrelated to FP8 inference failing on B200 hardware." + "reason": "Tokenization regression versus FP8 inference on B200; no common underlying bug." }, { "left": "issue:43232", "right": "issue:45834", "accept": false, - "reason": "Generation KV-cache update logic is different from Kosmos2.5 OCR index errors." + "reason": "Generation cache kwargs and Kosmos2.5 OCR index error are unrelated." }, { "left": "issue:43475", "right": "issue:46305", "accept": false, - "reason": "Missing `fpn_position_embeddings` in SAM3 output is unrelated to the MoonshineStreaming sliding-window off-by-one." + "reason": "SAM 3 video encoder attribute error and MoonshineStreaming window off-by-one are different models and code paths." }, { "left": "issue:44734", "right": "issue:46421", "accept": false, - "reason": "Wrong tensor indexing in `/v1/responses` cache continuation is a different bug from cache lazy initialization causing recompilation." + "reason": "Serve KV-cache indexing crash versus chunked-prefill recompilation; both cache-related but not the same defect." }, { "left": "issue:46209", "right": "issue:46431", "accept": false, - "reason": "FP8 inference failure on B200 is unrelated to missing files during processor save_pretrained." + "reason": "FP8 B200 inference failure and missing chat template error are unrelated." }, { "left": "issue:46305", "right": "issue:46655", "accept": false, - "reason": "Both are video/streaming related, but one is a window-mask off-by-one and the other is a padding-calculation bug." + "reason": "Both are video-related, but one is a streaming mask off-by-one and the other is padding calculation in a different processor." }, { "left": "issue:44263", "right": "issue:44360", "accept": false, - "reason": "Both mention DSA indexer, but they describe different functional issues (`torch.split` returns vs missing ReLU)." + "reason": "Same DSA/GlmMoe area, but one is torch.split return handling and the other claims a missing ReLU; too different to merge." }, { "left": "issue:44295", "right": "issue:44623", "accept": false, - "reason": "Reading a registered `position_ids` buffer is not the same as processor save_pretrained omitting files." + "reason": "Reading registered position_ids and processor.save_pretrained missing files are separate save/serialization problems." }, { "left": "issue:30990", "right": "issue:38175", "accept": false, - "reason": "Sentence Transformers load hanging and SigLIP zero-probability outputs are unrelated." + "reason": "Stuck loading Sentence Transformers versus zero probabilities in siglip2; unrelated symptoms." }, { "left": "issue:44263", "right": "issue:44373", "accept": false, - "reason": "Indexer runtime behavior and a wrong docstring for `position_ids` are not duplicates." + "reason": "Code bug versus docstring issue; not the same underlying change." }, { "left": "issue:45381", "right": "issue:46305", "accept": false, - "reason": "Qwen2.5-VL video `vision_position_ids` mismatch is a different problem from MoonshineStreaming mask off-by-one." + "reason": "Qwen2.5-VL vision_position_ids bug and MoonshineStreaming frame-window bug are different models and logic." }, { "left": "issue:43644", "right": "issue:46305", "accept": false, - "reason": "Non-persistent buffer junk in v5.0.0 is unrelated to the streaming sliding-window mask bug." + "reason": "Non-persistent buffer initialization in v5 versus streaming mask off-by-one; unrelated." }, { "left": "issue:42907", "right": "issue:46050", "accept": false, - "reason": "Dequantized Ministrals/Devstrals save failure is different from sinq quantization breakage after save/load." + "reason": "Dequantized save failure and sinq save/load breakage both involve quantization, but they are not the same concrete bug." }, { "left": "issue:42831", "right": "issue:45310", "accept": false, - "reason": "FineGrainedFP8 accuracy issues are not the same as a Qwen3.5 MoE from_pretrained error." + "reason": "FineGrainedFP8 accuracy issue and Qwen3.5 Moe from_pretrained error are unrelated." }, { "left": "issue:44279", "right": "issue:46328", "accept": false, - "reason": "Generic dependency issues are unrelated to the `SequenceFeatureExtractor.pad` NumPy conversion inefficiency." + "reason": "Generic dependency issue and SequenceFeatureExtractor.pad performance regression are unrelated." }, { "left": "issue:45092", "right": "issue:45357", "accept": false, - "reason": "Old InternVL2 meta-init incompatibility is different from Qwen3.5 `save_pretrained` writing incorrect visual encoder keys." + "reason": "Remote-code/meta-init incompatibility and incorrect visual encoder keys on save_pretrained are different regressions." }, { "left": "issue:44829", "right": "issue:46002", "accept": false, - "reason": "Flash Attention 3 training degeneration and NaNs after checkout/init are not the same failure." + "reason": "Flash Attention 3 training degeneration and NaNs after checkout+init do not describe the same bug." }, { "left": "issue:44261", "right": "issue:45092", "accept": false, - "reason": "MLA RMS norm epsilon precision discussion is unrelated to meta-initialization incompatibility for old checkpoints." + "reason": "MLA RMS norm epsilon precision issue and remote-code meta-init incompatibility are distinct model bugs." }, { "left": "issue:22355", "right": "issue:27957", "accept": false, - "reason": "Missing `transformers.onnx` module and XLMRoberta Flash Attention 2 support are unrelated." + "reason": "Importing transformers.onnx and XLMRoberta Flash Attention 2 are unrelated." } ] }, @@ -14530,202 +14518,202 @@ "issue:46172|issue:46762", "issue:44117|issue:44704" ], - "prepared_review_unit_hash": "0ded377b74926a3fb33179168834b9102621216548a54f3a8039422bee24a1a5", + "prepared_review_unit_hash": "a8473b8e70675ec56ea36f0201ac5a696d31f6bec954c6eea69694059a1ada43", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 35507, - "estimated_input_tokens": 8877, - "estimated_eval_tokens": 18010 + "serialized_chars": 35512, + "estimated_input_tokens": 8878, + "estimated_eval_tokens": 18012 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is a loose Transformers issue cluster with several broad import/tokenizer/model-regression reports, but the soft pairs are not true duplicates of the same concrete bug.", - "confidence": 0.63, - "canonical_issue_reason": "issue:45278 is the broadest umbrella report in the set (\u201cmany import errors after update\u201d), making it the best representative issue for the cluster\u2019s shared regression theme.", + "summary": "This cluster is largely heterogeneous: most items are unrelated model, tokenizer, import, or runtime regressions that only share broad Hugging Face/Transformers vocabulary. No soft pair looks like a clear same-bug duplicate.", + "confidence": 0.94, + "canonical_issue_reason": "No strong duplicate group emerges; if one issue must represent the set, 45278 is the broadest umbrella because it describes multiple import failures after an upgrade, but the cluster overall is not cohesive.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45278 best captures the cluster\u2019s main recurring pattern of post-upgrade import/API breakage, even though the cluster is not a tight duplicate set.", + "best_issue_reason": "45278 is the most general cross-cutting report in the set and best fits as a loose representative for import-regression complaints, though it is still not a true canonical duplicate target.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44483", "right": "issue:45406", "accept": false, - "reason": "Different failures: chat-completions request validation vs Gemma4Processor missing _tokenizer." + "reason": "Different problems: API request rejection vs a processor attribute error." }, { "left": "issue:41628", "right": "issue:43824", "accept": false, - "reason": "Both are import errors, but they are for different missing symbols and different model areas." + "reason": "Both are import errors, but for different missing symbols/models and likely different causes." }, { "left": "issue:42831", "right": "issue:44038", "accept": false, - "reason": "FP8 accuracy regression is unrelated to the Qwen3-VL-Moe bug." + "reason": "Unrelated: FP8 accuracy regression vs a Qwen3-VL-Moe version issue." }, { "left": "issue:43408", "right": "issue:46402", "accept": false, - "reason": "Same family (Sam3) but different concrete bugs: model-type warning vs text encoder weights not loading." + "reason": "Both mention sam3, but one is a model-type warning and the other is missing text-encoder weights." }, { "left": "issue:22355", "right": "issue:28218", "accept": false, - "reason": "Missing transformers.onnx module is unrelated to tokenizer spacing after added tokens." + "reason": "No overlap: missing transformers.onnx module vs tokenizer spacing behavior." }, { "left": "issue:42831", "right": "issue:45753", "accept": false, - "reason": "Different model-specific issues: FineGrainedFP8 accuracy vs Qwen3_5 infinite loop on one image." + "reason": "Different model families and symptoms; not the same bug." }, { "left": "issue:39401", "right": "issue:44568", "accept": false, - "reason": "Both concern tokenization, but one is offset mapping and the other is special-token insertion on a different model." + "reason": "Both are tokenizer issues, but offset mappings and special-token insertion are distinct code paths." }, { "left": "issue:27957", "right": "issue:30990", "accept": false, - "reason": "Flash Attention 2 on XLMRoberta is unrelated to Sentence Transformers loading hangs." + "reason": "Flash Attention 2 support vs a loading hang; too different to merge." }, { "left": "issue:42757", "right": "issue:45278", "accept": false, - "reason": "45278 is a broad import-error report, but not the same concrete bug as the specific is_offline_mode import failure." + "reason": "45278 is a broad import-regression report, but 42757 is a specific dependency symbol import failure." }, { "left": "issue:42886", "right": "issue:43066", "accept": false, - "reason": "Offline cache loading failure and wrong decoder type are different tokenizer bugs." + "reason": "Offline cache loading and decoder-type selection are unrelated tokenizer issues." }, { "left": "issue:42886", "right": "issue:42994", "accept": false, - "reason": "Cache/offline loading and quantized model saving are unrelated." + "reason": "Offline cache loading is not the same as quantized model saving." }, { "left": "issue:44188", "right": "issue:44373", "accept": false, - "reason": "Torch.compile attention-kernel divergence is unrelated to a wrong docstring." + "reason": "Attention-kernel divergence and a docstring issue are unrelated." }, { "left": "issue:43408", "right": "issue:46369", "accept": false, - "reason": "One is a model-type warning; the other is a memory overflow in sam3-litetext." + "reason": "Same family name, but model warning vs memory overflow are different failures." }, { "left": "issue:43232", "right": "issue:45593", "accept": false, - "reason": "Generation kwarg propagation and D-FINE auxiliary-loss behavior are different code paths." + "reason": "Generation kwargs handling and D-FINE auxiliary-loss behavior are unrelated." }, { "left": "issue:44261", "right": "issue:44373", "accept": false, - "reason": "Precision error in MLA layernorm is unrelated to a docstring typo." + "reason": "Precision/config bug vs documentation mismatch." }, { "left": "issue:43646", "right": "issue:46002", "accept": false, - "reason": "Custom model initialization breakage and NaNs after checkout/init are not the same bug." + "reason": "Custom model init breakage and NaNs on checkout/init do not share a concrete code path." }, { "left": "issue:42796", "right": "issue:43122", "accept": false, - "reason": "Specific BPE-marker handling on one tokenizer is not the same as general tokenization drift across versions." + "reason": "Both tokenize differently, but one is BPE-marker handling and the other is version-to-version tokenization drift." }, { "left": "issue:42907", "right": "issue:44945", "accept": false, - "reason": "Quantized model saving failure and pipeline-parallel output corruption are different failures." + "reason": "Model saving after dequantization vs incorrect pipeline-parallel output are different problems." }, { "left": "issue:30990", "right": "issue:33357", "accept": false, - "reason": "Loading hang in Sentence Transformers is unrelated to a MacOS bus error on a CLIP model." + "reason": "Loading hang and MacOS bus error are both runtime issues, but not the same bug." }, { "left": "issue:43576", "right": "issue:43784", "accept": false, - "reason": "Broken env command and NameError importing sentence-transformers are different regressions." + "reason": "Broken CLI env command and an import NameError in sentence-transformers are unrelated." }, { "left": "issue:42371", "right": "issue:44263", "accept": false, - "reason": "TF32 settings API guidance is unrelated to a torch.split indexing bug." + "reason": "TF32 settings guidance vs a torch.split indexing issue are unrelated." }, { "left": "issue:43756", "right": "issue:44060", "accept": false, - "reason": "Missing RoPE layers and incorrect tied-weights warning are distinct model-architecture issues." + "reason": "Different model-specific architecture/weighting issues; not the same code-path bug." }, { "left": "issue:39401", "right": "issue:44488", "accept": false, - "reason": "Tokenizer offset mapping bug and a model load failure are different problems." + "reason": "Tokenizer offset mapping bug vs model loading failure for a specific checkpoint." }, { "left": "issue:42907", "right": "issue:44964", "accept": false, - "reason": "Dequantized-save failure is not the same as failing to load a multimodal model." + "reason": "Quantized/dequantized saving failure and multimodal model loading failure are distinct." }, { "left": "issue:43976", "right": "issue:44945", "accept": false, - "reason": "Python version compatibility regression is unrelated to incorrect pipeline-parallel outputs." + "reason": "Python version compatibility and pipeline-parallel output correctness are unrelated." }, { "left": "issue:44295", "right": "issue:44297", "accept": false, - "reason": "Reading registered position_ids and tokenizer_class mismatch in saved config are different serialization bugs." + "reason": "Buffer read/access error vs tokenizer_config class mismatch are different serialization bugs." }, { "left": "issue:46172", "right": "issue:46762", "accept": false, - "reason": "DeepSeek tensor-transpose issue and Minimax numerical mismatch are different numerical/code-path problems." + "reason": "Both are attention-related numerics, but on different models and likely different kernels." }, { "left": "issue:44117", "right": "issue:44704", "accept": false, - "reason": "Tokenizer mapping returning None and AutoProcessor not forwarding kwargs are separate API issues." + "reason": "Tokenizer mapping lookup and AutoProcessor kwargs forwarding are separate API plumbing issues." } ] }, @@ -14832,226 +14820,256 @@ "issue:43611|issue:46002", "issue:43116|issue:43493" ], - "prepared_review_unit_hash": "4892d81352c3bc0c3ed9690cf93a4f7c0006b64722646fc7d51e37a33db64558", + "prepared_review_unit_hash": "533cab0528bab437feb1dba92e8894e96d125c5aefaa020850c8fd15a8e993e1", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 32, - "serialized_chars": 36988, - "estimated_input_tokens": 9247, - "estimated_eval_tokens": 18750 + "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 a grab bag of unrelated Transformers regressions, import errors, tokenizer quirks, and model-specific bugs. None of the soft pairs look like the same underlying issue closely enough to merge, so all soft edges are rejected.", - "confidence": 0.84, - "canonical_issue_reason": "issue:45278 is the broadest umbrella report here: multiple import failures after upgrading to Transformers 5.5.0, which best matches the cluster\u2019s many import/regression-style issues.", + "summary": "The cluster is highly heterogeneous: mostly unrelated import, tokenizer, multimodal, and model-specific regressions. None of the soft pairs look like the same underlying bug/change, so all candidate duplicate edges should be rejected.", + "confidence": 0.92, + "canonical_issue_reason": "Issue 43122 is the broadest version-regression tokenizer report among the set and is the closest thematic anchor for the tokenizer-heavy subset, though the overall cluster is still quite mixed.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45278 is the most representative issue overall because it captures the widest shared theme in the set and can subsume nearby import-regression reports better than the narrower, model-specific bugs.", + "best_issue_reason": "43122 is the best representative issue because it is a broad, user-facing tokenizer regression and sits nearest to the small tokenizer subgroup; the rest of the cluster is too diverse to support a tighter canonical choice.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43784", "right": "issue:45600", "accept": false, - "reason": "Both are import-related, but one is a sentence-transformers NameError and the other is a CI break from removed auto_mappings configs; different code paths and fixes." + "reason": "Both are import-related, but one is a missing symbol in sentence-transformers integration and the other is a removed config reference in auto_mappings; different code paths." }, { "left": "issue:42907", "right": "issue:43937", "accept": false, - "reason": "Different model-specific failures: save_pretrained on dequantized Ministrals/Devstrals vs invalid GenerationConfig for GLM-5." + "reason": "Unrelated bugs: dequantized model saving versus GLM-5 generation config validation." }, { "left": "issue:41628", "right": "issue:45278", "accept": false, - "reason": "41628 is one specific missing export (AutoImageProcessor), while 45278 is a broad summary of many import errors; not the same concrete bug." + "reason": "One is a specific missing AutoImageProcessor export; the other is a broad upgrade-induced import breakage report." }, { "left": "issue:44625", "right": "issue:44829", "accept": false, - "reason": "Qwen3.5 num_labels propagation and flash_attention_3 training degeneration are unrelated bugs in different subsystems." + "reason": "Config propagation for Qwen3.5 is unrelated to a flash_attention_3 training degeneration bug." }, { "left": "issue:44704", "right": "issue:45290", "accept": false, - "reason": "One is missing kwargs propagation in AutoProcessor.from_pretrained; the other is chat-template crashing on tool-call assistant messages." + "reason": "Processor keyword forwarding to cached_file is a different bug than chat template crashing on tool-call assistant messages." }, { "left": "issue:39401", "right": "issue:44521", "accept": false, - "reason": "Tokenizer offset_mapping errors and assistant_mask generation failures are both tokenizer-adjacent, but not the same underlying defect." + "reason": "Both involve tokenization, but one is offset mapping and the other is assistant mask generation for multimodal chat templates." }, { "left": "issue:44360", "right": "issue:44485", "accept": false, - "reason": "DSA indexer ReLU concern and GLM-5 RoPE implementation are unrelated topics." + "reason": "These are unrelated: a DSA indexer discussion versus a GLM-5 RoPE implementation report." }, { "left": "issue:43408", "right": "issue:43475", "accept": false, - "reason": "Both mention SAM 3 Video, but one is a model-type warning and the other is a missing attribute on encoder output; not the same bug." + "reason": "Same SAM 3 Video area, but a model-type warning and a missing output attribute are distinct problems." }, { "left": "issue:43122", "right": "issue:44521", "accept": false, - "reason": "Tokenizer behavior regression versus multimodal assistant_mask bug; different failure modes and fixes." + "reason": "Tokenizer regression versus multimodal chat-template assistant mask failure; not the same code path." }, { "left": "issue:43122", "right": "issue:46326", "accept": false, - "reason": "Both involve tokenization behavior, but one is general tokenization differences across versions and the other is continue_final_message for mistral_common tokenizers." + "reason": "Both tokenizer-related, but one is a v5 regression in tokenization behavior and the other is continue_final_message handling for mistral_common tokenizers." }, { "left": "issue:44038", "right": "issue:44297", "accept": false, - "reason": "Qwen3-VL-Moe runtime issue and tokenizer_config save_pretrained mismatch are separate problems." + "reason": "Qwen3-VL-Moe import issues are unrelated to tokenizer.save_pretrained metadata mismatch." }, { "left": "issue:42907", "right": "issue:43452", "accept": false, - "reason": "Dequantized model saving and gguf_file loading failures are different code paths." + "reason": "Saving dequantized models and gguf loading failures are different serialization/loading bugs." }, { "left": "issue:22355", "right": "issue:29159", "accept": false, - "reason": "Missing transformers.onnx module and slow/fast tokenizer inconsistency are unrelated." + "reason": "Missing transformers.onnx module is unrelated to slow/fast tokenizer inconsistency." }, { "left": "issue:43116", "right": "issue:45200", "accept": false, - "reason": "Example-script multi-label output bug and mm_token_type_ids defaulting issue are different behaviors in different workflows." + "reason": "Empty multi-label classification results and Gemma 4 mm_token_type_ids handling are different issues." }, { "left": "issue:39401", "right": "issue:46614", "accept": false, - "reason": "Both involve tokenization-related code, but one is offset mapping and the other is stale espeak backend language state." + "reason": "Both tokenizer bugs, but offset_mapping for Qwen3 and stale phonemizer backend state for Wav2Vec2 are unrelated." }, { "left": "issue:45216", "right": "issue:45478", "accept": false, - "reason": "Both affect Qwen3.5 around save/from_pretrained, but one is saved-checkpoint corruption and the other is a loading error for the Moe variant; too different to merge." + "reason": "Both mention Qwen3.5 and save/load regressions, but one is checkpoint saving correctness and the other is from_pretrained loading failure." }, { "left": "issue:44038", "right": "issue:44222", "accept": false, - "reason": "Different FP8/MoE/model-loading issues; no shared concrete bug path." + "reason": "Qwen3-VL-Moe import problems and FP8 save_pretrained issues are separate model/serialization bugs." }, { "left": "issue:44038", "right": "issue:44295", "accept": false, - "reason": "Qwen3-VL-Moe regression and position_ids buffer/read error are unrelated." + "reason": "The Qwen3-VL-Moe import bug is not the same as reading a registered position_ids buffer." }, { "left": "issue:43824", "right": "issue:44351", "accept": false, - "reason": "Both are missing-import errors, but for different symbols and likely different release breaks." + "reason": "Different missing imports: Qwen2_5_VLForConditionalGeneration versus HybridCache." }, { "left": "issue:43122", "right": "issue:44779", "accept": false, - "reason": "General tokenization regression across versions versus Deepseek-specific incorrect tokenizer results; not the same bug." + "reason": "Both are tokenizer regressions, but they affect different models and behavior surfaces; no clear shared bug." }, { "left": "issue:43278", "right": "issue:44945", "accept": false, - "reason": "Training/eval dtype drift and pipeline-parallel output corruption are distinct distributed-training issues." + "reason": "BF16-to-FP32 dtype drift in evaluation is unrelated to incorrect pipeline-parallel LLM output." }, { "left": "issue:42757", "right": "issue:44351", "accept": false, - "reason": "Missing is_offline_mode and missing HybridCache are separate import failures from different API removals." + "reason": "Different libraries and symbols: huggingface_hub offline mode import versus transformers HybridCache import." }, { "left": "issue:36296", "right": "issue:43116", "accept": false, - "reason": "Tensor parallel training bug and multi-label classification example issue are unrelated." + "reason": "Tensor parallel training instability is unrelated to the classification example script returning empty results." }, { "left": "issue:22355", "right": "issue:29127", "accept": false, - "reason": "ONNX import failure and LayoutLMv3 error-message clarity issue are unrelated." + "reason": "No transformers.onnx module is unrelated to the LayoutLMv3 error message clarity issue." }, { "left": "issue:42175", "right": "issue:44938", "accept": false, - "reason": "TensorFlow backend packaging and Python 3.14 load failure are different environment/release issues." + "reason": "Backend packaging for torch installs and Python 3.14 load failures are different environment issues." }, { "left": "issue:43493", "right": "issue:45753", "accept": false, - "reason": "SigLIP2 implementation discrepancy and Qwen3_5 image infinite loop are unrelated model bugs." + "reason": "Different model families and failures: SigLIP2 implementation discrepancy versus Qwen3_5 infinite loop on one image." }, { "left": "issue:44188", "right": "issue:46328", "accept": false, - "reason": "torch.compile attention kernel branching and SequenceFeatureExtractor.pad performance issue are different code paths." + "reason": "Attention-kernel divergence under torch.compile and SequenceFeatureExtractor.pad inefficiency are unrelated." }, { "left": "issue:42898", "right": "issue:46614", "accept": false, - "reason": "v5 tokenization space-cleanup behavior change and stale phonemizer backend state are different tokenizer issues." + "reason": "v5 clean_up_tokenization_spaces behavior change is not the same as stale phonemizer language state." }, { "left": "issue:43937", "right": "issue:44485", "accept": false, - "reason": "GLM-5 generation config validation and GLM-5 RoPE implementation are separate problems even if they affect the same model family." + "reason": "GenerationConfig validation for GLM-5 is unrelated to the GLM-5 RoPE implementation discussion." }, { "left": "issue:38617", "right": "issue:45278", "accept": false, - "reason": "A specific missing layer_type_validation import and a broad set of import errors after upgrade are not the same concrete bug." + "reason": "A specific missing layer_type_validation import and a broad import-error-on-upgrade report are not the same bug." }, { "left": "issue:43611", "right": "issue:46002", "accept": false, - "reason": "base_model_prefix loading regression and NaNs in classification heads after checkout/init are unrelated." + "reason": "base_model_prefix loading regression and NaNs in classification heads are different training/loading problems." }, { "left": "issue:43116", "right": "issue:43493", "accept": false, - "reason": "Example-script classification output bug and SigLIP2 implementation mismatch are unrelated." + "reason": "Classification example output emptiness and SigLIP2 implementation discrepancy do not share a concrete code path." + }, + { + "left": "issue:42757", + "right": "issue:42757", + "accept": false, + "reason": "This pair is not present in the input soft edges; no edge to accept." + }, + { + "left": "issue:42898", + "right": "issue:44521", + "accept": false, + "reason": "Tokenization-space cleanup behavior and multimodal assistant-mask generation are different features." + }, + { + "left": "issue:44704", + "right": "issue:45278", + "accept": false, + "reason": "AutoProcessor kwargs forwarding and broad import errors after upgrade are unrelated." + }, + { + "left": "issue:44038", + "right": "issue:44297", + "accept": false, + "reason": "Qwen3-VL-Moe import issues are unrelated to tokenizer config metadata mismatch." + }, + { + "left": "issue:43122", + "right": "issue:44521", + "accept": false, + "reason": "Tokenizer regression and multimodal assistant mask failure are not the same underlying bug." } ] }, @@ -15153,202 +15171,196 @@ "issue:43673|issue:46328", "issue:42673|issue:46655" ], - "prepared_review_unit_hash": "5cafc7d2ee8714fcdf981283afbf1b80b2d8f4153d35bac81500b44624b01f6e", + "prepared_review_unit_hash": "0e3f8c2fbc2031b41a314222aab6be10c5332229f4f9392c41963fe58115f84d", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 28, - "serialized_chars": 35261, - "estimated_input_tokens": 8816, - "estimated_eval_tokens": 17888 + "serialized_chars": 35258, + "estimated_input_tokens": 8815, + "estimated_eval_tokens": 17886 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The items are mostly unrelated standalone issues. None of the soft pairs look like the same underlying bug/change, so I would not merge any of them as duplicates. The closest-looking groups are tokenizer regressions and Qwen/Qwen3 family issues, but the specific failure modes differ.", + "summary": "This cluster is heterogeneous: the proposed soft links span unrelated bugs across tokenizers, generation, attention, training, vision, and docs. I would not merge any of the soft pairs as duplicates; no PRs are present in the cluster.", "confidence": 0.82, - "canonical_issue_reason": "issue:44779 is the clearest, most concrete user-facing regression among the set, with a specific model/version scope and reproducible tokenizer behavior. That said, this cluster does not appear to contain a true duplicate group.", + "canonical_issue_reason": "issue:33453 is the broadest, most generic tokenizer regression ticket here, so it is the best single issue to anchor the cluster if one must be chosen.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44779 is the best single anchor issue because it is specific, actionable, and clearly describes a regression rather than a broad symptom.", + "best_issue_reason": "issue:33453 is the most suitable representative because it is general enough to cover adjacent tokenizer-loading reports, whereas the other issues are more model-specific or on different code paths.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44355", "right": "issue:46231", "accept": false, - "reason": "Compiled-file runtime errors and a docs parameter-name mismatch are unrelated." + "reason": "Compiled-file execution errors and a docs parameter-name mismatch are unrelated." }, { "left": "issue:43976", "right": "issue:45446", "accept": false, - "reason": "Python-version compatibility and a PyTorch version check in flex_attention are different problems." + "reason": "Python version compatibility failure vs. a flex_attention import guard bug; different root causes." }, { "left": "issue:43066", "right": "issue:44779", "accept": false, - "reason": "Both are tokenizer-related regressions, but one is decoder type in v5 and the other is DeepSeek output correctness; different code paths." + "reason": "Both are tokenizer-related, but one is a decoder-type regression and the other is DeepSeek output correctness; not the same concrete bug/fix." }, { "left": "issue:44038", "right": "issue:45478", "accept": false, - "reason": "Same broad Qwen/Moe family, but different model versions and different failures (generic bug vs from_pretrained error)." + "reason": "Different Qwen model families and different from_pretrained failure modes." }, { "left": "issue:42673", "right": "issue:45072", "accept": false, - "reason": "VRAM leak in Qwen3ForCausalLM and dtype mismatches in unrelated models are not the same bug." + "reason": "VRAM leak in dataloader-thread usage vs. bfloat16 dtype mismatch in inference; unrelated." }, { "left": "issue:44625", "right": "issue:46002", "accept": false, - "reason": "Config propagation for num_labels and NaNs in classification heads are distinct symptoms with different likely causes." + "reason": "num_labels propagation bug and NaNs in classification heads are distinct issues." }, { "left": "issue:45083", "right": "issue:45446", "accept": false, - "reason": "A feature-extractor helper bug and a flex_attention import/version check are unrelated." + "reason": "Feature-length helper behavior and a PyTorch version check are unrelated." }, { "left": "issue:45357", "right": "issue:45478", "accept": false, - "reason": "Both involve Qwen3.5/Moe save/load paths, but one is incorrect saved keys while the other is a from_pretrained load error; not clearly the same defect." + "reason": "Both touch Qwen3.5, but one is save_pretrained key serialization and the other is a load-time from_pretrained error; not the same bug." }, { "left": "issue:42371", "right": "issue:43874", "accept": false, - "reason": "TF32 settings guidance and a missing image-processor method are unrelated." + "reason": "TF32 settings API guidance and a missing image-patch method are unrelated." }, { "left": "issue:42222", "right": "issue:43441", "accept": false, - "reason": "VitPose model breakage and a FlashAttention failure in Ministral are different model-specific issues." + "reason": "vitpose breakage and Ministral FlashAttention failures have different code paths and symptoms." }, { "left": "issue:43116", "right": "issue:43825", "accept": false, - "reason": "Example-script multi-label output bug and an incorrect pipeline() error message are different issues." + "reason": "Multi-label classification example failure vs. pipeline() translation error messaging; unrelated." }, { "left": "issue:44075", "right": "issue:44936", "accept": false, - "reason": "Unused SGD arguments and a train-then-evaluate failure do not point to the same code path." + "reason": "Optimizer SGD args not being used and trainer.evaluate() failing after training are different problems." }, { "left": "issue:43992", "right": "issue:44704", "accept": false, - "reason": "Missing embed_tokens weights in UMT5Encoder and kwargs not forwarded in AutoProcessor.from_pretrained are unrelated." + "reason": "Missing embed_tokens weight in UMT5Encoder and AutoProcessor kwarg forwarding are unrelated." }, { "left": "issue:44995", "right": "issue:46697", "accept": false, - "reason": "A stale cache crash in GlmMoeDsa and over-padding in Swin2SR image processing are different components." + "reason": "GlmMoe stale indexer cache vs. Swin2SR over-padding are different subsystems and bugs." }, { "left": "issue:28348", "right": "issue:46640", "accept": false, - "reason": "Both mention FlashAttention 2, but one is a GPT2 support request and the other is a generic open issue; not the same change/bug." + "reason": "FlashAttention 2 support request for GPT2 and a generic Flash Attention issue are not the same concrete change or bug." }, { "left": "issue:43066", "right": "issue:46710", "accept": false, - "reason": "Tokenizer decoder-type regression and a DeepSeek accuracy regression are different problems." + "reason": "Tokenizer decoder-type regression and a DeepSeek accuracy regression after upgrade are not the same underlying defect." }, { "left": "issue:46614", "right": "issue:46752", "accept": false, - "reason": "Stale phonemizer language state and apply_chat_template on empty input are unrelated." + "reason": "Stale phonemizer backend state and apply_chat_template() crashing on empty input are unrelated." }, { "left": "issue:44263", "right": "issue:46762", "accept": false, - "reason": "A torch.split return-value issue in GlmMoeDsa is unrelated to Minimax attention numerical differences." + "reason": "torch.split return-value handling in GlmMoeDsaIndexer and Minimax MSA numerical differences are different issues." }, { "left": "issue:43827", "right": "issue:45446", "accept": false, - "reason": "Documentation still referencing pipeline() and a flex_attention version check are unrelated." + "reason": "Docs still referencing pipeline() and a flex_attention import version check are unrelated." }, { "left": "issue:45092", "right": "issue:46762", "accept": false, - "reason": "Remote-code checkpoint meta-init incompatibility and numerical differences in Minimax attention are different failure modes." + "reason": "InternVL2 meta-init incompatibility and Minimax MSA numerical drift are unrelated." }, { "left": "issue:43064", "right": "issue:43673", "accept": false, - "reason": "FSDP2/PEFT optimizer-state corruption and missing GenerationMixin cache during chunked prefill are unrelated." + "reason": "Wrong optimizer state placement under FSDP2/PEFT and missing GenerationMixin cache during chunked_prefill are different code paths." }, { "left": "issue:42673", "right": "issue:42796", "accept": false, - "reason": "VRAM leakage in Qwen3ForCausalLM and tokenizer BPE-marker handling are different bugs." + "reason": "A VRAM leak in multithreaded loading and a tokenizer BPE-marker handling bug are unrelated." }, { "left": "issue:43937", "right": "issue:45563", "accept": false, - "reason": "Invalid GenerationConfig and a stale warning about num_return_sequences are not the same issue." + "reason": "GenerationConfig validation failure and a stale warning about num_return_sequences are different symptoms and fixes." }, { "left": "issue:33453", "right": "issue:39401", "accept": false, - "reason": "Generic tokenizer-loading regression and Qwen3 offset_mapping bug are different tokenizer issues." + "reason": "Tokenizer loading regression and wrong offset_mapping are related area-wise but not the same underlying bug." }, { "left": "issue:44380", "right": "issue:46077", "accept": false, - "reason": "GPT2 attention scaling under SDPA/FlashAttention and Gemma4 bidirectional-mask construction are different model/backend bugs." + "reason": "GPT2 attention scaling ignored under SDPA/FlashAttention and Gemma4 bidirectional mask construction are different model-specific issues." }, { "left": "issue:45356", "right": "issue:46614", "accept": false, - "reason": "Kimi tokenizer codec/regression issues and Wav2Vec2 phonemizer state leakage are unrelated tokenizer subsystems." - }, - { - "left": "issue:43673", - "right": "issue:46328", - "accept": false, - "reason": "Generation cache handling and SequenceFeatureExtractor padding performance are unrelated." + "reason": "Kimi tokenizer codec/warning regression and a stale phonemizer language backend are unrelated." }, { "left": "issue:42673", "right": "issue:46655", "accept": false, - "reason": "A VRAM leak in Qwen3ForCausalLM and a video padding calculation bug in SmolVLM are different issues." + "reason": "VRAM leak in multithreaded usage and incorrect video padding calculation are unrelated." } ] }, @@ -15452,214 +15464,214 @@ "issue:44933|issue:45278", "issue:38617|issue:44220" ], - "prepared_review_unit_hash": "28f2e89c0ddf9689b27bb58a1654fba357bccb9abe06a942cbdee34349f0628d", + "prepared_review_unit_hash": "3bbfbafbffa06188b8b79113d2fa47ab4bc420244b0ef07b22e5964feb18e10f", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 30, - "serialized_chars": 35935, - "estimated_input_tokens": 8984, - "estimated_eval_tokens": 18224 + "serialized_chars": 35930, + "estimated_input_tokens": 8983, + "estimated_eval_tokens": 18222 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is thematically related by transformer v5-era regressions, but the soft-paired items are not the same underlying bugs. I rejected all soft edges and picked the broad import-regression report as the best cluster anchor.", - "confidence": 0.76, - "canonical_issue_reason": "issue:45278 is the most representative cluster anchor: it describes a broad wave of import errors after upgrading from 4.57.0 to 5.5.0, which matches the general regression theme better than the narrower model-specific reports.", + "summary": "This cluster is highly heterogeneous: most items are unrelated model-loading, tokenizer, config, testing, or performance bugs. I found no soft-edge pair that clearly looks like the same underlying defect, so I rejected all candidate merges. The closest shared theme is import/API regression issues, where issue 45278 is the broadest representative.", + "confidence": 0.61, + "canonical_issue_reason": "issue:45278 is the broadest umbrella report in the cluster (multiple import errors after upgrading from 4.57.0 to 5.5.0) and is the best representative of the small import-regression subset.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45278 is the best single issue for this cluster because it is broad, user-facing, and plausibly encompasses multiple narrower import-breakage reports without being overly specific.", + "best_issue_reason": "issue:45278 best captures the only meaningful mini-theme here (import/API breakage); the rest of the cluster is too diverse to support a true duplicate canonical.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44291", "right": "issue:45071", "accept": false, - "reason": "Both are v5 regressions, but one is an init_empty_weights TypeError and the other is PretrainedConfig type-checking; different code paths and symptoms." + "reason": "Different bugs: one is an unexpected `_is_hf_initialized` argument during empty-weight model loading, the other is `PretrainedConfig` type checking. Same area, not same defect." }, { "left": "issue:42222", "right": "issue:46752", "accept": false, - "reason": "Vitpose model loading failures and empty-list chat template crashes are unrelated bugs in different subsystems." + "reason": "Completely different failures: broken vitpose models vs `apply_chat_template` crashing on an empty list." }, { "left": "issue:45405", "right": "issue:45446", "accept": false, - "reason": "A PEFT version pin issue and a PyTorch version check bug are both dependency-related, but they are not the same defect." + "reason": "PEFT version bump availability issue vs a PyTorch version check bug in flex attention." }, { "left": "issue:44117", "right": "issue:45588", "accept": false, - "reason": "TOKENIZER_MAPPING_NAMES returning None and a flash-attention crash on sink-less models are unrelated code paths." + "reason": "Tokenizer mapping fallback bug vs flash-attention crash on `s_aux=None`; unrelated code paths." }, { "left": "issue:44466", "right": "issue:46002", "accept": false, - "reason": "Inconsistent lm_head serialization and NaNs in classification heads are different model-state problems." + "reason": "Serialization of tied weights vs NaNs in classification heads; different symptoms and fixes." }, { "left": "issue:44410", "right": "issue:45310", "accept": false, - "reason": "Missing qwen3next projections and Qwen3.5 from_pretrained errors affect different model families and failure modes." + "reason": "Missing Qwen3Next projections vs Qwen3.5 MoE `from_pretrained` error; not the same underlying model bug." }, { "left": "issue:39401", "right": "issue:44779", "accept": false, - "reason": "Both are tokenizer bugs, but they concern different models and different tokenizer behaviors (offset mapping vs output correctness)." + "reason": "Both tokenizer-related, but one is offset mapping and the other is incorrect Deepseek tokenization results; no clear shared root cause." }, { "left": "issue:43937", "right": "issue:44977", "accept": false, - "reason": "GLM-5 generation config validation and Qwen3.5 flash-attention generation issues are distinct problems." + "reason": "Invalid `GenerationConfig` vs flash-attention generation failure; different failure modes." }, { "left": "issue:45071", "right": "issue:46512", "accept": false, - "reason": "PretrainedConfig type-checking breakage and stale config dtype after from_config are different behaviors in separate APIs." + "reason": "Type-checking regression vs stale `config.dtype` after `from_config`; not the same bug." }, { "left": "issue:45341", "right": "issue:45561", "accept": false, - "reason": "A testing_utils bug and a pytest-xdist race in patched testing utils are related only by file area, not by the same underlying bug." + "reason": "A small testing-utils bug vs an xdist race on `captured_info.txt`; different code paths." }, { "left": "issue:43450", "right": "issue:43986", "accept": false, - "reason": "Incorrect batched video-processor shapes and an AutoProcessor crash without torchvision are separate video-related failures." + "reason": "Batched video processor shape bug vs crash loading a video model without torchvision; separate issues." }, { "left": "issue:45341", "right": "issue:46328", "accept": false, - "reason": "A testing utility bug and a SequenceFeatureExtractor.pad performance issue are different problems." + "reason": "Testing-utils typo/bug vs `SequenceFeatureExtractor.pad` inefficiency; unrelated." }, { "left": "issue:44829", "right": "issue:45216", "accept": false, - "reason": "Degenerate training with flash_attention_3 and a Qwen3.5 save_pretrained regression are unrelated issues." + "reason": "Degenerate training with flash attention 3 vs incorrect checkpoint serialization for Qwen3.5; not mergeable as one fix." }, { "left": "issue:44117", "right": "issue:45081", "accept": false, - "reason": "A tokenizer mapping None bug and a Mistral regex patch AttributeError are different tokenizer internals." + "reason": "Tokenizer mapping `None` handling vs Mistral regex patch crash from missing `backend_tokenizer`; different concrete bugs." }, { "left": "issue:42617", "right": "issue:45341", "accept": false, - "reason": "3d_parallel.py execution failure and a testing_utils bug do not share the same underlying cause." + "reason": "3d_parallel.py runtime issue vs a testing-utils bug; no overlap beyond being in the repo." }, { "left": "issue:38617", "right": "issue:41628", "accept": false, - "reason": "Both are import errors, but they are for different missing symbols and different modules." + "reason": "Both are import errors, but for different missing symbols (`layer_type_validation` vs `AutoImageProcessor`) and likely different fixes." }, { "left": "issue:43408", "right": "issue:43742", "accept": false, - "reason": "A sam3_video/sam3_tracker model-type warning and a MobileLLM key error are unrelated loading issues." + "reason": "SAM3 tracker/video-model warning vs MobileLLM config/key error; unrelated model-loading problems." }, { "left": "issue:45362", "right": "issue:46328", "accept": false, - "reason": "Qwen3.5 chat crashes and a padding performance regression are not the same bug." + "reason": "Qwen3.5 chat crash vs a padding performance issue; different severity and cause." }, { "left": "issue:43122", "right": "issue:46298", "accept": false, - "reason": "Tokenization changes across versions and a noisy logging message are unrelated." + "reason": "Tokenizer behavior regression across versions vs noisy logging; no duplicate relationship." }, { "left": "issue:44877", "right": "issue:46231", "accept": false, - "reason": "Strict config rejection for granite_speech and a documentation parameter-name mismatch are different classes of issue." + "reason": "Strict config preventing granite_speech loading vs a documentation parameter-name mismatch." }, { "left": "issue:42222", "right": "issue:46178", "accept": false, - "reason": "Vitpose model breakage and a tensorflow_text backend-mapping error are unrelated import/loading issues." + "reason": "Vitpose model breakage vs backend mapping missing `tensorflow_text`; both loader-related but not the same bug." }, { "left": "issue:44484", "right": "issue:45310", "accept": false, - "reason": "A save_pretrained shard-size question and a Qwen3.5 from_pretrained error are unrelated." + "reason": "Shard-size question vs Qwen3.5 MoE loading error; no shared underlying defect." }, { "left": "issue:42371", "right": "issue:44261", "accept": false, - "reason": "TF32 behavior controls and MLA q_a_layernorm precision/config issues are distinct." + "reason": "TF32 API-setting guidance vs missing `rms_norm_eps` causing precision error; unrelated changes." }, { "left": "issue:39401", "right": "issue:46209", "accept": false, - "reason": "Qwen3 tokenizer offset mapping and FP8 inference failure on B200 are different model/runtime problems." + "reason": "Tokenizer offset mapping bug vs FP8 inference failure on B200; different subsystems." }, { "left": "issue:43937", "right": "issue:44496", "accept": false, - "reason": "Invalid GenerationConfig and unrecognized model config are separate validation failures." + "reason": "Invalid generation config vs unrecognized model/config `model_type`; both are config validation issues but not the same underlying problem." }, { "left": "issue:42371", "right": "issue:46617", "accept": false, - "reason": "TF32 API settings and trust_remote_code cache FileNotFoundError are unrelated." + "reason": "TF32 settings guidance vs trust-remote-code cache FileNotFoundError; unrelated." }, { "left": "issue:44410", "right": "issue:45478", "accept": false, - "reason": "Both mention Qwen3.5/Qwen3next-family loading, but the specific broken code paths and model variants differ." + "reason": "Missing projections in qwen3next vs Qwen3.5 MoE `from_pretrained` error; separate model-specific bugs." }, { "left": "issue:44261", "right": "issue:46762", "accept": false, - "reason": "A missing rms_norm_eps precision bug and numerical differences in Minimax MSA attention are not the same issue." + "reason": "Both mention numerical differences, but they are different models/implementations and the causes are not clearly the same." }, { "left": "issue:44933", "right": "issue:45278", "accept": false, - "reason": "A missing import from image_utils is a specific API break, while 45278 is a broad upgrade-wide import regression report." + "reason": "`image_utils` import problem may be one example of broader import breakage, but the artifacts do not clearly describe the same concrete bug." }, { "left": "issue:38617", "right": "issue:44220", "accept": false, - "reason": "A missing configuration_utils import and a _torch_extract_fbank_features issue are unrelated." + "reason": "Missing import symbol vs `_torch_extract_fbank_features()` issue; unrelated." } ] }, @@ -15767,20 +15779,20 @@ "issue:38617|issue:42757", "issue:44410|issue:45375" ], - "prepared_review_unit_hash": "d0767258c00a8eb0d599bb0b478a790ca08107a28d39466704626d3af4a599b1", + "prepared_review_unit_hash": "495d8368e6ed12bc415842324c028d55e8d73bead4247c650e91522b8263740c", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 33, - "serialized_chars": 37239, + "serialized_chars": 37240, "estimated_input_tokens": 9310, "estimated_eval_tokens": 18876 }, @@ -15789,219 +15801,216 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated v5/model-loading regressions with no accepted soft duplicates. The tokenizer reports (43066/43122) are the closest pair, but the packet does not justify merging them, so the duplicate claim stays conservative.", - "confidence": 0.57, - "canonical_issue_reason": "Issue 43122 is the best representative of the tokenizer-regression subgroup because it describes the user-visible version-to-version tokenization drift most directly; however, it is not strong enough evidence to merge it with 43066 as a proven duplicate.", + "summary": "The set is mostly a grab bag of superficially similar Transformers bugs (imports, model loading, FlashAttention, Qwen/GLM variants) rather than true duplicates. I would not merge any of the suggested soft pairs.", + "confidence": 0.73, + "canonical_issue_reason": "If one issue must anchor the cluster, issue 43097 is the broadest regression-style report with downstream references and a clear v5 impact; the rest are narrower, model-specific failures.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 43122 is the strongest anchor in this cluster for the tokenizer behavior-change thread, but the 43066 pair remains only soft similarity and should not be treated as confirmed duplicate evidence.", + "best_issue_reason": "Issue 43097 is the best representative because it describes a concrete, user-visible Transformers v5 regression in a general API area, rather than a highly model-specific or incidental bug.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44075", "right": "issue:45656", "accept": false, - "reason": "Different optimizer bugs: ignored SGD args vs double optimizer stepping under deepspeed." + "reason": "Both are optimizer-related, but one is about SGD args being ignored and the other about double optimizer steps under deepspeed; different bugs." }, { "left": "issue:44263", "right": "issue:44485", "accept": false, - "reason": "Different GLM internals and symptoms; no concrete shared code path." + "reason": "Different GLM internals: torch.split return handling in a DSA indexer vs a RoPE implementation discussion." }, { "left": "issue:43066", "right": "issue:43122", "accept": false, - "reason": "Tokenizer-related, but only soft similarity is present and deterministic_accept=false; not enough to confirm the same bug." + "reason": "Both touch tokenizer behavior in v5, but one reports a wrong decoder type and the other a tokenization regression; not the same concrete bug." }, { "left": "issue:43097", "right": "issue:44493", "accept": false, - "reason": "Both are v5 regressions, but one is API removal and the other is a position-id key issue." + "reason": "Unrelated regressions: removed tie_embeddings_and_encoder_decoder vs unexpected key errors on position ids." }, { "left": "issue:43673", "right": "issue:45341", "accept": false, - "reason": "Unrelated areas: generation cache/chunked prefill vs a testing utility bug." + "reason": "Chunked prefill cache handling is unrelated to a testing_utils bug." }, { "left": "issue:41628", "right": "issue:45071", "accept": false, - "reason": "Different failures: import error vs PreTrainedConfig type-checking regression." + "reason": "Importing AutoImageProcessor and PretrainedConfig type checking are different failure modes in different code paths." }, { "left": "issue:42222", "right": "issue:45542", "accept": false, - "reason": "Both mention loading/backend issues, but they are different model and TF/tensorboard problems." + "reason": "VitPose model breakage is unrelated to the TensorFlow/tensorboard backend detection issue." }, { "left": "issue:43931", "right": "issue:44410", "accept": false, - "reason": "Same broad Qwen3-VL family, but the reported failures are different code paths and symptoms." + "reason": "Both are Qwen-family issues, but one is a weight-shape mismatch on Qwen3-VL and the other is missing projections in qwen3next; different bugs." }, { "left": "issue:45071", "right": "issue:45542", "accept": false, - "reason": "Completely different regressions: type-checking vs TensorFlow backend detection." + "reason": "PretrainedConfig type checking and undefined tf backend errors are unrelated." }, { "left": "issue:44483", "right": "issue:45464", "accept": false, - "reason": "Both hit chat/completions, but one is a generic API rejection and the other is a model-specific streaming failure." + "reason": "One is a generic chat/completions API rejection, the other is a streaming inference failure on a specific Qwen model." }, { "left": "issue:44484", "right": "issue:45478", "accept": false, - "reason": "Unrelated: save_pretrained sharding-size question vs Qwen3.5-MoE loading error." + "reason": "Saving shard size behavior is unrelated to Qwen3.5-MoE from_pretrained loading errors." }, { "left": "issue:43874", "right": "issue:44261", "accept": false, - "reason": "Different multimodal bugs: missing image-patch method vs RMS norm epsilon precision issue." + "reason": "Different GLM issues: missing image patch count method vs missing rms_norm_eps in MLA layernorm." }, { "left": "issue:43012", "right": "issue:43577", "accept": false, - "reason": "Both are dtype-related, but one is a compile warning and the other is a load-time dtype mismatch." + "reason": "Both involve dtype/precision, but one is a PyTorch warning during compile and the other is incorrect dtype propagation in Blip2 loading." }, { "left": "issue:44360", "right": "issue:46172", "accept": false, - "reason": "Different model internals and symptoms; no clear shared code path." + "reason": "A missing ReLU in a DSA indexer is unrelated to the DeepSeek-v4 tensor transpose issue." }, { "left": "issue:44263", "right": "issue:46172", "accept": false, - "reason": "Different GLM/DeepSeek implementation bugs with no concrete overlap." + "reason": "torch.split return values in GlmMoeDsaIndexer and a DeepSeek-v4 comb tensor op are different bugs." }, { "left": "issue:27957", "right": "issue:43352", "accept": false, - "reason": "Both involve Flash Attention 2, but for different models and different failure modes." + "reason": "Both mention Flash Attention 2, but they are separate model-specific support gaps for different architectures." }, { "left": "issue:43931", "right": "issue:44038", "accept": false, - "reason": "Related Qwen3-VL variants, but the titles suggest distinct loading/shape issues rather than the same bug." + "reason": "Different Qwen variants and different symptoms; no evidence they share the same loading failure." }, { "left": "issue:43441", "right": "issue:44829", "accept": false, - "reason": "Flash-attention-related, but one is unsupported FA2 on a model and the other is degenerate training under FA3." + "reason": "One is a FlashAttention failure on Ministral-3, the other is degenerate training with flash_attention_3; different code paths and outcomes." }, { "left": "issue:43097", "right": "issue:43452", "accept": false, - "reason": "Unrelated regressions: embedding-tie removal vs gguf_file preload handling." + "reason": "Removed tie_embeddings support and gguf_file loading failures are unrelated." }, { "left": "issue:45003", "right": "issue:45341", "accept": false, - "reason": "modeling_utils sys.modules access bug is unrelated to the testing_utils issue." + "reason": "Unsafe sys.modules access in modeling_utils is unrelated to a testing_utils bug." }, { "left": "issue:43986", "right": "issue:45341", "accept": false, - "reason": "Processor/torchvision import crash is unrelated to the testing utility bug." + "reason": "Video-model loading without torchvision and a testing_utils issue are unrelated." }, { "left": "issue:44933", "right": "issue:46231", "accept": false, - "reason": "One is a missing import, the other is a documentation parameter mismatch." + "reason": "A missing import from image_utils is unrelated to a documentation parameter-name mismatch." }, { "left": "issue:44261", "right": "issue:44485", "accept": false, - "reason": "Both are GLM-related, but the precision/config epsilon bug is separate from the RoPE implementation discussion." + "reason": "Both are GLM-related, but the former is a numerical epsilon/config issue and the latter is about RoPE implementation." }, { "left": "issue:45753", "right": "issue:46369", "accept": false, - "reason": "Different model families and different failure modes: infinite loop vs memory overflow." + "reason": "An infinite-loop bug on a specific image is unrelated to memory overflow in sam3-litetext." }, { "left": "issue:44117", "right": "issue:44987", "accept": false, - "reason": "Both are from_pretrained/load regressions, but the titles point to different root causes." + "reason": "A tokenizer mapping None handling bug is not the same as loading failure for a specific repo." }, { "left": "issue:43531", "right": "issue:45341", "accept": false, - "reason": "Qwen3-MoE sliding_window bug is unrelated to the testing utility issue." + "reason": "Qwen3-MoE sliding_window behavior is unrelated to testing_utils." }, { "left": "issue:43874", "right": "issue:44263", "accept": false, - "reason": "Different failures in different multimodal/model components." + "reason": "Both are GLM46V/GLM-related, but they concern different missing methods and different failure points." }, { "left": "issue:43673", "right": "issue:46123", "accept": false, - "reason": "Generation cache handling vs pipeline partial-batch finalization are not the same bug." + "reason": "Generation cache missing in chunked_prefill is unrelated to MaskGenerationPipeline batch-finalization logic." }, { "left": "issue:45440", "right": "issue:46172", "accept": false, - "reason": "DeepseekV3MoE divergence and Deepseek-V4 tensor conversion are distinct implementation issues." + "reason": "DeepSeekV3MoE divergence from remote implementation is not the same as the DeepSeek-v4 comb dtype/transpose bug." }, { "left": "issue:43097", "right": "issue:45588", "accept": false, - "reason": "Embedding-tie removal and flash_attention s_aux crash are unrelated." + "reason": "Tie-embeddings removal and flash_attention.py crashing on s_aux=None are different regressions." }, { "left": "issue:43986", "right": "issue:45372", "accept": false, - "reason": "Both are import/loading crashes, but caused by different missing dependencies/modules." + "reason": "Both involve loading processors, but one is a missing torchvision dependency and the other is a missing mistral_common symbol." }, { "left": "issue:38617", "right": "issue:42757", "accept": false, - "reason": "Both are ImportErrors, but from different modules and likely different dependency/version breaks." + "reason": "Both are import errors, but they concern different modules and different missing symbols." }, { "left": "issue:44410", "right": "issue:45375", "accept": false, - "reason": "Both are Qwen3 family issues, but they refer to different missing fields/projections." + "reason": "Missing attention projections in qwen3next is unrelated to a missing deepstack_visual_indexes config field." } ] }, - "evaluator_result": { - "accept": false, - "feedback": "The summary overstates the evidence for the tokenizer pair. In the packet, every soft-pair verdict has deterministic_accept=false, including issue:43066 \u2194 issue:43122, so marking that pair accepted is not grounded. Please keep the duplicate claim conservative or reject the pair unless there is stronger packet evidence." - }, - "evaluator_used": true, - "retried": true, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null @@ -16104,238 +16113,238 @@ "issue:44077|issue:46365", "issue:45584|issue:46365" ], - "prepared_review_unit_hash": "ac71e646b107f6634bca3738a853deee80e3ae5a3ec5a8ebe727f457dcd23f9f", + "prepared_review_unit_hash": "223900239b6c1f853952f4fed5a6a948a0b3b7fee6a5fe81b321b1a5ae25bc44", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 34, - "serialized_chars": 37681, - "estimated_input_tokens": 9421, - "estimated_eval_tokens": 19098 + "serialized_chars": 37670, + "estimated_input_tokens": 9418, + "estimated_eval_tokens": 19092 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No soft pair looks like a true duplicate; the similarities are mostly broad subsystem overlap (v5, quantization, processors, tokenizers) rather than the same concrete bug. I\u2019d anchor the cluster on the active quantization/offloading regression in issue 43873, but it is not a duplicate of the other items.", - "confidence": 0.76, - "canonical_issue_reason": "Issue 43873 is the best anchor: it is open, has the most discussion/inbound references here, and describes a broad user-facing quantization/offloading failure with active investigation. The other issues are more specific, one-off regressions in different code paths.", + "summary": "The items are a heterogeneous set of unrelated Transformers bugs/regressions, so none of the soft pairs look like true duplicates. No PRs were provided. If a representative issue is needed, the most central unresolved report is the quantization/offloading bug.", + "confidence": 0.44, + "canonical_issue_reason": "issue:43873 is the most active open, user-facing bug in the list and the broadest reasonable anchor among the unrelated reports.", "canonical_pr_reason": null, - "best_issue_reason": "43873 has the clearest ongoing impact and the strongest triage signal, so it is the most suitable representative issue in this set even though it does not subsume the others.", + "best_issue_reason": "issue:43873 has the strongest combination of activity, open status, and general bug relevance, so it is the best representative issue here.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43064", "right": "issue:44315", "accept": false, - "reason": "Different failures: optimizer-state corruption under v5+FSDP2+PEFT vs Liger Kernel not being applied with model_init. Same ecosystem, not the same bug." + "reason": "Different failures: FSDP/PEFT optimizer state corruption vs Liger Kernel application during model_init." }, { "left": "issue:43097", "right": "issue:46050", "accept": false, - "reason": "One is a missing/removed config API in v5; the other is broken save/load for quantized sinq models. Different code paths and symptoms." + "reason": "Different save/load problems: removed tie_embeddings_and_encoder_decoder vs quantized model round-trip breakage." }, { "left": "issue:44038", "right": "issue:44410", "accept": false, - "reason": "Both involve Qwen-family models, but one is a v5 bug on Qwen3-VL-Moe while the other is missing projections in qwen3next layer 0. Not the same defect." + "reason": "Different models and symptoms: Qwen3-VL-Moe v5 bug vs qwen3next missing layer projections." }, { "left": "issue:43986", "right": "issue:44484", "accept": false, - "reason": "AutoProcessor/video-model loading without torchvision is unrelated to the 50GB shard-size default in save_pretrained()." + "reason": "Different code paths: AutoProcessor/video-model torchvision dependency crash vs save_pretrained shard-size default." }, { "left": "issue:46655", "right": "issue:46664", "accept": false, - "reason": "Both are SmolVLM video-processor bugs, but one is padding calculation and the other is the meaning of num_frames. Related area, distinct issues." + "reason": "Same subsystem but distinct bugs: padding math vs num_frames semantics in SmolVLM video processing." }, { "left": "issue:46298", "right": "issue:46752", "accept": false, - "reason": "Noisy logging vs apply_chat_template crashing on an empty list are unrelated behaviors." + "reason": "Noisy logging vs apply_chat_template IndexError are unrelated behaviors." }, { "left": "issue:42890", "right": "issue:46298", "accept": false, - "reason": "Missing set_seed in a model integration test is a test-flakiness issue, not a logging-message bug." + "reason": "Seed-related test flakiness is unrelated to an unspecified noisy logging message." }, { "left": "issue:38617", "right": "issue:44351", "accept": false, - "reason": "Both are import errors, but for different missing symbols: layer_type_validation vs HybridCache. Different breakages." + "reason": "Both are import errors, but for different symbols and modules with different root causes." }, { "left": "issue:46172", "right": "issue:46710", "accept": false, - "reason": "DeepSeek-V4 dtype/transpose bug is a concrete model implementation issue; the other is an accuracy regression after a version upgrade." + "reason": "Different models and failure modes: tensor transpose bug vs post-upgrade accuracy regression." }, { "left": "issue:43097", "right": "issue:43873", "accept": false, - "reason": "One concerns a removed config attribute in v5; the other is an offloading/quantization runtime failure. Same broad area, not the same bug." + "reason": "Both mention v5-era regressions, but one is a removed config API and the other is quantization/offloading." }, { "left": "issue:44062", "right": "issue:44117", "accept": false, - "reason": "Tokenizer AddedToken special-arg duplication is different from TOKENIZER_MAPPING_NAMES returning None during from_pretrained()." + "reason": "Tokenizer AddedToken constructor error is unrelated to TOKENIZER_MAPPING_NAMES returning None." }, { "left": "issue:44117", "right": "issue:46701", "accept": false, - "reason": "Tokenizer mapping failures are unrelated to the DiffusionGemma image-text-to-text pipeline breakage." + "reason": "Tokenizer mapping lookup failure is unrelated to a diffusion/image-text pipeline crash." }, { "left": "issue:44936", "right": "issue:46365", "accept": false, - "reason": "trainer.evaluate() failing after train is a lifecycle/runtime bug; import Trainer failing because datasets is installed is an import-time dependency issue." + "reason": "trainer.evaluate() after train() is a runtime state bug; importing Trainer depends on a package conflict." }, { "left": "issue:43502", "right": "issue:43992", "accept": false, - "reason": "local_files_only still causing network calls is unrelated to UMT5Encoder missing embed_tokens.weight on load." + "reason": "local_files_only network leakage is unrelated to missing embed_tokens.weight in UMT5Encoder loading." }, { "left": "issue:44704", "right": "issue:45103", "accept": false, - "reason": "AutoProcessor not forwarding kwargs to cached_file is unrelated to the auto_docstring AttributeError with future annotations." + "reason": "Both involve loading/doc plumbing, but cached_file kwarg forwarding and auto_docstring annotation parsing are separate." }, { "left": "issue:44380", "right": "issue:46421", "accept": false, - "reason": "GPT2 attention scaling ignored under SDPA/FlashAttention is a model backend bug, not the chunked-prefill recompilation issue." + "reason": "Attention scaling being ignored is unrelated to cache lazy initialization causing recompilation." }, { "left": "issue:43986", "right": "issue:44117", "accept": false, - "reason": "Video-model loading without torchvision and tokenizer mapping None are different failure modes in different subsystems." + "reason": "Video-model torchvision import failure is unrelated to tokenizer mapping None handling." }, { "left": "issue:44938", "right": "issue:46178", "accept": false, - "reason": "Python 3.14 import/load failure is not the same as an invalid BACKENDS_MAPPING entry for tensorflow_text." + "reason": "Python 3.14 load failure is a compatibility issue; tensorflow_text backend validation is a different backend-mapping bug." }, { "left": "issue:42907", "right": "issue:44655", "accept": false, - "reason": "Saving dequantized Ministrals/Devstrals and saving Pipeline objects are both save-pretrained topics but distinct bugs." + "reason": "Both touch saving, but one is dequantized model persistence and the other is Pipeline.save_pretrained support." }, { "left": "issue:42907", "right": "issue:45753", "accept": false, - "reason": "Quantized model save/load failure is unrelated to Qwen3_5 looping on a specific image." + "reason": "Dequantized save failure and Qwen3_5 image-triggered infinite loop are unrelated." }, { "left": "issue:43825", "right": "issue:44655", "accept": false, - "reason": "Incorrect translation-task messaging in pipeline() is unrelated to Pipeline save_pretrained support." + "reason": "Pipeline translation error messaging is unrelated to Pipeline.save_pretrained serialization support." }, { "left": "issue:43208", "right": "issue:43550", "accept": false, - "reason": "xLSTM training-size bugs are not the same as Bamba torch.compile+SDPA failure." + "reason": "xLSTM training blockers and Bamba torch.compile/SDPA failure are different model-specific issues." }, { "left": "issue:43012", "right": "issue:43550", "accept": false, - "reason": "A PyTorch precision warning on bfloat16 compilation is different from a compile-time SDPA model failure." + "reason": "PyTorch bfloat16 warning is not the same bug as torch.compile failing with SDPA." }, { "left": "issue:45712", "right": "issue:46298", "accept": false, - "reason": "Dummy classes leaking into dir(transformers) is a packaging/surface-area bug, not a logging issue." + "reason": "Leftover dummy classes/check_repo failures are unrelated to a noisy logging message." }, { "left": "issue:43012", "right": "issue:44368", "accept": false, - "reason": "Both mention warnings, but one is a float32 precision warning during compile and the other is a tie_word_embeddings config warning." + "reason": "Both are warnings, but they concern different configs and different warning sources." }, { "left": "issue:43992", "right": "issue:44484", "accept": false, - "reason": "Missing embed_tokens.weight during UMT5Encoder load is unrelated to the 50GB shard-size default in save_pretrained()." + "reason": "Missing checkpoint weights and max_shard_size default are not the same problem." }, { "left": "issue:43066", "right": "issue:43906", "accept": false, - "reason": "Wrong tokenizer decoder type in v5 is unrelated to an isolated reproduction of another issue." + "reason": "Wrong tokenizer decoder type is unrelated to reproducing issue #38071." }, { "left": "issue:43352", "right": "issue:43976", "accept": false, - "reason": "Nemotron FlashAttention support and Python version compatibility are different problems." + "reason": "Flash Attention support gating is unrelated to a Python version compatibility failure." }, { "left": "issue:43352", "right": "issue:44380", "accept": false, - "reason": "Lack of FlashAttention 2.0 support in Nemotron is distinct from GPT2 attention scaling being ignored under SDPA/FlashAttention." + "reason": "One is missing Flash Attention support; the other is GPT2 attention scaling being ignored under SDPA/FlashAttention." }, { "left": "issue:43097", "right": "issue:46298", "accept": false, - "reason": "A removed config API in v5 does not match a noisy logging-only complaint." + "reason": "A removed config API and a noisy logging message do not share an underlying bug." }, { "left": "issue:44162", "right": "issue:46172", "accept": false, - "reason": "ESM2 being broken broadly is not the same concrete bug as a DeepSeek dtype/transpose issue." + "reason": "ESM2 breakage is unrelated to the DeepSeek transpose bug." }, { "left": "issue:44380", "right": "issue:46132", "accept": false, - "reason": "GPT2 backend-specific scaling loss is unrelated to AttentionInterface.register changing function behavior." + "reason": "Ignored attention scaling and changed AttentionInterface.register behavior are different code-path problems." }, { "left": "issue:44077", "right": "issue:46365", "accept": false, - "reason": "patchtsmixer post_init policy and Trainer import failure due to datasets are unrelated." + "reason": "PatchTSMixer post_init policy and importing Trainer with datasets installed are unrelated." }, { "left": "issue:45584", "right": "issue:46365", "accept": false, - "reason": "Whisper empty-transcription generation failure is unrelated to an import-time Trainer/datasets problem." + "reason": "Whisper empty-transcription generation bug is unrelated to the Trainer import conflict." } ] }, @@ -16444,245 +16453,242 @@ "issue:36683|issue:44625", "issue:29870|issue:36296" ], - "prepared_review_unit_hash": "835ad73b76db09dce24f49c72422aec2268172ade17f597d331d154188739167", + "prepared_review_unit_hash": "038dbae25c8610cd449fffeb3b3f8d3f32131a97d615466b486d07d6d01abb5c", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 34, - "serialized_chars": 37638, - "estimated_input_tokens": 9410, - "estimated_eval_tokens": 19076 + "serialized_chars": 37644, + "estimated_input_tokens": 9411, + "estimated_eval_tokens": 19078 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a heterogeneous set of Transformers issues: import errors, config regressions, model-specific load/save bugs, and a few backend/runtime problems. None of the soft-similarity pairs are close enough to treat as the same underlying bug or change.", - "confidence": 0.83, - "canonical_issue_reason": "issue:36683 is the most central and actively discussed node in the set, with multiple soft links around it, but it is still a specific Gemma3Config vocab_size bug rather than a duplicate of the others.", + "summary": "Most items are unrelated Transformers bug reports. The only soft edge that looks like the same underlying regression is the CLIP v5 output bug; the rest should stay separate.", + "confidence": 0.58, + "canonical_issue_reason": "issue:43761 is the best representative of the only clear duplicate cluster: it names the v5 CLIPVisionModel.forward regression directly and is broader than the attentions-only symptom.", "canonical_pr_reason": null, - "best_issue_reason": "issue:36683 is the best representative issue here because it is the most concrete and well-discussed bug among a largely unrelated mix of regressions.", + "best_issue_reason": "issue:43761 is the most actionable and umbrella-like report for the CLIP regression, with a concrete model/flag and a clear failure mode.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44222", "right": "issue:45478", "accept": false, - "reason": "Both involve Qwen3.5 MoE/FP8, but one is save_pretrained and the other is from_pretrained; different code paths and failure modes." + "reason": "Same Qwen3.5 MoE family, but one is FP8 save_pretrained and the other is from_pretrained loading; different code paths." }, { "left": "issue:36683", "right": "issue:43582", "accept": false, - "reason": "Gemma3Config missing vocab_size is unrelated to an Apple Silicon caching_allocator_warmup TypeError." + "reason": "Gemma3 config attribute bug vs Apple Silicon allocator warmup TypeError; unrelated subsystems." }, { "left": "issue:44380", "right": "issue:45479", "accept": false, - "reason": "Attention scaling under SDPA/FlashAttention is a different bug from zero-loss sequence classification behavior." + "reason": "Attention-scaling backend bug and degenerate classification loss are different concrete failures." }, { "left": "issue:42175", "right": "issue:46365", "accept": false, - "reason": "Packaging/import problems, but one is TF backend omission for pip extras and the other is Trainer import shadowed by datasets." + "reason": "Packaging/backend install issue vs Trainer import shadowing; not the same bug." }, { "left": "issue:43064", "right": "issue:44380", "accept": false, - "reason": "FSDP2/PEFT optimizer-state corruption is not the same as GPT2 attention scaling being ignored." + "reason": "FSDP/PEFT optimizer-state corruption is unrelated to GPT2 attention scaling." }, { "left": "issue:42757", "right": "issue:43824", "accept": false, - "reason": "Both are ImportErrors, but they reference different missing symbols from different modules." + "reason": "Both are import errors, but for different symbols in different packages." }, { "left": "issue:42490", "right": "issue:43012", "accept": false, - "reason": "Local-path trailing-slash warning is unrelated to the bfloat16 compilation warning." + "reason": "Local path import warning and bfloat16 compile warning are distinct issues." }, { "left": "issue:43012", "right": "issue:44683", "accept": false, - "reason": "Both are compile-related, but one is a precision warning while the other is a torch>=2.9 flex_attention failure." + "reason": "One is a warning, the other a flex_attention compile failure; not the same bug." }, { "left": "issue:43208", "right": "issue:43526", "accept": false, - "reason": "xLSTM training bugs are unrelated to BeitImageProcessorFast reduce_labels returning the wrong label array." + "reason": "xLSTM training blockers and BeitImageProcessor label reduction are unrelated." }, { "left": "issue:44483", "right": "issue:44514", "accept": false, - "reason": "API request validation failure is different from batched chat-template processing crashing with padding=False." + "reason": "Chat/completions request rejection and batched processor crash are different code paths." }, { "left": "issue:44038", "right": "issue:44230", "accept": false, - "reason": "Qwen3-VL-Moe loading regression and fp8 inference support are different model/feature issues." + "reason": "Qwen3-VL-Moe loading regression vs FP8 inference support request; not one underlying bug." }, { "left": "issue:43618", "right": "issue:43761", - "accept": false, - "reason": "Both are CLIP regressions, but one is missing attentions and the other is hidden_states not being returned." - }, - { - "left": "issue:43618", - "right": "issue:44380", - "accept": false, - "reason": "CLIP output regression and GPT2 attention-scaling backend behavior are different code paths." + "accept": true, + "reason": "Both point to the same CLIP v5 forward/output regression where expected model outputs are no longer populated." }, { "left": "issue:38617", "right": "issue:43502", "accept": false, - "reason": "Import cleanup for layer_type_validation is unrelated to local_files_only still triggering API requests." + "reason": "Missing import symbol and unwanted network access with local_files_only=True are unrelated." }, { "left": "issue:43756", "right": "issue:44485", "accept": false, - "reason": "Smollm3 dropping RoPE layers and GLM-5 RoPE implementation discussion are not the same bug." + "reason": "Different RoPE/model-implementation issues in different model families." }, { "left": "issue:42994", "right": "issue:44655", "accept": false, - "reason": "Quantized model saving failure is distinct from saving Pipeline objects with save_pretrained." + "reason": "Quantized model serialization failure vs Pipeline.save_pretrained support are different save paths." }, { "left": "issue:43299", "right": "issue:43867", "accept": false, - "reason": "Qwen3VL-Moe loading regression is not the same as a generic state_dict-sorting load error." + "reason": "Different model-loading regressions: base_model_prefix handling vs sorted state_dict loading." }, { "left": "issue:44625", "right": "issue:46365", "accept": false, - "reason": "Qwen3.5 num_labels propagation is unrelated to the Trainer import issue caused by datasets." + "reason": "Config propagation bug vs Trainer import failure; no shared underlying change." }, { "left": "issue:45276", "right": "issue:46077", "accept": false, - "reason": "Embedding resize propagation and bidirectional attention mask construction are separate Gemma4 bugs." + "reason": "Gemma4 embedding resize bug and Gemma4 attention-mask bug affect different mechanisms." }, { "left": "issue:36683", "right": "issue:43475", "accept": false, - "reason": "Missing vocab_size in Gemma3Config is unrelated to missing fpn_position_embeddings in Sam3VisionEncoderOutput." + "reason": "Gemma3 vocab_size AttributeError and SAM3 fpn_position_embeddings AttributeError are model-specific and unrelated." }, { "left": "issue:36010", "right": "issue:42757", "accept": false, - "reason": "Different missing imports from different modules; not the same underlying defect." + "reason": "Different import regressions from different modules/symbols." }, { "left": "issue:43352", "right": "issue:43937", "accept": false, - "reason": "FlashAttention 2 unsupported for NemotronHForCausalLM is unrelated to invalid GenerationConfig for GLM-5." + "reason": "Flash Attention support limitation and invalid GenerationConfig are unrelated." }, { "left": "issue:44483", "right": "issue:46369", "accept": false, - "reason": "Chat/completions request rejection and SAM3 memory overflow are unrelated runtime failures." + "reason": "API request handling and SAM3 memory overflow are different problems." }, { "left": "issue:43611", "right": "issue:44655", "accept": false, - "reason": "base_model_prefix load regression is unrelated to pipeline serialization." + "reason": "base_model_prefix loading regression is unrelated to saving Pipeline objects." }, { "left": "issue:43901", "right": "issue:44655", "accept": false, - "reason": "Docs still mentioning return_all_scores is a documentation mismatch, not a save_pretrained bug." + "reason": "Documentation mismatch vs pipeline serialization support; not a duplicate bug." }, { "left": "issue:36683", "right": "issue:44077", "accept": false, - "reason": "Gemma3Config vocab_size and patchtsmixer post_init policy are unrelated." + "reason": "Gemma3 config attribute bug and patchtsmixer post_init policy are unrelated." }, { "left": "issue:43867", "right": "issue:45216", "accept": false, - "reason": "A sorted state_dict load error is a different bug from Qwen3.5 checkpoint corruption in save_pretrained." + "reason": "Both involve checkpoint loading/saving, but the symptoms and code paths differ too much to treat as one bug." }, { "left": "issue:43976", "right": "issue:44938", "accept": false, - "reason": "Both are Python-version compatibility issues, but they affect different Python versions and failure modes." + "reason": "Separate Python-version compatibility regressions." }, { "left": "issue:44625", "right": "issue:46077", "accept": false, - "reason": "num_labels propagation and bidirectional attention masks are separate Gemma4/Qwen3.5 issues." + "reason": "Qwen3.5 num_labels propagation and Gemma4 bidirectional masking are unrelated." }, { "left": "issue:46077", "right": "issue:46082", "accept": false, - "reason": "Gemma4 attention-mask behavior and LlamaConfig head_dim validation are different configuration bugs." + "reason": "Gemma4 attention-mask behavior and LlamaConfig head_dim validation are different issues." }, { "left": "issue:43618", "right": "issue:45216", "accept": false, - "reason": "CLIP attentions regression and Qwen3.5 save_pretrained regression do not share a concrete code path." + "reason": "CLIP output regression and Qwen3.5 checkpoint regression are unrelated." }, { "left": "issue:36683", "right": "issue:43525", "accept": false, - "reason": "Gemma3Config missing vocab_size is unrelated to Llama4Config missing pad_token_id." + "reason": "Different config attribute errors on different model families." }, { "left": "issue:36683", "right": "issue:44625", "accept": false, - "reason": "Gemma3Config vocab_size and Qwen3.5 num_labels propagation are different config-sync bugs." + "reason": "Different Qwen3.5/Gemma3 config propagation bugs." }, { "left": "issue:29870", "right": "issue:36296", "accept": false, - "reason": "Bart generate decoder_attention_mask initialization bug is unrelated to tensor-parallel training problems." + "reason": "BART generation mask error and tensor-parallel training bug are unrelated." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The clustering is grounded in the packet and stays conservative overall. The single accepted edge, issue:43618 \u2194 issue:43761, is well-supported as the same CLIP v5 output-regression family, while the other pairs are reasonably kept separate based on distinct titles/code paths." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -16780,214 +16786,214 @@ "issue:44734|issue:46617", "issue:43723|issue:44556" ], - "prepared_review_unit_hash": "95216ab8192ae7a18eecc3d9c5e99e5aefd498c69d8ad3d9ae3cb04e01a11723", + "prepared_review_unit_hash": "f650e39021f15998b3dacf93c207f3f6935a8e823270a9d8eb52edf56d2c61fc", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36279, - "estimated_input_tokens": 9070, - "estimated_eval_tokens": 18396 + "serialized_chars": 36272, + "estimated_input_tokens": 9068, + "estimated_eval_tokens": 18392 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated issues that only share broad themes like v5 regressions, import/load failures, and model-specific bugs. I did not find any soft-edge pair that looks like the same underlying defect, so all soft edges are rejected.", - "confidence": 0.46, - "canonical_issue_reason": "Issue 45278 is the best umbrella representative because it captures the broad post-upgrade import failure theme affecting many models, even though the other reports are narrower and not true duplicates.", + "summary": "This cluster is mostly a set of unrelated Transformers issues that only share broad themes like loading/import regressions, model-specific bugs, and test failures. I did not find any soft pair that clearly describes the same underlying bug or change, so all soft edges are rejected.", + "confidence": 0.84, + "canonical_issue_reason": "issue:45278 is the best umbrella candidate because it is the most general and directly captures the broad v5 upgrade/regression theme affecting multiple import/loading paths.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45278 is the strongest global representative for duplicate triage: it is recent, broad, and centered on the same general class of import/load regressions after upgrading to v5.", + "best_issue_reason": "issue:45278 is the most suitable representative issue for the cluster: it is broad enough to summarize the common v5 regression pattern without being overly model-specific.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43992", "right": "issue:45310", "accept": false, - "reason": "Both involve model loading errors, but one is a missing `embed_tokens.weight` path in UMT5Encoder while the other is a Qwen3.5 MoE `from_pretrained` failure; different root causes." + "reason": "Both concern model-loading failures, but for different models and different failure modes; not the same underlying bug." }, { "left": "issue:43064", "right": "issue:44492", "accept": false, - "reason": "FSDP/PEFT optimizer-state corruption is unrelated to a cache-strategy typo." + "reason": "Training optimizer-state corruption and a cache-strategy typo are unrelated." }, { "left": "issue:43010", "right": "issue:45278", "accept": false, - "reason": "`@torch.no_grad` on cache updates is a specific runtime correctness fix; 45278 is a broad import-error report after upgrade, not the same bug." + "reason": "One is about adding no_grad to cache update methods; the other is a broad import-error complaint after upgrade." }, { "left": "issue:44075", "right": "issue:46365", "accept": false, - "reason": "SGD arguments not being used is an optimizer configuration issue; importing Trainer failing due to `datasets` is a separate import-resolution problem." + "reason": "SGD args not being used and Trainer import failing due to datasets are different code paths and symptoms." }, { "left": "issue:29870", "right": "issue:33453", "accept": false, - "reason": "Bart `generate()` hits an unassigned `decoder_attention_mask`; tokenizer loading regression is a different code path." + "reason": "Bart generate decoder mask initialization and tokenizer loading regression are distinct bugs." }, { "left": "issue:44355", "right": "issue:44855", "accept": false, - "reason": "Compiled Python file errors are not the same as the Python 3.13 DebertaV2 `@torch.jit.script` parsing/IndentationError." + "reason": "Both are import-related errors, but one is about compiled Python files and the other is a torch.jit decorator parsing issue." }, { "left": "issue:44938", "right": "issue:45071", "accept": false, - "reason": "One is a Python 3.14 import/load failure, the other is a `PretrainedConfig` type-checking regression." + "reason": "Python version import failure and PretrainedConfig type checking are not the same defect." }, { "left": "issue:43867", "right": "issue:46202", "accept": false, - "reason": "Sorted `state_dict` loading and Gemma pipeline roundtrip on \u2581 are different failures." + "reason": "A state_dict sorting load error and a Gemma pipeline roundtrip issue are unrelated." }, { "left": "issue:36683", "right": "issue:43064", "accept": false, - "reason": "Missing `vocab_size` on `Gemma3Config` is unrelated to the FSDP2/PEFT optimizer-state bug." + "reason": "Gemma3Config missing vocab_size and optimizer-state corruption are unrelated." }, { "left": "issue:43867", "right": "issue:45375", "accept": false, - "reason": "A load-order/state_dict issue is not the same as Qwen3_5 vision config fields being dropped by `@strict`." + "reason": "Different model-loading problems with different missing fields; not the same bug." }, { "left": "issue:42890", "right": "issue:45561", "accept": false, - "reason": "Missing `set_seed()` causes flaky model tests; xdist workers racing on `captured_info.txt` is a different CI race." + "reason": "Seed-related flaky integration tests and xdist file-race failures are different test infrastructure issues." }, { "left": "issue:41669", "right": "issue:44246", "accept": false, - "reason": "Both concern import performance, but one is about removing `import *` in models and the other is a general slow-import report; not clearly the same underlying bug." + "reason": "Import-time performance complaints overlap broadly, but one is a proposal to remove import * and the other is a generic slow-import report; insufficient evidence of the same underlying issue." }, { "left": "issue:43010", "right": "issue:45850", "accept": false, - "reason": "A cache-update `no_grad` bug is not the same as a broad complaint that v5 breaks too many models." + "reason": "A no_grad cache-update fix and a generic 'breaks too many models' complaint are not the same concrete bug." }, { "left": "issue:36296", "right": "issue:43575", "accept": false, - "reason": "Tensor-parallel training bug and Qwen2-57B-A14B-Instruct OOM under TP are related only at a very high level; the concrete failures differ." + "reason": "Tensor-parallel training bug and tensor-parallel OOM on a specific model are different issues." }, { "left": "issue:43618", "right": "issue:44077", "accept": false, - "reason": "Missing CLIPOutput attentions and patchtsmixer `post_init` restrictions are unrelated model issues." + "reason": "CLIP attentions assignment and patchtsmixer post_init validation are unrelated model bugs." }, { "left": "issue:43010", "right": "issue:43976", "accept": false, - "reason": "Cache update `no_grad` is unrelated to the Python 3.9/3.10 compatibility problem." + "reason": "Cache no_grad decoration and a Python version support problem are unrelated." }, { "left": "issue:43867", "right": "issue:45357", "accept": false, - "reason": "A state_dict load-order issue is not the same as Qwen3.5 saving incorrect visual encoder keys." + "reason": "Different save/load regressions affecting different models and keys; not the same defect." }, { "left": "issue:43992", "right": "issue:45478", "accept": false, - "reason": "UMT5Encoder missing `embed_tokens.weight` and Qwen3.5 MoE `from_pretrained` failure are different model-specific regressions." + "reason": "Same general family of load errors, but distinct models and failures; not mergeable as one bug fix." }, { "left": "issue:28282", "right": "issue:33453", "accept": false, - "reason": "Missing PyTorch for AutoModel and a Bart tokenizer/eval bug are unrelated." + "reason": "Missing PyTorch import and tokenizer loading regression have no direct overlap." }, { "left": "issue:43618", "right": "issue:46285", "accept": false, - "reason": "Both mention CLIP, but one is missing attentions output and the other is an architectural inconsistency between two CLIP text model classes." + "reason": "CLIP attentions assignment and CLIP text-model structure inconsistency are different changes." }, { "left": "issue:43618", "right": "issue:44857", "accept": false, - "reason": "CLIP attentions regression and LwDetrImageLoss AMP crash are unrelated." + "reason": "CLIP output fields and LwDetr AMP crash are unrelated." }, { "left": "issue:44360", "right": "issue:45440", "accept": false, - "reason": "DSA indexer ReLU discussion has no overlap with DeepseekV3MoE implementation divergence." + "reason": "DSA indexer discussion and DeepSeekV3 implementation divergence are different topics." }, { "left": "issue:42890", "right": "issue:44964", "accept": false, - "reason": "Test seeding flakiness is unrelated to loading Phi-4 multimodal." + "reason": "A flaky test missing seeds and a multimodal model load failure are unrelated." }, { "left": "issue:42890", "right": "issue:44112", "accept": false, - "reason": "SAM HQ test seed flakiness and GraniteSpeech stale device override are distinct CI problems." + "reason": "These are different CI/test failures with different causes." }, { "left": "issue:42890", "right": "issue:45600", "accept": false, - "reason": "SAM HQ test flakiness and stale Sam3LiteText auto-mappings are different issues." + "reason": "A seed-related flaky test and broken auto_mappings references are unrelated CI issues." }, { "left": "issue:42898", "right": "issue:46664", "accept": false, - "reason": "Tokenization cleanup behavior change and SmolVLM `num_frames` semantics are unrelated." + "reason": "Tokenization cleanup behavior and video processor frame semantics are unrelated." }, { "left": "issue:43502", "right": "issue:44062", "accept": false, - "reason": "`local_files_only=True` still making API requests is not the same as `AddedToken(..., special=...)` getting duplicate keyword arguments." + "reason": "Offline-loading API leakage and AddedToken constructor duplication are unrelated tokenizer/model bugs." }, { "left": "issue:44222", "right": "issue:45125", "accept": false, - "reason": "FP8 `save_pretrained` handling for MoE is a serialization bug, while missing `_tp_plan` is a tensor-parallel metadata issue." + "reason": "FP8 save_pretrained and missing tensor-parallel plan are separate Qwen3.5 MoE issues." }, { "left": "issue:44734", "right": "issue:46617", "accept": false, - "reason": "KV-cache continuation indexing in `transformers serve` is unrelated to trust-remote-code cache file lookup failures." + "reason": "KV-cache continuation indexing and trust_remote_code cache loading failure are different code paths." }, { "left": "issue:43723", "right": "issue:44556", "accept": false, - "reason": "Tokenizer loading regression in v5 and checkpoint reload failures after upgrading versions are different load-path issues." + "reason": "Tokenizer-from-pretrained regression and checkpoint reload compatibility after upgrade are related only broadly to loading, not the same specific bug." } ] }, @@ -17101,284 +17107,281 @@ "issue:44315|issue:46209", "issue:44485|issue:46664" ], - "prepared_review_unit_hash": "df4eab729603e6f28b734720d0501b2a2ebca5567b618f94308897d0e654d915", + "prepared_review_unit_hash": "8eab4fca79db1345bc8e625ed7ba8837c657e8849e73cdeca1a4bdea137debba", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 40, - "serialized_chars": 38600, - "estimated_input_tokens": 9650, - "estimated_eval_tokens": 19556 + "serialized_chars": 38595, + "estimated_input_tokens": 9649, + "estimated_eval_tokens": 19554 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: most items are unrelated regressions, import errors, docs issues, or model-specific bugs. Only the RoFormer pair looks like a true duplicate; the rest should stay separate.", - "confidence": 0.87, - "canonical_issue_reason": "No single cluster-wide canonical issue. The only clear duplicate pair is issue 46143/46144, and 46144 is the better root-cause report because it names the attention-interface gap directly.", + "summary": "Loose cluster of largely unrelated Transformers issues, mostly around v5 regressions, loading/import failures, and a few model-specific bugs. No soft-edge pair looks like the same underlying bug/change, so all soft links are rejected.", + "confidence": 0.62, + "canonical_issue_reason": "issue:44556 is the clearest representative of the cluster\u2019s recurring theme: a concrete v4.57\u2192v5.x reload regression affecting users upgrading checkpoints.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 46144 is the strongest representative of the only real duplicate pair: it describes the underlying RoFormer attention-interface problem rather than just the symptom.", + "best_issue_reason": "issue:44556 is the best global representative because it is broad, user-facing, and directly captures the common upgrade-breakage pattern seen across the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44387", "right": "issue:44556", "accept": false, - "reason": "Both are v5 regressions, but one is int4 OOM/memory growth and the other is checkpoint reload failure; different symptoms and likely different code paths." + "reason": "Both mention v5 breakage, but one is int4 OOM/reserved-memory behavior and the other is checkpoint reload failure; different bugs and code paths." }, { "left": "issue:43824", "right": "issue:46617", "accept": false, - "reason": "ImportError for a missing Transformers symbol and FileNotFoundError loading a cached trust_remote_code model are unrelated load failures." + "reason": "One is a missing exported model class import error, the other is a cache FileNotFoundError for trust_remote_code; unrelated failure modes." }, { "left": "issue:43010", "right": "issue:43065", "accept": false, - "reason": "Cache update no_grad semantics and Sam3PixelDecoder dummy Conv2d are different model/runtime changes." + "reason": "Cache no_grad decoration and Sam3PixelDecoder dummy Conv2d are unrelated implementation issues." }, { "left": "issue:43874", "right": "issue:46617", "accept": false, - "reason": "A missing image-processor method in GLM46V is not the same bug as cached trust_remote_code file lookup failure." + "reason": "One is a missing image-processor method causing token counting to fail; the other is a cache loading FileNotFoundError." }, { "left": "issue:43502", "right": "issue:44485", "accept": false, - "reason": "local_files_only network leakage and GLM-5 RoPE implementation are unrelated issues." + "reason": "Local-files-only network leakage and GLM-5 RoPE implementation are different subsystems and bugs." }, { "left": "issue:43576", "right": "issue:46132", "accept": false, - "reason": "Broken env CLI behavior and AttentionInterface.register behavior are different features and code paths." + "reason": "A broken CLI command and a changed attention registration behavior are not the same underlying issue." }, { "left": "issue:45125", "right": "issue:45478", "accept": false, - "reason": "Both involve Qwen3.5 MoE, but one is missing tensor-parallel plan metadata and the other is a from_pretrained failure; not enough evidence of the same defect." + "reason": "Both involve Qwen3.5 MoE, but one is missing tensor-parallel plan metadata and the other is a from_pretrained load error; not enough evidence they are the same bug." }, { "left": "issue:30333", "right": "issue:33453", "accept": false, - "reason": "MLflow job-reporting failure and tokenizer-loading regression are unrelated." + "reason": "MLFlow reporting on failed jobs is unrelated to tokenizer-loading regression." }, { "left": "issue:44355", "right": "issue:44749", "accept": false, - "reason": "Compiled Python file errors and slower filtering after upgrade are different problems." + "reason": "Compiled Python file errors and slower data filtering after upgrade are unrelated." }, { "left": "issue:42548", "right": "issue:44351", "accept": false, - "reason": "Top-level import failures for different symbols usually indicate different export/compatibility bugs, not the same issue." + "reason": "Both are import errors, but for different symbols and likely different API breakages." }, { "left": "issue:45125", "right": "issue:45375", "accept": false, - "reason": "Both are Qwen3.5 MoE-family issues, but one is tensor-parallel metadata and the other is a config field being dropped; different code paths." + "reason": "Tensor-parallel plan metadata and missing config field in a different model are separate issues." }, { "left": "issue:46664", "right": "issue:46726", "accept": false, - "reason": "Video-frame count semantics and AutoProcessor backend-missing behavior are unrelated." + "reason": "Video num_frames semantics and missing-vision-backend AutoProcessor handling are different problems." }, { "left": "issue:43010", "right": "issue:43232", "accept": false, - "reason": "Cache update no_grad and generation kwargs update after sync_gpus are separate generation internals." + "reason": "Decorator placement for cache updates is unrelated to generation kwargs handling after sync_gpus." }, { "left": "issue:30990", "right": "issue:31795", "accept": false, - "reason": "Sentence Transformers loading hang and forward-API documentation confusion are not the same bug." + "reason": "SentenceTransformers loading stall and forward-doc confusion are not the same bug." }, { "left": "issue:43502", "right": "issue:44961", "accept": false, - "reason": "local_files_only network calls and an unrelated placeholder issue have no overlap." + "reason": "Remote API calls despite local_files_only and an unrelated placeholder issue are not the same." }, { "left": "issue:43502", "right": "issue:44222", "accept": false, - "reason": "Network requests despite local_files_only and FP8 moe save_pretrained behavior are different problems." + "reason": "Network access with local_files_only and FP8 save_pretrained MoE are unrelated." }, { "left": "issue:43827", "right": "issue:44961", "accept": false, - "reason": "Pipeline-removal docs cleanup and an unrelated issue title are not duplicates." + "reason": "Docs still referencing pipeline() and an unrelated placeholder issue do not match." }, { "left": "issue:30333", "right": "issue:36296", "accept": false, - "reason": "MLflow reporting on failed jobs is unrelated to tensor-parallel training bugs." + "reason": "MLFlow failure reporting and tensor-parallel training bug are distinct." }, { "left": "issue:29870", "right": "issue:31795", "accept": false, - "reason": "BART generation crash and docs confusion about model.forward inputs are unrelated." + "reason": "A generation-time UnboundLocalError in BART and a docs confusion issue are different." }, { "left": "issue:43299", "right": "issue:44556", "accept": false, - "reason": "Both are version-related loading regressions, but they affect different model families and failure modes." + "reason": "Qwen3VL/MoE loading breakage and generic checkpoint reload regression are both v5-related but not the same concrete bug." }, { "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 regression are unrelated." }, { "left": "issue:33453", "right": "issue:34689", "accept": false, - "reason": "Tokenizer-loading regression and Llama 3.2 Vision model-loading failure are distinct issues." + "reason": "Tokenizer-loading regression and Llama 3.2 Vision model-loading breakage are separate loading problems." }, { "left": "issue:44355", "right": "issue:44655", "accept": false, - "reason": "Compiled Python file errors and Pipeline.save_pretrained failure are unrelated." + "reason": "Compiled Python file errors and Pipeline.save_pretrained failure are different code paths." }, { "left": "issue:42757", "right": "issue:43576", "accept": false, - "reason": "Missing huggingface_hub import and broken transformers env CLI are different compatibility problems." + "reason": "Hugging Face Hub offline-mode import error and broken env command are unrelated." }, { "left": "issue:43576", "right": "issue:45071", "accept": false, - "reason": "CLI breakage and PretrainedConfig type-checking regression are unrelated." + "reason": "CLI env command breakage and config type-checking regression are different issues." }, { "left": "issue:44315", "right": "issue:44360", "accept": false, - "reason": "Liger Kernel application issue and DSA indexer ReLU discussion are unrelated." + "reason": "Liger kernel not applied with model_init and a DSA indexer ReLU discussion are unrelated." }, { "left": "issue:31795", "right": "issue:38617", "accept": false, - "reason": "Docs confusion in model.forward is unrelated to the missing layer_type_validation import." + "reason": "Forward-argument docs confusion and a missing configuration_utils import are different concerns." }, { "left": "issue:43576", "right": "issue:43824", "accept": false, - "reason": "Broken env CLI and missing Qwen2.5 VL export are different failures." + "reason": "Broken env command and missing model import are unrelated." }, { "left": "issue:44961", "right": "issue:45468", "accept": false, - "reason": "Uninformative title and Gemma-4 audio positional encoding bug are unrelated." + "reason": "Placeholder issue and Gemma-4 audio positional encoding bug have no evident relation." }, { "left": "issue:44351", "right": "issue:46617", "accept": false, - "reason": "HybridCache import failure and cached trust_remote_code FileNotFoundError are different issues." + "reason": "Importing HybridCache and trust_remote_code cache FileNotFoundError are different issues." }, { "left": "issue:44485", "right": "issue:44961", "accept": false, - "reason": "GLM-5 RoPE implementation discussion is unrelated to the placeholder issue." + "reason": "GLM-5 RoPE implementation and placeholder issue are unrelated." }, { "left": "issue:44908", "right": "issue:44961", "accept": false, - "reason": "Scheduler kwargs bug and unrelated placeholder title are not duplicates." + "reason": "Scheduler kwargs handling and placeholder issue are unrelated." }, { "left": "issue:38617", "right": "issue:43576", "accept": false, - "reason": "Missing import and broken env CLI are different breakages." + "reason": "Missing internal import and broken env command are different bug reports." }, { "left": "issue:36296", "right": "issue:43901", "accept": false, - "reason": "Tensor-parallel training bug and docs mismatch on return_all_scores are unrelated." + "reason": "Tensor-parallel training bug and docs mentioning return_all_scores are unrelated." }, { "left": "issue:46143", "right": "issue:46144", - "accept": true, - "reason": "Both describe the same RoFormer bug: kwargs are not flowing through because the attention implementation bypasses the attention interface." + "accept": false, + "reason": "Dropped kwargs in RoFormer methods and attention-interface integration are related area but not the same concrete bug." }, { "left": "issue:43475", "right": "issue:46617", "accept": false, - "reason": "SAM3 vision encoder attribute error and cached trust_remote_code file lookup failure are unrelated." + "reason": "SAM3 video attribute error and trust_remote_code cache FileNotFoundError are unrelated." }, { "left": "issue:44222", "right": "issue:44230", "accept": false, - "reason": "FP8 save_pretrained behavior and FP8 inference support are different code paths and likely need separate fixes." + "reason": "FP8 save_pretrained bug and FP8 inference support request are adjacent but distinct issues." }, { "left": "issue:41628", "right": "issue:42548", "accept": false, - "reason": "Missing AutoImageProcessor and missing PreTrainedModel exports are both import errors, but not the same underlying bug." + "reason": "AutoImageProcessor import failure and PreTrainedModel import failure are different symbols and likely different regressions." }, { "left": "issue:44315", "right": "issue:46209", "accept": false, - "reason": "Liger Kernel/model_init and B200 FP8 inference are unrelated GPU/model issues." + "reason": "Liger Kernel/model_init behavior and FP8 inference on B200 are unrelated." }, { "left": "issue:44485", "right": "issue:46664", "accept": false, - "reason": "GLM-5 RoPE implementation and SmolVLM num_frames semantics are unrelated." + "reason": "GLM-5 RoPE implementation and SmolVLM num_frames semantics are different model-specific bugs." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The assessment is grounded in the packet and stays conservative throughout. The lone accepted duplicate pair, issue:46143/46144, is well-supported by the titles, and the summary does not overclaim beyond the available evidence." - }, - "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 }, @@ -17487,289 +17490,286 @@ "issue:44315|issue:44829", "issue:44360|issue:45468" ], - "prepared_review_unit_hash": "3dc46dbadc1409cdc930fe1c7ae4ecd656a9c60e0d8a6847a518e70a38cfa05d", + "prepared_review_unit_hash": "45c0c2225e3bdc12de609ef0632aa821cff595f23e2e86777dcf72026108bb6c", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 41, - "serialized_chars": 39405, - "estimated_input_tokens": 9852, - "estimated_eval_tokens": 19960 + "serialized_chars": 39404, + "estimated_input_tokens": 9851, + "estimated_eval_tokens": 19958 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Conservative triage: issue:45003 remains the best representative of the lazy-loading/import-path regression, but none of the soft-edge pairs have enough evidence to be treated as duplicates. The two checkpoint/import-related pairs are only topical overlaps in this packet and should stay separate absent stronger linkage.", - "confidence": 0.79, - "canonical_issue_reason": "issue:45003 is the most concrete root-cause report in the set because it identifies unsafe sys.modules access in modeling_utils, which is more specific than the broader symptom-style lazy-loading report.", + "summary": "The cluster is mostly a loose collection of v5-regression and model-specific bug reports, not a set of true duplicates. I found no soft pairs that look safe to merge.", + "confidence": 0.42, + "canonical_issue_reason": "issue:45850 is the broadest umbrella report in the set, but it is still only a loose theme match rather than a true duplicate nucleus.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45003 is the strongest overall issue candidate because it describes a specific failing code path rather than a generic regression symptom.", + "best_issue_reason": "issue:45850 is the best global representative because it captures the widest version-regression complaint, even though the surrounding issues are too heterogeneous to collapse into one duplicate.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45362", "right": "issue:45850", "accept": false, - "reason": "Broad v5 regression vs a specific Qwen3.5 chat crash; not the same bug." + "reason": "Specific Qwen3.5 chat crash vs a broad \"many models broken\" complaint; not the same code-path." }, { "left": "issue:43673", "right": "issue:44492", "accept": false, - "reason": "Generation cache missing in chunked_prefill vs a cache-strategy typo; different issues." + "reason": "Generation chunked_prefill cache regression vs a typo in cache-strategy naming; different bugs." }, { "left": "issue:44273", "right": "issue:45542", "accept": false, - "reason": "Lazy loading failure vs TensorFlow backend error; unrelated symptoms." + "reason": "Lazy loading failure vs TensorFlow backend error from tensorboard-only install; unrelated symptoms." }, { "left": "issue:44360", "right": "issue:46664", "accept": false, - "reason": "DSA/ReLU discussion is unrelated to SmolVLM num_frames semantics." + "reason": "DSA indexer/ReLU discussion is unrelated to SmolVLM video num_frames semantics." }, { "left": "issue:44623", "right": "issue:45230", "accept": false, - "reason": "Missing processor save_pretrained files vs a generic bug report; too little evidence of sameness." + "reason": "Missing processor.save_pretrained files is too specific to merge with a generic bug report." }, { "left": "issue:44038", "right": "issue:45125", "accept": false, - "reason": "Qwen3-VL/MoE bug vs missing tensor-parallel plan; different concrete failures." + "reason": "Qwen3-VL-Moe support/inference issue vs missing _tp_plan for tensor parallelism; related model family but different fixes." }, { "left": "issue:44273", "right": "issue:45003", "accept": false, - "reason": "Both mention import/lazy-loading behavior, but the packet only shows topical similarity and no direct evidence they are the same underlying bug." + "reason": "Lazy loading problem vs unsafe sys.modules access in modeling_utils; different underlying failures." }, { "left": "issue:44393", "right": "issue:45071", "accept": false, - "reason": "2D bbox/output issue vs PretrainedConfig type-checking regression; different code paths." + "reason": "Qwen3-VL bbox hallucination/error vs PretrainedConfig type-checking regression; not the same bug." }, { "left": "issue:44393", "right": "issue:45753", "accept": false, - "reason": "Both are Qwen3-VL related, but one is bbox output hallucination and the other is an infinite loop on a specific image." + "reason": "Both touch Qwen3-VL-ish models, but one is bbox output errors and the other is an infinite-loop image case." }, { "left": "issue:41669", "right": "issue:43673", "accept": false, - "reason": "Import-star performance issue vs cache missing during chunked prefill; not the same underlying bug." + "reason": "Import-time slowdown from import * removal is unrelated to the missing GenerationMixin cache issue." }, { "left": "issue:43576", "right": "issue:44556", "accept": false, - "reason": "Broken env command vs checkpoint reload regression; unrelated features." + "reason": "Broken transformers env command vs checkpoint reload incompatibility after upgrade; different command paths." }, { "left": "issue:43525", "right": "issue:45850", "accept": false, - "reason": "Missing pad_token_id in Llama4Config is a specific config bug, not the generic v5 regression." + "reason": "Missing pad_token_id on Llama4Config is a specific config bug, not the broad regression theme." }, { "left": "issue:44556", "right": "issue:45092", "accept": false, - "reason": "Both are about v5-era checkpoint/loading incompatibilities, but the packet gives only broad topical overlap and no direct linkage to justify duplicate status." + "reason": "Both are upgrade-related, but one breaks checkpoint reload and the other breaks meta initialization for remote-code checkpoints." }, { "left": "issue:43572", "right": "issue:45850", "accept": false, - "reason": "StableLm pad_token_idx regression is model-specific, not the broad 'many models break' report." + "reason": "StableLmConfig pad_token_idx regression is model-specific and distinct from the broad \"many models\" complaint." }, { "left": "issue:44492", "right": "issue:45850", "accept": false, - "reason": "Cache-strategy typo is unrelated to the broad v5 regression complaint." + "reason": "Cache-strategy typo is documentation/label-level, not the same as general v5 breakage." }, { "left": "issue:45278", "right": "issue:45850", "accept": false, - "reason": "Generic import errors after upgrade vs generic breakage complaints; not a single concrete bug." + "reason": "General import errors after upgrade are broader than the specific model-breakage report, but still not the same underlying bug." }, { "left": "issue:42371", "right": "issue:44492", "accept": false, - "reason": "TF32 API guidance vs cache-strategy typo; unrelated." + "reason": "TF32 behavior API guidance is unrelated to cache-strategy typo." }, { "left": "issue:44230", "right": "issue:46095", "accept": false, - "reason": "FP8 inference support request vs weight-conversion question; different topics." + "reason": "FP8 inference support request vs weight-conversion-script question; different goals and code paths." }, { "left": "issue:41669", "right": "issue:46218", "accept": false, - "reason": "Import slowdown from import-star usage vs missing Qwen3_5ForCausalLM import; different symptoms and likely causes." + "reason": "Import slowdown from lazy-loading changes vs an actual missing module import error." }, { "left": "issue:43408", "right": "issue:44315", "accept": false, - "reason": "Model-type warning vs Liger Kernel application with model_init; unrelated." + "reason": "Model-type warning during instantiation vs Liger Kernel not applied with model_init; unrelated." }, { "left": "issue:42890", "right": "issue:44908", "accept": false, - "reason": "Seed-related test flakiness vs scheduler kwargs being ignored; different bugs." + "reason": "Test flakiness from missing set_seed calls vs scheduler kwargs not being passed." }, { "left": "issue:44246", "right": "issue:44273", "accept": false, - "reason": "Import is slow sometimes vs lazy loading not working properly; related area, but not enough to treat as the same concrete bug." + "reason": "Slow import and lazy loading issues are related broadly, but the reported failures and likely fixes differ." }, { "left": "issue:43576", "right": "issue:44908", "accept": false, - "reason": "Env command regression vs scheduler kwargs bug; unrelated." + "reason": "Broken env command and inverse_sqrt scheduler kwargs are unrelated." }, { "left": "issue:43352", "right": "issue:44393", "accept": false, - "reason": "Flash Attention 2.0 support error vs Qwen3-VL bbox output bug; different failures." + "reason": "Flash Attention 2 support check for Nemotron vs Qwen3-VL bbox output bug; different models and code paths." }, { "left": "issue:43976", "right": "issue:44393", "accept": false, - "reason": "Python version compatibility issue vs Qwen3-VL output bug; unrelated." + "reason": "Python version compatibility regression vs model output hallucination; no shared bug." }, { "left": "issue:44704", "right": "issue:44908", "accept": false, - "reason": "AutoProcessor kwargs forwarding vs scheduler kwargs handling; different APIs and code paths." + "reason": "AutoProcessor cached_file kwargs passthrough vs scheduler kwargs handling; different subsystems." }, { "left": "issue:44908", "right": "issue:45071", "accept": false, - "reason": "Scheduler kwargs regression vs PretrainedConfig type checking regression; unrelated." + "reason": "Scheduler kwargs bug vs PretrainedConfig type-checking regression." }, { "left": "issue:44297", "right": "issue:45230", "accept": false, - "reason": "Tokenizer class mismatch on save vs generic bug report; no duplicate evidence." + "reason": "Tokenizer save_pretrained mismatch is specific and not comparable to a generic bug report." }, { "left": "issue:44295", "right": "issue:45230", "accept": false, - "reason": "position_ids buffer access error vs generic bug report; unrelated." + "reason": "position_ids buffer access error is a concrete model bug, not a generic placeholder issue." }, { "left": "issue:41669", "right": "issue:44351", "accept": false, - "reason": "Import-performance issue vs HybridCache import failure; not the same underlying bug." + "reason": "Import performance regression vs cannot-import HybridCache; distinct failure modes." }, { "left": "issue:44492", "right": "issue:46328", "accept": false, - "reason": "Cache-strategy typo vs SequenceFeatureExtractor padding performance; unrelated." + "reason": "Cache-strategy typo and SequenceFeatureExtractor.pad performance issue are unrelated." }, { "left": "issue:41669", "right": "issue:45341", "accept": false, - "reason": "Import-performance regression vs a testing_utils bug; unrelated." + "reason": "Import-* performance regression vs a testing_utils bug; not the same underlying change." }, { "left": "issue:43976", "right": "issue:45850", "accept": false, - "reason": "Python version support issue vs broad v5 breakage report; not the same bug." + "reason": "Python support regression is different from the broad v5 breakage complaint." }, { "left": "issue:44393", "right": "issue:45478", "accept": false, - "reason": "Qwen3-VL bbox output issue vs Qwen3.5 MoE from_pretrained failure; different model paths." + "reason": "Qwen3-VL bbox output bug vs Qwen3.5-Moe from_pretrained error; different model/features." }, { "left": "issue:43976", "right": "issue:46218", "accept": false, - "reason": "Python version support vs import error for Qwen3_5ForCausalLM; unrelated." + "reason": "Python version compatibility issue vs module import failure for Qwen3_5ForCausalLM." }, { "left": "issue:44077", "right": "issue:44861", "accept": false, - "reason": "patchtsmixer post_init policy vs tied-weight key AttributeError; unrelated." + "reason": "post_init allowance change vs tied-weight-key AttributeError; different fixes." }, { "left": "issue:43827", "right": "issue:44908", "accept": false, - "reason": "Docs still referencing pipeline() vs scheduler kwargs handling; different concerns." + "reason": "Docs still referencing pipeline() vs scheduler kwargs bug; unrelated." }, { "left": "issue:31795", "right": "issue:33453", "accept": false, - "reason": "Documentation confusion vs tokenizer loading regression; unrelated." + "reason": "Forward-API documentation confusion vs tokenizer loading regression; not the same issue." }, { "left": "issue:43576", "right": "issue:44273", "accept": false, - "reason": "Env command breakage vs lazy loading failure; same broad area only, not same bug." + "reason": "Broken env command vs lazy loading not working properly; only loosely related to v5 behavior." }, { "left": "issue:44315", "right": "issue:44829", "accept": false, - "reason": "Liger Kernel/model_init issue vs flash_attention_3 degenerate training; different code paths." + "reason": "Liger Kernel not applied with model_init vs degenerate training under flash_attention_3; different code paths and fixes." }, { "left": "issue:44360", "right": "issue:45468", "accept": false, - "reason": "DSA/ReLU discussion is unrelated to Gemma-4 audio positional encoding." + "reason": "DSA/ReLU discussion is unrelated to Gemma4 audio positional encoding." } ] }, - "evaluator_result": { - "accept": false, - "feedback": "The writeup is generally careful, but it overcommits on two soft-edge pairs. In particular, treating issue:44273\u2013issue:45003 and issue:44556\u2013issue:45092 as accepted duplicates goes beyond the packet evidence, which only shows topical similarity and no direct linkage. The summary should stay more conservative and avoid claiming these are likely duplicate pairs without stronger grounding." - }, - "evaluator_used": true, - "retried": true, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null @@ -17881,64 +17881,64 @@ "issue:43298|issue:44230", "issue:43717|issue:46144" ], - "prepared_review_unit_hash": "1bb382372727aec223b4af5341a57f6bbb7329a87673db152db3079fd1fcce01", + "prepared_review_unit_hash": "22a49405135aebb24679d7c3bfc6daab5c0edbe0497faad8608658da68fccecd", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 43, - "serialized_chars": 39688, - "estimated_input_tokens": 9922, - "estimated_eval_tokens": 20100 + "serialized_chars": 39684, + "estimated_input_tokens": 9921, + "estimated_eval_tokens": 20098 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is highly heterogeneous: it mixes unrelated bug reports about loading, tokenizers, processors, attention kernels, training, and CLI/docs issues. The soft pairs are mostly keyword-similar but do not describe the same underlying defect or fix, so none should be merged as duplicates.", - "confidence": 0.96, - "canonical_issue_reason": "No canonical issue: the items do not share one underlying bug or change; the apparent overlaps are broad topical similarities only.", + "summary": "The cluster is heterogeneous and mostly contains unrelated model-loading, tokenizer, processor, and training/runtime bugs. There is no strong duplicate pair; the closest umbrella issue is the broad v5-regression complaint, but it is still not a true canonical duplicate set.", + "confidence": 0.82, + "canonical_issue_reason": "issue:45850 is the broadest umbrella topic in the set (\u201cSince 5.0 version it breaks too many models\u201d) and loosely matches several v5-regression reports, though the cluster is still not a real duplicate group.", "canonical_pr_reason": null, - "best_issue_reason": "No global best issue: there is no single report that safely represents the others because the cluster spans multiple unrelated code paths and products.", + "best_issue_reason": "issue:45850 is the most representative/high-level issue for this noisy cluster, but only as a loose umbrella rather than a true duplicate anchor.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43519", "right": "issue:46655", "accept": false, - "reason": "Different bugs in different components: Qwen3VL timestamp math vs SmolVLM video padding." + "reason": "Different bugs: Qwen3VL timestamp calculation vs video padding logic in SmolVLM." }, { "left": "issue:44273", "right": "issue:44908", "accept": false, - "reason": "Lazy loading failure and inverse_sqrt scheduler kwargs are unrelated subsystems." + "reason": "Lazy loading failure is unrelated to inverse_sqrt scheduler kwargs handling." }, { "left": "issue:44655", "right": "issue:44861", "accept": false, - "reason": "Saving Pipeline objects and tied-weight key handling are distinct code paths." + "reason": "Saving Pipeline objects and tied-weight-key crashes are different code paths." }, { "left": "issue:43576", "right": "issue:46218", "accept": false, - "reason": "Broken `transformers env` CLI and Qwen3_5 import failure are not the same issue." + "reason": "Broken env command is unrelated to importing Qwen3_5ForCausalLM." }, { "left": "issue:45230", "right": "issue:45310", "accept": false, - "reason": "Generic bug report vs a specific Qwen3.5 MoE from_pretrained error; no shared defect." + "reason": "Generic bug report vs a specific Qwen3.5 MoE from_pretrained error; not the same issue." }, { "left": "issue:43519", @@ -17950,223 +17950,223 @@ "left": "issue:43519", "right": "issue:44908", "accept": false, - "reason": "Processor timestamp math and scheduler kwarg handling are different areas." + "reason": "Video timestamp bug and scheduler kwargs bug are different subsystems." }, { "left": "issue:41669", "right": "issue:43704", "accept": false, - "reason": "Import-star cleanup and VRAM leak in Qwen3ForCausalLM are unrelated changes." + "reason": "Import-star cleanup/perf issue is unrelated to a Qwen3ForCausalLM VRAM leak." }, { "left": "issue:31795", "right": "issue:34689", "accept": false, - "reason": "Documentation confusion about forward args is not the same as Llama 3.2 Vision model loading breakage." + "reason": "Documentation confusion about inputs/past_key_values is not the same as a model-loading regression." }, { "left": "issue:46018", "right": "issue:46172", "accept": false, - "reason": "DeepSeek-V4 shared expert gating and a `comb.to(...).transpose(...)` bug are different model issues." + "reason": "DeepSeek-V4 gating logic and a tensor transpose in weight conversion are different defects." }, { "left": "issue:43704", "right": "issue:44485", "accept": false, - "reason": "VRAM leak in Qwen3ForCausalLM and GLM-5 RoPE implementation do not match." + "reason": "VRAM leak in Qwen3ForCausalLM is unrelated to GLM-5 RoPE math." }, { "left": "issue:44485", "right": "issue:45468", "accept": false, - "reason": "Both are model-specific, but they target different architectures and bugs." + "reason": "Two separate model families and two different positional-encoding bugs." }, { "left": "issue:43502", "right": "issue:43519", "accept": false, - "reason": "Unexpected network calls with `local_files_only` is unrelated to Qwen3VL timestamp calculation." + "reason": "local_files_only network requests do not match timestamp miscalculation." }, { "left": "issue:42371", "right": "issue:43704", "accept": false, - "reason": "TF32 behavior settings and Qwen3 VRAM leak are separate concerns." + "reason": "TF32 API guidance is unrelated to the Qwen3ForCausalLM memory leak." }, { "left": "issue:45850", "right": "issue:46218", "accept": false, - "reason": "A broad 5.0 regression complaint and a Qwen3_5 import error are not the same fix." + "reason": "Broad v5 breakage complaint is not the same as the Qwen3_5 import error." }, { "left": "issue:43519", "right": "issue:44623", "accept": false, - "reason": "Timestamp calculation and `processor.save_pretrained` missing files are different processor bugs." + "reason": "Processor save_pretrained file omission is unrelated to Qwen3VL timestamp math." }, { "left": "issue:43606", "right": "issue:45468", "accept": false, - "reason": "CPU offload device mismatch and Gemma-4 audio positional encoding are unrelated." + "reason": "CPU offload device mismatch and Gemma-4 audio positional encoding are different failures." }, { "left": "issue:44263", "right": "issue:45468", "accept": false, - "reason": "`torch.split` return handling in GlmMoeDsaIndexer is unrelated to Gemma-4 audio positional encoding." + "reason": "GlmMoeDsaIndexer split return values is unrelated to Gemma-4 audio positional encoding." }, { "left": "issue:44261", "right": "issue:45468", "accept": false, - "reason": "Missing `rms_norm_eps` for MLA q_a_layernorm is a different bug than Gemma-4 audio positional encoding." + "reason": "Missing rms_norm_eps in MLA q_a_layernorm is not the same as Gemma-4 audio RoPE/positional encoding." }, { "left": "issue:43298", "right": "issue:46095", "accept": false, - "reason": "Continuous batching audio support and DeepSeek-V4 weight conversion are not the same change." + "reason": "Continuous batching audio-model support and DeepSeek-V4 weight conversion are unrelated." }, { "left": "issue:43519", "right": "issue:43704", "accept": false, - "reason": "Qwen3VL timestamp math and Qwen3 VRAM leak are unrelated." + "reason": "Timestamp calculation bug and VRAM leak are distinct issues." }, { "left": "issue:44351", "right": "issue:46218", "accept": false, - "reason": "HybridCache import failure and Qwen3_5ForCausalLM import failure are separate missing-symbol issues." + "reason": "HybridCache import failure is a different missing-symbol problem than Qwen3_5ForCausalLM import failure." }, { "left": "issue:43352", "right": "issue:46218", "accept": false, - "reason": "NemotronH Flash Attention support and Qwen3_5 import failure do not share a concrete code path." + "reason": "Flash Attention support for NemotronH and a Qwen3_5 import error are unrelated." }, { "left": "issue:42913", "right": "issue:43066", "accept": false, - "reason": "Both mention tokenizer v5, but one is a general behavior difference and the other is a specific decoder-type bug." + "reason": "Both involve v5 tokenizer behavior, but one is a broad behavior regression and the other a specific decoder-type bug; not enough to merge." }, { "left": "issue:28348", "right": "issue:45250", "accept": false, - "reason": "GPT2-specific Flash Attention 2.0 support request is not the same as a generic FA2 topic." + "reason": "GPT2LMHeadModel FlashAttention-2 support request is not the same as a generic Flash Attention 2.0 issue." }, { "left": "issue:44908", "right": "issue:44998", "accept": false, - "reason": "Scheduler kwargs bug and an unrelated placeholder issue have no overlap." + "reason": "Scheduler kwargs bug is unrelated to the placeholder/garbage issue title." }, { "left": "issue:44492", "right": "issue:44998", "accept": false, - "reason": "Cache-strategy typo and an unrelated placeholder issue are not duplicates." + "reason": "Cache-strategy typo is unrelated to the unrelated title noise." }, { "left": "issue:43618", "right": "issue:44998", "accept": false, - "reason": "CLIPOutput attentions regression and an unrelated placeholder issue are not duplicates." + "reason": "CLIPOutput attentions regression is unrelated to the unrelated title noise." }, { "left": "issue:44998", "right": "issue:45850", "accept": false, - "reason": "The placeholder issue is unrelated to the broad Transformers 5.0 complaint." + "reason": "Placeholder/unemployment issue is not the same as the broad v5 breakage report." }, { "left": "issue:44938", "right": "issue:44998", "accept": false, - "reason": "Python 3.14 loading failure and an unrelated placeholder issue are not duplicates." + "reason": "Python 3.14 load failure is unrelated to the unrelated title noise." }, { "left": "issue:44861", "right": "issue:45290", "accept": false, - "reason": "Tied-weight-key crash and chat-template tool-call crash are different failures." + "reason": "Tied-weight key crash and apply_chat_template tool-call crash are different code paths." }, { "left": "issue:31515", "right": "issue:38617", "accept": false, - "reason": "Slow from_pretrained loading and missing layer_type_validation import are unrelated." + "reason": "Slow from_pretrained checkpoint loading is unrelated to the layer_type_validation import error." }, { "left": "issue:43525", "right": "issue:44861", "accept": false, - "reason": "Missing pad_token_id in Llama4Config is unrelated to tied-weight-key AttributeError." + "reason": "Missing pad_token_id on Llama4Config is unrelated to tied-weight-key handling." }, { "left": "issue:43572", "right": "issue:44861", "accept": false, - "reason": "Missing pad_token_idx in StableLmConfig is a config regression, not tied-weight handling." + "reason": "Missing pad_token_idx in StableLmConfig is unrelated to tied-weight-key handling." }, { "left": "issue:44230", "right": "issue:44623", "accept": false, - "reason": "FP8 inference support request and processor.save_pretrained file omission are unrelated." + "reason": "FP8 inference support request is unrelated to processor.save_pretrained missing files." }, { "left": "issue:28348", "right": "issue:43352", "accept": false, - "reason": "Both mention Flash Attention 2.0, but they target different models and separate support requests." + "reason": "Both mention Flash Attention 2.0, but they target different models and changes; not one fix." }, { "left": "issue:29870", "right": "issue:31515", "accept": false, - "reason": "BART generate decoder mask error and slow checkpoint loading are unrelated." + "reason": "Bart generate() decoder_attention_mask bug is unrelated to slow checkpoint loading." }, { "left": "issue:30333", "right": "issue:31515", "accept": false, - "reason": "MLFlow reporting on failed training and slow from_pretrained loading are unrelated." + "reason": "MLflow reporting on failed jobs is unrelated to from_pretrained performance." }, { "left": "issue:31515", "right": "issue:36296", "accept": false, - "reason": "Slow checkpoint loading and tensor-parallel training bug are different issues." + "reason": "Slow checkpoint loading and tensor-parallel training bug are unrelated." }, { "left": "issue:31515", "right": "issue:31795", "accept": false, - "reason": "Loading performance and forward-argument documentation confusion are not the same defect." + "reason": "Performance loading issue and documentation confusion are unrelated." }, { "left": "issue:44230", "right": "issue:45310", "accept": false, - "reason": "FP8 inference support request and Qwen3.5 MoE from_pretrained error are different topics." + "reason": "FP8 support request and Qwen3.5 MoE from_pretrained error are not the same concrete bug." }, { "left": "issue:43298", "right": "issue:44230", "accept": false, - "reason": "Continuous batching audio support and FP8 MoE inference support are separate feature requests." + "reason": "Audio continuous-batching support and Qwen3-VL/Qwen3.5 FP8 support are different feature areas." }, { "left": "issue:43717", "right": "issue:46144", "accept": false, - "reason": "Mamba init_weights usage and RoFormer attention-interface issue are unrelated model implementation topics." + "reason": "Mamba init_weights discussion is unrelated to RoFormer attention interface usage." } ] }, @@ -18212,20 +18212,20 @@ "issue:37934|issue:43668", "issue:42936|issue:43668" ], - "prepared_review_unit_hash": "3f58f6c59ec8a1f329d4f997588724cd7ad6b4d7dd6e1940da4beba54e3dfdd0", + "prepared_review_unit_hash": "3276d63282510663825ca4b10e39198d94989a497e03e7d48de477b4667b1a87", "component_budget": { "node_count": 483, "item_count": 483, "soft_pair_count": 1779, - "serialized_chars": 922196, - "estimated_input_tokens": 230549, - "estimated_eval_tokens": 461354 + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 }, "budget": { "node_count": 8, "item_count": 8, "soft_pair_count": 11, - "serialized_chars": 7773, + "serialized_chars": 7776, "estimated_input_tokens": 1944, "estimated_eval_tokens": 4144 }, @@ -18234,78 +18234,78 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: the issues span unrelated models, config typing, tokenizer warnings, and attention/loss plumbing. I don\u2019t see a single underlying bug/change that ties them together as duplicates.", - "confidence": 0.93, - "canonical_issue_reason": "No clear canonical issue: the items describe distinct problems in different subsystems, so any one issue would be misleading as a cluster representative.", - "canonical_pr_reason": "No canonical PR candidate is present in the provided items.", - "best_issue_reason": "None of the issues is a good global representative because there is no shared root cause across the set.", - "best_pr_reason": "No PRs were provided, so there is no global best PR.", + "summary": "These issues are largely unrelated across different models and subsystems (norms, loss wiring, config typing, tokenizer warnings, MoE gating, attention implementation). No duplicate cluster emerges, so no soft edge should be merged.", + "confidence": 0.92, + "canonical_issue_reason": "No single canonical issue stands out: the items describe distinct bugs and questions in separate code paths and models.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43240 is the best general anchor only because it has the strongest discussion/inbound reference signal, but it is not a duplicate of the others.", + "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:42936", "right": "issue:46018", "accept": false, - "reason": "Different bugs: Mask2Former ignore_value is unused after definition, while DeepSeek-V4 shared expert is not gated." + "reason": "Different models and failures: Mask2Former ignore_value handling vs DeepSeek-V4 MoE gating." }, { "left": "issue:43668", "right": "issue:44031", "accept": false, - "reason": "Unrelated concerns: ModernBERTConfig type-hint correctness vs tokenizer regex warning after v4.57.3." + "reason": "Unrelated bug types: ModernBERT config type hint vs tokenizer regex warning regression." }, { "left": "issue:42936", "right": "issue:43240", "accept": false, - "reason": "Different code paths: one is a model-specific ignore_value bug, the other is kwargs not being forwarded to a loss function." + "reason": "Mask2Former loss ignore_value bug is unrelated to kwargs not being forwarded into a loss function." }, { "left": "issue:43240", "right": "issue:46144", "accept": false, - "reason": "Loss-function kwargs forwarding and RoFormer attention-interface usage are unrelated implementation issues." + "reason": "Loss-kwargs forwarding and RoFormer attention-interface usage are different code paths and fixes." }, { "left": "issue:42936", "right": "issue:44031", "accept": false, - "reason": "One is a segmentation/model-loss parameter bug; the other is a tokenizer regex warning regression." + "reason": "Different subsystem and symptom: segmentation ignore_value vs tokenizer warning message." }, { "left": "issue:42936", "right": "issue:46144", "accept": false, - "reason": "These affect different components and failure modes; no shared underlying bug is evident." + "reason": "No shared underlying bug; one is a Mask2Former training target issue, the other an attention API usage issue." }, { "left": "issue:42936", "right": "issue:43813", "accept": false, - "reason": "Mask2Former ignore_value handling is not the same as a typo in a quantization operation reference." + "reason": "Mask2Former loss masking and a typo in a quantization attribute are unrelated." }, { "left": "issue:43240", "right": "issue:43668", "accept": false, - "reason": "Passing kwargs into loss computation is unrelated to a config type hint being incorrect." + "reason": "Passing kwargs to loss calculation is unrelated to an incorrect type hint in ModernBERTConfig." }, { "left": "issue:43668", "right": "issue:43813", "accept": false, - "reason": "Both are small correctness issues, but they concern different symbols and different files/problems." + "reason": "A type-hint correction and a string-attribute typo are not the same bug." }, { "left": "issue:37934", "right": "issue:43668", "accept": false, - "reason": "Llama4TextL2Norm RMS-norm question is unrelated to ModernBERTConfig norm_eps typing." + "reason": "Llama4TextL2Norm behavior question is unrelated to ModernBERTConfig typing." }, { "left": "issue:42936", "right": "issue:43668", "accept": false, - "reason": "No common root cause: model ignore_value usage vs config annotation mismatch." + "reason": "These concern different models and different failure modes; no shared code-path bug is evident." } ] }, @@ -18338,52 +18338,52 @@ "issue:10519|issue:11005", "issue:11005|issue:11307" ], - "prepared_review_unit_hash": "b4e3214cd83af9e934a11678af538820c9cd3023b91a35d21a97f56d3a8d6fae", + "prepared_review_unit_hash": "99a860ef127888c5a487f5a4acca4190bbcc7d515aaa97277fa10c3127883d70", "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 about unrelated features: tokenizer truncation behavior, ReduceLROnPlateau-like scheduling, and Wav2Vec2 word boundary offsets. There is no meaningful duplicate cluster here.", - "confidence": 0.98, - "canonical_issue_reason": null, + "summary": "All three artifacts are unrelated feature requests: truncation behavior in tokenization, ReduceLROnPlateau-like scheduler functionality, and Wav2Vec2 word boundary offsets. No duplicate cluster emerges from the titles or available metadata.", + "confidence": 0.99, + "canonical_issue_reason": "No canonical issue: the items describe different features in different parts of the library, so none represents the others.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "None of the issues are suitable as a global representative because they are not about the same underlying problem.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:10519", "right": "issue:11307", "accept": false, - "reason": "Different subsystems and problems: tokenizer truncation option vs Wav2Vec2 time offsets." + "reason": "Truncation-from-beginning vs Wav2Vec2 word offset extraction are unrelated requests in different code paths." }, { "left": "issue:10519", "right": "issue:11005", "accept": false, - "reason": "No overlap in functionality; one is about truncation strategy, the other about learning-rate scheduling." + "reason": "Token truncation behavior and ReduceLROnPlateau-like scheduling functionality are distinct, unrelated features." }, { "left": "issue:11005", "right": "issue:11307", "accept": false, - "reason": "Completely different requests: scheduler-like behavior vs extracting word start/end offsets from audio models." + "reason": "Learning-rate scheduler enhancements and Wav2Vec2 alignment offsets do not share the same underlying bug or change." } ] }, @@ -18415,12 +18415,12 @@ "issue:13244|issue:16618", "issue:15980|issue:16618" ], - "prepared_review_unit_hash": "926d579f22ebd1775f1ab594bef8b3a6fa23cf6f84eceb0767aeeb0b38eccac9", + "prepared_review_unit_hash": "afa98300a0847fa8ea8a628caf9411f8daf6668e4ebdbbb49717e001bcbaddce", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 2, - "serialized_chars": 2477, + "serialized_chars": 2478, "estimated_input_tokens": 620, "estimated_eval_tokens": 1496 }, @@ -18428,7 +18428,7 @@ "node_count": 3, "item_count": 3, "soft_pair_count": 2, - "serialized_chars": 2477, + "serialized_chars": 2478, "estimated_input_tokens": 620, "estimated_eval_tokens": 1496 }, @@ -18437,24 +18437,24 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These three issues are not duplicates: one is about Tapas tokenization mismatch, one is about private-model auth/error messaging, and one is a general model-loading failure for bert-base-uncased. The cluster is too loose for a single canonical issue.", - "confidence": 0.83, - "canonical_issue_reason": null, + "summary": "The cluster looks mixed rather than duplicate: one tokenizer discrepancy issue, one private-model/auth error issue, and one generic model-loading failure. I would not merge any of the soft pairs.", + "confidence": 0.91, + "canonical_issue_reason": "No clear canonical duplicate exists; 13244 is about Tapas tokenization, 15980 is about a bad login/private-model error message, and 16618 is about failing to load bert-base-uncased.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "Issue 16618 is the best standalone representative because it is the broadest and most directly about a model-loading failure, but it still does not match the other issues as the same bug.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:13244", "right": "issue:16618", "accept": false, - "reason": "Different failure modes: tokenization behavior vs model loading/import failure. No evidence of the same underlying bug or code path." + "reason": "Different problem domains: tokenizer output mismatch vs model loading failure; not the same underlying bug." }, { "left": "issue:15980", "right": "issue:16618", "accept": false, - "reason": "15980 is specifically about private-model download/auth error messages, while 16618 is a general inability to load bert-base-uncased. Related area, but not the same bug." + "reason": "Both mention model access/loading, but 15980 is specifically about unauthenticated private-model download messaging while 16618 is a generic load failure for bert-base-uncased." } ] }, @@ -18484,7 +18484,7 @@ "soft_pairs": [ "issue:26350|issue:33260" ], - "prepared_review_unit_hash": "f8a7b0237f4df2267d442ac1587bb9ff25648187079c582e9c93a9b709f12b36", + "prepared_review_unit_hash": "f9e609c556f9741817e8def7e24d755d16ea5d6dfce432507aa93ed4c8f60667", "component_budget": { "node_count": 2, "item_count": 2, @@ -18506,18 +18506,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are two unrelated feature-contribution issues: one for Flash Attention 2 support and one for GGUF support. They share only a broad \u201ccommunity contribution\u201d framing, so they should not be clustered as duplicates.", + "summary": "These are both broad community-contribution issues, but they target different features and code paths: Flash Attention 2 support vs GGUF support. They are not duplicates.", "confidence": 0.98, - "canonical_issue_reason": "Issue #33260 is the more specific and later task, with a clearer concrete scope (GGUF support for more architectures) and stronger discussion activity/inbound references, making it the better anchor if one issue must be chosen.", + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Issue #33260 is the best representative of its own topic because it states a concrete feature request and appears more developed in discussion than #26350.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:26350", "right": "issue:33260", "accept": false, - "reason": "Different underlying requests: Flash Attention 2 architecture support vs. GGUF architecture support. Same broad subsystem/theme, but not the same bug or change." + "reason": "Different underlying changes: one requests Flash Attention 2 support, the other requests GGUF support. Same general theme of adding architecture support, but not the same bug or change." } ] }, @@ -18576,184 +18576,184 @@ "issue:44370|issue:46226", "issue:46222|issue:46226" ], - "prepared_review_unit_hash": "2939a2d86eca5a1a213432af0717c41fdf7993379bf738a086be35c0a0a04262", + "prepared_review_unit_hash": "b0a5fc370acb74d29fbd3f92b4aa477b5ac98e45f8909f638317aa715eabca17", "component_budget": { "node_count": 8, "item_count": 8, "soft_pair_count": 24, - "serialized_chars": 13531, - "estimated_input_tokens": 3383, - "estimated_eval_tokens": 7022 + "serialized_chars": 13524, + "estimated_input_tokens": 3381, + "estimated_eval_tokens": 7018 }, "budget": { "node_count": 8, "item_count": 8, "soft_pair_count": 24, - "serialized_chars": 13531, - "estimated_input_tokens": 3383, - "estimated_eval_tokens": 7022 + "serialized_chars": 13524, + "estimated_input_tokens": 3381, + "estimated_eval_tokens": 7018 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Most items are separate i18n translation-request issues for different languages and should not be merged. The only clear duplicates are the near-identical generic placeholder translation issues (43834/44365/44370) and the exact duplicate Thai-titled pair (46222/46226).", - "confidence": 0.87, - "canonical_issue_reason": "issue:43834 is the earliest of the near-identical placeholder translation-request issues and is a reasonable representative for that duplicate subgroup.", + "summary": "This cluster is mostly a set of translation-request issues for different languages, so it is not a true duplicate cluster overall. The only pair that looks like a real duplicate is the two identical Thai issues; the rest are either different languages or generic template-like requests without enough evidence of the same underlying change.", + "confidence": 0.92, + "canonical_issue_reason": "issue:27088 is the most concrete, language-specific request and has the clearest discussion trail; the other items are either different languages or generic placeholder-style issues, so there is no stronger cluster anchor.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43834 is the best single issue representative here because it is the earliest closed placeholder-style i18n request and matches the duplicated template cluster more cleanly than the language-specific issues.", + "best_issue_reason": "issue:27088 is the best standalone representative because it is explicit, specific, and active, whereas most other items are either template-like or clearly different translation targets.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:27088", "right": "issue:44370", "accept": false, - "reason": "Different translation targets (Turkish vs generic placeholder); same broad i18n theme is not enough to merge." + "reason": "Different translation-request artifacts; no evidence they target the same language or underlying task." }, { "left": "issue:27088", "right": "issue:44365", "accept": false, - "reason": "Different translation targets; these are separate language requests." + "reason": "Different translation-request artifacts; the shared theme is only broad i18n/docs translation." }, { "left": "issue:27088", "right": "issue:43834", "accept": false, - "reason": "Turkish request vs generic placeholder template; not the same underlying issue." + "reason": "Different translation-request artifacts; similar wording is not enough to show the same request." }, { "left": "issue:27088", "right": "issue:38435", "accept": false, - "reason": "Different language requests (Turkish vs Romanian)." + "reason": "Different languages (Turkish vs Romanian), so these are distinct issues." }, { "left": "issue:27088", "right": "issue:28936", "accept": false, - "reason": "Different language requests (Turkish vs Spanish)." + "reason": "Different languages (Turkish vs Spanish), so these are distinct issues." }, { "left": "issue:28936", "right": "issue:43834", "accept": false, - "reason": "Spanish request vs generic placeholder template; not the same concrete issue." + "reason": "Spanish translation request vs a generic placeholder-style issue; no same-language evidence." }, { "left": "issue:28936", "right": "issue:44365", "accept": false, - "reason": "Different language requests; only broadly similar by topic." + "reason": "Spanish request and generic placeholder request are not the same concrete issue." }, { "left": "issue:28936", "right": "issue:38435", "accept": false, - "reason": "Different language requests (Spanish vs Romanian)." + "reason": "Different languages (Spanish vs Romanian), so these are distinct issues." }, { "left": "issue:28936", "right": "issue:44370", "accept": false, - "reason": "Different language requests; not a duplicate." + "reason": "Spanish request and generic placeholder request are not the same concrete issue." }, { "left": "issue:38435", "right": "issue:44370", "accept": false, - "reason": "Romanian request vs generic placeholder template; too vague to treat as the same issue." + "reason": "Romanian request vs generic placeholder-style issue; not enough to treat as the same request." }, { "left": "issue:38435", "right": "issue:44365", "accept": false, - "reason": "Romanian request vs generic placeholder template; different concrete request." + "reason": "Romanian request vs generic placeholder-style issue; similar template text is insufficient." }, { "left": "issue:38435", "right": "issue:43834", "accept": false, - "reason": "Different language/request specifics (Romanian vs placeholder template)." + "reason": "Romanian request vs generic placeholder-style issue; not the same concrete translation target." }, { "left": "issue:38435", "right": "issue:46222", "accept": false, - "reason": "Different language requests; Romanian and Thai-titled items are not the same issue." + "reason": "Different visible request text and no evidence of the same underlying translation target." }, { "left": "issue:38435", "right": "issue:46226", "accept": false, - "reason": "Different language requests; not a duplicate." + "reason": "Different visible request text and no evidence of the same underlying translation target." }, { "left": "issue:43834", "right": "issue:44370", - "accept": true, - "reason": "Same generic placeholder translation-request title/body pattern; near-identical duplicate." + "accept": false, + "reason": "These look like generic template variants, but there is no concrete evidence they are the same exact request." }, { "left": "issue:43834", "right": "issue:44365", - "accept": true, - "reason": "Same placeholder i18n request template with near-identical content; clear duplicate." + "accept": false, + "reason": "Template-like similarity alone is too weak to merge as duplicates." }, { "left": "issue:43834", "right": "issue:46222", "accept": false, - "reason": "Titles and visible request text differ; likely different language-specific issues despite body similarity." + "reason": "Generic placeholder-style issue vs a different visible request; no same-target evidence." }, { "left": "issue:43834", "right": "issue:46226", "accept": false, - "reason": "Different visible request/title; not safe to merge as the same concrete issue." + "reason": "Generic placeholder-style issue vs a different visible request; no same-target evidence." }, { "left": "issue:44365", "right": "issue:44370", - "accept": true, - "reason": "Same placeholder translation-request issue template and near-identical content." + "accept": false, + "reason": "Very similar template wording, but not enough evidence that they are the same underlying request." }, { "left": "issue:44365", "right": "issue:46222", "accept": false, - "reason": "Different visible request/title; body similarity alone is not enough here." + "reason": "Generic placeholder-style issue vs a different visible request; not enough to merge." }, { "left": "issue:44365", "right": "issue:46226", "accept": false, - "reason": "Different visible request/title; not clearly the same issue." + "reason": "Generic placeholder-style issue vs a different visible request; not enough to merge." }, { "left": "issue:44370", "right": "issue:46222", "accept": false, - "reason": "Different visible request/title; do not merge based on broad template similarity alone." + "reason": "Generic placeholder-style issue vs a different visible request; no concrete duplicate signal." }, { "left": "issue:44370", "right": "issue:46226", "accept": false, - "reason": "Different visible request/title; not the same underlying request." + "reason": "Generic placeholder-style issue vs a different visible request; no concrete duplicate signal." }, { "left": "issue:46222", "right": "issue:46226", "accept": true, - "reason": "Exact duplicate pair: same title and effectively identical content." + "reason": "Same visible Thai title and nearly identical metadata/body length strongly indicate the same underlying issue." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is well grounded in the packet and stays conservative. The duplicate claims for 43834/44365/44370 are supported by identical template titles plus very high similarity, and 46222/46226 is an exact duplicate pair. The rejections for language-specific issues are also appropriately cautious given the visible titles. No overstatement detected." + "feedback": "The assessment is grounded in the packet and stays conservative. It correctly treats the Turkish/Spanish/Romanian issues as distinct language-specific requests, treats the template-title issues as similar but not proven duplicates, and only accepts the Thai pair where the titles and metadata are effectively identical. No overstatement of evidence detected." }, "evaluator_used": true, "retried": false, @@ -18835,12 +18835,12 @@ "issue:42503|issue:46653", "issue:45027|issue:46653" ], - "prepared_review_unit_hash": "996535c74e56585b28e88eef6cd19dbdaea4ea89858b19fb6f86674d0a9dea4b", + "prepared_review_unit_hash": "60726fb48adde51efb445779676315b1850526e7d7d40e67acec1e5b207babd5", "component_budget": { "node_count": 18, "item_count": 18, "soft_pair_count": 40, - "serialized_chars": 24148, + "serialized_chars": 24146, "estimated_input_tokens": 6037, "estimated_eval_tokens": 12330 }, @@ -18848,7 +18848,7 @@ "node_count": 18, "item_count": 18, "soft_pair_count": 40, - "serialized_chars": 24148, + "serialized_chars": 24146, "estimated_input_tokens": 6037, "estimated_eval_tokens": 12330 }, @@ -18857,102 +18857,102 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is a loose set of model-request issues with only superficial title similarity. The soft edges mostly connect unrelated model additions or adjacent but distinct support requests, so none should be merged as duplicates.", - "confidence": 0.84, - "canonical_issue_reason": "issue #43671 is one of the most central and active nodes in the cluster, and its title is a clear representative of the recurring 'add model support' request pattern.", + "summary": "These items are mostly distinct model-support requests that share a generic 'add model' template. The apparent similarity is broad subsystem overlap, not the same underlying bug or change, so the soft duplicate links should be rejected.", + "confidence": 0.95, + "canonical_issue_reason": "No canonical duplicate issue: the cluster spans many different model requests and only shares superficial wording.", "canonical_pr_reason": null, - "best_issue_reason": "issue #43671 is the best single issue anchor here because it has high discussion activity and sits near the center of the soft-similarity graph, even though the cluster itself is not a duplicate set.", + "best_issue_reason": "issue:43671 is the clearest representative request in this set because it is explicit, model-specific, and has discussion activity, but it is still not a duplicate of the others.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:36817", "right": "issue:41211", "accept": false, - "reason": "EuroBert config support vs DEIMv2 are different model additions with separate implementation targets." + "reason": "Different models and different requested support; shared template only." }, { "left": "issue:36817", "right": "issue:43671", "accept": false, - "reason": "EuroBert config work and Qwen3-TTS support are unrelated model requests." + "reason": "EuroBert config vs Qwen3-TTS are unrelated model additions." }, { "left": "issue:36817", "right": "issue:36879", "accept": false, - "reason": "EuroBert and RF-DETR are different models and would not be merged into one fix." + "reason": "EuroBert and RF-DETR are distinct model requests." }, { "left": "issue:36879", "right": "issue:41211", "accept": false, - "reason": "RF-DETR and DEIMv2 are distinct model-add requests with no shared concrete bug." + "reason": "RF-DETR vs DEIMv2 are different model architectures." }, { "left": "issue:36879", "right": "issue:43909", "accept": false, - "reason": "RF-DETR and LFM2.5 Audio 1.5B are unrelated model support requests." + "reason": "Different targets: object detection vs audio model support." }, { "left": "issue:36879", "right": "issue:44121", "accept": false, - "reason": "Adding RF-DETR and adding the weight-sparse transformer are different model integrations." + "reason": "RF-DETR and sparse transformer request are unrelated." }, { "left": "issue:36879", "right": "issue:43837", "accept": false, - "reason": "RF-DETR and Qwen3-ASR are different model families and code paths." + "reason": "Different models and modalities; not the same change." }, { "left": "issue:36879", "right": "issue:43671", "accept": false, - "reason": "RF-DETR and Qwen3-TTS are unrelated model support requests." + "reason": "RF-DETR vs Qwen3-TTS are separate support requests." }, { "left": "issue:37015", "right": "issue:43837", "accept": false, - "reason": "NeoBERT and Qwen3-ASR are separate model additions, not the same underlying change." + "reason": "NeoBERT and Qwen3-ASR are different models and code paths." }, { "left": "issue:37015", "right": "issue:44121", "accept": false, - "reason": "NeoBERT support and the sparse transformer request are different features." + "reason": "NeoBERT vs weight-sparse transformer is not a duplicate request." }, { "left": "issue:37015", "right": "issue:43909", "accept": false, - "reason": "NeoBERT and LFM2.5 Audio 1.5B target different architectures and tasks." + "reason": "Different model families; only generic add-support wording overlaps." }, { "left": "issue:36879", "right": "issue:37015", "accept": false, - "reason": "RF-DETR and NeoBERT are unrelated model add requests." + "reason": "RF-DETR and NeoBERT are distinct model additions." }, { "left": "issue:37015", "right": "issue:42503", "accept": false, - "reason": "NeoBERT and ModernVBERT are distinct models, not one duplicate request." + "reason": "NeoBERT vs ModernVBERT are separate model requests." }, { "left": "issue:41115", "right": "issue:43837", "accept": false, - "reason": "MiniCPM3 architecture support and Qwen3-ASR support are separate model integrations." + "reason": "MiniCPM3 and Qwen3-ASR target different models and functionality." }, { "left": "issue:41115", "right": "issue:44121", "accept": false, - "reason": "MiniCPM3 and the sparse transformer request are different code-path changes." + "reason": "MiniCPM3 support vs sparse-transformer support are unrelated." }, { "left": "issue:41115", @@ -18964,145 +18964,139 @@ "left": "issue:41115", "right": "issue:45027", "accept": false, - "reason": "MiniCPM3 support and Voxtral TTS support are unrelated model additions." + "reason": "MiniCPM3 vs Voxtral TTS are different models and modalities." }, { "left": "issue:41211", "right": "issue:43671", "accept": false, - "reason": "DEIMv2 and Qwen3-TTS are different model requests with no shared concrete fix." + "reason": "DEIMv2 and Qwen3-TTS are unrelated model additions." }, { "left": "issue:42503", "right": "issue:43837", "accept": false, - "reason": "ModernVBERT models and Qwen3-ASR are unrelated additions." + "reason": "ModernVBERT vs Qwen3-ASR are different model-specific requests." }, { "left": "issue:42503", "right": "issue:43671", "accept": false, - "reason": "ModernVBERT and Qwen3-TTS are different model support requests." + "reason": "ModernVBERT and Qwen3-TTS are unrelated support requests." }, { "left": "issue:42503", "right": "issue:42738", "accept": false, - "reason": "ModernVBERT models vs BERT-like RoPE support is not the same change; one adds models, the other a positional encoding feature." + "reason": "Specific ModernVBERT support vs broad BERT-like RoPE feature request." }, { "left": "issue:42738", "right": "issue:45027", "accept": false, - "reason": "BERT-like RoPE support and Voxtral TTS support are unrelated." + "reason": "BERT-like RoPE support and Voxtral TTS are unrelated changes." }, { "left": "issue:42738", "right": "issue:46653", "accept": false, - "reason": "BERT-like RoPE support is unrelated to the Mirko Privitera issue." + "reason": "The latter is not a technical model request; no duplicate relation." }, { "left": "issue:43671", "right": "issue:43837", "accept": false, - "reason": "Qwen3-TTS and Qwen3-ASR are adjacent but distinct modalities, so they are not the same bug or change." + "reason": "Qwen3-TTS and Qwen3-ASR share a family but are different modalities and fixes." }, { "left": "issue:43671", - "right": "issue:45027", + "right": "issue:43909", "accept": false, - "reason": "Qwen3-TTS support and Voxtral TTS support target different models and implementations." + "reason": "Different audio models; not the same code change." }, { "left": "issue:43671", - "right": "issue:45177", + "right": "issue:44121", "accept": false, - "reason": "Qwen3-TTS and DeepSeek-OCR2 are different model requests." + "reason": "Qwen3-TTS vs sparse transformer are unrelated." }, { "left": "issue:43671", - "right": "issue:44121", + "right": "issue:45027", "accept": false, - "reason": "Qwen3-TTS support and the sparse transformer request are unrelated changes." + "reason": "Both are speech-related, but TTS model support is model-specific and not mergeable into one PR here." }, { "left": "issue:43671", - "right": "issue:43909", + "right": "issue:45177", "accept": false, - "reason": "Qwen3-TTS and LFM2.5 Audio 1.5B are different audio-model requests." + "reason": "TTS support request vs OCR model request are unrelated." }, { "left": "issue:43837", "right": "issue:43909", "accept": false, - "reason": "Qwen3-ASR and LFM2.5 Audio 1.5B are distinct audio-model support requests." + "reason": "Qwen3-ASR and LFM2.5 Audio are different model support tasks." }, { "left": "issue:43837", "right": "issue:45322", "accept": false, - "reason": "Qwen3-ASR support and EUPE are unrelated model additions." - }, - { - "left": "issue:43837", - "right": "issue:44121", - "accept": false, - "reason": "Qwen3-ASR and the sparse transformer request are different concrete changes." + "reason": "Speech ASR request vs EUPE vision-language encoder request are unrelated." }, { "left": "issue:43909", "right": "issue:44205", "accept": false, - "reason": "LFM2.5 Audio 1.5B and SAM3-LiteText are different models and tasks." + "reason": "Audio model support vs SAM3-LiteText are different model requests." }, { "left": "issue:43909", "right": "issue:46645", "accept": false, - "reason": "An audio-model addition is not the same as adding SWA+sink to Granite models." + "reason": "Model-add request vs Granite SWA+sink option change are different scopes." }, { "left": "issue:44121", "right": "issue:45322", "accept": false, - "reason": "Sparse transformer support and EUPE are separate model-request changes." + "reason": "Sparse transformer request and EUPE are unrelated." }, { "left": "issue:43837", "right": "issue:45027", "accept": false, - "reason": "Qwen3-ASR and Voxtral-4B-TTS are different speech-model support requests." + "reason": "ASR and TTS are different speech code paths; not the same fix." }, { "left": "issue:37015", "right": "issue:45027", "accept": false, - "reason": "NeoBERT and Voxtral TTS are unrelated model additions." + "reason": "NeoBERT support and Voxtral TTS support are unrelated." }, { "left": "issue:43837", "right": "issue:45177", "accept": false, - "reason": "Qwen3-ASR and DeepSeek-OCR2 are different models with different implementation needs." + "reason": "ASR model support vs OCR model support are different changes." }, { "left": "issue:45322", "right": "issue:46133", "accept": false, - "reason": "EUPE and TIPSv2 are different vision-language models; similar broad area, but not the same change." + "reason": "Both are vision-language models, but they are distinct model requests with different implementation needs." }, { "left": "issue:42503", "right": "issue:46653", "accept": false, - "reason": "ModernVBERT models and the Mirko Privitera issue are unrelated." + "reason": "ModernVBERT support is unrelated to the non-technical personal-name issue." }, { "left": "issue:45027", "right": "issue:46653", "accept": false, - "reason": "Voxtral TTS support and the Mirko Privitera issue do not describe the same underlying change." + "reason": "Voxtral TTS support and the personal-name issue are unrelated." } ] }, @@ -19132,40 +19126,40 @@ "soft_pairs": [ "issue:38740|issue:46443" ], - "prepared_review_unit_hash": "698fca07850c84424ac0688a332a55e525cc6bbf54c0409b8fdea50082525111", + "prepared_review_unit_hash": "c987aaf8f1024c3a532031bfa5303d231438909bacba21b7ed0f273fdeb63500", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1505, - "estimated_input_tokens": 377, - "estimated_eval_tokens": 1010 + "serialized_chars": 1504, + "estimated_input_tokens": 376, + "estimated_eval_tokens": 1008 }, "budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1505, - "estimated_input_tokens": 377, - "estimated_eval_tokens": 1010 + "serialized_chars": 1504, + "estimated_input_tokens": 376, + "estimated_eval_tokens": 1008 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two issues are not duplicates: one is a documentation request to add `pruna` as an optimization framework, while the other is a vision RoPE refactor. They do not describe the same underlying change or bug.", - "confidence": 0.97, - "canonical_issue_reason": "Issue #46443 is the more substantive technical item in the pair; #38740 is a docs-only request and unrelated to the refactor topic.", + "summary": "These two issues are not duplicates: one is a docs request to add Pruna as an optimization framework, the other is a Vision RoPE refactor. They concern different parts of the project and different kinds of work.", + "confidence": 0.96, + "canonical_issue_reason": "Issue #46443 is the better canonical representative because it is a concrete technical refactor topic; #38740 is a separate documentation-related request.", "canonical_pr_reason": null, - "best_issue_reason": "#46443 is the better representative issue only in the sense that it is the more concrete engineering task; however, it is not a duplicate of #38740.", + "best_issue_reason": "#46443 is the most suitable issue to anchor the cluster since it is the more substantive software-change topic and not just a docs addition request.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:38740", "right": "issue:46443", "accept": false, - "reason": "Different subjects and domains: docs entry for an optimization framework vs. a vision RoPE refactor. No shared target, filenames, or code path." + "reason": "Different subjects and artifact types: a docs enhancement for Pruna versus a Vision RoPE refactor. No evidence they describe the same underlying bug or change." } ] }, @@ -19195,12 +19189,12 @@ "soft_pairs": [ "issue:41084|issue:43037" ], - "prepared_review_unit_hash": "302acae6e976d5b0d559b52ae0fab5930e335641863188236e0c6f39ea43b2d9", + "prepared_review_unit_hash": "4da79feca56257751fb8425c42bf1254ffdd96a3aa0040f48168807ec9efc2c7", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1480, + "serialized_chars": 1479, "estimated_input_tokens": 370, "estimated_eval_tokens": 996 }, @@ -19208,7 +19202,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1480, + "serialized_chars": 1479, "estimated_input_tokens": 370, "estimated_eval_tokens": 996 }, @@ -19217,18 +19211,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two issues are not duplicates: one requests a generic \"Set Block Decoding\" change, while the other asks for \"DeepSeek v3.2 support.\" They touch different underlying problems and should remain separate.", - "confidence": 0.96, - "canonical_issue_reason": null, + "summary": "These two issues do not look like duplicates: one is about enabling block decoding, while the other asks for DeepSeek v3.2 model support. They are different feature requests with no shared targets or code context.", + "confidence": 0.94, + "canonical_issue_reason": "Issue #41084 is the better canonical issue only in the sense that it is more specific and likely to map to a concrete implementation task; however, it is not a duplicate of #43037.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43037 is the more concrete and actionable issue, but it is not a duplicate of issue:41084.", + "best_issue_reason": "#41084 is the most actionable issue in the pair because it states a narrower technical change. #43037 is a broader model-support request and does not subsume the block-decoding request.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:41084", "right": "issue:43037", "accept": false, - "reason": "Titles and implied scopes are unrelated: block decoding configuration vs model support for DeepSeek v3.2. No shared target, filename, or concrete bug path suggests the same underlying issue." + "reason": "Reject: the titles describe different underlying requests. 'Set Block Decoding' concerns a decoding behavior/feature, while 'DeepSeek v3.2 support' concerns adding support for a specific model version. No explicit linkage, shared targets, or evidence they refer to the same bug/change." } ] }, @@ -19258,40 +19252,40 @@ "soft_pairs": [ "issue:41306|issue:41826" ], - "prepared_review_unit_hash": "a0f1e8222a646ad364e5f71a10efe0ed153795fec94d63502ceea384d66058ae", + "prepared_review_unit_hash": "e3150bad90c7054af57d7420885ade23fad31d0f78bb88ba12471ce854fd5b3a", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1585, - "estimated_input_tokens": 397, - "estimated_eval_tokens": 1050 + "serialized_chars": 1584, + "estimated_input_tokens": 396, + "estimated_eval_tokens": 1048 }, "budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1585, - "estimated_input_tokens": 397, - "estimated_eval_tokens": 1050 + "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": "The two issues are related only at a broad theme level (memory-efficient model variants), but they describe different features for different modules: FLCE/TiledFusedLogitsLoss flag handling vs. integrating TiledMLP. They are not duplicates.", - "confidence": 0.93, - "canonical_issue_reason": "Issue #41306 is the more concrete, actionable report: it names a specific API path (`from_pretrained()`) and a specific missing flag for TiledFusedLogitsLoss/LigerFusedLinearCrossEntropy.", + "summary": "The two issues are both about memory-efficient model components, but they describe different feature requests and different code paths: FLCE/TiledFusedLogitsLoss support in `from_pretrained()` versus adding TiledMLP integration. They are not duplicates.", + "confidence": 0.96, + "canonical_issue_reason": "No canonical duplicate here; 41306 is a concrete `from_pretrained()` integration request for FLCE/TiledFusedLogitsLoss, while 41826 is a separate TiledMLP feature request.", "canonical_pr_reason": null, - "best_issue_reason": "Issue #41306 is the best representative because it is narrowly scoped and likely to map to a single fix; #41826 is a separate feature request for a different component.", + "best_issue_reason": "41306 is the more specific and actionable tracking issue, but it should still be kept separate from 41826 because the requested change is different.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:41306", "right": "issue:41826", "accept": false, - "reason": "Different underlying requests: one is about enabling a FLCE/TiledFusedLogitsLoss flag in `from_pretrained()`, the other is about integrating TiledMLP for lower memory use. They may share a theme, but not the same bug or change." + "reason": "Not the same underlying bug/change: one asks for a `from_pretrained()` flag for FLCE/TiledFusedLogitsLoss, the other for integrating TiledMLP to reduce memory footprint." } ] }, @@ -19321,7 +19315,7 @@ "soft_pairs": [ "issue:43598|issue:43600" ], - "prepared_review_unit_hash": "b585555e335a97947516661cfa5b3602d82fb84fed62a4e1283c508cb7d0b275", + "prepared_review_unit_hash": "0d1bb28870fffd4a041fc72e874888271b07439aca6913b4e8f0935e5e900be0", "component_budget": { "node_count": 2, "item_count": 2, @@ -19343,18 +19337,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are related only at a broad Trainer/customization level, but they describe different problems: one is about `remove_unused_column` behavior, the other about FLOPs estimation for custom trainers. They should not be merged as duplicates.", - "confidence": 0.91, - "canonical_issue_reason": "Neither issue is a clear canonical for the other; they cover different Trainer behaviors and fixing one would not resolve the other.", + "summary": "Both issues mention Trainer customization, but they target different problems: one is about `remove_unused_column` behavior, the other about FLOPS estimation for custom trainers. They are related in subsystem, not duplicates.", + "confidence": 0.9, + "canonical_issue_reason": "Issue #43598 is the broader, more central Trainer customization request and better represents the cluster theme.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 43598 is slightly more concrete and self-contained around a specific Trainer setting (`remove_unused_column`), so it is the better representative of the pair if one must be chosen.", + "best_issue_reason": "Issue #43598 is the best issue representative because it focuses on the underlying Trainer customization behavior rather than a narrower metrics-estimation concern.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43598", "right": "issue:43600", "accept": false, - "reason": "Both mention Trainer customizability, but one concerns unused-column handling and the other concerns FLOPs accounting. They are distinct bugs/features, not the same underlying issue." + "reason": "They concern different code paths and outcomes: `remove_unused_column` customizability vs FLOPS estimation for custom trainers. Similar theme, but not the same underlying bug or change." } ] }, @@ -19384,7 +19378,7 @@ "soft_pairs": [ "issue:43684|issue:44405" ], - "prepared_review_unit_hash": "68315b08333b0050879bb0d3a6ab7ade23584380976e8e6d6afe3d6a16705903", + "prepared_review_unit_hash": "603739588dcb4328e1aabd0637d9727dba7ace564cadc1126774670d475d7010", "component_budget": { "node_count": 2, "item_count": 2, @@ -19406,18 +19400,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two issues are about different Qwen model support gaps: one requests registration for Qwen3-Omni in AutoModel/AutoModelForConditionalGeneration, the other requests AutoModelForSequenceClassification support for Qwen3.5. They are related at a broad library level but not duplicates of the same underlying bug or change.", - "confidence": 0.95, - "canonical_issue_reason": null, + "summary": "These are related Qwen model-registration requests, but they target different API surfaces and model variants: one asks for AutoModel/AutoModelForConditionalGeneration registration for Qwen3-Omni, while the other asks for AutoModelForSequenceClassification support for Qwen3.5. They are not the same underlying bug/change.", + "confidence": 0.92, + "canonical_issue_reason": "Issue #43684 is the broader infrastructure request (model registration into AutoModel/AutoModelForConditionalGeneration) and is phrased as a general missing-registration problem, so it is the better canonical issue among the two.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "Issue #43684 is the best representative of the cluster because it describes a core registration gap rather than a narrower task-specific class addition.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43684", "right": "issue:44405", "accept": false, - "reason": "Different model families and different API surfaces (Qwen3-Omni registration vs Qwen3.5 sequence classification support); not the same underlying issue." + "reason": "Different model families and different heads/tasks: Qwen3-Omni registration for base/generative auto classes vs Qwen3.5 sequence classification support. Related, but not the same fix." } ] }, @@ -19454,12 +19448,12 @@ "issue:43845|issue:43859", "issue:43846|issue:43859" ], - "prepared_review_unit_hash": "bbc736afb87ad6b672bbca35f58ce4bf839037d3938649da9a0bdf644f0bc3ff", + "prepared_review_unit_hash": "6d367fd5ad4c7516e300c170a38d142b939052024ed20cec5accefa9127fa4bf", "component_budget": { "node_count": 4, "item_count": 4, "soft_pair_count": 6, - "serialized_chars": 4265, + "serialized_chars": 4267, "estimated_input_tokens": 1067, "estimated_eval_tokens": 2390 }, @@ -19467,7 +19461,7 @@ "node_count": 4, "item_count": 4, "soft_pair_count": 6, - "serialized_chars": 4265, + "serialized_chars": 4267, "estimated_input_tokens": 1067, "estimated_eval_tokens": 2390 }, @@ -19476,48 +19470,48 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster appears to be mostly low-information duplicate issue noise around the repeated title \"huggingface\"; one outlier titled \"Fett\" does not look related. No PRs are present.", - "confidence": 0.78, - "canonical_issue_reason": "Issue 43845 is the earliest of the near-identical \"huggingface\" issues and is the best representative of that duplicate-looking subset; 43846 and 43859 look like later repeats.", + "summary": "Four closed issues with extremely sparse content; three share the generic title \"huggingface\" and one is unrelated-looking \"Fett.\" There is not enough concrete signal to confirm any duplicate cluster beyond superficial title similarity, and there are no PRs.", + "confidence": 0.9, + "canonical_issue_reason": "Issue 43845 is the best of a weak set: it is one of the repeated \"huggingface\" reports and has the longest body among those near-duplicates, so it is the least arbitrary anchor. Still, the evidence is very thin.", "canonical_pr_reason": null, - "best_issue_reason": "43845 is the strongest issue candidate because it is the earliest repeated \"huggingface\" report and has the same title as the other likely duplicates.", + "best_issue_reason": "43845 is the most plausible representative issue because it matches the repeated title pattern and appears marginally more descriptive than the others. However, the cluster lacks actionable details, so it is only a weak canonical choice.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43835", "right": "issue:43845", "accept": false, - "reason": "Titles differ completely and there is no body/detail evidence tying \"Fett\" to the \"huggingface\" reports." + "reason": "Only superficial text similarity; \"Fett\" and \"huggingface\" do not clearly describe the same bug or request." }, { "left": "issue:43835", "right": "issue:43846", "accept": false, - "reason": "No concrete evidence of the same underlying issue; only generic soft similarity." + "reason": "No concrete shared problem statement or target; title similarity is insufficient to treat these as duplicates." }, { "left": "issue:43835", "right": "issue:43859", "accept": false, - "reason": "Different titles and no shared target or discussion details to support duplication." + "reason": "Different titles and no supporting detail linking them to the same underlying issue." }, { "left": "issue:43845", "right": "issue:43846", "accept": false, - "reason": "Same title, but the reports are empty/low-detail; not enough evidence to confirm the same bug or change." + "reason": "Both are generically titled \"huggingface,\" but there is no substantive body, target, or repro detail proving the same underlying bug." }, { "left": "issue:43845", "right": "issue:43859", "accept": false, - "reason": "Generic title match alone is insufficient to prove the same underlying issue." + "reason": "Shared generic wording alone is not enough; the sparse reports do not establish the same issue." }, { "left": "issue:43846", "right": "issue:43859", "accept": false, - "reason": "No substantive evidence beyond repeated wording; reject as duplicate candidate." + "reason": "The reports are too underspecified to confirm they refer to the same problem rather than unrelated mention of the same keyword." } ] }, @@ -19558,98 +19552,95 @@ "issue:44572|issue:44937", "issue:44573|issue:44937" ], - "prepared_review_unit_hash": "97dcd395bdae0526566df0291ba40e6fd89c2579af8e0f209c4b7fe46b0ffa56", + "prepared_review_unit_hash": "6ad56872d2c285fe966664734a39ebc406449f86954ad43416371314d1c88f9a", "component_budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 9, - "serialized_chars": 5788, - "estimated_input_tokens": 1447, - "estimated_eval_tokens": 3150 + "serialized_chars": 5794, + "estimated_input_tokens": 1449, + "estimated_eval_tokens": 3154 }, "budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 9, - "serialized_chars": 5788, - "estimated_input_tokens": 1447, - "estimated_eval_tokens": 3150 + "serialized_chars": 5794, + "estimated_input_tokens": 1449, + "estimated_eval_tokens": 3154 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is mostly low-signal spam/placeholder issues. The three \"\" reports are likely duplicates based on identical titles and similarly tiny bodies, but the evidence is not strong enough to call them definitively identical. The other two issues have different titles and do not appear to match that spam cluster.", - "confidence": 0.84, - "canonical_issue_reason": "issue:44371 is the earliest of the likely duplicate \"\" issues and is the best representative for that cluster.", + "summary": "This cluster looks like low-signal closed noise/spam rather than a real duplicate set. Three issues share the exact title \"\", but there is no substantive body or code-context evidence tying them to the same underlying bug. The remaining two issues are generic/unrelated titles and do not match the spam subgroup.", + "confidence": 0.88, + "canonical_issue_reason": "issue:44371 is the earliest of the repeated exact-title spam issues and is the cleanest representative of that subgroup.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44371 is the best single issue candidate because it is earliest among the likely duplicate spam reports and has matching minimal content.", + "best_issue_reason": "issue:44371 is the best overall representative because it matches two other items exactly by title and is earlier than the rest.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44371", "right": "issue:44572", - "accept": true, - "reason": "Likely duplicate pair: same placeholder title and similarly empty bodies, with no contrary detail in the packet." + "accept": false, + "reason": "Same placeholder title, but no body or context showing a shared underlying bug/change; looks like duplicate noise, not a real duplicate issue." }, { "left": "issue:44371", "right": "issue:44573", - "accept": true, - "reason": "Likely duplicate pair: same placeholder title and similarly empty bodies, with no contrary detail in the packet." + "accept": false, + "reason": "Exact same spam title, but there is no substantive evidence that these describe the same real problem." }, { "left": "issue:44371", "right": "issue:44418", "accept": false, - "reason": "Different title and different apparent intent; this looks like a documentation-related item, not the same placeholder spam report." + "reason": "Titles are unrelated and generic; no indication they refer to the same issue." }, { "left": "issue:44371", "right": "issue:44937", "accept": false, - "reason": "Different title and no concrete evidence they describe the same underlying issue." + "reason": "\"Check out \\\"Google LLC\\\"\" is unrelated to the spam placeholder issue." }, { "left": "issue:44418", "right": "issue:44572", "accept": false, - "reason": "A documentation enhancement request does not look like the same underlying issue as a placeholder spam report." + "reason": "Generic documentation suggestion versus spam placeholder; no shared underlying bug/change." }, { "left": "issue:44418", "right": "issue:44573", "accept": false, - "reason": "A documentation enhancement request does not look like the same underlying issue as a placeholder spam report." + "reason": "Generic documentation suggestion versus spam placeholder; no shared underlying bug/change." }, { "left": "issue:44572", "right": "issue:44573", - "accept": true, - "reason": "Likely duplicate pair: same placeholder title and similarly empty bodies, with no contrary detail in the packet." + "accept": false, + "reason": "Identical placeholder titles, but still only spam/noise with no real bug or change description to confirm a true duplicate." }, { "left": "issue:44572", "right": "issue:44937", "accept": false, - "reason": "Different title and no concrete evidence they describe the same underlying issue." + "reason": "Unrelated titles with no evidence of the same problem." }, { "left": "issue:44573", "right": "issue:44937", "accept": false, - "reason": "Different title and no concrete evidence they describe the same underlying issue." + "reason": "Unrelated titles with no evidence of the same problem." } ] }, - "evaluator_result": { - "accept": false, - "feedback": "The cluster-level direction is reasonable, but the response overstates the evidence. The packet only shows identical titles and very small bodies, plus soft_similarity; it does not prove the three issues have identical content or are clear/true duplicates. The pair verdicts should be phrased more conservatively as likely duplicates, not definitive ones." - }, - "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 }, @@ -19672,12 +19663,12 @@ "soft_pairs": [ "issue:45145|issue:45151" ], - "prepared_review_unit_hash": "4165118dff456224a224cfb606697cd3ee0633adee7605cce12b5f87e9a0aa9e", + "prepared_review_unit_hash": "491798b717d6a702e48ffb85074d3741d5e1af961e175a3413016eee02a2eaa2", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1645, + "serialized_chars": 1646, "estimated_input_tokens": 412, "estimated_eval_tokens": 1080 }, @@ -19685,7 +19676,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1645, + "serialized_chars": 1646, "estimated_input_tokens": 412, "estimated_eval_tokens": 1080 }, @@ -19694,24 +19685,24 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two issues are strong duplicate candidates: they share the exact title and very similar body lengths, and #45145 has an inbound reference. The metadata is still limited, so this is best treated as a likely duplicate pair rather than a proven one; #45145 is the better canonical candidate because it is the more connected report.", + "summary": "The two issues are highly similar on title and general shape, but the packet only gives soft similarity and does not provide enough evidence to conclusively treat them as duplicates. #45145 is still the better canonical representative because it has an inbound reference and is the more established thread.", "confidence": 0.79, - "canonical_issue_reason": "#45145 is the better canonical issue choice because it has the same title as #45151 and the only inbound reference, so it is the more connected instance if these are duplicates.", + "canonical_issue_reason": "#45145 is the better canonical issue because it has the same title as #45151 and already has an inbound reference, so it is the likelier existing discussion to retain.", "canonical_pr_reason": null, - "best_issue_reason": "#45145 is the strongest overall issue candidate for the same reason: identical title, slightly more linkage, and no metadata suggesting a different scope.", + "best_issue_reason": "#45145 is the best overall issue representative for this cluster because it has the stronger reference signal while still matching the same topic/title as #45151.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45145", "right": "issue:45151", "accept": false, - "reason": "Identical titles and similar body lengths suggest a possible duplicate, but the packet does not include body text, explicit links, filenames, or diff evidence to confidently establish the same underlying issue." + "reason": "The only explicit pair evidence is soft_similarity, and deterministic_accept is false. The issues share the same title and similar size, but the packet does not provide enough concrete evidence to confidently mark them as the same underlying issue." } ] }, "evaluator_result": { "accept": false, - "feedback": "The duplicate/canonical direction is plausible and grounded by the identical title, similar body lengths, and the inbound reference on #45145. However, the writeup overstates the evidence in a few places: phrases like \u201cmatches almost exactly,\u201d \u201chighly similar content,\u201d and \u201csame underlying issue\u201d go beyond what the packet shows, since there is no body text, explicit link, or filename evidence\u2014only soft similarity with deterministic_accept=false. The conclusion should be framed more conservatively." + "feedback": "The canonical-choice reasoning is mostly grounded, but the soft-edge verdict is too strong for the packet: the only pair evidence is `soft_similarity`, and `deterministic_accept` is false. The summary overstates certainty by saying the issues are duplicates with \u201chighly similar body content\u201d and no distinct scope, when the packet only directly supports same title, similar length, and soft similarity. Please keep this more conservative or mark the edge as unaccepted." }, "evaluator_used": true, "retried": true, @@ -19792,40 +19783,40 @@ "soft_pairs": [ "issue:45672|issue:45674" ], - "prepared_review_unit_hash": "f279be5c658ad608d5b4607166a6fc2b691d79f680312bc58eaf7ce1ca505174", + "prepared_review_unit_hash": "90f51230189b38d1f482db70426d7adca9f8885f589e8c7e14880d4338fcd8f0", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1673, - "estimated_input_tokens": 419, - "estimated_eval_tokens": 1094 + "serialized_chars": 1672, + "estimated_input_tokens": 418, + "estimated_eval_tokens": 1092 }, "budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1673, - "estimated_input_tokens": 419, - "estimated_eval_tokens": 1094 + "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": "The two issues both involve 4-bit inference, but they describe different failure modes and likely different fixes: one is a Gemma4/audio-module torch.finfo() TypeError on uint8 weights, the other is a BitsAndBytesConfig regression where llm_int8_skip_modules removes the default lm_head exclusion and triggers an AssertionError. They should not be merged as duplicates.", - "confidence": 0.94, - "canonical_issue_reason": "Issue #45674 is the better canonical issue because it states a broader, configuration-level regression with a clear trigger and likely shared impact across models.", + "summary": "These are related to 4-bit inference but describe two different bugs in different code paths, so they should not be deduplicated.", + "confidence": 0.96, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Issue #45674 is the strongest standalone issue: it names the exact problematic setting, the observed failure, and the inference mode affected.", + "best_issue_reason": "Issue 45674 is the better cluster representative because it describes a clearer, more general BitsAndBytesConfig regression with a concrete failure mode.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45672", "right": "issue:45674", "accept": false, - "reason": "Both are 4-bit inference bugs, but the root causes differ: torch.finfo() on uint8 weights in Gemma4 audio modules vs. llm_int8_skip_modules clearing the default lm_head exclusion and causing an AssertionError. They are not the same underlying bug." + "reason": "Different underlying bugs: one is a torch.finfo() TypeError on uint8 weights in audio modules, the other is llm_int8_skip_modules clearing the default lm_head exclusion and triggering an AssertionError in 4-bit inference." } ] }, @@ -19855,7 +19846,7 @@ "soft_pairs": [ "issue:45762|issue:45763" ], - "prepared_review_unit_hash": "a974efcb01c9fce9921fb171d031b48303d2b4b6321877dbb9a9127a2e74bd35", + "prepared_review_unit_hash": "02f7cc330a8a635bdc66d0c9e95a9c3d868a80f146905aacbe29e9adc53dafa0", "component_budget": { "node_count": 2, "item_count": 2, @@ -19877,18 +19868,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two closed issues are not duplicates: they are unrelated name/date strings with no technical content or shared bug report context, and likely spam or accidental submissions.", - "confidence": 0.98, - "canonical_issue_reason": null, + "summary": "These two closed issues do not appear to describe the same underlying bug or request; they are just unrelated name/date strings with no body or shared target.", + "confidence": 0.96, + "canonical_issue_reason": "No canonical issue: the titles are unrelated and there is no evidence of a shared problem, repro, or target.", "canonical_pr_reason": null, - "best_issue_reason": "No issue is a suitable canonical representative; the two issues are about different people and dates and do not describe the same underlying problem.", + "best_issue_reason": "Neither issue is a good duplicate anchor; if forced, 45762 is no more representative than 45763, but there is no real basis to choose one.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45762", "right": "issue:45763", "accept": false, - "reason": "The titles refer to different person-name/date strings and contain no shared bug, feature, or code-path context; they do not look like the same underlying issue." + "reason": "Do not merge: the titles are different personal-name/date entries with no body, links, files, or discussion indicating the same underlying issue." } ] }, @@ -19918,40 +19909,40 @@ "soft_pairs": [ "issue:45764|issue:45765" ], - "prepared_review_unit_hash": "7e311a3a46f912fa5e0e89a11ce589a31ed20012049aef706ca47ecae8e44708", + "prepared_review_unit_hash": "94d4fb9f8d8ee93318d18cc91698b3752c6381080ba77b90b2001150921d9272", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1481, - "estimated_input_tokens": 371, - "estimated_eval_tokens": 998 + "serialized_chars": 1480, + "estimated_input_tokens": 370, + "estimated_eval_tokens": 996 }, "budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1481, - "estimated_input_tokens": 371, - "estimated_eval_tokens": 998 + "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 closed issues appear unrelated: they reference different people and different dates, with no body text or linking evidence to suggest the same underlying report.", - "confidence": 0.96, - "canonical_issue_reason": "No canonical issue: the titles describe different individuals and different dates, and there is no supporting discussion or linkage indicating duplication.", + "summary": "The two issues are just distinct name/date placeholders and do not appear to describe the same underlying bug or request.", + "confidence": 0.97, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Neither issue is a strong global representative because there is no duplicate relationship; if forced, the earlier issue (#45764) is only marginally earlier, not meaningfully more canonical.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45764", "right": "issue:45765", "accept": false, - "reason": "Reject: the titles identify different people and different dates, and there is no additional evidence that they refer to the same underlying bug or request." + "reason": "The titles reference different people and different dates, with no shared technical context, files, or issue targets; this does not look like the same underlying issue." } ] }, @@ -19981,7 +19972,7 @@ "soft_pairs": [ "issue:45810|issue:46295" ], - "prepared_review_unit_hash": "bd5f9edaba3f78848c98b416ce4d470f76bc7ecb1f3508c662a693e5c59ac9c9", + "prepared_review_unit_hash": "84c955ba51257653795b5ba0362cfc9a7c55d3e1ccf6a305ed5dee884a142557", "component_budget": { "node_count": 2, "item_count": 2, @@ -20003,18 +19994,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two issues are not duplicates: one requests a new Qwen3_5 token-classification model class for value-model use, while the other requests FP8 training support. They address different features and code paths.", - "confidence": 0.98, - "canonical_issue_reason": "No canonical issue: the requests are unrelated in scope and target different functionality.", + "summary": "The two issues are unrelated feature requests: one asks for a new token-classification head for Qwen3_5, the other asks for FP8 training support. They should not be clustered as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "If a representative must be chosen, #46295 is the broader and more referenced feature request, so it is the better cluster placeholder.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45810", "right": "issue:46295", "accept": false, - "reason": "Different underlying changes: model-class addition for token classification vs FP8 training support. They do not describe the same bug or change." + "reason": "Different requests in different areas: model-head addition for token classification vs FP8 training support. No shared code-path or concrete duplicate bug/change." } ] }, @@ -20054,7 +20045,7 @@ "issue:45871|issue:45873", "issue:45872|issue:45873" ], - "prepared_review_unit_hash": "e64dfe0ae51de8f6a337f6473814dbd6b3cbecbb4cabbb49e6fdd471504865dc", + "prepared_review_unit_hash": "d53ca198c00e3878831813bdc61ab71d1a232805dfc7d039db65933ce50c542b", "component_budget": { "node_count": 5, "item_count": 5, @@ -20076,69 +20067,69 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The cluster contains two separate exact duplicate pairs\u201445870/45871 and 45872/45873\u2014plus one related but distinct issue (45869). There is no basis to rank one exact pair above the other; any representative choice between the two pairs is arbitrary.", - "confidence": 0.97, - "canonical_issue_reason": "Issue 45870 is a valid canonical representative only by arbitrary tie-break: it is one member of an exact-duplicate pair with 45871, but the packet also shows an equally exact duplicate pair at 45872/45873, so no pair is demonstrably stronger than the other.", + "summary": "This cluster contains two separate exact-duplicate issue pairs: 45870/45871 and 45872/45873. Issue 45869 is unrelated promotional dataset content and should not be merged with the DataForge requests.", + "confidence": 0.95, + "canonical_issue_reason": "45872 is the strongest representative issue: it is substantive, clearly scoped, and has an exact duplicate (45873). The 45870/45871 pair is also duplicated, but it is a more generic example request.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45870 is suitable as a representative duplicate issue because it has an exact twin (45871). That said, 45872 is equally suitable as a representative of its own exact-duplicate pair, so this is a tie rather than a strictly better choice.", + "best_issue_reason": "45872 is the best single issue to keep because it is the clearest, most specific request among the set and stands in an exact duplicate pair.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45869", "right": "issue:45870", "accept": false, - "reason": "Related financial/DataForge topic, but the requests are different: a dataset bundle versus a prediction example." + "reason": "Different underlying content: a dataset-bundle promotion versus an example request for financial time-series prediction." }, { "left": "issue:45869", "right": "issue:45871", "accept": false, - "reason": "Related financial/DataForge topic, but the requests are different: a dataset bundle versus a prediction example." + "reason": "Different underlying content: a dataset-bundle promotion versus an example request for financial time-series prediction." }, { "left": "issue:45870", "right": "issue:45871", "accept": true, - "reason": "Exact title/body match with no visible substantive difference; this is a true duplicate pair." + "reason": "Exact duplicate issue pair with the same title and apparent same request." }, { "left": "issue:45870", "right": "issue:45872", "accept": false, - "reason": "Same broad area, but not the same concrete request; the scraping-service component makes it a different issue." + "reason": "Related domain, but different request: example prediction workflow versus time series plus scraping service." }, { "left": "issue:45870", "right": "issue:45873", "accept": false, - "reason": "Same broad area, but not the same concrete request; the scraping-service component makes it a different issue." + "reason": "Related domain, but different request: example prediction workflow versus time series plus scraping service." }, { "left": "issue:45871", "right": "issue:45872", "accept": false, - "reason": "Same broad area, but different underlying request: generic prediction example versus prediction plus scraping service." + "reason": "Related domain, but different request: example prediction workflow versus time series plus scraping service." }, { "left": "issue:45871", "right": "issue:45873", "accept": false, - "reason": "Same broad area, but not the same concrete request; the scraping-service component makes it a different issue." + "reason": "Related domain, but different request: example prediction workflow versus time series plus scraping service." }, { "left": "issue:45872", "right": "issue:45873", "accept": true, - "reason": "Exact title/body match with matching scope and no visible substantive difference; this is a true duplicate pair." + "reason": "Exact duplicate issue pair with the same title and apparent same request." } ] }, "evaluator_result": { - "accept": false, - "feedback": "The duplicate-pair judgments are mostly conservative, but the canonical selection overstates what the packet supports. The packet shows two equally exact duplicate pairs (45870/45871 and 45872/45873), so there is no grounded basis for calling 45872/45873 the uniquely \"strongest\" pair or 45872 the \"more descriptive\" version of its twin. Please soften that to an arbitrary representative choice or note the tie explicitly." + "accept": true, + "feedback": "The summary is grounded in the packet and the soft-edge decisions are conservative. The two exact-title, same-length pairs (45870/45871 and 45872/45873) are reasonable duplicate calls, and the 45869 rejections are well-supported by the titles. No PR-pair merge issue applies here." }, "evaluator_used": true, - "retried": true, + "retried": false, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -20162,7 +20153,7 @@ "soft_pairs": [ "issue:46439|issue:46441" ], - "prepared_review_unit_hash": "04c88ef446d7f51c81ae6471c63e5eca76bd12beacc5a3e4651c761736dfcaf3", + "prepared_review_unit_hash": "fca36374313af495e35f992bc69a0dc1adc93c8867bc8bbf3ac92119c71cd34d", "component_budget": { "node_count": 2, "item_count": 2, @@ -20184,18 +20175,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two open issues both involve static cache and linear-attention, but they describe different failures: one is a shape-corruption/runtime error during `early_initialization`, the other is a Qwen3.5 `KeyError: 'linear_attention'`. They are related in area, but not clearly the same underlying bug.", - "confidence": 0.8, - "canonical_issue_reason": "Issue 46439 is the better cluster representative because it states a broader, more root-cause-oriented failure mode (`early_initialization` corrupting linear-attention shapes) rather than a single model-specific crash symptom.", + "summary": "Both issues involve `cache_implementation=\"static\"` and `linear_attention`, but they describe different failure modes: one is a shape-corruption `RuntimeError` during early initialization, the other is a `KeyError: 'linear_attention'` in Qwen3.5. Related, but not clearly the same duplicate.", + "confidence": 0.81, + "canonical_issue_reason": "Issue 46439 is the broader, more general report of the static-cache/linear-attention initialization bug and better captures the core code-path problem.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 46439 is the strongest tracker for the cluster since it describes the likely underlying static-cache/linear-attention problem in a general way and names the problematic initialization path.", + "best_issue_reason": "Issue 46439 is the better representative because it points at the underlying static-cache corruption path rather than a model-specific `KeyError` symptom.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:46439", "right": "issue:46441", "accept": false, - "reason": "Both mention static cache and linear-attention, but the reported bugs differ materially: 46439 is about wrong shapes from `early_initialization`, while 46441 is a `KeyError` for a missing `linear_attention` entry on Qwen3.5. That is not enough evidence to treat them as the same bug." + "reason": "Same subsystem, but the concrete failures differ: 46439 reports wrong-shape corruption and `RuntimeError`, while 46441 reports a missing `linear_attention` entry causing `KeyError`. Not enough to treat as the same underlying bug." } ] }, @@ -20225,7 +20216,7 @@ "soft_pairs": [ "issue:46551|issue:46552" ], - "prepared_review_unit_hash": "03bbf7cd7369a0c099cf7b1979fc3dce31d9df6a7e86e15c6f8d08f4e2866616", + "prepared_review_unit_hash": "512215a01f854c372a69e7eb38f497297d21e1a5b6928c769d1fdcf630f75080", "component_budget": { "node_count": 2, "item_count": 2, @@ -20247,24 +20238,24 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two issues are effectively identical reports of the same Wav2Vec2CTCTokenizer v5 special-token bug; issue 46552 is the better canonical tracker because it is open and already has inbound references.", + "summary": "The two issues are effectively identical reports of the same Wav2Vec2CTCTokenizer v5 regression, so they should be treated as duplicates.", "confidence": 0.99, - "canonical_issue_reason": "Issue 46552 is open, already referenced, and matches 46551 exactly in title and apparent content, making it the better living tracker.", + "canonical_issue_reason": "issue:46552 is the better surviving report because it is open and already has inbound references, making it the more useful hub for the duplicate cluster.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 46552 is the most suitable representative because it is open and appears to be the active discussion thread for the same bug.", + "best_issue_reason": "issue:46552 best represents the cluster for ongoing tracking and discussion; it matches the other issue exactly and is the active report.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:46551", "right": "issue:46552", "accept": true, - "reason": "Same title and same underlying tokenizer bug; these are duplicate reports of the exact same issue." + "reason": "Exact same title and same underlying bug description; these are duplicate reports of the same tokenizer regression." } ] }, "evaluator_result": { "accept": true, - "feedback": "The conclusions are grounded in the packet: the issues have identical titles, identical body length, and very strong similarity evidence (jaccard 1.0). Choosing the open issue 46552 as the canonical tracker is conservative and supported by its inbound references. No overstatement beyond the available evidence." + "feedback": "The conclusion is grounded in the packet: both issues have the same title, identical body length, and a perfect soft-similarity score, with 46552 open and carrying inbound references. Treating 46552 as the canonical duplicate hub is a conservative, well-supported choice." }, "evaluator_used": true, "retried": false, @@ -20291,12 +20282,12 @@ "soft_pairs": [ "issue:46657|issue:46692" ], - "prepared_review_unit_hash": "f66ee13e2f78ff21c30bfdea60345e3baad4918d4ca8c103c3bc7211c5341449", + "prepared_review_unit_hash": "d447cb0095c500d27c06ca4c483e3c9f07832b33607d8f8e98b69c86a32bcc3e", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1565, + "serialized_chars": 1568, "estimated_input_tokens": 392, "estimated_eval_tokens": 1040 }, @@ -20304,7 +20295,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1565, + "serialized_chars": 1568, "estimated_input_tokens": 392, "estimated_eval_tokens": 1040 }, @@ -20313,18 +20304,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are not duplicates: one reports a tensor shape mismatch during object-detection training, the other reports an out-of-memory failure after several epochs in RT-DETR v2. They describe different failure modes and likely different root causes.", - "confidence": 0.96, - "canonical_issue_reason": "issue #46657 is the cleaner representative report: it describes a concrete runtime error with a specific symptom and training context.", + "summary": "These two issues are not the same underlying bug: one reports a tensor-size mismatch during object detection training, while the other reports an out-of-memory failure after several epochs in RT-DETR v2 training.", + "confidence": 0.93, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue #46657 is the best issue anchor for this cluster because it is more specific and actionable, while #46692 is a separate OOM symptom.", + "best_issue_reason": "Neither issue is a clean canonical duplicate target; if one must be surfaced, #46657 is the more concrete, immediately reproducible failure report.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:46657", "right": "issue:46692", "accept": false, - "reason": "Different bug types: tensor size mismatch versus OOM after epochs. They do not appear to be the same underlying defect." + "reason": "Different failure modes and likely different code paths: tensor shape mismatch vs memory exhaustion after epochs. Shared object-detection training context is too broad to treat as the same bug." } ] }, @@ -20386,160 +20377,178 @@ "pull_request:44827|pull_request:45655", "pull_request:43251|pull_request:43254" ], - "prepared_review_unit_hash": "d194a74868ace96562135b095939233a60863ae0a0d0adfca57f3b724c3a1ff8", + "prepared_review_unit_hash": "67e4cc8c91cbfd41cd58dee8c75a04dd390f9afb69fb563482fdcdb6aa38fcd9", "component_budget": { "node_count": 15, "item_count": 15, "soft_pair_count": 20, - "serialized_chars": 40208, - "estimated_input_tokens": 10052, - "estimated_eval_tokens": 20360 + "serialized_chars": 40203, + "estimated_input_tokens": 10051, + "estimated_eval_tokens": 20358 }, "budget": { "node_count": 15, "item_count": 15, "soft_pair_count": 20, - "serialized_chars": 40208, - "estimated_input_tokens": 10052, - "estimated_eval_tokens": 20360 + "serialized_chars": 40203, + "estimated_input_tokens": 10051, + "estimated_eval_tokens": 20358 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is mostly false-positive similarity across unrelated PRs. The only true duplicate pair is the two `fixed_cross_entropy` patches, which both address the same `loss_utils.py` API gap.", - "confidence": 0.94, + "summary": "This cluster is mostly a loose set of unrelated PRs grouped by broad soft similarity. The only clear duplicate-like pair is the two `fixed_cross_entropy` PRs, which both target the same issue and same code path; the rest are too different in model, subsystem, or intent to merge together.", + "confidence": 0.41, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 43251 is the best canonical representative for the duplicate pair: it is directly issue-linked, narrowly scoped to `fixed_cross_entropy`, and cleanly describes the concrete fix of forwarding cross-entropy kwargs.", + "canonical_pr_reason": "PR 43251 is the clearest focused bug-fix PR in the cluster: it has an explicit issue target, a single-file change, and directly addresses the `cross_entropy` kwargs forwarding problem.", "best_issue_reason": null, - "best_pr_reason": "PR 43251 is the strongest overall representative of the only real duplicate here because it is the more explicit, issue-targeted implementation of the same loss-function change.", + "best_pr_reason": "PR 43251 is the best representative for triage because it is the most concrete and scoped fix, while the other PRs are either unrelated feature additions, docs/tests, or broad refactors.", "soft_edge_verdicts": [ { "left": "pull_request:39895", "right": "pull_request:43424", "accept": false, - "reason": "Different changes: VideoPrism model addition vs. an executorch dynamic-shape export test." + "reason": "Different features: VideoPrism model addition vs. executorch dynamic-shape test. Same broad ML stack, but not the same underlying change." }, { "left": "pull_request:41855", "right": "pull_request:43823", "accept": false, - "reason": "Unrelated scopes: Mistral tokenizer methods vs. adding the MobileLLM model." + "reason": "Tokenizer method additions vs. adding a new MobileLLM model. Both are model-related, but they fix unrelated gaps." }, { "left": "pull_request:41855", "right": "pull_request:43424", "accept": false, - "reason": "Tokenizer API work and executorch export testing are unrelated code paths." + "reason": "A tokenizer API patch and an executorch export test are unrelated changes." }, { "left": "pull_request:42668", "right": "pull_request:43823", "accept": false, - "reason": "Processor auto-loading robustness changes are not the same change as adding MobileLLM." + "reason": "Processor auto-loading robustness and adding MobileLLM are different changes with different code paths." }, { "left": "pull_request:43251", "right": "pull_request:43823", "accept": false, - "reason": "Loss-function kwargs fix vs. adding a new model; no shared concrete bug or change." + "reason": "`fixed_cross_entropy` kwargs forwarding is unrelated to adding a new model implementation." }, { "left": "pull_request:43251", "right": "pull_request:44827", "accept": false, - "reason": "Both touch `loss_utils.py` only indirectly via similarity; one fixes cross-entropy kwargs, the other is Mistral4 test/model cleanup." + "reason": "Both touch training/model code, but one is a loss utility fix and the other is Mistral4 test/model wiring; not the same bug or change." }, { "left": "pull_request:43424", "right": "pull_request:44827", "accept": false, - "reason": "Executorch export test coverage and Mistral4 test fixes are different issues." + "reason": "Executorch export dynamic-shape testing and Mistral4 test fixes are separate test concerns." }, { "left": "pull_request:43424", "right": "pull_request:43823", "accept": false, - "reason": "Dynamic-shape export testing is unrelated to MobileLLM model introduction." + "reason": "Dynamic-shape export testing does not match a new model addition." }, { "left": "pull_request:43823", "right": "pull_request:45655", "accept": false, - "reason": "New MobileLLM model code and config-resolution ordering fix are distinct changes." + "reason": "Model addition vs. `cls.config` resolution order fix; unrelated code paths." }, { "left": "pull_request:43823", "right": "pull_request:43995", "accept": false, - "reason": "Adding MobileLLM and refactoring Falcon output collection are different model-specific changes." + "reason": "New MobileLLM model and Falcon output-interface refactor are different model implementations with no shared bug." }, { "left": "pull_request:43995", "right": "pull_request:45213", "accept": false, - "reason": "Falcon refactor and the model-creation skill are unrelated; one changes model code, the other adds tooling/docs." + "reason": "Falcon refactor and model-creation skill tooling are not the same change." }, { "left": "pull_request:43995", "right": "pull_request:44827", "accept": false, - "reason": "Different model families and different problems; not mergeable as one PR." + "reason": "Falcon standardized-output refactor and Mistral4 test fixes are unrelated model internals." }, { "left": "pull_request:44827", "right": "pull_request:45170", "accept": false, - "reason": "Mistral4 test fixes and CLIP-family `layrnorm` renaming are unrelated bugs." + "reason": "Mistral4 test/model fixes and a CLIP-like weight-renaming conversion patch address different problems." }, { "left": "pull_request:44827", "right": "pull_request:46267", "accept": false, - "reason": "Mistral4 test work and chat-response parsing updates are different subsystems and changes." + "reason": "Mistral4 test fixes and chat response parsing updates are unrelated subsystems." }, { "left": "pull_request:44827", "right": "pull_request:46744", "accept": false, - "reason": "Model test fixes are not the same change as MPS documentation updates." + "reason": "Mistral4 test/model changes are unrelated to Apple Silicon safetensors docs." }, { "left": "pull_request:44827", "right": "pull_request:46048", "accept": false, - "reason": "Mistral4 test fixes and adding a TDT loss kernel are separate features/bugs." + "reason": "Mistral4 model/test work and adding a TDT loss kernel are distinct changes." }, { "left": "pull_request:44827", "right": "pull_request:45213", "accept": false, - "reason": "Mistral4 test fixes are unrelated to the new-model creation skill/tooling PR." + "reason": "A model/test fix and a repository skill/documentation PR are not the same underlying change." }, { "left": "pull_request:43823", "right": "pull_request:45213", "accept": false, - "reason": "New model implementation and the model-creation skill are different artifacts, not the same bug/change." + "reason": "Adding MobileLLM and adding a model-creation skill are different deliverables." }, { "left": "pull_request:44827", "right": "pull_request:45655", "accept": false, - "reason": "Mistral4 test changes and config-resolution ordering fix are distinct issues." + "reason": "Mistral4 test work and class config resolution ordering are unrelated fixes." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46048", + "accept": false, + "reason": "A Mistral4 test/model PR and a kernel-backed TDT loss addition are different features." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46267", + "accept": false, + "reason": "Mistral4 fixes do not duplicate chat parsing updates." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46744", + "accept": false, + "reason": "Mistral4 fixes and MPS loading docs are unrelated." }, { "left": "pull_request:43251", "right": "pull_request:43254", "accept": true, - "reason": "Both PRs fix the same `fixed_cross_entropy` gap in `loss_utils.py` by forwarding supported `nn.functional.cross_entropy` kwargs; they are mergeable into one change." + "reason": "Both PRs fix the same `fixed_cross_entropy` gap in `loss_utils.py` for issue 43240, and the changes are alternative variants of the same code-path fix." } ] }, "evaluator_result": { "accept": true, - "feedback": "The evaluation is grounded in the packet and stays conservative: the only clearly mergeable duplicate pair is PR 43251/43254, which share the same issue target and same `fixed_cross_entropy` kwargs fix. The other pair rejections are consistent with the diffs and titles and do not overclaim duplication." + "feedback": "The summary is grounded in the packet: it stays conservative, reflects that most pairwise links are only soft similarity, and correctly identifies the two `fixed_cross_entropy` PRs as the only mergeable duplicate-like pair with the same issue/code path. The other verdicts are appropriately rejected as unrelated without overstating evidence." }, "evaluator_used": true, "retried": false, @@ -20566,12 +20575,12 @@ "soft_pairs": [ "pull_request:40546|pull_request:42781" ], - "prepared_review_unit_hash": "b4b6ecb38f6468ca84b3367ed54fd9df3efb6015a397eec8a94385d3f550e5ac", + "prepared_review_unit_hash": "0d3a8637fcbc3aa5c54dbb4f54acc73a25a69d4b37f2271a98158ec0ffe8c9f1", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 7223, + "serialized_chars": 7224, "estimated_input_tokens": 1806, "estimated_eval_tokens": 3868 }, @@ -20579,7 +20588,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 7223, + "serialized_chars": 7224, "estimated_input_tokens": 1806, "estimated_eval_tokens": 3868 }, @@ -20588,18 +20597,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two PRs are related but not duplicates: #40546 is the initial VibeVoice integration, while #42781 is a separate VibeVoice Realtime follow-up with additional acoustic-tokenizer and generation changes.", - "confidence": 0.91, + "summary": "Both PRs are about VibeVoice, but they are not the same change: #40546 is the base VibeVoice integration, while #42781 adds a separate realtime extension with acoustic-tokenizer and generation/auto-mapping updates.", + "confidence": 0.93, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical duplicate PR; #40546 is the base model integration and #42781 is a distinct realtime extension built on top of it.", + "canonical_pr_reason": "#40546 is the foundational VibeVoice implementation and broader model integration; it reads as the main base change rather than the later realtime extension.", "best_issue_reason": null, - "best_pr_reason": "#40546 is the best representative because it is the foundational VibeVoice implementation with the broadest initial scope and substantial review activity.", + "best_pr_reason": "#40546 is the best representative of the cluster because it introduces the core VibeVoice package/docs and anchors the model integration, while #42781 is a follow-on realtime-specific expansion.", "soft_edge_verdicts": [ { "left": "pull_request:40546", "right": "pull_request:42781", "accept": false, - "reason": "Reject: they share the VibeVoice subsystem and some filenames, but #42781 adds realtime/acoustic-tokenizer functionality and different code-path changes rather than fixing the same concrete issue as #40546." + "reason": "Reject: they share the VibeVoice namespace and some files, but #40546 implements the base model, whereas #42781 adds realtime-specific acoustic tokenizer and generation changes. The concrete code paths differ enough that these would not plausibly be merged into one PR." } ] }, @@ -20629,12 +20638,12 @@ "soft_pairs": [ "pull_request:40755|pull_request:41224" ], - "prepared_review_unit_hash": "6c505f0b0181094ff973a69484864ab175ce089c791d50ad08b966dc40ad1073", + "prepared_review_unit_hash": "f016eb055b7c9dc9c597b02966551d99e05ce70c90b714a5c4799e7646eae9da", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4674, + "serialized_chars": 4675, "estimated_input_tokens": 1169, "estimated_eval_tokens": 2594 }, @@ -20642,7 +20651,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4674, + "serialized_chars": 4675, "estimated_input_tokens": 1169, "estimated_eval_tokens": 2594 }, @@ -20651,8 +20660,8 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two pull requests are unrelated: one adds TimesFM covariate forecasting, the other adds DINOv3 image classification support. They do not fix the same code path or represent a duplicate change.", - "confidence": 0.99, + "summary": "The two pull requests are unrelated feature additions in different model areas (TimesFM covariates vs. DINOv3 image classification), so they should not be clustered as duplicates.", + "confidence": 0.98, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, @@ -20662,7 +20671,7 @@ "left": "pull_request:40755", "right": "pull_request:41224", "accept": false, - "reason": "Different models and different feature areas (TimesFM covariates vs DINOv3 image classification); not the same underlying change and would not be merged into one PR." + "reason": "Different underlying changes and code paths: one adds TimesFM covariate forecasting support, the other adds DINOv3 image classification support. They do not appear mergeable into a single PR." } ] }, @@ -20755,7 +20764,7 @@ "soft_pairs": [ "pull_request:42975|pull_request:42976" ], - "prepared_review_unit_hash": "f484c8b211526436abc32c6b281df4540ec39e5b95be2fed3ff9c48e225c56cb", + "prepared_review_unit_hash": "c270010a5e1d275e35aeba58bc60513d91e0aa90a98d3cf60ae3562d8e76045e", "component_budget": { "node_count": 2, "item_count": 2, @@ -20777,18 +20786,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both items are maintenance PRs that update GitHub Actions versions, but they address different dependencies and different compatibility concerns. They overlap in workflow files, yet they are not the same underlying change and would not be merged as one PR.", - "confidence": 0.94, + "summary": "Two workflow-maintenance PRs in the same area, but they update different GitHub Actions families: one is a broad Node 24 compatibility sweep, the other is a docker/build-push-action version bump. Related, but not the same underlying change.", + "confidence": 0.92, "canonical_issue_reason": null, - "canonical_pr_reason": "No clear canonical PR: PR 42975 is focused on Node 24 compatibility for several core actions (checkout/cache/upload-artifact), while PR 42976 updates docker/build-push-action and a few other workflow actions to latest versions. They are related maintenance, not one duplicate change.", + "canonical_pr_reason": "pull_request:42975 is the broader, more representative workflow-actions upgrade: it updates multiple core actions across many workflows for Node 24 compatibility, making it the better cluster anchor.", "best_issue_reason": null, - "best_pr_reason": "Neither PR is a strong global representative of the cluster because the cluster is mixed. If forced, 42975 is slightly broader and more central to the shared workflow maintenance theme, but it does not subsume 42976's Docker action update.", + "best_pr_reason": "pull_request:42975 is the best single PR to represent the cluster because it covers the main cross-workflow action-compatibility update rather than a narrower docker-action bump.", "soft_edge_verdicts": [ { "left": "pull_request:42975", "right": "pull_request:42976", "accept": false, - "reason": "Different action families and different fix goals: one targets Node 24 compatibility for checkout/cache/upload-artifact, the other updates docker/build-push-action and related workflow actions. They share CI workflow files but not the same concrete bug or change." + "reason": "Reject: both are GitHub Actions upgrade PRs, but they target different concrete dependencies and code paths. 42975 updates checkout/cache/upload-artifact for Node 24 compatibility; 42976 updates docker/build-push-action. They could be merged in principle as a broad maintenance sweep, but they are not the same underlying fix/change." } ] }, @@ -20834,106 +20843,106 @@ "pull_request:45490|pull_request:46705", "pull_request:46562|pull_request:46658" ], - "prepared_review_unit_hash": "2d5397765c26f1f16e1f389d7c0188178af53fc3c50d6f46520b34dac168c24a", + "prepared_review_unit_hash": "490abf623a6ceed6a81a816f766bc3c943c5ce828368b77a4bcc49a3046c9a1a", "component_budget": { "node_count": 7, "item_count": 7, "soft_pair_count": 12, - "serialized_chars": 24616, - "estimated_input_tokens": 6154, - "estimated_eval_tokens": 12564 + "serialized_chars": 24621, + "estimated_input_tokens": 6156, + "estimated_eval_tokens": 12568 }, "budget": { "node_count": 7, "item_count": 7, "soft_pair_count": 12, - "serialized_chars": 24616, - "estimated_input_tokens": 6154, - "estimated_eval_tokens": 12564 + "serialized_chars": 24621, + "estimated_input_tokens": 6156, + "estimated_eval_tokens": 12568 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Most PRs in this cluster are unrelated model-addition or infrastructure changes (Molmo, Molmo2, CTSM, distributed config). The only clear duplicate/same-change pair is the Step 3.7 Flash work: PR #46562 and PR #46658. I\u2019d treat #46562 as the cleaner canonical representative.", - "confidence": 0.94, + "summary": "This cluster is mostly a set of unrelated model-addition PRs plus one closely related Step 3.7 Flash pair. The only strong duplicate-like match is the WIP Step 3 7 Flash PR versus the main Step 3.7 Flash model PR; the rest share broad auto-registration/docs patterns but are distinct features/models.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #46562 is the best canonical representative of the duplicate pair: it is the main user-facing Step 3.7 Flash model addition with docs/tests, while #46658 looks like a WIP/supporting implementation and conversion script PR for the same model.", + "canonical_pr_reason": "46562 is the best representative PR: it contains the actual Step 3.7 Flash model implementation, docs, and tests. The nearby 46658 PR is a separate WIP/mini-model conversion harness rather than a distinct user-facing change.", "best_issue_reason": null, - "best_pr_reason": "PR #46562 is the strongest overall representative of the cluster because it is the most complete and directly product-facing Step 3.7 Flash change. The other PRs are separate model additions or unrelated config work.", + "best_pr_reason": "46562 is the strongest global representative because it is the complete model-addition PR with the broadest user-facing scope; the other PRs are separate model additions or infrastructure changes.", "soft_edge_verdicts": [ { "left": "pull_request:42978", "right": "pull_request:43448", "accept": false, - "reason": "Both are model-addition PRs, but they add different models (ViT NEPA vs Molmo) and do not target the same code change." + "reason": "Both add different models (ViT NEPA vs Molmo). Shared docs/auto-registration files are incidental, not the same change." }, { "left": "pull_request:42978", "right": "pull_request:43451", "accept": false, - "reason": "Different model additions: ViT NEPA versus Molmo2, with distinct files and implementation paths." + "reason": "ViT NEPA and Molmo2 are unrelated model additions; overlap is only in shared registration/docs plumbing." }, { "left": "pull_request:42978", "right": "pull_request:45490", "accept": false, - "reason": "Different model families and model code; only broad Transformers registry/docs overlap." + "reason": "Different model families and use cases; no evidence they fix the same underlying code-path or feature." }, { "left": "pull_request:42978", "right": "pull_request:46705", "accept": false, - "reason": "Unrelated changes: a model addition versus distributed configuration support." + "reason": "Model-addition PR vs distributed-config infrastructure PR; unrelated scope." }, { "left": "pull_request:43448", "right": "pull_request:43451", "accept": false, - "reason": "Molmo and Molmo2 are separate model implementations and should not be merged as duplicates." + "reason": "Molmo and Molmo2 are related by name, but they are separate model implementations with different files and behavior." }, { "left": "pull_request:43448", "right": "pull_request:45490", "accept": false, - "reason": "Distinct model additions for different architectures; shared repo plumbing is not enough to call them the same change." + "reason": "Molmo and CTSM are different model additions; shared auto/docs changes do not make them duplicates." }, { "left": "pull_request:43448", "right": "pull_request:46705", "accept": false, - "reason": "Molmo model support and distributed config changes are unrelated." + "reason": "Molmo model addition is unrelated to adding distributed config support." }, { "left": "pull_request:43451", "right": "pull_request:45490", "accept": false, - "reason": "Molmo2 and CTSM are different models with different code paths and test coverage." + "reason": "Molmo2 and CTSM are distinct model ports; no shared concrete bug or feature implementation." }, { "left": "pull_request:43451", "right": "pull_request:46705", "accept": false, - "reason": "Molmo2 model work is not the same underlying change as distributed config support." + "reason": "Molmo2 model support and distributed config are different areas of the codebase." }, { "left": "pull_request:45490", "right": "pull_request:46562", "accept": false, - "reason": "Both are model additions, but they add different models (CTSM vs Step 3.7 Flash) and touch different implementation areas." + "reason": "CTSM and Step 3.7 Flash are different model additions with no shared code-path problem." }, { "left": "pull_request:45490", "right": "pull_request:46705", "accept": false, - "reason": "CTSM model support is unrelated to distributed configuration work." + "reason": "CTSM model support is unrelated to distributed configuration plumbing." }, { "left": "pull_request:46562", "right": "pull_request:46658", "accept": true, - "reason": "These appear to be the same Step 3.7 Flash change: both center on the same model name/code path, and #46658\u2019s WIP scripts/mini-model artifacts look like supporting conversion/validation work for the PR #46562 model implementation." + "reason": "Both are about Step 3.7 Flash and the same model code-path; the WIP mini-model/conversion harness plausibly belongs in the same PR as the model addition." } ] }, @@ -20986,136 +20995,130 @@ "pull_request:46579|pull_request:46719", "pull_request:43665|pull_request:46719" ], - "prepared_review_unit_hash": "a621087015b5ecc5c4ec8fa0d8874223c4d3c750a5676a0042bd74565e1dccd2", + "prepared_review_unit_hash": "432ad07f793cb9600e16cca94126cde3f78308fe5c5e1522b1d3cef600ea5473", "component_budget": { "node_count": 9, "item_count": 9, "soft_pair_count": 17, - "serialized_chars": 26388, - "estimated_input_tokens": 6597, - "estimated_eval_tokens": 13450 + "serialized_chars": 26393, + "estimated_input_tokens": 6599, + "estimated_eval_tokens": 13454 }, "budget": { "node_count": 9, "item_count": 9, "soft_pair_count": 17, - "serialized_chars": 26388, - "estimated_input_tokens": 6597, - "estimated_eval_tokens": 13450 + "serialized_chars": 26393, + "estimated_input_tokens": 6599, + "estimated_eval_tokens": 13454 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is a loose mix of unrelated PRs that all look like generic \"fix\" changes, but most are clearly separate: Whisper tests, backward-compat aliasing, doctest/typo cleanup, CI/test-only edits, generation watermark validation, FP8 dequantization, and MSA attention layout fixes. I do not see any true duplicates here.", - "confidence": 0.92, + "summary": "This cluster is not duplicate-coherent: it mixes unrelated PRs on Whisper tests, backward-compat shims, model test stubs, FP8 dequantization, doctest cleanup, generation validation, CI memory limits, typo fixes, and sparse attention.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #44815 is the strongest representative because it makes a concrete code-path fix in model loading / FP8 dequantization with supporting runtime and model changes, rather than mainly touching tests, docs, typos, or CI.", + "canonical_pr_reason": "PR 46579 is the broadest substantive code-fix PR here; the others are narrower, test/doc-only, infra-only, or clearly unrelated.", "best_issue_reason": null, - "best_pr_reason": "PR #44815 is the best global representative: it has the clearest substantive bug fix and the most explicit runtime impact across related code paths, while the others are either orthogonal or mostly non-functional cleanup.", + "best_pr_reason": "PR 46579 is the strongest representative because it makes concrete code-path changes across multiple core modules, while the other PRs are mostly isolated cleanup or unrelated fixes.", "soft_edge_verdicts": [ { "left": "pull_request:43291", "right": "pull_request:43665", "accept": false, - "reason": "Whisper tokenizer/test fixes are unrelated to the CLIP/ViT test edits; no shared underlying bug or change." + "reason": "Different areas and failure modes: Whisper tokenizer/test fix vs. CLIP/ViT test stubs; no shared underlying bug." }, { "left": "pull_request:43291", "right": "pull_request:44815", "accept": false, - "reason": "Whisper decode behavior and FP8 dequantization are different subsystems and different failure modes." + "reason": "Whisper decoding/tests and FP8 dequantization are unrelated code paths." }, { "left": "pull_request:43291", "right": "pull_request:46380", "accept": false, - "reason": "Whisper test cleanup and generation watermark validation are separate issues with no common code path." + "reason": "Whisper test fix vs. generation watermarking validation; different subsystems and bugs." }, { "left": "pull_request:43498", "right": "pull_request:46579", "accept": false, - "reason": "Backward-compat alias for tie_weights is a concrete API fix; typo cleanup across many files is not the same change." + "reason": "Backward-compat alias for tie_weights is unrelated to broad typo cleanup across many files." }, { "left": "pull_request:43498", "right": "pull_request:46719", "accept": false, - "reason": "Tie-weights compatibility and sparse attention/model edits are unrelated fixes." + "reason": "tie_weights compatibility shim and sparse-attention/layout fix are unrelated." }, { "left": "pull_request:43665", "right": "pull_request:46539", "accept": false, - "reason": "Bogus CLIP/ViT test edits do not match the CI infrastructure change for Phimoe integration tests." + "reason": "Bogus model test edits are not the same as CI memory-limit/test-environment changes." }, { "left": "pull_request:43665", "right": "pull_request:44815", "accept": false, - "reason": "Test-file sabotage/placeholder assertions are not the same underlying bug as the FP8 dequantization fix." + "reason": "Test-only edits for CLIP/ViT do not match FP8 loading/dequant logic." }, { "left": "pull_request:43665", "right": "pull_request:46380", "accept": false, - "reason": "These touch different areas and the CLIP/ViT test edits do not indicate the same bug as generation validation." + "reason": "Different targets: model test stubs versus generation config validation." }, { "left": "pull_request:43665", "right": "pull_request:46579", "accept": false, - "reason": "Test-only edits and code typo cleanup are not the same fix." + "reason": "Model test changes and multi-file typo cleanup are not the same underlying change." }, { "left": "pull_request:43665", "right": "pull_request:46719", "accept": false, - "reason": "CLIP/ViT test changes are unrelated to MSA attention / MiniMax model fixes." + "reason": "Test stubs for CLIP/ViT are unrelated to sparse attention / Minimax-DeepSeek code." }, { "left": "pull_request:44815", "right": "pull_request:46539", "accept": false, - "reason": "FP8 dequantization/model-loading changes are unrelated to CI memory patching and Phimoe integration tests." + "reason": "FP8 dequantization fixes and Phimoe CI harness changes address different problems." }, { "left": "pull_request:44815", "right": "pull_request:46380", "accept": false, - "reason": "Both are real bug fixes, but they address different code paths: FP8 dequantization versus generation watermark config validation." + "reason": "Different code paths: FP8 quantization/loading versus generation watermarking validation." }, { "left": "pull_request:44815", "right": "pull_request:46579", "accept": false, - "reason": "FP8 runtime fix and cross-repo typo cleanup are not the same change." + "reason": "Only broad subsystem overlap in core/model-loading; the concrete fixes are different (FP8 dequant vs typo cleanup)." }, { "left": "pull_request:45114", "right": "pull_request:46579", "accept": false, - "reason": "Doctest documentation updates are separate from code typo cleanup." + "reason": "Doctest documentation cleanup is not the same as code typo fixes across core modules." }, { "left": "pull_request:46380", "right": "pull_request:46539", "accept": false, - "reason": "Generation config validation and CI/Phimoe integration test infrastructure are unrelated." - }, - { - "left": "pull_request:46380", - "right": "pull_request:46579", - "accept": false, - "reason": "Generation watermark validation fix is distinct from broad typo-only edits." + "reason": "Generation config validation and CI memory patching are unrelated." }, { "left": "pull_request:46579", "right": "pull_request:46719", "accept": false, - "reason": "Typo cleanup across many files is not the same underlying fix as sparse attention layout changes and related model code." + "reason": "Typo cleanup across many modules is not the same concrete bug as the sparse-attention reshape/layout fix." } ] }, @@ -21148,12 +21151,12 @@ "pull_request:43488|pull_request:43532", "pull_request:43532|pull_request:46747" ], - "prepared_review_unit_hash": "ad1e1850615e1a7fddc9d0709f87a6a6140333967e5f9f49ee8f61c491047db5", + "prepared_review_unit_hash": "c4a36a22a4341ff21320797c3aa9e463a7109eb67013e425f9e696dea838cbf6", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 6349, + "serialized_chars": 6350, "estimated_input_tokens": 1588, "estimated_eval_tokens": 3432 }, @@ -21161,7 +21164,7 @@ "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 6349, + "serialized_chars": 6350, "estimated_input_tokens": 1588, "estimated_eval_tokens": 3432 }, @@ -21170,30 +21173,30 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Three open '[don't merge]' PRs that are only loosely similar in title/theme, but they change different code paths: packaging/setup, CI/workflow plumbing, and a security-gate workflow. No duplicate relationship is supported by the diffs, so all soft edges should be rejected.", - "confidence": 0.93, + "summary": "Three unrelated \"[don't merge]\" PRs: one tweaks setup/versioning and a test, one rewires several CI/comment/testing utilities, and one changes the PR security gate workflow. They share superficial repo/CI similarity but do not address the same concrete bug or change.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR: the PRs are unrelated validation/bot/workflow changes rather than the same underlying fix or feature.", + "canonical_pr_reason": "PR 43532 is the most representative cluster item because it has the broadest and most explicit CI-related diff plus the most discussion, but it is still not a duplicate of the others.", "best_issue_reason": null, - "best_pr_reason": "No global best PR stands out; each PR targets a different file set and purpose, so none is a clear representative duplicate target.", + "best_pr_reason": "PR 43532 is the best single representative of the cluster due to its larger surface area and clearer workflow/testing context, making it the most informative anchor for review triage.", "soft_edge_verdicts": [ { "left": "pull_request:43488", "right": "pull_request:46747", "accept": false, - "reason": "Different changes: setup/dependency/version-bump bot check vs a workflow trigger change. Same 'don't merge' label is not enough to indicate the same bug or change." + "reason": "Different changes: setup/version/dependency and VIT test tweaks versus a PR security gate workflow switch. Same broad repo, not the same underlying bug or fix." }, { "left": "pull_request:43488", "right": "pull_request:43532", "accept": false, - "reason": "They only share an incidental test file; one PR edits packaging/versioning, the other rewrites several workflow files and testing utilities. Not the same underlying change." + "reason": "The diffs target different areas: packaging/versioning and a test file versus CI workflows, testing utilities, and notifications. No shared concrete code-path problem." }, { "left": "pull_request:43532", "right": "pull_request:46747", "accept": false, - "reason": "Both touch workflow-related files, but they modify different workflows for different purposes. They do not appear mergeable as one fix and are not the same code-path problem." + "reason": "Both touch CI-related files, but one rewrites several comment/test workflows while the other only changes the security gate trigger. They are not the same change and would not plausibly merge into one PR." } ] }, @@ -21223,12 +21226,12 @@ "soft_pairs": [ "pull_request:43612|pull_request:43613" ], - "prepared_review_unit_hash": "1933f0a6e0296429c7676add8f08a5d60b24782bfa020713ac8b4efc7b85e13c", + "prepared_review_unit_hash": "b4e1634d6da8cd4030bc5e95f307f712ad9acf0761fc304187d1364439f43da1", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 6082, + "serialized_chars": 6083, "estimated_input_tokens": 1521, "estimated_eval_tokens": 3298 }, @@ -21236,7 +21239,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 6082, + "serialized_chars": 6083, "estimated_input_tokens": 1521, "estimated_eval_tokens": 3298 }, @@ -21245,18 +21248,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two separate feature PRs in the same multimodal/pipeline area: one adds Promptable Concept Segmentation, the other adds Promptable Visual Segmentation. They share boilerplate docs/registration changes, but they target different tasks, different pipeline classes, and different model/test integrations.", - "confidence": 0.95, + "summary": "These are related but distinct PRs: one adds a Promptable Concept Segmentation pipeline, the other adds a Promptable Visual Segmentation pipeline. They share docs/registry plumbing, but the actual pipeline code, model coverage, and tests target different tasks and code paths.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "No true duplicate here; if forced to pick one representative, PR 43613 is the broader pipeline addition with more model/test integrations and is slightly more central to the cluster.", + "canonical_pr_reason": "No clear duplicate canonical PR; the two PRs implement different segmentation tasks with different pipeline modules and model integrations.", "best_issue_reason": null, - "best_pr_reason": "PR 43613 is the better representative of the cluster because it spans more related codepaths and support files, while 43612 is a separate, narrower feature PR.", + "best_pr_reason": "PR #43613 is the broader/more general feature addition, covering multiple segmentation backends and corresponding tests/docs, so it is the better representative of the cluster.", "soft_edge_verdicts": [ { "left": "pull_request:43612", "right": "pull_request:43613", "accept": false, - "reason": "Different underlying changes: Promptable Concept Segmentation vs Promptable Visual Segmentation. The shared files are mostly framework registration/docs, but the pipeline classes, task docs, and model/test coverage are distinct, so they should not be merged as duplicates." + "reason": "Different underlying features: concept segmentation vs visual segmentation. Shared docs/registration changes are incidental; the pipeline implementations and model/test coverage are not the same concrete change and would not plausibly merge into one PR." } ] }, @@ -21286,7 +21289,7 @@ "soft_pairs": [ "pull_request:43651|pull_request:43663" ], - "prepared_review_unit_hash": "9f813ec9ce25a4399ca0cd282dc64bb3a82f540ea180e5d91725ac7728aaa23d", + "prepared_review_unit_hash": "8aa0d836135130b29b3942456726bbdb4e9e312a3a178889339493aed123aa64", "component_budget": { "node_count": 2, "item_count": 2, @@ -21308,18 +21311,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are two separate trainer.py extension-point PRs, both aimed at custom trainers but changing different code paths: gradient-accumulation loss scaling vs. signature/column filtering. They are not duplicates.", - "confidence": 0.95, + "summary": "Both pull requests modify `src/transformers/trainer.py`, but they address different trainer extension points: 43651 changes gradient-accumulation loss scaling, while 43663 refactors signature-column filtering. They are related by file and subsystem, not by the same underlying bug or code path.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 43651 is the more central representative of the cluster because it changes the training_step loss-scaling behavior and adds an explicit hook for custom trainers; PR 43663 is a distinct trainer helper for column filtering.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR 43651 is the best single representative of the cluster, but only as a broader trainer extensibility change; it is not the same underlying change as PR 43663.", + "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "pull_request:43651", "right": "pull_request:43663", "accept": false, - "reason": "Different underlying changes: 43651 adds _loss_is_scaled_for_ga and adjusts gradient-accumulation loss normalization, while 43663 adds _get_signature_columns to customize dataset column filtering. Same file and similar extensibility theme, but separate code paths and not mergeable as one PR." + "reason": "Different concrete changes in the same file: one fixes loss scaling during training, the other adds an overridable method for column filtering. They do not look mergeable into one PR or the same bug fix." } ] }, @@ -21349,7 +21352,7 @@ "soft_pairs": [ "pull_request:43656|pull_request:43842" ], - "prepared_review_unit_hash": "a347ba59ec84f257e2f91bb5e34e34dfa8566cbb236b86d075e34f0ff54bf9b5", + "prepared_review_unit_hash": "9a2682decb04f4e94521f25b18a23bb1c3324814599b004252d30acc3e92185c", "component_budget": { "node_count": 2, "item_count": 2, @@ -21371,18 +21374,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two distinct open PRs touch the same CLI file but fix different NameError/type-annotation problems in different code paths; they should not be clustered as duplicates.", - "confidence": 0.96, + "summary": "Two open PRs touch the same CLI file, but they fix different annotation/runtime typing problems in separate code paths. They are related by symptom class, not duplicates of the same bug.", + "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #43842 is the more explicit and narrowly scoped fix: it targets a TypeAdapter NameError when pydantic is absent and shows the exact typing/import adjustment needed.", + "canonical_pr_reason": "No clear canonical PR: 43656 and 43842 address different underlying NameError causes in the same module, so they should stay separate rather than be collapsed into one canonical change.", "best_issue_reason": null, - "best_pr_reason": "PR #43842 is the best representative because its title and diff clearly identify a concrete runtime failure and a minimal fix path, making it the stronger triage anchor.", + "best_pr_reason": "43842 is the tighter representative for the cluster because it directly targets the explicit TypeAdapter/pydantic NameError with a minimal scope and an issue link, whereas 43656 is a broader typing cleanup/future-annotations change.", "soft_edge_verdicts": [ { "left": "pull_request:43656", "right": "pull_request:43842", "accept": false, - "reason": "Reject: both edit src/transformers/cli/serve.py, but #43656 fixes a different typing/annotation NameError issue than #43842. The diffs address separate failure modes and are not mergeable as one PR." + "reason": "Reject: these are distinct fixes. 43656 changes module annotations/import typing behavior broadly; 43842 specifically guards TypeAdapter behind TYPE_CHECKING/quotes the annotation to avoid a pydantic runtime import/name error. Same file, different underlying bug." } ] }, @@ -21528,8 +21531,8 @@ "pull_request:44071|pull_request:44085", "pull_request:44059|pull_request:44086", "pull_request:44085|pull_request:44086", - "pull_request:44059|pull_request:44722", "pull_request:44098|pull_request:44101", + "pull_request:44059|pull_request:44722", "pull_request:44018|pull_request:44071", "pull_request:44017|pull_request:44161", "pull_request:44017|pull_request:44722", @@ -21560,8 +21563,8 @@ "pull_request:44018|pull_request:44129", "pull_request:43997|pull_request:44074", "pull_request:44010|pull_request:44018", - "pull_request:44071|pull_request:44129", "pull_request:43997|pull_request:44129", + "pull_request:44071|pull_request:44129", "pull_request:43996|pull_request:44161", "pull_request:43996|pull_request:44017", "pull_request:44010|pull_request:44044", @@ -21574,20 +21577,20 @@ "pull_request:44044|pull_request:44129", "pull_request:44044|pull_request:44154" ], - "prepared_review_unit_hash": "9fd6d278fc6f037c53461b052cca3ccab49a0a0742e91e9a2f63fd001ee48db9", + "prepared_review_unit_hash": "f4f0e9ae733ab8c27da6a967662a33d887a6f18c95035a5dc6ca2dde728f754f", "component_budget": { "node_count": 28, "item_count": 28, "soft_pair_count": 78, - "serialized_chars": 84898, - "estimated_input_tokens": 21225, - "estimated_eval_tokens": 42706 + "serialized_chars": 84896, + "estimated_input_tokens": 21224, + "estimated_eval_tokens": 42704 }, "budget": { "node_count": 27, "item_count": 27, "soft_pair_count": 72, - "serialized_chars": 75095, + "serialized_chars": 75093, "estimated_input_tokens": 18774, "estimated_eval_tokens": 37804 }, @@ -21596,81 +21599,315 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly standardized output-tracing refactors across many model families. The only true near-duplicate line is the GPT-Neo and GPT-J/CodeGen iterations, where later PRs clearly narrow or extend the same code-path rewrite.", - "confidence": 0.88, + "summary": "This cluster is mostly a batch of model-by-model refactors for standardized output tracing. Most pairs are only thematically related and should not be merged as duplicates; the clear duplicate groups are the GPT-J/CodeGen PRs and the GPT-Neo PRs.", + "confidence": 0.89, "canonical_issue_reason": null, - "canonical_pr_reason": "44722 is the broadest and latest GPT-J/CodeGen output-tracing refactor, covering both gptj and codegen and subsuming the earlier GPT-J-only iterations.", + "canonical_pr_reason": "PR 44066 is the best canonical PR because it explicitly targets the tracking issue, makes a concrete GPT-J/CodeGen output-tracing refactor, and is the most central duplicate candidate among the GPT-family changes.", "best_issue_reason": null, - "best_pr_reason": "44722 is the strongest representative PR because it is the most complete GPT-J/CodeGen iteration of the same standardized output-tracing change.", + "best_pr_reason": "PR 44066 is the strongest overall representative of the cluster: explicit issue linkage, coherent code-path changes, and broad enough scope to stand in for the shared GPT-J/CodeGen tracing work.", "soft_edge_verdicts": [ { - "left": "pull_request:44066", + "left": "pull_request:43996", "right": "pull_request:44085", - "accept": true, - "reason": "Same GPT-J/modeling_gptj.py output-tracing refactor; 44085 is a narrower iteration of the same change." + "accept": false, + "reason": "Different model families and code paths (CVT/FNet vs GPT-J); same broad refactor theme is not enough." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44044", + "accept": false, + "reason": "Different models and implementations (CVT/FNet vs DeBERTa); not the same underlying change." }, { "left": "pull_request:44066", - "right": "pull_request:44722", + "right": "pull_request:44085", "accept": true, - "reason": "Same GPT-J/CodeGen output-tracing rewrite; 44722 is the broader/latest iteration of the same code-path change." + "reason": "Both target GPT-J output tracing in the same file/code path; 44085 looks like a narrower variant of the same refactor." }, { - "left": "pull_request:44085", + "left": "pull_request:44007", + "right": "pull_request:44072", + "accept": false, + "reason": "RegNet and EfficientNet are separate model implementations; only the refactor style overlaps." + }, + { + "left": "pull_request:44072", "right": "pull_request:44722", - "accept": true, - "reason": "Same GPT-J/modeling_gptj.py refactor; 44722 extends the same change to CodeGen as well." + "accept": false, + "reason": "EfficientNet vs GPT-J/CodeGen are unrelated concrete code paths." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44072", + "accept": false, + "reason": "GPT-J/CodeGen output tracing and EfficientNet output tracing are different model-specific changes." + }, + { + "left": "pull_request:44013", + "right": "pull_request:44044", + "accept": false, + "reason": "MobileNetV2 and DeBERTa touch unrelated modules; not the same bug/change." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-J/CodeGen vs MGP-STR are different model refactors." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44071", + "accept": false, + "reason": "Different model families and different forward paths; same standardization effort only." }, { "left": "pull_request:44018", "right": "pull_request:44068", "accept": true, - "reason": "Same GPT-Neo output-tracing rewrite in the same file; one PR is a narrower/updated version of the other." + "reason": "Both are GPT-Neo tracing refactors on the same model file/test surface, so they appear to be the same change." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44068", + "accept": false, + "reason": "GPT-J/CodeGen vs GPT-Neo are distinct model changes." + }, + { + "left": "pull_request:44019", + "right": "pull_request:44722", + "accept": false, + "reason": "ResNet and GPT-J/CodeGen are unrelated code paths." + }, + { + "left": "pull_request:44086", + "right": "pull_request:44722", + "accept": false, + "reason": "MGP-STR and GPT-J/CodeGen are separate model refactors." + }, + { + "left": "pull_request:44019", + "right": "pull_request:44085", + "accept": false, + "reason": "ResNet and GPT-J target different implementations and behaviors." }, { "left": "pull_request:44018", - "right": "pull_request:44066", + "right": "pull_request:44019", "accept": false, - "reason": "GPT-Neo vs GPT-J/CodeGen are different model implementations; only the generic refactor theme matches." + "reason": "GPT-Neo and ResNet are unrelated models; not mergeable as one PR." }, { - "left": "pull_request:44066", - "right": "pull_request:44068", + "left": "pull_request:44019", + "right": "pull_request:44071", "accept": false, - "reason": "GPT-J/CodeGen and GPT-Neo are different code paths, so these are related but not duplicates." + "reason": "ResNet and MPT do not share the same concrete bug/change." }, { "left": "pull_request:44068", "right": "pull_request:44722", "accept": false, - "reason": "GPT-Neo and GPT-J/CodeGen are distinct model families with different forward paths." + "reason": "GPT-Neo vs GPT-J/CodeGen are different model paths." }, { - "left": "pull_request:44066", + "left": "pull_request:44068", + "right": "pull_request:44116", + "accept": false, + "reason": "GPT-Neo and Flaubert/XLM are unrelated model implementations." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44116", + "accept": false, + "reason": "Different models and different output-tracing implementations." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44722", + "accept": false, + "reason": "MPT and GPT-J/CodeGen are unrelated concrete changes." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44085", + "accept": false, + "reason": "GPT-Neo and GPT-J are separate model-specific refactors." + }, + { + "left": "pull_request:44018", "right": "pull_request:44086", "accept": false, - "reason": "GPT-J/CodeGen and MGP-STR are unrelated model code paths." + "reason": "GPT-Neo and MGP-STR are unrelated modules." }, { - "left": "pull_request:44066", - "right": "pull_request:44071", + "left": "pull_request:44068", + "right": "pull_request:44086", "accept": false, - "reason": "GPT-J/CodeGen and MPT are separate model refactors, not one mergeable change." + "reason": "GPT-Neo vs MGP-STR are different underlying changes." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44086", + "accept": false, + "reason": "MPT and MGP-STR are different model code paths." + }, + { + "left": "pull_request:44059", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-2 and MGP-STR are unrelated implementations." }, { "left": "pull_request:44085", "right": "pull_request:44086", "accept": false, - "reason": "GPT-J and MGP-STR are unrelated implementations." + "reason": "GPT-J and MGP-STR are unrelated model changes." + }, + { + "left": "pull_request:44098", + "right": "pull_request:44101", + "accept": false, + "reason": "ViLT and Flaubert/XLM are different model families; same refactor theme only." + }, + { + "left": "pull_request:44059", + "right": "pull_request:44071", + "accept": false, + "reason": "GPT-2 and MPT are different model-specific output changes." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44101", + "accept": false, + "reason": "MPNet and Flaubert/XLM do not share the same concrete bug/change." + }, + { + "left": "pull_request:44074", + "right": "pull_request:44101", + "accept": false, + "reason": "TextNet and Flaubert/XLM are unrelated model implementations." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44017", + "accept": false, + "reason": "RegNet and SegFormer are separate model code paths." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44056", + "accept": false, + "reason": "RegNet and MPNet are unrelated changes." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44076", + "accept": false, + "reason": "MPT and ImageGPT are different models; not the same fix." + }, + { + "left": "pull_request:44074", + "right": "pull_request:44161", + "accept": false, + "reason": "TextNet and LongT5 are unrelated concrete changes." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44018", + "accept": false, + "reason": "RegNet and GPT-Neo are separate model refactors." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44161", + "accept": false, + "reason": "MPNet and LongT5 are different model implementations." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44161", + "accept": false, + "reason": "RegNet and LongT5 are unrelated code paths." + }, + { + "left": "pull_request:44074", + "right": "pull_request:44161", + "accept": false, + "reason": "TextNet and LongT5 are separate model-specific changes." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44017", + "accept": false, + "reason": "CVT/FNet and SegFormer are different models; only the tracing pattern overlaps." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44074", + "accept": false, + "reason": "CVT/FNet and TextNet are unrelated concrete changes." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44071", + "accept": false, + "reason": "SqueezeBERT and MPT are unrelated model implementations." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44073", + "accept": false, + "reason": "SqueezeBERT and VisualBert are distinct model code paths." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44722", + "accept": true, + "reason": "Both are GPT-J/CodeGen standardized output-tracing refactors on the same underlying code paths, so they are plausible duplicates." + }, + { + "left": "pull_request:44085", + "right": "pull_request:44722", + "accept": true, + "reason": "Both touch GPT-J output tracing in the same file and are close enough to be one combined PR; 44722 just adds the CodeGen companion change." + }, + { + "left": "pull_request:44085", + "right": "pull_request:44129", + "accept": false, + "reason": "GPT-J and SpeechT5 are unrelated models; not the same underlying change." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44161", + "accept": false, + "reason": "CVT/FNet and LongT5 are different model-specific refactors." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44007", + "accept": false, + "reason": "CVT/FNet and RegNet are separate model code paths." + }, + { + "left": "pull_request:44044", + "right": "pull_request:44129", + "accept": false, + "reason": "DeBERTa and SpeechT5 are unrelated model implementations." + }, + { + "left": "pull_request:44044", + "right": "pull_request:44154", + "accept": false, + "reason": "DeBERTa and VITS are different concrete changes despite similar refactor wording." } ] }, "evaluator_result": { - "accept": false, - "feedback": "The summary is broadly grounded, but several soft-edge verdicts are too conservative relative to the packet. In particular, the GPT-J/CodeGen line has strong same-file, same-target overlap: 44066\u219444722 and 44085\u219444722 are marked deterministic_accept=true in the packet, yet were rejected as merely related/different-model work. Those look like mergeable iterations of the same output-tracing refactor, not separate bugfixes. Because the verdicts miss clear duplicates/near-duplicates, I can\u2019t accept this evaluation as written." + "accept": true, + "feedback": "The result is grounded in the packet: the only clearly mergeable duplicate groups are GPT-J/CodeGen (44066, 44085, 44722) and GPT-Neo (44018, 44068), and the other false verdicts stay conservative by requiring the same concrete model/code path rather than just the same refactor theme. The canonical/best-PR choice of 44066 is also supported by its explicit issue link and broader GPT-J/CodeGen scope." }, "evaluator_used": true, - "retried": true, + "retried": false, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -21705,14 +21942,14 @@ "pull_request:43997|pull_request:44010", "pull_request:44010|pull_request:44129" ], - "prepared_review_unit_hash": "8e2aff89041ff0e8a2166bb9f53d1f1ac55973f266fe6d99f51a69631a1ba770", + "prepared_review_unit_hash": "d9516d39a30b89f919c27cc64cfbbe51d7aa234509ac72d0dfb5b9844d22e884", "component_budget": { "node_count": 28, "item_count": 28, "soft_pair_count": 78, - "serialized_chars": 84898, - "estimated_input_tokens": 21225, - "estimated_eval_tokens": 42706 + "serialized_chars": 84896, + "estimated_input_tokens": 21224, + "estimated_eval_tokens": 42704 }, "budget": { "node_count": 8, @@ -21727,48 +21964,48 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These are parallel, model-specific refactors to standardized output tracing/capture_outputs across different architectures, not duplicate bug fixes or mergeable single-change PRs.", - "confidence": 0.95, + "summary": "These PRs are all architecture-specific refactors toward standardized output tracing/capture, but they affect different model code paths and are not duplicate fixes/changes. They should remain separate.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": "No single PR is a true canonical duplicate: each PR touches a different model and implements its own output-tracing refactor, so they should remain separate.", + "canonical_pr_reason": "PR #44073 is a clean representative of the cluster: it applies the standardized capture_outputs/can_return_tuple pattern in a single model with a clear, focused refactor and explicit output-tracing setup.", "best_issue_reason": null, - "best_pr_reason": "44129 is the best representative only in the sense that it is the most concrete and scoped, with an explicit issue target; however, it is still not a duplicate of the others.", + "best_pr_reason": "PR #44073 is the best representative PR for this cluster because it is well-scoped, clearly titled, and directly demonstrates the standardized output-tracing pattern without being broader than the others.", "soft_edge_verdicts": [ { "left": "pull_request:44024", "right": "pull_request:44076", "accept": false, - "reason": "Same broad output-tracing theme, but different models (FocalNet vs ImageGPT) and different code paths; not one mergeable change." + "reason": "Both are output-tracing refactors, but FocalNet and ImageGPT are different models with different forward paths and output wiring; not the same underlying change." }, { "left": "pull_request:44073", "right": "pull_request:44074", "accept": false, - "reason": "Both refactor output collection, but VisualBert and TextNet are separate implementations with different forward signatures and tests." + "reason": "VisualBert vs TextNet are separate model implementations; these are similar refactors, not one concrete shared bug or patch." }, { "left": "pull_request:44056", "right": "pull_request:44073", "accept": false, - "reason": "Different model families and different refactor details; only the general output-tracing cleanup overlaps." + "reason": "MPNet and VisualBert each add model-specific _can_record_outputs/capture_outputs plumbing in different files and code paths; not duplicates." }, { "left": "pull_request:44056", "right": "pull_request:44074", "accept": false, - "reason": "MPNet and TextNet are unrelated model-specific changes, not the same underlying bug or change." + "reason": "MPNet and TextNet are distinct architectures with separate refactors; similarity is only at the pattern level." }, { "left": "pull_request:43997", "right": "pull_request:44010", "accept": false, - "reason": "RegNet and SqueezeBert are distinct model implementations; shared standardized-output wording is not enough to merge them." + "reason": "RegNet and SqueezeBert touch unrelated model implementations; both are standardization work, but not the same bug or change." }, { "left": "pull_request:44010", "right": "pull_request:44129", "accept": false, - "reason": "Both are output-tracing refactors, but they target different models and different concrete code paths, so they are not duplicates." + "reason": "SqueezeBert and SpeechT5 are different model stacks and the PRs address model-specific output tracing changes, not one mergeable fix." } ] }, @@ -21848,274 +22085,274 @@ "pull_request:44028|pull_request:44030", "pull_request:44029|pull_request:44030" ], - "prepared_review_unit_hash": "334ff9718f2ea879309c14d54fe95b17bdb8e07f803e5662d270756cb8affc73", + "prepared_review_unit_hash": "774f3bde0e2832123e9fd7d70575a3bfe3f717b1d5139aef6afb22892cdc2cfc", "component_budget": { "node_count": 13, "item_count": 13, "soft_pair_count": 40, - "serialized_chars": 40181, - "estimated_input_tokens": 10046, - "estimated_eval_tokens": 20348 + "serialized_chars": 40177, + "estimated_input_tokens": 10045, + "estimated_eval_tokens": 20346 }, "budget": { "node_count": 13, "item_count": 13, "soft_pair_count": 40, - "serialized_chars": 40181, - "estimated_input_tokens": 10046, - "estimated_eval_tokens": 20348 + "serialized_chars": 40177, + "estimated_input_tokens": 10045, + "estimated_eval_tokens": 20346 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is a set of similar-but-independent PRs that refactor output tracing across different model implementations. They share a common helper/decorator theme, but the affected code paths are model-specific, so they should not be treated as duplicates.", + "summary": "These are all related output-tracing refactors, but each PR targets a different model family and code path, so they are not duplicates.", "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 44025 is a good anchor because it is a clear, self-contained output-tracing refactor with the same overall pattern as the rest of the cluster and touches a related model pair in one change.", + "canonical_pr_reason": "No canonical PR: the cluster is a set of similar-but-separate model-specific refactors, not one underlying fix/change.", "best_issue_reason": null, - "best_pr_reason": "PR 44025 is the best representative of the cluster theme: it cleanly shows the tracing refactor pattern, is scoped to one model family, and is more substantial than the single-file variants.", + "best_pr_reason": "PR 44025 is the best single representative because it shows the refactor pattern clearly across two closely related models, but it still isn't a duplicate of the others.", "soft_edge_verdicts": [ { "left": "pull_request:43998", "right": "pull_request:44002", "accept": false, - "reason": "Different model families (timm_backbone vs upernet); same refactor style, but not the same underlying code path or fix." + "reason": "Different model families and code paths; same refactor theme only." }, { "left": "pull_request:43998", "right": "pull_request:44000", "accept": false, - "reason": "Unrelated model implementations and forward paths; only the tracing refactor pattern is shared." + "reason": "Unrelated model implementations; not the same underlying bug or change." }, { "left": "pull_request:43998", "right": "pull_request:44001", "accept": false, - "reason": "Different models (timm_backbone vs univnet); not a single concrete bug/change." + "reason": "Different architectures; only shares the output-tracing refactor pattern." }, { "left": "pull_request:43998", "right": "pull_request:44003", "accept": false, - "reason": "Different subsystems and model code paths; too broad to be a duplicate PR." + "reason": "Separate model code paths; too broad to be one duplicate PR." }, { "left": "pull_request:43998", "right": "pull_request:44029", "accept": false, - "reason": "Different architectures (backbone vs RWKV); same output-tracing theme but not the same fix." + "reason": "Different model implementation; not a mergeable duplicate." }, { "left": "pull_request:43999", "right": "pull_request:44001", "accept": false, - "reason": "Mobilenet V1 and UnivNet are unrelated implementations; these are separate refactors." + "reason": "Different model families; same refactoring style is not enough." }, { "left": "pull_request:43999", "right": "pull_request:44002", "accept": false, - "reason": "Different model families and different forward logic; no shared concrete bug." + "reason": "Different code paths and model goals; not the same fix." }, { "left": "pull_request:43999", "right": "pull_request:44025", "accept": false, - "reason": "Both refactor tracing, but they apply to different model families and code paths." + "reason": "Different model families; no shared concrete bug/change." }, { "left": "pull_request:43999", "right": "pull_request:44026", "accept": false, - "reason": "MobileNet V1 output handling is unrelated to vision-encoder-decoder tracing changes." + "reason": "Separate wrappers/models; not the same underlying change." }, { "left": "pull_request:43999", "right": "pull_request:44027", "accept": false, - "reason": "Different model wrappers and forward paths; not mergeable as one PR." + "reason": "Different encoder-decoder implementation; only superficially similar." }, { "left": "pull_request:44000", "right": "pull_request:44001", "accept": false, - "reason": "Vision-text dual encoder and UnivNet are unrelated; the shared pattern is only superficial." + "reason": "Distinct model implementations; both are refactors, not duplicates." }, { "left": "pull_request:44000", "right": "pull_request:44002", "accept": false, - "reason": "Different model code paths; same decorator/refactor style is not enough for duplication." + "reason": "Different models and files; same modernization pattern only." }, { "left": "pull_request:44000", "right": "pull_request:44003", "accept": false, - "reason": "Vision-text dual encoder vs Mamba/Falcon-Mamba are distinct changes in unrelated modules." + "reason": "Different architecture and code path; not one concrete change." }, { "left": "pull_request:44000", "right": "pull_request:44025", "accept": false, - "reason": "Different model families and different output-shaping behavior; not the same underlying change." + "reason": "Different model families; no shared bug to collapse." }, { "left": "pull_request:44001", "right": "pull_request:44002", "accept": false, - "reason": "UnivNet and UperNet are unrelated models; no shared concrete bug or code path." + "reason": "Separate model-specific refactors; not the same PR-worthy fix." }, { "left": "pull_request:44001", "right": "pull_request:44003", "accept": false, - "reason": "Different architectures and runtime paths; too broad to merge as one PR." + "reason": "Different sequence-model implementations; too broad for duplication." }, { "left": "pull_request:44001", "right": "pull_request:44025", "accept": false, - "reason": "Same refactor motif, but different model families and distinct implementation details." + "reason": "Different models; only the tracing refactor theme matches." }, { "left": "pull_request:44002", "right": "pull_request:44029", "accept": false, - "reason": "UperNet segmentation tracing and RWKV output handling are unrelated changes." + "reason": "Unrelated models and code paths; not the same underlying issue." }, { "left": "pull_request:44002", "right": "pull_request:44003", "accept": false, - "reason": "Different subsystems and model behavior; not a duplicate fix." + "reason": "Different architectures; same refactor label is insufficient." }, { "left": "pull_request:44003", "right": "pull_request:44028", "accept": false, - "reason": "Falcon-Mamba/Mamba and SuperPoint are unrelated code paths despite the shared tracing refactor theme." + "reason": "Falcon/Mamba refactor vs SuperPoint refactor are unrelated changes." }, { "left": "pull_request:44003", "right": "pull_request:44004", "accept": false, - "reason": "Different model code paths (Mamba/Falcon-Mamba vs CodeGen) and different tracing behavior; not one bug." + "reason": "Different model code and behavior; not a duplicate fix." }, { "left": "pull_request:43998", "right": "pull_request:44004", "accept": false, - "reason": "Timm backbone and CodeGen are unrelated implementations; the shared helper usage is not enough." + "reason": "Different model families and implementation details; only broad similarity." }, { "left": "pull_request:44001", "right": "pull_request:44004", "accept": false, - "reason": "UnivNet and CodeGen fix different model-specific output handling." + "reason": "Unrelated model code paths; not mergeable as one PR." }, { "left": "pull_request:44002", "right": "pull_request:44004", "accept": false, - "reason": "UperNet and CodeGen are distinct model changes with no common concrete defect." + "reason": "Different models; same general cleanup pattern only." }, { "left": "pull_request:44000", "right": "pull_request:44004", "accept": false, - "reason": "Different model families and different forward/output semantics; not the same PR." + "reason": "Distinct model implementations; no shared concrete bug." }, { "left": "pull_request:44025", "right": "pull_request:44026", "accept": false, - "reason": "Both are tracing refactors, but they target different encoder/decoder model families and different code paths." + "reason": "Depth Anything vs Vision-Encoder-Decoder are separate model families." }, { "left": "pull_request:44025", "right": "pull_request:44027", "accept": false, - "reason": "Depth-Anything and Speech-Encoder-Decoder are unrelated models; only the refactor pattern is shared." + "reason": "Different model wrappers and code paths; not the same fix." }, { "left": "pull_request:44025", "right": "pull_request:44028", "accept": false, - "reason": "Different model implementations and output-capturing details; not a single underlying change." + "reason": "Separate vision model implementations; same pattern only." }, { "left": "pull_request:44025", "right": "pull_request:44029", "accept": false, - "reason": "Depth-Anything and RWKV are unrelated architectures; these are separate refactors." + "reason": "Different architecture and module behavior; not a duplicate." }, { "left": "pull_request:44025", "right": "pull_request:44030", "accept": false, - "reason": "Different model families and forward signatures; same theme, not same fix." + "reason": "Different model code paths; same refactor theme is too broad." }, { "left": "pull_request:44026", "right": "pull_request:44029", "accept": false, - "reason": "Vision-encoder-decoder and RWKV differ in architecture and code path; no duplicate bug." + "reason": "Different models and runtime behavior; not the same change." }, { "left": "pull_request:44026", "right": "pull_request:44027", "accept": false, - "reason": "Vision encoder-decoder and speech encoder-decoder are parallel wrappers, but still distinct PRs with separate code paths." + "reason": "Encoder-decoder wrappers for different modalities; not duplicates." }, { "left": "pull_request:44026", "right": "pull_request:44028", "accept": false, - "reason": "Different models and different output-tracing implementations." + "reason": "Different model families; no shared concrete bug." }, { "left": "pull_request:44026", "right": "pull_request:44030", "accept": false, - "reason": "Vision-encoder-decoder and DPR are unrelated changes." + "reason": "Different model implementations; only broad similarity." }, { "left": "pull_request:44027", "right": "pull_request:44028", "accept": false, - "reason": "Speech encoder-decoder and SuperPoint are unrelated model code paths." + "reason": "Speech encoder-decoder vs SuperPoint are unrelated code paths." }, { "left": "pull_request:44027", "right": "pull_request:44029", "accept": false, - "reason": "Different model families and output semantics; not mergeable as one fix." + "reason": "Different model families; not the same underlying issue." }, { "left": "pull_request:44027", "right": "pull_request:44030", "accept": false, - "reason": "Speech encoder-decoder and DPR are different model-specific refactors." + "reason": "Separate implementations with different behavior; not a duplicate." }, { "left": "pull_request:44028", "right": "pull_request:44029", "accept": false, - "reason": "SuperPoint and RWKV are unrelated architectures; the shared helper pattern is superficial." + "reason": "Different model types; same tracing pattern only." }, { "left": "pull_request:44028", "right": "pull_request:44030", "accept": false, - "reason": "SuperPoint and DPR touch unrelated code paths and do not fix the same bug." + "reason": "Unrelated model code paths; too broad to accept." }, { "left": "pull_request:44029", "right": "pull_request:44030", "accept": false, - "reason": "RWKV and DPR are different models with distinct output-handling logic; not a duplicate PR." + "reason": "Different models and outputs; not the same fix." } ] }, @@ -22145,12 +22382,12 @@ "soft_pairs": [ "pull_request:44662|pull_request:45401" ], - "prepared_review_unit_hash": "059834f36c02ef89fb78c280ccf97908cef463d74984f4125e338e5638e488f8", + "prepared_review_unit_hash": "f99ab1f0a5e666ba59b13557a9adb7941322d07507e545a1a9c8981a088481f0", "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 }, @@ -22158,7 +22395,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 6552, + "serialized_chars": 6551, "estimated_input_tokens": 1638, "estimated_eval_tokens": 3532 }, @@ -22167,18 +22404,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two PRs are unrelated model additions that share only generic Transformers boilerplate and auto-registration files. They do not appear to fix the same code-path problem or be mergeable into one PR.", - "confidence": 0.95, + "summary": "These are two different model-integration PRs: one adds PenguinVL, the other adds Voxtral TTS. They overlap in shared Transformers plumbing, but they do not appear to fix the same underlying code path or change set.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45401 is the cleaner representative: it is tightly scoped to Voxtral TTS, has an explicit issue target, and includes a focused implementation/test set. PR 44662 is a separate PenguinVL model addition with different scope and no issue target.", + "canonical_pr_reason": "No single canonical PR for a duplicate cluster: the two PRs target different models and different integration work.", "best_issue_reason": null, - "best_pr_reason": "PR 45401 is the best single representative of the cluster because it is more self-contained and better linked to a concrete tracking issue; PR 44662 is a distinct model implementation rather than a duplicate change.", + "best_pr_reason": "PR #45401 is the better representative only because it is more narrowly scoped and has an explicit issue target, but it is still unrelated to #44662.", "soft_edge_verdicts": [ { "left": "pull_request:44662", "right": "pull_request:45401", "accept": false, - "reason": "Different models and different changes: PenguinVL vs Voxtral TTS. Shared files are only broad registry/boilerplate locations, not the same underlying bug or feature." + "reason": "Different features for different models (PenguinVL vs Voxtral TTS); shared auto/registry files are framework boilerplate, not evidence of the same bug or mergeable change." } ] }, @@ -22208,12 +22445,12 @@ "soft_pairs": [ "pull_request:44979|pull_request:45363" ], - "prepared_review_unit_hash": "5eae6c5fa14ecef93d6d03f24e2971950d2108dd5b6c219c2ab6e1438c99ba32", + "prepared_review_unit_hash": "ae042fdf44ec47ea55765f264ebfe72fad97c1eea3204eb4b1b26b834e4bf00e", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4407, + "serialized_chars": 4408, "estimated_input_tokens": 1102, "estimated_eval_tokens": 2460 }, @@ -22221,7 +22458,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4407, + "serialized_chars": 4408, "estimated_input_tokens": 1102, "estimated_eval_tokens": 2460 }, @@ -22230,18 +22467,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These PRs are related to model/kernel fusion, but they target different layers and are not the same change: #44979 introduces a new Module Fusion API, while #45363 wires up n-to-1 kernel fusion through KernelConfig and integration plumbing. They should not be merged as duplicates.", - "confidence": 0.89, + "summary": "These PRs are related thematically around fusion, but they implement different layers and use cases: one introduces a generic Module Fusion API, while the other adds a specific n-to-1 kernel fusion path via KernelConfig and integration changes. They are not duplicates.", + "confidence": 0.93, "canonical_issue_reason": null, - "canonical_pr_reason": "#44979 is the better anchor for the cluster because it introduces the core Module Fusion API and tests; the other PR is a separate, later kernel-config integration on top of that broader foundation.", + "canonical_pr_reason": "PR #44979 is the cleaner foundational change: it adds a new module fusion API and tests without the extra integration surface of the draft kernel-config PR.", "best_issue_reason": null, - "best_pr_reason": "#44979 is the most representative and substantial PR in the cluster: it adds the central new abstraction rather than a usage example / integration path. #45363 is draft and narrower in scope.", + "best_pr_reason": "PR #44979 is the best representative artifact in this cluster because it is more self-contained, non-draft, and establishes the core abstraction.", "soft_edge_verdicts": [ { "left": "pull_request:44979", "right": "pull_request:45363", "accept": false, - "reason": "Different concrete changes: #44979 adds a standalone module fusion API, while #45363 adds kernel-config-based n-to-1 fusion plumbing and an example. They overlap conceptually but do not appear to fix the same underlying bug or implement the same feature branch." + "reason": "Different underlying changes: #44979 defines a generic module fusion API, while #45363 implements specific n-to-1 kernel fusion through KernelConfig and model-loading integrations. They would not plausibly be merged into one PR." } ] }, @@ -22271,12 +22508,12 @@ "soft_pairs": [ "pull_request:45086|pull_request:45317" ], - "prepared_review_unit_hash": "66f2cefb078f67cc2285f63b7cc5b8d2c48f5f1cec30d73ac4ed9fd54afae85d", + "prepared_review_unit_hash": "bc4452be8c2e014f0625fabf3ac559c57d3eda8bb09d993d9b14454f37e138fd", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4413, + "serialized_chars": 4414, "estimated_input_tokens": 1104, "estimated_eval_tokens": 2464 }, @@ -22284,7 +22521,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4413, + "serialized_chars": 4414, "estimated_input_tokens": 1104, "estimated_eval_tokens": 2464 }, @@ -22293,18 +22530,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both pull requests implement the same fix for the Mistral regex AttributeError in `_patch_mistral_regex` by switching to `tokenizer.pre_tokenizer`; PR 45317 is the stronger canonical choice because it adds tests and has the clearer, more complete patch.", - "confidence": 0.98, + "summary": "Both PRs fix the same AttributeError in `_patch_mistral_regex` by switching from `backend_tokenizer.pre_tokenizer` to `tokenizer.pre_tokenizer`; `45317` is the more complete version because it adds test coverage.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45317 is the best canonical PR: it targets the same underlying bug as PR 45086, includes the same code-path fix, and also adds test coverage with a clearer title/description.", + "canonical_pr_reason": "PR 45317 is the stronger canonical candidate: it addresses the same concrete bug as 45086 and also includes a test change, making it the more complete fix.", "best_issue_reason": null, - "best_pr_reason": "PR 45317 is the best overall PR candidate because it fixes the same concrete AttributeError, is more complete than 45086, and includes tests that validate the behavior.", + "best_pr_reason": "PR 45317 is the best PR candidate because it fixes the same code-path bug and adds validation in tests, whereas 45086 is the minimal patch only.", "soft_edge_verdicts": [ { "left": "pull_request:45086", "right": "pull_request:45317", "accept": true, - "reason": "Same underlying bug and same code change in the same file: both replace `backend_tokenizer.pre_tokenizer` with `tokenizer.pre_tokenizer` to fix the Mistral regex AttributeError. PR 45317 just adds tests and a slightly broader wrapper around the same fix, so these look like duplicate/overlapping PRs that could be merged into one." + "reason": "Same underlying bug and same fix in the same file, both tied to issue 45081; 45317 is essentially a superset with test coverage." } ] }, @@ -22334,7 +22571,7 @@ "soft_pairs": [ "pull_request:45115|pull_request:45569" ], - "prepared_review_unit_hash": "f84ef91ea25f8cecb5606f6bcc8ad45564bb9de14d8eae02e4fe516808469e83", + "prepared_review_unit_hash": "bcefe9c0f546358a88049052d477e1672f70b8fa6bb5f26dcad260958eea0a8a", "component_budget": { "node_count": 2, "item_count": 2, @@ -22356,18 +22593,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs concern Nemotron H, but they are different changes: one refactors NemotronH to inherit GraniteMoeHybrid and updates conversion mapping, while the other introduces the dense/sparse NemotronH variants and broader model/docs wiring. They share files and subsystem, but not the same concrete fix.", - "confidence": 0.91, + "summary": "These PRs are related NemotronH architecture changes, but they are not the same fix/change: #45115 is a refactor toward GraniteMoeHybrid-compatible conversion and inheritance, while #45569 is a broader implementation split introducing NemotronHDense/NemotronHSparse plus docs, auto-mapping, and tests.", + "confidence": 0.82, "canonical_issue_reason": null, - "canonical_pr_reason": "No issue links are present. If treating one PR as the cluster representative, PR #45569 is the broader, more complete NemotronH family change, while #45115 is a narrower refactor/conversion-mapping update.", + "canonical_pr_reason": "#45569 is the more complete and representative NemotronH code change: it adds the dense/sparse model packages, updates auto mappings and docs, and includes test coverage, so it better captures the final architecture work than the narrower refactor in #45115.", "best_issue_reason": null, - "best_pr_reason": "PR #45569 is the best representative because it covers the major model-structure change, adds dense/sparse variants, updates docs and auto-mappings, and appears to subsume the larger product-facing work.", + "best_pr_reason": "#45569 is the best candidate to anchor this cluster because it is the broader end-state implementation and subsumes the model-structure direction that #45115 is working toward.", "soft_edge_verdicts": [ { "left": "pull_request:45115", "right": "pull_request:45569", "accept": false, - "reason": "Shared NemotronH files are not enough here: #45115 is a narrow inheritance/conversion-mapping refactor, while #45569 introduces the dense/sparse split and associated docs/model wiring. They do not look mergeable as one PR for the same underlying bug or change." + "reason": "They share the NemotronH area and some files, but the concrete changes differ: #45115 is a conversion/inheritance refactor, while #45569 introduces separate dense and sparse model variants with docs and tests. They do not look like one mergeable PR for the same underlying bug or change." } ] }, @@ -22397,7 +22634,7 @@ "soft_pairs": [ "pull_request:45332|pull_request:45333" ], - "prepared_review_unit_hash": "8a809bde9c61ba4cd33734c4a28f9ac72caebd63204f7b7fad8073170e01d4dd", + "prepared_review_unit_hash": "e5cfa185df3e02d4bd4128c514a77e9b18815c4cc633eaa7f65a0890eda62ea0", "component_budget": { "node_count": 2, "item_count": 2, @@ -22419,18 +22656,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are related feature PRs in the same heterogeneous-models effort, but they address different layers of the stack: one introduces per-layer config support, the other adds broader heterogeneous modeling/cache behavior built on top of that config. They are not the same concrete change and should not be deduplicated.", - "confidence": 0.91, + "summary": "The two PRs are closely related parts of a larger heterogeneous-model effort, but they address different layers of the stack: one adds per-layer config support, the other adds per-layer modeling/cache behavior. They are not the same underlying change and would not be merged as a single duplicate PR.", + "confidence": 0.92, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45333 is the more foundational change because it introduces the per-layer configuration machinery that the other PR appears to rely on.", + "canonical_pr_reason": "PR 45333 is the more foundational change because it introduces per-layer configuration support, which the modeling PR builds on.", "best_issue_reason": null, - "best_pr_reason": "PR 45333 is the best representative of the cluster since it defines the core config abstraction, whereas PR 45332 is a larger downstream modeling/cache integration built around it.", + "best_pr_reason": "PR 45333 is the best representative of this cluster because it defines the core heterogeneous-config API and is the enabling piece for the broader feature.", "soft_edge_verdicts": [ { "left": "pull_request:45332", "right": "pull_request:45333", "accept": false, - "reason": "Related but distinct: 45333 adds heterogeneous per-layer config support, while 45332 adds heterogeneous model/cache/modeling integration on top of that. They touch overlapping files, but not the same underlying bug or one mergeable fix." + "reason": "Related feature work, but not the same concrete change: 45333 adds heterogeneous per-layer config, while 45332 implements heterogeneous model/cache behavior on top of that. They are complementary, not duplicates." } ] }, @@ -22460,7 +22697,7 @@ "soft_pairs": [ "pull_request:45415|pull_request:45425" ], - "prepared_review_unit_hash": "4cfa7ef6e23255d7b4c911eaa7472e01268129090e0dda81948c5ea13d110442", + "prepared_review_unit_hash": "7dd0e54a547a87e57bab0076f600577ceb8c84997b0964575750743cb2c519bd", "component_budget": { "node_count": 2, "item_count": 2, @@ -22482,18 +22719,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs are typing-related, but they address different changes: #45415 is a broad repository-wide type-checking sweep, while #45425 adds typing helpers/protocols and targeted fixes around modeling_utils/peft/flash-attention. The overlap in shared files looks incidental, not the same underlying change.", - "confidence": 0.93, + "summary": "Both pull requests are typing-related and overlap in a few files, but they address different changesets: #45415 is a broad repository-wide type-checking sweep, while #45425 adds shared typing protocol/helpers for specific integrations. They are not the same underlying bug/change and should not be deduplicated.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical duplicate PR: these are separate typing efforts with different scopes and code paths, so neither should be treated as the same change.", + "canonical_pr_reason": "No strong duplicate canonical PR here; if forced, #45425 is the narrower, more concrete typing helper change, while #45415 is a broader type-checking sweep.", "best_issue_reason": null, - "best_pr_reason": "#45425 is the more focused representative of the cluster because it introduces specific typing infrastructure and targeted fixes, whereas #45415 is a broad sweeping type-check pass.", + "best_pr_reason": "#45425 is the better representative of the cluster\u2019s more specific typing work because it introduces shared protocol/type aliases tied to particular code paths, whereas #45415 is a wide mechanical type-checking pass across many files.", "soft_edge_verdicts": [ { "left": "pull_request:45415", "right": "pull_request:45425", "accept": false, - "reason": "Reject: they share some touched files, but #45415 is a broad type-checking cleanup across many modules and #45425 adds specific typing aliases/protocols for modeling_utils and related integrations. Different underlying changes, not mergeable as one PR." + "reason": "Both are typing-related and share some touched files, but #45415 is a broad mypy/pyright cleanup across many modules, while #45425 adds specific typing protocols/aliases for modeling_utils and PEFT-related code. Different scope and not mergeable as one duplicate change." } ] }, @@ -22523,7 +22760,7 @@ "soft_pairs": [ "pull_request:45513|pull_request:46084" ], - "prepared_review_unit_hash": "7c65754d8a470aac9977914572f8d7b44490bb8fc53409fac660f891abee65c1", + "prepared_review_unit_hash": "78ee2bf4b42f4af3d46f543da3f62dfc83b517bba01f1db5e89f0e2dc04f4326", "component_budget": { "node_count": 2, "item_count": 2, @@ -22545,18 +22782,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two open PRs with superficially similar cache/seq_len handling, but they target different model families and different code paths: Qwen/Olmo linear attention vs Mamba2 CUDA kernels. They should not be deduplicated.", - "confidence": 0.93, + "summary": "These are both cache/seq_len>1 bug-fix PRs, but they target different model families and different execution paths. The similarity is only at a broad symptom level, not the same underlying change.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #46084 is the more clearly scoped candidate: it has an explicit issue target (#46032) and a focused fix for Mamba2 cache behavior when seq_len > 1.", + "canonical_pr_reason": "PR #45513 is the broader representative: it fixes multi-token cached forward across several Qwen3-family model files and includes the underlying control-flow change for cached continuation.", "best_issue_reason": null, - "best_pr_reason": "PR #46084 is the strongest standalone representative because it is narrowly targeted, explicitly linked to an issue, and addresses a concrete cache/sequence-length bug in one code path.", + "best_pr_reason": "PR #45513 is the best representative of the cluster because it spans multiple related model implementations and addresses the shared cached-forward logic directly.", "soft_edge_verdicts": [ { "left": "pull_request:45513", "right": "pull_request:46084", "accept": false, - "reason": "Different underlying bugs in different models/code paths. #45513 fixes multi-token cached forward for Qwen/Olmo linear attention, while #46084 fixes Mamba2 use_cache handling for seq_len > 1." + "reason": "Different concrete code paths and model families: Qwen3/OLMo hybrid linear attention cached forward vs Mamba2 CUDA kernel cache handling. They share the general theme of seq_len>1 cache bugs, but not the same underlying fix." } ] }, @@ -22586,7 +22823,7 @@ "soft_pairs": [ "pull_request:45548|pull_request:45662" ], - "prepared_review_unit_hash": "1d134ff2d24f8b52f1f996aaa44df7de3489ae9a74f5704eccab9a1134f94213", + "prepared_review_unit_hash": "159284bfeb8921cfbd03edfdd74e67a8aa897bf17eb18e7c53e01887fcdf7b2a", "component_budget": { "node_count": 2, "item_count": 2, @@ -22608,18 +22845,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These PRs are related by expert parallelism, but they fix different loading-time bugs in different code paths: one preserves EP config/has_ep for DeepSpeed ZeRO-3 via accelerate launch, the other prevents expert parameters from being overwritten in FSDP2/rank-0 broadcast. Not duplicates.", - "confidence": 0.95, + "summary": "These PRs are related by expert-parallel loading, but they fix different failure modes on different code paths: one addresses EP + DeepSpeed ZeRO-3/accelerate launch config handling, the other addresses EP + FSDP2 expert parameters being overwritten during sharded loading. They are not duplicates.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45662 is the better representative because it targets a concrete EP parameter-loading failure mode with a broader, more explicit fix across the loading stack.", + "canonical_pr_reason": "PR #45662 is the stronger representative of the cluster because it targets a concrete loading corruption bug and includes the broader set of loader-side fixes around sharded parameter handling; PR #45548 is narrower and focused on config serialization plus an EP capability flag for a different launch path.", "best_issue_reason": null, - "best_pr_reason": "PR 45662 is the strongest standalone PR for the cluster: it addresses a specific, user-visible EP bug and includes the most complete code-path fix.", + "best_pr_reason": "PR #45662 is the better standalone PR for a canonical bug-fix anchor since it fixes a specific expert-parameter overwrite path with direct loading consequences, whereas #45548 is a separate accelerate/ZeRO-3 compatibility fix.", "soft_edge_verdicts": [ { "left": "pull_request:45548", "right": "pull_request:45662", "accept": false, - "reason": "Both concern EP, but 45548 fixes config preservation/has_ep for DeepSpeed ZeRO-3 with accelerate launch, while 45662 fixes expert tensors being overwritten in FSDP2/rank-0 broadcast. Different bugs and different mechanisms." + "reason": "Both mention EP and touch modeling_utils, but they fix different bugs: #45548 is about serializing distributed_config and exposing has_ep for DeepSpeed ZeRO-3/accelerate launch, while #45662 fixes post-shard wrapping to prevent experts from being overwritten under FSDP2. Different concrete code paths and not safely mergeable into one PR." } ] }, @@ -22649,12 +22886,12 @@ "soft_pairs": [ "pull_request:45639|pull_request:45645" ], - "prepared_review_unit_hash": "87d99e0499be3d097649f2fc1173b8c7e0da5f57df73707d4ea0e500d88ae731", + "prepared_review_unit_hash": "8822e831a7c5a45aa89bb32e2b69f5aca680fa2d296f394f2b3f00a94c8cff3e", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4589, + "serialized_chars": 4590, "estimated_input_tokens": 1148, "estimated_eval_tokens": 2552 }, @@ -22662,7 +22899,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4589, + "serialized_chars": 4590, "estimated_input_tokens": 1148, "estimated_eval_tokens": 2552 }, @@ -22671,18 +22908,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs address the same xdist collision around `captured_info` debug artifacts, with PR #45645 being the broader end-to-end fix and PR #45639 a narrower subset of the same underlying change.", - "confidence": 0.89, + "summary": "Both PRs address the same xdist collision problem around `captured_info` debug artifacts in `testing_utils.py`; PR #45645 is the broader, more complete version that also updates CI log collection.", + "confidence": 0.92, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #45645 is the stronger canonical choice because it covers the core xdist-safe file handling plus the CI/workflow log aggregation needed to consume multiple `captured_info*.txt` files, making it the more complete fix for the issue.", + "canonical_pr_reason": "PR #45645 is the better canonical PR because it covers the core xdist-safe `captured_info` fix plus the downstream CI/workflow changes needed to preserve and collect the split log files.", "best_issue_reason": null, - "best_pr_reason": "PR #45645 best represents the full solution to the captured-info collision problem; PR #45639 appears to implement only the file-writing side of that fix.", + "best_pr_reason": "PR #45645 is the strongest representative of the cluster: same underlying bug as #45639, but with the full end-to-end handling across test utilities and CI artifact collection.", "soft_edge_verdicts": [ { "left": "pull_request:45639", "right": "pull_request:45645", "accept": true, - "reason": "Both PRs fix the same concrete bug: `captured_info` artifacts colliding under xdist. They share the same issue target and overlapping code paths in `testing_utils.py`; #45645 simply extends the fix to CI artifact collection and related plumbing, so they could plausibly be merged into one PR." + "reason": "Same underlying xdist `captured_info` collision fix in `testing_utils.py`; #45645 appears to extend and supersede the narrower implementation in #45639." } ] }, @@ -22712,7 +22949,7 @@ "soft_pairs": [ "pull_request:45720|pull_request:46259" ], - "prepared_review_unit_hash": "23744d238d4a05cf4ed19262d2c01aa014c87e01baff9dbeb20a6fb9ca6929aa", + "prepared_review_unit_hash": "899ac2bbbbea35d3e3eb4dec737526106082b63916ca4bee3c24b866322ce783", "component_budget": { "node_count": 2, "item_count": 2, @@ -22734,18 +22971,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are two unrelated linter-rule PRs that happen to touch overlapping model files, but they implement different behaviors: TRF003 adjusts forward return handling, while TRF018 rewrites weight initialization logic.", - "confidence": 0.97, + "summary": "These are both transformers modeling lint/chore PRs, but they address different rules and different code paths: PR #45720 adds `can_return_tuple`/return-dict handling across model forwards, while PR #46259 changes `_init_weights` behavior for a broad set of models. The overlap in filenames looks incidental, not duplicate work.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45720 is the cleaner representative because it has a narrowly defined, concrete change set around `can_return_tuple`/`return_dict` handling and shows the exact code-path being standardized.", + "canonical_pr_reason": "PR #45720 is the better anchor because it is a narrower, more concrete code change with a clearly scoped rule implementation (TRF003) and a consistent forward-path modification pattern.", "best_issue_reason": null, - "best_pr_reason": "PR 45720 is the best overall anchor for the cluster since its intent is more explicit and its change is easier to summarize as one coherent modeling-rule implementation.", + "best_pr_reason": "PR #45720 is the best representative of the cluster because it is more focused and easier to triage as a single change, whereas PR #46259 targets a different lint rule and a different initialization path.", "soft_edge_verdicts": [ { "left": "pull_request:45720", "right": "pull_request:46259", "accept": false, - "reason": "Do not merge: they target different linter rules and different code paths. One changes model forward return semantics, the other changes parameter initialization behavior; overlapping model filenames alone is not enough." + "reason": "Reject: they implement different linter rules on different code paths (`forward` return handling vs `_init_weights` initialization). Shared model filenames do not indicate the same underlying fix." } ] }, @@ -22841,7 +23078,7 @@ "soft_pairs": [ "pull_request:45932|pull_request:46100" ], - "prepared_review_unit_hash": "500313808a234f135d929d0d8fc96dfc42b03794447753affe7dc9e87a55cd0b", + "prepared_review_unit_hash": "6a0648c14b70b48510cfeb02f3611e3ba8fd1ef446db9ba1e57fff3455b72f78", "component_budget": { "node_count": 2, "item_count": 2, @@ -22863,18 +23100,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are unrelated pull requests: one changes model/workflow code and the other is a test migration. They do not look like the same underlying change and should not be deduplicated.", - "confidence": 0.94, + "summary": "The two pull requests are not duplicates: one is a Grafana/test CI and model-configuration miscellany PR, while the other is a test migration focused on kernel and vision-model tests. They touch different changesets and are not plausibly one merged fix/change.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45932 is the broader, more substantive code change and is the better representative of the cluster; PR 46100 is primarily a test migration with no clear overlap in the actual fix.", + "canonical_pr_reason": "pull_request:46100 is the more coherent, self-contained change set (test migration and kernel test additions), making it the better cluster representative.", "best_issue_reason": null, - "best_pr_reason": "PR 45932 is the best PR representative because it targets production/model and CI files, whereas PR 46100 is focused on test scaffolding and migration rather than the same concrete bug or change.", + "best_pr_reason": "pull_request:46100 is the best representative PR because it has a single, clear purpose and a tighter scope than pull_request:45932, which mixes CI/workflow and unrelated model edits.", "soft_edge_verdicts": [ { "left": "pull_request:45932", "right": "pull_request:46100", "accept": false, - "reason": "Different work items: 45932 modifies model code and CI secrets, while 46100 adds/changes tests and DDP config. They do not appear to fix the same concrete problem or be mergeable as one PR." + "reason": "They are different kinds of changes affecting different areas: 45932 is a mixed workflow/model tweak PR, while 46100 is a test migration with new kernel and vision tests. They do not look like the same underlying bug or change and would not sensibly be merged into one PR." } ] }, @@ -22904,40 +23141,40 @@ "soft_pairs": [ "pull_request:46126|pull_request:46269" ], - "prepared_review_unit_hash": "aa3ec574eb67586225f03d5b26642de9a716ec53dc05251042d0145fc80db637", + "prepared_review_unit_hash": "0d8835e117242dab959ba7b91e349ae861aedb2433b6d18efcd95132845634ce", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 5525, - "estimated_input_tokens": 1382, - "estimated_eval_tokens": 3020 + "serialized_chars": 5522, + "estimated_input_tokens": 1381, + "estimated_eval_tokens": 3018 }, "budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 5525, - "estimated_input_tokens": 1382, - "estimated_eval_tokens": 3020 + "serialized_chars": 5522, + "estimated_input_tokens": 1381, + "estimated_eval_tokens": 3018 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These PRs overlap in distributed files, but they are not the same underlying change: #46126 is a focused EP/trainer integration, while #46269 is a broad distributed branch base touching many unrelated areas. Reject the duplicate link.", - "confidence": 0.95, + "summary": "These are not duplicates: one PR is a focused EP/MoE + Trainer integration change, while the other is a broad distributed-branch base refactor touching many unrelated distributed components. The shared filenames are too generic to imply the same underlying fix.", + "confidence": 0.93, "canonical_issue_reason": null, - "canonical_pr_reason": "#46126 is the more focused, self-contained PR around EP routing and trainer integration; its diff is coherent and specific compared with the broad umbrella changes in #46269.", + "canonical_pr_reason": "PR #46126 is the more canonical candidate because it is narrowly scoped to EP and Trainer integration on top of DistributedConfig, with a concrete, coherent feature intent.", "best_issue_reason": null, - "best_pr_reason": "#46126 is the best representative artifact here because it targets one concrete feature/change, whereas #46269 is a large branch-base PR spanning multiple distributed subsystems.", + "best_pr_reason": "PR #46126 is the best standalone PR because it targets a specific code path and feature set, whereas PR #46269 is a broad integration/base branch with many disparate changes.", "soft_edge_verdicts": [ { "left": "pull_request:46126", "right": "pull_request:46269", "accept": false, - "reason": "Shared filenames alone are not enough: #46126 implements EP-specific routing/trainer integration, while #46269 is a broad distributed branch base with many additional changes. They do not fix the same concrete code-path problem." + "reason": "Reject: despite overlapping distributed filenames, the PRs have different goals and scopes. #46126 adds EP/Trainer integration and router/tensor-parallel behavior, while #46269 is a large distributed branch base touching FSDP, generation, integrations, and model loading. They do not look like the same concrete bug or mergeable single fix." } ] }, @@ -23003,208 +23240,208 @@ "pull_request:46504|pull_request:46563", "pull_request:46504|pull_request:46716" ], - "prepared_review_unit_hash": "3442f31b832421924081f2ce1c4aed8581966b87cfe8441032244b5e9d4dd48c", + "prepared_review_unit_hash": "ee65201344f36eea5359dae4ec2081b14ec88c60fe3e93ec67946de5030cb12d", "component_budget": { "node_count": 10, "item_count": 10, "soft_pair_count": 29, - "serialized_chars": 25922, - "estimated_input_tokens": 6481, - "estimated_eval_tokens": 13218 + "serialized_chars": 25926, + "estimated_input_tokens": 6482, + "estimated_eval_tokens": 13220 }, "budget": { "node_count": 10, "item_count": 10, "soft_pair_count": 29, - "serialized_chars": 25922, - "estimated_input_tokens": 6481, - "estimated_eval_tokens": 13218 + "serialized_chars": 25926, + "estimated_input_tokens": 6482, + "estimated_eval_tokens": 13220 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is mostly unrelated GitHub Actions maintenance PRs. There are two loose themes\u2014workflow hardening around pr-ci-security-gate.yml and assorted one-off CI tweaks\u2014but no pair appears to be the same underlying change.", - "confidence": 0.97, + "summary": "These PRs are all workflow-maintenance changes, but they split into unrelated subtopics: several distinct CI workflow tweaks and a separate set of different security-gate edits. The shared filenames/title pattern creates superficial similarity, but the concrete changes do not match well enough to treat any pair as duplicates.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46554 is the most substantive representative: it makes concrete security-gate logic changes in .github/workflows/pr-ci-security-gate.yml and sits at the center of the nearest soft-similarity neighborhood. It is still not a duplicate of the others.", + "canonical_pr_reason": "No true duplicate PR stands out; PR 46554 is the best representative of the cluster because it makes the most substantive security-gate workflow change.", "best_issue_reason": null, - "best_pr_reason": "46554 is the best overall representative because it changes actual security-gate behavior rather than just permissions, comments, or minor workflow plumbing. The cluster is too heterogeneous to treat as one duplicate set.", + "best_pr_reason": "PR 46554 is the strongest representative of the set: it has concrete runtime behavior changes in the security-gate workflow, whereas most others are smaller, unrelated workflow tweaks.", "soft_edge_verdicts": [ { "left": "pull_request:46138", "right": "pull_request:46554", "accept": false, - "reason": "Different workflows and different fixes; same general CI/workflow maintenance theme is not enough to make these the same change." + "reason": "Different workflows and unrelated changes; this is only a broad CI-maintenance similarity." }, { "left": "pull_request:46138", "right": "pull_request:46558", "accept": false, - "reason": "Unrelated workflow changes: self-comment stdin handling versus security-gate permissions/approval logic." + "reason": "Different workflows and different fixes; no shared underlying bug or change." }, { "left": "pull_request:46138", "right": "pull_request:46563", "accept": false, - "reason": "One updates self-comment-ci input handling; the other only adds explanatory comments to the security-gate workflow." + "reason": "Different workflows and unrelated edits; both are workflow chores, not duplicates." }, { "left": "pull_request:46138", "right": "pull_request:46716", "accept": false, - "reason": "Different workflow and different concern: stdin handling for comments versus secrets scoping in pr-ci-security-gate.yml." + "reason": "Different workflows and different behavior changes; not the same PR." }, { "left": "pull_request:46138", "right": "pull_request:46504", "accept": false, - "reason": "Both use printf-style shell handling, but they fix different workflows and different code paths." + "reason": "Different workflow files and different shell-handling changes; only the maintenance style matches." }, { "left": "pull_request:46432", "right": "pull_request:46503", "accept": false, - "reason": "SSH-runner hardening is unrelated to adding workflow_dispatch to benchmark_v2." + "reason": "Different workflow files and unrelated CI behavior changes." }, { "left": "pull_request:46432", "right": "pull_request:46501", "accept": false, - "reason": "Different workflow files and different fixes: SSH runner changes versus anti-slop permissions placement." + "reason": "Different workflow files and different operational concerns; not duplicates." }, { "left": "pull_request:46432", "right": "pull_request:46502", "accept": false, - "reason": "SSH-runner updates are not the same change as checking out the base ref for assign-reviewers." + "reason": "Different workflow files and different changes; same subsystem is too broad." }, { "left": "pull_request:46432", "right": "pull_request:46554", "accept": false, - "reason": "SSH-runner secret handling and pinned download changes are unrelated to security-gate file checks." + "reason": "SSH runner hardening vs PR CI security-gate logic; unrelated code paths." }, { "left": "pull_request:46432", "right": "pull_request:46558", "accept": false, - "reason": "Different workflows and different problems; no shared code-path change." + "reason": "Different workflow areas and different fixes; no single underlying change." }, { "left": "pull_request:46501", "right": "pull_request:46503", "accept": false, - "reason": "Anti-slop permissions and benchmark trigger conditions are different workflow behaviors." + "reason": "Both touch workflows, but they change different jobs and behaviors." }, { "left": "pull_request:46501", "right": "pull_request:46502", "accept": false, - "reason": "Different workflow files and different fixes: anti-slop permissions versus reviewer assignment checkout ref." + "reason": "Same broad workflow-maintenance theme, but unrelated workflow files and changes." }, { "left": "pull_request:46501", "right": "pull_request:46504", "accept": false, - "reason": "Anti-slop permission scoping is unrelated to build-ci-docker-images shell input handling." + "reason": "Different workflow files and different fixes; not the same bug/change." }, { "left": "pull_request:46501", "right": "pull_request:46554", "accept": false, - "reason": "Different workflows and different issues; only broad GitHub Actions overlap." + "reason": "Unrelated workflow updates; one is anti-slop, the other is PR CI security gating." }, { "left": "pull_request:46502", "right": "pull_request:46503", "accept": false, - "reason": "Reviewer checkout behavior and benchmark dispatch triggering are unrelated changes." + "reason": "Different workflow files and different trigger/behavior changes." }, { "left": "pull_request:46502", "right": "pull_request:46504", "accept": false, - "reason": "Different workflow files and different concrete fixes; not the same bug or change." + "reason": "Different workflow files and different shell/input handling changes." }, { "left": "pull_request:46502", "right": "pull_request:46554", "accept": false, - "reason": "assign-reviewers checkout ref and security-gate workflow scanning solve different problems." + "reason": "Assign-reviewers vs PR CI security-gate are unrelated workflow changes." }, { "left": "pull_request:46503", "right": "pull_request:46504", "accept": false, - "reason": "Benchmark dispatch support and build-ci-docker-images printf handling are distinct workflow changes." + "reason": "Different workflow files and different edits; same CI domain is too broad." }, { "left": "pull_request:46503", "right": "pull_request:46554", "accept": false, - "reason": "No shared underlying bug: benchmark trigger logic versus security-gate checks." + "reason": "Benchmark trigger logic vs PR CI security checks are unrelated." }, { "left": "pull_request:46504", "right": "pull_request:46554", "accept": false, - "reason": "Both are workflow tweaks, but they target different scripts and different failure modes." + "reason": "Build-image trigger handling vs PR CI security-gate logic are different changes." }, { "left": "pull_request:46504", "right": "pull_request:46558", "accept": false, - "reason": "build-ci-docker-images input handling is unrelated to pr-ci-security-gate permissions and approvals." + "reason": "Different workflow files and different behavior changes; not duplicates." }, { "left": "pull_request:46554", "right": "pull_request:46558", "accept": false, - "reason": "Same file, but different concerns: one adds file-change/security checks, the other changes permissions and approval gating." + "reason": "Same file, but different concerns: workflow-file checks vs permissions/approval gating." }, { "left": "pull_request:46554", "right": "pull_request:46563", "accept": false, - "reason": "Security-gate logic changes versus comment-only documentation updates are not the same change." + "reason": "Same file, but one changes gate logic and the other adds explanatory comments; not the same fix." }, { "left": "pull_request:46554", - "right": "46567", + "right": "pull_request:46716", "accept": false, - "reason": "No matching artifact; the intended pair appears to be 46554 vs 46716, which are different security-gate hardening changes." + "reason": "Same file, but different security-gate adjustments: checks vs secrets handling." }, { "left": "pull_request:46558", "right": "pull_request:46563", "accept": false, - "reason": "Permissions/approval logic and explanatory comments are not the same underlying fix." + "reason": "Both in the same workflow file, but they address different aspects of the gate." }, { "left": "pull_request:46558", "right": "pull_request:46716", "accept": false, - "reason": "Different security-gate hardening concerns: job permissions versus secrets inheritance." + "reason": "Permissions/approval gating vs secrets scoping are different changes." }, { "left": "pull_request:46563", "right": "pull_request:46716", "accept": false, - "reason": "Comment-only clarification is not the same as changing secrets exposure in the reusable workflow." + "reason": "Comment-only clarification vs secrets handling; not the same underlying change." }, { "left": "pull_request:46504", "right": "pull_request:46563", "accept": false, - "reason": "build-ci-docker-images shell handling and security-gate comments are unrelated." + "reason": "Different workflow files and unrelated CI maintenance changes." }, { "left": "pull_request:46504", "right": "pull_request:46716", "accept": false, - "reason": "Different workflow files and different changes; no duplicate evidence beyond broad similarity." + "reason": "Different workflow files and different security/CI concerns; not duplicates." } ] }, @@ -23234,7 +23471,7 @@ "soft_pairs": [ "pull_request:46152|pull_request:46207" ], - "prepared_review_unit_hash": "317a65d39ecd3afd2e36c39a21c3f52a8e6eaa3b413eae75109779c83534a533", + "prepared_review_unit_hash": "0b07ab9c2abd60b54bd9d9c0ae14502e8aa1b79f7618f0aa131e30b1f3b103ff", "component_budget": { "node_count": 2, "item_count": 2, @@ -23256,18 +23493,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two open PRs both remove direct file-open source inspection, but they do so in different code paths and with different mechanisms. They are related in theme, not a duplicate fix.", - "confidence": 0.89, + "summary": "Both pull requests are about eliminating brittle source-file open() access, but they fix different code paths in different modules, so they are not duplicates and should not be merged together.", + "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #46207 is the cleaner representative because it targets the specific _can_set_attn_implementation heuristic in modeling_utils, includes tests, and has a narrowly defined behavioral change.", + "canonical_pr_reason": "PR 46207 is the stronger standalone change: it targets a specific source-reading heuristic in `modeling_utils.py`, includes a focused test update, and is more concrete than the import-utils-only fix in 46152.", "best_issue_reason": null, - "best_pr_reason": "PR #46207 is the best cluster representative: it is the more explicit and self-contained fix, with a concrete source-reading replacement and accompanying tests.", + "best_pr_reason": "PR 46207 is the best representative of this cluster because it has a clearly scoped runtime behavior change and test coverage, whereas 46152 addresses a separate utility-path implementation.", "soft_edge_verdicts": [ { "left": "pull_request:46152", "right": "pull_request:46207", "accept": false, - "reason": "Different code paths and different implementation details: #46152 changes import_utils path resolution via importlib.resources, while #46207 changes modeling_utils source inspection via inspect.getsource. They share a broad goal, not the same underlying bug/change." + "reason": "Both avoid reading source via `open()`, but 46152 changes `import_utils._resolve_traversable` while 46207 changes `modeling_utils._can_set_attn_implementation`; these are different concrete bugs/code paths and would not plausibly be merged into one PR." } ] }, @@ -23297,7 +23534,7 @@ "soft_pairs": [ "pull_request:46154|pull_request:46196" ], - "prepared_review_unit_hash": "736003696cd355b9e5a766ea463066193898cda2ac36b77d3362c23b76a88785", + "prepared_review_unit_hash": "bc902838473adbcaf4340697a008df6eba12291f1f94ef38d48ebd8842bb990a", "component_budget": { "node_count": 2, "item_count": 2, @@ -23319,18 +23556,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These PRs both add SDPA dispatch support, but for different model families and code paths: T5 vs wav2vec2/audio models. They are related in theme, not duplicates, so the soft edge should be rejected.", - "confidence": 0.93, + "summary": "Both pull requests add SDPA attention dispatch, but they target different model families and code paths: T5 versus wav2vec2/audio models. They are thematically similar but not the same underlying change, so they should not be merged as duplicates.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: the two changes target different architectures and implementation paths, so they are not the same underlying fix.", + "canonical_pr_reason": "No single canonical PR: 46154 is a T5-specific SDPA dispatch change, while 46196 is a separate, broader audio-model SDPA dispatch effort across multiple wav2vec2-related architectures.", "best_issue_reason": null, - "best_pr_reason": "PR 46196 is the broader representative of the SDPA-dispatch theme, covering many audio model implementations and related docs/tests, but it is still not a duplicate of PR 46154.", + "best_pr_reason": "If one PR must be highlighted as the broader representative of this SDPA rollout, 46196 is the more comprehensive change set; however, it is still not a duplicate of 46154.", "soft_edge_verdicts": [ { "left": "pull_request:46154", "right": "pull_request:46196", "accept": false, - "reason": "Same feature theme, but different model families and concrete code paths; they would not plausibly be merged into one PR without conflating separate implementations." + "reason": "Different concrete implementations and model families. T5 attention dispatch and wav2vec2/audio attention dispatch are separate code paths, so they are not mergeable as one PR." } ] }, @@ -23426,12 +23663,12 @@ "soft_pairs": [ "pull_request:46310|pull_request:46546" ], - "prepared_review_unit_hash": "1c2eaea4aeb0fed76698de3e900d83bbdb0e31ce87da600535f6b16e41d8aef3", + "prepared_review_unit_hash": "0580782e39049fb5fc7bcb9dbb2b9172024f2b945066699300fa26ef2b88dd1e", "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 }, @@ -23439,7 +23676,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4850, + "serialized_chars": 4851, "estimated_input_tokens": 1213, "estimated_eval_tokens": 2682 }, @@ -23448,18 +23685,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs fix the same off-by-one right-window boundary bug in sliding window masking; 46546 is the broader, more complete patch.", - "confidence": 0.9, + "summary": "Both PRs fix the same off-by-one bug in the sliding-window right boundary (`-dist <= right_window_size` instead of `<`) for Moonshine streaming. PR 46546 is the more complete change set because it also adds the parallel Gemma4 fix and tests, while 46310 is the narrower Moonshine-only patch.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46546 is the better canonical change because it covers the same Moonshine fix plus the related Gemma4 implementation and adds test coverage.", + "canonical_pr_reason": "PR 46546 is the better canonical PR: it contains the Moonshine streaming boundary fix plus the related Gemma4 update and tests, making it the more complete implementation of the same bug fix.", "best_issue_reason": null, - "best_pr_reason": "PR 46546 is the strongest representative because it is the later, broader fix for the same boundary bug and appears more complete.", + "best_pr_reason": "PR 46546 is the strongest representative of the fix because it addresses the boundary bug across all affected code paths and includes test coverage.", "soft_edge_verdicts": [ { "left": "pull_request:46310", "right": "pull_request:46546", "accept": true, - "reason": "Same underlying off-by-one boundary bug in the sliding-window mask; 46546 extends the same fix to a related implementation and tests, so these can be treated as duplicates." + "reason": "Both PRs correct the same right-window off-by-one boundary in the same inner_mask logic; 46546 is just a broader, more complete version of the same fix." } ] }, @@ -23489,7 +23726,7 @@ "soft_pairs": [ "pull_request:46376|pull_request:46417" ], - "prepared_review_unit_hash": "737573dba5a9464ec49b0b9a8b0a7452ceb0f98b3949bb484455354f662759ad", + "prepared_review_unit_hash": "cf8c6d7e2c3963198e26621f078a33a72c38829a191779487ffc6a613a30c922", "component_budget": { "node_count": 2, "item_count": 2, @@ -23511,18 +23748,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two pull requests are unrelated: one adds multimodal support to continuous batching, the other adds the HunYuan VL model stack and docs. The soft similarity signal is a false positive.", + "summary": "The two pull requests are unrelated: one adds multimodal support to continuous batching internals, while the other adds the HunYuan VL model, docs, processors, and tests. They do not look like the same underlying change or bug.", "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46417 is the more self-contained, complete model addition with docs/tests, while PR 46376 is a draft feature expansion in a different codepath.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR 46417 is the best representative of the cluster because it is a fully scoped standalone addition; PR 46376 is draft and addresses a separate batching subsystem.", + "best_pr_reason": "PR #46417 is the more complete, self-contained feature addition with docs and tests, so it is the better representative PR for the cluster, even though it is not a duplicate of #46376.", "soft_edge_verdicts": [ { "left": "pull_request:46376", "right": "pull_request:46417", "accept": false, - "reason": "Reject: these PRs target different features and code paths (continuous batching multimodal support vs. HunYuan VL model implementation), so they are not the same underlying change." + "reason": "Different subsystems and goals: continuous batching multimodal runtime changes vs a new model implementation/docs/tests. No shared filenames, targets, or concrete code-path overlap suggesting a single fix." } ] }, @@ -23552,40 +23789,40 @@ "soft_pairs": [ "pull_request:46574|pull_request:46580" ], - "prepared_review_unit_hash": "2419e65a753f6fc6be7bf9ff1568ff572ddf8cc552063e67a999b9bc5bec5aa9", + "prepared_review_unit_hash": "fb7049813207caa05053baee268914f7423f6aeb7188471c4972a6475caa8327", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4408, - "estimated_input_tokens": 1102, - "estimated_eval_tokens": 2460 + "serialized_chars": 4410, + "estimated_input_tokens": 1103, + "estimated_eval_tokens": 2462 }, "budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4408, - "estimated_input_tokens": 1102, - "estimated_eval_tokens": 2460 + "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": "The two pull requests are related only at a very broad tokenizer-loading level, but they fix different bugs in different code paths: CodeLlama decode stripping vs. DeepSeek checkpoint-to-tokenizer-class misclassification. They should not be deduplicated.", - "confidence": 0.98, + "summary": "These are two unrelated tokenizer-loading fixes: one changes CodeLlama decode behavior to preserve leading spaces, the other fixes DeepSeek-Coder/DeepSeek-LLM auto-tokenizer selection for certain checkpoints. They are not the same bug or mergeable change.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46580 is the better representative of the cluster because it addresses tokenizer-class resolution in auto-loading with a targeted checkpoint-prefix override, which is the more general and explicit fix among the two PRs.", + "canonical_pr_reason": "No single canonical PR: the two pull requests fix different code paths and different underlying failures.", "best_issue_reason": null, - "best_pr_reason": "PR 46580 is the strongest cluster representative: it has a clear, scoped tokenizer-loading fix, explicit checkpoint targeting, and a direct issue link. PR 46574 is a separate CodeLlama decode bug.", + "best_pr_reason": "PR #46580 is the better representative of the cluster only in the sense that it addresses a broader auto-tokenizer misclassification problem; however, it is not a duplicate of #46574.", "soft_edge_verdicts": [ { "left": "pull_request:46574", "right": "pull_request:46580", "accept": false, - "reason": "Different underlying problems and code paths: 46574 fixes CodeLlama decoder stripping on decode, while 46580 fixes DeepSeek models loading under the wrong tokenizer class via auto-tokenizer mapping." + "reason": "Different failures and fixes: CodeLlama decoder stripping spaces on decode vs. DeepSeek checkpoint-to-tokenizer-class selection in auto tokenization." } ] }, @@ -23615,12 +23852,12 @@ "soft_pairs": [ "pull_request:46603|pull_request:46604" ], - "prepared_review_unit_hash": "9c5321306beef45c6e5bf74a9ca145c1e8c1502d651c78a36b6533442c02c640", + "prepared_review_unit_hash": "b0dee0e19f87d47edd1c689847fa8d395df21f1b801f3e141a86bd42b11975a0", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 5494, + "serialized_chars": 5495, "estimated_input_tokens": 1374, "estimated_eval_tokens": 3004 }, @@ -23628,7 +23865,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 5494, + "serialized_chars": 5495, "estimated_input_tokens": 1374, "estimated_eval_tokens": 3004 }, @@ -23637,18 +23874,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two related Mistral PRs touch the same tokenizer area, but they implement different changes: one moves MistralConverter into the Mistral integration package, while the other adds tekken tokenizer detection/conversion/save utilities. They share files and subsystem, but not the same concrete fix.", - "confidence": 0.93, + "summary": "Both PRs touch the Mistral tokenizer integration, but they address different changes: #46603 is a refactor that moves `MistralConverter` into the new integration package, while #46604 adds tekken tokenizer detection, conversion, and save utilities. They are related codebase areas, not the same fix/change.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46604 is the better representative because it is the more substantial feature change in the Mistral tokenizer path, adding tekken detection and save/conversion utilities rather than just relocating an existing converter.", + "canonical_pr_reason": "PR #46604 is the more substantive and representative Mistral change: it introduces new tokenizer behavior and utilities, whereas #46603 is mainly a code relocation/refactor.", "best_issue_reason": null, - "best_pr_reason": "PR 46604 best captures the cluster\u2019s main user-facing functionality around Mistral tokenizer support. PR 46603 is a smaller refactor/split of converter code and is not the clearest anchor for the underlying change set.", + "best_pr_reason": "PR #46604 best represents the cluster because it covers the broader functional work in the Mistral integration and is more likely to be the primary user-facing change.", "soft_edge_verdicts": [ { "left": "pull_request:46603", "right": "pull_request:46604", "accept": false, - "reason": "Different concrete changes: 46603 is a code move/refactor for MistralConverter, while 46604 implements tekken tokenizer detection, conversion, and save utilities. Same subsystem and files, but not the same mergeable PR." + "reason": "Reject: they share the Mistral integration area and files, but #46603 is a package move/refactor while #46604 implements tekken tokenizer detection/conversion/save utilities. Different concrete changes, not one underlying bug or mergeable single PR." } ] }, @@ -23670,56 +23907,48 @@ "source": "llm", "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", - "cluster_id": "cluster-46741-3", + "cluster_id": "cluster-46741-2", "nodes": [ "pull_request:46741", - "pull_request:46743", "pull_request:46748" ], "soft_pairs": [ - "pull_request:46741|pull_request:46748", - "pull_request:46743|pull_request:46748" + "pull_request:46741|pull_request:46748" ], - "prepared_review_unit_hash": "8e5c22b0746e8049346750ede51fa34a182b09d3e774e8eef2fd1def0e26636a", + "prepared_review_unit_hash": "10614e98502db49184d921c7bf8bb3ac912ff1424cf806e251a5d4e898bed70a", "component_budget": { - "node_count": 3, - "item_count": 3, - "soft_pair_count": 2, - "serialized_chars": 6685, - "estimated_input_tokens": 1672, - "estimated_eval_tokens": 3600 + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4434, + "estimated_input_tokens": 1109, + "estimated_eval_tokens": 2474 }, "budget": { - "node_count": 3, - "item_count": 3, - "soft_pair_count": 2, - "serialized_chars": 6685, - "estimated_input_tokens": 1672, - "estimated_eval_tokens": 3600 + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4434, + "estimated_input_tokens": 1109, + "estimated_eval_tokens": 2474 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The three pull requests are about different code paths: Zamba2/Nemotron-H chunked-prefill handling, CPU fallback for grouped_mm, and offloaded cache prefetching for hybrid models. They are not duplicates of the same underlying fix.", - "confidence": 0.95, + "summary": "These two PRs address different bugs in different code paths: one fixes Mamba2-style chunked-prefill/speculative decoding in specific model implementations, the other fixes cache prefetch/device-mismatch behavior for offloaded hybrid models. They are not duplicates.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #46748 is the clearest self-contained bug fix in the set: it targets cache prefetch behavior in hybrid models and includes a test file, making it the best representative artifact here.", + "canonical_pr_reason": "PR #46748 is the better canonical representative because it has a single, focused fix in cache.prefetch plus tests, making it more self-contained than the model-specific chunked-prefill changes in #46741.", "best_issue_reason": null, - "best_pr_reason": "PR #46748 is the most representative overall because it fixes a concrete runtime mismatch in cache offloading, with an accompanying test, and is easier to interpret as a standalone bug fix than the broader model or MoE changes.", + "best_pr_reason": "PR #46748 is the strongest standalone PR in the cluster: it targets one concrete cache/offload bug with a narrow implementation surface and validation coverage.", "soft_edge_verdicts": [ { "left": "pull_request:46741", "right": "pull_request:46748", "accept": false, - "reason": "Different underlying problems and code paths: #46741 fixes Mamba2 mixer chunked-prefill/speculative decoding in specific model implementations, while #46748 fixes cache prefetch/offload device mismatch in hybrid models." - }, - { - "left": "pull_request:46743", - "right": "pull_request:46748", - "accept": false, - "reason": "These address unrelated subsystems: #46743 changes grouped_mm CPU fallback logic in MoE, while #46748 changes cache prefetching for hybrid-model offloaded caches." + "reason": "Different underlying problems: #46741 fixes chunked-prefill/speculative decoding behavior in Nemotron-H/Zamba2 Mamba2 mixer code, while #46748 fixes offloaded cache prefetch/device mismatch in cache_utils for hybrid models." } ] }, diff --git a/analysis/current/manifest.json b/analysis/current/manifest.json index 5c36507843979d9c0634667f25aa22d358b6cff0..e2432aa3270b897f8dc4703d6ff22a829d1d3c22 100644 --- a/analysis/current/manifest.json +++ b/analysis/current/manifest.json @@ -1,8 +1,8 @@ { - "analysis_id": "hybrid-model-20260619t060024z", + "analysis_id": "hybrid-model-20260619t120025z", "archived_artifacts": { - "hybrid": "snapshots/20260619T060024Z/analysis-runs/hybrid-model-20260619t060024z/analysis-report-hybrid.json", - "hybrid_reviews": "snapshots/20260619T060024Z/analysis-runs/hybrid-model-20260619t060024z/analysis-report-hybrid.llm-reviews.json" + "hybrid": "snapshots/20260619T120025Z/analysis-runs/hybrid-model-20260619t120025z/analysis-report-hybrid.json", + "hybrid_reviews": "snapshots/20260619T120025Z/analysis-runs/hybrid-model-20260619t120025z/analysis-report-hybrid.llm-reviews.json" }, "artifacts": { "hybrid": "analysis/current/analysis-report-hybrid.json", @@ -10,9 +10,9 @@ }, "channel": "canonical", "model": null, - "published_at": "2026-06-19T06:10:39Z", + "published_at": "2026-06-19T12:09:57Z", "repo": "huggingface/transformers", "schema_version": 1, - "snapshot_id": "20260619T060024Z", + "snapshot_id": "20260619T120025Z", "variant": "hybrid" } diff --git a/snapshots/20260619T120025Z/analysis-runs/hybrid-model-20260619t120025z/analysis-report-hybrid.json b/snapshots/20260619T120025Z/analysis-runs/hybrid-model-20260619t120025z/analysis-report-hybrid.json new file mode 100644 index 0000000000000000000000000000000000000000..aa53524ac62c8adb7bc3f5a8cd7a04818556691e --- /dev/null +++ b/snapshots/20260619T120025Z/analysis-runs/hybrid-model-20260619t120025z/analysis-report-hybrid.json @@ -0,0 +1,2220 @@ +{ + "schema_version": "1.0", + "repo": "huggingface/transformers", + "snapshot_id": "20260619T120025Z", + "generated_at": "2026-06-19T12:09:55Z", + "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": 0.811, + "size_similarity": 0.9, + "file_overlap": 0.75, + "area_overlap": 0.882, + "patch_similarity": 0.648, + "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": [ + [ + 737, + 744 + ] + ], + "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": [ + [ + 301, + 307 + ], + [ + 742, + 748 + ], + [ + 1360, + 1366 + ] + ], + "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, + 150 + ] + ], + "right_ranges": [ + [ + 1, + 126 + ] + ] + }, + { + "filename": "src/transformers/models/minicpm3/modeling_minicpm3.py", + "left_ranges": [ + [ + 1, + 575 + ] + ], + "right_ranges": [ + [ + 1, + 522 + ] + ] + }, + { + "filename": "src/transformers/models/minicpm3/modular_minicpm3.py", + "left_ranges": [ + [ + 1, + 354 + ] + ], + "right_ranges": [ + [ + 1, + 342 + ] + ] + }, + { + "filename": "tests/models/minicpm3/test_modeling_minicpm3.py", + "left_ranges": [ + [ + 1, + 135 + ] + ], + "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-46562-2", + "summary": "Cluster of 2 related pull requests linked by soft_similarity.", + "status": "open", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 46562, + "issue_numbers": [], + "pr_numbers": [ + 46562, + 46658 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 46562, + "right_pr_number": 46658, + "code_similarity": 0.154, + "size_similarity": 0.38, + "file_overlap": 0.024, + "area_overlap": 0.0, + "patch_similarity": 0.468, + "shared_filenames": [ + "src/transformers/models/auto/auto_mappings.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/models/auto/auto_mappings.py", + "left_ranges": [ + [ + 441, + 447 + ], + [ + 591, + 598 + ], + [ + 823, + 829 + ], + [ + 883, + 889 + ], + [ + 1082, + 1088 + ], + [ + 1205, + 1211 + ] + ], + "right_ranges": [ + [ + 362, + 368 + ], + [ + 795, + 801 + ] + ] + } + ] + } + ] + }, + { + "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-46562-2", + "canonical_pr_number": 46562, + "duplicate_pr_numbers": [ + 46658 + ], + "target_issue_number": null, + "reason": "PRs in cluster-46562-2 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/20260619T120025Z/analysis-runs/hybrid-model-20260619t120025z/analysis-report-hybrid.llm-reviews.json b/snapshots/20260619T120025Z/analysis-runs/hybrid-model-20260619t120025z/analysis-report-hybrid.llm-reviews.json new file mode 100644 index 0000000000000000000000000000000000000000..5e866d53fa9c93b68bd20275fd8b6c9b9693b0c5 --- /dev/null +++ b/snapshots/20260619T120025Z/analysis-runs/hybrid-model-20260619t120025z/analysis-report-hybrid.llm-reviews.json @@ -0,0 +1,23963 @@ +{ + "schema_version": "1.0", + "repo": "huggingface/transformers", + "snapshot_id": "20260619T120025Z", + "generated_at": "2026-06-19T12:09:55Z", + "model": "gpt-5.4-mini?service_tier=flex", + "reviews": [ + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "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:24643|issue:30277", + "issue:34928|issue:44805", + "issue:45859|issue:45902", + "issue:43638|issue:45161", + "issue:34928|issue:36331", + "issue:43596|issue:45161", + "issue:43716|issue:45507", + "issue:44292|issue:45237", + "issue:45874|issue:46294", + "issue:39290|issue:45081", + "issue: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": "ca75d09a52212709ce0521add24d0d42711a16545bfe67df427db791afd42007", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 62, + "serialized_chars": 45723, + "estimated_input_tokens": 11431, + "estimated_eval_tokens": 23118 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a grab-bag of unrelated bug reports. The clearest duplicate cluster is the DeepSpeed ZeRO-3 Bert index-out-of-bounds reports (43596/43638); a second likely duplicate pair is the Qwen3.5 MoE rotary-embedding/device-materialization reports (45859/45902). The torch 2.4.1 incompatibility reports (45800/46515) also look duplicate-like, but most other pairs are just broad topical similarity.", + "confidence": 0.82, + "canonical_issue_reason": "issue:43596 is the strongest representative of the cluster: it states the concrete ZeRO-3 \"index 0 out of bounds\" failure clearly, and it has the highest-confidence soft duplicate match to 43638 with essentially the same error path and initialization context.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43596 is the best issue candidate overall because it is specific, well-scoped, and sits at the center of the only clearly duplicated bug pair in the set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44322", + "right": "issue:45464", + "accept": false, + "reason": "Different failures: Qwen3_5Config missing num_attention_heads vs chat/completions streaming error. Same family, not the same bug." + }, + { + "left": "issue:45237", + "right": "issue:45685", + "accept": false, + "reason": "Both are MoE/GPU-related, but one is AMD GPU execution and the other is an MPS histogram op issue. Different code paths." + }, + { + "left": "issue:43716", + "right": "issue:45237", + "accept": false, + "reason": "Mistral-3 image dtype mismatch and GPT-OSS AMD GPU failure are unrelated problems." + }, + { + "left": "issue:43828", + "right": "issue:45237", + "accept": false, + "reason": "Autocast dtype mismatch in Phi-tiny-MoE is not the same issue as GPT-OSS on AMD GPUs." + }, + { + "left": "issue:45237", + "right": "issue:46468", + "accept": false, + "reason": "GPT-OSS AMD GPU failure and a broken Nemotron example are too different to be duplicates." + }, + { + "left": "issue:45237", + "right": "issue:45507", + "accept": false, + "reason": "AMD GPU failure in GPT-OSS and GraniteMoEHybrid invalid method are different model/code-path issues." + }, + { + "left": "issue:30277", + "right": "issue:35545", + "accept": false, + "reason": "DeepSpeed ZeRO-3 collective mismatch and ModernBERT ONNX export error are unrelated." + }, + { + "left": "issue:35545", + "right": "issue:42915", + "accept": false, + "reason": "Both are model-export/config issues, but ModernBERT ONNX export and Qwen3Moe FP8 failure are different bugs." + }, + { + "left": "issue:35545", + "right": "issue:41720", + "accept": false, + "reason": "ModernBERT ONNX export and Qwen3 auto device mapping cudaErrorAssert are unrelated." + }, + { + "left": "issue:45901", + "right": "issue:46015", + "accept": false, + "reason": "Table-question-answering crash and outdated RAG examples are different issues; one is a runtime bug, the other docs." + }, + { + "left": "issue:44509", + "right": "issue:44512", + "accept": false, + "reason": "Both are docs regressions after v5, but they cover different removed topics and should not be merged as the same change." + }, + { + "left": "issue:35545", + "right": "issue:36010", + "accept": false, + "reason": "ModernBERT ONNX export and GenerationMixin import error are different failure modes." + }, + { + "left": "issue:44918", + "right": "issue:46165", + "accept": false, + "reason": "TRL SFT embedding unpacking and resume-training with trust_remote_code are different training issues." + }, + { + "left": "issue:44805", + "right": "issue:45823", + "accept": false, + "reason": "Mask-shape IndexError and Gemma4 PLE device mismatch are different symptoms and code paths." + }, + { + "left": "issue:43638", + "right": "issue:44805", + "accept": false, + "reason": "Both involve tensor/index errors, but 43638 is ZeRO-3/Bert init while 44805 is a mask/tensor shape mismatch." + }, + { + "left": "issue:43638", + "right": "issue:46165", + "accept": false, + "reason": "ZeRO-3 Bert init error and resume-training/remote-code problem are unrelated." + }, + { + "left": "issue:43596", + "right": "issue:44805", + "accept": false, + "reason": "Different IndexError classes: 43596 is ZeRO-3 empty-dimension init, 44805 is mask-vs-tensor shape mismatch." + }, + { + "left": "issue:44805", + "right": "issue:46294", + "accept": false, + "reason": "Mask/tensor shape mismatch and Qwen3.5 video token/feature mismatch are different multimodal bugs." + }, + { + "left": "issue:44805", + "right": "issue:45161", + "accept": false, + "reason": "Index/shape mismatch in one issue and GPT-OSS only-TP failure in the other are unrelated." + }, + { + "left": "issue:46165", + "right": "issue:46468", + "accept": false, + "reason": "Resume-training with remote code and broken Nemotron example are not the same bug." + }, + { + "left": "issue:44560", + "right": "issue:46165", + "accept": false, + "reason": "Qwen3-vl video StopIteration and resume-training remote-code failure are unrelated." + }, + { + "left": "issue:43596", + "right": "issue:46165", + "accept": false, + "reason": "ZeRO-3 Bert init error and remote-code resume failure are unrelated." + }, + { + "left": "issue:34634", + "right": "issue:35545", + "accept": false, + "reason": "BarkProcessor voice_preset and ModernBERT ONNX export are unrelated." + }, + { + "left": "issue:30277", + "right": "issue:43638", + "accept": false, + "reason": "Both involve DeepSpeed, but collective mismatch and Bert ZeRO-3 index error are different failures." + }, + { + "left": "issue:43596", + "right": "issue:43638", + "accept": true, + "reason": "Same underlying ZeRO-3 initialization bug: both report the same out-of-bounds index error on Bert models during deepspeed zero3 init." + }, + { + "left": "issue:43039", + "right": "issue:45161", + "accept": false, + "reason": "Liger Kernel cross-entropy being called and GPT-OSS TP failure are not the same issue." + }, + { + "left": "issue:43975", + "right": "issue:44457", + "accept": false, + "reason": "Incorrect detokenization and LoRA merge/reload output mismatch are different problems." + }, + { + "left": "issue:44292", + "right": "issue:44457", + "accept": false, + "reason": "Qwen-3-8B NVFP4 runtime error and LoRA merge/save inconsistency are unrelated." + }, + { + "left": "issue:44841", + "right": "issue:46661", + "accept": false, + "reason": "Processor failure for Voxtral and remote-code model_class config_class error are different loading paths." + }, + { + "left": "issue:24643", + "right": "issue:30277", + "accept": false, + "reason": "Both mention DeepSpeed training, but weight dimensionality and collective mismatch are distinct bugs." + }, + { + "left": "issue:34928", + "right": "issue:44805", + "accept": false, + "reason": "FSDP activation-checkpointing size mismatch and mask-shape IndexError are not the same failure." + }, + { + "left": "issue:45859", + "right": "issue:45902", + "accept": true, + "reason": "Same underlying Qwen3.5 MoE rotary-embedding/device handling bug: both center on Qwen3_5MoeTextRotaryEmbedding and its inv_freq behavior under device materialization/offload." + }, + { + "left": "issue:43638", + "right": "issue:45161", + "accept": false, + "reason": "ZeRO-3 Bert index error and GPT-OSS TP failure are unrelated." + }, + { + "left": "issue:34928", + "right": "issue:36331", + "accept": false, + "reason": "FSDP checkpointing shape mismatch and CustomTrainer.compute_loss signature mismatch are different bugs." + }, + { + "left": "issue:43596", + "right": "issue:45161", + "accept": false, + "reason": "ZeRO-3 Bert init error and GPT-OSS TP issue are unrelated." + }, + { + "left": "issue:43716", + "right": "issue:45507", + "accept": false, + "reason": "Mistral-3 image dtype mismatch and GraniteMoEHybrid invalid method are different issues." + }, + { + "left": "issue:44292", + "right": "issue:45237", + "accept": false, + "reason": "Qwen-3-8B NVFP4 error and GPT-OSS AMD GPU failure are different model/runtime problems." + }, + { + "left": "issue:45874", + "right": "issue:46294", + "accept": false, + "reason": "Gemma4 reverse embedding lookup and Qwen3.5 video token/feature mismatch are unrelated." + }, + { + "left": "issue:39290", + "right": "issue:45081", + "accept": false, + "reason": "Gemma3 sliding_window_pattern/vLLM error and Mistral tokenizer regex crash are different code paths." + }, + { + "left": "issue:45823", + "right": "issue:45874", + "accept": false, + "reason": "Both are Gemma4-related, but device mismatch during forward and expensive reverse embedding lookup are different bugs." + }, + { + "left": "issue:45800", + "right": "issue:46515", + "accept": true, + "reason": "Both report the same torch 2.4.1 incompatibility; 46515 is a broader restatement of the version-compatibility problem in 45800." + }, + { + "left": "issue:43425", + "right": "issue:44292", + "accept": false, + "reason": "Torch 2.10 incompatibility and Qwen-3-8B NVFP4 runtime error are unrelated." + }, + { + "left": "issue:43827", + "right": "issue:44512", + "accept": false, + "reason": "Both are docs regressions, but one is pipeline-task references and the other is the removed transformers run command." + }, + { + "left": "issue:43828", + "right": "issue:44292", + "accept": false, + "reason": "Autocast dtype mismatch and Qwen-3-8B NVFP4 runtime error are not the same issue." + }, + { + "left": "issue:43541", + "right": "issue:43716", + "accept": false, + "reason": "Mixtral grouped_mm torch dynamo tracing error and Mistral-3 image dtype mismatch are different failures." + }, + { + "left": "issue:43716", + "right": "issue:46468", + "accept": false, + "reason": "Mistral-3 image dtype mismatch and broken Nemotron example are unrelated." + }, + { + "left": "issue:43716", + "right": "issue:45685", + "accept": false, + "reason": "Mistral-3 image preprocessing dtype mismatch and MPS histogram op error are different problems." + }, + { + "left": "issue:44292", + "right": "issue:44918", + "accept": false, + "reason": "Qwen-3-8B NVFP4 runtime error and TRL SFT embedding unpacking failure are unrelated." + }, + { + "left": "issue:43716", + "right": "issue:44292", + "accept": false, + "reason": "Mistral-3 image dtype mismatch and Qwen-3-8B NVFP4 runtime error are different bugs." + }, + { + "left": "issue:45507", + "right": "issue:45685", + "accept": false, + "reason": "GraniteMoEHybrid invalid method and MPS histogram_mps not implemented are unrelated." + }, + { + "left": "issue:44918", + "right": "issue:45161", + "accept": false, + "reason": "TRL SFT embedding unpacking and GPT-OSS only-TP failure are different issues." + }, + { + "left": "issue:43039", + "right": "issue:44918", + "accept": false, + "reason": "Liger Kernel cross_entropy being called and TRL SFT embedding unpacking are unrelated." + }, + { + "left": "issue:41762", + "right": "issue:43638", + "accept": false, + "reason": "Both are ZeRO-3 index errors, but Gemma3 conditional generation and Bert init are not the same concrete bug." + }, + { + "left": "issue:43828", + "right": "issue:45685", + "accept": false, + "reason": "Autocast dtype mismatch and MPS histogram op error are different runtime issues." + }, + { + "left": "issue:44918", + "right": "issue:45685", + "accept": false, + "reason": "TRL SFT embedding unpacking and MPS histogram op error are unrelated." + }, + { + "left": "issue:43039", + "right": "issue:43828", + "accept": false, + "reason": "Liger Kernel cross_entropy usage and autocast dtype mismatch are different problems." + }, + { + "left": "issue:30277", + "right": "issue:36331", + "accept": false, + "reason": "DeepSpeed collective mismatch and compute_loss signature mismatch are unrelated." + }, + { + "left": "issue:34928", + "right": "issue:36010", + "accept": false, + "reason": "FSDP activation-checkpointing mismatch and GenerationMixin import error are different issues." + }, + { + "left": "issue:45237", + "right": "issue:46015", + "accept": false, + "reason": "GPT-OSS AMD GPU failure and outdated RAG examples are unrelated." + }, + { + "left": "issue:43039", + "right": "issue:44805", + "accept": false, + "reason": "Liger Kernel cross_entropy issue and mask-shape IndexError are different bugs." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The analyst summary is well grounded in the packet and stays conservative. The accepted soft edges are supported by closely matching titles and issue themes: 43596/43638 are near-duplicate ZeRO-3 Bert index errors, 45859/45902 are clearly the same Qwen3.5 MoE rotary-embedding/materialization bug family, and 45800/46515 are plausibly the same torch 2.4.1 incompatibility report. The remaining rejections are appropriately cautious 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": 23, + "review_unit_index": 2, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "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:44387", + "issue:44512", + "issue:44560", + "issue:44589", + "issue:44841", + "issue:44863", + "issue:45005", + "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:43872|issue:46304", + "issue:44292|issue:46015", + "issue:39290|issue:43531", + "issue:41093|issue:43638", + "issue:43792|issue:45901", + "issue:43854|issue:43866", + "issue:43425|issue:45070", + "issue:43975|issue:45901", + "issue: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:36010|issue:39290", + "issue:43541|issue:43828", + "issue:44841|issue:45084", + "issue:43404|issue:43866", + "issue:41093|issue:41762", + "issue:45399|issue:46661", + "issue:43572|issue:45070", + "issue:45800|issue:45823", + "issue:43296|issue:43366", + "issue:44387|issue:45005" + ], + "prepared_review_unit_hash": "43c5730fe49c0898215644f682f2f8eedc4a8a40f20628fdc61d6ecbe4904abc", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 41, + "serialized_chars": 39793, + "estimated_input_tokens": 9949, + "estimated_eval_tokens": 20154 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly heterogeneous, with only a couple of true duplicate-looking subthreads. The clearest match is the AutoTokenizer wrong-token-ID regression, and there is also a plausible duplicate pair around the DeepSpeed ZeRO-3 index-out-of-bounds load failure.", + "confidence": 0.77, + "canonical_issue_reason": "Issue 45920 is the broadest and most representative of the tokenizer regression family: it describes the wrong-token-ID bug across multiple model families, subsuming the narrower Granite-only report.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45920 is the best single issue to anchor the cluster because it is the most general, clearly regression-focused, and spans the widest set of affected models.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43866", + "right": "issue:46402", + "accept": false, + "reason": "Different model families and failures: corrupted checkpoint vs. missing text-encoder weights." + }, + { + "left": "issue:45800", + "right": "issue:46304", + "accept": false, + "reason": "Both mention compatibility, but they are different incompatibilities and different code paths." + }, + { + "left": "issue:43296", + "right": "issue:46304", + "accept": false, + "reason": "Unrelated symptoms: vLLM load failure for PaddleOCR-VL vs. PyTorch USE_DISTRIBUTED=0 support breakage." + }, + { + "left": "issue:41762", + "right": "issue:43596", + "accept": true, + "reason": "Same ZeRO-3 load-time index-out-of-bounds failure; only the model types differ." + }, + { + "left": "issue:39290", + "right": "issue:41720", + "accept": false, + "reason": "Different problems: missing config attribute in vLLM vs. CUDA assert with auto device mapping." + }, + { + "left": "issue:43901", + "right": "issue:44512", + "accept": false, + "reason": "Both are docs-related, but they reference different outdated behaviors and do not share the same bug." + }, + { + "left": "issue:44589", + "right": "issue:46304", + "accept": false, + "reason": "Completely different failures: missing Float8 storage vs. distributed build/runtime incompatibility." + }, + { + "left": "issue:45237", + "right": "issue:45901", + "accept": false, + "reason": "Different subsystems and unrelated crash modes." + }, + { + "left": "issue:43039", + "right": "issue:44560", + "accept": false, + "reason": "Unrelated: Liger Kernel cross-entropy path vs. video chat-template StopIteration." + }, + { + "left": "issue:45823", + "right": "issue:46294", + "accept": false, + "reason": "Different model families and different device/chat-template failures." + }, + { + "left": "issue:43975", + "right": "issue:46015", + "accept": false, + "reason": "Different scope: detokenization regression vs. outdated RAG examples." + }, + { + "left": "issue:44841", + "right": "issue:45823", + "accept": false, + "reason": "Different concrete failures: processor load bug vs. device-mismatch bug." + }, + { + "left": "issue:44863", + "right": "issue:46015", + "accept": false, + "reason": "NemotronH checkpoint loading is unrelated to RAG docs." + }, + { + "left": "issue:45020", + "right": "issue:46304", + "accept": false, + "reason": "Remote-code regression and distributed build support are different bugs." + }, + { + "left": "issue:45812", + "right": "issue:45920", + "accept": true, + "reason": "Same underlying AutoTokenizer wrong-token-ID regression; 45920 is the broader report covering more models." + }, + { + "left": "issue:43866", + "right": "issue:44863", + "accept": false, + "reason": "Corrupted checkpoint vs. model-loading support issue; not the same bug." + }, + { + "left": "issue:43872", + "right": "issue:46304", + "accept": false, + "reason": "bitsandbytes constructor breakage is unrelated to USE_DISTRIBUTED=0 support." + }, + { + "left": "issue:44292", + "right": "issue:46015", + "accept": false, + "reason": "Different topics and failure modes." + }, + { + "left": "issue:39290", + "right": "issue:43531", + "accept": false, + "reason": "Both mention sliding-window, but the config-attribute error and the Qwen3-MoE sliding-window issue are not clearly the same bug." + }, + { + "left": "issue:41093", + "right": "issue:43638", + "accept": false, + "reason": "Both are IndexError reports, but the mask-shape mismatch is a different failure from the ZeRO-3 empty-tensor issue." + }, + { + "left": "issue:43792", + "right": "issue:45901", + "accept": false, + "reason": "Different tasks and unrelated crash causes." + }, + { + "left": "issue:43854", + "right": "issue:43866", + "accept": false, + "reason": "Different model load failures with no shared root cause." + }, + { + "left": "issue:43425", + "right": "issue:45070", + "accept": false, + "reason": "Version incompatibility vs. pydantic/config-field regression; unrelated." + }, + { + "left": "issue:43975", + "right": "issue:45901", + "accept": false, + "reason": "Detokenization regression and table QA crash are unrelated." + }, + { + "left": "issue:43596", + "right": "issue:45800", + "accept": false, + "reason": "ZeRO-3 load-time IndexError and torch/transformers version incompatibility are different issues." + }, + { + "left": "issue:43792", + "right": "issue:45070", + "accept": false, + "reason": "Whisper runtime failure is unrelated to the pydantic PretrainedConfig regression." + }, + { + "left": "issue:45362", + "right": "issue:45464", + "accept": false, + "reason": "Both involve Qwen3.5, but the chat crash and streaming API failure are not clearly the same code-path bug." + }, + { + "left": "issue:44292", + "right": "issue:45901", + "accept": false, + "reason": "Unrelated model-specific error vs. table QA task crash." + }, + { + "left": "issue:39290", + "right": "issue:45823", + "accept": false, + "reason": "Different model/config problems; no shared underlying bug is evident." + }, + { + "left": "issue:45464", + "right": "issue:45685", + "accept": false, + "reason": "Chat/streaming failure and MPS MoE histogram error are unrelated." + }, + { + "left": "issue:43541", + "right": "issue:45685", + "accept": false, + "reason": "Both are numeric/runtime issues, but in different kernels and backends." + }, + { + "left": "issue:36010", + "right": "issue:39290", + "accept": false, + "reason": "ImportError for GenerationMixin is unrelated to the Gemma3 sliding_window_pattern error." + }, + { + "left": "issue:43541", + "right": "issue:43828", + "accept": false, + "reason": "Both are dtype-related, but they affect different models and ops." + }, + { + "left": "issue:44841", + "right": "issue:45084", + "accept": false, + "reason": "Processor failure and template-compilation TypeError are different bugs." + }, + { + "left": "issue:43404", + "right": "issue:43866", + "accept": false, + "reason": "Untied lm_head weights and corrupted checkpoint are unrelated." + }, + { + "left": "issue:41093", + "right": "issue:41762", + "accept": false, + "reason": "Same general IndexError wording, but the mask-size mismatch is not the ZeRO-3 empty-tensor loading bug." + }, + { + "left": "issue:45399", + "right": "issue:46661", + "accept": false, + "reason": "Both mention remote-code-ish flows, but the fallback gating bug and missing config_class AttributeError are different code paths." + }, + { + "left": "issue:43572", + "right": "issue:45070", + "accept": false, + "reason": "Missing pad_token_idx and pydantic PretrainedConfig field regression are unrelated." + }, + { + "left": "issue:45800", + "right": "issue:45823", + "accept": false, + "reason": "Torch/version compatibility and Gemma4 device mismatch are separate problems." + }, + { + "left": "issue:43296", + "right": "issue:43366", + "accept": false, + "reason": "Different model support gaps: PaddleOCR-VL in vLLM vs. GGUF gpt-oss support." + }, + { + "left": "issue:44387", + "right": "issue:45005", + "accept": false, + "reason": "Int4 OOM regression and tied-weights translation-model issues are unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The two accepted soft pairs are reasonably supported by the issue titles, and the rest of the pairwise rejections avoid overstating similarity." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 3, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "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:44479", + "issue:44492", + "issue:44512", + "issue:44530", + "issue:44589", + "issue:44805", + "issue:44863", + "issue:44918", + "issue:44960", + "issue:45005", + "issue:45020", + "issue:45070", + "issue:45161", + "issue:45183", + "issue:45399", + "issue:45464", + "issue:45507", + "issue:45632", + "issue:45698", + "issue:45759", + "issue:45800", + "issue:45902", + "issue:46294", + "issue:46463", + "issue:46489", + "issue:46491", + "issue:46640", + "issue:46661" + ], + "soft_pairs": [ + "issue:44805|issue:44918", + "issue:43541|issue:45161", + "issue:43828|issue:45464", + "issue:43366|issue:45464", + "issue:44291|issue:45070", + "issue:43531|issue:45070", + "issue:43541|issue:44589", + "issue:43716|issue:46294", + "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:43716|issue:44918", + "issue:39186|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:44492|issue:44512", + "issue:44918|issue:45507", + "issue:45183|issue:46463", + "issue:46489|issue:46491", + "issue:43975|issue:44292", + "issue:44292|issue:45507" + ], + "prepared_review_unit_hash": "423d94ffa025f77241acf84406bd4459c157fd746124a4eada6eb16e84772efa", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 31, + "serialized_chars": 36258, + "estimated_input_tokens": 9065, + "estimated_eval_tokens": 18386 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a heterogeneous issue set with a few near-duplicate themes around Qwen video validation and input-validation error messaging. I only accept soft edges where the titles point to the same concrete code path; most pairs are just the same broad subsystem or adjacent version regressions, not true duplicates.", + "confidence": 0.72, + "canonical_issue_reason": "No single issue truly represents the whole cluster; if one must be chosen, issue:45020 is the broadest version-regression umbrella among the set.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45020 is the most generally representative issue because it spans recent transformers-version breakage rather than one narrow model symptom, though it is still not a duplicate of most others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44805", + "right": "issue:44918", + "accept": false, + "reason": "Both are shape/runtime failures, but they affect different code paths: generic mask indexing vs TRL SFT embedding unpacking." + }, + { + "left": "issue:43541", + "right": "issue:45161", + "accept": false, + "reason": "Different problems: grouped_mm tracing failure in Mixtral vs GPT-OSS MoE tensor-parallel behavior." + }, + { + "left": "issue:43828", + "right": "issue:45464", + "accept": false, + "reason": "Unrelated symptoms and paths: autocast dtype mismatch in Phi-tiny-MoE vs chat/completions streaming failure in Qwen3.5." + }, + { + "left": "issue:43366", + "right": "issue:45464", + "accept": false, + "reason": "GGUF gpt-oss architecture support and Qwen3.5 streaming inference are different issues." + }, + { + "left": "issue:44291", + "right": "issue:45070", + "accept": false, + "reason": "Both are v5 compatibility regressions, but one is init_empty_weights argument handling and the other is PretrainedConfig/pydantic field behavior." + }, + { + "left": "issue:43531", + "right": "issue:45070", + "accept": false, + "reason": "Sliding-window handling for Qwen3-MoE is unrelated to the pydantic config break in v5.4.0." + }, + { + "left": "issue:43541", + "right": "issue:44589", + "accept": false, + "reason": "Different low-level runtime failures: dynamo tracing/grouped_mm vs Float8 storage lookup." + }, + { + "left": "issue:43716", + "right": "issue:46294", + "accept": false, + "reason": "Both are multimodal, but one is dtype mismatch in image preprocessing and the other is Qwen3.5 video-token alignment." + }, + { + "left": "issue:45020", + "right": "issue:45800", + "accept": false, + "reason": "Both mention broad compatibility regressions, but the concrete failures are different: remote_code loading vs torch/transformers version incompatibility." + }, + { + "left": "issue:41093", + "right": "issue:43596", + "accept": false, + "reason": "Both are IndexErrors, but the causes are different: mask-length mismatch vs DeepSpeed ZeRO3 init with empty tensors." + }, + { + "left": "issue:44479", + "right": "issue:46294", + "accept": true, + "reason": "Same underlying Qwen video-input regression: feature/token mismatch across Qwen3.5 video flows." + }, + { + "left": "issue:43054", + "right": "issue:43994", + "accept": false, + "reason": "Different model families and symptoms: siglip2 embedding quality vs SigLIP auto-model/pipeline results." + }, + { + "left": "issue:45759", + "right": "issue:45902", + "accept": false, + "reason": "Different composite-model bugs: config unwrapping in from_config vs uninitialized inv_freq after meta\u2192to_empty materialization." + }, + { + "left": "issue:46640", + "right": "issue:46661", + "accept": false, + "reason": "Flash Attention 2 issue and remote-code model_class/config_class failure are separate loading paths." + }, + { + "left": "issue:43716", + "right": "issue:44918", + "accept": false, + "reason": "Different failure modes: image preprocessor dtype mismatch vs TRL SFT unpacking of embeddings." + }, + { + "left": "issue:39186", + "right": "issue:44918", + "accept": false, + "reason": "FSDP 2-D weight assertion and Qwen3.5 SFT embedding unpacking are unrelated." + }, + { + "left": "issue:40990", + "right": "issue:44960", + "accept": false, + "reason": "Perplexity anomaly on gpt-oss and GLM5 model issue do not share the same bug." + }, + { + "left": "issue:45632", + "right": "issue:45698", + "accept": false, + "reason": "Both involve remote-code loading, but one is cache-path collision and the other is loading the wrong custom module after save_pretrained." + }, + { + "left": "issue:43645", + "right": "issue:45399", + "accept": false, + "reason": "Jupyter custom-model initialization and flash-attn fallback gating are different breakages." + }, + { + "left": "issue:43404", + "right": "issue:45005", + "accept": false, + "reason": "Tied lm_head weights in Mistral3 and translation-model tied weights are not the same concrete bug." + }, + { + "left": "issue:42915", + "right": "issue:45399", + "accept": false, + "reason": "FineGrainedFP8Config failure in Qwen3Moe is unrelated to flash-attn fallback blocking." + }, + { + "left": "issue:44530", + "right": "issue:45005", + "accept": false, + "reason": "PagedAttentionCache linear_attention crash and translation tied-weights regression are unrelated." + }, + { + "left": "issue:43873", + "right": "issue:45005", + "accept": false, + "reason": "Quantization/offloading behavior and tied-weight handling are different code paths." + }, + { + "left": "issue:44360", + "right": "issue:44512", + "accept": false, + "reason": "DSA indexer/ReLU discussion is unrelated to a docs mention of the removed transformers run command." + }, + { + "left": "issue:44863", + "right": "issue:45005", + "accept": false, + "reason": "NemotronH checkpoint loading and translation tied-weights are different model-loading bugs." + }, + { + "left": "issue:44492", + "right": "issue:44512", + "accept": false, + "reason": "A typo in cache strategies is unrelated to documentation about a removed command." + }, + { + "left": "issue:44918", + "right": "issue:45507", + "accept": false, + "reason": "Qwen3.5 SFT embedding unpacking and GraniteMoEHybrid invalid method calls are different failures." + }, + { + "left": "issue:45183", + "right": "issue:46463", + "accept": true, + "reason": "Same validation/error-message family: misleading input validation for decoder/input selection, with T5 as a specific instance." + }, + { + "left": "issue:46489", + "right": "issue:46491", + "accept": false, + "reason": "Both are tokenizer regressions, but they affect different model tokenizers and different symptoms." + }, + { + "left": "issue:43975", + "right": "issue:44292", + "accept": false, + "reason": "DeepSeek tokenizer detokenization and init_empty_weights argument handling are unrelated." + }, + { + "left": "issue:44292", + "right": "issue:45507", + "accept": false, + "reason": "init_empty_weights loading error and GraniteMoEHybrid invalid method calls are different issues." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: it correctly emphasizes that the set is mostly heterogeneous, and the two accepted soft edges are conservative and supported by the issue titles (Qwen video regression; input-validation/error-message regression). The rejected pairs are generally framed as distinct code paths or symptoms rather than true duplicates, which is appropriately cautious." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 4, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "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:43872|issue:44589", + "issue:43828|issue:44928", + "issue:45920|issue:46489", + "issue:43866|issue:44403", + "issue:43054|issue:45701", + "issue:38175|issue:43054", + "issue:43950|issue:45902", + "issue:41553|issue:43883", + "issue:44910|issue:45859", + "issue:43054|issue:43493", + "issue:42947|issue:44387", + "issue:45663|issue:45902", + "issue:41553|issue:43881", + "issue:44403|issue:44451", + "issue:43975|issue:45520", + "issue:44403|issue:44488", + "issue:24643|issue:28530", + "issue:42947|issue:43856", + "issue:43643|issue:45020", + "issue:41762|issue:42915", + "issue:44991|issue:45399", + "issue:43824|issue:43872", + "issue:45685|issue:45823", + "issue:43782|issue:44960", + "issue:43881|issue:43883", + "issue:45020|issue:45823", + "issue:45081|issue:45399", + "issue:44661|issue:45020", + "issue:42491|issue:44368", + "issue:43828|issue:45823" + ], + "prepared_review_unit_hash": "976611d02977e5ef075ab0c324480841c5efd43647f84724d4f268240f2063ab", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 34, + "serialized_chars": 38116, + "estimated_input_tokens": 9529, + "estimated_eval_tokens": 19314 + }, + "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 strongest overlap is the non-persistent-buffer regression pair (43950/44534), which looks like the same underlying bug or a very close re-report, but the packet does not prove duplicate status; the rest should remain separate.", + "confidence": 0.78, + "canonical_issue_reason": "issue:43950 is the clearest representative for the non-persistent-buffer regression: it is more specific and detailed than 44534, which appears to be a shorter re-report of the same behavior.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43950 is the best representative issue for this cluster because it most completely describes the strongest near-duplicate bug report while the other items are clearly different problems.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43950", + "right": "issue:44534", + "accept": true, + "reason": "Same non-persistent-buffer from_pretrained regression; 44534 reads like a shorter re-report of the same underlying bug." + }, + { + "left": "issue:39692", + "right": "issue:43054", + "accept": false, + "reason": "Both mention SigLIP2, but one is a doc/example mismatch plus quantization failure and the other is a text-embedding quality complaint; not the same bug." + }, + { + "left": "issue:43644", + "right": "issue:43950", + "accept": false, + "reason": "Different failures: missing fields with trust_remote_code versus non-persistent-buffer corruption." + }, + { + "left": "issue:36331", + "right": "issue:41762", + "accept": false, + "reason": "Unrelated code paths: custom trainer API break versus DeepSpeed/Gemma loading IndexError." + }, + { + "left": "issue:43872", + "right": "issue:44589", + "accept": false, + "reason": "Different incompatibilities: bitsandbytes Int8Params kwargs versus missing Float8 storage type." + }, + { + "left": "issue:43828", + "right": "issue:44928", + "accept": false, + "reason": "Different models and symptoms: autocast dtype mismatch versus Qwen3.5 position_ids / SDPA fallback instability." + }, + { + "left": "issue:45920", + "right": "issue:46489", + "accept": false, + "reason": "Both involve tokenizer regressions, but 46489 is a model-specific gap in a prior fix, not the same concrete bug as the broader multi-model token-ID regression." + }, + { + "left": "issue:43866", + "right": "issue:44403", + "accept": false, + "reason": "Checkpoint corruption versus generic loading noise; not the same underlying issue." + }, + { + "left": "issue:43054", + "right": "issue:45701", + "accept": false, + "reason": "SigLIP2 embedding quality issues are unrelated to generic tokenizer version-dependent changes." + }, + { + "left": "issue:38175", + "right": "issue:43054", + "accept": false, + "reason": "Same model family, but the titles describe different symptoms and no shared concrete failure mode is evident." + }, + { + "left": "issue:43950", + "right": "issue:45902", + "accept": false, + "reason": "Both involve non-persistent buffers, but 45902 is about uninitialized memory after meta/to_empty materialization, which is a distinct bug." + }, + { + "left": "issue:41553", + "right": "issue:43883", + "accept": false, + "reason": "Different models and failure types: bad AutoTokenizer error for Voxtral versus missing all_tied_weights_keys on Molmo." + }, + { + "left": "issue:44910", + "right": "issue:45859", + "accept": false, + "reason": "Flash-Attention crash from misread position_ids is unrelated to CPU-offload incompatibility in Qwen3_5Moe rotary embedding." + }, + { + "left": "issue:43054", + "right": "issue:43493", + "accept": false, + "reason": "Related SigLIP2 area, but one is a performance complaint and the other is an implementation-vs-JAX discrepancy; not clearly the same bug from the titles alone." + }, + { + "left": "issue:42947", + "right": "issue:44387", + "accept": false, + "reason": "Different training/memory issues: gradient checkpointing with LoRA versus higher reserved memory under int4 quantization." + }, + { + "left": "issue:45663", + "right": "issue:45902", + "accept": false, + "reason": "Gemma-4 FSDP2 shared_kv_states bug is unrelated to Qwen3_5Moe rotary embedding initialization." + }, + { + "left": "issue:41553", + "right": "issue:43881", + "accept": false, + "reason": "Different models and failure types; only a broad loading-context overlap." + }, + { + "left": "issue:44403", + "right": "issue:44451", + "accept": false, + "reason": "Generic loading noise versus inability to load a specific model are too vague to treat as the same bug." + }, + { + "left": "issue:43975", + "right": "issue:45520", + "accept": false, + "reason": "Tokenizer detokenization bug for DeepSeek-Coder v1 is different from the flash_attn import issue on Python 3.13." + }, + { + "left": "issue:44403", + "right": "issue:44488", + "accept": false, + "reason": "Same broad loading theme, but the specific model failure is not the same as the generic noise report." + }, + { + "left": "issue:24643", + "right": "issue:28530", + "accept": false, + "reason": "Completely different training errors: DeepSpeed weight shape versus early-stopping metric selection." + }, + { + "left": "issue:42947", + "right": "issue:43856", + "accept": false, + "reason": "Both are training/memory-related, but they describe different problems and code paths." + }, + { + "left": "issue:43643", + "right": "issue:45020", + "accept": false, + "reason": "A missing-fields object returned by trust_remote_code is not the same as a broader remote_code breakage report." + }, + { + "left": "issue:41762", + "right": "issue:42915", + "accept": false, + "reason": "Different model stacks and error types: Gemma3 loading IndexError versus Qwen3Moe FP8 failure." + }, + { + "left": "issue:44991", + "right": "issue:45399", + "accept": false, + "reason": "Tokenizer loading failure and flash-attn2 fallback gating are unrelated." + }, + { + "left": "issue:43824", + "right": "issue:43872", + "accept": false, + "reason": "Importing a missing Qwen model class is unrelated to bitsandbytes parameter initialization incompatibility." + }, + { + "left": "issue:45685", + "right": "issue:45823", + "accept": false, + "reason": "MPS histogram support for Int and Gemma4 device mismatch are different bugs." + }, + { + "left": "issue:43782", + "right": "issue:44960", + "accept": false, + "reason": "Different models and failure modes: Qwen3VL weight_only loading versus GLM5." + }, + { + "left": "issue:43881", + "right": "issue:43883", + "accept": false, + "reason": "Different models and errors; only both are load-time failures." + }, + { + "left": "issue:45020", + "right": "issue:45823", + "accept": false, + "reason": "Broad remote_code breakage is not the same concrete issue as a Gemma4 device_map forward mismatch." + }, + { + "left": "issue:45081", + "right": "issue:45399", + "accept": false, + "reason": "Tokenizer regex patch crash is unrelated to flash-attn2 fallback checks." + }, + { + "left": "issue:44661", + "right": "issue:45020", + "accept": false, + "reason": "Tokenizer mapping-name edge case is a narrower implementation bug than the broad remote_code regression report." + }, + { + "left": "issue:42491", + "right": "issue:44368", + "accept": false, + "reason": "LoRA model compatibility across HF versions is unrelated to the tie_word_embeddings warning in Qwen3.5 LoRA training." + }, + { + "left": "issue:43828", + "right": "issue:45823", + "accept": false, + "reason": "Different models and distinct runtime failures: dtype mismatch versus device mismatch." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The cluster summary is mostly grounded, but it overstates the evidence by calling 43950/44534 the \"only confirmed duplicate bug\" / \"one true duplicate subgroup.\" The packet only supports that these two issues are very similar and likely duplicates; it does not confirm duplicate status. Please soften that language to reflect uncertainty." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 5, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "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:43827|issue:44509", + "issue:45488|issue:46489", + "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:41553|issue:44991", + "issue:42947|issue:45923", + "issue:36331|issue:41093", + "issue:43940|issue:45313", + "issue:43854|issue:44863", + "issue:44479|issue:44560", + "issue:41553|issue:44488", + "issue:44928|issue:45588", + "issue:41093|issue:44805", + "issue:24643|issue:30064", + "issue:44530|issue:44863", + "issue:44863|issue:46402", + "issue:43366|issue:43828", + "issue:43596|issue:46515", + "issue:43366|issue:43531", + "issue:45325|issue:45902", + "issue:45663|issue:45823" + ], + "prepared_review_unit_hash": "2043b62a7a42937acdbd3aab0729e02ab3ea30d90fd4562f8e032fc1ae7a24fc", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36630, + "estimated_input_tokens": 9158, + "estimated_eval_tokens": 18572 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a heterogeneous cluster of transformer bugs across model loading, tokenizers, DeepSpeed, and docs. I only accept a few pairs that look like the same underlying regression, and reject the rest as merely similar subsystems or generic error patterns.", + "confidence": 0.69, + "canonical_issue_reason": "issue:44291 is the clearest representative of the duplicated `_is_hf_initialized` / `init_empty_weights` loading regression, and it is broader than the bitsandbytes-specific report.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44291 has the most actionable, concrete failure description among the accepted issue duplicates and maps to a specific code-path regression.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45020", + "right": "issue:45356", + "accept": false, + "reason": "Both are tokenizer/model-loading regressions, but they target different concrete failures and model families; too broad to be the same bug." + }, + { + "left": "issue:45715", + "right": "issue:46315", + "accept": false, + "reason": "One is about `convert_ids_to_tokens`, the other `convert_tokens_to_ids`; similar performance theme, different code paths and symptom." + }, + { + "left": "issue:43827", + "right": "issue:44509", + "accept": true, + "reason": "Both report stale docs still referencing pipeline tasks removed in v5; same documentation cleanup issue." + }, + { + "left": "issue:45488", + "right": "issue:46489", + "accept": false, + "reason": "Both involve tokenizer correctness, but the model families and failure mechanisms differ; not enough evidence of one underlying bug." + }, + { + "left": "issue:43404", + "right": "issue:43950", + "accept": false, + "reason": "Distinct regressions: tied weights in a multimodal model vs non-persistent buffer corruption during `from_pretrained()`." + }, + { + "left": "issue:45759", + "right": "issue:46661", + "accept": false, + "reason": "Different loading bugs: composite config unwrapping vs remote-code `config_class` lookup failure." + }, + { + "left": "issue:43872", + "right": "issue:44291", + "accept": true, + "reason": "Both describe the same `_is_hf_initialized` argument regression during model init/loading, with 44291 giving the broader root-cause report." + }, + { + "left": "issue:42915", + "right": "issue:43872", + "accept": false, + "reason": "Different issues: FineGrainedFP8/Qwen3Moe behavior vs unexpected `_is_hf_initialized` argument handling." + }, + { + "left": "issue:45020", + "right": "issue:46661", + "accept": false, + "reason": "Both are generic loading regressions, but they concern different APIs and failure modes." + }, + { + "left": "issue:43792", + "right": "issue:44220", + "accept": true, + "reason": "Both point to the same Whisper feature-extraction / fbank pipeline failure and likely duplicate the same underlying bug." + }, + { + "left": "issue:43257", + "right": "issue:45305", + "accept": false, + "reason": "Both involve DeepSpeed, but one is weight loading for Qwen3 MoE and the other is gradient averaging with GAS; different bugs." + }, + { + "left": "issue:41553", + "right": "issue:44991", + "accept": false, + "reason": "Both are tokenizer-loading problems, but they affect different models and appear to be separate regressions." + }, + { + "left": "issue:42947", + "right": "issue:45923", + "accept": false, + "reason": "Checkpointing-related, but one is PEFT LoRA ineffectiveness and the other is a missing capability flag on a remote-code model." + }, + { + "left": "issue:36331", + "right": "issue:41093", + "accept": false, + "reason": "Different training API and tensor-shape errors; not the same underlying problem." + }, + { + "left": "issue:43940", + "right": "issue:45313", + "accept": false, + "reason": "Both are ZeRO-3 load failures for Qwen variants, but the model targets and missing-weight patterns differ enough to treat separately." + }, + { + "left": "issue:43854", + "right": "issue:44863", + "accept": false, + "reason": "Unrelated loading failures for different architectures." + }, + { + "left": "issue:44479", + "right": "issue:44560", + "accept": false, + "reason": "Both involve Qwen video paths, but one is a broad regression across multiple models and the other is a specific StopIteration case; insufficiently same bug." + }, + { + "left": "issue:44928", + "right": "issue:45588", + "accept": false, + "reason": "Different symptoms and code paths: RLHF NaNs from position IDs vs flash-attention attribute crash on `s_aux=None`." + }, + { + "left": "issue:41093", + "right": "issue:44805", + "accept": true, + "reason": "These are the same generic indexed-tensor vs mask-shape mismatch error pattern and look like duplicate reports of the same bug class." + }, + { + "left": "issue:24643", + "right": "issue:30064", + "accept": false, + "reason": "Training with DeepSpeed vs image processor void-segmentation-map handling are unrelated." + }, + { + "left": "issue:44530", + "right": "issue:44863", + "accept": false, + "reason": "Different subsystems and failures: PagedAttentionCache linear_attention crash vs checkpoint loading on NemotronH." + }, + { + "left": "issue:44863", + "right": "issue:46402", + "accept": false, + "reason": "Both are model-loading issues, but for different models and different failure modes." + }, + { + "left": "issue:43366", + "right": "issue:43828", + "accept": false, + "reason": "Unrelated: GGUF/gpt-oss architecture support vs autocast dtype mismatch on Phi-tiny-MoE." + }, + { + "left": "issue:43596", + "right": "issue:46515", + "accept": false, + "reason": "Index error with ZeRO-3 init and BertModel is not the same as torch 2.4.1 incompatibility." + }, + { + "left": "issue:43366", + "right": "issue:43531", + "accept": false, + "reason": "Different Qwen/GGUF concerns; one is architecture support, the other a sliding_window bug." + }, + { + "left": "issue:45325", + "right": "issue:45902", + "accept": false, + "reason": "Both concern Qwen rotary/position handling, but one is video rope scaling and the other is uninitialized memory after materialization." + }, + { + "left": "issue:45663", + "right": "issue:45823", + "accept": false, + "reason": "Same model family, but different bugs: per-layer cache rebuild KeyError vs device mismatch in forward." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays mostly conservative. The accepted edges are reasonably supported by near-identical symptoms or clearly shared regressions, and the rejects generally avoid collapsing merely similar subsystems into one bug. The only slightly soft spot is the 41093/44805 pairing, which is a bit broad on title similarity alone, but still not an overreach given the identical error pattern." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 6, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "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:43366|issue:44589", + "issue:43716|issue:44560", + "issue:44155|issue:45584", + "issue:43824|issue:43957", + "issue:43950|issue:45834", + "issue:42915|issue:43957", + "issue:44560|issue:45685", + "issue:43645|issue:43646", + "issue:44560|issue:44918", + "issue:43550|issue:44910", + "issue:44683|issue:46515", + "issue:42617|issue:43366", + "issue:36010|issue:42915", + "issue:45812|issue:46489", + "issue:45823|issue:46515", + "issue:44560|issue:44805", + "issue:44661|issue:45823", + "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:43644|issue:44534", + "issue:38175|issue:39692", + "issue:41720|issue:42915", + "issue:42915|issue:44756", + "issue:42491|issue:45276", + "issue:44451|issue:44488", + "issue:44960|issue:44977", + "issue:46459|issue:46512" + ], + "prepared_review_unit_hash": "57b268b45842e58c1294cacd8687c27f0b721f306436b8ad82e2644b3bd43fa8", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 34, + "serialized_chars": 37775, + "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 similarity cluster: a few true duplicates around Transformers v5 loading/init regressions and repeated model-loading failures, but most pairs are unrelated model-specific bugs that only share broad subsystem vocabulary.", + "confidence": 0.82, + "canonical_issue_reason": "43644 is the clearest anchor for the core framework regression: it states the non-persistent-buffer corruption bug in general terms, and 44534 is a near-exact duplicate.", + "canonical_pr_reason": null, + "best_issue_reason": "Best overall representative because it captures the underlying Transformers v5 regression at the framework level rather than a single model symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43638", + "right": "issue:46515", + "accept": false, + "reason": "Different failures: Deepspeed ZeRO3/BERT index error vs a general torch 2.4.1 incompatibility claim." + }, + { + "left": "issue:43366", + "right": "issue:44589", + "accept": false, + "reason": "Unrelated problem areas: GGUF/gpt-oss support vs missing Float8 storage object during loading." + }, + { + "left": "issue:43716", + "right": "issue:44560", + "accept": false, + "reason": "Different model paths and symptoms: image preprocessor dtype mismatch vs Qwen3-vl video StopIteration." + }, + { + "left": "issue:44155", + "right": "issue:45584", + "accept": false, + "reason": "Different modalities and bugs: AudioFlamingo embedding/token leakage vs Whisper empty-transcription handling." + }, + { + "left": "issue:43824", + "right": "issue:43957", + "accept": false, + "reason": "Import error for a specific class vs meta-device loading breakage; not the same underlying bug." + }, + { + "left": "issue:43950", + "right": "issue:45834", + "accept": false, + "reason": "Both are loading-related, but one is about non-persistent buffers and the other is a long-OCR index error in Kosmos2.5." + }, + { + "left": "issue:42915", + "right": "issue:43957", + "accept": false, + "reason": "Qwen3Moe FP8 failure is unrelated to meta-device loading of some models." + }, + { + "left": "issue:44560", + "right": "issue:45685", + "accept": false, + "reason": "Different subsystems: Qwen3-vl video inference vs MPS histogram support in MoE." + }, + { + "left": "issue:43645", + "right": "issue:43646", + "accept": true, + "reason": "Same custom-model initialization regression in Transformers 5; the second issue is just a shorter rewording of the first." + }, + { + "left": "issue:44560", + "right": "issue:44918", + "accept": false, + "reason": "Both involve Qwen3-ish workflows, but one is video StopIteration and the other is TRL embedding unpacking." + }, + { + "left": "issue:43550", + "right": "issue:44910", + "accept": false, + "reason": "Different model families and code paths: SDPA/torch.compile on Bamba vs flash-attention position_id handling on Qwen3.5." + }, + { + "left": "issue:44683", + "right": "issue:46515", + "accept": false, + "reason": "torch>=2.9 flex_attention compile failure is unrelated to a torch 2.4.1 incompatibility report." + }, + { + "left": "issue:42617", + "right": "issue:43366", + "accept": false, + "reason": "3d_parallel.py runtime problem is unrelated to GGUF gpt-oss support." + }, + { + "left": "issue:36010", + "right": "issue:42915", + "accept": false, + "reason": "Importing GenerationMixin from transformers.generation is unrelated to Qwen3Moe FP8 loading." + }, + { + "left": "issue:45812", + "right": "issue:46489", + "accept": false, + "reason": "Both are tokenizer regressions, but they target different model families and likely different code paths." + }, + { + "left": "issue:45823", + "right": "issue:46515", + "accept": false, + "reason": "Gemma4 device-mismatch with device_map=auto is unrelated to a torch 2.4.1 incompatibility." + }, + { + "left": "issue:44560", + "right": "issue:44805", + "accept": false, + "reason": "Different concrete bugs: Qwen3-vl video handling vs generic mask/tensor shape mismatch." + }, + { + "left": "issue:44661", + "right": "issue:45823", + "accept": false, + "reason": "Model-registration logic for tokenizer mappings is unrelated to Gemma4 PLE device placement." + }, + { + "left": "issue:38175", + "right": "issue:43994", + "accept": false, + "reason": "Same model family, but one reports zero probabilities and the other nonsensical outputs; not enough to treat as the same bug." + }, + { + "left": "issue:43632", + "right": "issue:45834", + "accept": false, + "reason": "_is_hf_initialized flag regression is unrelated to Kosmos2.5 OCR index errors." + }, + { + "left": "issue:44877", + "right": "issue:45030", + "accept": false, + "reason": "Both mention config validation, but they are different model/config failures with no clear shared bug." + }, + { + "left": "issue:43638", + "right": "issue:44661", + "accept": false, + "reason": "Deepspeed ZeRO3/BERT initialization error is unrelated to tokenizer-mapping registration." + }, + { + "left": "issue:43596", + "right": "issue:44661", + "accept": false, + "reason": "Different failure modes: empty-shape index error during ZeRO3 init vs add-new-model-like tokenizer mapping issue." + }, + { + "left": "issue:43317", + "right": "issue:44756", + "accept": false, + "reason": "Device-map/offload loading bug is unrelated to mmap OOM on Strix Halo." + }, + { + "left": "issue:44661", + "right": "issue:44805", + "accept": false, + "reason": "Model-registration bug and tensor mask shape error are unrelated." + }, + { + "left": "issue:42491", + "right": "issue:43720", + "accept": false, + "reason": "Qwen3_moe LoRA compatibility across branches is unrelated to BitNet packed-weight loading." + }, + { + "left": "issue:43644", + "right": "issue:44534", + "accept": true, + "reason": "Near-exact duplicate reports of the same non-persistent-buffer junk regression in Transformers v5." + }, + { + "left": "issue:38175", + "right": "issue:39692", + "accept": false, + "reason": "Both concern SigLIP2, but one is an inference-output anomaly and the other is a docs/example mismatch plus quantization error." + }, + { + "left": "issue:41720", + "right": "issue:42915", + "accept": false, + "reason": "CUDA assert under auto device mapping is unrelated to Qwen3Moe FP8 configuration failures." + }, + { + "left": "issue:42915", + "right": "issue:44756", + "accept": false, + "reason": "FP8 config failure and mmap/OOM on Strix Halo are different issues." + }, + { + "left": "issue:42491", + "right": "issue:45276", + "accept": false, + "reason": "LoRA branch compatibility for qwen3_moe is unrelated to Gemma4 token-embedding resize behavior." + }, + { + "left": "issue:44451", + "right": "issue:44488", + "accept": true, + "reason": "Both report the current/latest version failing to load closely related BERT checkpoints, with the same loading symptom." + }, + { + "left": "issue:44960", + "right": "issue:44977", + "accept": false, + "reason": "GLM5 issue and Qwen3.5 flash-attention generation bug are separate model-specific regressions." + }, + { + "left": "issue:46459", + "right": "issue:46512", + "accept": false, + "reason": "One is about dtype being ignored for composite checkpoint loading; the other is stale config.dtype after from_config." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The clustering judgment is grounded in the packet. The two accepted pairs are well supported by near-identical titles and clearly overlapping regression descriptions, and the rejected pairs are generally conservative rather than overclaiming duplicates. The summary also stays appropriately cautious about most pairs being broad-similarity noise rather than true duplicates." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 7, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "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:43644|issue:43645", + "issue:44488|issue:44991", + "issue:43856|issue:43873", + "issue:43856|issue:44387", + "issue:43257|issue:43854", + "issue:39692|issue:43994", + "issue:43854|issue:43931", + "issue:42491|issue:45292", + "issue:43708|issue:44075", + "issue:42491|issue:43931", + "issue:44960|issue:46050", + "issue:44164|issue:44291", + "issue:45488|issue:45812", + "issue:45362|issue:45406", + "issue:45276|issue:45335", + "issue:44910|issue:45588", + "issue:44402|issue:44488", + "issue:43425|issue:43957", + "issue:43404|issue:43454", + "issue:43873|issue:44387", + "issue:39692|issue:44960", + "issue:41720|issue:44155", + "issue:42947|issue:45704", + "issue:45663|issue:45759", + "issue:44291|issue:44756", + "issue:36010|issue:39404", + "issue:45663|issue:45859" + ], + "prepared_review_unit_hash": "07e9941bd651b84fcf530a0b43fb5229349dfcbe6c476d6df249226970135739", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 35909, + "estimated_input_tokens": 8978, + "estimated_eval_tokens": 18212 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly a grab bag of unrelated Transformers issues. The soft pairs are generally same-subsystem but not the same underlying bug; none look safe to merge as duplicates.", + "confidence": 0.88, + "canonical_issue_reason": "issue:45276 is the closest representative because it matches the strongest near-duplicate pair and captures a concrete, recurring API regression around `resize_token_embeddings` failing to update nested embeddings.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45276 is the best stand-in for the cluster: it is specific, clearly stated, and sits in the tightest soft-similarity pair among the issues listed.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45313", + "right": "issue:45357", + "accept": false, + "reason": "Both are Qwen3.5 multimodal regressions, but one is ZeRO-3 language_model loading and the other is incorrect visual encoder keys on save; different code paths." + }, + { + "left": "issue:45507", + "right": "issue:46468", + "accept": false, + "reason": "Different models and failures: GraniteMoEHybrid invalid method call vs Nemotron example breakage; no clear shared underlying bug." + }, + { + "left": "issue:43644", + "right": "issue:43645", + "accept": false, + "reason": "Both mention Transformers 5.0 breakage, but one is non-persistent buffer initialization and the other is custom-model initialization in Jupyter; distinct bugs." + }, + { + "left": "issue:44488", + "right": "issue:44991", + "accept": false, + "reason": "Both are tokenizer/model-loading failures, but for different models and likely different causes; not the same concrete issue." + }, + { + "left": "issue:43856", + "right": "issue:43873", + "accept": false, + "reason": "One is Qwen3 MoE training memory inefficiency, the other is quantization offloading behavior; related area but not the same bug." + }, + { + "left": "issue:43856", + "right": "issue:44387", + "accept": false, + "reason": "Both involve memory, but one is MoE training inefficiency and the other is int4 quantization reserving extra CUDA memory; different failure modes." + }, + { + "left": "issue:43257", + "right": "issue:43854", + "accept": false, + "reason": "Different models and load paths: Qwen3 MoE accelerate/deepspeed conversion vs GLM-4.7-Flash unit test loading." + }, + { + "left": "issue:39692", + "right": "issue:43994", + "accept": false, + "reason": "Both are SigLIP2-related, but one is a broken documentation example and the other is incorrect model/pipeline outputs; not clearly the same bug." + }, + { + "left": "issue:43854", + "right": "issue:43931", + "accept": false, + "reason": "Unrelated model-loading failures for different models; no shared code-path evidence." + }, + { + "left": "issue:42491", + "right": "issue:45292", + "accept": false, + "reason": "Qwen3 MoE hf4.x\u2192hf5.x LoRA compatibility is unrelated to `resize_token_embeddings` not updating output embeddings." + }, + { + "left": "issue:43708", + "right": "issue:44075", + "accept": false, + "reason": "Trainer checkpoint-step math and SGD argument handling are separate training bugs." + }, + { + "left": "issue:42491", + "right": "issue:43931", + "accept": false, + "reason": "Different models and failure mechanisms; not a duplicate." + }, + { + "left": "issue:44960", + "right": "issue:46050", + "accept": false, + "reason": "GLM5 issue vs sinq quantized save/load corruption; different model families and bug types." + }, + { + "left": "issue:44164", + "right": "issue:44291", + "accept": false, + "reason": "Both involve loading/saving internals, but extra_state handling and `init_empty_weights`/_is_hf_initialized are distinct bugs." + }, + { + "left": "issue:45488", + "right": "issue:45812", + "accept": false, + "reason": "Both are tokenizer regressions in v5, but one is LlamaTokenizer pre-tokenizer override for DeepSeek and the other is AutoTokenizer ID corruption for Granite; different root causes." + }, + { + "left": "issue:45362", + "right": "issue:45406", + "accept": false, + "reason": "Different entrypoints and models: Qwen3.5 chat crash vs Gemma4Processor missing `_tokenizer` in transformers serve." + }, + { + "left": "issue:45276", + "right": "issue:45335", + "accept": false, + "reason": "Very similar symptom, but model families and affected submodules differ; insufficient evidence they are the same underlying bug." + }, + { + "left": "issue:44910", + "right": "issue:45588", + "accept": false, + "reason": "Both touch flash attention, but one is Qwen3.5 packed-sequence misinterpretation and the other is `s_aux=None` AttributeError for sink-less models; distinct issues." + }, + { + "left": "issue:44402", + "right": "issue:44488", + "accept": false, + "reason": "Tokenizer vocab-size mismatch vs inability to load a different model; not the same bug." + }, + { + "left": "issue:43425", + "right": "issue:43957", + "accept": false, + "reason": "Torch 2.10 compatibility and `torch.device('meta')` loading breakage are separate compatibility problems." + }, + { + "left": "issue:43404", + "right": "issue:43454", + "accept": false, + "reason": "Both mention untied `lm_head`, but they affect different model classes and there is no evidence of one shared defect across them." + }, + { + "left": "issue:43873", + "right": "issue:44387", + "accept": false, + "reason": "Both are memory/quantization related, but one is offloading behavior and the other is increased reserved memory under int4 quantization; not the same bug." + }, + { + "left": "issue:39692", + "right": "issue:44960", + "accept": false, + "reason": "SigLIP2 documentation errors are unrelated to a GLM5 model-loading issue." + }, + { + "left": "issue:41720", + "right": "issue:44155", + "accept": false, + "reason": "Qwen3 auto device mapping CUDA assert and AudioFlamingo3 token leakage are unrelated." + }, + { + "left": "issue:42947", + "right": "issue:45704", + "accept": false, + "reason": "Gradient checkpointing with PEFT LoRA and T5 Apex RMSNorm memory leak are different problems." + }, + { + "left": "issue:45663", + "right": "issue:45759", + "accept": false, + "reason": "Gemma-4 FSDP2 shared_kv_states KeyError and Qwen 3.5/3.6 text_config unwrapping are different model/config bugs." + }, + { + "left": "issue:44291", + "right": "issue:44756", + "accept": false, + "reason": "Unexpected `_is_hf_initialized` handling and Strix Halo mmap OOM are unrelated." + }, + { + "left": "issue:36010", + "right": "issue:39404", + "accept": false, + "reason": "ImportError for `GenerationMixin` and Whisper pipeline `return_language` regression are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 8, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "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:43957", + "issue:44155", + "issue:44164", + "issue:44292", + "issue:44387", + "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:44387|issue:44756", + "issue:43957|issue:44756", + "issue:43335|issue:45663", + "issue:44155|issue:45823", + "issue:42947|issue:43381", + "issue:44863|issue:46468", + "issue:45276|issue:45292", + "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:43828|issue:43957", + "issue:45292|issue:45335", + "issue:43646|issue:43708", + "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:39404|issue:46665", + "issue:44530|issue:45084", + "issue:44683|issue:45823", + "issue:43425|issue:45823", + "issue:43425|issue:43716", + "issue:44589|issue:45084", + "issue:44479|issue:44805" + ], + "prepared_review_unit_hash": "f31190a2c05b2fd86aaeca6f173d54fc9ae5f52848ac825bab59af6f1a014ea8", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 34, + "serialized_chars": 37886, + "estimated_input_tokens": 9472, + "estimated_eval_tokens": 19200 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a heterogeneous cluster of Transformers issues with a few loose thematic overlaps. The only clear duplicate family is the resize_token_embeddings / output_embeddings regression across model variants (Gemma4, generic output_embeddings, T5Gemma); the rest are different bugs or different code paths.", + "confidence": 0.81, + "canonical_issue_reason": "issue:45292 is the broadest, most generic formulation of the resize_token_embeddings/output_embeddings regression, making it the best anchor for the one clear duplicate family in this set.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45292 is the best representative issue overall because it states the underlying bug cleanly and subsumes the more model-specific variants.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44292", + "right": "issue:44912", + "accept": false, + "reason": "Both are quantization/runtime loading failures, but they involve different models, quantization modes, and likely different code paths." + }, + { + "left": "issue:44387", + "right": "issue:44756", + "accept": false, + "reason": "One is int4 memory growth/OOM, the other is mmap-related OOM on Strix Halo; same symptom, different cause." + }, + { + "left": "issue:43957", + "right": "issue:44756", + "accept": false, + "reason": "These are separate OOM/loading regressions: torch.device('meta') model loading vs disabling mmap on Strix Halo." + }, + { + "left": "issue:43335", + "right": "issue:45663", + "accept": false, + "reason": "Sparse-layer construction in SwitchTransformers and Gemma4 FSDP2 shared_kv_states are unrelated bugs." + }, + { + "left": "issue:44155", + "right": "issue:45823", + "accept": false, + "reason": "AudioFlamingo3 batched inference leakage and Gemma4 PLE device mismatch are different model-specific forward-path issues." + }, + { + "left": "issue:42947", + "right": "issue:43381", + "accept": false, + "reason": "Both mention gradient checkpointing, but one is LoRA ineffectiveness and the other is eval-mode incompatibility." + }, + { + "left": "issue:44863", + "right": "issue:46468", + "accept": false, + "reason": "NemotronH checkpoint loading and a broken official Nemotron example are not the same concrete bug." + }, + { + "left": "issue:45276", + "right": "issue:45292", + "accept": true, + "reason": "Same underlying regression: resize_token_embeddings fails to update output-related embeddings; 45276 is just a Gemma4-specific manifestation." + }, + { + "left": "issue:45685", + "right": "issue:46468", + "accept": false, + "reason": "An MPS histogram kernel limitation is unrelated to the Nemotron example failure." + }, + { + "left": "issue:44479", + "right": "issue:44877", + "accept": false, + "reason": "Video-input regression in Qwen VL models vs strict config blocking granite_speech loading are different issues." + }, + { + "left": "issue:44164", + "right": "issue:44756", + "accept": false, + "reason": "save/from_pretrained extra_state handling is unrelated to the Strix Halo mmap OOM." + }, + { + "left": "issue:43381", + "right": "issue:44928", + "accept": false, + "reason": "Eval-mode gradient checkpointing incompatibility and Qwen3.5 RLHF NaNs from 3D position_ids are different failures." + }, + { + "left": "issue:43854", + "right": "issue:45216", + "accept": false, + "reason": "GLM-4.7-Flash test loading failure and Qwen3.5 save_pretrained regression are separate model/save-path bugs." + }, + { + "left": "issue:45588", + "right": "issue:45663", + "accept": false, + "reason": "flash_attention.py crashing on s_aux=None is unrelated to Gemma4 FSDP2 shared_kv_states rebuilds." + }, + { + "left": "issue:45684", + "right": "issue:45698", + "accept": false, + "reason": "Both involve custom model save/load flows, but one is read-only permission propagation and the other is loading the wrong custom module." + }, + { + "left": "issue:42796", + "right": "issue:46326", + "accept": false, + "reason": "Tokenizer BPE marker handling and continue_final_message behavior are different tokenizer regressions." + }, + { + "left": "issue:43828", + "right": "issue:43957", + "accept": false, + "reason": "Different root causes: autocast dtype mismatch for Phi-tiny-MoE vs meta-device loading breakage." + }, + { + "left": "issue:45292", + "right": "issue:45335", + "accept": true, + "reason": "Same resize_token_embeddings regression; 45335 is a T5Gemma-specific case where decoder.embed_tokens is not updated." + }, + { + "left": "issue:43646", + "right": "issue:43708", + "accept": false, + "reason": "Custom model initialization under Transformers 5.0.0 and Trainer resume max_steps calculation are unrelated." + }, + { + "left": "issue:44402", + "right": "issue:44451", + "accept": false, + "reason": "Tokenizer vocab-size mismatch for one model and inability to load another model are not the same bug." + }, + { + "left": "issue:43645", + "right": "issue:44164", + "accept": false, + "reason": "Notebook-defined custom model initialization and extra_state save/load handling are different code paths." + }, + { + "left": "issue:44589", + "right": "issue:44756", + "accept": false, + "reason": "Float8 storage lookup failure is unrelated to mmap-induced OOM." + }, + { + "left": "issue:43425", + "right": "issue:43828", + "accept": false, + "reason": "Torch 2.10 incompatibility is broader environment breakage, not the same as a specific autocast dtype mismatch." + }, + { + "left": "issue:45812", + "right": "issue:46315", + "accept": false, + "reason": "Wrong Granite token IDs and slow convert_tokens_to_ids are distinct tokenizer issues, one correctness and one performance." + }, + { + "left": "issue:46315", + "right": "issue:46491", + "accept": false, + "reason": "Performance regression in token conversion is unrelated to CodeLlama round-trip spacing behavior." + }, + { + "left": "issue:39404", + "right": "issue:45701", + "accept": false, + "reason": "Whisper return_language pipeline breakage and tokenization changes in version upgrades are not the same bug." + }, + { + "left": "issue:43611", + "right": "issue:43645", + "accept": false, + "reason": "Both concern Transformers 5.0.0 custom-model behavior, but one is base_model_prefix loading and the other is notebook-defined model initialization." + }, + { + "left": "issue:39404", + "right": "issue:46665", + "accept": false, + "reason": "Whisper pipeline return_language and from_pretrained inside Python classes are unrelated." + }, + { + "left": "issue:44530", + "right": "issue:45084", + "accept": false, + "reason": "PagedAttentionCache linear_attention group handling and compile-non-template-node errors are different failures." + }, + { + "left": "issue:44683", + "right": "issue:45823", + "accept": false, + "reason": "Compiled flex_attention on torch>=2.9 and Gemma4 device mismatch are unrelated." + }, + { + "left": "issue:43425", + "right": "issue:45823", + "accept": false, + "reason": "Torch version incompatibility is not the same concrete bug as Gemma4 PLE device mismatch." + }, + { + "left": "issue:43425", + "right": "issue:43716", + "accept": false, + "reason": "General torch incompatibility and Mistral-3 image-preprocessor dtype mismatch are different problems." + }, + { + "left": "issue:44589", + "right": "issue:45084", + "accept": false, + "reason": "Float8 storage object lookup and non-template node compilation errors are unrelated." + }, + { + "left": "issue:44479", + "right": "issue:44805", + "accept": false, + "reason": "Qwen VL video regression and mask-shape IndexError do not look like the same underlying issue." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet, and the two accepted links are well supported by the titles and soft-similarity signal: issue:45276/45292 and issue:45292/45335 are clearly the same resize_token_embeddings/output_embeddings family. The other soft-edge rejections stay conservative and do not overclaim identity across clearly different model- or code-path-specific failures." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 9, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:32090", + "issue:36032", + "issue:39404", + "issue:42796", + "issue:42915", + "issue:43299", + "issue:43404", + "issue:43425", + "issue:43452", + "issue:43541", + "issue:43596", + "issue:43611", + "issue:43643", + "issue:43644", + "issue:43716", + "issue:43856", + "issue:43901", + "issue:43906", + "issue:43931", + "issue:43994", + "issue:44079", + "issue:44164", + "issue:44265", + "issue:44451", + "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:43906|issue:44530", + "issue:43901|issue:44509", + "issue:43425|issue:44918", + "issue:43994|issue:44079", + "issue:44265|issue:44898", + "issue:44671|issue:44991", + "issue:36032|issue:43452", + "issue:39404|issue:44912", + "issue:44164|issue:45759", + "issue:46050|issue:46326", + "issue:43299|issue:43931", + "issue:43906|issue:46326", + "issue:44991|issue:45081", + "issue:43856|issue:44756", + "issue:43596|issue:44683", + "issue:44164|issue:44479", + "issue:44488|issue:45081", + "issue:44488|issue:44987", + "issue:43611|issue:43644", + "issue:44462|issue:46489", + "issue:42796|issue:43404", + "issue:43716|issue:45084", + "issue:36032|issue:42915", + "issue:46285|issue:46402", + "issue:44737|issue:45859", + "issue:32090|issue:39404", + "issue:44661|issue:44683", + "issue:43643|issue:44530", + "issue:43541|issue:45084", + "issue:42915|issue:44530", + "issue:44451|issue:44987" + ], + "prepared_review_unit_hash": "cdfe49bc87154dfd7cc9dc23f402d608644b81ed43166a0af4ac267b8a6f8793", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 37109, + "estimated_input_tokens": 9278, + "estimated_eval_tokens": 18812 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a heterogeneous cluster of mostly unrelated Transformers issues (tokenizer/load regressions, multimodal model loading, quantization, compile/runtime errors, and docs). None of the soft pairs look like the same concrete underlying bug/change, so all pairings should stay separate.", + "confidence": 0.93, + "canonical_issue_reason": "Issue #43611 is the broadest representative of the cluster\u2019s recurring theme of Transformers v5 loading regressions, but the overall set is still too mixed to treat as a single duplicate target.", + "canonical_pr_reason": null, + "best_issue_reason": "#43611 is the most general issue here; it\u2019s a clear v5 model-loading regression and is closer to the dominant pattern than the more model-specific reports. Still, this cluster does not form a strong duplicate group.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43906", + "right": "issue:44530", + "accept": false, + "reason": "One is an isolated reproduction of a previously reported issue; the other is a Qwen3.5 PagedAttentionCache crash. Different code paths and symptoms." + }, + { + "left": "issue:43901", + "right": "issue:44509", + "accept": false, + "reason": "Both are docs-related, but they cover different pipeline API mismatches: return_all_scores vs removed task names in v5." + }, + { + "left": "issue:43425", + "right": "issue:44918", + "accept": false, + "reason": "Torch 2.10 compatibility is unrelated to the Qwen3.5 TRL SFT embedding-unpacking failure." + }, + { + "left": "issue:43994", + "right": "issue:44079", + "accept": false, + "reason": "SigLIP2 incorrect outputs and ModelOutput key assignment bugs are different problems in different layers." + }, + { + "left": "issue:44265", + "right": "issue:44898", + "accept": false, + "reason": "torch.export failures with torch_compilable_check are unrelated to Perceiver non-default-resolution interpolation failures." + }, + { + "left": "issue:44671", + "right": "issue:44991", + "accept": false, + "reason": "CamemBERT masked-LM prediction regression and EMBEDDIA tokenizer loading failure are distinct tokenizer/model issues." + }, + { + "left": "issue:36032", + "right": "issue:43452", + "accept": false, + "reason": "T5Tokenizer special-token method conflict and gguf_file loading breakage are unrelated tokenizer/load paths." + }, + { + "left": "issue:39404", + "right": "issue:44912", + "accept": false, + "reason": "Whisper pipeline return_language regression is unrelated to MXFP4 quantization fallback behavior." + }, + { + "left": "issue:44164", + "right": "issue:45759", + "accept": false, + "reason": "extra_state save/load handling and AutoModelForCausalLM config unwrapping are different serialization/config bugs." + }, + { + "left": "issue:46050", + "right": "issue:46326", + "accept": false, + "reason": "sinq quantization save/load corruption and mistral_common continue_final_message behavior are unrelated." + }, + { + "left": "issue:43299", + "right": "issue:43931", + "accept": false, + "reason": "A generic Qwen3VL MoE loading break in v5.0.0.dev0 is not the same as a specific weight-shape mismatch on a Qwen3-VL model." + }, + { + "left": "issue:43906", + "right": "issue:46326", + "accept": false, + "reason": "An isolated reproduction report is not the same bug as mistral_common tokenizer continue_final_message handling." + }, + { + "left": "issue:44991", + "right": "issue:45081", + "accept": false, + "reason": "Different tokenizer-loading failures affecting different models and different internals." + }, + { + "left": "issue:43856", + "right": "issue:44756", + "accept": false, + "reason": "Qwen3 MoE training memory inefficiency and Strix Halo mmap OOM avoidance are unrelated performance issues." + }, + { + "left": "issue:43596", + "right": "issue:44683", + "accept": false, + "reason": "DeepSpeed ZeRO-3/BertModel index error is unrelated to compiled flex_attention on torch >=2.9." + }, + { + "left": "issue:44164", + "right": "issue:44479", + "accept": false, + "reason": "extra_state serialization and Qwen video-input regression are separate bugs." + }, + { + "left": "issue:44488", + "right": "issue:45081", + "accept": false, + "reason": "Loading sleng-bert and patch_mistral_regex crashes are different tokenizer-loading failures." + }, + { + "left": "issue:44488", + "right": "issue:44987", + "accept": false, + "reason": "These are both model-loading issues, but they affect different repositories and likely different tokenizer/model code paths." + }, + { + "left": "issue:43611", + "right": "issue:43644", + "accept": false, + "reason": "Both mention Transformers v5, but one is base_model_prefix loading and the other is non-persistent buffer initialization\u2014different defects." + }, + { + "left": "issue:44462", + "right": "issue:46489", + "accept": false, + "reason": "Ignoring tokenizer.json on load and DeepSeek-Coder tokenizer output regressions are not the same underlying problem." + }, + { + "left": "issue:42796", + "right": "issue:43404", + "accept": false, + "reason": "Tokenizer BPE-marker handling and Mistral3 lm_head weight tying are unrelated." + }, + { + "left": "issue:43716", + "right": "issue:45084", + "accept": false, + "reason": "Image preprocessor/model dtype mismatch is unrelated to a template compilation TypeError." + }, + { + "left": "issue:36032", + "right": "issue:42915", + "accept": false, + "reason": "A T5 tokenizer method-name conflict is unrelated to Qwen3Moe FineGrainedFP8Config failure." + }, + { + "left": "issue:46285", + "right": "issue:46402", + "accept": false, + "reason": "CLIP structure inconsistency and Sam3 text-encoder weight loading are different model-specific regressions." + }, + { + "left": "issue:44737", + "right": "issue:45859", + "accept": false, + "reason": "Both are device/offload related, but XLNet CPU arange placement and Qwen3_5Moe rotary embedding CPU-offload compatibility are different code paths." + }, + { + "left": "issue:32090", + "right": "issue:39404", + "accept": false, + "reason": "Trainer gpu_broadcast_one NoneType error and Whisper return_language regression are unrelated." + }, + { + "left": "issue:44661", + "right": "issue:44683", + "accept": false, + "reason": "TOKENIZER_MAPPING_NAMES add-new-model-like behavior is unrelated to compiled flex_attention failures." + }, + { + "left": "issue:43643", + "right": "issue:44530", + "accept": false, + "reason": "trust_remote_code config field loss and Qwen3.5 PagedAttentionCache crashes are separate issues." + }, + { + "left": "issue:43541", + "right": "issue:45084", + "accept": false, + "reason": "Torch Dynamo grouped_mm tracing failure and template-node compilation TypeError are different compile-time bugs." + }, + { + "left": "issue:42915", + "right": "issue:44530", + "accept": false, + "reason": "Qwen3 MoE FP8 config loading and Qwen3.5 cache group-type crash are distinct model/runtime failures." + }, + { + "left": "issue:44451", + "right": "issue:44987", + "accept": false, + "reason": "Loading a specific BERT checkpoint and loading physical-intelligence/fast are unrelated model-load regressions." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 10, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:28530", + "issue:34567", + "issue:36032", + "issue:40444", + "issue:40990", + "issue:42175", + "issue:42915", + "issue:43089", + "issue:43262", + "issue:43317", + "issue:43377", + "issue:43404", + "issue:43421", + "issue:43452", + "issue:43482", + "issue:43582", + "issue:43632", + "issue:43644", + "issue:43645", + "issue:43646", + "issue:43742", + "issue:43746", + "issue:43819", + "issue:43883", + "issue:43994", + "issue:44206", + "issue:44220", + "issue:44291", + "issue:44403", + "issue:44451", + "issue:44464", + "issue:44466", + "issue:44488", + "issue:44534", + "issue:44589", + "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:43582|issue:43632", + "issue:40990|issue:43421", + "issue:43644|issue:43646", + "issue:34567|issue:40444", + "issue:44206|issue:44220", + "issue:44811|issue:45584", + "issue:42175|issue:46178", + "issue:43883|issue:44488", + "issue:40444|issue:46326", + "issue:36032|issue:43645", + "issue:43452|issue:43482", + "issue:45084|issue:45507", + "issue:43317|issue:43482", + "issue:44987|issue:44991", + "issue:43262|issue:43377", + "issue:43746|issue:46326", + "issue:42915|issue:45084", + "issue:43883|issue:44991", + "issue:44403|issue:45081", + "issue:44464|issue:45987", + "issue:43646|issue:44534", + "issue:43645|issue:44534", + "issue:43404|issue:44466", + "issue:44849|issue:45325", + "issue:43377|issue:43819", + "issue:42915|issue:44291", + "issue:43994|issue:44610", + "issue:44451|issue:45081", + "issue:44987|issue:45081", + "issue:28530|issue:34567", + "issue:36032|issue:43742", + "issue:42915|issue:44589", + "issue:43089|issue:45987" + ], + "prepared_review_unit_hash": "4487abee1eba01967a7bb59a08ca760f09ccd379a155cbbb84ec54016596af3d", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 33, + "serialized_chars": 37692, + "estimated_input_tokens": 9423, + "estimated_eval_tokens": 19102 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose set of Transformers v5 regression reports across unrelated subsystems. I don\u2019t see any soft pair that clearly describes the same concrete bug, so I would not merge any of them as duplicates.", + "confidence": 0.24, + "canonical_issue_reason": "Issue 43646 is the broadest and most representative v5 regression report in the set, but the cluster is heterogeneous and not a strong duplicate group.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43646 is the best single representative for this cluster because it is general, version-scoped, and closer to the shared v5 regression theme than the more model-specific reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43582", + "right": "issue:43632", + "accept": false, + "reason": "Different failures: AppleSilicon allocator warmup TypeError vs _is_hf_initialized flag regression." + }, + { + "left": "issue:40990", + "right": "issue:43421", + "accept": false, + "reason": "Unrelated bugs: perplexity quality issue vs tokenizer post-processor update behavior." + }, + { + "left": "issue:43644", + "right": "issue:43646", + "accept": false, + "reason": "Both are v5 regressions, but one is buffer initialization corruption and the other is custom model init breakage; not clearly the same concrete bug." + }, + { + "left": "issue:34567", + "right": "issue:40444", + "accept": false, + "reason": "Trainer token-count tracking vs multimodal iterable dataset batching; different code paths." + }, + { + "left": "issue:44206", + "right": "issue:44220", + "accept": false, + "reason": "Both touch audio features, but one is a center-arg crash and the other is a different fbank helper issue; not enough to treat as one bug." + }, + { + "left": "issue:44811", + "right": "issue:45584", + "accept": false, + "reason": "Whisper decode skip_special_tokens vs empty-transcription generation after align_special_tokens; separate symptoms." + }, + { + "left": "issue:42175", + "right": "issue:46178", + "accept": false, + "reason": "Different backend-installation problems: missing tensorflow in extras vs unknown tensorflow_text backend mapping." + }, + { + "left": "issue:43883", + "right": "issue:44488", + "accept": false, + "reason": "Different model-loading failures for different models and likely different root causes." + }, + { + "left": "issue:40444", + "right": "issue:46326", + "accept": false, + "reason": "Different audio-multimodal failures: iterable dataset with multiple images vs mistral_common continue_final_message handling." + }, + { + "left": "issue:36032", + "right": "issue:43645", + "accept": false, + "reason": "Tokenizer method-name conflict vs Jupyter custom model initialization regression; unrelated." + }, + { + "left": "issue:43452", + "right": "issue:43482", + "accept": false, + "reason": "Both are GGUF-related, but one is a general from_pretrained breakage and the other is a specific Qwen2.5-GGUF v5 issue." + }, + { + "left": "issue:45084", + "right": "issue:45507", + "accept": false, + "reason": "Template compilation error vs GraniteMoEHybrid invalid method call; different failures." + }, + { + "left": "issue:43317", + "right": "issue:43482", + "accept": false, + "reason": "Dequantized model offload failure vs Qwen2.5-GGUF loading failure; not the same bug." + }, + { + "left": "issue:44987", + "right": "issue:44991", + "accept": false, + "reason": "Both are load failures on different models; too broad to be duplicates." + }, + { + "left": "issue:43262", + "right": "issue:43377", + "accept": false, + "reason": "Audio chat-template sampling-rate default vs MIMI batching/padding-mask inconsistency; unrelated." + }, + { + "left": "issue:43746", + "right": "issue:46326", + "accept": false, + "reason": "PEFT adapter local checkpoint loading vs mistral_common message handling; distinct issues." + }, + { + "left": "issue:42915", + "right": "issue:45084", + "accept": false, + "reason": "FineGrainedFP8Config failure and template compilation failure are unrelated." + }, + { + "left": "issue:43883", + "right": "issue:44991", + "accept": false, + "reason": "Different models and different loading paths; no shared concrete bug." + }, + { + "left": "issue:44403", + "right": "issue:45081", + "accept": false, + "reason": "General loading noise vs Mistral regex patch crash; not the same problem." + }, + { + "left": "issue:44464", + "right": "issue:45987", + "accept": false, + "reason": "Chunked generation inconsistency vs StaticCache return-type contract bug; different code paths." + }, + { + "left": "issue:43646", + "right": "issue:44534", + "accept": false, + "reason": "Both mention v5 and initialization-like behavior, but buffer corruption is not the same concrete bug as custom model initialization breakage." + }, + { + "left": "issue:43645", + "right": "issue:44534", + "accept": false, + "reason": "Jupyter notebook custom-model init breakage vs non-persistent buffer junk; separate regressions." + }, + { + "left": "issue:43404", + "right": "issue:44466", + "accept": false, + "reason": "Different lm_head tying/serialization symptoms on different models and scenarios." + }, + { + "left": "issue:44849", + "right": "issue:45325", + "accept": false, + "reason": "Different model-specific output/position-id bugs with no clear shared code path." + }, + { + "left": "issue:43377", + "right": "issue:43819", + "accept": false, + "reason": "Both are audio model issues, but batched-vs-single padding mask behavior is not the same as DAC latent/STE mismatch." + }, + { + "left": "issue:42915", + "right": "issue:44291", + "accept": false, + "reason": "Different initialization/quantization failures; only loosely related by v5-era API changes." + }, + { + "left": "issue:43994", + "right": "issue:44610", + "accept": false, + "reason": "Different models and distinct processor/model input-size problems." + }, + { + "left": "issue:44451", + "right": "issue:45081", + "accept": false, + "reason": "Different tokenizer/model loading failures with different models and mechanics." + }, + { + "left": "issue:44987", + "right": "issue:45081", + "accept": false, + "reason": "Different targets: physical-intelligence/fast load regression vs Mistral tokenizer regex crash." + }, + { + "left": "issue:28530", + "right": "issue:34567", + "accept": false, + "reason": "Early stopping metric selection vs trainer token counter update; unrelated." + }, + { + "left": "issue:36032", + "right": "issue:43742", + "accept": false, + "reason": "Tokenizer class-name conflict vs model loading KeyError; different failure modes." + }, + { + "left": "issue:42915", + "right": "issue:44589", + "accept": false, + "reason": "FineGrainedFP8Config failure vs missing Float8 storage type; only loosely related by quantization support." + }, + { + "left": "issue:43089", + "right": "issue:45987", + "accept": false, + "reason": "Generation performance overhead vs cache sequence-length return type; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 11, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:33666", + "issue:35707", + "issue:36032", + "issue:40444", + "issue:40990", + "issue:41553", + "issue:41720", + "issue:42796", + "issue:43296", + "issue:43322", + "issue:43381", + "issue:43425", + "issue:43482", + "issue:43525", + "issue:43531", + "issue:43540", + "issue:43582", + "issue:43611", + "issue:43632", + "issue:43746", + "issue:43792", + "issue:44062", + "issue:44079", + "issue:44220", + "issue:44265", + "issue:44291", + "issue:44403", + "issue:44479", + "issue:44534", + "issue:44589", + "issue:44610", + "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:46315", + "issue:46491", + "issue:46710" + ], + "soft_pairs": [ + "issue:44062|issue:44987", + "issue:44265|issue:44610", + "issue:43322|issue:43540", + "issue:44479|issue:45325", + "issue:45216|issue:45357", + "issue:43531|issue:44877", + "issue:36032|issue:41720", + "issue:44991|issue:46315", + "issue:43425|issue:44403", + "issue:43296|issue:43531", + "issue:44220|issue:44683", + "issue:43611|issue:44534", + "issue:40444|issue:45325", + "issue:45812|issue:46491", + "issue:43632|issue:44756", + "issue:44403|issue:44991", + "issue:43792|issue:44589", + "issue:43381|issue:45923", + "issue:42796|issue:43746", + "issue:44291|issue:44589", + "issue:33666|issue:40444", + "issue:44079|issue:45663", + "issue:40990|issue:44811", + "issue:42796|issue:46710", + "issue:43482|issue:43525", + "issue:35707|issue:40444", + "issue:43632|issue:44534", + "issue:43482|issue:45538", + "issue:41553|issue:43582", + "issue:43531|issue:44291" + ], + "prepared_review_unit_hash": "60f813d8bb283f311e0d6dfde92db059a99359924b45602ede2912669994641c", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 35940, + "estimated_input_tokens": 8985, + "estimated_eval_tokens": 18226 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous transformer/model issues, not one tight duplicate set. The only plausible duplicate-like pair is the Qwen3.5 save_pretrained regression thread; the rest are different bugs in different code paths.", + "confidence": 0.79, + "canonical_issue_reason": "issue:45216 is the clearest anchor: it describes a concrete Qwen3.5 save_pretrained regression and already has downstream references, making it the best representative of the only near-duplicate thread.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45812 is the strongest standalone triage candidate overall because it is broad ('all Granite models'), clearly a silent v4\u2192v5 regression, and has the highest apparent external impact.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44062", + "right": "issue:44987", + "accept": false, + "reason": "Both involve tokenizer loading failures, but the symptoms and likely code paths differ: one is an AddedToken API conflict, the other is a model-specific load failure on v5.1." + }, + { + "left": "issue:44265", + "right": "issue:44610", + "accept": false, + "reason": "Different subsystems and bugs: torch.export/torch_compilable_check versus an OmDet-Turbo image-size mismatch in the processor." + }, + { + "left": "issue:43322", + "right": "issue:43540", + "accept": false, + "reason": "Both are multimodal loading/processing problems, but one is a Llava Next segfault and the other is a Qwen3OmniMoe video ValueError; not the same bug." + }, + { + "left": "issue:44479", + "right": "issue:45325", + "accept": false, + "reason": "Related Qwen VL/video/rope area, but one is a broad v5.3 video regression across models and the other is a specific still-image temporal position_ids scaling bug." + }, + { + "left": "issue:45216", + "right": "issue:45357", + "accept": true, + "reason": "Same model family and same code path: Qwen3.5 save_pretrained producing incorrect checkpoints/visual encoder keys across versions." + }, + { + "left": "issue:43531", + "right": "issue:44877", + "accept": false, + "reason": "Different failures: Qwen3-MoE sliding_window handling versus strict config blocking granite_speech loading." + }, + { + "left": "issue:36032", + "right": "issue:41720", + "accept": false, + "reason": "Unrelated issues: a T5 tokenizer method conflict versus Qwen3 auto device mapping causing CUDA assert." + }, + { + "left": "issue:44991", + "right": "issue:46315", + "accept": false, + "reason": "Both are tokenizer regressions, but one is a load-time failure for a specific tokenizer and the other is a performance regression in convert_tokens_to_ids; different bug." + }, + { + "left": "issue:43425", + "right": "issue:44403", + "accept": false, + "reason": "No shared underlying change: Torch 2.10 compatibility versus extra loading noise." + }, + { + "left": "issue:43296", + "right": "issue:43531", + "accept": false, + "reason": "Different models and different errors: PaddleOCR-VL/vLLM loading versus Qwen3-MoE sliding_window behavior." + }, + { + "left": "issue:44220", + "right": "issue:44683", + "accept": false, + "reason": "Different code paths: _torch_extract_fbank_features versus compiled flex_attention on newer torch." + }, + { + "left": "issue:43611", + "right": "issue:44534", + "accept": false, + "reason": "One is a model-loading break with base_model_prefix; the other is a v5 buffer-initialization regression." + }, + { + "left": "issue:40444", + "right": "issue:45325", + "accept": false, + "reason": "Both are Qwen VL-related, but one is iterable-dataset/multi-image finetuning failure and the other is a rope/position_ids bug; not the same defect." + }, + { + "left": "issue:45812", + "right": "issue:46491", + "accept": false, + "reason": "Both are tokenizer regressions, but they affect different model families and different semantics: Granite token IDs versus CodeLlama space preservation." + }, + { + "left": "issue:43632", + "right": "issue:44756", + "accept": false, + "reason": "No meaningful overlap: _is_hf_initialized flag handling versus mmap/OOM behavior on Strix Halo." + }, + { + "left": "issue:44403", + "right": "issue:44991", + "accept": false, + "reason": "Different symptoms and targets: generic loading noise versus a specific tokenizer load failure." + }, + { + "left": "issue:43792", + "right": "issue:44589", + "accept": false, + "reason": "Completely different failure modes: Whisper model run failure versus missing Float8 storage object." + }, + { + "left": "issue:43381", + "right": "issue:45923", + "accept": false, + "reason": "Related to gradient checkpointing, but one is eval-mode usage semantics and the other is a missing supports_gradient_checkpointing flag on a trust_remote_code model." + }, + { + "left": "issue:42796", + "right": "issue:43746", + "accept": false, + "reason": "Tokenizer BPE marker handling and GraniteSpeech PEFT local checkpoint loading are unrelated." + }, + { + "left": "issue:44291", + "right": "issue:44589", + "accept": false, + "reason": "Different backend issues: init_empty_weights/_is_hf_initialized argument handling versus Float8 storage lookup." + }, + { + "left": "issue:33666", + "right": "issue:40444", + "accept": false, + "reason": "Both are Qwen VL training/fine-tuning related, but the concrete failures are different and not the same underlying bug." + }, + { + "left": "issue:44079", + "right": "issue:45663", + "accept": false, + "reason": "Different mechanisms: ModelOutput key assignment bug versus Gemma4 FSDP2 shared_kv_states rebuilding error." + }, + { + "left": "issue:40990", + "right": "issue:44811", + "accept": false, + "reason": "No duplicate relationship: perplexity regression on GPT-OSS versus Whisper batch_decode special-token handling." + }, + { + "left": "issue:42796", + "right": "issue:46710", + "accept": false, + "reason": "Tokenizer BPE-marker bug and DeepSeek accuracy regression are unrelated." + }, + { + "left": "issue:43482", + "right": "issue:43525", + "accept": false, + "reason": "Different model/config problems: GGUF loading under transformers v5 versus missing pad_token_id on Llama4Config." + }, + { + "left": "issue:35707", + "right": "issue:40444", + "accept": false, + "reason": "Both mention generation/fine-tuning workflows, but one is inputs_embeds/past_key_values progressive generation and the other is multimodal iterable dataset loading." + }, + { + "left": "issue:43632", + "right": "issue:44534", + "accept": false, + "reason": "Unrelated regressions: _is_hf_initialized flag versus non-persistent buffer initialization." + }, + { + "left": "issue:43482", + "right": "issue:45538", + "accept": false, + "reason": "Different tokenizer/model-loading issues: GGUF loading failure versus CLIPTokenizer max length behavior." + }, + { + "left": "issue:41553", + "right": "issue:43582", + "accept": false, + "reason": "Completely different areas: AutoTokenizer error messaging for Voxtral versus Apple Silicon caching allocator TypeError." + }, + { + "left": "issue:43531", + "right": "issue:44291", + "accept": false, + "reason": "Different bugs in different layers: Qwen3-MoE sliding_window behavior versus init_empty_weights argument handling." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded and conservative overall. The broad Granite tokenizer regression is a reasonable best standalone issue, the Qwen3.5 save_pretrained thread is the clearest canonical anchor, and the accepted pair is well-supported by the packet. The rejected soft-edge pairs stay appropriately cautious and do not overclaim duplication." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 12, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:40444", + "issue:41720", + "issue:42175", + "issue:42796", + "issue:42915", + "issue:43296", + "issue:43317", + "issue:43377", + "issue:43421", + "issue:43454", + "issue:43482", + "issue:43493", + "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:44568", + "issue:44829", + "issue:44877", + "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:43421|issue:44568", + "issue:44079|issue:44534", + "issue:44829|issue:45479", + "issue:41720|issue:45081", + "issue:44462|issue:45812", + "issue:43818|issue:46402", + "issue:43818|issue:45357", + "issue:43611|issue:43646", + "issue:43377|issue:45491", + "issue:43644|issue:44079", + "issue:42175|issue:43296", + "issue:43749|issue:44079", + "issue:44265|issue:45072", + "issue:43792|issue:43957", + "issue:45325|issue:45381", + "issue:43454|issue:44466", + "issue:43493|issue:43994", + "issue:40444|issue:45381", + "issue:45325|issue:45834", + "issue:42915|issue:43482", + "issue:44291|issue:44877", + "issue:43632|issue:45072", + "issue:43873|issue:45701", + "issue:43701|issue:45325", + "issue:42796|issue:45381", + "issue:43296|issue:45800", + "issue:44479|issue:45823" + ], + "prepared_review_unit_hash": "8774ee2b0987a2361dc348146d6d8924a8a01ac6358cbeb27294e458b4827311", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36624, + "estimated_input_tokens": 9156, + "estimated_eval_tokens": 18568 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose collection of unrelated or only loosely related Transformers regressions, many around v5 loading/initialization/tokenizer behavior. I did not find any soft pair that looks like the same underlying bug/change, so this should not be merged as duplicate triage.", + "confidence": 0.93, + "canonical_issue_reason": "If a single issue must represent the cluster, issue 43611 is the broadest v5 loading-regression umbrella; however, the cluster is not a true duplicate set and the items are still distinct bugs.", + "canonical_pr_reason": null, + "best_issue_reason": "43611 is the most representative of the repeated Transformers v5 model-loading breakage theme, but it is still only an umbrella candidate rather than an exact duplicate of the others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43317", + "right": "issue:43856", + "accept": false, + "reason": "Both concern offloading/quantization performance or failure modes, but they describe different symptoms and code paths." + }, + { + "left": "issue:42175", + "right": "issue:45542", + "accept": false, + "reason": "Both are backend-installation packaging issues, but one is about pip extras missing TensorFlow while the other is about tensorboard/TF backend detection." + }, + { + "left": "issue:43421", + "right": "issue:44568", + "accept": false, + "reason": "Runtime special-token/post-processor update behavior is unrelated to add_special_tokens BOS/EOS insertion regression." + }, + { + "left": "issue:44079", + "right": "issue:44534", + "accept": false, + "reason": "Both mention ModelOutput/non-persistent buffer handling, but they affect different APIs and failure modes." + }, + { + "left": "issue:44829", + "right": "issue:45479", + "accept": false, + "reason": "These are different training regressions: flash_attention_3 classification degeneration vs single-label num_labels=1 zero-loss behavior." + }, + { + "left": "issue:41720", + "right": "issue:45081", + "accept": false, + "reason": "One is a CUDA assert in Qwen3 device mapping; the other is a tokenizer regex patch crash." + }, + { + "left": "issue:44462", + "right": "issue:45812", + "accept": false, + "reason": "Both involve tokenizer loading/serialization, but one ignores tokenizer.json and the other returns wrong token IDs for Granite models." + }, + { + "left": "issue:43818", + "right": "issue:46402", + "accept": false, + "reason": "Video-LLaVA tower architecture corruption is unrelated to Sam3 text-encoder weights not loading." + }, + { + "left": "issue:43818", + "right": "issue:45357", + "accept": false, + "reason": "Both are vision/model-loading issues, but one is Video-LLaVA architecture state and the other is Qwen3.5 save_pretrained key serialization." + }, + { + "left": "issue:43611", + "right": "issue:43646", + "accept": false, + "reason": "Both are v5 loading regressions, but one is base_model_prefix loading and the other is custom model initialization." + }, + { + "left": "issue:43377", + "right": "issue:45491", + "accept": false, + "reason": "MIMI encoder padding-mask batching bug is not the same as Gemma3 sliding-window NaN embeddings." + }, + { + "left": "issue:43644", + "right": "issue:44079", + "accept": false, + "reason": "Non-persistent buffer junk filling is a different defect from ModelOutput key assignment when previous value was None." + }, + { + "left": "issue:42175", + "right": "issue:43296", + "accept": false, + "reason": "Package-extras TensorFlow omission is unrelated to PaddleOCR-VL loading in vLLM/transformers." + }, + { + "left": "issue:43749", + "right": "issue:44079", + "accept": false, + "reason": "FSDP CPU RAM efficient loading is a distributed-loading issue, not a ModelOutput bookkeeping bug." + }, + { + "left": "issue:44265", + "right": "issue:45072", + "accept": false, + "reason": "torch.export.export with torch_compilable_check is unrelated to dtype mismatches in bfloat16 inference." + }, + { + "left": "issue:43792", + "right": "issue:43957", + "accept": false, + "reason": "Whisper model runtime failure and meta-device loading breakage are different model-loading problems." + }, + { + "left": "issue:45325", + "right": "issue:45381", + "accept": false, + "reason": "Both are Qwen2.5-VL video-position issues, but one is still-image temporal scaling while the other is video input vision_position_ids regression." + }, + { + "left": "issue:43454", + "right": "issue:44466", + "accept": false, + "reason": "AyaVision lm_head tying failure and inconsistent lm_head serialization are related only at a broad weights level, not the same bug." + }, + { + "left": "issue:43493", + "right": "issue:43994", + "accept": false, + "reason": "SigLIP2 HF-vs-JAX discrepancy is not the same as wrong AutoModel/pipeline outputs for siglip2-base." + }, + { + "left": "issue:40444", + "right": "issue:45381", + "accept": false, + "reason": "Multimodal finetuning with multiple images per prompt is unrelated to Qwen2.5-VL video position-id regression." + }, + { + "left": "issue:45325", + "right": "issue:45834", + "accept": false, + "reason": "Qwen2.5-VL temporal position scaling and Kosmos2.5 OCR index errors are different code paths and symptoms." + }, + { + "left": "issue:42915", + "right": "issue:43482", + "accept": false, + "reason": "Qwen3Moe FP8 failure and Qwen2.5-GGUF loading under v5 are distinct model/config incompatibilities." + }, + { + "left": "issue:44291", + "right": "issue:44877", + "accept": false, + "reason": "Unexpected _is_hf_initialized argument in init_empty_weights is unrelated to strict config blocking granite_speech." + }, + { + "left": "issue:43632", + "right": "issue:45072", + "accept": false, + "reason": "_is_hf_initialized flag breakage is a loading/init regression, not a dtype mismatch in inference." + }, + { + "left": "issue:43873", + "right": "issue:45701", + "accept": false, + "reason": "Quantization offloading behavior and version-dependent tokenization are different classes of bugs." + }, + { + "left": "issue:43701", + "right": "issue:45325", + "accept": false, + "reason": "Resume-from-checkpoint key mismatch is unrelated to Qwen2.5-VL rope index scaling." + }, + { + "left": "issue:42796", + "right": "issue:45381", + "accept": false, + "reason": "Tokenizer BPE marker handling and Qwen2.5-VL video position IDs are separate tokenizer vs multimodal-vision bugs." + }, + { + "left": "issue:43296", + "right": "issue:45800", + "accept": false, + "reason": "PaddleOCR-VL load failure in vLLM/transformers is not the same as torch 2.4.1 versus transformers 5.8.0 incompatibility." + }, + { + "left": "issue:44479", + "right": "issue:45823", + "accept": false, + "reason": "Video-input regression in Qwen-family models and Gemma4 PLE device mismatch are unrelated model families and failure mechanisms." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 13, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:34634", + "issue:35707", + "issue:36331", + "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: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:43606|issue:44265", + "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:41720|issue:45823", + "issue:44186|issue:44265", + "issue:43377|issue:44610", + "issue:46305|issue:46315", + "issue:43257|issue:45923", + "issue:44568|issue:45920", + "issue:43454|issue:45072", + "issue:42796|issue:46209", + "issue:43317|issue:43873", + "issue:43454|issue:43720", + "issue:43377|issue:44898", + "issue:40990|issue:42947", + "issue:42617|issue:42915", + "issue:43749|issue:44534", + "issue:43572|issue:45759", + "issue:43322|issue:46050", + "issue:43526|issue:43883", + "issue:41720|issue:43322", + "issue:36331|issue:42617", + "issue:43881|issue:44991", + "issue:44387|issue:45704" + ], + "prepared_review_unit_hash": "124facf1541ebf7269f694ffeb7d8b8b9095f013c593c5a818992584256ad044", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36830, + "estimated_input_tokens": 9208, + "estimated_eval_tokens": 18672 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly distinct bug reports across different models, tokenizers, loading paths, and training/inference features. The only loose thematic overlap is around offloading/quantization, but none of the soft pairs are clearly the same underlying bug or mergeable as one PR.", + "confidence": 0.82, + "canonical_issue_reason": "issue:43873 is the broadest representative of the offloading/quantization subgroup and has enough discussion/activity to serve as the best umbrella issue.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43873 best captures the common loader/offload failure theme present in the nearest related reports, while still being broader than any single model-specific incident.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43606", + "right": "issue:44265", + "accept": false, + "reason": "Different failure classes: CPU-offload device mismatch for Bark vs export/tracing failure with torch_compilable_check." + }, + { + "left": "issue:43262", + "right": "issue:43746", + "accept": false, + "reason": "Audio chat-template sampling-rate default bug vs a PEFT/local-checkpoint loading problem; no shared code path." + }, + { + "left": "issue:40990", + "right": "issue:45584", + "accept": false, + "reason": "Unrelated symptoms and components: perplexity regression on gpt-oss vs Whisper empty-transcription generation failure." + }, + { + "left": "issue:35707", + "right": "issue:41720", + "accept": false, + "reason": "Progressive generation with inputs_embeds/past_key_values is a different path from auto device-mapping cuda assert failures." + }, + { + "left": "issue:43881", + "right": "issue:44488", + "accept": false, + "reason": "Both are loading failures, but for different models and likely different root causes; not the same bug." + }, + { + "left": "issue:34634", + "right": "issue:35707", + "accept": false, + "reason": "BarkProcessor voice_preset issue is unrelated to generation with inputs_embeds and past_key_values." + }, + { + "left": "issue:43454", + "right": "issue:43550", + "accept": false, + "reason": "Weight-tying/generation corruption in AyaVision is unrelated to Bamba torch.compile+SDPA failures." + }, + { + "left": "issue:43606", + "right": "issue:43819", + "accept": false, + "reason": "Both mention latent/model loading behavior, but one is CPU offload mismatch and the other is DAC latent conversion mismatch." + }, + { + "left": "issue:39692", + "right": "issue:40444", + "accept": false, + "reason": "Doc-example model/processor mismatch and quantization failure is unrelated to iterable-dataset multi-image finetuning." + }, + { + "left": "issue:41720", + "right": "issue:45823", + "accept": false, + "reason": "Both involve device_map='auto', but one is a Qwen3 cuda assert and the other is a Gemma4 PLE forward device mismatch." + }, + { + "left": "issue:44186", + "right": "issue:44265", + "accept": false, + "reason": "Tokenizer crashes on NER/padding are unrelated to torch.export exportability issues." + }, + { + "left": "issue:43377", + "right": "issue:44610", + "accept": false, + "reason": "MIMI batching/padding-mask correctness is unrelated to OmDet-Turbo processor/model input-size mismatch." + }, + { + "left": "issue:46305", + "right": "issue:46315", + "accept": false, + "reason": "Streaming mask off-by-one and tokenizer performance regression are different bugs in different code paths." + }, + { + "left": "issue:43257", + "right": "issue:45923", + "accept": false, + "reason": "Accelerate+deepspeed MOE weight conversion failure is unrelated to a missing gradient-checkpointing flag." + }, + { + "left": "issue:44568", + "right": "issue:45920", + "accept": false, + "reason": "Both are tokenizer regressions, but they affect different APIs and models with different symptoms." + }, + { + "left": "issue:43454", + "right": "issue:45072", + "accept": false, + "reason": "Weight tying causing garbage generation is not the same as bf16 dtype mismatches in inference wrappers." + }, + { + "left": "issue:42796", + "right": "issue:46209", + "accept": false, + "reason": "Tokenizer BPE-marker handling and FP8 inference on B200 are unrelated." + }, + { + "left": "issue:43317", + "right": "issue:43873", + "accept": false, + "reason": "Both touch offloading/quantization, but one is a specific dequantized model load failure and the other is a broad offloading complaint without enough evidence of the same code-path." + }, + { + "left": "issue:43454", + "right": "issue:43720", + "accept": false, + "reason": "Missing lm_head weight tying and packed-weight unpacking during accelerate loading are different model-loading bugs." + }, + { + "left": "issue:43377", + "right": "issue:44898", + "accept": false, + "reason": "Padding-mask correctness in MIMI is unrelated to Perceiver resolution/interpolation failures." + }, + { + "left": "issue:40990", + "right": "issue:42947", + "accept": false, + "reason": "Perplexity regression on a model is unrelated to gradient-checkpointing behavior with PEFT LoRA." + }, + { + "left": "issue:42617", + "right": "issue:42915", + "accept": false, + "reason": "3d_parallel runtime failure and Qwen3Moe FineGrainedFP8Config failure are different execution paths." + }, + { + "left": "issue:43749", + "right": "issue:44534", + "accept": false, + "reason": "CPU RAM efficient loading breakage and non-persistent buffer junk are distinct v5 regressions." + }, + { + "left": "issue:43572", + "right": "issue:45759", + "accept": false, + "reason": "Missing pad_token_idx in StableLmConfig is unrelated to unwrapping text_config in composite Qwen multimodal configs." + }, + { + "left": "issue:43322", + "right": "issue:46050", + "accept": false, + "reason": "Llava Next loading segfault and sinq save/load breakage are different failure modes." + }, + { + "left": "issue:43526", + "right": "issue:43883", + "accept": false, + "reason": "BeitImageProcessorFast label reduction bug is unrelated to missing all_tied_weights_keys on MolmoForCausalLM." + }, + { + "left": "issue:41720", + "right": "issue:43322", + "accept": false, + "reason": "Qwen3 device-map assert and Llava Next segfault are both loading-related but not the same concrete bug." + }, + { + "left": "issue:36331", + "right": "issue:42617", + "accept": false, + "reason": "CustomTrainer compute_loss signature breakage is unrelated to 3d_parallel execution failure." + }, + { + "left": "issue:43881", + "right": "issue:44991", + "accept": false, + "reason": "Two tokenizer/model-loading regressions, but for different models and likely different fixes." + }, + { + "left": "issue:44387", + "right": "issue:45704", + "accept": false, + "reason": "Quantization OOM from reserved memory growth is unrelated to an apex RMSNorm memory leak." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 14, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "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:43531", + "issue:43550", + "issue:43572", + "issue:43577", + "issue:43606", + "issue:43646", + "issue:43708", + "issue:43720", + "issue:43873", + "issue:44008", + "issue:44186", + "issue:44361", + "issue:44423", + "issue:44466", + "issue:44610", + "issue:44671", + "issue:44743", + "issue:44756", + "issue:44877", + "issue:44898", + "issue:44977", + "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:41720|issue:44977", + "issue:36331|issue:42915", + "issue:44466|issue:45357", + "issue:44877|issue:44991", + "issue:43450|issue:45701", + "issue:40990|issue:45701", + "issue:43065|issue:46402", + "issue:41720|issue:42617", + "issue:43873|issue:44756", + "issue:44186|issue:45072", + "issue:26413|issue:28530", + "issue:43550|issue:43720", + "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:42915|issue:43317", + "issue:44008|issue:45823", + "issue:44610|issue:44898", + "issue:36010|issue:42617", + "issue:44423|issue:45406", + "issue:45632|issue:45684", + "issue:41720|issue:45161", + "issue:43531|issue:43572", + "issue:28530|issue:33666", + "issue:44671|issue:44743", + "issue:42915|issue:46291", + "issue:36331|issue:41720", + "issue:44186|issue:44361" + ], + "prepared_review_unit_hash": "968c754f131bbd8ba603fc4c03b14911a97b15be46b44aa8641eb095ab697f97", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 31, + "serialized_chars": 36504, + "estimated_input_tokens": 9126, + "estimated_eval_tokens": 18508 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These are broad semantic-similarity matches, but the paired items mostly describe different concrete bugs or model-specific regressions. I reject all soft edges; there are no PRs in this set. The most representative issue is the detailed Trainer regression in #43708.", + "confidence": 0.94, + "canonical_issue_reason": "issue:43708 is the most concrete, actionable regression report here, with a clearly identified code path (`resume_from_checkpoint` / `max_steps`) and a specific reproduction condition.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43708 best represents the cluster because it is well-scoped, reproducible, and more clearly described than the other candidates.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:41720", + "right": "issue:44977", + "accept": false, + "reason": "Different failures: one is a CUDA assert during auto device mapping, the other is flash-attention generation behavior for Qwen3.5." + }, + { + "left": "issue:36331", + "right": "issue:42915", + "accept": false, + "reason": "Unrelated bugs: a Trainer API signature change versus a Qwen3MoE FP8 model failure." + }, + { + "left": "issue:44466", + "right": "issue:45357", + "accept": false, + "reason": "Both involve serialization in v5, but one is tied `lm_head.weight` behavior and the other is wrong visual encoder keys; not the same bug." + }, + { + "left": "issue:44877", + "right": "issue:44991", + "accept": false, + "reason": "Different loading failures affecting different model assets and code paths: strict config handling vs tokenizer loading." + }, + { + "left": "issue:43450", + "right": "issue:45701", + "accept": false, + "reason": "Video batch-shape bug is unrelated to a version-dependent tokenization change." + }, + { + "left": "issue:40990", + "right": "issue:45701", + "accept": false, + "reason": "High perplexity on a specific language model is not the same underlying issue as version-dependent tokenization differences." + }, + { + "left": "issue:43065", + "right": "issue:46402", + "accept": false, + "reason": "Different model components and failures: a dummy Conv2d in Sam3PixelDecoder versus missing text encoder weights in Sam3Model." + }, + { + "left": "issue:41720", + "right": "issue:42617", + "accept": false, + "reason": "Qwen3 auto device-mapping CUDA assertion and inability to run `3d_parallel.py` are distinct runtime problems." + }, + { + "left": "issue:43873", + "right": "issue:44756", + "accept": false, + "reason": "Quantization/offload behavior and Strix Halo mmap OOM are related only broadly by memory usage, not the same defect." + }, + { + "left": "issue:44186", + "right": "issue:45072", + "accept": false, + "reason": "Tokenizer crash on NER/batched padding is unrelated to bfloat16 dtype mismatches in inference." + }, + { + "left": "issue:26413", + "right": "issue:28530", + "accept": false, + "reason": "Checkpoint resume epoch-iterator failure and early-stopping metric lookup are separate Trainer features with different failure modes." + }, + { + "left": "issue:43550", + "right": "issue:43720", + "accept": false, + "reason": "One is a torch.compile/SDPA failure in Bamba; the other is an Accelerate loading issue for packed BitNet weights." + }, + { + "left": "issue:43577", + "right": "issue:43708", + "accept": false, + "reason": "BLIP2 dtype propagation and Trainer checkpoint-resume max-step calculation are unrelated." + }, + { + "left": "issue:43404", + "right": "issue:45357", + "accept": false, + "reason": "Weight tying in Mistral3 and incorrect saved visual encoder keys are different serialization/model-structure bugs." + }, + { + "left": "issue:43550", + "right": "issue:43606", + "accept": false, + "reason": "Both are model/runtime failures, but the code paths differ: SDPA compilation versus CPU-offload device mismatch." + }, + { + "left": "issue:43646", + "right": "issue:46665", + "accept": false, + "reason": "Both concern custom model usage, but one is a Transformers 5.0 initialization breakage and the other is inconsistent `from_pretrained` behavior inside Python classes." + }, + { + "left": "issue:43295", + "right": "issue:45701", + "accept": false, + "reason": "Processor/tokenizer regression in v4.57.5 is not the same bug as general tokenization changes across versions." + }, + { + "left": "issue:44186", + "right": "issue:44610", + "accept": false, + "reason": "Different preprocessing bugs: LayoutLMv2 tokenizer crashes versus OmDet processor output-size mismatch." + }, + { + "left": "issue:42915", + "right": "issue:43317", + "accept": false, + "reason": "Qwen3Moe FP8 failure and dequantized model offload loading failure are different quantization/offload issues." + }, + { + "left": "issue:44008", + "right": "issue:45823", + "accept": false, + "reason": "Gemma3n variable-name collision in forward is unrelated to Gemma4 PLE device mismatch under `device_map=auto`." + }, + { + "left": "issue:44610", + "right": "issue:44898", + "accept": false, + "reason": "Both are vision-input shape problems, but they affect different models and the symptoms are not the same underlying bug." + }, + { + "left": "issue:36010", + "right": "issue:42617", + "accept": false, + "reason": "ImportError for `GenerationMixin` and failure to run `3d_parallel.py` are unrelated." + }, + { + "left": "issue:44423", + "right": "issue:45406", + "accept": false, + "reason": "Both hit `transformers serve`, but one is a `'str' object has no attribute 'to'` crash and the other is a missing `_tokenizer` attribute; different failures." + }, + { + "left": "issue:45632", + "right": "issue:45684", + "accept": false, + "reason": "Cache-path collision for local trust-remote-code models and read-only permission propagation during save_pretrained are unrelated." + }, + { + "left": "issue:43531", + "right": "issue:43572", + "accept": false, + "reason": "Qwen3-MoE sliding-window behavior and missing `pad_token_idx` in StableLmConfig are different configuration/model bugs." + }, + { + "left": "issue:28530", + "right": "issue:33666", + "accept": false, + "reason": "Early stopping metric resolution and Qwen2-VL multi-GPU training are not the same issue." + }, + { + "left": "issue:44671", + "right": "issue:44743", + "accept": false, + "reason": "CamemBERT masked-LM prediction regression and Qwen3.5 recurrent-state reset are unrelated model bugs." + }, + { + "left": "issue:42915", + "right": "issue:46291", + "accept": false, + "reason": "Qwen3Moe FP8 failure and kernels import requiring version/revision are different dependency/runtime problems." + }, + { + "left": "issue:36331", + "right": "issue:41720", + "accept": false, + "reason": "Trainer compute_loss signature mismatch is unrelated to a Qwen3 auto-device CUDA assertion." + }, + { + "left": "issue:44186", + "right": "issue:44361", + "accept": false, + "reason": "Different tokenizer failures affecting different models and tasks; not the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 15, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "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:43526", + "issue:43531", + "issue:43540", + "issue:43550", + "issue:43606", + "issue:43644", + "issue:43653", + "issue:43720", + "issue:43746", + "issue:43782", + "issue:43873", + "issue:43881", + "issue:44077", + "issue:44387", + "issue:44451", + "issue:44492", + "issue:44509", + "issue:44534", + "issue:44589", + "issue:44625", + "issue:44913", + "issue:44991", + "issue:45059", + "issue:45357", + "issue:45406", + "issue:45464", + "issue:45479", + "issue:45701", + "issue:45704", + "issue:45759", + "issue:45859", + "issue:46402", + "issue:46665" + ], + "soft_pairs": [ + "issue:43526|issue:43881", + "issue:43316|issue:44913", + "issue:42915|issue:43322", + "issue:28530|issue:30064", + "issue:33357|issue:35707", + "issue:43540|issue:43782", + "issue:30064|issue:34634", + "issue:43873|issue:45704", + "issue:43550|issue:43653", + "issue:44387|issue:45701", + "issue:44492|issue:44509", + "issue:29159|issue:36032", + "issue:43606|issue:43720", + "issue:44534|issue:46665", + "issue:43531|issue:45759", + "issue:41720|issue:43746", + "issue:42490|issue:43329", + "issue:43388|issue:45059", + "issue:43606|issue:43653", + "issue:45406|issue:45464", + "issue:34567|issue:35707", + "issue:34634|issue:41720", + "issue:45357|issue:46402", + "issue:43531|issue:44589", + "issue:44451|issue:44991", + "issue:45759|issue:45859", + "issue:44625|issue:45479", + "issue:44077|issue:44625", + "issue:43644|issue:46665" + ], + "prepared_review_unit_hash": "6fbd2ab79d1d091db3747769dd09470e9f48f6f179a6601d83be68017e303394", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36369, + "estimated_input_tokens": 9093, + "estimated_eval_tokens": 18442 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is over-broad and mostly contains unrelated issue reports. The only clear duplicate-looking pair is the two Transformers v5 non-persistent buffer junk reports; the rest differ by model, subsystem, or bug class.", + "confidence": 0.88, + "canonical_issue_reason": "issue:44534 is the clearest representative of the one true duplicate pair here: it uses the broader, version-agnostic wording and matches issue:43644\u2019s buffer-junk report. The rest of the cluster is not one cohesive issue.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44534 is the best single issue to anchor the duplicate pair because it is the more general and later wording of the same underlying bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43526", + "right": "issue:43881", + "accept": false, + "reason": "Different components and failures: BeitImageProcessorFast label reduction vs glm-4v-9b loading failure." + }, + { + "left": "issue:43316", + "right": "issue:44913", + "accept": false, + "reason": "Both are config-related, but one is Gemma3TextConfig API mismatch and the other is GPTNeoX rotary_pct persistence." + }, + { + "left": "issue:42915", + "right": "issue:43322", + "accept": false, + "reason": "Both are loading failures, but for different models and likely different root causes (Qwen3Moe FP8 vs Llava Next segfault)." + }, + { + "left": "issue:28530", + "right": "issue:30064", + "accept": false, + "reason": "Early stopping metric lookup bug vs image processor void segmentation map handling; unrelated subsystems." + }, + { + "left": "issue:33357", + "right": "issue:35707", + "accept": false, + "reason": "MacOS CLIP bus error vs generation with inputs_embeds/past_key_values; not the same bug." + }, + { + "left": "issue:43540", + "right": "issue:43782", + "accept": false, + "reason": "Both involve Qwen multimodal models, but one is video input processing and the other is from_pretrained weight_only loading." + }, + { + "left": "issue:30064", + "right": "issue:34634", + "accept": false, + "reason": "Different image processor/tokenizer bugs; one is segmentation map void handling, the other BarkProcessor voice_preset behavior." + }, + { + "left": "issue:43873", + "right": "issue:45704", + "accept": false, + "reason": "Both mention memory/offloading-related problems, but one is quantization offloading and the other is apex FusedRMSNorm leak." + }, + { + "left": "issue:43550", + "right": "issue:43653", + "accept": false, + "reason": "Different models and failure modes: Bamba torch.compile SDPA vs BigBirdTokenizer special-token decode issue." + }, + { + "left": "issue:44387", + "right": "issue:45701", + "accept": false, + "reason": "Int4 quantization OOM vs tokenization changes across versions; not the same underlying defect." + }, + { + "left": "issue:44492", + "right": "issue:44509", + "accept": false, + "reason": "A typo in cache strategies is unrelated to removed pipeline docs." + }, + { + "left": "issue:29159", + "right": "issue:36032", + "accept": false, + "reason": "Tokenizer inconsistency vs T5Tokenizer add_special_tokens name conflict; different API problems." + }, + { + "left": "issue:43606", + "right": "issue:43720", + "accept": false, + "reason": "Both CI/loading issues, but Bark CPU offload device mismatch is not the same as BitNet packed-weight unpacking." + }, + { + "left": "issue:44534", + "right": "issue:46665", + "accept": false, + "reason": "One is a specific v5 buffer initialization bug; the other is a generic from_pretrained behavior inconsistency." + }, + { + "left": "issue:43531", + "right": "issue:45759", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior vs from_config not unwrapping text_config; different code paths." + }, + { + "left": "issue:41720", + "right": "issue:43746", + "accept": false, + "reason": "Qwen3 auto device mapping CUDA assert vs GraniteSpeech PEFT checkpoint loading; not the same issue." + }, + { + "left": "issue:42490", + "right": "issue:43329", + "accept": false, + "reason": "Trailing-slash import warning vs multimodal token counting video-branch bug; unrelated." + }, + { + "left": "issue:43388", + "right": "issue:45059", + "accept": false, + "reason": "Metric gathering label truncation vs SAM3 text/bbox behavior; different abstractions." + }, + { + "left": "issue:43606", + "right": "issue:43653", + "accept": false, + "reason": "Both are CI failures, but one is offload/device mismatch and the other is tokenizer special-token registration." + }, + { + "left": "issue:45406", + "right": "issue:45464", + "accept": false, + "reason": "Gemma4Processor missing _tokenizer in serve vs Qwen3.5 streaming chat/completions failure; different failures." + }, + { + "left": "issue:34567", + "right": "issue:35707", + "accept": false, + "reason": "TrainerState token counting bug vs generation with inputs_embeds/past_key_values; no shared root cause." + }, + { + "left": "issue:34634", + "right": "issue:41720", + "accept": false, + "reason": "Bark voice_preset issue and Qwen3 device-mapping CUDA assert are unrelated." + }, + { + "left": "issue:45357", + "right": "issue:46402", + "accept": false, + "reason": "Qwen3.5 save_pretrained visual encoder key regression vs Sam3Model text encoder weights not loaded; different save/load paths." + }, + { + "left": "issue:43531", + "right": "issue:44589", + "accept": false, + "reason": "Sliding-window Qwen3-MoE bug vs float8 storage lookup error; different model/runtime issues." + }, + { + "left": "issue:44451", + "right": "issue:44991", + "accept": false, + "reason": "Both are model/tokenizer load failures, but for different checkpoints and likely different tokenizer incompatibilities." + }, + { + "left": "issue:45759", + "right": "issue:45859", + "accept": false, + "reason": "One is AutoModelForCausalLM config unwrapping; the other is CPU offload compatibility in Qwen3_5MoeTextRotaryEmbedding." + }, + { + "left": "issue:44625", + "right": "issue:45479", + "accept": false, + "reason": "num_labels propagation in Qwen3.5 config is distinct from degenerate zero-loss behavior in sequence classification." + }, + { + "left": "issue:44077", + "right": "issue:44625", + "accept": false, + "reason": "PatchTSMixer post_init allowance and Qwen3.5 num_labels propagation are unrelated config issues." + }, + { + "left": "issue:43644", + "right": "issue:44534", + "accept": true, + "reason": "Same underlying Transformers v5 bug: non-persistent buffers being filled with junk; titles and symptoms match closely." + }, + { + "left": "issue:43644", + "right": "issue:46665", + "accept": false, + "reason": "Buffer initialization bug is unrelated to from_pretrained behavior inside Python classes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 16, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:30064", + "issue:36010", + "issue:39692", + "issue:41720", + "issue:42886", + "issue:43381", + "issue:43421", + "issue:43531", + "issue:43596", + "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:44220", + "issue:44368", + "issue:44451", + "issue:44514", + "issue:44545", + "issue:44610", + "issue:44671", + "issue:44843", + "issue:44898", + "issue:44918", + "issue:44936", + "issue:44987", + "issue:45042", + "issue:45072", + "issue:45081", + "issue:45356", + "issue:45357", + "issue:45381", + "issue:45704", + "issue:45759", + "issue:45907", + "issue:45923", + "issue:46032", + "issue:46050", + "issue:46463", + "issue:46655", + "issue:46693" + ], + "soft_pairs": [ + "issue:44062|issue:44451", + "issue:44898|issue:45072", + "issue:44987|issue:45042", + "issue:44368|issue:45357", + "issue:43381|issue:43844", + "issue:39692|issue:44610", + "issue:41720|issue:45704", + "issue:45704|issue:45923", + "issue:43531|issue:43645", + "issue:46655|issue:46693", + "issue:43746|issue:46050", + "issue:43381|issue:45704", + "issue:43421|issue:43927", + "issue:44918|issue:44936", + "issue:42886|issue:44843", + "issue:45759|issue:45923", + "issue:44514|issue:44545", + "issue:46032|issue:46463", + "issue:39692|issue:43873", + "issue:43746|issue:43873", + "issue:43994|issue:44671", + "issue:43421|issue:43653", + "issue:30064|issue:36010", + "issue:43782|issue:45381", + "issue:43421|issue:44186", + "issue:45081|issue:45356", + "issue:45356|issue:45907", + "issue:43720|issue:43819", + "issue:43653|issue:43720", + "issue:43596|issue:44220" + ], + "prepared_review_unit_hash": "76926e5e3b0c6b04db2736d16353650835d81344e2786c5d819f8da8ce84ee48", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 37070, + "estimated_input_tokens": 9268, + "estimated_eval_tokens": 18792 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly heterogeneous across unrelated tokenizer, vision, training, quantization, and loading bugs. Only one clear duplicate pair stands out: the two Qwen2_5_VLProcessor batched-input padding=False crash reports.", + "confidence": 0.93, + "canonical_issue_reason": "issue:44545 is the best canonical issue because it is an exact duplicate of issue:44514 (same title and same batched-input crash with padding=False).", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44545 is the strongest issue representative for the only true duplicate pair in this cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44062", + "right": "issue:44451", + "accept": false, + "reason": "Different problems: AddedToken constructor special-arg error vs model loading failure for ScandiBERT." + }, + { + "left": "issue:44898", + "right": "issue:45072", + "accept": false, + "reason": "Different vision regressions: Perceiver input-resolution handling vs bfloat16 dtype mismatch in SwitchTransformers/TimmWrapperModel." + }, + { + "left": "issue:44987", + "right": "issue:45042", + "accept": false, + "reason": "One is a model-loading regression; the other is a PIL backend processor dependency issue. Not the same bug." + }, + { + "left": "issue:44368", + "right": "issue:45357", + "accept": false, + "reason": "Different code paths: tie_word_embeddings warning vs incorrect visual encoder keys saved by Qwen3.5." + }, + { + "left": "issue:43381", + "right": "issue:43844", + "accept": false, + "reason": "Gradient-checkpointing eval-mode guard and ZeRO-3 gradient blow-up are distinct failures." + }, + { + "left": "issue:39692", + "right": "issue:44610", + "accept": false, + "reason": "Both involve processor/model mismatch, but for different models and different defects." + }, + { + "left": "issue:41720", + "right": "issue:45704", + "accept": false, + "reason": "Qwen3 auto device mapping CUDA assert is unrelated to the T5 apex RMSNorm memory leak." + }, + { + "left": "issue:45704", + "right": "issue:45923", + "accept": false, + "reason": "Memory leak in T5 RMSNorm is unrelated to a missing supports_gradient_checkpointing flag." + }, + { + "left": "issue:43531", + "right": "issue:43645", + "accept": false, + "reason": "Sliding-window behavior in Qwen3-MoE is unrelated to custom-model initialization in Jupyter notebooks." + }, + { + "left": "issue:46655", + "right": "issue:46693", + "accept": false, + "reason": "Video padding calculation bug vs long-context FA2 performance regression; different underlying issues." + }, + { + "left": "issue:43746", + "right": "issue:46050", + "accept": false, + "reason": "Local checkpoint loading with PEFT adapters is not the same as sinq quantized save/load corruption." + }, + { + "left": "issue:43381", + "right": "issue:45704", + "accept": false, + "reason": "Different training/runtime bugs: eval-mode checkpointing restriction vs T5 memory leak." + }, + { + "left": "issue:43421", + "right": "issue:43927", + "accept": false, + "reason": "Both touch token handling, but one is runtime post-processor refresh and the other is config persistence of custom token IDs." + }, + { + "left": "issue:44918", + "right": "issue:44936", + "accept": false, + "reason": "Different trainer failures: SFT embedding unpacking vs evaluate() failing after train()." + }, + { + "left": "issue:42886", + "right": "issue:44843", + "accept": false, + "reason": "Both are offline-related, but one is generic cache loading and the other is a specific mistral-regex model_info call path." + }, + { + "left": "issue:45759", + "right": "issue:45923", + "accept": false, + "reason": "Composite-config unwrapping in from_config is unrelated to a missing gradient-checkpointing flag." + }, + { + "left": "issue:44514", + "right": "issue:44545", + "accept": true, + "reason": "Exact duplicate: same title and same batched-input crash when padding=False." + }, + { + "left": "issue:46032", + "right": "issue:46463", + "accept": false, + "reason": "Mamba2 use_cache correctness bug is unrelated to a T5 error-message validation issue." + }, + { + "left": "issue:39692", + "right": "issue:43873", + "accept": false, + "reason": "Different topics: SigLIP2 doc example errors vs quantization offloading behavior." + }, + { + "left": "issue:43746", + "right": "issue:43873", + "accept": false, + "reason": "PEFT checkpoint loading and quantization offloading are different code paths and different failures." + }, + { + "left": "issue:43994", + "right": "issue:44671", + "accept": false, + "reason": "Different models and symptoms: SigLIP2 nonsensical outputs vs CamemBERT masked-LM regression." + }, + { + "left": "issue:43421", + "right": "issue:43653", + "accept": false, + "reason": "Both involve special tokens, but one is backend post-processor updates and the other is BigBird mask-token registration/decode output." + }, + { + "left": "issue:30064", + "right": "issue:36010", + "accept": false, + "reason": "Image segmentation-map processing and GenerationMixin import failure are unrelated." + }, + { + "left": "issue:43782", + "right": "issue:45381", + "accept": false, + "reason": "Both are Qwen VL issues, but one is weight_only loading and the other is incorrect video vision_position_ids." + }, + { + "left": "issue:43596", + "right": "issue:44220", + "accept": false, + "reason": "Deepspeed ZeRO-3/BertModel init error is unrelated to _torch_extract_fbank_features()." + }, + { + "left": "issue:45081", + "right": "issue:45356", + "accept": false, + "reason": "Both mention fix_mistral_regex, but one crashes on backend_tokenizer access while the other is a codec-handling regression with a warning issue." + }, + { + "left": "issue:45356", + "right": "issue:45907", + "accept": false, + "reason": "Tokenizer regression in Kimi-K2.5 is unrelated to list_repo_files not forwarding revision." + }, + { + "left": "issue:43720", + "right": "issue:43819", + "accept": false, + "reason": "BitNet packed-weight loading and DAC latent reconstruction mismatch are different bugs." + }, + { + "left": "issue:43653", + "right": "issue:43720", + "accept": false, + "reason": "BigBird special-token decode bug is unrelated to BitNet accelerate-loading unpack behavior." + }, + { + "left": "issue:43596", + "right": "issue:44220", + "accept": false, + "reason": "Different subsystems and failure modes; not a duplicate pair." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The cluster summary is well grounded in the packet: it identifies one genuinely duplicate pair (44514/44545) and the remaining soft-edge rejections are cautious and supported by the issue titles. No overstatement beyond the evidence is apparent." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 17, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:30064", + "issue:34634", + "issue:36010", + "issue:36331", + "issue:39692", + "issue:41720", + "issue:43295", + "issue:43322", + "issue:43335", + "issue:43531", + "issue:43540", + "issue:43550", + "issue:43611", + "issue:43618", + "issue:43643", + "issue:43644", + "issue:43653", + "issue:43697", + "issue:43749", + "issue:43756", + "issue:43819", + "issue:43927", + "issue:44008", + "issue:44062", + "issue:44186", + "issue:44220", + "issue:44442", + "issue:44448", + "issue:44485", + "issue:44488", + "issue:44509", + "issue:44661", + "issue:44671", + "issue:44821", + "issue:44843", + "issue:44877", + "issue:44898", + "issue:44991", + "issue:45072", + "issue:45081", + "issue:45084", + "issue:45381", + "issue:45584", + "issue:45685", + "issue:45834", + "issue:46294", + "issue:46665", + "issue:46693" + ], + "soft_pairs": [ + "issue:43550|issue:45072", + "issue:43295|issue:45834", + "issue:43550|issue:43927", + "issue:43540|issue:46294", + "issue:44062|issue:44488", + "issue:45381|issue:46693", + "issue:43618|issue:43697", + "issue:43531|issue:43643", + "issue:43611|issue:46665", + "issue:44220|issue:44661", + "issue:43653|issue:43819", + "issue:44186|issue:44898", + "issue:30064|issue:36331", + "issue:43643|issue:44877", + "issue:43335|issue:43756", + "issue:41720|issue:44991", + "issue:43550|issue:43819", + "issue:44821|issue:45084", + "issue:43540|issue:44008", + "issue:44843|issue:45081", + "issue:39692|issue:45834", + "issue:44485|issue:44509", + "issue:44448|issue:44671", + "issue:44186|issue:44442", + "issue:43644|issue:43749", + "issue:43322|issue:45685", + "issue:34634|issue:36010", + "issue:45584|issue:45834" + ], + "prepared_review_unit_hash": "14175961c4a33ba78afe31bbfc1a23ef4836be96f17040943f466629a68e4cf8", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36082, + "estimated_input_tokens": 9021, + "estimated_eval_tokens": 18298 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous; the soft pairs look like different bugs across unrelated models/features rather than true duplicates. No soft edge should be merged.", + "confidence": 0.26, + "canonical_issue_reason": "If a single issue must anchor the cluster, issue 43611 is the broadest user-facing regression report: a Transformers 5.0.0 loading break that can generalize across multiple model classes.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43611 is the best representative because it is a broad core regression report rather than a narrow model-specific symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43550", + "right": "issue:45072", + "accept": false, + "reason": "Different failure modes: torch.compile+SDPA on Bamba vs bfloat16 dtype mismatch in SwitchTransformers/TimmWrapperModel." + }, + { + "left": "issue:43295", + "right": "issue:45834", + "accept": false, + "reason": "Unrelated domains: processor.tokenizer regression vs Kosmos2.5 OCR length/indexing error." + }, + { + "left": "issue:43550", + "right": "issue:43927", + "accept": false, + "reason": "Different code paths and symptoms: compiled SDPA failure vs config save/load losing token IDs." + }, + { + "left": "issue:43540", + "right": "issue:46294", + "accept": false, + "reason": "Both involve Qwen variants, but one is sliding_window/MoE and the other is video chat-template token/feature mismatch." + }, + { + "left": "issue:44062", + "right": "issue:44488", + "accept": false, + "reason": "Token construction bug vs tokenizer loading failure for a specific model; not the same underlying defect." + }, + { + "left": "issue:45381", + "right": "issue:46693", + "accept": false, + "reason": "Different regressions: Qwen2.5-VL video position IDs vs FA2 long-context performance degradation." + }, + { + "left": "issue:43618", + "right": "issue:43697", + "accept": false, + "reason": "Both are v5 model regressions, but on different architectures and different observable failures." + }, + { + "left": "issue:43531", + "right": "issue:43643", + "accept": false, + "reason": "Qwen3-MoE sliding_window bug is unrelated to trust_remote_code config field loss." + }, + { + "left": "issue:43611", + "right": "issue:46665", + "accept": false, + "reason": "Both touch loading, but one is a base_model_prefix regression in v5 and the other is inconsistent from_pretrained behavior inside Python classes." + }, + { + "left": "issue:44220", + "right": "issue:44661", + "accept": false, + "reason": "Audio feature extraction bug vs add-new-model script failure in tokenizer mapping; different subsystems." + }, + { + "left": "issue:43653", + "right": "issue:43819", + "accept": false, + "reason": "Tokenizer special-token registration bug vs DAC latent reconstruction mismatch; no shared code path." + }, + { + "left": "issue:44186", + "right": "issue:44898", + "accept": false, + "reason": "Tokenizer/NER padding crash vs Perceiver image classification resize/interpolation failure." + }, + { + "left": "issue:30064", + "right": "issue:36331", + "accept": false, + "reason": "Image segmentation-map processing failure is unrelated to Trainer.compute_loss signature changes." + }, + { + "left": "issue:43643", + "right": "issue:44877", + "accept": false, + "reason": "Both involve config loading, but one loses custom fields under trust_remote_code and the other is strict config rejection of granite_speech." + }, + { + "left": "issue:43335", + "right": "issue:43756", + "accept": false, + "reason": "Different model configuration bugs: sparse layer creation vs RoPE layer dropping." + }, + { + "left": "issue:41720", + "right": "issue:44991", + "accept": false, + "reason": "CUDA assert during Qwen3 device mapping is unrelated to tokenizer loading for EMBEDDIA/est-roberta." + }, + { + "left": "issue:43550", + "right": "issue:43819", + "accept": false, + "reason": "SDPA/compile runtime failure and DAC.from_latents mismatch are distinct bugs." + }, + { + "left": "issue:44821", + "right": "issue:45084", + "accept": false, + "reason": "AutoImageProcessor URL loading and template compilation errors are unrelated." + }, + { + "left": "issue:43540", + "right": "issue:44008", + "accept": false, + "reason": "Video-input validation error in Qwen3OmniMoe vs variable-name collision in Gemma3n forward." + }, + { + "left": "issue:44843", + "right": "issue:45081", + "accept": false, + "reason": "Both mention _patch_mistral_regex, but one is offline model_info access and the other is backend_tokenizer attribute access; separate defects." + }, + { + "left": "issue:39692", + "right": "issue:45834", + "accept": false, + "reason": "Doc/example errors for SigLIP2 vs long OCR input failure in Kosmos2.5." + }, + { + "left": "issue:44485", + "right": "issue:44509", + "accept": false, + "reason": "GLM-5 RoPE implementation discussion is unrelated to docs still listing removed pipeline tasks." + }, + { + "left": "issue:44448", + "right": "issue:44671", + "accept": false, + "reason": "Different model-specific output regressions in v5, not the same underlying change." + }, + { + "left": "issue:44186", + "right": "issue:44442", + "accept": false, + "reason": "LayoutLMv2 tokenizer padding/NER crash and FastSpeech2Conformer AutoTokenizer loading failure are different tokenizer issues." + }, + { + "left": "issue:43644", + "right": "issue:43749", + "accept": false, + "reason": "Non-persistent buffer junk values and FSDP CPU RAM efficient loading failure are separate loading/runtime problems." + }, + { + "left": "issue:43322", + "right": "issue:45685", + "accept": false, + "reason": "Llava Next loading segfault is unrelated to MPS histogram support for Int in MoE." + }, + { + "left": "issue:34634", + "right": "issue:36010", + "accept": false, + "reason": "BarkProcessor voice_preset behavior and GenerationMixin import error are unrelated." + }, + { + "left": "issue:45584", + "right": "issue:45834", + "accept": false, + "reason": "Whisper empty-transcription generation bug is unrelated to Kosmos2.5 long OCR input indexing." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 18, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:28218", + "issue:29159", + "issue:34634", + "issue:36331", + "issue:42175", + "issue:43388", + "issue:43421", + "issue:43540", + "issue:43577", + "issue:43582", + "issue:43606", + "issue:43645", + "issue:43716", + "issue:43720", + "issue:43749", + "issue:43782", + "issue:43818", + "issue:43825", + "issue:43827", + "issue:43950", + "issue:43994", + "issue:44008", + "issue:44291", + "issue:44479", + "issue:44561", + "issue:44610", + "issue:44671", + "issue:44805", + "issue:44811", + "issue:44936", + "issue:44977", + "issue:44991", + "issue:45020", + "issue:45072", + "issue:45081", + "issue:45183", + "issue:45381", + "issue:45479", + "issue:45701", + "issue:45834", + "issue:45987", + "issue:46032", + "issue:46192", + "issue:46291", + "issue:46294", + "issue:46463", + "issue:46512", + "issue:46665" + ], + "soft_pairs": [ + "issue:43606|issue:45072", + "issue:45183|issue:45479", + "issue:43716|issue:44008", + "issue:43388|issue:46032", + "issue:44479|issue:45834", + "issue:43950|issue:45987", + "issue:43582|issue:46512", + "issue:43825|issue:43994", + "issue:44991|issue:46291", + "issue:43577|issue:43818", + "issue:46294|issue:46463", + "issue:43782|issue:44977", + "issue:44561|issue:45020", + "issue:43421|issue:44811", + "issue:44805|issue:44936", + "issue:43540|issue:45381", + "issue:44610|issue:45072", + "issue:43720|issue:45072", + "issue:43825|issue:43827", + "issue:43749|issue:44291", + "issue:28218|issue:29159", + "issue:43645|issue:46665", + "issue:43577|issue:43582", + "issue:45987|issue:46192", + "issue:44671|issue:45701", + "issue:42175|issue:45081", + "issue:34634|issue:36331" + ], + "prepared_review_unit_hash": "cde82b580fd372ade7235eabb839ae372b4c1ebfd4d5785888d85c8815c2bf2b", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 36340, + "estimated_input_tokens": 9085, + "estimated_eval_tokens": 18426 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is a loose mix of unrelated Transformers issues; none of the soft pairs look like the same underlying bug/change, so no merges are justified. The best single anchor is the highly discussed FSDP loading regression issue, but even that is only a representative issue, not a duplicate hub.", + "confidence": 0.89, + "canonical_issue_reason": "issue:43749 is the strongest central anchor by discussion activity and inbound references, and it is a concrete regression report with substantial debugging context.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43749 is the most suitable global representative because it has the most engagement and a clearly scoped, actionable failure mode.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43606", + "right": "issue:45072", + "accept": false, + "reason": "Both mention mismatch errors, but they affect different models and different paths: CPU offload/device placement vs bfloat16 dtype handling." + }, + { + "left": "issue:45183", + "right": "issue:45479", + "accept": false, + "reason": "One is about input_ids/inputs_embeds validation messaging; the other is a loss/computation bug in sequence classification. Different problems." + }, + { + "left": "issue:43716", + "right": "issue:44008", + "accept": false, + "reason": "Different model families and different root causes: image-preprocessor dtype mismatch vs an audio variable-name collision." + }, + { + "left": "issue:43388", + "right": "issue:46032", + "accept": false, + "reason": "Different subsystems and failure modes: metrics gathering on the last batch vs incorrect Mamba2 cache results." + }, + { + "left": "issue:44479", + "right": "issue:45834", + "accept": false, + "reason": "Both are vision/video related, but they describe different models and distinct regressions." + }, + { + "left": "issue:43950", + "right": "issue:45987", + "accept": false, + "reason": "from_pretrained buffer corruption and StaticCache return-type issues are unrelated bugs." + }, + { + "left": "issue:43582", + "right": "issue:46512", + "accept": false, + "reason": "Apple Silicon warmup TypeError is unrelated to from_config leaving config.dtype stale." + }, + { + "left": "issue:43825", + "right": "issue:43994", + "accept": false, + "reason": "An error-message regression is not the same as a model output correctness bug." + }, + { + "left": "issue:44991", + "right": "issue:46291", + "accept": false, + "reason": "Tokenizer-loading failure for a specific model and kernels package import failure are different breakages." + }, + { + "left": "issue:43577", + "right": "issue:43818", + "accept": false, + "reason": "Blip2 dtype-loading regression and Video-LLaVA architecture/weight-sharing issue do not share a concrete code path." + }, + { + "left": "issue:46294", + "right": "issue:46463", + "accept": false, + "reason": "One concerns Qwen video chat template feature alignment; the other is T5 decoder-input validation messaging." + }, + { + "left": "issue:43782", + "right": "issue:44977", + "accept": false, + "reason": "Qwen3VL weight-only loading and Qwen3.5 flash-attention generation failures are different issues." + }, + { + "left": "issue:44561", + "right": "issue:45020", + "accept": false, + "reason": "Both involve trust_remote_code/remote_code breakage, but the specific removed API and the broader compatibility regression are not the same bug." + }, + { + "left": "issue:43421", + "right": "issue:44811", + "accept": false, + "reason": "Runtime special-token post-processing update and Whisper skip_special_tokens handling are separate tokenizer behaviors." + }, + { + "left": "issue:44805", + "right": "issue:44936", + "accept": false, + "reason": "Mask shape IndexError and trainer.evaluate() failing after train are unrelated training/runtime issues." + }, + { + "left": "issue:43540", + "right": "issue:45381", + "accept": false, + "reason": "Different Qwen video regressions with different symptoms and likely different fixes." + }, + { + "left": "issue:44610", + "right": "issue:45072", + "accept": false, + "reason": "Processor image-size mismatch and dtype mismatch in inference are unrelated." + }, + { + "left": "issue:43720", + "right": "issue:45072", + "accept": false, + "reason": "Packed-weight unpacking during accelerate loading is a different failure mode from dtype mismatches." + }, + { + "left": "issue:43825", + "right": "issue:43827", + "accept": false, + "reason": "One is an error message bug; the other is stale documentation after pipeline removals." + }, + { + "left": "issue:43749", + "right": "issue:44291", + "accept": false, + "reason": "These are both loading-related, but one is FSDP CPU RAM-efficient loading and the other is init_empty_weights argument handling." + }, + { + "left": "issue:28218", + "right": "issue:29159", + "accept": false, + "reason": "Tokenizer spacing after added tokens and slow/fast tokenizer inconsistency are too broad and not clearly the same underlying bug." + }, + { + "left": "issue:43645", + "right": "issue:46665", + "accept": false, + "reason": "Jupyter custom-model initialization and from_pretrained-in-classes inconsistencies are different usage scenarios." + }, + { + "left": "issue:43577", + "right": "issue:43582", + "accept": false, + "reason": "Blip2 dtype persistence and Apple Silicon allocator warmup TypeError are unrelated." + }, + { + "left": "issue:45987", + "right": "issue:46192", + "accept": false, + "reason": "A cache-length contract bug and text_kwargs/return_tensors handling are different API issues." + }, + { + "left": "issue:44671", + "right": "issue:45701", + "accept": false, + "reason": "CamemBERT masked-LM regression and version-dependent tokenization changes are not the same code-path problem." + }, + { + "left": "issue:42175", + "right": "issue:45081", + "accept": false, + "reason": "Backend dependency packaging and Mistral regex patch/tokenizer attribute access are unrelated." + }, + { + "left": "issue:34634", + "right": "issue:36331", + "accept": false, + "reason": "BarkProcessor voice_preset and CustomTrainer.compute_loss signature errors are entirely separate issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 19, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:36010", + "issue:36331", + "issue:39692", + "issue:42175", + "issue:42490", + "issue:43257", + "issue:43262", + "issue:43329", + "issue:43452", + "issue:43526", + "issue:43531", + "issue:43550", + "issue:43582", + "issue:43638", + "issue:43645", + "issue:43653", + "issue:43697", + "issue:43720", + "issue:43742", + "issue:43782", + "issue:43819", + "issue:43906", + "issue:43927", + "issue:44060", + "issue:44368", + "issue:44554", + "issue:44589", + "issue:44610", + "issue:44857", + "issue:44912", + "issue:44936", + "issue:44991", + "issue:45003", + "issue:45020", + "issue:45042", + "issue:45081", + "issue:45084", + "issue:45325", + "issue:45372", + "issue:45406", + "issue:45538", + "issue:45600", + "issue:46209", + "issue:46285", + "issue:46291", + "issue:46326", + "issue:46697", + "issue:46728" + ], + "soft_pairs": [ + "issue:45003|issue:46285", + "issue:43653|issue:44912", + "issue:46697|issue:46728", + "issue:43582|issue:44857", + "issue:43262|issue:43819", + "issue:44589|issue:45042", + "issue:43742|issue:45020", + "issue:43782|issue:45325", + "issue:45081|issue:45406", + "issue:45372|issue:46326", + "issue:43550|issue:44912", + "issue:45084|issue:45538", + "issue:44060|issue:44368", + "issue:43526|issue:43697", + "issue:43329|issue:45600", + "issue:43526|issue:44554", + "issue:43452|issue:44991", + "issue:43257|issue:43329", + "issue:36010|issue:36331", + "issue:36010|issue:43645", + "issue:43906|issue:46209", + "issue:43819|issue:43927", + "issue:43531|issue:45042", + "issue:43638|issue:44936", + "issue:42175|issue:46291", + "issue:43720|issue:44610", + "issue:42490|issue:43257", + "issue:39692|issue:44912" + ], + "prepared_review_unit_hash": "3171dd26f9576afb0a038c7ac239502b504594adbde96a2297f821da8ebb4d8e", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36578, + "estimated_input_tokens": 9145, + "estimated_eval_tokens": 18546 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab bag of mostly unrelated Transformers issues; the soft pairs are only superficially similar and do not appear to describe the same underlying bug or change.", + "confidence": 0.89, + "canonical_issue_reason": "issue:45020 is the best canonical issue because it is broad, user-facing, and describes a core regression class ('recent versions break models using remote_code') that can subsume multiple model-loading failures.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45020 is the strongest representative issue in the set due to its broad scope and centrality to model-loading regressions.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45003", + "right": "issue:46285", + "accept": false, + "reason": "Both touch model internals, but one is about unsafe sys.modules access and the other about CLIPTextModel structure changes; different bugs and code paths." + }, + { + "left": "issue:43653", + "right": "issue:44912", + "accept": false, + "reason": "BigBirdTokenizer special-token decoding and MXFP4 quantization/loading fallback are unrelated." + }, + { + "left": "issue:46697", + "right": "issue:46728", + "accept": false, + "reason": "Both are image-processor sizing issues, but they affect different models and different calculations (over-padding vs patch-count math)." + }, + { + "left": "issue:43582", + "right": "issue:44857", + "accept": false, + "reason": "Apple Silicon allocator warmup TypeError and CUDA float16 AMP crash are different runtime/platform failures." + }, + { + "left": "issue:43262", + "right": "issue:43819", + "accept": false, + "reason": "Audio chat-template sampling-rate defaulting and DAC latent/forward mismatch are unrelated audio bugs." + }, + { + "left": "issue:44589", + "right": "issue:45042", + "accept": false, + "reason": "A missing Float8 storage type and a PIL backend torchvision dependency regression are not the same problem." + }, + { + "left": "issue:43742", + "right": "issue:45020", + "accept": false, + "reason": "A specific key error while loading one model is not the same as the broad remote_code regression." + }, + { + "left": "issue:43782", + "right": "issue:45325", + "accept": false, + "reason": "One is a Qwen3VL weight_only load error; the other is a Qwen2.5-VL rope-index scaling bug. Different model families and failures." + }, + { + "left": "issue:45081", + "right": "issue:45406", + "accept": false, + "reason": "Both involve tokenizer/processor internals, but one is a Mistral regex patch crash and the other is a Gemma4 processor missing _tokenizer; not the same bug." + }, + { + "left": "issue:45372", + "right": "issue:46326", + "accept": false, + "reason": "Both mention mistral_common, but one is an import failure for ReasoningEffort and the other is a continue_final_message behavior bug." + }, + { + "left": "issue:43550", + "right": "issue:44912", + "accept": false, + "reason": "torch.compile+SDPA failure and MXFP4 quantization fallback are unrelated." + }, + { + "left": "issue:45084", + "right": "issue:45538", + "accept": false, + "reason": "Template compilation TypeError and CLIPTokenizer max_length semantics are different issues." + }, + { + "left": "issue:44060", + "right": "issue:44368", + "accept": false, + "reason": "They are both about Qwen weight-tying warnings, but one reports an incorrect tie to dt_bias while the other is a repeated warning during fine-tuning; not clearly the same underlying defect." + }, + { + "left": "issue:43526", + "right": "issue:43697", + "accept": false, + "reason": "BeitImageProcessorFast label reduction and RTDetrV2 output drift in Transformers v5 are unrelated." + }, + { + "left": "issue:43329", + "right": "issue:45600", + "accept": false, + "reason": "A multimodal-token counting bug and stale auto_mappings config references are different code paths." + }, + { + "left": "issue:43526", + "right": "issue:44554", + "accept": false, + "reason": "Different components and failure modes: image-label reduction vs MPS attention correctness." + }, + { + "left": "issue:43452", + "right": "issue:44991", + "accept": false, + "reason": "Both are loading-related, but gguf_file incompatibility and a tokenizer failure for a specific model are not the same issue." + }, + { + "left": "issue:43257", + "right": "issue:43329", + "accept": false, + "reason": "Qwen3 MoE weight conversion under accelerate/deepspeed and multimodal video token counting are unrelated." + }, + { + "left": "issue:36010", + "right": "issue:36331", + "accept": false, + "reason": "Importing GenerationMixin and CustomTrainer.compute_loss signature changes are separate API breakages." + }, + { + "left": "issue:36010", + "right": "issue:43645", + "accept": false, + "reason": "GenerationMixin import failure and Jupyter custom-model initialization regression are distinct issues." + }, + { + "left": "issue:43906", + "right": "issue:46209", + "accept": false, + "reason": "An isolated reproduction of an older issue and an FP8 inference problem on B200 are not the same bug." + }, + { + "left": "issue:43819", + "right": "issue:43927", + "accept": false, + "reason": "DAC latent reconstruction mismatch and DiaConfig token-ID persistence are unrelated." + }, + { + "left": "issue:43531", + "right": "issue:45042", + "accept": false, + "reason": "Qwen3-MoE sliding-window behavior and PIL backend torchvision dependency regression have no common underlying cause." + }, + { + "left": "issue:43638", + "right": "issue:44936", + "accept": false, + "reason": "Zero3/Bert training crash and evaluate-after-train failure are different trainer failures." + }, + { + "left": "issue:42175", + "right": "issue:46291", + "accept": false, + "reason": "Packaging TensorFlow with '.[torch]' and kernels import/version requirements are separate dependency issues." + }, + { + "left": "issue:43720", + "right": "issue:44610", + "accept": false, + "reason": "BitNet packed-weight loading and OmDet-Turbo processor image-size mismatch are unrelated." + }, + { + "left": "issue:42490", + "right": "issue:43257", + "accept": false, + "reason": "A trailing-slash local-path import warning is unrelated to Qwen3 MoE accelerate/deepspeed conversion." + }, + { + "left": "issue:39692", + "right": "issue:44912", + "accept": false, + "reason": "SigLIP2 documentation errors and git-oss-20b MXFP4 quantization fallback are different problems despite both mentioning quantization." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 20, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:28282", + "issue:29159", + "issue:29870", + "issue:34567", + "issue:38175", + "issue:40990", + "issue:42915", + "issue:43257", + "issue:43262", + "issue:43329", + "issue:43450", + "issue:43454", + "issue:43525", + "issue:43540", + "issue:43577", + "issue:43606", + "issue:43643", + "issue:43650", + "issue:43696", + "issue:43742", + "issue:43746", + "issue:43749", + "issue:43854", + "issue:43906", + "issue:43927", + "issue:43931", + "issue:44291", + "issue:44387", + "issue:44479", + "issue:44610", + "issue:44617", + "issue:44756", + "issue:44811", + "issue:44821", + "issue:44912", + "issue:44977", + "issue:44987", + "issue:45042", + "issue:45459", + "issue:45538", + "issue:45759", + "issue:45803", + "issue:45859", + "issue:46291", + "issue:46421", + "issue:46424", + "issue:46701" + ], + "soft_pairs": [ + "issue:38175|issue:44610", + "issue:43746|issue:44811", + "issue:43643|issue:45759", + "issue:44387|issue:44617", + "issue:43854|issue:44977", + "issue:43257|issue:43696", + "issue:43257|issue:43931", + "issue:43262|issue:44811", + "issue:46421|issue:46424", + "issue:42915|issue:43742", + "issue:43329|issue:44479", + "issue:45459|issue:45803", + "issue:43329|issue:43450", + "issue:43749|issue:44756", + "issue:44291|issue:45042", + "issue:42915|issue:44821", + "issue:43606|issue:45859", + "issue:34567|issue:40990", + "issue:43650|issue:43906", + "issue:43540|issue:46701", + "issue:43927|issue:44912", + "issue:43454|issue:43577", + "issue:43525|issue:46291", + "issue:28282|issue:29870", + "issue:29159|issue:34567", + "issue:44987|issue:45538" + ], + "prepared_review_unit_hash": "d6c4cfae12711efa71fb74a7c786713d7694989a536ef722234b668e31e6acfc", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 34989, + "estimated_input_tokens": 8748, + "estimated_eval_tokens": 17752 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is mostly heterogeneous: loading, tokenizer, quantization, multimodal/video, cache, and memory issues are all mixed together. No PRs are present. The only near-looking pair is the video-path issues, but they still read as separate regressions rather than one duplicate fix.", + "confidence": 0.93, + "canonical_issue_reason": "Issue 43329 is the most concrete, code-path-specific report in the set: it names undefined helpers in the video branch of `_get_num_multimodal_tokens`. It is the best single anchor for the small video-regression subset, though the overall cluster is too mixed for a true universal canonical.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43329 is the strongest representative issue because it has the clearest failing helper/function and the most actionable root cause among the items, but it only fits a narrow subset of the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:38175", + "right": "issue:44610", + "accept": false, + "reason": "Different bugs: one is zero-probability behavior in SigLIP2, the other is an OmDet-Turbo processor/model input-size mismatch." + }, + { + "left": "issue:43746", + "right": "issue:44811", + "accept": false, + "reason": "Separate code paths: GraniteSpeech PEFT/local-checkpoint loading vs Whisper `batch_decode()` skipping special-token handling." + }, + { + "left": "issue:43643", + "right": "issue:45759", + "accept": false, + "reason": "Both concern config/model loading, but one is `trust_remote_code` dropping fields in `AutoConfig`, while the other is `AutoModelForCausalLM.from_config` not unwrapping `text_config` for composite Qwen configs." + }, + { + "left": "issue:44387", + "right": "issue:44617", + "accept": false, + "reason": "Both mention OOM, but the underlying problems are different: int4 quantization reserved-memory growth vs a Sam3Video CUDA-memory failure." + }, + { + "left": "issue:43854", + "right": "issue:44977", + "accept": false, + "reason": "Different model-specific failures: GLM-4.7-Flash loading in tests vs Qwen3.5 generation with flash-attention." + }, + { + "left": "issue:43257", + "right": "issue:43696", + "accept": false, + "reason": "Qwen3 MOE weight-conversion/loading under accelerate+deepspeed is unrelated to a generic GPT-oss-20b CUDA OOM." + }, + { + "left": "issue:43257", + "right": "issue:43931", + "accept": false, + "reason": "Both involve model loading, but the symptoms and models differ: MOE weight conversion vs Qwen3-VL shape mismatch." + }, + { + "left": "issue:43262", + "right": "issue:44811", + "accept": false, + "reason": "Different audio-processing bugs: chat-template sampling-rate defaulting vs Whisper processor decode parameters." + }, + { + "left": "issue:46421", + "right": "issue:46424", + "accept": false, + "reason": "Same broad cache area, but distinct defects: spurious recompilation from lazy cache init vs `max_cache_length` only applying to Quantized Cache." + }, + { + "left": "issue:42915", + "right": "issue:43742", + "accept": false, + "reason": "Unrelated failures: Qwen3 MOE + FineGrainedFP8Config vs a key error loading MobileLLM-125M." + }, + { + "left": "issue:43329", + "right": "issue:44479", + "accept": false, + "reason": "Both are video/multimodal regressions, but 43329 is a specific undefined-variable/helper bug in `_get_num_multimodal_tokens` while 44479 is a broader v5.3.0 video-input regression across several models; not enough evidence they are the same fix." + }, + { + "left": "issue:45459", + "right": "issue:45803", + "accept": false, + "reason": "Both involve swallowed errors, but the functions and failure modes differ: protobuf decode fallback masking tokenizer errors vs bare `except:` in `FuyuBatchFeature.convert_to_tensors()`." + }, + { + "left": "issue:43329", + "right": "issue:43450", + "accept": false, + "reason": "Different video bugs: token counting in the multimodal helper vs incorrect batched video processor output shape." + }, + { + "left": "issue:43749", + "right": "issue:44756", + "accept": false, + "reason": "Separate memory/loading problems: FSDP CPU RAM efficient loading is broken, while Strix Halo mmap is being disabled to avoid OOM." + }, + { + "left": "issue:44291", + "right": "issue:45042", + "accept": false, + "reason": "Unrelated issues: `init_empty_weights` passes an unexpected `_is_hf_initialized` argument, whereas PIL image processors incorrectly require torchvision." + }, + { + "left": "issue:43606", + "right": "issue:45859", + "accept": false, + "reason": "Both mention CPU offload, but the model/code path is different: suno/bark-small device mismatch vs Qwen3_5MoeTextRotaryEmbedding incompatibility." + }, + { + "left": "issue:34567", + "right": "issue:40990", + "accept": false, + "reason": "No relation between TrainerState token-count tracking and GPT-oss-20b perplexity on WikiText-2." + }, + { + "left": "issue:43650", + "right": "issue:43906", + "accept": false, + "reason": "`ADD THE DATA` is not a meaningful duplicate target, and the other item is an isolated reproduction of a separate issue." + }, + { + "left": "issue:43540", + "right": "issue:46701", + "accept": false, + "reason": "Different multimodal failures: Qwen3OmniMoe video processing ValueError vs DiffusionGemma image-text-to-text pipeline malfunction." + }, + { + "left": "issue:43927", + "right": "issue:44912", + "accept": false, + "reason": "Separate concerns: DiaConfig losing custom token IDs on save/load vs MXFP4 quantization falling back to bf16." + }, + { + "left": "issue:43454", + "right": "issue:43577", + "accept": false, + "reason": "Different model-loading bugs: AyaVision weight tying/generation corruption vs BLIP-2 dtype propagation remaining float32." + }, + { + "left": "issue:43525", + "right": "issue:46291", + "accept": false, + "reason": "No overlap: missing `pad_token_id` in Llama4Config vs pipeline import failing because kernels v0.15 requires version/revision." + }, + { + "left": "issue:28282", + "right": "issue:29870", + "accept": false, + "reason": "Unrelated: missing PyTorch for AutoModel import vs a Bart generation/evaluation `UnboundLocalError`." + }, + { + "left": "issue:29159", + "right": "issue:34567", + "accept": false, + "reason": "Tokenizer slow/fast inconsistency is unrelated to TrainerState's `num_input_tokens_seen` not updating." + }, + { + "left": "issue:44987", + "right": "issue:45538", + "accept": false, + "reason": "Different bugs in different areas: loading `physical-intelligence/fast` under transformers>=5.1.0 vs CLIPTokenizer's `model_max_length` constant." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 21, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:24643", + "issue:27957", + "issue:28282", + "issue:29127", + "issue:29159", + "issue:33357", + "issue:34567", + "issue:34689", + "issue:38175", + "issue:41720", + "issue:42175", + "issue:42915", + "issue:43317", + "issue:43404", + "issue:43425", + "issue:43452", + "issue:43540", + "issue:43577", + "issue:43582", + "issue:43697", + "issue:43742", + "issue:43749", + "issue:43824", + "issue:44062", + "issue:44291", + "issue:44479", + "issue:44488", + "issue:44560", + "issue:44568", + "issue:44821", + "issue:44849", + "issue:44898", + "issue:44987", + "issue:44991", + "issue:45127", + "issue:45137", + "issue:45356", + "issue:45357", + "issue:45381", + "issue:45405", + "issue:45488", + "issue:45538", + "issue:45701", + "issue:45907", + "issue:46291", + "issue:46512", + "issue:46693" + ], + "soft_pairs": [ + "issue:44568|issue:45356", + "issue:28282|issue:34689", + "issue:43404|issue:45127", + "issue:43425|issue:46291", + "issue:42915|issue:45538", + "issue:43697|issue:45357", + "issue:45137|issue:45405", + "issue:33357|issue:34567", + "issue:42175|issue:43824", + "issue:43742|issue:44291", + "issue:28282|issue:29127", + "issue:44479|issue:46693", + "issue:43317|issue:45538", + "issue:44062|issue:44991", + "issue:43577|issue:46512", + "issue:42915|issue:43749", + "issue:43452|issue:43582", + "issue:44849|issue:45701", + "issue:41720|issue:43540", + "issue:27957|issue:28282", + "issue:43540|issue:44560", + "issue:44291|issue:44821", + "issue:24643|issue:29159", + "issue:42175|issue:44488", + "issue:43540|issue:44479", + "issue:44987|issue:45907", + "issue:44479|issue:45381", + "issue:38175|issue:44898", + "issue:44568|issue:45488" + ], + "prepared_review_unit_hash": "971ffbef58123bbe74ac086d2b321e3411bbe628c6e5be54252cd9e267355bd8", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 35858, + "estimated_input_tokens": 8965, + "estimated_eval_tokens": 18186 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a loose similarity cluster of unrelated Transformers regressions; only the Qwen video-input reports look like the same underlying bug.", + "confidence": 0.74, + "canonical_issue_reason": "issue:44479 is the broadest and most representative report in the one clearly matching subgroup: it covers the Qwen video-input regression across multiple variants, while 45381 narrows to qwen2.5-vl and its vision_position_ids symptom.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44479 is the strongest global representative: it describes the broadest concrete failure mode among the items and best captures the shared Qwen video regression that appears to recur in the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44568", + "right": "issue:45356", + "accept": false, + "reason": "Different tokenizer bugs: mdeberta special-token insertion vs Kimi-K2.5 codec/regex regression." + }, + { + "left": "issue:28282", + "right": "issue:34689", + "accept": false, + "reason": "Generic PyTorch-missing import error vs a model-specific Llama 3.2 vision loading regression." + }, + { + "left": "issue:43404", + "right": "issue:45127", + "accept": false, + "reason": "Different model-path bugs: untied lm_head in Mistral3 vs LoRA merge collapse after vocab extension." + }, + { + "left": "issue:43425", + "right": "issue:46291", + "accept": false, + "reason": "Torch version incompatibility is unrelated to the kernels v0.15 revision/version import failure." + }, + { + "left": "issue:42915", + "right": "issue:45538", + "accept": false, + "reason": "Qwen3Moe FP8 config failure and CLIPTokenizer max-length bug are unrelated." + }, + { + "left": "issue:43697", + "right": "issue:45357", + "accept": false, + "reason": "RTDetrV2 output drift in v5 is unrelated to Qwen3.5 save_pretrained visual-key corruption." + }, + { + "left": "issue:45137", + "right": "issue:45405", + "accept": false, + "reason": "DeepSpeed ZeRO3 deque error is unrelated to the PEFT minimum-version bump." + }, + { + "left": "issue:33357", + "right": "issue:34567", + "accept": false, + "reason": "MacOS CLIP bus error and TrainerState token-count bookkeeping are different subsystems and failures." + }, + { + "left": "issue:42175", + "right": "issue:43824", + "accept": false, + "reason": "Packaging/backend install issue vs missing Qwen2_5_VL import symbol; not the same bug." + }, + { + "left": "issue:43742", + "right": "issue:44291", + "accept": false, + "reason": "Different load-time problems: MobileLLM key error vs unexpected _is_hf_initialized arg in init_empty_weights." + }, + { + "left": "issue:28282", + "right": "issue:29127", + "accept": false, + "reason": "Missing-PyTorch ImportError is unrelated to LayoutLMv3 box-validation messaging." + }, + { + "left": "issue:44479", + "right": "issue:46693", + "accept": false, + "reason": "Video-input regression is unrelated to the FA2 long-context performance regression." + }, + { + "left": "issue:43317", + "right": "issue:45538", + "accept": false, + "reason": "device_map/offload dequantization failure is unrelated to CLIPTokenizer max_length." + }, + { + "left": "issue:44062", + "right": "issue:44991", + "accept": false, + "reason": "Both are tokenizer-related, but one is AddedToken argument duplication and the other is a model-loading regression for est-roberta." + }, + { + "left": "issue:43577", + "right": "issue:46512", + "accept": false, + "reason": "Both mention dtype, but one is BLIP2 module casting and the other is stale config.dtype from from_config." + }, + { + "left": "issue:42915", + "right": "issue:43749", + "accept": false, + "reason": "Qwen3Moe FP8 config failure and FSDP CPU-RAM-efficient-loading break are different code paths." + }, + { + "left": "issue:43452", + "right": "issue:43582", + "accept": false, + "reason": "gguf loading failure and Apple Silicon caching_allocator_warmup TypeError are unrelated." + }, + { + "left": "issue:44849", + "right": "issue:45701", + "accept": false, + "reason": "Qwen3.5 hidden-states bug is unrelated to a general tokenization-version change report." + }, + { + "left": "issue:41720", + "right": "issue:43540", + "accept": false, + "reason": "Auto device-mapping CUDA assert and Qwen3OmniMoe video ValueError are different failures." + }, + { + "left": "issue:27957", + "right": "issue:28282", + "accept": false, + "reason": "FlashAttention2 support for XLMRoberta is unrelated to missing PyTorch installation." + }, + { + "left": "issue:43540", + "right": "issue:44560", + "accept": false, + "reason": "Both are video-input failures, but different Qwen model families and error modes with no clear shared code path." + }, + { + "left": "issue:44291", + "right": "issue:44821", + "accept": false, + "reason": "init_empty_weights TypeError and AutoImageProcessor URL loading are unrelated load paths." + }, + { + "left": "issue:24643", + "right": "issue:29159", + "accept": false, + "reason": "DeepSpeed weight-shape RuntimeError and tokenizer slow/fast inconsistency are unrelated." + }, + { + "left": "issue:42175", + "right": "issue:44488", + "accept": false, + "reason": "TensorFlow not included in a torch install extra is unrelated to loading a specific model." + }, + { + "left": "issue:43540", + "right": "issue:44479", + "accept": false, + "reason": "Both are video-related, but different Qwen model stacks and distinct failures; not enough to call them the same bug." + }, + { + "left": "issue:44987", + "right": "issue:45907", + "accept": false, + "reason": "Loading a specific model in transformers>=5.1.0 is unrelated to list_repo_files needing revision forwarding." + }, + { + "left": "issue:44479", + "right": "issue:45381", + "accept": true, + "reason": "Both point to the same Qwen video-input regression; 45381\u2019s wrong vision_position_ids symptom fits the broader 44479 failure mode." + }, + { + "left": "issue:38175", + "right": "issue:44898", + "accept": false, + "reason": "Siglip2 zero probabilities and Perceiver non-default-resolution failure are different vision-model bugs." + }, + { + "left": "issue:44568", + "right": "issue:45488", + "accept": false, + "reason": "mdeberta special-token insertion is unrelated to the LlamaTokenizer ByteLevel/Metaspace override regression." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The only accepted soft pair, issue:44479 vs issue:45381, is plausibly the same Qwen video-input regression and looks mergeable as a single bugfix. The other verdicts are appropriately cautious and do not overstate relatedness." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 22, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:24643", + "issue:28282", + "issue:29870", + "issue:32090", + "issue:34567", + "issue:36331", + "issue:43257", + "issue:43404", + "issue:43452", + "issue:43526", + "issue:43531", + "issue:43540", + "issue:43577", + "issue:43716", + "issue:43742", + "issue:43756", + "issue:43906", + "issue:44062", + "issue:44488", + "issue:44568", + "issue:44589", + "issue:44625", + "issue:44737", + "issue:44792", + "issue:44821", + "issue:44877", + "issue:44913", + "issue:44964", + "issue:45042", + "issue:45072", + "issue:45081", + "issue:45325", + "issue:45357", + "issue:45405", + "issue:45479", + "issue:45588", + "issue:45600", + "issue:45823", + "issue:45920", + "issue:45987", + "issue:46121", + "issue:46399", + "issue:46463", + "issue:46468", + "issue:46491", + "issue:46693", + "issue:46701" + ], + "soft_pairs": [ + "issue:43257|issue:43756", + "issue:34567|issue:36331", + "issue:44792|issue:44964", + "issue:44062|issue:45081", + "issue:43742|issue:44589", + "issue:44737|issue:46463", + "issue:45042|issue:45081", + "issue:43716|issue:46701", + "issue:43756|issue:46399", + "issue:43531|issue:44821", + "issue:43756|issue:45325", + "issue:44568|issue:46491", + "issue:43577|issue:45072", + "issue:45405|issue:45600", + "issue:43906|issue:46468", + "issue:45823|issue:46463", + "issue:24643|issue:29870", + "issue:43404|issue:45987", + "issue:44737|issue:45588", + "issue:43452|issue:44488", + "issue:43906|issue:46693", + "issue:44625|issue:44877", + "issue:43526|issue:45479", + "issue:45357|issue:45920", + "issue:43540|issue:46693", + "issue:44913|issue:46121", + "issue:28282|issue:32090" + ], + "prepared_review_unit_hash": "c13e9d1690635973771fce1eb4d245e7d07c1c278ce88118aa5c5a70635f4676", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 27, + "serialized_chars": 35593, + "estimated_input_tokens": 8899, + "estimated_eval_tokens": 18054 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is heterogeneous: most pairs share only superficial subsystem overlap (tokenizers, multimodal models, configs, dtype, loading) but describe different concrete failures and fixes. I found no true duplicate pair among the soft edges.", + "confidence": 0.96, + "canonical_issue_reason": "issue:45920 is the strongest representative issue: it is a broad, high-impact v5 regression affecting multiple model families and clearly describes a concrete user-facing tokenizer failure.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45920 is the best global issue candidate because it has the broadest scope and clearest regression framing, making it the most representative triage anchor for this set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43257", + "right": "issue:43756", + "accept": false, + "reason": "Both mention model-specific loading/regression problems, but one is about Qwen3 MoE weight conversion under accelerate+deepspeed and the other about Smollm3 RoPE layer dropping; different components and fixes." + }, + { + "left": "issue:34567", + "right": "issue:36331", + "accept": false, + "reason": "Both involve Trainer behavior, but num_input_tokens_seen not updating is a state/metrics bug while compute_loss getting num_items_in_batch is an API signature incompatibility." + }, + { + "left": "issue:44792", + "right": "issue:44964", + "accept": false, + "reason": "Both are multimodal model failures, but one is a failed janus image generation test and the other is a Phi-4-multimodal model loading error; not the same code path." + }, + { + "left": "issue:44062", + "right": "issue:45081", + "accept": false, + "reason": "Both are tokenizer-related crashes, but one is AddedToken duplication with special tokens and the other is a Mistral regex patch/backend_tokenizer attribute crash; distinct bugs." + }, + { + "left": "issue:43742", + "right": "issue:44589", + "accept": false, + "reason": "These are separate loading failures: MobileLLM key lookup vs float8 storage deserialization; no shared root cause is evident." + }, + { + "left": "issue:44737", + "right": "issue:46463", + "accept": false, + "reason": "XLNet CPU allocation in rotary positional encoding and T5 decoder-input validation error messaging are unrelated." + }, + { + "left": "issue:45042", + "right": "issue:45081", + "accept": false, + "reason": "One is an image-processor dependency issue requiring torchvision, the other is a tokenizer regex patch crash; different subsystems and fixes." + }, + { + "left": "issue:43716", + "right": "issue:46701", + "accept": false, + "reason": "Both touch multimodal generation, but one is a dtype mismatch in Mistral-3 image preprocessing and the other is a DiffusionGemma image-text-to-text pipeline failure." + }, + { + "left": "issue:43756", + "right": "issue:46399", + "accept": false, + "reason": "Both reference RoPE/kernel behavior, but Smollm3 drops RoPE layers while Qwen3.5 fails to register apply_rotary_pos_emb under use_kernels=True; not the same defect." + }, + { + "left": "issue:43531", + "right": "issue:44821", + "accept": false, + "reason": "Qwen3-MoE sliding-window behavior and AutoImageProcessor URL loading are unrelated." + }, + { + "left": "issue:43756", + "right": "issue:45325", + "accept": false, + "reason": "Both involve Qwen3.5/Qwen2.5-VL and positional handling, but one is about RoPE layer dropping and the other about temporal position_ids scaling for still images." + }, + { + "left": "issue:44568", + "right": "issue:46491", + "accept": false, + "reason": "Tokenizer regressions in different models: mdeberta special-token addition vs CodeLlama leading-space round-trip loss; separate tokenization logic." + }, + { + "left": "issue:43577", + "right": "issue:45072", + "accept": false, + "reason": "Both are dtype-mismatch reports, but they affect different architectures and code paths (BLIP2 qformer vs SwitchTransformers/TimmWrapperModel)." + }, + { + "left": "issue:45405", + "right": "issue:45600", + "accept": false, + "reason": "One is an external dependency version pin mismatch, the other is stale auto_mappings entries causing CI breakage; different maintenance issues." + }, + { + "left": "issue:43906", + "right": "issue:46468", + "accept": false, + "reason": "An isolated reproduction of one prior issue and an official Nemotron example breakage are not the same bug without stronger evidence." + }, + { + "left": "issue:45823", + "right": "issue:46463", + "accept": false, + "reason": "Gemma4 device-map mismatch during forward and T5 decoder input validation messaging are unrelated." + }, + { + "left": "issue:24643", + "right": "issue:29870", + "accept": false, + "reason": "DeepSpeed training weight-shape runtime error and Bart generate decoder_attention_mask unbound variable are different failures in different pipelines." + }, + { + "left": "issue:43404", + "right": "issue:45987", + "accept": false, + "reason": "Mistral3 lm_head weight tying and StaticCache sequence-length return type are unrelated model/utility bugs." + }, + { + "left": "issue:44737", + "right": "issue:45588", + "accept": false, + "reason": "XLNet positional encoding CPU placement and flash_attention s_aux=None crash are different code paths." + }, + { + "left": "issue:43452", + "right": "issue:44488", + "accept": false, + "reason": "Both involve model/tokenizer loading, but gguf_file regression in from_pretrained is not the same as a specific model no longer loading." + }, + { + "left": "issue:43906", + "right": "issue:46693", + "accept": false, + "reason": "An issue reproduction and a performance regression in long-context FA2 generation are not the same underlying bug." + }, + { + "left": "issue:44625", + "right": "issue:44877", + "accept": false, + "reason": "Both are config-related, but one is num_labels propagation in Qwen3.5 and the other is strict config blocking granite_speech loading." + }, + { + "left": "issue:43526", + "right": "issue:45479", + "accept": false, + "reason": "BeitImageProcessorFast reduce_labels behavior and sequence-classification zero-loss with num_labels=1 are unrelated label-handling bugs." + }, + { + "left": "issue:45357", + "right": "issue:45920", + "accept": false, + "reason": "One is incorrect visual encoder keys saved for Qwen3.5, the other is wrong token IDs from AutoTokenizer across several models; different failure modes." + }, + { + "left": "issue:43540", + "right": "issue:46693", + "accept": false, + "reason": "Qwen3OmniMoe video input processing and FA2 long-context performance regression do not share a concrete code-path failure." + }, + { + "left": "issue:44913", + "right": "issue:46121", + "accept": false, + "reason": "GPTNeoX rotary_pct persistence and JSON-loaded ignore_keys_at_rope_validation type handling are distinct configuration bugs." + }, + { + "left": "issue:28282", + "right": "issue:32090", + "accept": false, + "reason": "PyTorch missing ImportError for AutoModel and Trainer _gpu_broadcast_one NoneType type error are completely different 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": 23, + "review_unit_index": 23, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:29159", + "issue:33357", + "issue:38175", + "issue:41762", + "issue:42898", + "issue:43295", + "issue:43452", + "issue:43525", + "issue:43531", + "issue:43577", + "issue:43643", + "issue:43646", + "issue:43653", + "issue:43688", + "issue:43697", + "issue:43720", + "issue:43742", + "issue:43749", + "issue:43756", + "issue:43819", + "issue:43994", + "issue:44361", + "issue:44442", + "issue:44464", + "issue:44466", + "issue:44493", + "issue:44568", + "issue:44743", + "issue:44811", + "issue:44843", + "issue:44898", + "issue:44936", + "issue:45005", + "issue:45200", + "issue:45216", + "issue:45356", + "issue:45372", + "issue:45375", + "issue:45542", + "issue:45704", + "issue:45823", + "issue:46050", + "issue:46170", + "issue:46178", + "issue:46402", + "issue:46463", + "issue:46491", + "issue:46701" + ], + "soft_pairs": [ + "issue:43720|issue:44898", + "issue:43697|issue:45216", + "issue:43643|issue:45375", + "issue:44466|issue:45005", + "issue:45200|issue:46463", + "issue:41762|issue:44936", + "issue:44361|issue:44442", + "issue:44493|issue:46050", + "issue:43688|issue:43749", + "issue:43653|issue:44361", + "issue:43525|issue:46701", + "issue:43819|issue:44811", + "issue:43994|issue:44898", + "issue:43452|issue:44843", + "issue:38175|issue:43295", + "issue:44936|issue:45823", + "issue:45372|issue:46170", + "issue:43531|issue:43742", + "issue:45542|issue:46178", + "issue:42898|issue:46491", + "issue:43646|issue:43749", + "issue:43653|issue:44568", + "issue:43577|issue:43697", + "issue:43452|issue:43577", + "issue:43295|issue:45356", + "issue:44464|issue:44743", + "issue:29159|issue:33357", + "issue:43756|issue:46402", + "issue:43688|issue:45704" + ], + "prepared_review_unit_hash": "22a96c90be7313ede2b0f825a040f4c5886a85f80a7106199ddf636a678c4bc1", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36738, + "estimated_input_tokens": 9185, + "estimated_eval_tokens": 18626 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly heterogeneous transformer issues with two tokenizer-related themes. I\u2019m keeping issue:42898 as the broad tokenizer-regression representative, but I\u2019m rejecting the 42898\u219446491 duplicate claim because the evidence is too title-level and the concrete failures differ.", + "confidence": 0.71, + "canonical_issue_reason": "Issue:42898 is the broadest tokenizer-regression report in the set and the cleanest cluster representative for the v5 tokenization-behavior theme.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue:42898 is the best general representative because it describes the upstream tokenizer cleanup-behavior change directly, even though the cluster is otherwise mostly unrelated.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43720", + "right": "issue:44898", + "accept": false, + "reason": "Different subsystems and failure modes: BitNet packed-weight loading vs Perceiver resolution handling." + }, + { + "left": "issue:43697", + "right": "issue:45216", + "accept": false, + "reason": "RTDetrV2 output drift in v5 and a Qwen3.5 save_pretrained regression are different bugs." + }, + { + "left": "issue:43643", + "right": "issue:45375", + "accept": false, + "reason": "Both mention missing fields, but the specific loading paths and root causes differ." + }, + { + "left": "issue:44466", + "right": "issue:45005", + "accept": true, + "reason": "Both describe the same tied-weights/save_pretrained serialization regression in v5." + }, + { + "left": "issue:45200", + "right": "issue:46463", + "accept": false, + "reason": "Gemma4 mm_token_type_ids handling is unrelated to T5 decoder-input validation messaging." + }, + { + "left": "issue:41762", + "right": "issue:44936", + "accept": false, + "reason": "DeepSpeed ZeRO-3 loading IndexError is a different path from trainer.evaluate() failing after train()." + }, + { + "left": "issue:44361", + "right": "issue:44442", + "accept": false, + "reason": "Different tokenizer problems: MLukeTokenizer task AttributeError vs AutoTokenizer loading failure." + }, + { + "left": "issue:44493", + "right": "issue:46050", + "accept": false, + "reason": "Unexpected position-id keys and sinq quantization save/load breakage are unrelated issues." + }, + { + "left": "issue:43688", + "right": "issue:43749", + "accept": false, + "reason": "Aux-loss normalization in MoE models is unrelated to FSDP CPU RAM efficient loading." + }, + { + "left": "issue:43653", + "right": "issue:44361", + "accept": false, + "reason": "BigBirdTokenizer special-token registration and MLukeTokenizer task AttributeError are different bugs." + }, + { + "left": "issue:43525", + "right": "issue:46701", + "accept": false, + "reason": "Llama4Config pad_token_id and DiffusionGemma image-text pipeline failure are unrelated." + }, + { + "left": "issue:43819", + "right": "issue:44811", + "accept": false, + "reason": "DAC latent reconstruction mismatch and Whisper batch_decode skip_special_tokens handling are different issues." + }, + { + "left": "issue:43994", + "right": "issue:44898", + "accept": false, + "reason": "SigLIP2 nonsensical outputs are unrelated to Perceiver non-default-resolution failure." + }, + { + "left": "issue:43452", + "right": "issue:44843", + "accept": false, + "reason": "gguf_file loading breakage is a different code path from the offline-mode model_info() call." + }, + { + "left": "issue:38175", + "right": "issue:43295", + "accept": false, + "reason": "SigLIP2 zero probabilities and processor.tokenizer regression are different model/pipeline bugs." + }, + { + "left": "issue:44936", + "right": "issue:45823", + "accept": false, + "reason": "Trainer lifecycle failure is unrelated to Gemma4 device-mismatch during forward." + }, + { + "left": "issue:45372", + "right": "issue:46170", + "accept": false, + "reason": "Gemma4 processor import dependency breakage is unrelated to the Windows ROCm torch.distributed issue." + }, + { + "left": "issue:43531", + "right": "issue:43742", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior and MobileLLM loading KeyError are distinct problems." + }, + { + "left": "issue:45542", + "right": "issue:46178", + "accept": false, + "reason": "TensorBoard-without-TF backend setup and tensorflow_text backend mapping are separate dependency/backend issues." + }, + { + "left": "issue:42898", + "right": "issue:46491", + "accept": false, + "reason": "The titles point to a related tokenizer-regression theme, but not enough to treat them as the same concrete bug from title-level evidence alone: cleanup-behavior change vs. CodeLlama-specific leading-space stripping." + }, + { + "left": "issue:43646", + "right": "issue:43749", + "accept": false, + "reason": "Custom model initialization breakage is unrelated to FSDP CPU RAM efficient loading." + }, + { + "left": "issue:43653", + "right": "issue:44568", + "accept": false, + "reason": "Both involve tokenizer special tokens, but one is mask-token registration and the other is BOS/EOS insertion; not the same bug." + }, + { + "left": "issue:43577", + "right": "issue:46402", + "accept": false, + "reason": "Blip2 dtype propagation and Sam3Model text-encoder weight loading are different loading bugs." + }, + { + "left": "issue:43452", + "right": "issue:43577", + "accept": false, + "reason": "gguf_file loading failures are unrelated to Blip2 dtype remaining float32." + }, + { + "left": "issue:43295", + "right": "issue:45356", + "accept": false, + "reason": "processor.tokenizer/images regression is unrelated to the Kimi-K2.5 codec-handling issue." + }, + { + "left": "issue:44464", + "right": "issue:44743", + "accept": false, + "reason": "Chunked generation with compiled forward and Qwen3 recurrent-state reset are different generation bugs." + }, + { + "left": "issue:29159", + "right": "issue:33357", + "accept": false, + "reason": "Tokenizer slow/fast inconsistency is unrelated to the MacOS bus error on a CLIP community model." + }, + { + "left": "issue:43756", + "right": "issue:46402", + "accept": false, + "reason": "Smollm3 RoPE-layer dropping and Sam3Model text-encoder loading are unrelated model-definition bugs." + }, + { + "left": "issue:43688", + "right": "issue:45704", + "accept": false, + "reason": "MoE auxiliary-loss normalization is unrelated to T5's apex.FusedRMSNorm memory leak." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The cluster summary is broadly grounded, but one soft-edge verdict is too assertive for the packet: `issue:42898` vs `issue:46491`. The titles support a related tokenizer regression theme, but not enough to confidently call them the same bug from title-level evidence alone (`clean_up_tokenization_spaces` behavior change vs. a CodeLlama-specific encode\u2192decode leading-space strip). That makes the duplicate claim less conservative than required. The other judgments look more defensible." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 24, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:32090", + "issue:33666", + "issue:36010", + "issue:36331", + "issue:39692", + "issue:41720", + "issue:43262", + "issue:43278", + "issue:43295", + "issue:43493", + "issue:43504", + "issue:43525", + "issue:43550", + "issue:43572", + "issue:43577", + "issue:43606", + "issue:43643", + "issue:43653", + "issue:43688", + "issue:43720", + "issue:43742", + "issue:43782", + "issue:43825", + "issue:43927", + "issue:44361", + "issue:44464", + "issue:44466", + "issue:44479", + "issue:44484", + "issue:44554", + "issue:44561", + "issue:44610", + "issue:44625", + "issue:44704", + "issue:44734", + "issue:44821", + "issue:44843", + "issue:44964", + "issue:44977", + "issue:44991", + "issue:45405", + "issue:45440", + "issue:45491", + "issue:45701", + "issue:46002", + "issue:46050", + "issue:46612", + "issue:46693" + ], + "soft_pairs": [ + "issue:43653|issue:43927", + "issue:43688|issue:45701", + "issue:44843|issue:44991", + "issue:44479|issue:44625", + "issue:45491|issue:46002", + "issue:43550|issue:44361", + "issue:44484|issue:44704", + "issue:33666|issue:36010", + "issue:43688|issue:45440", + "issue:43742|issue:44821", + "issue:44734|issue:46612", + "issue:32090|issue:33666", + "issue:44466|issue:46050", + "issue:41720|issue:44464", + "issue:44464|issue:44977", + "issue:43504|issue:43550", + "issue:43504|issue:43720", + "issue:44977|issue:46693", + "issue:43504|issue:44610", + "issue:43577|issue:44554", + "issue:43720|issue:43927", + "issue:33666|issue:36331", + "issue:43643|issue:44561", + "issue:43278|issue:43688", + "issue:43606|issue:43927", + "issue:43782|issue:46693", + "issue:43262|issue:43295", + "issue:39692|issue:43493", + "issue:44964|issue:45405", + "issue:43720|issue:44361", + "issue:43493|issue:43825", + "issue:43525|issue:43572" + ], + "prepared_review_unit_hash": "19caa81b5de2a0e1e1bb07aa6fd6925db72e662ca288d2ce267897b911be8e6e", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 37417, + "estimated_input_tokens": 9355, + "estimated_eval_tokens": 18966 + }, + "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 across tokenizers, configs, generation, multimodal inputs, loading, and training. I would reject every soft edge; none look like the same underlying fix. If one issue has to serve as a loose representative, 45701 is the broadest and most referenced item.", + "confidence": 0.89, + "canonical_issue_reason": "No clear canonical duplicate target; the set mixes unrelated issues, so there is no single issue that obviously anchors the cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45701 is the broadest and most referenced report in the set, so it is the best loose representative if one must be chosen.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43653", + "right": "issue:43927", + "accept": false, + "reason": "Different bugs: BigBird special-token registration vs DiaConfig losing custom token IDs after save/load." + }, + { + "left": "issue:43688", + "right": "issue:45701", + "accept": false, + "reason": "Aux-loss normalization bug is unrelated to tokenization changes." + }, + { + "left": "issue:44843", + "right": "issue:44991", + "accept": false, + "reason": "Both affect tokenizer loading, but one is offline model_info patching and the other is an est-roberta tokenizer regression." + }, + { + "left": "issue:44479", + "right": "issue:44625", + "accept": false, + "reason": "Video-input regression across Qwen VL variants vs Qwen3.5 num_labels propagation; different code paths." + }, + { + "left": "issue:45491", + "right": "issue:46002", + "accept": false, + "reason": "Gemma3 sliding-window NaNs are unrelated to classification-head NaNs after checkout/init." + }, + { + "left": "issue:43550", + "right": "issue:44361", + "accept": false, + "reason": "Bamba torch.compile/SDPA failure and MLukeTokenizer AttributeError are distinct problems." + }, + { + "left": "issue:44484", + "right": "issue:44704", + "accept": false, + "reason": "max_shard_size behavior is unrelated to AutoProcessor kwargs forwarding." + }, + { + "left": "issue:33666", + "right": "issue:36010", + "accept": false, + "reason": "Multi-GPU Qwen2-VL training and GenerationMixin import failure are unrelated." + }, + { + "left": "issue:43688", + "right": "issue:45440", + "accept": false, + "reason": "Aux-loss normalization and DeepseekV3MoE implementation divergence are different issues." + }, + { + "left": "issue:43742", + "right": "issue:44821", + "accept": false, + "reason": "MobileLLM load key error is unrelated to AutoImageProcessor URL loading." + }, + { + "left": "issue:44734", + "right": "issue:46612", + "accept": false, + "reason": "Both involve generation/cache, but one is KV-cache continuation indexing and the other is beam-search cache reorder skipping for specific models." + }, + { + "left": "issue:32090", + "right": "issue:33666", + "accept": false, + "reason": "Trainer GPU broadcast NoneType error is unrelated to Qwen2-VL multi-GPU training." + }, + { + "left": "issue:44466", + "right": "issue:46050", + "accept": false, + "reason": "lm_head serialization/tied-weights behavior is unrelated to sinq quantized save/load breakage." + }, + { + "left": "issue:41720", + "right": "issue:44464", + "accept": false, + "reason": "Qwen3 auto device-map cuda assert and compiled chunked-generation inconsistency are different generation bugs." + }, + { + "left": "issue:44464", + "right": "issue:44977", + "accept": false, + "reason": "Chunked generation with compiled forward is not the same as Qwen3.5 flash-attention generation failure." + }, + { + "left": "issue:43504", + "right": "issue:43550", + "accept": false, + "reason": "Legacy-field loading for Beit segmentation and Bamba SDPA torch.compile failure are unrelated." + }, + { + "left": "issue:43504", + "right": "issue:43720", + "accept": false, + "reason": "Legacy-field model loading and BitNet packed-weight unpacking are different loading issues." + }, + { + "left": "issue:44977", + "right": "issue:46693", + "accept": false, + "reason": "Generation correctness with flash-attn is not the same as a performance regression with FA2 long-context generation." + }, + { + "left": "issue:43504", + "right": "issue:44610", + "accept": false, + "reason": "Beit pretrained preset loading and OmDet processor/model size mismatch are unrelated." + }, + { + "left": "issue:43577", + "right": "issue:44554", + "accept": false, + "reason": "Blip2 dtype loading drift and MPS attention correctness are separate bugs." + }, + { + "left": "issue:43720", + "right": "issue:43927", + "accept": false, + "reason": "BitNet packed-weight loading and DiaConfig token-id persistence have no shared code path." + }, + { + "left": "issue:33666", + "right": "issue:36331", + "accept": false, + "reason": "Multi-GPU training and CustomTrainer compute_loss signature mismatch are distinct trainer issues." + }, + { + "left": "issue:43643", + "right": "issue:44561", + "accept": false, + "reason": "Both touch trust_remote_code, but one is missing config fields and the other is removal of is_torch_fx_available." + }, + { + "left": "issue:43278", + "right": "issue:43688", + "accept": false, + "reason": "Embedding dtype drift during evaluate is unrelated to auxiliary-loss normalization." + }, + { + "left": "issue:43606", + "right": "issue:43927", + "accept": false, + "reason": "CPU offload device mismatch and custom token-ID loss after save/load are unrelated." + }, + { + "left": "issue:43782", + "right": "issue:46693", + "accept": false, + "reason": "Qwen3VL weight_only loading error is unrelated to FA2 long-context performance regression." + }, + { + "left": "issue:43262", + "right": "issue:43295", + "accept": false, + "reason": "Audio chat-template sample-rate default and processor.tokenizer regression are different processor bugs." + }, + { + "left": "issue:39692", + "right": "issue:43493", + "accept": false, + "reason": "SigLIP2 docs/example errors and HF-vs-JAX implementation divergence are related to the same model family but not the same bug." + }, + { + "left": "issue:44964", + "right": "issue:45405", + "accept": false, + "reason": "Phi-4 multimodal load failure and PEFT version pinning are unrelated." + }, + { + "left": "issue:43720", + "right": "issue:44361", + "accept": false, + "reason": "BitNet weight loading and MLukeTokenizer task errors are unrelated." + }, + { + "left": "issue:43493", + "right": "issue:43825", + "accept": false, + "reason": "SigLIP2 implementation mismatch and pipeline translation error messaging are unrelated." + }, + { + "left": "issue:43525", + "right": "issue:43572", + "accept": false, + "reason": "Missing pad_token_id and missing pad_token_idx are similar-shaped but distinct model/config regressions." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 25, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:28218", + "issue:33357", + "issue:33666", + "issue:39692", + "issue:42796", + "issue:43257", + "issue:43278", + "issue:43295", + "issue:43299", + "issue:43334", + "issue:43452", + "issue:43493", + "issue:43504", + "issue:43572", + "issue:43575", + "issue:43606", + "issue:43646", + "issue:43701", + "issue:43824", + "issue:43844", + "issue:43927", + "issue:43994", + "issue:44361", + "issue:44451", + "issue:44466", + "issue:44488", + "issue:44514", + "issue:44554", + "issue:44568", + "issue:44589", + "issue:44625", + "issue:44792", + "issue:44843", + "issue:44849", + "issue:44977", + "issue:45290", + "issue:45357", + "issue:45405", + "issue:45520", + "issue:45584", + "issue:45812", + "issue:45823", + "issue:46050", + "issue:46192", + "issue:46209", + "issue:46365", + "issue:46489", + "issue:46710" + ], + "soft_pairs": [ + "issue:44554|issue:46365", + "issue:44514|issue:45290", + "issue:44625|issue:45584", + "issue:44977|issue:46209", + "issue:43452|issue:44451", + "issue:45823|issue:46050", + "issue:28218|issue:33357", + "issue:43994|issue:46192", + "issue:42796|issue:44977", + "issue:43278|issue:43844", + "issue:43493|issue:43844", + "issue:39692|issue:43295", + "issue:44849|issue:46209", + "issue:33357|issue:33666", + "issue:43646|issue:44466", + "issue:44568|issue:46489", + "issue:46209|issue:46710", + "issue:43278|issue:43701", + "issue:44792|issue:45405", + "issue:43575|issue:45405", + "issue:43927|issue:44361", + "issue:44849|issue:45357", + "issue:43257|issue:43299", + "issue:44589|issue:45520", + "issue:43824|issue:44589", + "issue:44488|issue:44843", + "issue:44568|issue:45812", + "issue:43504|issue:43606", + "issue:43334|issue:43572" + ], + "prepared_review_unit_hash": "4f1614f9717c6b0e4db502c6d59193883e2b5d3f5638cdf80ae3f2a6069528cf", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36441, + "estimated_input_tokens": 9111, + "estimated_eval_tokens": 18478 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose collection of mostly unrelated Transformers bug reports, with many tokenizer/config/model-loading regressions but no pair that clearly represents the same underlying issue. The closest matches still differ in trigger, affected model, or code path, so I would not merge any soft edge.", + "confidence": 0.88, + "canonical_issue_reason": "issue:45812 is the broadest, most representative report in the set: a silent tokenizer regression affecting an entire model family with clear v4\u2192v5 impact and multiple references.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45812 best fits the cluster because it is concrete, widely scoped, and regression-focused, making it a reasonable anchor for this group of mostly tokenizer/model-compatibility reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44554", + "right": "issue:46365", + "accept": false, + "reason": "Different subsystems and failures: MPS attention correctness vs. Trainer import resolution." + }, + { + "left": "issue:44514", + "right": "issue:45290", + "accept": false, + "reason": "Both involve apply_chat_template, but the crash conditions and code paths differ materially." + }, + { + "left": "issue:44625", + "right": "issue:45584", + "accept": false, + "reason": "Unrelated models and bugs: Qwen3.5 config propagation vs Whisper empty-transcription handling." + }, + { + "left": "issue:44977", + "right": "issue:46209", + "accept": false, + "reason": "Different model/runtime issues: flash-attention behavior vs FP8 inference on B200." + }, + { + "left": "issue:43452", + "right": "issue:44451", + "accept": false, + "reason": "Both are load failures, but for different models and seemingly different causes." + }, + { + "left": "issue:45823", + "right": "issue:46050", + "accept": false, + "reason": "Device-map forward mismatch is not the same as quantized save/load corruption." + }, + { + "left": "issue:28218", + "right": "issue:33357", + "accept": false, + "reason": "Tokenizer spacing bug vs MacOS bus error; no shared underlying defect." + }, + { + "left": "issue:43994", + "right": "issue:46192", + "accept": false, + "reason": "SigLIP2 output quality issue is unrelated to kwargs precedence/return_tensors behavior." + }, + { + "left": "issue:42796", + "right": "issue:44977", + "accept": false, + "reason": "Tokenizer BPE marker handling and Qwen3.5 flash-attention generation failure are unrelated." + }, + { + "left": "issue:43278", + "right": "issue:43844", + "accept": false, + "reason": "Training/eval dtype mismatch differs from ZeRO-3 gradient inflation behavior." + }, + { + "left": "issue:43493", + "right": "issue:43844", + "accept": false, + "reason": "SigLIP2 implementation mismatch is unrelated to DeepSpeed gradient regression." + }, + { + "left": "issue:39692", + "right": "issue:43295", + "accept": false, + "reason": "Doc example mistakes are not the same as a processor/tokenizer regression in custom code." + }, + { + "left": "issue:44849", + "right": "issue:46209", + "accept": false, + "reason": "Different failure modes: hidden_states output bug vs FP8 inference support issue." + }, + { + "left": "issue:33357", + "right": "issue:33666", + "accept": false, + "reason": "MacOS bus error on a CLIP model and Qwen2-VL multi-GPU training are unrelated." + }, + { + "left": "issue:43646", + "right": "issue:44466", + "accept": false, + "reason": "Generic custom-model init breakage is not the same as device-dependent serialization of tied weights." + }, + { + "left": "issue:44568", + "right": "issue:46489", + "accept": false, + "reason": "Different tokenizer regressions affecting different model families and symptoms." + }, + { + "left": "issue:46209", + "right": "issue:46710", + "accept": false, + "reason": "Both mention performance/regression, but one is FP8 inference and the other is model accuracy after upgrade." + }, + { + "left": "issue:43278", + "right": "issue:43701", + "accept": false, + "reason": "Checkpoint resume key mismatch is unrelated to training/eval dtype changes." + }, + { + "left": "issue:44792", + "right": "issue:45405", + "accept": false, + "reason": "A failing Janus image-generation test is unrelated to an unavailable PEFT version bump." + }, + { + "left": "issue:43575", + "right": "issue:45405", + "accept": false, + "reason": "Tensor-parallel OOM and dependency version pinning are entirely different issues." + }, + { + "left": "issue:43927", + "right": "issue:44361", + "accept": false, + "reason": "Different model/config bugs: custom token IDs lost on save/load vs tokenizer AttributeError on tasks." + }, + { + "left": "issue:44849", + "right": "issue:45357", + "accept": false, + "reason": "Hidden-state output bug and incorrect visual encoder keys are different Qwen3.5 regressions." + }, + { + "left": "issue:43257", + "right": "issue:43299", + "accept": false, + "reason": "Both involve Qwen MoE loading, but one is accelerate/deepspeed conversion and the other is a dev-version load failure; not the same bug." + }, + { + "left": "issue:44589", + "right": "issue:45520", + "accept": false, + "reason": "Float8 storage lookup and flash_attn import_utils KeyError are unrelated runtime/environment issues." + }, + { + "left": "issue:43824", + "right": "issue:44589", + "accept": false, + "reason": "Import error for a VL class and float8 storage resolution failure do not share a code path." + }, + { + "left": "issue:44488", + "right": "issue:44843", + "accept": false, + "reason": "Model loading failure for a specific BERT model is not the same as offline regex patch behavior in AutoTokenizer." + }, + { + "left": "issue:44568", + "right": "issue:45812", + "accept": false, + "reason": "Different tokenizer regressions: missing BOS/EOS insertion vs wrong IDs for Granite models." + }, + { + "left": "issue:43504", + "right": "issue:43606", + "accept": false, + "reason": "Legacy-field model loading and CPU-offload device mismatch are unrelated." + }, + { + "left": "issue:43334", + "right": "issue:43572", + "accept": false, + "reason": "Similar symptom of missing config attributes, but different fields, models, and failure contexts." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 26, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:15354", + "issue:28218", + "issue:29159", + "issue:29942", + "issue:30333", + "issue:32090", + "issue:34567", + "issue:38175", + "issue:42490", + "issue:43065", + "issue:43334", + "issue:43504", + "issue:43572", + "issue:43606", + "issue:43653", + "issue:43696", + "issue:43844", + "issue:43873", + "issue:43937", + "issue:43994", + "issue:44112", + "issue:44291", + "issue:44423", + "issue:44464", + "issue:44493", + "issue:44554", + "issue:44734", + "issue:44869", + "issue:44964", + "issue:44977", + "issue:45084", + "issue:45200", + "issue:45229", + "issue:45245", + "issue:45446", + "issue:45520", + "issue:45584", + "issue:45588", + "issue:45759", + "issue:45823", + "issue:46050", + "issue:46285", + "issue:46369", + "issue:46424", + "issue:46701", + "issue:46710", + "issue:46752" + ], + "soft_pairs": [ + "issue:43334|issue:45759", + "issue:43606|issue:44112", + "issue:43572|issue:44291", + "issue:45245|issue:46710", + "issue:44493|issue:46285", + "issue:43504|issue:43653", + "issue:44464|issue:46424", + "issue:43937|issue:46701", + "issue:44464|issue:45584", + "issue:45200|issue:45229", + "issue:45229|issue:45823", + "issue:44869|issue:46752", + "issue:43844|issue:44964", + "issue:43873|issue:46050", + "issue:43994|issue:44493", + "issue:29942|issue:30333", + "issue:45588|issue:46050", + "issue:29159|issue:32090", + "issue:42490|issue:43696", + "issue:45084|issue:45520", + "issue:44977|issue:46701", + "issue:43065|issue:46369", + "issue:28218|issue:34567", + "issue:44554|issue:45446", + "issue:15354|issue:38175", + "issue:44423|issue:44734" + ], + "prepared_review_unit_hash": "a1e0111eb58313cf27baadfc64462ac8b01c68736acca28e1e424e88c1336f5f", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 34993, + "estimated_input_tokens": 8749, + "estimated_eval_tokens": 17754 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is a grab-bag of unrelated transformer bugs across tokenizers, model loading, generation, quantization, serving, and CI. None of the soft pairs appear to be the same underlying issue, so I reject all of them.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43334", + "right": "issue:45759", + "accept": false, + "reason": "Both involve Qwen multimodal config/loading, but one is a missing `pad_token_id` attribute and the other is `from_config` not unwrapping `text_config`; different bugs and fixes." + }, + { + "left": "issue:43606", + "right": "issue:44112", + "accept": false, + "reason": "Different models and failure modes: Bark CPU offload device mismatch vs a stale GraniteSpeech CI test." + }, + { + "left": "issue:43572", + "right": "issue:44291", + "accept": false, + "reason": "Both are load-time regressions after 5.0, but one is a missing config field and the other is an unexpected `_is_hf_initialized` argument; not the same code path." + }, + { + "left": "issue:45245", + "right": "issue:46710", + "accept": false, + "reason": "Category-count runtime error versus an accuracy regression in a specific model; unrelated symptoms and causes." + }, + { + "left": "issue:44493", + "right": "issue:46285", + "accept": false, + "reason": "Both mention structural/model-output changes, but one is an unexpected key warning and the other is a CLIP text model API inconsistency." + }, + { + "left": "issue:43504", + "right": "issue:43653", + "accept": false, + "reason": "Different components and bugs: BEiT legacy-field loading failure vs BigBirdTokenizer special-token registration/empty decode issue." + }, + { + "left": "issue:44464", + "right": "issue:46424", + "accept": false, + "reason": "Both touch generation/cache behavior, but one is compiled-forward inconsistency and the other is `max_cache_length` handling for quantized cache; not the same defect." + }, + { + "left": "issue:43937", + "right": "issue:46701", + "accept": false, + "reason": "Different models and failure modes: invalid `GenerationConfig` vs a pipeline failure in DiffusionGemmaGenerationOutput." + }, + { + "left": "issue:44464", + "right": "issue:45584", + "accept": false, + "reason": "Chunked generation with compiled forward is unrelated to Whisper failing on empty transcription after `align_special_tokens`." + }, + { + "left": "issue:45200", + "right": "issue:45229", + "accept": false, + "reason": "One is a missing default for `mm_token_type_ids`; the other is a multi-GPU CUDA OOM. Different problems entirely." + }, + { + "left": "issue:45229", + "right": "issue:45823", + "accept": false, + "reason": "Both are Gemma4 inference issues, but OOM from multi-GPU versus a device mismatch in forward are distinct bugs." + }, + { + "left": "issue:44869", + "right": "issue:46752", + "accept": false, + "reason": "Both are edge-case crashes, but one is Whisper timestamp decoding on a trailing replacement character and the other is `apply_chat_template` on an empty list." + }, + { + "left": "issue:43844", + "right": "issue:44964", + "accept": false, + "reason": "Different model/training stack issue versus a model-loading failure; no shared underlying bug." + }, + { + "left": "issue:43873", + "right": "issue:46050", + "accept": false, + "reason": "Both involve quantization/offloading, but one is offloading not working as expected and the other is save/load corruption after `sinq`; not mergeable as one fix." + }, + { + "left": "issue:43994", + "right": "issue:44493", + "accept": false, + "reason": "SIGLIP2 bad outputs and unexpected position-id key warnings are different regressions in different paths." + }, + { + "left": "issue:29942", + "right": "issue:30333", + "accept": false, + "reason": "Flash Attention 2 test failures are unrelated to MLFlow not reporting failed training jobs." + }, + { + "left": "issue:45588", + "right": "issue:46050", + "accept": false, + "reason": "A flash-attention integration crash with `s_aux=None` is unrelated to quantized save/load breakage." + }, + { + "left": "issue:29159", + "right": "issue:32090", + "accept": false, + "reason": "Tokenizer slow/fast inconsistency and a trainer `_gpu_broadcast_one` TypeError are unrelated." + }, + { + "left": "issue:42490", + "right": "issue:43696", + "accept": false, + "reason": "Import-path warning from trailing slash and a CUDA OOM are not the same issue." + }, + { + "left": "issue:45084", + "right": "issue:45520", + "accept": false, + "reason": "Template-node compilation error and a Python 3.13 `flash_attn` import KeyError are unrelated." + }, + { + "left": "issue:44977", + "right": "issue:46701", + "accept": false, + "reason": "Qwen3.5 flash-attention generation problems do not match a DiffusionGemma pipeline failure." + }, + { + "left": "issue:43065", + "right": "issue:46369", + "accept": false, + "reason": "Same broad SAM3 area, but dummy `nn.Conv2d` in one component is not the same as memory overflow in `sam3-litetext`." + }, + { + "left": "issue:28218", + "right": "issue:34567", + "accept": false, + "reason": "Tokenizer spacing behavior and `TrainerState.num_input_tokens_seen` not updating are different features with different symptoms." + }, + { + "left": "issue:44554", + "right": "issue:45446", + "accept": false, + "reason": "MPS attention correctness with mismatched head dims is unrelated to an AuxRequest PyTorch version check in flex_attention." + }, + { + "left": "issue:15354", + "right": "issue:38175", + "accept": false, + "reason": "TorchScript GeneratorExp export failure and SIGLIP2 zero-probability outputs are unrelated bugs." + }, + { + "left": "issue:44423", + "right": "issue:44734", + "accept": false, + "reason": "Both are `transformers serve` crashes, but one is multimodal continuous batching and the other is KV-cache continuation tensor indexing; different code paths 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": 1, + "component_count": 23, + "review_unit_index": 27, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:29870", + "issue:29942", + "issue:37428", + "issue:38175", + "issue:39692", + "issue:41720", + "issue:42490", + "issue:43232", + "issue:43278", + "issue:43295", + "issue:43404", + "issue:43450", + "issue:43493", + "issue:43540", + "issue:43550", + "issue:43572", + "issue:43575", + "issue:43653", + "issue:43701", + "issue:43824", + "issue:43844", + "issue:43856", + "issue:43873", + "issue:43957", + "issue:44112", + "issue:44186", + "issue:44466", + "issue:44554", + "issue:44610", + "issue:44849", + "issue:44987", + "issue:45072", + "issue:45127", + "issue:45198", + "issue:45200", + "issue:45356", + "issue:45381", + "issue:45406", + "issue:45520", + "issue:45588", + "issue:45759", + "issue:45823", + "issue:46121", + "issue:46178", + "issue:46192", + "issue:46209", + "issue:46710" + ], + "soft_pairs": [ + "issue:43493|issue:43701", + "issue:41720|issue:46710", + "issue:45406|issue:46121", + "issue:43404|issue:46710", + "issue:45072|issue:45198", + "issue:37428|issue:43957", + "issue:45588|issue:45759", + "issue:43295|issue:43450", + "issue:43540|issue:45200", + "issue:45200|issue:45823", + "issue:44186|issue:46192", + "issue:43295|issue:43873", + "issue:45823|issue:46192", + "issue:43295|issue:44610", + "issue:29870|issue:29942", + "issue:45520|issue:46178", + "issue:43572|issue:44987", + "issue:43450|issue:43575", + "issue:43550|issue:43844", + "issue:43232|issue:44849", + "issue:43572|issue:43957", + "issue:44554|issue:45356", + "issue:37428|issue:43824", + "issue:42490|issue:43493", + "issue:44466|issue:45127", + "issue:38175|issue:43493", + "issue:39692|issue:46192", + "issue:43653|issue:44112", + "issue:43278|issue:43856", + "issue:45381|issue:46209" + ], + "prepared_review_unit_hash": "b5cdc853d90e2bac5ee276d272edd0fcd4a90f2d90d05a2200ef25be3e4f6b11", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 36397, + "estimated_input_tokens": 9100, + "estimated_eval_tokens": 18456 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous: the soft-similarity links span unrelated bugs across generation, tokenizers, vision/video processors, quantization, and config/loading issues. I would keep 43295 as the best anchor only because it is the most connected and representative-looking issue in the set, but there is no true coherent duplicate cluster here.", + "confidence": 0.64, + "canonical_issue_reason": "No strong true canonical emerges from the titles alone; if one anchor is needed, issue 43295 is the best hub because it has the most soft neighbors and a clear, concrete regression report.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43295 is the best representative by connectivity and specificity: it is a concrete regression in processor/tokenizer behavior and is the most central item among the soft-similarity candidates.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43493", + "right": "issue:43701", + "accept": false, + "reason": "Different problems: SigLIP2 model-output discrepancy vs a resume_from_checkpoint key mismatch." + }, + { + "left": "issue:41720", + "right": "issue:46710", + "accept": false, + "reason": "Different failures: device-map CUDA assert on A800 vs accuracy regression after version upgrade." + }, + { + "left": "issue:45406", + "right": "issue:46121", + "accept": false, + "reason": "Gemma4 processor attribute regression and rope-validation TypeError are unrelated code paths." + }, + { + "left": "issue:43404", + "right": "issue:46710", + "accept": false, + "reason": "Different model families and symptoms; tied lm_head bug vs DeepSeek accuracy regression." + }, + { + "left": "issue:45072", + "right": "issue:45198", + "accept": false, + "reason": "Different areas: dtype mismatches in inference vs Wav2Vec2 save/tokenization failure." + }, + { + "left": "issue:37428", + "right": "issue:43957", + "accept": false, + "reason": "One is a flash-attention import symbol issue; the other is a meta-device model-loading regression." + }, + { + "left": "issue:45588", + "right": "issue:45759", + "accept": false, + "reason": "Flash-attention sink handling crash and composite-config text_config unwrapping are unrelated." + }, + { + "left": "issue:43295", + "right": "issue:43450", + "accept": false, + "reason": "Both involve processors, but one is tokenizer access/images handling and the other is batched video shape; distinct bugs." + }, + { + "left": "issue:43540", + "right": "issue:45200", + "accept": false, + "reason": "Video-input processing error in Qwen3OmniMoe vs Gemma4 mm_token_type_ids defaulting; not the same issue." + }, + { + "left": "issue:45200", + "right": "issue:45823", + "accept": false, + "reason": "Both are Gemma4-related, but one is missing mm_token_type_ids and the other is a device_map forward mismatch." + }, + { + "left": "issue:44186", + "right": "issue:46192", + "accept": false, + "reason": "Tokenizer crash on NER/batched padding is different from silently ignored return_tensors handling." + }, + { + "left": "issue:43295", + "right": "issue:43873", + "accept": false, + "reason": "Processor/tokenizer regression and offloading-with-quantization are separate concerns." + }, + { + "left": "issue:45823", + "right": "issue:46192", + "accept": false, + "reason": "Gemma4 device mismatch and return_tensors precedence are unrelated bugs." + }, + { + "left": "issue:43295", + "right": "issue:44610", + "accept": false, + "reason": "Custom processor/tokenizer regression vs OmDet-Turbo image-size mismatch are different failures." + }, + { + "left": "issue:29870", + "right": "issue:29942", + "accept": false, + "reason": "Bart generate() UnboundLocalError and failing Flash Attention 2 tests do not describe the same bug." + }, + { + "left": "issue:45520", + "right": "issue:46178", + "accept": false, + "reason": "Flash-attn import KeyError and tensorflow_text backend-mapping error are different import-system issues." + }, + { + "left": "issue:43572", + "right": "issue:44987", + "accept": false, + "reason": "StableLmConfig pad_token_idx regression and loading physical-intelligence/fast are unrelated." + }, + { + "left": "issue:43450", + "right": "issue:43575", + "accept": false, + "reason": "Batched video processor shape bug and tensor-parallel OOM are not the same underlying problem." + }, + { + "left": "issue:43550", + "right": "issue:43844", + "accept": false, + "reason": "torch.compile+SDPA failure in Bamba is distinct from ZeRO-3 gradient growth during training." + }, + { + "left": "issue:43232", + "right": "issue:44849", + "accept": false, + "reason": "Generation kwargs after sync_gpus and output_hidden_states regression are separate code paths." + }, + { + "left": "issue:43572", + "right": "issue:43957", + "accept": false, + "reason": "Config pad token issue is unrelated to meta-device model loading breakage." + }, + { + "left": "issue:44554", + "right": "issue:45356", + "accept": false, + "reason": "MPS attention correctness bug and Kimi tokenizer codec regression are unrelated." + }, + { + "left": "issue:37428", + "right": "issue:43824", + "accept": false, + "reason": "Different import errors for different symbols/models; no shared concrete bug." + }, + { + "left": "issue:42490", + "right": "issue:43493", + "accept": false, + "reason": "Trailing-slash local model path warning and SigLIP2 implementation discrepancy are unrelated." + }, + { + "left": "issue:44466", + "right": "issue:45127", + "accept": false, + "reason": "Tied-weight serialization inconsistency and LoRA merge collapse are different weight-handling bugs." + }, + { + "left": "issue:38175", + "right": "issue:43493", + "accept": false, + "reason": "Close in model family, but zero probabilities on a checkpoint vs HF-vs-JAX discrepancy are not clearly the same concrete bug." + }, + { + "left": "issue:39692", + "right": "issue:46192", + "accept": false, + "reason": "SigLIP2 doc-example/model-processor issues and return_tensors precedence are unrelated." + }, + { + "left": "issue:43653", + "right": "issue:44112", + "accept": false, + "reason": "BigBirdTokenizer special-token registration and GraniteSpeech CI stale device override are different issues." + }, + { + "left": "issue:43278", + "right": "issue:43856", + "accept": false, + "reason": "Embedding dtype drift in eval and Qwen3 MoE memory inefficiency are unrelated." + }, + { + "left": "issue:45381", + "right": "issue:46209", + "accept": false, + "reason": "Qwen2.5-VL video position IDs and FP8 inference on B200 are different hardware/model 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": 23, + "review_unit_index": 28, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:28218", + "issue:35141", + "issue:38175", + "issue:41720", + "issue:42915", + "issue:43262", + "issue:43334", + "issue:43335", + "issue:43388", + "issue:43450", + "issue:43452", + "issue:43525", + "issue:43526", + "issue:43550", + "issue:43606", + "issue:43819", + "issue:43824", + "issue:43827", + "issue:43901", + "issue:43950", + "issue:43975", + "issue:44112", + "issue:44206", + "issue:44442", + "issue:44479", + "issue:44521", + "issue:44561", + "issue:44610", + "issue:44792", + "issue:44857", + "issue:44898", + "issue:45083", + "issue:45198", + "issue:45200", + "issue:45290", + "issue:45325", + "issue:45412", + "issue:45520", + "issue:45563", + "issue:45584", + "issue:45593", + "issue:45759", + "issue:46170", + "issue:46192", + "issue:46209", + "issue:46234", + "issue:46285", + "issue:46512" + ], + "soft_pairs": [ + "issue:44521|issue:45290", + "issue:28218|issue:35141", + "issue:42915|issue:43824", + "issue:43950|issue:46285", + "issue:43452|issue:46512", + "issue:43824|issue:43975", + "issue:43334|issue:43525", + "issue:43262|issue:43450", + "issue:44442|issue:45198", + "issue:44792|issue:45563", + "issue:44479|issue:46192", + "issue:44898|issue:45412", + "issue:43819|issue:44112", + "issue:43827|issue:43901", + "issue:45200|issue:45325", + "issue:44610|issue:44857", + "issue:43388|issue:43526", + "issue:44561|issue:46234", + "issue:43335|issue:43550", + "issue:43819|issue:45083", + "issue:44857|issue:45593", + "issue:41720|issue:46209", + "issue:44206|issue:45584", + "issue:45759|issue:46285", + "issue:45520|issue:46170", + "issue:43606|issue:45083", + "issue:38175|issue:45412" + ], + "prepared_review_unit_hash": "404dc1decb55664b032d4fb88f00caf98dd3cffff40e9e0a67879f6cc8ed9454", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 36148, + "estimated_input_tokens": 9037, + "estimated_eval_tokens": 18330 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: the soft pairs span unrelated tokenizer, multimodal processor, config, loading, docs, and hardware-specific bugs. I do not see any pair that clearly describes the same underlying defect, so all soft edges are rejected. No PRs are present.", + "confidence": 0.94, + "canonical_issue_reason": "issue:44521 is the most discussion-heavy, actively open issue in the set and is a reasonable representative node for the cluster, but it is not a duplicate of the other issues.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44521 is the best overall representative issue by activity and recency; it has the strongest signal among otherwise unrelated issues, but it should not be treated as semantically merged with the others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44521", + "right": "issue:45290", + "accept": false, + "reason": "Both concern apply_chat_template, but one is about multimodal assistant masks and the other is a tokenize=True crash with tool-call assistant messages; different failure modes and code paths." + }, + { + "left": "issue:28218", + "right": "issue:35141", + "accept": false, + "reason": "Tokenizer spacing after added tokens is unrelated to embedding reinitialization during post_init when tie_word_embedding=False." + }, + { + "left": "issue:42915", + "right": "issue:43824", + "accept": false, + "reason": "Different models and failures: Qwen3MoE FP8 inference vs Qwen2.5-VL import error." + }, + { + "left": "issue:43950", + "right": "issue:46285", + "accept": false, + "reason": "Non-persistent buffer corruption in from_pretrained is unrelated to CLIP text model structural inconsistency." + }, + { + "left": "issue:43452", + "right": "issue:46512", + "accept": false, + "reason": "gguf_file loading breakage and stale config.dtype after from_config(dtype=...) are separate loading/configuration bugs." + }, + { + "left": "issue:43824", + "right": "issue:43975", + "accept": false, + "reason": "Import failure for Qwen2.5-VL is unrelated to incorrect detokenization for deepseek-coder." + }, + { + "left": "issue:43334", + "right": "issue:43525", + "accept": false, + "reason": "Both mention pad_token_id, but they are different model families with likely model-specific config issues, not the same concrete bug." + }, + { + "left": "issue:43262", + "right": "issue:43450", + "accept": false, + "reason": "Audio processor sampling-rate defaulting and video batch shape handling are different processor bugs." + }, + { + "left": "issue:44442", + "right": "issue:45198", + "accept": false, + "reason": "FastSpeech2ConformerTokenizer loading failure is unrelated to Wav2Vec2 save_pretrained/tokenization failure." + }, + { + "left": "issue:44792", + "right": "issue:45563", + "accept": false, + "reason": "Janus image-generation test failure and a stale paged-generate warning are different generation-path issues." + }, + { + "left": "issue:44479", + "right": "issue:46192", + "accept": false, + "reason": "Qwen multimodal video-input regression is unrelated to return_tensors being ignored when text_kwargs is passed." + }, + { + "left": "issue:44898", + "right": "issue:45412", + "accept": false, + "reason": "Perceiver non-default-resolution failure and RT-DETR memory release are unrelated." + }, + { + "left": "issue:43819", + "right": "issue:44112", + "accept": false, + "reason": "DAC.from_latents mismatch and a GraniteSpeech CI test failure are different audio/model behaviors." + }, + { + "left": "issue:43827", + "right": "issue:43901", + "accept": false, + "reason": "Both are docs issues, but they concern different pipeline APIs and different deprecations." + }, + { + "left": "issue:45200", + "right": "issue:45325", + "accept": false, + "reason": "Gemma 4 mm_token_type_ids handling is unrelated to Qwen2.5-VL rope index scaling." + }, + { + "left": "issue:44610", + "right": "issue:44857", + "accept": false, + "reason": "OmDet-Turbo input-size mismatch and LwDetrImageLoss AMP crash are different detection bugs." + }, + { + "left": "issue:43388", + "right": "issue:43526", + "accept": false, + "reason": "Both involve labels, but one is gather_for_metrics truncation and the other is BeitImageProcessorFast reduce_labels returning the wrong shape." + }, + { + "left": "issue:44561", + "right": "issue:46234", + "accept": false, + "reason": "Different missing utility functions with different downstream breakages; not the same code-path problem." + }, + { + "left": "issue:43335", + "right": "issue:43550", + "accept": false, + "reason": "SwitchTransformers sparse-layer construction and Bamba torch.compile/SDPA failure are unrelated." + }, + { + "left": "issue:43819", + "right": "issue:45083", + "accept": false, + "reason": "DAC latent decoding mismatch is unrelated to qwen3_omni_moe feature-length helper behavior." + }, + { + "left": "issue:44857", + "right": "issue:45593", + "accept": false, + "reason": "AMP crash in LwDetrImageLoss is unrelated to D-FINE auxiliary-loss handling." + }, + { + "left": "issue:41720", + "right": "issue:46209", + "accept": false, + "reason": "Different model/hardware inference failures: Qwen3 auto device mapping on A800 vs FP8 inference on B200." + }, + { + "left": "issue:44206", + "right": "issue:45584", + "accept": false, + "reason": "Unsupported center arg in LasrFeatureExtractor is unrelated to Whisper empty-transcription failure after align_special_tokens." + }, + { + "left": "issue:45759", + "right": "issue:46285", + "accept": false, + "reason": "Qwen multimodal from_config text_config unwrapping and CLIP text-model structure divergence are different model-initialization issues." + }, + { + "left": "issue:45520", + "right": "issue:46170", + "accept": false, + "reason": "Python 3.13 flash_attn import_utils KeyError and Windows ROCm torch.distributed import/type errors are different platform-specific issues." + }, + { + "left": "issue:43606", + "right": "issue:45083", + "accept": false, + "reason": "CPU offload device mismatch in bark-small is unrelated to qwen3_omni_moe feature-length helper behavior." + }, + { + "left": "issue:38175", + "right": "issue:45412", + "accept": false, + "reason": "Siglip2 zero probabilities and RT-DETR memory release are unrelated bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 29, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:28282", + "issue:29127", + "issue:29942", + "issue:32090", + "issue:33357", + "issue:34689", + "issue:35141", + "issue:37428", + "issue:42490", + "issue:43278", + "issue:43335", + "issue:43425", + "issue:43504", + "issue:43550", + "issue:43606", + "issue:43611", + "issue:43653", + "issue:43720", + "issue:43819", + "issue:43825", + "issue:43854", + "issue:43931", + "issue:44112", + "issue:44351", + "issue:44448", + "issue:44493", + "issue:44514", + "issue:44521", + "issue:44743", + "issue:44779", + "issue:44855", + "issue:44857", + "issue:45083", + "issue:45127", + "issue:45200", + "issue:45245", + "issue:45290", + "issue:45310", + "issue:45478", + "issue:45588", + "issue:45593", + "issue:45759", + "issue:45859", + "issue:46326", + "issue:46431", + "issue:46512", + "issue:46566" + ], + "soft_pairs": [ + "issue:37428|issue:43425", + "issue:43335|issue:43653", + "issue:43550|issue:44112", + "issue:29942|issue:32090", + "issue:28282|issue:33357", + "issue:43611|issue:43931", + "issue:44448|issue:44743", + "issue:43335|issue:43504", + "issue:45759|issue:46512", + "issue:44855|issue:45083", + "issue:43278|issue:43825", + "issue:43720|issue:44112", + "issue:43335|issue:45083", + "issue:29127|issue:32090", + "issue:43335|issue:44112", + "issue:43335|issue:45593", + "issue:37428|issue:44351", + "issue:44514|issue:44521", + "issue:45588|issue:45859", + "issue:34689|issue:35141", + "issue:46431|issue:46566", + "issue:45290|issue:46326", + "issue:45127|issue:45245", + "issue:42490|issue:43854", + "issue:45200|issue:45588", + "issue:45310|issue:45478", + "issue:43606|issue:44857", + "issue:43819|issue:44493", + "issue:44448|issue:44779" + ], + "prepared_review_unit_hash": "5995cea2ef3885348e665c17a1bdfc0899317e8965b653c6fa633664fed6056c", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 36062, + "estimated_input_tokens": 9016, + "estimated_eval_tokens": 18288 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly heterogeneous; the only clear duplicate pair is the Qwen3.5 MoE from_pretrained regression (45310/45478). The other soft links are mostly same-subsystems but distinct bugs and should stay separate.", + "confidence": 0.88, + "canonical_issue_reason": "issue:45310 is the earlier and clearer representative of the only exact duplicate pair here; issue:45478 has the same title and describes the same Qwen3.5 MoE from_pretrained failure on transformers>=5.4.0.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45310 is the best cluster representative because it is the strongest confirmed duplicate and the rest of the set is too diverse to unify cleanly under one issue.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:37428", + "right": "issue:43425", + "accept": false, + "reason": "Different failures: missing internal flash-attn helper import vs a Torch 2.10 compatibility report." + }, + { + "left": "issue:43335", + "right": "issue:43653", + "accept": false, + "reason": "Different models and bugs: SwitchTransformers sparse-layer config vs BigBirdTokenizer special-token decoding." + }, + { + "left": "issue:43550", + "right": "issue:44112", + "accept": false, + "reason": "Bamba SDPA/torch.compile failure and GraniteSpeech CI stale-device test are unrelated code paths." + }, + { + "left": "issue:29942", + "right": "issue:32090", + "accept": false, + "reason": "Flash Attention 2 test failures and Trainer _gpu_broadcast_one NoneType errors are distinct issues." + }, + { + "left": "issue:28282", + "right": "issue:33357", + "accept": false, + "reason": "PyTorch-missing ImportError vs Mac bus error loading CLIP are different root causes." + }, + { + "left": "issue:43611", + "right": "issue:43931", + "accept": false, + "reason": "base_model_prefix loading regression and Qwen3-VL weight-shape mismatch are not the same bug." + }, + { + "left": "issue:44448", + "right": "issue:44743", + "accept": false, + "reason": "Pegasus v4/v5 output drift and Qwen3_5 recurrent-state reset are different model behaviors." + }, + { + "left": "issue:43335", + "right": "issue:43504", + "accept": false, + "reason": "SwitchTransformers sparse-layer creation and Beit legacy-field loading are unrelated." + }, + { + "left": "issue:45759", + "right": "issue:46512", + "accept": false, + "reason": "Both involve from_config, but one is text_config unwrapping for multimodal configs and the other is stale config.dtype after construction." + }, + { + "left": "issue:44855", + "right": "issue:45083", + "accept": false, + "reason": "Python 3.13 TorchScript indentation parsing error and qwen3_omni output-length helper behavior are unrelated." + }, + { + "left": "issue:43278", + "right": "issue:43825", + "accept": false, + "reason": "BF16-to-FP32 dtype regression during evaluate and an incorrect pipeline error message are different problems." + }, + { + "left": "issue:43720", + "right": "issue:44112", + "accept": false, + "reason": "BitNet packed-weight unloading bug and GraniteSpeech stale-device CI failure do not share the same code path." + }, + { + "left": "issue:43335", + "right": "issue:45083", + "accept": false, + "reason": "SwitchTransformers sparse-layer config bug is unrelated to qwen3_omni feature-length computation." + }, + { + "left": "issue:29127", + "right": "issue:32090", + "accept": false, + "reason": "LayoutLMV3 error-message clarity and Trainer broadcast NoneType failure are different issues." + }, + { + "left": "issue:43335", + "right": "issue:44112", + "accept": false, + "reason": "SwitchTransformers config bug and GraniteSpeech CI stale-device test are unrelated." + }, + { + "left": "issue:43335", + "right": "issue:45593", + "accept": false, + "reason": "SwitchTransformers sparse-layer creation and D-FINE auxiliary-loss handling are different model bugs." + }, + { + "left": "issue:37428", + "right": "issue:44351", + "accept": false, + "reason": "Both are import errors, but they concern different missing symbols and likely different fixes." + }, + { + "left": "issue:44514", + "right": "issue:44521", + "accept": false, + "reason": "Both are apply_chat_template issues, but one is a batched padding=False crash and the other is incorrect assistant_masks for multimodal inputs." + }, + { + "left": "issue:45588", + "right": "issue:45859", + "accept": false, + "reason": "flash_attention.py s_aux=None crash and Qwen3_5 rotary embedding CPU-offload incompatibility are different failures." + }, + { + "left": "issue:34689", + "right": "issue:35141", + "accept": false, + "reason": "Model loading breakage for Llama 3.2 Vision and token-embedding reinitialization after resize are separate bugs." + }, + { + "left": "issue:46431", + "right": "issue:46566", + "accept": false, + "reason": "Missing chat template error and DiffusionGemma meta-device loading issue are unrelated." + }, + { + "left": "issue:45290", + "right": "issue:46326", + "accept": false, + "reason": "Both involve chat templating, but tool-call/no-content crashes and mistral_common continue_final_message behavior are not the same bug." + }, + { + "left": "issue:45127", + "right": "issue:45245", + "accept": false, + "reason": "LoRA merge collapse with extended vocab and category-count overflow are unrelated." + }, + { + "left": "issue:42490", + "right": "issue:43854", + "accept": false, + "reason": "Trailing-slash local-path warning and GLM-4.7-Flash loading failure are distinct issues." + }, + { + "left": "issue:45200", + "right": "issue:45588", + "accept": false, + "reason": "mm_token_type_ids defaulting and flash_attention s_aux handling are unrelated." + }, + { + "left": "issue:45310", + "right": "issue:45478", + "accept": true, + "reason": "Same title and same Qwen3.5 MoE from_pretrained regression on transformers>=5.4.0; clear duplicate." + }, + { + "left": "issue:43606", + "right": "issue:44857", + "accept": false, + "reason": "CPU offload device mismatch for bark-small and float16 AMP CUDA crash in LwDetrImageLoss are different bugs." + }, + { + "left": "issue:43819", + "right": "issue:44493", + "accept": false, + "reason": "DAC from_latents/STE mismatch and unexpected position-id keys affect different code paths." + }, + { + "left": "issue:44448", + "right": "issue:44779", + "accept": false, + "reason": "Pegasus v4/v5 output differences and Deepseek tokenizer v5 regression are separate tokenizer/model issues." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded overall. The only clearly duplicated pair in the packet is 45310/45478, which have identical titles and matching Qwen3.5 MoE from_pretrained regression descriptions. The remaining soft links are handled conservatively as distinct bugs, with reasons tied to different models/code paths rather than overstating duplication." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 30, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:10000", + "issue:15354", + "issue:26413", + "issue:27957", + "issue:28218", + "issue:29870", + "issue:29942", + "issue:31501", + "issue:32090", + "issue:33290", + "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:43701", + "issue:43927", + "issue:44206", + "issue:44360", + "issue:44448", + "issue:44493", + "issue:44561", + "issue:44617", + "issue:44855", + "issue:44857", + "issue:44977", + "issue:45072", + "issue:45083", + "issue:45229", + "issue:45290", + "issue:45381", + "issue:45459", + "issue:45588", + "issue:45698", + "issue:45823", + "issue:45920", + "issue:46192", + "issue:46234", + "issue:46399", + "issue:46697" + ], + "soft_pairs": [ + "issue:39692|issue:43526", + "issue:26413|issue:31501", + "issue:45459|issue:46234", + "issue:28218|issue:32090", + "issue:39692|issue:44493", + "issue:44617|issue:46697", + "issue:44206|issue:45823", + "issue:43701|issue:45083", + "issue:44857|issue:45072", + "issue:32090|issue:33357", + "issue:42490|issue:43575", + "issue:44360|issue:44561", + "issue:43550|issue:45083", + "issue:38175|issue:44493", + "issue:29942|issue:35141", + "issue:27957|issue:34689", + "issue:43278|issue:43329", + "issue:29870|issue:32090", + "issue:44448|issue:44977", + "issue:45920|issue:46192", + "issue:40444|issue:43526", + "issue:43278|issue:43381", + "issue:10000|issue:15354", + "issue:44977|issue:45229", + "issue:45290|issue:45381", + "issue:44855|issue:46399", + "issue:29870|issue:33290", + "issue:44857|issue:45588", + "issue:43927|issue:45698" + ], + "prepared_review_unit_hash": "b597d41b7b2b7dee9e8624538671f5712989373fed191624e88cd9749ea9cb75", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36477, + "estimated_input_tokens": 9120, + "estimated_eval_tokens": 18496 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab-bag of unrelated Transformers issues across tokenization, model loading, training, Flash Attention, JIT/export, dtype, and OOM behavior. None of the soft pairs look like the same underlying bug or change, so I reject all edges.", + "confidence": 0.97, + "canonical_issue_reason": "No clear canonical issue: the items span many unrelated subsystems and failure modes rather than one duplicate bug.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45920 is the best representative only because it is the broadest, highest-signal regression report in the set; it still is not a duplicate of the others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:39692", + "right": "issue:43526", + "accept": false, + "reason": "SigLIP2 docs/model-processor example errors and BeitImageProcessorFast reduce_labels are different bugs in different components." + }, + { + "left": "issue:26413", + "right": "issue:31501", + "accept": false, + "reason": "Both involve training startup, but one is resume_from_checkpoint epoch handling while the other is a FineWeb dataset/training launch failure." + }, + { + "left": "issue:45459", + "right": "issue:46234", + "accept": false, + "reason": "One is a tokenizer exception-masking problem, the other is a missing import_utils feature flag; unrelated fixes." + }, + { + "left": "issue:28218", + "right": "issue:32090", + "accept": false, + "reason": "Tokenizer added-space behavior and Trainer _gpu_broadcast_one NoneType are separate code paths and symptoms." + }, + { + "left": "issue:39692", + "right": "issue:44493", + "accept": false, + "reason": "Doc/example mismatch for SigLIP2 is not the same as the position_id unexpected-key regression." + }, + { + "left": "issue:44617", + "right": "issue:46697", + "accept": false, + "reason": "CUDA OOM in Sam3Video and Swin2SR over-padding are distinct model/image-processing issues." + }, + { + "left": "issue:44206", + "right": "issue:45823", + "accept": false, + "reason": "Unsupported center arg in LasrFeatureExtractor and Gemma4 device_map mismatch are different regressions." + }, + { + "left": "issue:43701", + "right": "issue:45083", + "accept": false, + "reason": "resume_from_checkpoint key mismatch is unrelated to qwen3_omni_moe feature-extraction length logic." + }, + { + "left": "issue:44857", + "right": "issue:45072", + "accept": false, + "reason": "Both mention precision, but an AMP crash in LwDetrImageLoss is not the same bug as bfloat16 dtype mismatches in inference." + }, + { + "left": "issue:32090", + "right": "issue:33357", + "accept": false, + "reason": "Trainer NoneType broadcast failure and MacOS bus error on CLIP are different failure modes and subsystems." + }, + { + "left": "issue:42490", + "right": "issue:43575", + "accept": false, + "reason": "Trailing-slash import warning and tp-related OOM when loading Qwen2-57B are unrelated." + }, + { + "left": "issue:44360", + "right": "issue:44561", + "accept": false, + "reason": "DSA indexer missing ReLU and removal of is_torch_fx_available breaking trust_remote_code models are different API/code changes." + }, + { + "left": "issue:43550", + "right": "issue:45083", + "accept": false, + "reason": "Bamba torch.compile/SDPA failure is not the same as qwen3_omni_moe feature-length behavior." + }, + { + "left": "issue:38175", + "right": "issue:44493", + "accept": false, + "reason": "Unexpected zero probabilities in SigLIP2 and unexpected position_id keys are separate model-level regressions." + }, + { + "left": "issue:29942", + "right": "issue:35141", + "accept": false, + "reason": "Flash Attention 2 test failures and post_init reinitializing output embeddings are unrelated issues." + }, + { + "left": "issue:27957", + "right": "issue:34689", + "accept": false, + "reason": "XLMRoberta Flash Attention 2 and Llama 3.2 Vision model loading on 4.46.2 are different bugs." + }, + { + "left": "issue:43278", + "right": "issue:43329", + "accept": false, + "reason": "Training/evaluate embedding dtype drift and undefined video-branch variables in _get_num_multimodal_tokens are unrelated." + }, + { + "left": "issue:29870", + "right": "issue:32090", + "accept": false, + "reason": "Bart generate decoder_attention_mask UnboundLocalError and _gpu_broadcast_one NoneType are distinct problems." + }, + { + "left": "issue:44448", + "right": "issue:44977", + "accept": false, + "reason": "Pegasus v4/v5 output difference and Qwen3.5 flash-attention generation failure are not the same underlying bug." + }, + { + "left": "issue:45920", + "right": "issue:46192", + "accept": false, + "reason": "Both touch tokenizer APIs, but wrong token IDs and return_tensors being ignored are different behaviors with different fixes." + }, + { + "left": "issue:40444", + "right": "issue:43526", + "accept": false, + "reason": "Qwen2.5-VL multi-image iterable dataset finetuning failure and BeitImageProcessorFast reduce_labels are unrelated." + }, + { + "left": "issue:43278", + "right": "issue:43381", + "accept": false, + "reason": "Embedding dtype change in eval and gradient checkpointing being disallowed in eval mode are separate eval-mode issues." + }, + { + "left": "issue:10000", + "right": "issue:15354", + "accept": false, + "reason": "German DistilBertModel issue and GeneratorExp unsupported by torch.jit.script are unrelated historical bugs." + }, + { + "left": "issue:44977", + "right": "issue:45229", + "accept": false, + "reason": "Qwen3.5 flash-attention generation and Gemma4 multi-GPU CUDA OOM are different classes of problem." + }, + { + "left": "issue:45290", + "right": "issue:45381", + "accept": false, + "reason": "Chat template tool-call crash and qwen2.5-vl video position_id bug affect different formatting/code paths." + }, + { + "left": "issue:44855", + "right": "issue:46399", + "accept": false, + "reason": "Both involve decorator/JIT-related errors, but the Python 3.13 IndentationError and missing-decorator registration bug are different failures." + }, + { + "left": "issue:29870", + "right": "issue:33290", + "accept": false, + "reason": "Bart generate crash and Adafactor OOM in deepspeed are unrelated training/inference issues." + }, + { + "left": "issue:44857", + "right": "issue:45588", + "accept": false, + "reason": "LwDetrImageLoss AMP crash and flash_attention.py s_aux=None AttributeError are separate code paths." + }, + { + "left": "issue:43927", + "right": "issue:45698", + "accept": false, + "reason": "DiaConfig losing custom token IDs on save/load and from_pretrained loading the wrong custom module are both persistence bugs but not the same one." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 31, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:25251", + "issue:29127", + "issue:29870", + "issue:32090", + "issue:33290", + "issue:33357", + "issue:34567", + "issue:34689", + "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:43873", + "issue:43976", + "issue:44186", + "issue:44361", + "issue:44448", + "issue:44466", + "issue:44561", + "issue:44568", + "issue:44743", + "issue:44779", + "issue:44871", + "issue:44977", + "issue:45198", + "issue:45229", + "issue:45290", + "issue:45372", + "issue:45406", + "issue:45440", + "issue:45491", + "issue:45563", + "issue:45588", + "issue:45803", + "issue:45823", + "issue:46121", + "issue:46655" + ], + "soft_pairs": [ + "issue:45290|issue:46655", + "issue:37428|issue:42175", + "issue:32090|issue:33290", + "issue:43873|issue:45229", + "issue:44561|issue:44779", + "issue:25251|issue:32090", + "issue:44448|issue:44466", + "issue:43976|issue:45440", + "issue:42490|issue:43873", + "issue:43278|issue:43575", + "issue:45588|issue:45823", + "issue:41720|issue:45229", + "issue:44568|issue:45491", + "issue:43381|issue:43701", + "issue:33357|issue:39692", + "issue:44977|issue:45563", + "issue:29127|issue:34567", + "issue:44871|issue:44977", + "issue:45198|issue:45803", + "issue:43653|issue:44743", + "issue:45372|issue:45588", + "issue:44361|issue:45198", + "issue:44186|issue:45198", + "issue:43611|issue:46121", + "issue:42915|issue:42994", + "issue:29870|issue:34689", + "issue:43412|issue:45588", + "issue:45372|issue:45406" + ], + "prepared_review_unit_hash": "adece12eb045bcf399c048a9f1614dbd6683447e2466c8d918e728e6b94015f7", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 35729, + "estimated_input_tokens": 8933, + "estimated_eval_tokens": 18122 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is very heterogeneous: the paired items share only broad Transformers-related wording, not the same underlying bug or fix. No soft pair should be merged; they span unrelated models, tokenizers, generation paths, packaging, and runtime errors.", + "confidence": 0.84, + "canonical_issue_reason": "If a single issue must anchor the cluster, issue 43873 is the most central by activity and inbound references and it covers a broad runtime concern (quantization/offloading). Still, it is not a true representative of the many unrelated bugs in this set.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43873 is the best available anchor because it has the most discussion/inbound references and a broad, user-facing runtime impact. It is still only a loose fit, not a duplicate umbrella for the rest.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45290", + "right": "issue:46655", + "accept": false, + "reason": "Assistant chat-template tool-call crash and video processor padding bug are different code paths and different failure modes." + }, + { + "left": "issue:37428", + "right": "issue:42175", + "accept": false, + "reason": "One is a flash-attention import error; the other is a packaging/backend dependency issue for TensorFlow." + }, + { + "left": "issue:32090", + "right": "issue:33290", + "accept": false, + "reason": "A NoneType broadcast error in Trainer is unrelated to Deepspeed Adafactor OOM." + }, + { + "left": "issue:43873", + "right": "issue:45229", + "accept": false, + "reason": "Quantization/offloading behavior is a different problem from Gemma4 multi-GPU CUDA OOM." + }, + { + "left": "issue:44561", + "right": "issue:44779", + "accept": false, + "reason": "Removal of a utility function breaking trust_remote_code models is unrelated to DeepSeek tokenizer output regressions." + }, + { + "left": "issue:25251", + "right": "issue:32090", + "accept": false, + "reason": "Pipeline top_k nesting behavior is unrelated to the GPU broadcast NoneType error." + }, + { + "left": "issue:44448", + "right": "issue:44466", + "accept": false, + "reason": "Both involve v5 behavior changes, but one is generation output drift and the other is serialization of tied weights; not the same fix." + }, + { + "left": "issue:43976", + "right": "issue:45440", + "accept": false, + "reason": "Python version compatibility and DeepSeekV3 implementation divergence are unrelated issues." + }, + { + "left": "issue:42490", + "right": "issue:43873", + "accept": false, + "reason": "Local-path import warning has nothing to do with quantization/offloading behavior." + }, + { + "left": "issue:43278", + "right": "issue:43575", + "accept": false, + "reason": "Embedding dtype drift in evaluate and Qwen2 tp OOM are different runtime problems." + }, + { + "left": "issue:45588", + "right": "issue:45823", + "accept": false, + "reason": "Flash-attention sinkless-model crash and Gemma4 PLE device mismatch are distinct forward-pass bugs." + }, + { + "left": "issue:41720", + "right": "issue:45229", + "accept": false, + "reason": "Qwen3 auto device mapping cuda assert and Gemma4 multi-GPU OOM are separate model-specific failures." + }, + { + "left": "issue:44568", + "right": "issue:45491", + "accept": false, + "reason": "Tokenizer special-token handling and Gemma3 sliding-window NaNs are unrelated." + }, + { + "left": "issue:43381", + "right": "issue:43701", + "accept": false, + "reason": "Eval-mode gradient checkpointing restriction and resume checkpoint key mismatch are different training-state issues." + }, + { + "left": "issue:33357", + "right": "issue:39692", + "accept": false, + "reason": "MacOS bus error in CLIP loading and SigLIP2 documentation/example errors do not share the same bug." + }, + { + "left": "issue:44977", + "right": "issue:45563", + "accept": false, + "reason": "Flash-attention generation failure and a stale warning in paged generate are not the same change." + }, + { + "left": "issue:29127", + "right": "issue:34567", + "accept": false, + "reason": "LayoutLMv3 error clarity and TrainerState token counter not updating are unrelated." + }, + { + "left": "issue:44871", + "right": "issue:44977", + "accept": false, + "reason": "Gemma eos_token_id config mismatch is a different issue from Qwen3.5 flash-attention generation failure." + }, + { + "left": "issue:45198", + "right": "issue:45803", + "accept": false, + "reason": "Wav2Vec2 save/tokenization failure and a bare-except cleanup in FuyuBatchFeature are unrelated." + }, + { + "left": "issue:43653", + "right": "issue:44743", + "accept": false, + "reason": "BigBirdTokenizer special-token registration and Qwen3_5 recurrent-state reset are different tokenizer/model behaviors." + }, + { + "left": "issue:45372", + "right": "issue:45588", + "accept": false, + "reason": "Gemma4 processor import failure and flash-attention s_aux=None crash are separate errors in different layers." + }, + { + "left": "issue:44361", + "right": "issue:45198", + "accept": false, + "reason": "MLukeTokenizer AttributeError and Wav2Vec2 save/tokenization failure do not indicate the same underlying bug." + }, + { + "left": "issue:44186", + "right": "issue:45198", + "accept": false, + "reason": "LayoutLMv2 NER/padding crash and Wav2Vec2 save/tokenization fail are unrelated tokenizer issues." + }, + { + "left": "issue:43611", + "right": "issue:46121", + "accept": false, + "reason": "base_model_prefix loading regression and rope-validation JSON list type error affect different loading/validation paths." + }, + { + "left": "issue:42915", + "right": "issue:42994", + "accept": false, + "reason": "Qwen3Moe FP8 config failure and quantized model saving failure are different quantization workflows." + }, + { + "left": "issue:29870", + "right": "issue:34689", + "accept": false, + "reason": "BART generate decoder mask bug and Llama 3.2 model loading break are unrelated model-specific regressions." + }, + { + "left": "issue:43412", + "right": "issue:45588", + "accept": false, + "reason": "ExecuTorch export problems for Gemma3n and a flash-attention runtime crash are different fixes." + }, + { + "left": "issue:45372", + "right": "issue:45406", + "accept": false, + "reason": "Gemma4 processor import error and transformers serve missing _tokenizer are distinct processor-loading/serving issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 32, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:27957", + "issue:29127", + "issue:29942", + "issue:30064", + "issue:33290", + "issue:33357", + "issue:34689", + "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:45372", + "issue:45412", + "issue:45440", + "issue:45459", + "issue:45491", + "issue:45563", + "issue:45593", + "issue:45712", + "issue:45834", + "issue:46121", + "issue:46369", + "issue:46399", + "issue:46693" + ], + "soft_pairs": [ + "issue:44206|issue:44479", + "issue:29942|issue:30064", + "issue:43756|issue:45440", + "issue:43653|issue:44442", + "issue:45459|issue:45712", + "issue:45491|issue:46693", + "issue:43065|issue:45412", + "issue:29127|issue:33357", + "issue:43065|issue:43994", + "issue:45245|issue:45563", + "issue:45372|issue:45459", + "issue:44265|issue:45198", + "issue:43575|issue:43701", + "issue:45440|issue:45593", + "issue:43066|issue:44462", + "issue:29127|issue:35141", + "issue:42490|issue:43381", + "issue:43994|issue:45412", + "issue:45412|issue:46369", + "issue:43421|issue:45491", + "issue:43575|issue:44387", + "issue:44442|issue:45072", + "issue:42915|issue:43931", + "issue:43526|issue:45834", + "issue:43526|issue:46693", + "issue:43824|issue:44279", + "issue:43526|issue:44186", + "issue:27957|issue:29127", + "issue:27957|issue:33290", + "issue:46121|issue:46399", + "issue:43994|issue:46369", + "issue:43334|issue:43531", + "issue:33290|issue:34689" + ], + "prepared_review_unit_hash": "542a67a1ef1a3181ef9a7ce847a71f549446d4720270a522579ec8391ad16085", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 33, + "serialized_chars": 37514, + "estimated_input_tokens": 9379, + "estimated_eval_tokens": 19014 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly a grab-bag of unrelated Transformers issues; I don\u2019t see any soft pair that looks like the same underlying bug/change. No PRs are present. If one issue must anchor the cluster, the broad and active FA2 regression report is the best fit.", + "confidence": 0.9, + "canonical_issue_reason": "issue:46693 is the most active, broadly scoped open bug report and is the best representative anchor if this cluster must be summarized around one issue.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:46693 has the strongest combination of discussion, current relevance, and generalized failure mode compared with the other highly specific reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44206", + "right": "issue:44479", + "accept": false, + "reason": "Different regressions: unsupported feature-extractor arg vs video input handling for Qwen VL models." + }, + { + "left": "issue:29942", + "right": "issue:30064", + "accept": false, + "reason": "Unrelated areas: Flash Attention 2 test failures vs segmentation-map processing in image processor." + }, + { + "left": "issue:43756", + "right": "issue:45440", + "accept": false, + "reason": "Different model families and failure modes; RoPE-layer mismatch is not the same as DeepSeek V3 implementation divergence." + }, + { + "left": "issue:43653", + "right": "issue:44442", + "accept": false, + "reason": "Tokenizer special-token registration bug vs AutoTokenizer loading a specific tokenizer file; not the same code path." + }, + { + "left": "issue:45459", + "right": "issue:45712", + "accept": false, + "reason": "One is tokenizer error masking without protobuf, the other is leftover dummy classes leaking into module namespace." + }, + { + "left": "issue:45491", + "right": "issue:46693", + "accept": false, + "reason": "Both mention FA2/long-context, but one is NaN embeddings in Gemma3 and the other is a performance regression; not the same concrete bug." + }, + { + "left": "issue:43065", + "right": "issue:45412", + "accept": false, + "reason": "Dummy Conv2d in Sam3PixelDecoder is unrelated to RT-DETR memory not being released." + }, + { + "left": "issue:29127", + "right": "issue:33357", + "accept": false, + "reason": "Clarity of an error message in LayoutLMv3 vs a MacOS bus error in CLIP loading are unrelated." + }, + { + "left": "issue:43065", + "right": "issue:43994", + "accept": false, + "reason": "Different models and problems: dummy layer construction vs nonsensical SigLIP2 outputs." + }, + { + "left": "issue:45245", + "right": "issue:45563", + "accept": false, + "reason": "Category-limit runtime error is unrelated to a stale warning in paged generate()." + }, + { + "left": "issue:45372", + "right": "issue:45459", + "accept": false, + "reason": "Dependency import breakage for Gemma 4 processor is unrelated to protobuf-based tokenizer error handling." + }, + { + "left": "issue:44265", + "right": "issue:45198", + "accept": false, + "reason": "torch.export failure with torch_compilable_check is unrelated to Wav2Vec2 save/tokenization failures." + }, + { + "left": "issue:43575", + "right": "issue:43701", + "accept": false, + "reason": "TP OOM when loading a Qwen model is unrelated to resume_from_checkpoint key mismatch." + }, + { + "left": "issue:45440", + "right": "issue:45593", + "accept": false, + "reason": "DeepSeekV3 implementation divergence is unrelated to auxiliary-loss behavior in D-FINE." + }, + { + "left": "issue:43066", + "right": "issue:44462", + "accept": false, + "reason": "Wrong tokenizer decoder type in v5 is unrelated to AutoTokenizer ignoring tokenizer.json." + }, + { + "left": "issue:29127", + "right": "issue:35141", + "accept": false, + "reason": "Error-message clarity in LayoutLMv3 is unrelated to embedding reinitialization after resize." + }, + { + "left": "issue:42490", + "right": "issue:43381", + "accept": false, + "reason": "Local-path import warning and gradient-checkpointing-in-eval are separate behavioral issues." + }, + { + "left": "issue:43994", + "right": "issue:45412", + "accept": false, + "reason": "SigLIP2 inference quality issue is unrelated to RT-DETR memory retention on deletion." + }, + { + "left": "issue:45412", + "right": "issue:46369", + "accept": false, + "reason": "Memory leak on GC and memory overflow when calling sam3-litetext are different failure modes with different triggers." + }, + { + "left": "issue:43421", + "right": "issue:45491", + "accept": false, + "reason": "Runtime special-token post-processing bug is unrelated to Gemma3 sliding-window NaNs." + }, + { + "left": "issue:43575", + "right": "issue:44387", + "accept": false, + "reason": "Both are memory-related, but TP OOM and int4 reserved-memory regression are not the same code-path problem." + }, + { + "left": "issue:44442", + "right": "issue:45072", + "accept": false, + "reason": "Tokenizer loading failure is unrelated to bfloat16 dtype mismatches in inference." + }, + { + "left": "issue:42915", + "right": "issue:43931", + "accept": false, + "reason": "FP8 config failure for Qwen3Moe is unrelated to Qwen3-VL weight-shape mismatch." + }, + { + "left": "issue:43526", + "right": "issue:45834", + "accept": false, + "reason": "Label reduction bug in BeitImageProcessorFast is unrelated to Kosmos2.5 OCR index errors." + }, + { + "left": "issue:43526", + "right": "issue:46693", + "accept": false, + "reason": "Different subsystems and symptoms: image-label reduction vs FA2 long-context performance regression." + }, + { + "left": "issue:43824", + "right": "issue:44279", + "accept": false, + "reason": "Missing Qwen2.5-VL import is not the same as a generic dependency issue." + }, + { + "left": "issue:43526", + "right": "issue:44186", + "accept": false, + "reason": "BeitImageProcessorFast label handling and LayoutLMv2Tokenizer NER/padding crashes are unrelated." + }, + { + "left": "issue:27957", + "right": "issue:29127", + "accept": false, + "reason": "Flash Attention 2 support for XLM-Roberta is unrelated to LayoutLMv3 error-message clarity." + }, + { + "left": "issue:27957", + "right": "issue:33290", + "accept": false, + "reason": "FA2 model support and deepspeed adafactor OOM are separate issues." + }, + { + "left": "issue:46121", + "right": "issue:46399", + "accept": false, + "reason": "RoPE validation TypeError and Qwen3.5 kernel registration failure are different bugs." + }, + { + "left": "issue:43994", + "right": "issue:46369", + "accept": false, + "reason": "Different models and failure modes: nonsensical outputs vs memory overflow." + }, + { + "left": "issue:43334", + "right": "issue:43531", + "accept": false, + "reason": "Qwen3-VL pad_token_id loading error is unrelated to Qwen3-MoE sliding_window behavior." + }, + { + "left": "issue:33290", + "right": "issue:34689", + "accept": false, + "reason": "Optimizer OOM in deepspeed is unrelated to Llama 3.2 Vision model loading regression." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 33, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:27957", + "issue:29870", + "issue:33290", + "issue:35141", + "issue:42175", + "issue:42490", + "issue:43262", + "issue:43316", + "issue:43381", + "issue:43493", + "issue:43525", + "issue:43643", + "issue:43646", + "issue:43720", + "issue:43723", + "issue:43749", + "issue:43761", + "issue:43827", + "issue:43873", + "issue:43931", + "issue:44062", + "issue:44077", + "issue:44079", + "issue:44206", + "issue:44442", + "issue:44479", + "issue:44514", + "issue:44743", + "issue:44811", + "issue:44871", + "issue:44898", + "issue:44913", + "issue:44977", + "issue:45003", + "issue:45216", + "issue:45245", + "issue:45250", + "issue:45325", + "issue:45381", + "issue:45440", + "issue:45479", + "issue:45588", + "issue:45600", + "issue:45859", + "issue:45907", + "issue:46234", + "issue:46640", + "issue:46697" + ], + "soft_pairs": [ + "issue:45250|issue:46640", + "issue:33290|issue:35141", + "issue:43761|issue:45859", + "issue:45588|issue:45600", + "issue:43381|issue:43873", + "issue:44514|issue:45381", + "issue:43646|issue:44077", + "issue:43262|issue:44206", + "issue:44977|issue:45245", + "issue:44743|issue:44811", + "issue:44442|issue:44898", + "issue:42490|issue:44743", + "issue:45003|issue:45440", + "issue:43720|issue:44442", + "issue:42175|issue:46234", + "issue:27957|issue:29870", + "issue:43316|issue:43525", + "issue:43723|issue:45479", + "issue:43493|issue:43827", + "issue:44077|issue:44479", + "issue:45325|issue:46697", + "issue:43873|issue:44871", + "issue:44062|issue:45907", + "issue:43761|issue:44079", + "issue:43749|issue:43931", + "issue:43931|issue:45216", + "issue:43643|issue:44913" + ], + "prepared_review_unit_hash": "97d75f214290700cd9651ca03509696005d77115cab15e1eac2783e7edd4ef7d", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 35844, + "estimated_input_tokens": 8961, + "estimated_eval_tokens": 18178 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is noisy: the soft pairs only share broad subsystem vocabulary and do not describe the same concrete bug or change, so none should be merged.", + "confidence": 0.93, + "canonical_issue_reason": "issue:43749 is the strongest anchor in this set: it has the most discussion/inbound references and a detailed, high-signal report, making it the best representative issue for the cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43749 is the most informative and central issue here; if one issue must represent the cluster, it is the clearest choice.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45250", + "right": "issue:46640", + "accept": false, + "reason": "Both mention Flash Attention 2, but one is a generic topic stub and the other is a specific flash-attn bug report; not the same underlying issue." + }, + { + "left": "issue:33290", + "right": "issue:35141", + "accept": false, + "reason": "OOM with Adafactor/DeepSpeed and embedding reinitialization after resize are unrelated problems." + }, + { + "left": "issue:43761", + "right": "issue:45859", + "accept": false, + "reason": "CLIPVision hidden_states regression and Qwen3_5MoE CPU offload incompatibility affect different code paths and symptoms." + }, + { + "left": "issue:45588", + "right": "issue:45600", + "accept": false, + "reason": "A flash_attention AttributeError on s_aux=None is unrelated to auto_mappings referencing removed configs." + }, + { + "left": "issue:43381", + "right": "issue:43873", + "accept": false, + "reason": "Gradient checkpointing in eval mode and quantization offloading are separate behaviors with different failure modes." + }, + { + "left": "issue:44514", + "right": "issue:45381", + "accept": false, + "reason": "Batched apply_chat_template padding behavior and Qwen2.5-VL video position_ids are different bugs." + }, + { + "left": "issue:43646", + "right": "issue:44077", + "accept": false, + "reason": "Custom model initialization breakage and a specific patchtsmixer post_init API issue are not the same code-path problem." + }, + { + "left": "issue:43262", + "right": "issue:44206", + "accept": false, + "reason": "Audio chat template sampling-rate defaults and LaserFeatureExtractor center-arg crashes are unrelated." + }, + { + "left": "issue:44977", + "right": "issue:45245", + "accept": false, + "reason": "Qwen3.5 flash-attention generation and a category-cardinality runtime error are unrelated." + }, + { + "left": "issue:44743", + "right": "issue:44811", + "accept": false, + "reason": "Recurrent-state reset in modular_qwen3_5 and Whisper batch_decode skip_special_tokens are different components." + }, + { + "left": "issue:44442", + "right": "issue:44898", + "accept": false, + "reason": "Tokenizer loading failure and Perceiver image-size handling are unrelated model-specific bugs." + }, + { + "left": "issue:42490", + "right": "issue:44743", + "accept": false, + "reason": "Local path import warning and Qwen3_5 recurrent-state reset do not share the same bug." + }, + { + "left": "issue:45003", + "right": "issue:45440", + "accept": false, + "reason": "Unsafe sys.modules access and DeepseekV3MoE divergence are not the same defect." + }, + { + "left": "issue:43720", + "right": "issue:44442", + "accept": false, + "reason": "BitNet packed-weight loading and tokenizer loading for FastSpeech2Conformer are unrelated." + }, + { + "left": "issue:42175", + "right": "issue:46234", + "accept": false, + "reason": "Missing tensorflow in a pip extra and a missing import_utils helper are different packaging/import issues." + }, + { + "left": "issue:27957", + "right": "issue:29870", + "accept": false, + "reason": "XLMRoberta Flash Attention 2 support and BART decoder_attention_mask assignment are unrelated." + }, + { + "left": "issue:43316", + "right": "issue:43525", + "accept": false, + "reason": "A generic config API discrepancy and Llama4Config missing pad_token_id are not the same underlying bug." + }, + { + "left": "issue:43723", + "right": "issue:45479", + "accept": false, + "reason": "Tokenizer loading in v5 and zero-loss sequence classification are unrelated regressions." + }, + { + "left": "issue:43493", + "right": "issue:43827", + "accept": false, + "reason": "SigLIP2 JAX/HF divergence and docs still referencing pipeline() are different problem classes." + }, + { + "left": "issue:44077", + "right": "issue:44479", + "accept": false, + "reason": "PatchTSMixer post_init policy and Qwen video-input regression are unrelated." + }, + { + "left": "issue:45325", + "right": "issue:46697", + "accept": false, + "reason": "Qwen2.5-VL rope-index scaling and Swin2SR over-padding are different image/video preprocessing bugs." + }, + { + "left": "issue:43873", + "right": "issue:44871", + "accept": false, + "reason": "Quantization offloading and Gemma-3 eos_token_id inconsistency are unrelated." + }, + { + "left": "issue:44062", + "right": "issue:45907", + "accept": false, + "reason": "AddedToken special-arg duplication and list_repo_files revision forwarding are unrelated API bugs." + }, + { + "left": "issue:43761", + "right": "issue:44079", + "accept": false, + "reason": "CLIPVisionModel hidden_states regression and ModelOutput key assignment are different internal behaviors." + }, + { + "left": "issue:43749", + "right": "issue:43931", + "accept": false, + "reason": "FSDP CPU RAM efficient loading and Qwen3-VL shape mismatch are unrelated loading failures." + }, + { + "left": "issue:43931", + "right": "issue:45216", + "accept": false, + "reason": "Model shape mismatch and incorrect save_pretrained output are different Qwen3-related regressions, not one bug." + }, + { + "left": "issue:43643", + "right": "issue:44913", + "accept": false, + "reason": "trust_remote_code missing fields and GPTNeoX rotary_pct reload behavior are distinct config-handling issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 34, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:16998", + "issue:22355", + "issue:25251", + "issue:27957", + "issue:28282", + "issue:32090", + "issue:42175", + "issue:42617", + "issue:42907", + "issue:43232", + "issue:43295", + "issue:43316", + "issue:43388", + "issue:43452", + "issue:43550", + "issue:43575", + "issue:43606", + "issue:43646", + "issue:43701", + "issue:43716", + "issue:43742", + "issue:43761", + "issue:43825", + "issue:44077", + "issue:44188", + "issue:44190", + "issue:44442", + "issue:44464", + "issue:44496", + "issue:44534", + "issue:44568", + "issue:44855", + "issue:44871", + "issue:44913", + "issue:44987", + "issue:45137", + "issue:45479", + "issue:45584", + "issue:45698", + "issue:45759", + "issue:45803", + "issue:46050", + "issue:46178", + "issue:46234", + "issue:46291", + "issue:46365", + "issue:46421", + "issue:46506" + ], + "soft_pairs": [ + "issue:43316|issue:45759", + "issue:44855|issue:45137", + "issue:44987|issue:46178", + "issue:43761|issue:44188", + "issue:44190|issue:44496", + "issue:43232|issue:43388", + "issue:44442|issue:45803", + "issue:42907|issue:45698", + "issue:43701|issue:46506", + "issue:42617|issue:43575", + "issue:43575|issue:43742", + "issue:27957|issue:32090", + "issue:25251|issue:28282", + "issue:43550|issue:43701", + "issue:44568|issue:45479", + "issue:46234|issue:46291", + "issue:43606|issue:43701", + "issue:44464|issue:46421", + "issue:16998|issue:22355", + "issue:43295|issue:43825", + "issue:43295|issue:44871", + "issue:44871|issue:45584", + "issue:43316|issue:43716", + "issue:42175|issue:43316", + "issue:44077|issue:44534", + "issue:44871|issue:46050", + "issue:43646|issue:44913", + "issue:43452|issue:44496", + "issue:45803|issue:46365" + ], + "prepared_review_unit_hash": "f685435ffbbfcc7a0df298f72517b3cea4de3c701ec28139ea307d8e209b712f", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36449, + "estimated_input_tokens": 9113, + "estimated_eval_tokens": 18482 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: the paired issues span unrelated loading, generation, multimodal, backend, and distributed-training problems. None of the soft pairs looks like the same underlying bug or change, so I would not merge any of them.", + "confidence": 0.62, + "canonical_issue_reason": "Issue 45698 is the most active open report and the clearest concrete save/load regression in the set, so it is the best single anchor issue here.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45698 has the strongest discussion signal and a well-defined reproduction path, making it the most useful representative issue among these items.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43316", + "right": "issue:45759", + "accept": false, + "reason": "Both are multimodal-config related, but they hit different code paths: Gemma3TextConfig API mismatch vs from_config not unwrapping text_config." + }, + { + "left": "issue:44855", + "right": "issue:45137", + "accept": false, + "reason": "Unrelated failures: Python 3.13 TorchScript parsing on import vs a DeepSpeed ZeRO3 deque IndexError." + }, + { + "left": "issue:44987", + "right": "issue:46178", + "accept": false, + "reason": "Both are load/import problems, but one is a model-specific v5.1.0 regression and the other is a missing tensorflow_text backend mapping." + }, + { + "left": "issue:43761", + "right": "issue:44188", + "accept": false, + "reason": "CLIP hidden_states regression and torch.compile attention-kernel divergence are different bugs." + }, + { + "left": "issue:44190", + "right": "issue:44496", + "accept": false, + "reason": "Example script dataset loading failure and unrecognized-model config loading failure are unrelated." + }, + { + "left": "issue:43232", + "right": "issue:43388", + "accept": false, + "reason": "Generation kwargs sync bug vs gather_for_metrics dropping tuple labels; different subsystems and symptoms." + }, + { + "left": "issue:44442", + "right": "issue:45803", + "accept": false, + "reason": "Tokenizer class loading failure and bare-except tensor conversion bug are separate issues." + }, + { + "left": "issue:42907", + "right": "issue:45698", + "accept": false, + "reason": "Both involve save/load behavior, but one is dequantized-model persistence and the other is wrong custom module restoration." + }, + { + "left": "issue:43701", + "right": "issue:46506", + "accept": false, + "reason": "Checkpoint resume key mismatch is unrelated to CPU/CUDA output divergence in PPDocLayoutV3." + }, + { + "left": "issue:42617", + "right": "issue:43575", + "accept": false, + "reason": "3d_parallel.py failure and Qwen TP OOM are different distributed-loading problems." + }, + { + "left": "issue:43575", + "right": "issue:43742", + "accept": false, + "reason": "OOM on a large TP load and a key error on MobileLLM loading are not the same bug." + }, + { + "left": "issue:27957", + "right": "issue:32090", + "accept": false, + "reason": "Flash Attention 2 support issue and Trainer _gpu_broadcast_one NoneType error are unrelated." + }, + { + "left": "issue:25251", + "right": "issue:28282", + "accept": false, + "reason": "Pipeline output-shape bug and missing PyTorch ImportError are entirely different." + }, + { + "left": "issue:43550", + "right": "issue:43701", + "accept": false, + "reason": "torch.compile SDPA failure and resume_from_checkpoint key mismatch do not share a concrete code path." + }, + { + "left": "issue:44568", + "right": "issue:45479", + "accept": false, + "reason": "Tokenizer special-token regression and degenerate single-label classification loss are different bugs." + }, + { + "left": "issue:46234", + "right": "issue:46291", + "accept": false, + "reason": "Missing import_utils helper and kernels v0.15 import requirement are separate optional-dependency issues." + }, + { + "left": "issue:43606", + "right": "issue:43701", + "accept": false, + "reason": "CPU offload device mismatch and checkpoint key mismatch are unrelated." + }, + { + "left": "issue:44464", + "right": "issue:46421", + "accept": false, + "reason": "Both mention chunked generation, but one is output inconsistency under compiled forward while the other is cache-init causing recompilation; not the same failure." + }, + { + "left": "issue:16998", + "right": "issue:22355", + "accept": false, + "reason": "Model_max_length question and missing transformers.onnx module are unrelated." + }, + { + "left": "issue:43295", + "right": "issue:43825", + "accept": false, + "reason": "Custom processor/tokenizer regression and pipeline error-message regression are different issues." + }, + { + "left": "issue:43295", + "right": "issue:44871", + "accept": false, + "reason": "Processor/tokenizer image-passing regression and Gemma3 eos_token_id mismatch affect different models and code paths." + }, + { + "left": "issue:44871", + "right": "issue:45584", + "accept": false, + "reason": "Gemma3 eos_token config inconsistency is not the same as Whisper empty-transcription failure after align_special_tokens." + }, + { + "left": "issue:43316", + "right": "issue:43716", + "accept": false, + "reason": "Both are multimodal, but one is a config API discrepancy and the other is an image-preprocessor dtype mismatch." + }, + { + "left": "issue:42175", + "right": "issue:43316", + "accept": false, + "reason": "Installation/backend packaging issue vs Gemma3 config API discrepancy." + }, + { + "left": "issue:44077", + "right": "issue:44534", + "accept": false, + "reason": "PatchTSMixer post_init policy and non-persistent buffer junk are unrelated v5 issues." + }, + { + "left": "issue:44871", + "right": "issue:46050", + "accept": false, + "reason": "EOS-token configuration mismatch is not the same as quantized model save/load corruption." + }, + { + "left": "issue:43646", + "right": "issue:44913", + "accept": false, + "reason": "Broad custom-model initialization breakage and GPTNeoX rotary_pct reload reset are different problems." + }, + { + "left": "issue:43452", + "right": "issue:44496", + "accept": false, + "reason": "gguf_file loader breakage and missing model_type in config.json are separate model-loading failures." + }, + { + "left": "issue:45803", + "right": "issue:46365", + "accept": false, + "reason": "Bare-except in Fuyu conversion and Trainer import blocked by datasets are unrelated import/runtime issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 35, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "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:44162", + "issue:44190", + "issue:44479", + "issue:44514", + "issue:44534", + "issue:44610", + "issue:44617", + "issue:44734", + "issue:44743", + "issue:44792", + "issue:44913", + "issue:44964", + "issue:44977", + "issue:45127", + "issue:45200", + "issue:45406", + "issue:45542", + "issue:45600", + "issue:45632", + "issue:45698", + "issue:45803", + "issue:46082", + "issue:46170", + "issue:46178", + "issue:46202", + "issue:46234", + "issue:46369", + "issue:46424", + "issue:46640", + "issue:46697" + ], + "soft_pairs": [ + "issue:44162|issue:44792", + "issue:45803|issue:46170", + "issue:43404|issue:43441", + "issue:45632|issue:46424", + "issue:44610|issue:46369", + "issue:43650|issue:46202", + "issue:43761|issue:44479", + "issue:43572|issue:46178", + "issue:45542|issue:46234", + "issue:43540|issue:44514", + "issue:43065|issue:44617", + "issue:43504|issue:45698", + "issue:45127|issue:45200", + "issue:43388|issue:43688", + "issue:44792|issue:44977", + "issue:43352|issue:46640", + "issue:44792|issue:45600", + "issue:43673|issue:44743", + "issue:41950|issue:43475", + "issue:43525|issue:43531", + "issue:44190|issue:44964", + "issue:44534|issue:44913", + "issue:35141|issue:42490", + "issue:44734|issue:45406", + "issue:46082|issue:46697" + ], + "prepared_review_unit_hash": "b513df637688cf5909dbcd0f558a2a0b1e3e5c8b7efe6b32228be15d2a7b0592", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 25, + "serialized_chars": 35153, + "estimated_input_tokens": 8789, + "estimated_eval_tokens": 17834 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These soft pairs are mostly unrelated false positives across different models, pipelines, and code paths. I would reject every soft edge; there is no single duplicate bug cluster here.", + "confidence": 0.95, + "canonical_issue_reason": "No single canonical issue: the items span unrelated bugs, models, and subsystems rather than one underlying defect.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45698 is the most broadly impactful open bug in the set, but it still does not match the others well enough to serve as a duplicate anchor.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44162", + "right": "issue:44792", + "accept": false, + "reason": "Different bugs: ESM2 breakage vs. Janus image-generation test failure." + }, + { + "left": "issue:45803", + "right": "issue:46170", + "accept": false, + "reason": "Unrelated causes: bare except in Fuyu tensor conversion vs Windows ROCm torch.distributed import error." + }, + { + "left": "issue:43404", + "right": "issue:43441", + "accept": false, + "reason": "Mistral3 lm_head tying bug and Ministral-3 FlashAttention support failure are different code paths." + }, + { + "left": "issue:45632", + "right": "issue:46424", + "accept": false, + "reason": "Cache-path collision and max_cache_length handling are separate cache issues." + }, + { + "left": "issue:44610", + "right": "issue:46369", + "accept": false, + "reason": "Processor/model input-size mismatch is unrelated to SAM3-LiteText memory overflow." + }, + { + "left": "issue:43650", + "right": "issue:46202", + "accept": false, + "reason": "No clear shared bug; one title is non-technical while the other is a Gemma tokenizer roundtrip failure." + }, + { + "left": "issue:43761", + "right": "issue:44479", + "accept": false, + "reason": "CLIP hidden_states regression and Qwen video-input regression are different failures." + }, + { + "left": "issue:43572", + "right": "issue:46178", + "accept": false, + "reason": "Missing pad_token_idx in StableLmConfig is unrelated to tensorflow_text backend mapping." + }, + { + "left": "issue:45542", + "right": "issue:46234", + "accept": false, + "reason": "Different import/backend problems: tensorboard-only TF backend vs missing import_utils symbol." + }, + { + "left": "issue:43540", + "right": "issue:44514", + "accept": false, + "reason": "Qwen3OmniMoe video processing error and Qwen2_5_VL batched chat-template crash are distinct." + }, + { + "left": "issue:43065", + "right": "issue:44617", + "accept": false, + "reason": "Dummy Conv2d in Sam3PixelDecoder is not the same as SAM3Video CUDA OOM." + }, + { + "left": "issue:43504", + "right": "issue:45698", + "accept": false, + "reason": "Legacy field loading in BEiT and wrong custom module loading after save_pretrained are different bugs." + }, + { + "left": "issue:45127", + "right": "issue:45200", + "accept": false, + "reason": "LoRA merge collapse with tied embeddings is unrelated to Gemma4 mm_token_type_ids defaulting." + }, + { + "left": "issue:43388", + "right": "issue:43688", + "accept": false, + "reason": "Gather truncation in metric collection and auxiliary-loss normalization are separate issues." + }, + { + "left": "issue:44792", + "right": "issue:44977", + "accept": false, + "reason": "Janus test failure and Qwen3.5 FlashAttention generation bug do not share the same root cause." + }, + { + "left": "issue:43352", + "right": "issue:46640", + "accept": false, + "reason": "Too generic to merge: specific Nemotron FA2 unsupported issue vs a vague Flash Attention 2 report." + }, + { + "left": "issue:44792", + "right": "issue:45600", + "accept": false, + "reason": "Janus image-generation failure is unrelated to Sam3LiteText auto-mapping/CI breakage." + }, + { + "left": "issue:43673", + "right": "issue:44743", + "accept": false, + "reason": "Missing GenerationMixin cache in chunked prefill and recurrent-state reset with cache are different cache bugs." + }, + { + "left": "issue:41950", + "right": "issue:43475", + "accept": false, + "reason": "Video-classification processor lookup and missing fpn_position_embeddings in SAM3 are unrelated." + }, + { + "left": "issue:43525", + "right": "issue:43531", + "accept": false, + "reason": "Missing pad_token_id in Llama4Config is not the same as the Qwen3-MoE sliding_window issue." + }, + { + "left": "issue:44190", + "right": "issue:44964", + "accept": false, + "reason": "Local dataset loading failure and Phi-4 multimodal loading failure are unrelated." + }, + { + "left": "issue:44534", + "right": "issue:44913", + "accept": false, + "reason": "Non-persistent buffer junk and rotary_pct reload regression are different config/state bugs." + }, + { + "left": "issue:35141", + "right": "issue:42490", + "accept": false, + "reason": "Embedding resize reinitialization and trailing-slash import warning are unrelated loading behaviors." + }, + { + "left": "issue:44734", + "right": "issue:45406", + "accept": false, + "reason": "KV-cache continuation tensor indexing and Gemma4Processor missing _tokenizer are different serve failures." + }, + { + "left": "issue:46082", + "right": "issue:46697", + "accept": false, + "reason": "LlamaConfig head_dim validation is unrelated to Swin2SR over-padding." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 36, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:25251", + "issue:27957", + "issue:29870", + "issue:35141", + "issue:39692", + "issue:41720", + "issue:43065", + "issue:43335", + "issue:43526", + "issue:43550", + "issue:43644", + "issue:43673", + "issue:43761", + "issue:43827", + "issue:43931", + "issue:43994", + "issue:44060", + "issue:44077", + "issue:44112", + "issue:44190", + "issue:44242", + "issue:44464", + "issue:44479", + "issue:44610", + "issue:44617", + "issue:44734", + "issue:44779", + "issue:44855", + "issue:44877", + "issue:44913", + "issue:44933", + "issue:45127", + "issue:45245", + "issue:45357", + "issue:45362", + "issue:45381", + "issue:45563", + "issue:45584", + "issue:45588", + "issue:45600", + "issue:45632", + "issue:46369", + "issue:46421", + "issue:46489", + "issue:46617", + "issue:46655", + "issue:46697" + ], + "soft_pairs": [ + "issue:45632|issue:46617", + "issue:43931|issue:44060", + "issue:44855|issue:45588", + "issue:43827|issue:43994", + "issue:43673|issue:46421", + "issue:43761|issue:44242", + "issue:29870|issue:35141", + "issue:44617|issue:45584", + "issue:45563|issue:45584", + "issue:43644|issue:44913", + "issue:44933|issue:45362", + "issue:25251|issue:27957", + "issue:43335|issue:44855", + "issue:44479|issue:44734", + "issue:45381|issue:46369", + "issue:44077|issue:44877", + "issue:43550|issue:44190", + "issue:41720|issue:45245", + "issue:43335|issue:44190", + "issue:43065|issue:44610", + "issue:44464|issue:45563", + "issue:45127|issue:45357", + "issue:39692|issue:44190", + "issue:45381|issue:46655", + "issue:44112|issue:45600", + "issue:43526|issue:46697", + "issue:44617|issue:46369", + "issue:44779|issue:46489" + ], + "prepared_review_unit_hash": "25d6542070ee4375c0e6afd4e70c1237556d205f60a2626edb97a9343d93da60", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "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": "The cluster is heterogeneous: most items are distinct model-, processor-, config-, or generation-path bugs, and none of the soft-edge pairs look like the same underlying change closely enough to merge.", + "confidence": 0.95, + "canonical_issue_reason": "issue:46489 is the most detailed and actionable report in the set, with a concrete DeepSeek-Coder v1 v5 regression and an explicit link to a gap in a prior fix, making it the strongest single anchor even though the cluster is not truly duplicate-heavy.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:46489 is the best standalone issue to keep as a representative because it has the clearest root-cause framing and reproduction context.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45632", + "right": "issue:46617", + "accept": false, + "reason": "Both involve trust_remote_code/cache, but one is a leaf-directory collision for local models while the other is a FileNotFoundError during cache load; different failure modes." + }, + { + "left": "issue:43931", + "right": "issue:44060", + "accept": false, + "reason": "Different model families and symptoms: load-time weight-shape mismatch versus a tied-weights warning bug in Qwen3-Next." + }, + { + "left": "issue:44855", + "right": "issue:45588", + "accept": false, + "reason": "Unrelated problems: Python 3.13/JIT parsing import error versus a flash-attention runtime crash on sink-less models." + }, + { + "left": "issue:43827", + "right": "issue:43994", + "accept": false, + "reason": "One is stale docs still mentioning pipeline(), the other is a SigLIP2 model/pipeline correctness issue; not the same bug." + }, + { + "left": "issue:43673", + "right": "issue:46421", + "accept": false, + "reason": "Both touch chunked prefill/cache, but one is missing cache in v5.0.0 and the other is lazy-init causing recompilation; distinct code-path issues." + }, + { + "left": "issue:43761", + "right": "issue:44242", + "accept": false, + "reason": "Different subsystems: hidden-states propagation in CLIPVisionModel versus omission of load-balancing loss when router logits are disabled." + }, + { + "left": "issue:29870", + "right": "issue:35141", + "accept": false, + "reason": "Both are generation/model-init regressions, but one is an unbound decoder_attention_mask path and the other is embedding reinitialization after resize; not the same fix." + }, + { + "left": "issue:44617", + "right": "issue:45584", + "accept": false, + "reason": "CUDA OOM in Sam3Video is unrelated to Whisper empty-transcription handling after align_special_tokens." + }, + { + "left": "issue:45563", + "right": "issue:45584", + "accept": false, + "reason": "A stale warning in paged generate() is a different bug from Whisper failing on empty transcription." + }, + { + "left": "issue:43644", + "right": "issue:44913", + "accept": false, + "reason": "A broad non-persistent-buffer initialization regression is not the same as GPTNeoXConfig rotary_pct losing its value on reload." + }, + { + "left": "issue:44933", + "right": "issue:45362", + "accept": false, + "reason": "Importing a nonexistent image_utils symbol and Qwen3.5 chat crashing are unrelated." + }, + { + "left": "issue:25251", + "right": "issue:27957", + "accept": false, + "reason": "Pipeline top_k nesting behavior and XLMRoberta Flash Attention 2 support are unrelated issues." + }, + { + "left": "issue:43335", + "right": "issue:44855", + "accept": false, + "reason": "A sparse-layer config bug and a Python 3.13 decorator/comment parsing failure are different problems." + }, + { + "left": "issue:44479", + "right": "issue:44734", + "accept": false, + "reason": "Both are generation-related, but one is a Qwen VL video-input regression and the other is a server KV-cache indexing crash." + }, + { + "left": "issue:45381", + "right": "issue:46369", + "accept": false, + "reason": "Different model families and failures: Qwen2.5-VL video position IDs versus Sam3-LiteText memory overflow." + }, + { + "left": "issue:44077", + "right": "issue:44877", + "accept": false, + "reason": "Both are config-related, but optional post_init in patchtsmixer and strict granite_speech config loading are not the same change." + }, + { + "left": "issue:43550", + "right": "issue:44190", + "accept": false, + "reason": "A torch.compile+SDPA model failure is unrelated to loading a local dataset in a training script." + }, + { + "left": "issue:41720", + "right": "issue:45245", + "accept": false, + "reason": "Qwen3 auto device mapping cudaErrorAssert and the 2^24 categories runtime error are unrelated." + }, + { + "left": "issue:43335", + "right": "issue:44190", + "accept": false, + "reason": "A SwitchTransformers sparse-layer config bug is unrelated to the dataset-loading script failure." + }, + { + "left": "issue:43065", + "right": "issue:44610", + "accept": false, + "reason": "Sam3PixelDecoder dummy Conv2d and OmDet-Turbo processor size mismatch are different model/processor issues." + }, + { + "left": "issue:44464", + "right": "issue:45563", + "accept": false, + "reason": "Compiled-forward chunked-generation inconsistency and a stale warning about num_return_sequences are distinct bugs." + }, + { + "left": "issue:45127", + "right": "issue:45357", + "accept": false, + "reason": "LoRA merge collapse with extended vocab and incorrect visual encoder keys in save_pretrained are separate problems." + }, + { + "left": "issue:39692", + "right": "issue:44190", + "accept": false, + "reason": "SigLIP2 documentation errors and local dataset loading in a training script are unrelated." + }, + { + "left": "issue:45381", + "right": "issue:46655", + "accept": false, + "reason": "Both are video-related, but wrong vision_position_ids and incorrect padding calculation are different processor bugs." + }, + { + "left": "issue:44112", + "right": "issue:45600", + "accept": false, + "reason": "CI failures in GraniteSpeech and Sam3LiteText auto_mappings breakage have different root causes." + }, + { + "left": "issue:43526", + "right": "issue:46697", + "accept": false, + "reason": "One image processor drops labels, the other over-pads images; same subsystem, different bug." + }, + { + "left": "issue:44617", + "right": "issue:46369", + "accept": false, + "reason": "Same broad SAM3 family, but CUDA OOM in Sam3Video and memory overflow in Sam3-LiteText are not the same underlying issue." + }, + { + "left": "issue:44779", + "right": "issue:46489", + "accept": false, + "reason": "Both are DeepSeek tokenizer regressions, but 46489 is a specific DeepSeek-Coder v1 follow-up to an earlier fix gap, not the same report." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 37, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:25251", + "issue:27957", + "issue:29870", + "issue:32090", + "issue:35141", + "issue:36246", + "issue:38617", + "issue:41628", + "issue:41950", + "issue:42175", + "issue:42796", + "issue:43295", + "issue:43404", + "issue:43479", + "issue:43540", + "issue:43643", + "issue:43650", + "issue:43854", + "issue:44016", + "issue:44112", + "issue:44186", + "issue:44220", + "issue:44246", + "issue:44279", + "issue:44462", + "issue:44496", + "issue:44521", + "issue:44734", + "issue:44737", + "issue:44938", + "issue:44964", + "issue:45003", + "issue:45216", + "issue:45245", + "issue:45276", + "issue:45290", + "issue:45375", + "issue:45381", + "issue:45542", + "issue:45584", + "issue:45600", + "issue:45632", + "issue:45803", + "issue:45907", + "issue:46082", + "issue:46121", + "issue:46234", + "issue:46752" + ], + "soft_pairs": [ + "issue:45290|issue:46752", + "issue:43295|issue:45276", + "issue:44737|issue:45803", + "issue:43540|issue:44734", + "issue:44938|issue:45542", + "issue:45276|issue:45584", + "issue:44521|issue:45381", + "issue:25251|issue:29870", + "issue:44462|issue:45907", + "issue:45632|issue:46234", + "issue:44246|issue:45003", + "issue:43854|issue:44964", + "issue:44964|issue:45600", + "issue:43643|issue:44496", + "issue:44112|issue:44964", + "issue:27957|issue:35141", + "issue:46082|issue:46121", + "issue:42175|issue:44279", + "issue:43404|issue:45216", + "issue:45245|issue:45584", + "issue:42796|issue:44521", + "issue:36246|issue:41628", + "issue:44186|issue:44734", + "issue:43650|issue:44220", + "issue:36246|issue:38617", + "issue:32090|issue:35141", + "issue:43479|issue:45375", + "issue:41950|issue:44016" + ], + "prepared_review_unit_hash": "44b0314d34ebac571b68863b70fb068261c56a306d5273c3383aa31192543e3b", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "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 is a loose cluster of mostly unrelated Transformers issues. The only mild theme is chat-template/multimodal edge cases, but the paired reports still describe different failures, so I rejected all soft merges.", + "confidence": 0.39, + "canonical_issue_reason": "issue:44521 is the strongest representative: it is a clearly described, actively discussed multimodal `apply_chat_template` bug with multiple inbound references.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44521 is the best single issue to anchor the set because it has the broadest relevance and clearest reproduction among the nearby chat-template reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45290", + "right": "issue:46752", + "accept": false, + "reason": "Both are `apply_chat_template` crashes, but one is about tool-call assistant messages with no content and the other is an empty-list `IndexError`; different edge cases." + }, + { + "left": "issue:43295", + "right": "issue:45276", + "accept": false, + "reason": "These hit different code paths: custom model code accessing `processor.tokenizer`/image passing vs `resize_token_embeddings` not updating Gemma4 output embeddings." + }, + { + "left": "issue:44737", + "right": "issue:45803", + "accept": false, + "reason": "One is an XLNet device placement bug in `torch.arange`; the other is a bare-`except` swallowing errors in Fuyu tensor conversion." + }, + { + "left": "issue:43540", + "right": "issue:44734", + "accept": false, + "reason": "Both mention video/generation areas, but one is a Qwen3OmniMoe video-input validation error and the other is a server KV-cache indexing bug." + }, + { + "left": "issue:44938", + "right": "issue:45542", + "accept": false, + "reason": "Different runtime/import failures: Python 3.14 load failure vs tensorboard installed without TensorFlow causing a backend error." + }, + { + "left": "issue:45276", + "right": "issue:45584", + "accept": false, + "reason": "One is a Gemma4 embedding-resize regression; the other is Whisper failing on empty transcription after token alignment." + }, + { + "left": "issue:44521", + "right": "issue:45381", + "accept": false, + "reason": "Both involve multimodal chat/video behavior, but one is zeroed assistant masks and the other is wrong `vision_position_ids` for qwen2.5-vl video inputs." + }, + { + "left": "issue:25251", + "right": "issue:29870", + "accept": false, + "reason": "Different bugs in different paths: pipeline `top_k` nesting vs Bart `generate()` using `decoder_attention_mask` before assignment." + }, + { + "left": "issue:44462", + "right": "issue:45907", + "accept": false, + "reason": "One is `AutoTokenizer` ignoring `tokenizer.json`; the other is `list_repo_files` not forwarding `revision`." + }, + { + "left": "issue:45632", + "right": "issue:46234", + "accept": false, + "reason": "A local trust-remote-code cache collision is unrelated to a missing `is_falcon_mamba_ssm_available` import utility." + }, + { + "left": "issue:44246", + "right": "issue:45003", + "accept": false, + "reason": "Import slowness during `import transformers` is not the same as unsafe `sys.modules[]` access in `modeling_utils`." + }, + { + "left": "issue:43854", + "right": "issue:44964", + "accept": false, + "reason": "Distinct model-loading problems: GLM-4.7-Flash test failure vs Phi-4 multimodal load failure on latest transformers." + }, + { + "left": "issue:44964", + "right": "issue:45600", + "accept": false, + "reason": "One is a model load regression; the other is `auto_mappings.py` referencing removed Sam3LiteText configs and breaking CI." + }, + { + "left": "issue:43643", + "right": "issue:44496", + "accept": false, + "reason": "`trust_remote_code=True` dropping fields from `AutoConfig` is different from an unrecognized model missing `model_type`." + }, + { + "left": "issue:44112", + "right": "issue:44964", + "accept": false, + "reason": "A stale CI override test failure is unrelated to a model loading error for Phi-4 multimodal." + }, + { + "left": "issue:27957", + "right": "issue:35141", + "accept": false, + "reason": "XLMRoberta Flash Attention support and token-embedding resize/post_init reinitialization are unrelated issues." + }, + { + "left": "issue:46082", + "right": "issue:46121", + "accept": false, + "reason": "Different validation paths: LlamaConfig head_dim rejection vs rope-parameter conversion failing on a JSON-loaded list." + }, + { + "left": "issue:42175", + "right": "issue:44279", + "accept": false, + "reason": "One is a packaging/backend extra issue (`.[torch]` missing TensorFlow backend support); the other is a generic dependency complaint." + }, + { + "left": "issue:43404", + "right": "issue:45216", + "accept": false, + "reason": "Mistral3 tied-weights behavior and Qwen3.5 checkpoint save corruption are separate model-specific regressions." + }, + { + "left": "issue:45245", + "right": "issue:45584", + "accept": false, + "reason": "Category-cardinality limits and Whisper empty-transcription handling are unrelated failures." + }, + { + "left": "issue:42796", + "right": "issue:44521", + "accept": false, + "reason": "Tokenizer BPE-marker handling is a tokenizer bug, not the multimodal assistant-mask issue in `apply_chat_template`." + }, + { + "left": "issue:36246", + "right": "issue:41628", + "accept": false, + "reason": "Both are import problems around image processors, but one is a specific missing Qwen2.5-VL symbol and the other is a top-level `AutoImageProcessor` import failure." + }, + { + "left": "issue:44186", + "right": "issue:44734", + "accept": false, + "reason": "LayoutLMv2 tokenizer padding/NER crashes are not the same as server KV-cache continuation indexing." + }, + { + "left": "issue:43650", + "right": "issue:44220", + "accept": false, + "reason": "The titles and repros are unrelated; one is non-technical noise, the other is `_torch_extract_fbank_features()`." + }, + { + "left": "issue:36246", + "right": "issue:38617", + "accept": false, + "reason": "Different missing imports from different modules; no shared code path." + }, + { + "left": "issue:32090", + "right": "issue:35141", + "accept": false, + "reason": "Trainer GPU broadcast of `None` is unrelated to embedding resize/post-init reinitialization." + }, + { + "left": "issue:43479", + "right": "issue:45375", + "accept": false, + "reason": "Different config bugs in different models: default vision/audio config initialization vs missing `deepstack_visual_indexes`." + }, + { + "left": "issue:41950", + "right": "issue:44016", + "accept": false, + "reason": "A video-classification pipeline processor lookup bug is unrelated to a notebook syntax error." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 38, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:39692", + "issue:42907", + "issue:43295", + "issue:43299", + "issue:43381", + "issue:43441", + "issue:43526", + "issue:43531", + "issue:43577", + "issue:43673", + "issue:43824", + "issue:43901", + "issue:43976", + "issue:44016", + "issue:44062", + "issue:44220", + "issue:44279", + "issue:44295", + "issue:44327", + "issue:44336", + "issue:44496", + "issue:44521", + "issue:44683", + "issue:44779", + "issue:44792", + "issue:44933", + "issue:44987", + "issue:45042", + "issue:45083", + "issue:45216", + "issue:45335", + "issue:45405", + "issue:45446", + "issue:45584", + "issue:45588", + "issue:45593", + "issue:45600", + "issue:45759", + "issue:46082", + "issue:46123", + "issue:46369", + "issue:46399", + "issue:46424", + "issue:46506", + "issue:46515", + "issue:46693", + "issue:46752" + ], + "soft_pairs": [ + "issue:44295|issue:46506", + "issue:43531|issue:45600", + "issue:43673|issue:46424", + "issue:44016|issue:44220", + "issue:43976|issue:44987", + "issue:44683|issue:45446", + "issue:43295|issue:45335", + "issue:45083|issue:46506", + "issue:43526|issue:43901", + "issue:43381|issue:45593", + "issue:42907|issue:43441", + "issue:43901|issue:44496", + "issue:44496|issue:45042", + "issue:43976|issue:44279", + "issue:43673|issue:44336", + "issue:43441|issue:45588", + "issue:44496|issue:45759", + "issue:44016|issue:44062", + "issue:44792|issue:46424", + "issue:43824|issue:44933", + "issue:43299|issue:45216", + "issue:44327|issue:46752", + "issue:46123|issue:46369", + "issue:44279|issue:46515", + "issue:44521|issue:46693", + "issue:45335|issue:45584", + "issue:46082|issue:46399", + "issue:44279|issue:45405", + "issue:43577|issue:45216", + "issue:39692|issue:43901", + "issue:43295|issue:44779", + "issue:43976|issue:45405" + ], + "prepared_review_unit_hash": "e2ef941e3b21b94cc03e2cd8eb3edd3628c06980a09fa70017f4919f376c9170", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 32, + "serialized_chars": 36645, + "estimated_input_tokens": 9162, + "estimated_eval_tokens": 18580 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is a heterogeneous mix of unrelated Transformers issues across different models, pipelines, and utilities. None of the soft edges look like true duplicates or the same fixable code-path, so I would not merge any of them into a single canonical artifact.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44295", + "right": "issue:46506", + "accept": false, + "reason": "Different bugs: position_ids buffer access vs CPU/CUDA output divergence in PPDocLayoutV3." + }, + { + "left": "issue:43531", + "right": "issue:45600", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior is unrelated to stale auto_mappings references for Sam3LiteText." + }, + { + "left": "issue:43673", + "right": "issue:46424", + "accept": false, + "reason": "Cache API regression during chunked_prefill is distinct from max_cache_length being honored only for Quantized Cache." + }, + { + "left": "issue:44016", + "right": "issue:44220", + "accept": false, + "reason": "Notebook syntax error and _torch_extract_fbank_features are unrelated failures." + }, + { + "left": "issue:43976", + "right": "issue:44987", + "accept": false, + "reason": "Python-version support issue is unrelated to loading physical-intelligence/fast." + }, + { + "left": "issue:44683", + "right": "issue:45446", + "accept": false, + "reason": "Both touch flex_attention, but one is a torch>=2.9 compile failure and the other is a bad version gate for AuxRequest import." + }, + { + "left": "issue:43295", + "right": "issue:45335", + "accept": false, + "reason": "Different model families and code paths: processor/tokenizer regression vs t5gemma embedding resize bug." + }, + { + "left": "issue:45083", + "right": "issue:46506", + "accept": false, + "reason": "Helper-length logic in qwen3_omni_moe is unrelated to PPDocLayoutV3 CPU/CUDA discrepancies." + }, + { + "left": "issue:43526", + "right": "issue:43901", + "accept": false, + "reason": "BeitImageProcessorFast label reduction bug is unrelated to TextClassificationPipeline docs mismatch." + }, + { + "left": "issue:43381", + "right": "issue:45593", + "accept": false, + "reason": "Gradient checkpointing in eval mode and D-FINE auxiliary-loss behavior are different issues." + }, + { + "left": "issue:42907", + "right": "issue:43441", + "accept": false, + "reason": "Dequantized save failure for Ministral/Devstral is not the same as a flash-attention load/runtime bug." + }, + { + "left": "issue:43901", + "right": "issue:44496", + "accept": false, + "reason": "A docs/behavior mismatch in a pipeline is unrelated to an unrecognized model config." + }, + { + "left": "issue:44496", + "right": "issue:45042", + "accept": false, + "reason": "Unrecognized model loading and PIL image processor torchvision requirements are separate problems." + }, + { + "left": "issue:43976", + "right": "issue:44279", + "accept": false, + "reason": "General Python support incompatibility is not the same as a dependency issue with transformers." + }, + { + "left": "issue:43673", + "right": "issue:44336", + "accept": false, + "reason": "Generation cache handling and ANSI output in loading_report are unrelated." + }, + { + "left": "issue:43441", + "right": "issue:45588", + "accept": false, + "reason": "Both involve flash_attention, but they affect different failures/models and would not merge into one fix." + }, + { + "left": "issue:44496", + "right": "issue:45759", + "accept": false, + "reason": "Unrecognized model config and text_config unwrapping for composite multimodal configs are different loading paths." + }, + { + "left": "issue:44016", + "right": "issue:44062", + "accept": false, + "reason": "Notebook syntax error is unrelated to AddedToken special-argument duplication." + }, + { + "left": "issue:44792", + "right": "issue:46424", + "accept": false, + "reason": "Janus image-generation test failure is unrelated to cache max length handling." + }, + { + "left": "issue:43824", + "right": "issue:44933", + "accept": false, + "reason": "Missing Qwen2_5_VLForConditionalGeneration import is unrelated to a nonexistent image_utils import." + }, + { + "left": "issue:43299", + "right": "issue:45216", + "accept": false, + "reason": "Both are Qwen regressions, but one breaks model loading and the other breaks save_pretrained checkpoint saving." + }, + { + "left": "issue:44327", + "right": "issue:46752", + "accept": false, + "reason": "QA decode_spans top-k bounds error is unrelated to apply_chat_template on an empty list." + }, + { + "left": "issue:46123", + "right": "issue:46369", + "accept": false, + "reason": "Final partial batch dropping results and Sam3-LiteText memory overflow are different pipeline issues." + }, + { + "left": "issue:44279", + "right": "issue:46515", + "accept": false, + "reason": "A vague dependency issue is not the same as torch 2.4.1 incompatibility." + }, + { + "left": "issue:44521", + "right": "issue:46693", + "accept": false, + "reason": "Multimodal assistant mask generation and FA2 long-context performance regression are unrelated." + }, + { + "left": "issue:45335", + "right": "issue:45584", + "accept": false, + "reason": "Token embedding resize behavior in t5gemma is unrelated to Whisper empty-transcription generation." + }, + { + "left": "issue:46082", + "right": "issue:46399", + "accept": false, + "reason": "LlamaConfig head_dim validation and Qwen3.5 kernel registration failure are different code paths." + }, + { + "left": "issue:44279", + "right": "issue:45405", + "accept": false, + "reason": "General dependency complaints do not match the specific unreleased MIN_PEFT_VERSION bump." + }, + { + "left": "issue:43577", + "right": "issue:45216", + "accept": false, + "reason": "Blip2 dtype propagation and Qwen3.5 checkpoint saving are unrelated regressions." + }, + { + "left": "issue:39692", + "right": "issue:43901", + "accept": false, + "reason": "SigLIP2 doc-example errors are unrelated to TextClassificationPipeline return_all_scores docs." + }, + { + "left": "issue:43295", + "right": "issue:44779", + "accept": false, + "reason": "Processor/tokenizer regression for multimodal inputs is not the same as Deepseek tokenizer output changes." + }, + { + "left": "issue:43976", + "right": "issue:45405", + "accept": false, + "reason": "Python-version compatibility and PEFT minimum-version bump are unrelated dependency issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 39, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:33290", + "issue:41950", + "issue:42796", + "issue:42994", + "issue:43299", + "issue:43317", + "issue:43441", + "issue:43479", + "issue:43525", + "issue:43650", + "issue:43653", + "issue:43854", + "issue:44016", + "issue:44075", + "issue:44242", + "issue:44261", + "issue:44263", + "issue:44291", + "issue:44351", + "issue:44361", + "issue:44368", + "issue:44462", + "issue:44466", + "issue:44623", + "issue:44683", + "issue:44749", + "issue:44779", + "issue:45020", + "issue:45081", + "issue:45103", + "issue:45200", + "issue:45276", + "issue:45278", + "issue:45357", + "issue:45372", + "issue:45542", + "issue:45593", + "issue:45600", + "issue:45656", + "issue:45712", + "issue:45812", + "issue:45823", + "issue:45834", + "issue:45907", + "issue:46082", + "issue:46209", + "issue:46512" + ], + "soft_pairs": [ + "issue:44075|issue:44368", + "issue:46082|issue:46512", + "issue:44351|issue:45278", + "issue:43299|issue:45357", + "issue:41950|issue:45020", + "issue:42796|issue:43650", + "issue:44779|issue:45812", + "issue:43299|issue:43854", + "issue:44242|issue:45200", + "issue:43479|issue:43525", + "issue:44466|issue:45276", + "issue:41950|issue:45823", + "issue:43653|issue:45593", + "issue:44683|issue:45542", + "issue:43650|issue:46209", + "issue:44016|issue:44291", + "issue:44462|issue:45081", + "issue:43441|issue:45834", + "issue:44361|issue:45593", + "issue:45103|issue:45372", + "issue:42994|issue:43317", + "issue:33290|issue:45656", + "issue:45081|issue:45907", + "issue:45600|issue:45712", + "issue:44261|issue:44263", + "issue:44623|issue:44749" + ], + "prepared_review_unit_hash": "6c3acb213129c96977e50a5ccf8618e7b4e7b0a3cbff280def924f24bbf8e44c", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 35021, + "estimated_input_tokens": 8756, + "estimated_eval_tokens": 17768 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous and does not resolve into a single duplicate family. The soft pairs are mostly topical overlaps (tokenizers, configs, deepspeed, multimodal, serialization) but each points to a different concrete bug or regression, so none should be merged as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "issue:44261 is the cleanest concrete bug report with a focused code path and clear symptom, but it should only serve as a loose representative anchor; the cluster as a whole is not a duplicate set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44075", + "right": "issue:44368", + "accept": false, + "reason": "Different problems: unused SGD args in optimizer handling vs a tie_word_embeddings warning during Qwen fine-tuning." + }, + { + "left": "issue:46082", + "right": "issue:46512", + "accept": false, + "reason": "Both are config-related, but one is about LlamaConfig head_dim validation and the other about dtype being left stale in from_config." + }, + { + "left": "issue:44351", + "right": "issue:45278", + "accept": false, + "reason": "45278 is a broad import-regression report; 44351 is a specific HybridCache import failure. Not the same concrete bug." + }, + { + "left": "issue:43299", + "right": "issue:45357", + "accept": false, + "reason": "Different Qwen multimodal issues: model loading failure vs incorrect save_pretrained visual encoder keys." + }, + { + "left": "issue:41950", + "right": "issue:45020", + "accept": false, + "reason": "Different subsystems and symptoms: pipeline processor lookup vs remote_code breakage in recent transformers versions." + }, + { + "left": "issue:42796", + "right": "issue:43650", + "accept": false, + "reason": "Tokenizer BPE-marker handling is unrelated to the generic 'ADD THE DATA' issue." + }, + { + "left": "issue:44779", + "right": "issue:45812", + "accept": false, + "reason": "Both are tokenizer regressions, but they affect different model families and appear to be separate bugs." + }, + { + "left": "issue:43299", + "right": "issue:43854", + "accept": false, + "reason": "Different multimodal model failures with different loaders and symptoms; not the same underlying defect." + }, + { + "left": "issue:44242", + "right": "issue:45200", + "accept": false, + "reason": "Different model logic: MoE load-balancing loss gating vs mm_token_type_ids defaults for Gemma 4 text-only fine-tuning." + }, + { + "left": "issue:43479", + "right": "issue:43525", + "accept": false, + "reason": "Different config initialization bugs on different classes and fields (vision/audio defaults vs missing pad_token_id)." + }, + { + "left": "issue:44466", + "right": "issue:45276", + "accept": false, + "reason": "Both touch tied/embedding behavior, but one is about serialization depending on device and the other about resize_token_embeddings propagation." + }, + { + "left": "issue:41950", + "right": "issue:45823", + "accept": false, + "reason": "Both mention multimodal models, but one is a processor lookup bug and the other is a device_map forward device mismatch." + }, + { + "left": "issue:43653", + "right": "issue:45593", + "accept": false, + "reason": "Tokenizer special-token registration is unrelated to D-FINE auxiliary-loss behavior." + }, + { + "left": "issue:44683", + "right": "issue:45542", + "accept": false, + "reason": "Compiled flex_attention on torch>=2.9 has no relation to the TensorFlow backend import error." + }, + { + "left": "issue:43650", + "right": "issue:46209", + "accept": false, + "reason": "Completely different issues: a generic 'ADD THE DATA' ticket vs FP8 inference failure on B200." + }, + { + "left": "issue:44016", + "right": "issue:44291", + "accept": false, + "reason": "Notebook syntax error and init_empty_weights TypeError are unrelated failures." + }, + { + "left": "issue:44462", + "right": "issue:45081", + "accept": false, + "reason": "Both involve tokenizers, but one is AutoTokenizer ignoring tokenizer.json and the other is a Mistral regex patch crash." + }, + { + "left": "issue:43441", + "right": "issue:45834", + "accept": false, + "reason": "Different model-specific runtime failures: FlashAttention on Ministral-3 vs long OCR index error in Kosmos2.5." + }, + { + "left": "issue:44361", + "right": "issue:45593", + "accept": false, + "reason": "Tokenizer AttributeError and D-FINE loss logic are unrelated." + }, + { + "left": "issue:45103", + "right": "issue:45372", + "accept": false, + "reason": "Docstring helper crash from future annotations is unrelated to Gemma 4 processor loading/import failure." + }, + { + "left": "issue:42994", + "right": "issue:43317", + "accept": false, + "reason": "Both are quantization-related, but one is saving a quantized model and the other is loading a dequantized model with device_map offload." + }, + { + "left": "issue:33290", + "right": "issue:45656", + "accept": false, + "reason": "Both mention deepspeed/optimizers, but one is OOM with Adafactor and the other is an optimizer.step being called twice." + }, + { + "left": "issue:45081", + "right": "issue:45907", + "accept": false, + "reason": "Tokenizer regex crash and list_repo_files revision forwarding are unrelated code paths." + }, + { + "left": "issue:45600", + "right": "issue:45712", + "accept": false, + "reason": "Both are repo-maintenance/CI-ish, but one is stale auto_mappings references and the other is leftover dummy classes leaking into imports." + }, + { + "left": "issue:44261", + "right": "issue:44263", + "accept": false, + "reason": "Completely different bugs: MLA layernorm epsilon handling vs torch.split return values in a GLM-MoE indexer." + }, + { + "left": "issue:44623", + "right": "issue:44749", + "accept": false, + "reason": "save_pretrained missing files is unrelated to the performance regression after upgrading Transformers." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 40, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:41628", + "issue:42175", + "issue:42491", + "issue:43299", + "issue:43441", + "issue:43525", + "issue:43526", + "issue:43531", + "issue:43643", + "issue:43673", + "issue:43723", + "issue:43761", + "issue:43784", + "issue:43824", + "issue:43881", + "issue:43901", + "issue:43976", + "issue:44016", + "issue:44162", + "issue:44188", + "issue:44242", + "issue:44246", + "issue:44297", + "issue:44336", + "issue:44373", + "issue:44496", + "issue:44589", + "issue:44623", + "issue:44671", + "issue:44683", + "issue:44779", + "issue:45003", + "issue:45103", + "issue:45362", + "issue:45397", + "issue:45542", + "issue:45561", + "issue:45563", + "issue:45593", + "issue:45800", + "issue:45823", + "issue:46291", + "issue:46399", + "issue:46614", + "issue:46655", + "issue:46665", + "issue:46726" + ], + "soft_pairs": [ + "issue:42175|issue:43976", + "issue:44623|issue:45397", + "issue:44671|issue:44779", + "issue:43901|issue:44246", + "issue:43441|issue:44683", + "issue:43673|issue:43761", + "issue:45542|issue:45800", + "issue:43723|issue:44162", + "issue:43824|issue:43976", + "issue:46291|issue:46399", + "issue:43525|issue:43881", + "issue:44016|issue:45823", + "issue:43525|issue:43643", + "issue:44589|issue:45362", + "issue:44188|issue:44336", + "issue:46655|issue:46726", + "issue:44336|issue:44373", + "issue:41628|issue:43976", + "issue:43784|issue:45542", + "issue:44246|issue:44496", + "issue:44297|issue:45397", + "issue:45561|issue:45563", + "issue:46614|issue:46665", + "issue:45103|issue:46399", + "issue:44242|issue:45593", + "issue:43526|issue:45003", + "issue:42491|issue:43299", + "issue:43531|issue:45362" + ], + "prepared_review_unit_hash": "6a64ba4aa55eacc29588d5d7083aeb0415e019c36df03f18f3b75625392e28ed", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 35373, + "estimated_input_tokens": 8844, + "estimated_eval_tokens": 17944 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a loose mix of Transformers v5-era regressions around loading/imports, tokenizers/processors, and model-specific runtime bugs. The suggested soft pairs mostly share broad subsystem language but not the same concrete underlying defect, so none should be merged as duplicates.", + "confidence": 0.27, + "canonical_issue_reason": "Issue 43723 is the broadest user-facing loading regression in the set and best represents the cluster\u2019s repeated v5 import/load failures, even though the cluster is not a tight duplicate set.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43723 is the best single representative issue because it describes a top-level API loading failure that is easy to understand and closest to the dominant theme of the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:42175", + "right": "issue:43976", + "accept": false, + "reason": "One is a backend-install packaging problem; the other is a Python-version compatibility issue. Different failure modes." + }, + { + "left": "issue:44623", + "right": "issue:45397", + "accept": false, + "reason": "Missing files from processor.save_pretrained and a zero3 from_pretrained crash on Gemma-4 are unrelated bugs." + }, + { + "left": "issue:44671", + "right": "issue:44779", + "accept": false, + "reason": "CamemBERT MLM prediction corruption and Deepseek tokenizer corruption affect different code paths and models." + }, + { + "left": "issue:43901", + "right": "issue:44246", + "accept": false, + "reason": "Docs mismatch for return_all_scores is not the same as intermittent slow import of transformers." + }, + { + "left": "issue:43441", + "right": "issue:44683", + "accept": false, + "reason": "Ministral FlashAttention failure and compiled flex_attention failure on torch>=2.9 are different kernel/backend issues." + }, + { + "left": "issue:43673", + "right": "issue:43761", + "accept": false, + "reason": "Chunked_prefill cache regression and CLIPVisionModel hidden_states regression are distinct model-path bugs." + }, + { + "left": "issue:45542", + "right": "issue:45800", + "accept": false, + "reason": "One is a missing TensorFlow backend after an extras install; the other is a version incompatibility between torch and transformers." + }, + { + "left": "issue:43723", + "right": "issue:44162", + "accept": false, + "reason": "Generic tokenizer loading failure and ESM2 model breakage are different concrete problems." + }, + { + "left": "issue:43824", + "right": "issue:43976", + "accept": false, + "reason": "Missing Qwen2.5-VL import symbol and Python-version incompatibility are unrelated." + }, + { + "left": "issue:46291", + "right": "issue:46399", + "accept": false, + "reason": "pipelines import failure from kernels versioning and a missing decorator in Qwen3.5 rotary-pos-emb registration are different bugs." + }, + { + "left": "issue:43525", + "right": "issue:43881", + "accept": false, + "reason": "Llama4Config pad_token_id AttributeError and glm-4v loading failure do not point to the same root cause." + }, + { + "left": "issue:44016", + "right": "issue:45823", + "accept": false, + "reason": "Notebook syntax error and Gemma4 device-map forward mismatch are unrelated." + }, + { + "left": "issue:43525", + "right": "issue:43643", + "accept": false, + "reason": "A missing config attribute in Llama4 and trust_remote_code dropping fields are both config-related, but they are not the same bug." + }, + { + "left": "issue:44589", + "right": "issue:45362", + "accept": false, + "reason": "Float8 storage lookup failure and Qwen3.5 chat crash are unrelated runtime issues." + }, + { + "left": "issue:44188", + "right": "issue:44336", + "accept": false, + "reason": "torch.compile attention-kernel divergence and ANSI codes in loading_report are completely different." + }, + { + "left": "issue:46655", + "right": "issue:46726", + "accept": false, + "reason": "Video padding calculation in SmolVLM and missing vision-backend handling in AutoProcessor are separate processor bugs." + }, + { + "left": "issue:44336", + "right": "issue:44373", + "accept": false, + "reason": "ANSI code emission and wrong docstring content are unrelated documentation/reporting issues." + }, + { + "left": "issue:41628", + "right": "issue:43976", + "accept": false, + "reason": "AutoImageProcessor import error and Python3.9/3.10 compatibility are different import/version issues." + }, + { + "left": "issue:43784", + "right": "issue:45542", + "accept": false, + "reason": "sentence-transformers import NameError and undefined tf backend after extras install do not share a root cause." + }, + { + "left": "issue:44246", + "right": "issue:44496", + "accept": false, + "reason": "Performance on import and unrecognized model/config parsing failure are unrelated." + }, + { + "left": "issue:44297", + "right": "issue:45397", + "accept": false, + "reason": "Tokenizer serialization metadata mismatch and Gemma-4 from_pretrained failure are different code paths." + }, + { + "left": "issue:45561", + "right": "issue:45563", + "accept": false, + "reason": "pytest-xdist file race and stale num_return_sequences warning are unrelated testing/generation issues." + }, + { + "left": "issue:46614", + "right": "issue:46665", + "accept": false, + "reason": "Stale phonemizer backend state and inconsistent from_pretrained behavior inside Python classes are not the same defect." + }, + { + "left": "issue:45103", + "right": "issue:46399", + "accept": false, + "reason": "auto_docstring future-annotations crash and missing decorator registration in Qwen3.5 are unrelated." + }, + { + "left": "issue:44242", + "right": "issue:45593", + "accept": false, + "reason": "Missing load-balancing loss in one MoE model and missing auxiliary losses in D-FINE are model-specific, not one bug." + }, + { + "left": "issue:43526", + "right": "issue:45003", + "accept": false, + "reason": "BeitImageProcessorFast label reduction bug and unsafe sys.modules access are unrelated." + }, + { + "left": "issue:42491", + "right": "issue:43299", + "accept": false, + "reason": "Both involve Qwen3 MoE, but one is a LoRA compatibility issue and the other is a direct model-loading regression; not the same concrete bug." + }, + { + "left": "issue:43531", + "right": "issue:45362", + "accept": false, + "reason": "Sliding-window behavior in Qwen3-MoE and Qwen3.5 chat crashes are different symptoms and code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 41, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:36246", + "issue:41950", + "issue:42757", + "issue:42886", + "issue:43257", + "issue:43262", + "issue:43295", + "issue:43475", + "issue:43540", + "issue:43550", + "issue:43596", + "issue:43673", + "issue:43901", + "issue:44162", + "issue:44220", + "issue:44242", + "issue:44246", + "issue:44295", + "issue:44336", + "issue:44351", + "issue:44451", + "issue:44479", + "issue:44485", + "issue:44496", + "issue:44561", + "issue:44734", + "issue:44749", + "issue:44779", + "issue:44857", + "issue:44871", + "issue:44877", + "issue:45003", + "issue:45137", + "issue:45200", + "issue:45292", + "issue:45375", + "issue:45397", + "issue:45561", + "issue:45600", + "issue:45759", + "issue:45859", + "issue:46132", + "issue:46369", + "issue:46399", + "issue:46424", + "issue:46506", + "issue:46655", + "issue:46752" + ], + "soft_pairs": [ + "issue:43540|issue:46752", + "issue:44246|issue:44871", + "issue:44295|issue:45397", + "issue:44336|issue:44485", + "issue:44451|issue:46752", + "issue:44336|issue:46424", + "issue:42886|issue:46132", + "issue:43550|issue:46506", + "issue:41950|issue:44220", + "issue:44779|issue:45292", + "issue:36246|issue:42757", + "issue:36246|issue:44351", + "issue:43901|issue:45003", + "issue:43257|issue:45137", + "issue:45003|issue:45561", + "issue:44734|issue:46424", + "issue:44496|issue:45003", + "issue:43540|issue:46655", + "issue:45859|issue:46399", + "issue:43673|issue:44561", + "issue:44162|issue:45200", + "issue:44479|issue:46655", + "issue:43475|issue:46369", + "issue:44242|issue:44857", + "issue:45375|issue:45759", + "issue:43262|issue:44246", + "issue:44295|issue:44749", + "issue:45137|issue:45600", + "issue:36246|issue:43596", + "issue:44877|issue:45375", + "issue:44561|issue:46399", + "issue:43295|issue:45292" + ], + "prepared_review_unit_hash": "aa146d244bb03480c92de4f923b112c16e5860d497780618eb96bd14e2fcabe1", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 37244, + "estimated_input_tokens": 9311, + "estimated_eval_tokens": 18878 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: most items are standalone Transformers bug reports across unrelated areas (imports, configs, deepspeed, multimodal/video, caching, docs). I did not find any soft-edge pair that clearly describes the same underlying issue, so all candidate merges should be rejected.", + "confidence": 0.27, + "canonical_issue_reason": "Issue 44479 is the broadest representative of the strongest visible theme in this set: a versioned multimodal/video regression affecting multiple Qwen model families. It is more cluster-like than the many narrow one-off failures, though the overall cluster remains quite mixed.", + "canonical_pr_reason": null, + "best_issue_reason": "44479 is the best single issue to anchor the cluster because it names a wide regression, spans several related model variants, and is more general than the other candidates.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43540", + "right": "issue:46752", + "accept": false, + "reason": "Both mention input processing, but one is a Qwen3OmniMoe video-input ValueError and the other is an empty-list apply_chat_template crash; different code paths and failure modes." + }, + { + "left": "issue:44246", + "right": "issue:44871", + "accept": false, + "reason": "Import-time slowness and Gemma eos_token_id inconsistency are unrelated bugs." + }, + { + "left": "issue:44295", + "right": "issue:45397", + "accept": false, + "reason": "A position_ids buffer access error and a Gemma-4 ZeRO3 from_pretrained bug are different issues." + }, + { + "left": "issue:44336", + "right": "issue:44485", + "accept": false, + "reason": "ANSI output in loading_report and GLM-5 RoPE implementation are unrelated." + }, + { + "left": "issue:44451", + "right": "issue:46752", + "accept": false, + "reason": "ScandiBERT loading failure and empty-list chat-template handling are different failures." + }, + { + "left": "issue:44336", + "right": "issue:46424", + "accept": false, + "reason": "Terminal ANSI formatting and cache max length handling do not share the same underlying bug." + }, + { + "left": "issue:42886", + "right": "issue:46132", + "accept": false, + "reason": "Offline tokenizer cache loading and AttentionInterface.register behavior are unrelated subsystems." + }, + { + "left": "issue:43550", + "right": "issue:46506", + "accept": false, + "reason": "torch.compile with SDPA on Bamba and CPU/CUDA divergence in PPDocLayoutV3 are distinct model/runtime problems." + }, + { + "left": "issue:41950", + "right": "issue:44220", + "accept": false, + "reason": "Video-classification image-processor lookup and audio fbank extraction are different pipeline bugs." + }, + { + "left": "issue:44779", + "right": "issue:45292", + "accept": false, + "reason": "DeepSeek tokenizer correctness and resize_token_embeddings not updating output embeddings are different code paths." + }, + { + "left": "issue:36246", + "right": "issue:42757", + "accept": false, + "reason": "Qwen2_5_VLImageProcessor import failure and missing is_offline_mode import are separate missing-symbol regressions." + }, + { + "left": "issue:36246", + "right": "issue:44351", + "accept": false, + "reason": "Both are import errors, but for different exported symbols/classes and different root causes." + }, + { + "left": "issue:43901", + "right": "issue:45003", + "accept": false, + "reason": "Pipeline docs mismatch and unsafe sys.modules access are unrelated." + }, + { + "left": "issue:43257", + "right": "issue:45137", + "accept": false, + "reason": "Both involve DeepSpeed, but one is Qwen3 MOE weight conversion and the other is a ZeRO3 deque/index error; not the same bug." + }, + { + "left": "issue:45003", + "right": "issue:45561", + "accept": false, + "reason": "Unsafe sys.modules access and xdist captured_info.txt race are different concurrency/robustness issues." + }, + { + "left": "issue:44734", + "right": "issue:46424", + "accept": false, + "reason": "KV-cache continuation indexing and cache max_length configuration concern different cache logic." + }, + { + "left": "issue:44496", + "right": "issue:45003", + "accept": false, + "reason": "Unrecognized model config loading and sys.modules access are unrelated." + }, + { + "left": "issue:43540", + "right": "issue:46655", + "accept": false, + "reason": "Both are video-related, but Qwen3OmniMoe video-input handling and SmolVLM padding calculation are not the same defect." + }, + { + "left": "issue:45859", + "right": "issue:46399", + "accept": false, + "reason": "CPU-offload incompatibility in rotary embedding and a missing registration decorator for kernels are different Qwen3.5 issues." + }, + { + "left": "issue:43673", + "right": "issue:44561", + "accept": false, + "reason": "Chunked-prefill cache regression and removal of is_torch_fx_available break different behaviors." + }, + { + "left": "issue:44162", + "right": "issue:45200", + "accept": false, + "reason": "ESM2 breakage and Gemma 4 mm_token_type_ids defaulting are unrelated model-specific bugs." + }, + { + "left": "issue:44479", + "right": "issue:46655", + "accept": false, + "reason": "Both are video-input regressions, but they affect different model families and point to different preprocessing bugs." + }, + { + "left": "issue:43475", + "right": "issue:46369", + "accept": false, + "reason": "SAM 3 video encoder output attribute mismatch and SAM3-litetext memory overflow are different issues." + }, + { + "left": "issue:44242", + "right": "issue:44857", + "accept": false, + "reason": "MoE load-balancing loss omission and LwDetrImageLoss AMP crash are unrelated." + }, + { + "left": "issue:45375", + "right": "issue:45759", + "accept": false, + "reason": "Both touch Qwen multimodal configs, but one is strict-field dropping and the other is from_config not unwrapping text_config; different fixes." + }, + { + "left": "issue:43262", + "right": "issue:44246", + "accept": false, + "reason": "Audio chat-template sampling-rate default and import slowness are unrelated." + }, + { + "left": "issue:44295", + "right": "issue:44749", + "accept": false, + "reason": "A buffer access IndexError and a broad slowdown regression are not the same bug." + }, + { + "left": "issue:45137", + "right": "issue:45600", + "accept": false, + "reason": "DeepSpeed ZeRO3 deque failure and Sam3LiteText auto_mappings CI breakage are unrelated." + }, + { + "left": "issue:36246", + "right": "issue:43596", + "accept": false, + "reason": "Qwen2_5_VLImageProcessor import error and ZeRO3/BertModel empty-dimension IndexError are different failures." + }, + { + "left": "issue:44877", + "right": "issue:45375", + "accept": false, + "reason": "A generic strict-config loading failure and a missing Qwen3_5MoeVisionConfig field are related only at a very high level, not the same underlying bug." + }, + { + "left": "issue:44561", + "right": "issue:46399", + "accept": false, + "reason": "Removal of is_torch_fx_available and a missing decorator for apply_rotary_pos_emb are unrelated breakages." + }, + { + "left": "issue:43295", + "right": "issue:45292", + "accept": false, + "reason": "Processor.tokenizer/image-passing regression and resize_token_embeddings output-embedding mismatch are different issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 42, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:29870", + "issue:30333", + "issue:42886", + "issue:43065", + "issue:43352", + "issue:43408", + "issue:43475", + "issue:43540", + "issue:43653", + "issue:43723", + "issue:43906", + "issue:43931", + "issue:44038", + "issue:44060", + "issue:44077", + "issue:44162", + "issue:44188", + "issue:44368", + "issue:44466", + "issue:44479", + "issue:44483", + "issue:44623", + "issue:44877", + "issue:44977", + "issue:44987", + "issue:44995", + "issue:45200", + "issue:45229", + "issue:45250", + "issue:45292", + "issue:45335", + "issue:45397", + "issue:45412", + "issue:45459", + "issue:45563", + "issue:45584", + "issue:45588", + "issue:45656", + "issue:45987", + "issue:46082", + "issue:46123", + "issue:46170", + "issue:46202", + "issue:46421", + "issue:46506", + "issue:46710", + "issue:46726" + ], + "soft_pairs": [ + "issue:45292|issue:45584", + "issue:44466|issue:45335", + "issue:44623|issue:46726", + "issue:44038|issue:46506", + "issue:44987|issue:45459", + "issue:29870|issue:30333", + "issue:44479|issue:46170", + "issue:43352|issue:45250", + "issue:44060|issue:44077", + "issue:45412|issue:46123", + "issue:43653|issue:44162", + "issue:44368|issue:44977", + "issue:44995|issue:45200", + "issue:44877|issue:46170", + "issue:43475|issue:43540", + "issue:42886|issue:45229", + "issue:43065|issue:43408", + "issue:45397|issue:45656", + "issue:43931|issue:44368", + "issue:44368|issue:45563", + "issue:44995|issue:45588", + "issue:43723|issue:46710", + "issue:45987|issue:46082", + "issue:44188|issue:46202", + "issue:43906|issue:46202", + "issue:44483|issue:46202", + "issue:44188|issue:46421" + ], + "prepared_review_unit_hash": "4bf917949490516578e0f6b691a22fa99e6a9e578d6e9cd96c68032f0895a648", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 27, + "serialized_chars": 35198, + "estimated_input_tokens": 8800, + "estimated_eval_tokens": 17856 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All suggested soft pairs look like false positives: they span unrelated failures across tokenization, model loading, generation, cache behavior, DeepSpeed, vision/video models, and config handling. No single issue stands out as a credible duplicate hub for the cluster.", + "confidence": 0.95, + "canonical_issue_reason": "No coherent canonical issue emerges; the items are too heterogeneous and the soft pairs do not share the same underlying bug or code path.", + "canonical_pr_reason": null, + "best_issue_reason": "No issue is a strong global representative here because the cluster is dominated by unrelated or only loosely similar bugs.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45292", + "right": "issue:45584", + "accept": false, + "reason": "Resize-token-embeddings/output-embeddings bug vs empty-transcription Whisper failure; unrelated code paths." + }, + { + "left": "issue:44466", + "right": "issue:45335", + "accept": false, + "reason": "Both mention embeddings, but one is tied-weight serialization by device and the other is decoder.embed_tokens not updating on resize; different bugs." + }, + { + "left": "issue:44623", + "right": "issue:46726", + "accept": false, + "reason": "save_pretrained missing files vs load-time AutoProcessor error when a vision backend is missing; different lifecycle stages." + }, + { + "left": "issue:44038", + "right": "issue:46506", + "accept": false, + "reason": "Qwen3-VL-Moe loading bug vs CPU/CUDA output divergence in PPDocLayoutV3; unrelated models and symptoms." + }, + { + "left": "issue:44987", + "right": "issue:45459", + "accept": false, + "reason": "Model loading regression vs tokenizer error masking when protobuf is absent; different failure modes." + }, + { + "left": "issue:29870", + "right": "issue:30333", + "accept": false, + "reason": "Generation UnboundLocalError vs MLflow reporting not updated after failed training; unrelated." + }, + { + "left": "issue:44479", + "right": "issue:46170", + "accept": false, + "reason": "Video-input regression in Qwen VL models vs Windows ROCm import/dependency failure; not the same bug." + }, + { + "left": "issue:43352", + "right": "issue:45250", + "accept": false, + "reason": "Specific model lacking Flash Attention 2 support vs a generic flash-attention topic; too broad to be the same concrete issue." + }, + { + "left": "issue:44060", + "right": "issue:44077", + "accept": false, + "reason": "Qwen3-Next tied-weights warning vs patchtsmixer post_init policy; unrelated subsystems." + }, + { + "left": "issue:45412", + "right": "issue:46123", + "accept": false, + "reason": "Memory leak on delete vs partial-batch result dropping in a pipeline; different problems." + }, + { + "left": "issue:43653", + "right": "issue:44162", + "accept": false, + "reason": "BigBirdTokenizer special-token registration bug vs ESM2 breakage; unrelated." + }, + { + "left": "issue:44368", + "right": "issue:44977", + "accept": false, + "reason": "Tied-embeddings warning in Qwen3.5 fine-tuning vs flash-attention generation regression; different bugs." + }, + { + "left": "issue:44995", + "right": "issue:45200", + "accept": false, + "reason": "Stale indexer cache crash vs Gemma 4 token-type-id defaulting; unrelated." + }, + { + "left": "issue:44877", + "right": "issue:46170", + "accept": false, + "reason": "Strict config loading failure vs torch.distributed import issue on Windows/ROCm; unrelated." + }, + { + "left": "issue:43475", + "right": "issue:43540", + "accept": false, + "reason": "SAM3 vision-encoder attribute error vs Qwen3OmniMoe video ValueError; different models and code paths." + }, + { + "left": "issue:42886", + "right": "issue:45229", + "accept": false, + "reason": "Offline tokenizer cache load failure vs Gemma4 multi-GPU OOM; unrelated." + }, + { + "left": "issue:43065", + "right": "issue:43408", + "accept": false, + "reason": "Dummy Conv2d implementation vs model-type mismatch warning; related family, but not the same bug." + }, + { + "left": "issue:45397", + "right": "issue:45656", + "accept": false, + "reason": "DeepSpeed zero3 from_pretrained issue vs optimizer stepping twice; not the same concrete failure." + }, + { + "left": "issue:43931", + "right": "issue:44368", + "accept": false, + "reason": "Model weight-shape mismatch on load vs tied-weights warning; different symptoms and causes." + }, + { + "left": "issue:44368", + "right": "issue:45563", + "accept": false, + "reason": "Tied-weights warning vs stale num_return_sequences warning in paged generate; unrelated." + }, + { + "left": "issue:44995", + "right": "issue:45588", + "accept": false, + "reason": "Stale cache crash vs flash_attention AttributeError on s_aux=None; different code paths." + }, + { + "left": "issue:43723", + "right": "issue:46710", + "accept": false, + "reason": "Tokenizer loading issue in v5 vs accuracy regression on a specific model after upgrade; not duplicates." + }, + { + "left": "issue:45987", + "right": "issue:46082", + "accept": false, + "reason": "Cache length return-type contract vs LlamaConfig head_dim validation; unrelated." + }, + { + "left": "issue:44188", + "right": "issue:46202", + "accept": false, + "reason": "torch.compile attention-kernel divergence vs Gemma pipeline roundtrip failure; different bugs." + }, + { + "left": "issue:43906", + "right": "issue:46202", + "accept": false, + "reason": "An isolated reproduction of another issue is not enough to show the same underlying bug as the Gemma roundtrip failure." + }, + { + "left": "issue:44483", + "right": "issue:46202", + "accept": false, + "reason": "Chat-completions request rejection vs Gemma pipeline roundtrip issue; unrelated." + }, + { + "left": "issue:44188", + "right": "issue:46421", + "accept": false, + "reason": "Both touch cache/compile behavior, but one is attention-kernel divergence and the other is lazy cache init causing recompilation; different concrete 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": 23, + "review_unit_index": 43, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:28282", + "issue:30990", + "issue:39692", + "issue:41628", + "issue:42371", + "issue:42994", + "issue:43065", + "issue:43232", + "issue:43441", + "issue:43475", + "issue:43482", + "issue:43525", + "issue:43643", + "issue:43646", + "issue:43673", + "issue:43701", + "issue:43723", + "issue:43986", + "issue:43994", + "issue:44038", + "issue:44162", + "issue:44297", + "issue:44361", + "issue:44623", + "issue:44749", + "issue:44869", + "issue:44871", + "issue:44933", + "issue:44964", + "issue:45092", + "issue:45137", + "issue:45230", + "issue:45310", + "issue:45356", + "issue:45375", + "issue:45381", + "issue:45397", + "issue:45478", + "issue:45584", + "issue:45656", + "issue:45685", + "issue:46123", + "issue:46132", + "issue:46170", + "issue:46326", + "issue:46431", + "issue:46726", + "issue:46752" + ], + "soft_pairs": [ + "issue:45685|issue:46431", + "issue:45584|issue:46170", + "issue:43065|issue:46123", + "issue:44361|issue:44749", + "issue:44297|issue:44623", + "issue:43994|issue:46726", + "issue:43646|issue:44162", + "issue:44749|issue:45656", + "issue:44871|issue:46132", + "issue:43701|issue:44749", + "issue:42994|issue:43482", + "issue:44623|issue:45656", + "issue:43441|issue:46752", + "issue:43232|issue:43673", + "issue:43475|issue:43643", + "issue:45375|issue:45478", + "issue:28282|issue:30990", + "issue:41628|issue:44933", + "issue:44964|issue:45137", + "issue:43986|issue:46726", + "issue:45310|issue:45375", + "issue:44869|issue:45356", + "issue:42994|issue:43525", + "issue:44038|issue:44623", + "issue:45230|issue:45397", + "issue:39692|issue:46726", + "issue:42371|issue:45092", + "issue:43475|issue:45381", + "issue:43723|issue:46326" + ], + "prepared_review_unit_hash": "fff10ed3d4884b951d60a06382fa2e89879a2900f362bde0bb903710484d8508", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36157, + "estimated_input_tokens": 9040, + "estimated_eval_tokens": 18336 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab-bag of Transformers v5 regression reports, with only a couple of plausible duplicate pairs around Qwen3.5-Moe config loading and AutoProcessor backend-missing handling. Most other pairs are different bug paths and should stay separate.", + "confidence": 0.74, + "canonical_issue_reason": "issue:43646 is the broadest, most central regression report here: it describes Transformers 5.0 breaking custom model initialization, which best matches the cluster\u2019s overall v5-loading/regression theme.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43646 is the best single issue to anchor this cluster because it is the most umbrella-like report and can represent the shared v5 initialization/loading regression theme better than the narrower model-specific bugs.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45685", + "right": "issue:46431", + "accept": false, + "reason": "Different problems: an MPS histogram backend error in MoE code vs an empty-list apply_chat_template crash." + }, + { + "left": "issue:45584", + "right": "issue:46170", + "accept": false, + "reason": "Unrelated: Whisper empty-transcription generation failure vs Windows/ROCm torch.distributed import dependency issue." + }, + { + "left": "issue:43065", + "right": "issue:46123", + "accept": false, + "reason": "Both involve segmentation-related code, but one is a dummy Conv2d in Sam3PixelDecoder and the other is a MaskGenerationPipeline batching bug; different code paths and fixes." + }, + { + "left": "issue:44361", + "right": "issue:44749", + "accept": false, + "reason": "Unrelated: MLukeTokenizer task attribute error vs a broad performance regression after upgrading Transformers." + }, + { + "left": "issue:44297", + "right": "issue:44623", + "accept": false, + "reason": "Both are save_pretrained-related, but one is tokenizer metadata mismatch and the other is missing processor files; not the same bug." + }, + { + "left": "issue:43994", + "right": "issue:46726", + "accept": false, + "reason": "Different symptoms and targets: SigLIP2 bad outputs with AutoModel/pipeline vs AutoProcessor's missing-image-processor error when vision backend is absent." + }, + { + "left": "issue:43646", + "right": "issue:44162", + "accept": false, + "reason": "Different subsystems: custom model initialization regression vs ESM2 breakage." + }, + { + "left": "issue:44749", + "right": "issue:45656", + "accept": false, + "reason": "Performance slowdown after upgrade vs optimizer step being called twice under deepspeed; not the same failure mode." + }, + { + "left": "issue:44871", + "right": "issue:46132", + "accept": false, + "reason": "Gemma eos_token_id configuration inconsistency vs AttentionInterface.register behavior change; unrelated." + }, + { + "left": "issue:43701", + "right": "issue:44749", + "accept": false, + "reason": "Checkpoint key mismatch vs filtering performance regression; different bugs." + }, + { + "left": "issue:42994", + "right": "issue:43482", + "accept": false, + "reason": "Quantized model saving failure vs Qwen2.5-GGUF loading failure; save/load and model-specific issues differ." + }, + { + "left": "issue:44623", + "right": "issue:45656", + "accept": false, + "reason": "Processor.save_pretrained missing files vs optimizer double-step in deepspeed; unrelated." + }, + { + "left": "issue:43441", + "right": "issue:46752", + "accept": false, + "reason": "FlashAttention failure for Ministral-3 vs apply_chat_template empty-list IndexError; different code paths." + }, + { + "left": "issue:43232", + "right": "issue:43673", + "accept": false, + "reason": "Both are generation-related, but one is sync_gpus/_update_model_kwargs_for_generation and the other is chunked_prefill cache missing; too different to merge." + }, + { + "left": "issue:43475", + "right": "issue:43643", + "accept": false, + "reason": "Sam3VisionEncoderOutput missing attribute vs AutoConfig trust_remote_code missing fields; different model families and failure points." + }, + { + "left": "issue:45375", + "right": "issue:45478", + "accept": true, + "reason": "Same underlying Qwen3.5-Moe from_pretrained/config regression: one report identifies the missing strict-dropped field, the other reports the resulting load failure on the same model family." + }, + { + "left": "issue:28282", + "right": "issue:30990", + "accept": false, + "reason": "ImportError for missing PyTorch vs Sentence Transformers loading stuck; not the same issue." + }, + { + "left": "issue:41628", + "right": "issue:44933", + "accept": false, + "reason": "Both are import-related, but they concern different missing symbols/modules and different API surfaces; not clearly the same bug." + }, + { + "left": "issue:44964", + "right": "issue:45137", + "accept": false, + "reason": "Phi-4 multimodal model loading failure vs DeepSpeed ZeRO3 deque IndexError; unrelated." + }, + { + "left": "issue:43986", + "right": "issue:46726", + "accept": true, + "reason": "Both are AutoProcessor failures caused by missing vision-related optional dependencies, with the same underlying class of backend-detection/error-reporting bug." + }, + { + "left": "issue:39692", + "right": "issue:46726", + "accept": false, + "reason": "SigLIP2 documentation/model/processor mismatch plus quantization issue vs generic missing-vision-backend processor error; different bugs." + }, + { + "left": "issue:42371", + "right": "issue:45092", + "accept": false, + "reason": "TF32 API settings warning vs old InternVL2 remote-code/meta-init incompatibility; unrelated." + }, + { + "left": "issue:43475", + "right": "issue:45381", + "accept": false, + "reason": "Different SAM3 vs Qwen2.5-VL issues with no shared concrete failure path." + }, + { + "left": "issue:43723", + "right": "issue:46326", + "accept": false, + "reason": "Tokenizer loading regression vs mistral_common continue_final_message bug; different areas." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet and stays conservative about the soft edges. The two accepted pairs are plausibly mergeable into single PRs for the same bugfix class (Qwen3.5-Moe config/load regression and AutoProcessor missing-vision-backend handling), and the rejected pairs are appropriately treated as separate bugs rather than overstated duplicates." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 44, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:30333", + "issue:30990", + "issue:33290", + "issue:35141", + "issue:36010", + "issue:41628", + "issue:42175", + "issue:42673", + "issue:42831", + "issue:42898", + "issue:42994", + "issue:43232", + "issue:43278", + "issue:43377", + "issue:43479", + "issue:43504", + "issue:43704", + "issue:43784", + "issue:43824", + "issue:43873", + "issue:43976", + "issue:44038", + "issue:44188", + "issue:44222", + "issue:44242", + "issue:44351", + "issue:44496", + "issue:44568", + "issue:44623", + "issue:44792", + "issue:44869", + "issue:44933", + "issue:44945", + "issue:45081", + "issue:45103", + "issue:45137", + "issue:45310", + "issue:45406", + "issue:45561", + "issue:45600", + "issue:45712", + "issue:45753", + "issue:46077", + "issue:46123", + "issue:46202", + "issue:46399", + "issue:46693", + "issue:46701" + ], + "soft_pairs": [ + "issue:43784|issue:46399", + "issue:43232|issue:44188", + "issue:43873|issue:44945", + "issue:45406|issue:46202", + "issue:44945|issue:45753", + "issue:41628|issue:44351", + "issue:46202|issue:46701", + "issue:42994|issue:43824", + "issue:42673|issue:43704", + "issue:45081|issue:45103", + "issue:43278|issue:45137", + "issue:44038|issue:45310", + "issue:36010|issue:41628", + "issue:44222|issue:45310", + "issue:42831|issue:44222", + "issue:43232|issue:44792", + "issue:42175|issue:44933", + "issue:43504|issue:45561", + "issue:44496|issue:45561", + "issue:30333|issue:33290", + "issue:45561|issue:45600", + "issue:43976|issue:44933", + "issue:30990|issue:35141", + "issue:42175|issue:45712", + "issue:42898|issue:44568", + "issue:43377|issue:44242", + "issue:46077|issue:46123", + "issue:44623|issue:45310", + "issue:43479|issue:44869", + "issue:42673|issue:46693" + ], + "prepared_review_unit_hash": "6c265b5263425e093d09e003061d69334649e67339d39f039c8df2c40eadb82a", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "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 cluster is mostly a mix of unrelated Transformers issues; only one soft pair is a clear duplicate. The exact duplicate is the Qwen3ForCausalLM VRAM leak report, where the earlier issue is the better anchor.", + "confidence": 0.92, + "canonical_issue_reason": "issue:42673 is the best canonical issue because it is the earlier, fuller report for the exact duplicate pair and has more discussion/inbound references than issue:43704.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:42673 is the strongest representative in the set: it is an exact duplicate of issue:43704 and has the richest signal for the underlying bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43784", + "right": "issue:46399", + "accept": false, + "reason": "Both are import/runtime failures, but they concern different symbols and code paths (sentence-transformers import vs Qwen3.5 rotary registration)." + }, + { + "left": "issue:43232", + "right": "issue:44188", + "accept": false, + "reason": "Different generation-related bugs: one is sync_gpus/_update_model_kwargs_for_generation, the other is attention kernel divergence under torch.compile." + }, + { + "left": "issue:43873", + "right": "issue:44945", + "accept": false, + "reason": "One is a quantization/offloading problem, the other is incorrect output under pipeline parallelism; not the same underlying defect." + }, + { + "left": "issue:45406", + "right": "issue:46202", + "accept": false, + "reason": "Both involve Gemma, but one is a processor attribute crash and the other is a token roundtrip failure; different bugs." + }, + { + "left": "issue:44945", + "right": "issue:45753", + "accept": false, + "reason": "Different models and symptoms: pipeline parallelism output corruption vs a Qwen3_5 image-specific infinite loop." + }, + { + "left": "issue:41628", + "right": "issue:44351", + "accept": false, + "reason": "Both are import errors, but for different exported names and likely different missing-export causes." + }, + { + "left": "issue:46202", + "right": "issue:46701", + "accept": false, + "reason": "Gemma tokenizer roundtrip failure is not the same as the DiffusionGemma image-text-to-text pipeline failure." + }, + { + "left": "issue:42994", + "right": "issue:43824", + "accept": false, + "reason": "Unrelated failures: quantized model saving vs missing Qwen2_5_VLForConditionalGeneration import." + }, + { + "left": "issue:42673", + "right": "issue:43704", + "accept": true, + "reason": "Exact same bug and identical title: Qwen3ForCausalLM leaks VRAM in multiple dataloader threads." + }, + { + "left": "issue:45081", + "right": "issue:45103", + "accept": false, + "reason": "Different crash sites and behaviors: tokenizer regex patching vs auto-docstring kwargs processing." + }, + { + "left": "issue:43278", + "right": "issue:45137", + "accept": false, + "reason": "Embedding dtype drift and a DeepSpeed ZeRO3 deque IndexError are unrelated." + }, + { + "left": "issue:44038", + "right": "issue:45310", + "accept": false, + "reason": "Both mention Qwen3.x model variants, but the reported failures and version contexts differ; not the same concrete bug." + }, + { + "left": "issue:36010", + "right": "issue:41628", + "accept": false, + "reason": "Both are import errors from transformers, but they are distinct missing symbols and distinct fixes." + }, + { + "left": "issue:44222", + "right": "issue:45310", + "accept": false, + "reason": "FP8 save_pretrained behavior is different from a Qwen3.5 from_pretrained import/load failure." + }, + { + "left": "issue:42831", + "right": "issue:44222", + "accept": false, + "reason": "FineGrainedFP8 accuracy issues and FP8 save_pretrained failures are different problem classes." + }, + { + "left": "issue:43232", + "right": "issue:44792", + "accept": false, + "reason": "Generation kwargs handling and a janus image generation test failure are not the same bug." + }, + { + "left": "issue:42175", + "right": "issue:44933", + "accept": false, + "reason": "Packaging/backend dependency resolution and an image_utils import failure are unrelated." + }, + { + "left": "issue:43504", + "right": "issue:45561", + "accept": false, + "reason": "Pretrained model config loading and a pytest-xdist file race are completely different issues." + }, + { + "left": "issue:44496", + "right": "issue:45561", + "accept": false, + "reason": "Model config/model_type resolution and a test utility race condition are unrelated." + }, + { + "left": "issue:30333", + "right": "issue:33290", + "accept": false, + "reason": "MLFlow reporting after failed training and a deepspeed OOM are different failure modes." + }, + { + "left": "issue:45561", + "right": "issue:45600", + "accept": false, + "reason": "A parallel test file race and stale auto_mappings references are not the same bug." + }, + { + "left": "issue:43976", + "right": "issue:44933", + "accept": false, + "reason": "Python version compatibility and a missing image_utils import are unrelated." + }, + { + "left": "issue:30990", + "right": "issue:35141", + "accept": false, + "reason": "Stuck loading vs token embedding reinitialization are different issues." + }, + { + "left": "issue:42175", + "right": "issue:45712", + "accept": false, + "reason": "Both touch packaging/dummy backend behavior, but one is missing tensorflow in extras and the other is leftover dummy classes leaking into dir()." + }, + { + "left": "issue:42898", + "right": "issue:44568", + "accept": false, + "reason": "Both concern tokenizer behavior changes, but they affect different APIs and symptoms." + }, + { + "left": "issue:43377", + "right": "issue:44242", + "accept": false, + "reason": "Different MoE-related bugs: padding-mask batching mismatch vs missing load-balancing loss when router logits are disabled." + }, + { + "left": "issue:46077", + "right": "issue:46123", + "accept": false, + "reason": "Bidirectional attention mask construction and partial-batch pipeline bookkeeping are unrelated." + }, + { + "left": "issue:44623", + "right": "issue:45310", + "accept": false, + "reason": "Missing processor save files and Qwen3.5 from_pretrained errors are different code paths." + }, + { + "left": "issue:43479", + "right": "issue:44869", + "accept": false, + "reason": "Config default initialization and Whisper decode crashing on a trailing replacement character are unrelated." + }, + { + "left": "issue:42673", + "right": "issue:46693", + "accept": false, + "reason": "VRAM leak in dataloader threads and a long-context FA2 performance regression are different performance problems." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The cluster summary is well grounded in the packet: 42673 and 43704 have identical titles and are a clear duplicate pair, and the canonical choice is conservative and supported by earlier date plus richer discussion/body. The soft-edge rejections are also appropriately cautious and do not overclaim equivalence from title-level similarity alone." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 45, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:22355", + "issue:27957", + "issue:28282", + "issue:30333", + "issue:34689", + "issue:35141", + "issue:42222", + "issue:42796", + "issue:42831", + "issue:42898", + "issue:43116", + "issue:43295", + "issue:43479", + "issue:43644", + "issue:43723", + "issue:43784", + "issue:43824", + "issue:43874", + "issue:44016", + "issue:44060", + "issue:44075", + "issue:44079", + "issue:44190", + "issue:44222", + "issue:44295", + "issue:44297", + "issue:44360", + "issue:44448", + "issue:44568", + "issue:44623", + "issue:44734", + "issue:44779", + "issue:44933", + "issue:45092", + "issue:45103", + "issue:45125", + "issue:45200", + "issue:45245", + "issue:45356", + "issue:45405", + "issue:45656", + "issue:45712", + "issue:45812", + "issue:46132", + "issue:46231", + "issue:46665", + "issue:46762" + ], + "soft_pairs": [ + "issue:44016|issue:46231", + "issue:42898|issue:45812", + "issue:43723|issue:44568", + "issue:42222|issue:45200", + "issue:43479|issue:44079", + "issue:42898|issue:44779", + "issue:43784|issue:43824", + "issue:44734|issue:45103", + "issue:43874|issue:45092", + "issue:27957|issue:30333", + "issue:30333|issue:34689", + "issue:42898|issue:45356", + "issue:42898|issue:43295", + "issue:43116|issue:44190", + "issue:44060|issue:45125", + "issue:45405|issue:45712", + "issue:44295|issue:45656", + "issue:30333|issue:35141", + "issue:22355|issue:28282", + "issue:44933|issue:45712", + "issue:44075|issue:46231", + "issue:44222|issue:44297", + "issue:44360|issue:46762", + "issue:44448|issue:46132", + "issue:42831|issue:44623", + "issue:44297|issue:45656", + "issue:42796|issue:43723", + "issue:43479|issue:43644", + "issue:44222|issue:44623", + "issue:45245|issue:46132", + "issue:43479|issue:46665" + ], + "prepared_review_unit_hash": "308a6781a2b6a267258e66ff99972b45a946b4919ce9fabebb7327baa8824d6f", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 31, + "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 mostly a loose set of unrelated issues that only share high-level transformers/tokenizer/regression language. I did not find any soft pair that looks like the same underlying bug or change, so all pairwise duplicates should be rejected.", + "confidence": 0.46, + "canonical_issue_reason": "Issue 42898 is the broadest representative of the weakest shared theme here: a Transformers v5 tokenizer behavior regression. It is still not a true duplicate of the others, but it is the closest umbrella issue among the set.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 42898 is the best single anchor because it describes a core tokenizer regression with broad impact, whereas the other issues are mostly model-specific, import/config, docs, or unrelated training/runtime bugs.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44016", + "right": "issue:46231", + "accept": false, + "reason": "Notebook syntax error and a docs parameter-name mismatch are unrelated." + }, + { + "left": "issue:42898", + "right": "issue:45812", + "accept": false, + "reason": "Both involve v5 tokenizer regressions, but they are different concrete bugs affecting different behaviors/models." + }, + { + "left": "issue:43723", + "right": "issue:44568", + "accept": false, + "reason": "Generic AutoTokenizer loading issue vs missing BOS/EOS insertion for a specific tokenizer; not the same bug." + }, + { + "left": "issue:42222", + "right": "issue:45200", + "accept": false, + "reason": "Different model families and different config defaults; no shared code-path bug." + }, + { + "left": "issue:43479", + "right": "issue:44079", + "accept": false, + "reason": "Multimodal config initialization and ModelOutput key assignment are unrelated internals." + }, + { + "left": "issue:42898", + "right": "issue:44779", + "accept": false, + "reason": "Both are tokenizer regressions in v5, but the symptoms and affected model behavior differ." + }, + { + "left": "issue:43784", + "right": "issue:43824", + "accept": false, + "reason": "Different import errors from different modules and causes." + }, + { + "left": "issue:44734", + "right": "issue:45103", + "accept": false, + "reason": "Serving KV-cache indexing bug and auto_docstring annotation crash are unrelated." + }, + { + "left": "issue:43874", + "right": "issue:45092", + "accept": false, + "reason": "Two different multimodal model-loading failures with different root causes." + }, + { + "left": "issue:27957", + "right": "issue:30333", + "accept": false, + "reason": "Flash Attention 2 support and MLFlow reporting on failed jobs are unrelated." + }, + { + "left": "issue:30333", + "right": "issue:34689", + "accept": false, + "reason": "Training-job reporting and Llama vision model loading are separate problems." + }, + { + "left": "issue:42898", + "right": "issue:45356", + "accept": false, + "reason": "Both are tokenizer regressions, but they concern different tokenization/codecs and warning behavior." + }, + { + "left": "issue:42898", + "right": "issue:43295", + "accept": false, + "reason": "General tokenizer cleanup behavior vs processor/tokenizer access regression are different issues." + }, + { + "left": "issue:43116", + "right": "issue:44190", + "accept": false, + "reason": "Example-script multi-label output bug and local dataset loading are unrelated." + }, + { + "left": "issue:44060", + "right": "issue:45125", + "accept": false, + "reason": "Different Qwen model bugs: tied-weights warning vs missing tensor-parallel plan." + }, + { + "left": "issue:45405", + "right": "issue:45712", + "accept": false, + "reason": "Released-version dependency bump and leftover dummy classes are unrelated maintenance issues." + }, + { + "left": "issue:44295", + "right": "issue:45656", + "accept": false, + "reason": "Buffer registration/read bug and deepspeed optimizer-step duplication are different runtime problems." + }, + { + "left": "issue:30333", + "right": "issue:35141", + "accept": false, + "reason": "MLFlow failure reporting vs embedding resize/post_init reinitialization are unrelated." + }, + { + "left": "issue:22355", + "right": "issue:28282", + "accept": false, + "reason": "Two distinct import/dependency errors with different missing modules/libraries." + }, + { + "left": "issue:44933", + "right": "issue:45712", + "accept": false, + "reason": "Import exposure from image_utils and dummy_pt_objects cleanup are different packaging issues." + }, + { + "left": "issue:44075", + "right": "issue:46231", + "accept": false, + "reason": "Optimizer behavior bug and documentation typo are unrelated." + }, + { + "left": "issue:44222", + "right": "issue:44297", + "accept": false, + "reason": "FP8 save_pretrained issue and tokenizer config mismatch are different save-path bugs." + }, + { + "left": "issue:44360", + "right": "issue:46762", + "accept": false, + "reason": "DSA indexer discussion and Minimax attention numerical differences are unrelated." + }, + { + "left": "issue:44448", + "right": "issue:46132", + "accept": false, + "reason": "Pegasus v4/v5 output regression and AttentionInterface.register behavior change are different code paths." + }, + { + "left": "issue:42831", + "right": "issue:44623", + "accept": false, + "reason": "FP8 accuracy issue and processor save_pretrained missing files are not the same bug." + }, + { + "left": "issue:44297", + "right": "issue:45656", + "accept": false, + "reason": "Tokenizer save metadata mismatch has no relation to deepspeed optimizer stepping." + }, + { + "left": "issue:42796", + "right": "issue:43723", + "accept": false, + "reason": "Both are tokenizer-related, but one is a model-specific BPE-marker bug and the other is a generic loading issue." + }, + { + "left": "issue:43479", + "right": "issue:43644", + "accept": false, + "reason": "Config default initialization and non-persistent buffer filling are unrelated." + }, + { + "left": "issue:44222", + "right": "issue:44623", + "accept": false, + "reason": "Different save_pretrained failures affecting different artifact types." + }, + { + "left": "issue:45245", + "right": "issue:46132", + "accept": false, + "reason": "Category-count runtime error and attention registration behavior change are unrelated." + }, + { + "left": "issue:43479", + "right": "issue:46665", + "accept": false, + "reason": "Multimodal config initialization and from_pretrained behavior inside classes are different 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": 23, + "review_unit_index": 46, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:22355", + "issue:27957", + "issue:30990", + "issue:38175", + "issue:41628", + "issue:42757", + "issue:42831", + "issue:42907", + "issue:43122", + "issue:43232", + "issue:43475", + "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:44484", + "issue:44485", + "issue:44623", + "issue:44734", + "issue:44829", + "issue:44977", + "issue:45071", + "issue:45092", + "issue:45103", + "issue:45310", + "issue:45357", + "issue:45381", + "issue:45834", + "issue:46002", + "issue:46050", + "issue:46082", + "issue:46132", + "issue:46209", + "issue:46305", + "issue:46328", + "issue:46421", + "issue:46424", + "issue:46431", + "issue:46655", + "issue:46710" + ], + "soft_pairs": [ + "issue:41628|issue:42757", + "issue:45071|issue:46082", + "issue:44373|issue:44485", + "issue:46132|issue:46710", + "issue:44297|issue:45310", + "issue:43723|issue:44977", + "issue:44222|issue:44484", + "issue:44190|issue:46082", + "issue:43232|issue:46424", + "issue:44206|issue:45103", + "issue:44263|issue:45092", + "issue:43122|issue:46209", + "issue:43232|issue:45834", + "issue:43475|issue:46305", + "issue:44734|issue:46421", + "issue:46209|issue:46431", + "issue:46305|issue:46655", + "issue:44263|issue:44360", + "issue:44295|issue:44623", + "issue:30990|issue:38175", + "issue:44263|issue:44373", + "issue:45381|issue:46305", + "issue:43644|issue:46305", + "issue:42907|issue:46050", + "issue:42831|issue:45310", + "issue:44279|issue:46328", + "issue:45092|issue:45357", + "issue:44829|issue:46002", + "issue:44261|issue:45092", + "issue:22355|issue:27957" + ], + "prepared_review_unit_hash": "ad815562c3f38bca37d13e0a9fd156c050c1e3223f10f43634c75f20f3f1bd9d", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36306, + "estimated_input_tokens": 9077, + "estimated_eval_tokens": 18410 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is a grab bag of superficially similar Transformers issues, mostly around v5 regressions, save/load, generation, cache, or model-specific bugs, but the paired artifacts do not describe the same underlying defect. No soft edge is a safe duplicate merge.", + "confidence": 0.34, + "canonical_issue_reason": "issue:44263 is the best anchor for this cluster because it is the most concrete code-path bug and sits near the only small related GlmMoeDsaIndexer/DSA subset; it is more substantive than the nearby docstring-only item.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44263 is the strongest representative issue in the set: specific, technical, and the only one with adjacent near-match discussion, while the rest are mostly isolated model- or version-specific reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:41628", + "right": "issue:42757", + "accept": false, + "reason": "Both are import errors, but for different symbols and different dependency layers; not the same bug." + }, + { + "left": "issue:45071", + "right": "issue:46082", + "accept": false, + "reason": "One is type-checking for PretrainedConfig, the other is LlamaConfig head_dim validation; unrelated failures." + }, + { + "left": "issue:44373", + "right": "issue:44485", + "accept": false, + "reason": "Docstring mismatch versus RoPE implementation bug; different problem types and code paths." + }, + { + "left": "issue:46132", + "right": "issue:46710", + "accept": false, + "reason": "AttentionInterface registration behavior and DeepSeek accuracy regression are not the same defect." + }, + { + "left": "issue:44297", + "right": "issue:45310", + "accept": false, + "reason": "Tokenizer save_pretrained metadata mismatch versus Qwen3.5 from_pretrained failure; too different and model-specific." + }, + { + "left": "issue:43723", + "right": "issue:44977", + "accept": false, + "reason": "Tokenizer loading regression and flash-attention generation issue are unrelated." + }, + { + "left": "issue:44222", + "right": "issue:44484", + "accept": false, + "reason": "FP8 save_pretrained bug versus max_shard_size default discussion; not a duplicate." + }, + { + "left": "issue:44190", + "right": "issue:46082", + "accept": false, + "reason": "Dataset loading in a training script versus LlamaConfig head_dim validation; no shared code bug." + }, + { + "left": "issue:43232", + "right": "issue:46424", + "accept": false, + "reason": "Both mention generation/cache, but one is sync_gpus kwargs handling and the other cache length honoring; different paths." + }, + { + "left": "issue:44206", + "right": "issue:45103", + "accept": false, + "reason": "Feature extractor center-arg crash and auto_docstring annotation crash are unrelated." + }, + { + "left": "issue:44263", + "right": "issue:45092", + "accept": false, + "reason": "GlmMoeDsaIndexer split handling and remote-code/meta-init incompatibility are different issues." + }, + { + "left": "issue:43122", + "right": "issue:46209", + "accept": false, + "reason": "Tokenization regression versus FP8 inference on B200; no common underlying bug." + }, + { + "left": "issue:43232", + "right": "issue:45834", + "accept": false, + "reason": "Generation cache kwargs and Kosmos2.5 OCR index error are unrelated." + }, + { + "left": "issue:43475", + "right": "issue:46305", + "accept": false, + "reason": "SAM 3 video encoder attribute error and MoonshineStreaming window off-by-one are different models and code paths." + }, + { + "left": "issue:44734", + "right": "issue:46421", + "accept": false, + "reason": "Serve KV-cache indexing crash versus chunked-prefill recompilation; both cache-related but not the same defect." + }, + { + "left": "issue:46209", + "right": "issue:46431", + "accept": false, + "reason": "FP8 B200 inference failure and missing chat template error are unrelated." + }, + { + "left": "issue:46305", + "right": "issue:46655", + "accept": false, + "reason": "Both are video-related, but one is a streaming mask off-by-one and the other is padding calculation in a different processor." + }, + { + "left": "issue:44263", + "right": "issue:44360", + "accept": false, + "reason": "Same DSA/GlmMoe area, but one is torch.split return handling and the other claims a missing ReLU; too different to merge." + }, + { + "left": "issue:44295", + "right": "issue:44623", + "accept": false, + "reason": "Reading registered position_ids and processor.save_pretrained missing files are separate save/serialization problems." + }, + { + "left": "issue:30990", + "right": "issue:38175", + "accept": false, + "reason": "Stuck loading Sentence Transformers versus zero probabilities in siglip2; unrelated symptoms." + }, + { + "left": "issue:44263", + "right": "issue:44373", + "accept": false, + "reason": "Code bug versus docstring issue; not the same underlying change." + }, + { + "left": "issue:45381", + "right": "issue:46305", + "accept": false, + "reason": "Qwen2.5-VL vision_position_ids bug and MoonshineStreaming frame-window bug are different models and logic." + }, + { + "left": "issue:43644", + "right": "issue:46305", + "accept": false, + "reason": "Non-persistent buffer initialization in v5 versus streaming mask off-by-one; unrelated." + }, + { + "left": "issue:42907", + "right": "issue:46050", + "accept": false, + "reason": "Dequantized save failure and sinq save/load breakage both involve quantization, but they are not the same concrete bug." + }, + { + "left": "issue:42831", + "right": "issue:45310", + "accept": false, + "reason": "FineGrainedFP8 accuracy issue and Qwen3.5 Moe from_pretrained error are unrelated." + }, + { + "left": "issue:44279", + "right": "issue:46328", + "accept": false, + "reason": "Generic dependency issue and SequenceFeatureExtractor.pad performance regression are unrelated." + }, + { + "left": "issue:45092", + "right": "issue:45357", + "accept": false, + "reason": "Remote-code/meta-init incompatibility and incorrect visual encoder keys on save_pretrained are different regressions." + }, + { + "left": "issue:44829", + "right": "issue:46002", + "accept": false, + "reason": "Flash Attention 3 training degeneration and NaNs after checkout+init do not describe the same bug." + }, + { + "left": "issue:44261", + "right": "issue:45092", + "accept": false, + "reason": "MLA RMS norm epsilon precision issue and remote-code meta-init incompatibility are distinct model bugs." + }, + { + "left": "issue:22355", + "right": "issue:27957", + "accept": false, + "reason": "Importing transformers.onnx and XLMRoberta Flash Attention 2 are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 47, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "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: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:44488", + "issue:44568", + "issue:44704", + "issue:44945", + "issue:44964", + "issue:45278", + "issue:45406", + "issue:45593", + "issue:45753", + "issue:46002", + "issue:46172", + "issue:46369", + "issue:46402", + "issue:46762" + ], + "soft_pairs": [ + "issue:44483|issue:45406", + "issue:41628|issue:43824", + "issue:42831|issue:44038", + "issue:43408|issue:46402", + "issue:22355|issue:28218", + "issue:42831|issue:45753", + "issue:39401|issue:44568", + "issue:27957|issue:30990", + "issue:42757|issue:45278", + "issue:42886|issue:43066", + "issue:42886|issue:42994", + "issue:44188|issue:44373", + "issue:43408|issue:46369", + "issue:43232|issue:45593", + "issue:44261|issue:44373", + "issue:43646|issue:46002", + "issue:42796|issue:43122", + "issue:42907|issue:44945", + "issue:30990|issue:33357", + "issue:43576|issue:43784", + "issue:42371|issue:44263", + "issue:43756|issue:44060", + "issue:39401|issue:44488", + "issue:42907|issue:44964", + "issue:43976|issue:44945", + "issue:44295|issue:44297", + "issue:46172|issue:46762", + "issue:44117|issue:44704" + ], + "prepared_review_unit_hash": "a8473b8e70675ec56ea36f0201ac5a696d31f6bec954c6eea69694059a1ada43", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 35512, + "estimated_input_tokens": 8878, + "estimated_eval_tokens": 18012 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is largely heterogeneous: most items are unrelated model, tokenizer, import, or runtime regressions that only share broad Hugging Face/Transformers vocabulary. No soft pair looks like a clear same-bug duplicate.", + "confidence": 0.94, + "canonical_issue_reason": "No strong duplicate group emerges; if one issue must represent the set, 45278 is the broadest umbrella because it describes multiple import failures after an upgrade, but the cluster overall is not cohesive.", + "canonical_pr_reason": null, + "best_issue_reason": "45278 is the most general cross-cutting report in the set and best fits as a loose representative for import-regression complaints, though it is still not a true canonical duplicate target.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44483", + "right": "issue:45406", + "accept": false, + "reason": "Different problems: API request rejection vs a processor attribute error." + }, + { + "left": "issue:41628", + "right": "issue:43824", + "accept": false, + "reason": "Both are import errors, but for different missing symbols/models and likely different causes." + }, + { + "left": "issue:42831", + "right": "issue:44038", + "accept": false, + "reason": "Unrelated: FP8 accuracy regression vs a Qwen3-VL-Moe version issue." + }, + { + "left": "issue:43408", + "right": "issue:46402", + "accept": false, + "reason": "Both mention sam3, but one is a model-type warning and the other is missing text-encoder weights." + }, + { + "left": "issue:22355", + "right": "issue:28218", + "accept": false, + "reason": "No overlap: missing transformers.onnx module vs tokenizer spacing behavior." + }, + { + "left": "issue:42831", + "right": "issue:45753", + "accept": false, + "reason": "Different model families and symptoms; not the same bug." + }, + { + "left": "issue:39401", + "right": "issue:44568", + "accept": false, + "reason": "Both are tokenizer issues, but offset mappings and special-token insertion are distinct code paths." + }, + { + "left": "issue:27957", + "right": "issue:30990", + "accept": false, + "reason": "Flash Attention 2 support vs a loading hang; too different to merge." + }, + { + "left": "issue:42757", + "right": "issue:45278", + "accept": false, + "reason": "45278 is a broad import-regression report, but 42757 is a specific dependency symbol import failure." + }, + { + "left": "issue:42886", + "right": "issue:43066", + "accept": false, + "reason": "Offline cache loading and decoder-type selection are unrelated tokenizer issues." + }, + { + "left": "issue:42886", + "right": "issue:42994", + "accept": false, + "reason": "Offline cache loading is not the same as quantized model saving." + }, + { + "left": "issue:44188", + "right": "issue:44373", + "accept": false, + "reason": "Attention-kernel divergence and a docstring issue are unrelated." + }, + { + "left": "issue:43408", + "right": "issue:46369", + "accept": false, + "reason": "Same family name, but model warning vs memory overflow are different failures." + }, + { + "left": "issue:43232", + "right": "issue:45593", + "accept": false, + "reason": "Generation kwargs handling and D-FINE auxiliary-loss behavior are unrelated." + }, + { + "left": "issue:44261", + "right": "issue:44373", + "accept": false, + "reason": "Precision/config bug vs documentation mismatch." + }, + { + "left": "issue:43646", + "right": "issue:46002", + "accept": false, + "reason": "Custom model init breakage and NaNs on checkout/init do not share a concrete code path." + }, + { + "left": "issue:42796", + "right": "issue:43122", + "accept": false, + "reason": "Both tokenize differently, but one is BPE-marker handling and the other is version-to-version tokenization drift." + }, + { + "left": "issue:42907", + "right": "issue:44945", + "accept": false, + "reason": "Model saving after dequantization vs incorrect pipeline-parallel output are different problems." + }, + { + "left": "issue:30990", + "right": "issue:33357", + "accept": false, + "reason": "Loading hang and MacOS bus error are both runtime issues, but not the same bug." + }, + { + "left": "issue:43576", + "right": "issue:43784", + "accept": false, + "reason": "Broken CLI env command and an import NameError in sentence-transformers are unrelated." + }, + { + "left": "issue:42371", + "right": "issue:44263", + "accept": false, + "reason": "TF32 settings guidance vs a torch.split indexing issue are unrelated." + }, + { + "left": "issue:43756", + "right": "issue:44060", + "accept": false, + "reason": "Different model-specific architecture/weighting issues; not the same code-path bug." + }, + { + "left": "issue:39401", + "right": "issue:44488", + "accept": false, + "reason": "Tokenizer offset mapping bug vs model loading failure for a specific checkpoint." + }, + { + "left": "issue:42907", + "right": "issue:44964", + "accept": false, + "reason": "Quantized/dequantized saving failure and multimodal model loading failure are distinct." + }, + { + "left": "issue:43976", + "right": "issue:44945", + "accept": false, + "reason": "Python version compatibility and pipeline-parallel output correctness are unrelated." + }, + { + "left": "issue:44295", + "right": "issue:44297", + "accept": false, + "reason": "Buffer read/access error vs tokenizer_config class mismatch are different serialization bugs." + }, + { + "left": "issue:46172", + "right": "issue:46762", + "accept": false, + "reason": "Both are attention-related numerics, but on different models and likely different kernels." + }, + { + "left": "issue:44117", + "right": "issue:44704", + "accept": false, + "reason": "Tokenizer mapping lookup and AutoProcessor kwargs forwarding are separate API plumbing issues." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 48, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "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:43611", + "issue:43784", + "issue:43824", + "issue:43937", + "issue:44038", + "issue:44188", + "issue:44222", + "issue:44295", + "issue:44297", + "issue:44351", + "issue:44360", + "issue:44485", + "issue:44521", + "issue:44625", + "issue:44704", + "issue:44779", + "issue:44829", + "issue:44938", + "issue:44945", + "issue:45200", + "issue:45216", + "issue:45278", + "issue:45290", + "issue:45478", + "issue:45600", + "issue:45753", + "issue:46002", + "issue:46326", + "issue:46328", + "issue:46614" + ], + "soft_pairs": [ + "issue:43784|issue:45600", + "issue:42907|issue:43937", + "issue:41628|issue:45278", + "issue:44625|issue:44829", + "issue:44704|issue:45290", + "issue:39401|issue:44521", + "issue:44360|issue:44485", + "issue:43408|issue:43475", + "issue:43122|issue:44521", + "issue:43122|issue:46326", + "issue:44038|issue:44297", + "issue:42907|issue:43452", + "issue:22355|issue:29159", + "issue:43116|issue:45200", + "issue:39401|issue:46614", + "issue:45216|issue:45478", + "issue:44038|issue:44222", + "issue:44038|issue:44295", + "issue:43824|issue:44351", + "issue:43122|issue:44779", + "issue:43278|issue:44945", + "issue:42757|issue:44351", + "issue:36296|issue:43116", + "issue:22355|issue:29127", + "issue:42175|issue:44938", + "issue:43493|issue:45753", + "issue:44188|issue:46328", + "issue:42898|issue:46614", + "issue:43937|issue:44485", + "issue:38617|issue:45278", + "issue:43611|issue:46002", + "issue:43116|issue:43493" + ], + "prepared_review_unit_hash": "533cab0528bab437feb1dba92e8894e96d125c5aefaa020850c8fd15a8e993e1", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "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": "The cluster is highly heterogeneous: mostly unrelated import, tokenizer, multimodal, and model-specific regressions. None of the soft pairs look like the same underlying bug/change, so all candidate duplicate edges should be rejected.", + "confidence": 0.92, + "canonical_issue_reason": "Issue 43122 is the broadest version-regression tokenizer report among the set and is the closest thematic anchor for the tokenizer-heavy subset, though the overall cluster is still quite mixed.", + "canonical_pr_reason": null, + "best_issue_reason": "43122 is the best representative issue because it is a broad, user-facing tokenizer regression and sits nearest to the small tokenizer subgroup; the rest of the cluster is too diverse to support a tighter canonical choice.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43784", + "right": "issue:45600", + "accept": false, + "reason": "Both are import-related, but one is a missing symbol in sentence-transformers integration and the other is a removed config reference in auto_mappings; different code paths." + }, + { + "left": "issue:42907", + "right": "issue:43937", + "accept": false, + "reason": "Unrelated bugs: dequantized model saving versus GLM-5 generation config validation." + }, + { + "left": "issue:41628", + "right": "issue:45278", + "accept": false, + "reason": "One is a specific missing AutoImageProcessor export; the other is a broad upgrade-induced import breakage report." + }, + { + "left": "issue:44625", + "right": "issue:44829", + "accept": false, + "reason": "Config propagation for Qwen3.5 is unrelated to a flash_attention_3 training degeneration bug." + }, + { + "left": "issue:44704", + "right": "issue:45290", + "accept": false, + "reason": "Processor keyword forwarding to cached_file is a different bug than chat template crashing on tool-call assistant messages." + }, + { + "left": "issue:39401", + "right": "issue:44521", + "accept": false, + "reason": "Both involve tokenization, but one is offset mapping and the other is assistant mask generation for multimodal chat templates." + }, + { + "left": "issue:44360", + "right": "issue:44485", + "accept": false, + "reason": "These are unrelated: a DSA indexer discussion versus a GLM-5 RoPE implementation report." + }, + { + "left": "issue:43408", + "right": "issue:43475", + "accept": false, + "reason": "Same SAM 3 Video area, but a model-type warning and a missing output attribute are distinct problems." + }, + { + "left": "issue:43122", + "right": "issue:44521", + "accept": false, + "reason": "Tokenizer regression versus multimodal chat-template assistant mask failure; not the same code path." + }, + { + "left": "issue:43122", + "right": "issue:46326", + "accept": false, + "reason": "Both tokenizer-related, but one is a v5 regression in tokenization behavior and the other is continue_final_message handling for mistral_common tokenizers." + }, + { + "left": "issue:44038", + "right": "issue:44297", + "accept": false, + "reason": "Qwen3-VL-Moe import issues are unrelated to tokenizer.save_pretrained metadata mismatch." + }, + { + "left": "issue:42907", + "right": "issue:43452", + "accept": false, + "reason": "Saving dequantized models and gguf loading failures are different serialization/loading bugs." + }, + { + "left": "issue:22355", + "right": "issue:29159", + "accept": false, + "reason": "Missing transformers.onnx module is unrelated to slow/fast tokenizer inconsistency." + }, + { + "left": "issue:43116", + "right": "issue:45200", + "accept": false, + "reason": "Empty multi-label classification results and Gemma 4 mm_token_type_ids handling are different issues." + }, + { + "left": "issue:39401", + "right": "issue:46614", + "accept": false, + "reason": "Both tokenizer bugs, but offset_mapping for Qwen3 and stale phonemizer backend state for Wav2Vec2 are unrelated." + }, + { + "left": "issue:45216", + "right": "issue:45478", + "accept": false, + "reason": "Both mention Qwen3.5 and save/load regressions, but one is checkpoint saving correctness and the other is from_pretrained loading failure." + }, + { + "left": "issue:44038", + "right": "issue:44222", + "accept": false, + "reason": "Qwen3-VL-Moe import problems and FP8 save_pretrained issues are separate model/serialization bugs." + }, + { + "left": "issue:44038", + "right": "issue:44295", + "accept": false, + "reason": "The Qwen3-VL-Moe import bug is not the same as reading a registered position_ids buffer." + }, + { + "left": "issue:43824", + "right": "issue:44351", + "accept": false, + "reason": "Different missing imports: Qwen2_5_VLForConditionalGeneration versus HybridCache." + }, + { + "left": "issue:43122", + "right": "issue:44779", + "accept": false, + "reason": "Both are tokenizer regressions, but they affect different models and behavior surfaces; no clear shared bug." + }, + { + "left": "issue:43278", + "right": "issue:44945", + "accept": false, + "reason": "BF16-to-FP32 dtype drift in evaluation is unrelated to incorrect pipeline-parallel LLM output." + }, + { + "left": "issue:42757", + "right": "issue:44351", + "accept": false, + "reason": "Different libraries and symbols: huggingface_hub offline mode import versus transformers HybridCache import." + }, + { + "left": "issue:36296", + "right": "issue:43116", + "accept": false, + "reason": "Tensor parallel training instability is unrelated to the classification example script returning empty results." + }, + { + "left": "issue:22355", + "right": "issue:29127", + "accept": false, + "reason": "No transformers.onnx module is unrelated to the LayoutLMv3 error message clarity issue." + }, + { + "left": "issue:42175", + "right": "issue:44938", + "accept": false, + "reason": "Backend packaging for torch installs and Python 3.14 load failures are different environment issues." + }, + { + "left": "issue:43493", + "right": "issue:45753", + "accept": false, + "reason": "Different model families and failures: SigLIP2 implementation discrepancy versus Qwen3_5 infinite loop on one image." + }, + { + "left": "issue:44188", + "right": "issue:46328", + "accept": false, + "reason": "Attention-kernel divergence under torch.compile and SequenceFeatureExtractor.pad inefficiency are unrelated." + }, + { + "left": "issue:42898", + "right": "issue:46614", + "accept": false, + "reason": "v5 clean_up_tokenization_spaces behavior change is not the same as stale phonemizer language state." + }, + { + "left": "issue:43937", + "right": "issue:44485", + "accept": false, + "reason": "GenerationConfig validation for GLM-5 is unrelated to the GLM-5 RoPE implementation discussion." + }, + { + "left": "issue:38617", + "right": "issue:45278", + "accept": false, + "reason": "A specific missing layer_type_validation import and a broad import-error-on-upgrade report are not the same bug." + }, + { + "left": "issue:43611", + "right": "issue:46002", + "accept": false, + "reason": "base_model_prefix loading regression and NaNs in classification heads are different training/loading problems." + }, + { + "left": "issue:43116", + "right": "issue:43493", + "accept": false, + "reason": "Classification example output emptiness and SigLIP2 implementation discrepancy do not share a concrete code path." + }, + { + "left": "issue:42757", + "right": "issue:42757", + "accept": false, + "reason": "This pair is not present in the input soft edges; no edge to accept." + }, + { + "left": "issue:42898", + "right": "issue:44521", + "accept": false, + "reason": "Tokenization-space cleanup behavior and multimodal assistant-mask generation are different features." + }, + { + "left": "issue:44704", + "right": "issue:45278", + "accept": false, + "reason": "AutoProcessor kwargs forwarding and broad import errors after upgrade are unrelated." + }, + { + "left": "issue:44038", + "right": "issue:44297", + "accept": false, + "reason": "Qwen3-VL-Moe import issues are unrelated to tokenizer config metadata mismatch." + }, + { + "left": "issue:43122", + "right": "issue:44521", + "accept": false, + "reason": "Tokenizer regression and multimodal assistant mask failure are not the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 49, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:28348", + "issue:33453", + "issue:39401", + "issue:42222", + "issue:42371", + "issue:42673", + "issue:42796", + "issue:43064", + "issue:43066", + "issue:43116", + "issue:43441", + "issue:43673", + "issue:43825", + "issue:43827", + "issue:43874", + "issue:43937", + "issue:43976", + "issue:43992", + "issue:44038", + "issue:44075", + "issue:44263", + "issue:44355", + "issue:44380", + "issue:44625", + "issue:44704", + "issue:44779", + "issue:44936", + "issue:44995", + "issue:45072", + "issue:45083", + "issue:45092", + "issue:45356", + "issue:45357", + "issue:45446", + "issue:45478", + "issue:45563", + "issue:46002", + "issue:46077", + "issue:46231", + "issue:46328", + "issue:46614", + "issue:46640", + "issue:46655", + "issue:46697", + "issue:46710", + "issue:46752", + "issue:46762" + ], + "soft_pairs": [ + "issue:44355|issue:46231", + "issue:43976|issue:45446", + "issue:43066|issue:44779", + "issue:44038|issue:45478", + "issue:42673|issue:45072", + "issue:44625|issue:46002", + "issue:45083|issue:45446", + "issue:45357|issue:45478", + "issue:42371|issue:43874", + "issue:42222|issue:43441", + "issue:43116|issue:43825", + "issue:44075|issue:44936", + "issue:43992|issue:44704", + "issue:44995|issue:46697", + "issue:28348|issue:46640", + "issue:43066|issue:46710", + "issue:46614|issue:46752", + "issue:44263|issue:46762", + "issue:43827|issue:45446", + "issue:45092|issue:46762", + "issue:43064|issue:43673", + "issue:42673|issue:42796", + "issue:43937|issue:45563", + "issue:33453|issue:39401", + "issue:44380|issue:46077", + "issue:45356|issue:46614", + "issue:43673|issue:46328", + "issue:42673|issue:46655" + ], + "prepared_review_unit_hash": "0e3f8c2fbc2031b41a314222aab6be10c5332229f4f9392c41963fe58115f84d", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 35258, + "estimated_input_tokens": 8815, + "estimated_eval_tokens": 17886 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: the proposed soft links span unrelated bugs across tokenizers, generation, attention, training, vision, and docs. I would not merge any of the soft pairs as duplicates; no PRs are present in the cluster.", + "confidence": 0.82, + "canonical_issue_reason": "issue:33453 is the broadest, most generic tokenizer regression ticket here, so it is the best single issue to anchor the cluster if one must be chosen.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:33453 is the most suitable representative because it is general enough to cover adjacent tokenizer-loading reports, whereas the other issues are more model-specific or on different code paths.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44355", + "right": "issue:46231", + "accept": false, + "reason": "Compiled-file execution errors and a docs parameter-name mismatch are unrelated." + }, + { + "left": "issue:43976", + "right": "issue:45446", + "accept": false, + "reason": "Python version compatibility failure vs. a flex_attention import guard bug; different root causes." + }, + { + "left": "issue:43066", + "right": "issue:44779", + "accept": false, + "reason": "Both are tokenizer-related, but one is a decoder-type regression and the other is DeepSeek output correctness; not the same concrete bug/fix." + }, + { + "left": "issue:44038", + "right": "issue:45478", + "accept": false, + "reason": "Different Qwen model families and different from_pretrained failure modes." + }, + { + "left": "issue:42673", + "right": "issue:45072", + "accept": false, + "reason": "VRAM leak in dataloader-thread usage vs. bfloat16 dtype mismatch in inference; unrelated." + }, + { + "left": "issue:44625", + "right": "issue:46002", + "accept": false, + "reason": "num_labels propagation bug and NaNs in classification heads are distinct issues." + }, + { + "left": "issue:45083", + "right": "issue:45446", + "accept": false, + "reason": "Feature-length helper behavior and a PyTorch version check are unrelated." + }, + { + "left": "issue:45357", + "right": "issue:45478", + "accept": false, + "reason": "Both touch Qwen3.5, but one is save_pretrained key serialization and the other is a load-time from_pretrained error; not the same bug." + }, + { + "left": "issue:42371", + "right": "issue:43874", + "accept": false, + "reason": "TF32 settings API guidance and a missing image-patch method are unrelated." + }, + { + "left": "issue:42222", + "right": "issue:43441", + "accept": false, + "reason": "vitpose breakage and Ministral FlashAttention failures have different code paths and symptoms." + }, + { + "left": "issue:43116", + "right": "issue:43825", + "accept": false, + "reason": "Multi-label classification example failure vs. pipeline() translation error messaging; unrelated." + }, + { + "left": "issue:44075", + "right": "issue:44936", + "accept": false, + "reason": "Optimizer SGD args not being used and trainer.evaluate() failing after training are different problems." + }, + { + "left": "issue:43992", + "right": "issue:44704", + "accept": false, + "reason": "Missing embed_tokens weight in UMT5Encoder and AutoProcessor kwarg forwarding are unrelated." + }, + { + "left": "issue:44995", + "right": "issue:46697", + "accept": false, + "reason": "GlmMoe stale indexer cache vs. Swin2SR over-padding are different subsystems and bugs." + }, + { + "left": "issue:28348", + "right": "issue:46640", + "accept": false, + "reason": "FlashAttention 2 support request for GPT2 and a generic Flash Attention issue are not the same concrete change or bug." + }, + { + "left": "issue:43066", + "right": "issue:46710", + "accept": false, + "reason": "Tokenizer decoder-type regression and a DeepSeek accuracy regression after upgrade are not the same underlying defect." + }, + { + "left": "issue:46614", + "right": "issue:46752", + "accept": false, + "reason": "Stale phonemizer backend state and apply_chat_template() crashing on empty input are unrelated." + }, + { + "left": "issue:44263", + "right": "issue:46762", + "accept": false, + "reason": "torch.split return-value handling in GlmMoeDsaIndexer and Minimax MSA numerical differences are different issues." + }, + { + "left": "issue:43827", + "right": "issue:45446", + "accept": false, + "reason": "Docs still referencing pipeline() and a flex_attention import version check are unrelated." + }, + { + "left": "issue:45092", + "right": "issue:46762", + "accept": false, + "reason": "InternVL2 meta-init incompatibility and Minimax MSA numerical drift are unrelated." + }, + { + "left": "issue:43064", + "right": "issue:43673", + "accept": false, + "reason": "Wrong optimizer state placement under FSDP2/PEFT and missing GenerationMixin cache during chunked_prefill are different code paths." + }, + { + "left": "issue:42673", + "right": "issue:42796", + "accept": false, + "reason": "A VRAM leak in multithreaded loading and a tokenizer BPE-marker handling bug are unrelated." + }, + { + "left": "issue:43937", + "right": "issue:45563", + "accept": false, + "reason": "GenerationConfig validation failure and a stale warning about num_return_sequences are different symptoms and fixes." + }, + { + "left": "issue:33453", + "right": "issue:39401", + "accept": false, + "reason": "Tokenizer loading regression and wrong offset_mapping are related area-wise but not the same underlying bug." + }, + { + "left": "issue:44380", + "right": "issue:46077", + "accept": false, + "reason": "GPT2 attention scaling ignored under SDPA/FlashAttention and Gemma4 bidirectional mask construction are different model-specific issues." + }, + { + "left": "issue:45356", + "right": "issue:46614", + "accept": false, + "reason": "Kimi tokenizer codec/warning regression and a stale phonemizer language backend are unrelated." + }, + { + "left": "issue:42673", + "right": "issue:46655", + "accept": false, + "reason": "VRAM leak in multithreaded usage and incorrect video padding calculation are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 50, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:38617", + "issue:39401", + "issue:41628", + "issue:42222", + "issue:42371", + "issue:42617", + "issue:43122", + "issue:43408", + "issue:43450", + "issue:43742", + "issue:43937", + "issue:43986", + "issue:44117", + "issue:44220", + "issue:44261", + "issue:44291", + "issue:44410", + "issue:44466", + "issue:44484", + "issue:44496", + "issue:44779", + "issue:44829", + "issue:44877", + "issue:44933", + "issue:44977", + "issue:45071", + "issue:45081", + "issue:45216", + "issue:45278", + "issue:45310", + "issue:45341", + "issue:45362", + "issue:45405", + "issue:45446", + "issue:45478", + "issue:45561", + "issue:45588", + "issue:46002", + "issue:46178", + "issue:46209", + "issue:46231", + "issue:46298", + "issue:46328", + "issue:46512", + "issue:46617", + "issue:46752", + "issue:46762" + ], + "soft_pairs": [ + "issue:44291|issue:45071", + "issue:42222|issue:46752", + "issue:45405|issue:45446", + "issue:44117|issue:45588", + "issue:44466|issue:46002", + "issue:44410|issue:45310", + "issue:39401|issue:44779", + "issue:43937|issue:44977", + "issue:45071|issue:46512", + "issue:45341|issue:45561", + "issue:43450|issue:43986", + "issue:45341|issue:46328", + "issue:44829|issue:45216", + "issue:44117|issue:45081", + "issue:42617|issue:45341", + "issue:38617|issue:41628", + "issue:43408|issue:43742", + "issue:45362|issue:46328", + "issue:43122|issue:46298", + "issue:44877|issue:46231", + "issue:42222|issue:46178", + "issue:44484|issue:45310", + "issue:42371|issue:44261", + "issue:39401|issue:46209", + "issue:43937|issue:44496", + "issue:42371|issue:46617", + "issue:44410|issue:45478", + "issue:44261|issue:46762", + "issue:44933|issue:45278", + "issue:38617|issue:44220" + ], + "prepared_review_unit_hash": "3bbfbafbffa06188b8b79113d2fa47ab4bc420244b0ef07b22e5964feb18e10f", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 35930, + "estimated_input_tokens": 8983, + "estimated_eval_tokens": 18222 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: most items are unrelated model-loading, tokenizer, config, testing, or performance bugs. I found no soft-edge pair that clearly looks like the same underlying defect, so I rejected all candidate merges. The closest shared theme is import/API regression issues, where issue 45278 is the broadest representative.", + "confidence": 0.61, + "canonical_issue_reason": "issue:45278 is the broadest umbrella report in the cluster (multiple import errors after upgrading from 4.57.0 to 5.5.0) and is the best representative of the small import-regression subset.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45278 best captures the only meaningful mini-theme here (import/API breakage); the rest of the cluster is too diverse to support a true duplicate canonical.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44291", + "right": "issue:45071", + "accept": false, + "reason": "Different bugs: one is an unexpected `_is_hf_initialized` argument during empty-weight model loading, the other is `PretrainedConfig` type checking. Same area, not same defect." + }, + { + "left": "issue:42222", + "right": "issue:46752", + "accept": false, + "reason": "Completely different failures: broken vitpose models vs `apply_chat_template` crashing on an empty list." + }, + { + "left": "issue:45405", + "right": "issue:45446", + "accept": false, + "reason": "PEFT version bump availability issue vs a PyTorch version check bug in flex attention." + }, + { + "left": "issue:44117", + "right": "issue:45588", + "accept": false, + "reason": "Tokenizer mapping fallback bug vs flash-attention crash on `s_aux=None`; unrelated code paths." + }, + { + "left": "issue:44466", + "right": "issue:46002", + "accept": false, + "reason": "Serialization of tied weights vs NaNs in classification heads; different symptoms and fixes." + }, + { + "left": "issue:44410", + "right": "issue:45310", + "accept": false, + "reason": "Missing Qwen3Next projections vs Qwen3.5 MoE `from_pretrained` error; not the same underlying model bug." + }, + { + "left": "issue:39401", + "right": "issue:44779", + "accept": false, + "reason": "Both tokenizer-related, but one is offset mapping and the other is incorrect Deepseek tokenization results; no clear shared root cause." + }, + { + "left": "issue:43937", + "right": "issue:44977", + "accept": false, + "reason": "Invalid `GenerationConfig` vs flash-attention generation failure; different failure modes." + }, + { + "left": "issue:45071", + "right": "issue:46512", + "accept": false, + "reason": "Type-checking regression vs stale `config.dtype` after `from_config`; not the same bug." + }, + { + "left": "issue:45341", + "right": "issue:45561", + "accept": false, + "reason": "A small testing-utils bug vs an xdist race on `captured_info.txt`; different code paths." + }, + { + "left": "issue:43450", + "right": "issue:43986", + "accept": false, + "reason": "Batched video processor shape bug vs crash loading a video model without torchvision; separate issues." + }, + { + "left": "issue:45341", + "right": "issue:46328", + "accept": false, + "reason": "Testing-utils typo/bug vs `SequenceFeatureExtractor.pad` inefficiency; unrelated." + }, + { + "left": "issue:44829", + "right": "issue:45216", + "accept": false, + "reason": "Degenerate training with flash attention 3 vs incorrect checkpoint serialization for Qwen3.5; not mergeable as one fix." + }, + { + "left": "issue:44117", + "right": "issue:45081", + "accept": false, + "reason": "Tokenizer mapping `None` handling vs Mistral regex patch crash from missing `backend_tokenizer`; different concrete bugs." + }, + { + "left": "issue:42617", + "right": "issue:45341", + "accept": false, + "reason": "3d_parallel.py runtime issue vs a testing-utils bug; no overlap beyond being in the repo." + }, + { + "left": "issue:38617", + "right": "issue:41628", + "accept": false, + "reason": "Both are import errors, but for different missing symbols (`layer_type_validation` vs `AutoImageProcessor`) and likely different fixes." + }, + { + "left": "issue:43408", + "right": "issue:43742", + "accept": false, + "reason": "SAM3 tracker/video-model warning vs MobileLLM config/key error; unrelated model-loading problems." + }, + { + "left": "issue:45362", + "right": "issue:46328", + "accept": false, + "reason": "Qwen3.5 chat crash vs a padding performance issue; different severity and cause." + }, + { + "left": "issue:43122", + "right": "issue:46298", + "accept": false, + "reason": "Tokenizer behavior regression across versions vs noisy logging; no duplicate relationship." + }, + { + "left": "issue:44877", + "right": "issue:46231", + "accept": false, + "reason": "Strict config preventing granite_speech loading vs a documentation parameter-name mismatch." + }, + { + "left": "issue:42222", + "right": "issue:46178", + "accept": false, + "reason": "Vitpose model breakage vs backend mapping missing `tensorflow_text`; both loader-related but not the same bug." + }, + { + "left": "issue:44484", + "right": "issue:45310", + "accept": false, + "reason": "Shard-size question vs Qwen3.5 MoE loading error; no shared underlying defect." + }, + { + "left": "issue:42371", + "right": "issue:44261", + "accept": false, + "reason": "TF32 API-setting guidance vs missing `rms_norm_eps` causing precision error; unrelated changes." + }, + { + "left": "issue:39401", + "right": "issue:46209", + "accept": false, + "reason": "Tokenizer offset mapping bug vs FP8 inference failure on B200; different subsystems." + }, + { + "left": "issue:43937", + "right": "issue:44496", + "accept": false, + "reason": "Invalid generation config vs unrecognized model/config `model_type`; both are config validation issues but not the same underlying problem." + }, + { + "left": "issue:42371", + "right": "issue:46617", + "accept": false, + "reason": "TF32 settings guidance vs trust-remote-code cache FileNotFoundError; unrelated." + }, + { + "left": "issue:44410", + "right": "issue:45478", + "accept": false, + "reason": "Missing projections in qwen3next vs Qwen3.5 MoE `from_pretrained` error; separate model-specific bugs." + }, + { + "left": "issue:44261", + "right": "issue:46762", + "accept": false, + "reason": "Both mention numerical differences, but they are different models/implementations and the causes are not clearly the same." + }, + { + "left": "issue:44933", + "right": "issue:45278", + "accept": false, + "reason": "`image_utils` import problem may be one example of broader import breakage, but the artifacts do not clearly describe the same concrete bug." + }, + { + "left": "issue:38617", + "right": "issue:44220", + "accept": false, + "reason": "Missing import symbol vs `_torch_extract_fbank_features()` issue; unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 51, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:27957", + "issue:38617", + "issue:41628", + "issue:42222", + "issue:42757", + "issue:43012", + "issue:43066", + "issue:43097", + "issue:43122", + "issue:43352", + "issue:43441", + "issue:43452", + "issue:43531", + "issue:43577", + "issue:43673", + "issue:43874", + "issue:43931", + "issue:43986", + "issue:44038", + "issue:44075", + "issue:44117", + "issue:44261", + "issue:44263", + "issue:44360", + "issue:44410", + "issue:44483", + "issue:44484", + "issue:44485", + "issue:44493", + "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:45656", + "issue:45753", + "issue:46123", + "issue:46172", + "issue:46231", + "issue:46369" + ], + "soft_pairs": [ + "issue:44075|issue:45656", + "issue:44263|issue:44485", + "issue:43066|issue:43122", + "issue:43097|issue:44493", + "issue:43673|issue:45341", + "issue:41628|issue:45071", + "issue:42222|issue:45542", + "issue:43931|issue:44410", + "issue:45071|issue:45542", + "issue:44483|issue:45464", + "issue:44484|issue:45478", + "issue:43874|issue:44261", + "issue:43012|issue:43577", + "issue:44360|issue:46172", + "issue:44263|issue:46172", + "issue:27957|issue:43352", + "issue:43931|issue:44038", + "issue:43441|issue:44829", + "issue:43097|issue:43452", + "issue:45003|issue:45341", + "issue:43986|issue:45341", + "issue:44933|issue:46231", + "issue:44261|issue:44485", + "issue:45753|issue:46369", + "issue:44117|issue:44987", + "issue:43531|issue:45341", + "issue:43874|issue:44263", + "issue:43673|issue:46123", + "issue:45440|issue:46172", + "issue:43097|issue:45588", + "issue:43986|issue:45372", + "issue:38617|issue:42757", + "issue:44410|issue:45375" + ], + "prepared_review_unit_hash": "495d8368e6ed12bc415842324c028d55e8d73bead4247c650e91522b8263740c", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 33, + "serialized_chars": 37240, + "estimated_input_tokens": 9310, + "estimated_eval_tokens": 18876 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is mostly a grab bag of superficially similar Transformers bugs (imports, model loading, FlashAttention, Qwen/GLM variants) rather than true duplicates. I would not merge any of the suggested soft pairs.", + "confidence": 0.73, + "canonical_issue_reason": "If one issue must anchor the cluster, issue 43097 is the broadest regression-style report with downstream references and a clear v5 impact; the rest are narrower, model-specific failures.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43097 is the best representative because it describes a concrete, user-visible Transformers v5 regression in a general API area, rather than a highly model-specific or incidental bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44075", + "right": "issue:45656", + "accept": false, + "reason": "Both are optimizer-related, but one is about SGD args being ignored and the other about double optimizer steps under deepspeed; different bugs." + }, + { + "left": "issue:44263", + "right": "issue:44485", + "accept": false, + "reason": "Different GLM internals: torch.split return handling in a DSA indexer vs a RoPE implementation discussion." + }, + { + "left": "issue:43066", + "right": "issue:43122", + "accept": false, + "reason": "Both touch tokenizer behavior in v5, but one reports a wrong decoder type and the other a tokenization regression; not the same concrete bug." + }, + { + "left": "issue:43097", + "right": "issue:44493", + "accept": false, + "reason": "Unrelated regressions: removed tie_embeddings_and_encoder_decoder vs unexpected key errors on position ids." + }, + { + "left": "issue:43673", + "right": "issue:45341", + "accept": false, + "reason": "Chunked prefill cache handling is unrelated to a testing_utils bug." + }, + { + "left": "issue:41628", + "right": "issue:45071", + "accept": false, + "reason": "Importing AutoImageProcessor and PretrainedConfig type checking are different failure modes in different code paths." + }, + { + "left": "issue:42222", + "right": "issue:45542", + "accept": false, + "reason": "VitPose model breakage is unrelated to the TensorFlow/tensorboard backend detection issue." + }, + { + "left": "issue:43931", + "right": "issue:44410", + "accept": false, + "reason": "Both are Qwen-family issues, but one is a weight-shape mismatch on Qwen3-VL and the other is missing projections in qwen3next; different bugs." + }, + { + "left": "issue:45071", + "right": "issue:45542", + "accept": false, + "reason": "PretrainedConfig type checking and undefined tf backend errors are unrelated." + }, + { + "left": "issue:44483", + "right": "issue:45464", + "accept": false, + "reason": "One is a generic chat/completions API rejection, the other is a streaming inference failure on a specific Qwen model." + }, + { + "left": "issue:44484", + "right": "issue:45478", + "accept": false, + "reason": "Saving shard size behavior is unrelated to Qwen3.5-MoE from_pretrained loading errors." + }, + { + "left": "issue:43874", + "right": "issue:44261", + "accept": false, + "reason": "Different GLM issues: missing image patch count method vs missing rms_norm_eps in MLA layernorm." + }, + { + "left": "issue:43012", + "right": "issue:43577", + "accept": false, + "reason": "Both involve dtype/precision, but one is a PyTorch warning during compile and the other is incorrect dtype propagation in Blip2 loading." + }, + { + "left": "issue:44360", + "right": "issue:46172", + "accept": false, + "reason": "A missing ReLU in a DSA indexer is unrelated to the DeepSeek-v4 tensor transpose issue." + }, + { + "left": "issue:44263", + "right": "issue:46172", + "accept": false, + "reason": "torch.split return values in GlmMoeDsaIndexer and a DeepSeek-v4 comb tensor op are different bugs." + }, + { + "left": "issue:27957", + "right": "issue:43352", + "accept": false, + "reason": "Both mention Flash Attention 2, but they are separate model-specific support gaps for different architectures." + }, + { + "left": "issue:43931", + "right": "issue:44038", + "accept": false, + "reason": "Different Qwen variants and different symptoms; no evidence they share the same loading failure." + }, + { + "left": "issue:43441", + "right": "issue:44829", + "accept": false, + "reason": "One is a FlashAttention failure on Ministral-3, the other is degenerate training with flash_attention_3; different code paths and outcomes." + }, + { + "left": "issue:43097", + "right": "issue:43452", + "accept": false, + "reason": "Removed tie_embeddings support and gguf_file loading failures are unrelated." + }, + { + "left": "issue:45003", + "right": "issue:45341", + "accept": false, + "reason": "Unsafe sys.modules access in modeling_utils is unrelated to a testing_utils bug." + }, + { + "left": "issue:43986", + "right": "issue:45341", + "accept": false, + "reason": "Video-model loading without torchvision and a testing_utils issue are unrelated." + }, + { + "left": "issue:44933", + "right": "issue:46231", + "accept": false, + "reason": "A missing import from image_utils is unrelated to a documentation parameter-name mismatch." + }, + { + "left": "issue:44261", + "right": "issue:44485", + "accept": false, + "reason": "Both are GLM-related, but the former is a numerical epsilon/config issue and the latter is about RoPE implementation." + }, + { + "left": "issue:45753", + "right": "issue:46369", + "accept": false, + "reason": "An infinite-loop bug on a specific image is unrelated to memory overflow in sam3-litetext." + }, + { + "left": "issue:44117", + "right": "issue:44987", + "accept": false, + "reason": "A tokenizer mapping None handling bug is not the same as loading failure for a specific repo." + }, + { + "left": "issue:43531", + "right": "issue:45341", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior is unrelated to testing_utils." + }, + { + "left": "issue:43874", + "right": "issue:44263", + "accept": false, + "reason": "Both are GLM46V/GLM-related, but they concern different missing methods and different failure points." + }, + { + "left": "issue:43673", + "right": "issue:46123", + "accept": false, + "reason": "Generation cache missing in chunked_prefill is unrelated to MaskGenerationPipeline batch-finalization logic." + }, + { + "left": "issue:45440", + "right": "issue:46172", + "accept": false, + "reason": "DeepSeekV3MoE divergence from remote implementation is not the same as the DeepSeek-v4 comb dtype/transpose bug." + }, + { + "left": "issue:43097", + "right": "issue:45588", + "accept": false, + "reason": "Tie-embeddings removal and flash_attention.py crashing on s_aux=None are different regressions." + }, + { + "left": "issue:43986", + "right": "issue:45372", + "accept": false, + "reason": "Both involve loading processors, but one is a missing torchvision dependency and the other is a missing mistral_common symbol." + }, + { + "left": "issue:38617", + "right": "issue:42757", + "accept": false, + "reason": "Both are import errors, but they concern different modules and different missing symbols." + }, + { + "left": "issue:44410", + "right": "issue:45375", + "accept": false, + "reason": "Missing attention projections in qwen3next is unrelated to a missing deepstack_visual_indexes config field." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 52, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:38617", + "issue:42890", + "issue:42907", + "issue:43012", + "issue:43064", + "issue:43066", + "issue:43097", + "issue:43208", + "issue:43352", + "issue:43502", + "issue:43550", + "issue:43825", + "issue:43873", + "issue:43906", + "issue:43976", + "issue:43986", + "issue:43992", + "issue:44038", + "issue:44062", + "issue:44077", + "issue:44117", + "issue:44162", + "issue:44315", + "issue:44351", + "issue:44368", + "issue:44380", + "issue:44410", + "issue:44484", + "issue:44655", + "issue:44704", + "issue:44936", + "issue:44938", + "issue:45103", + "issue:45584", + "issue:45712", + "issue:45753", + "issue:46050", + "issue:46132", + "issue:46172", + "issue:46178", + "issue:46298", + "issue:46365", + "issue:46421", + "issue:46655", + "issue:46664", + "issue:46701", + "issue:46710", + "issue:46752" + ], + "soft_pairs": [ + "issue:43064|issue:44315", + "issue:43097|issue:46050", + "issue:44038|issue:44410", + "issue:43986|issue:44484", + "issue:46655|issue:46664", + "issue:46298|issue:46752", + "issue:42890|issue:46298", + "issue:38617|issue:44351", + "issue:46172|issue:46710", + "issue:43097|issue:43873", + "issue:44062|issue:44117", + "issue:44117|issue:46701", + "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:42907|issue:44655", + "issue:42907|issue:45753", + "issue:43825|issue:44655", + "issue:43208|issue:43550", + "issue:43012|issue:43550", + "issue:45712|issue:46298", + "issue:43012|issue:44368", + "issue:43992|issue:44484", + "issue:43066|issue:43906", + "issue:43352|issue:43976", + "issue:43352|issue:44380", + "issue:43097|issue:46298", + "issue:44162|issue:46172", + "issue:44380|issue:46132", + "issue:44077|issue:46365", + "issue:45584|issue:46365" + ], + "prepared_review_unit_hash": "223900239b6c1f853952f4fed5a6a948a0b3b7fee6a5fe81b321b1a5ae25bc44", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 34, + "serialized_chars": 37670, + "estimated_input_tokens": 9418, + "estimated_eval_tokens": 19092 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The items are a heterogeneous set of unrelated Transformers bugs/regressions, so none of the soft pairs look like true duplicates. No PRs were provided. If a representative issue is needed, the most central unresolved report is the quantization/offloading bug.", + "confidence": 0.44, + "canonical_issue_reason": "issue:43873 is the most active open, user-facing bug in the list and the broadest reasonable anchor among the unrelated reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43873 has the strongest combination of activity, open status, and general bug relevance, so it is the best representative issue here.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43064", + "right": "issue:44315", + "accept": false, + "reason": "Different failures: FSDP/PEFT optimizer state corruption vs Liger Kernel application during model_init." + }, + { + "left": "issue:43097", + "right": "issue:46050", + "accept": false, + "reason": "Different save/load problems: removed tie_embeddings_and_encoder_decoder vs quantized model round-trip breakage." + }, + { + "left": "issue:44038", + "right": "issue:44410", + "accept": false, + "reason": "Different models and symptoms: Qwen3-VL-Moe v5 bug vs qwen3next missing layer projections." + }, + { + "left": "issue:43986", + "right": "issue:44484", + "accept": false, + "reason": "Different code paths: AutoProcessor/video-model torchvision dependency crash vs save_pretrained shard-size default." + }, + { + "left": "issue:46655", + "right": "issue:46664", + "accept": false, + "reason": "Same subsystem but distinct bugs: padding math vs num_frames semantics in SmolVLM video processing." + }, + { + "left": "issue:46298", + "right": "issue:46752", + "accept": false, + "reason": "Noisy logging vs apply_chat_template IndexError are unrelated behaviors." + }, + { + "left": "issue:42890", + "right": "issue:46298", + "accept": false, + "reason": "Seed-related test flakiness is unrelated to an unspecified noisy logging message." + }, + { + "left": "issue:38617", + "right": "issue:44351", + "accept": false, + "reason": "Both are import errors, but for different symbols and modules with different root causes." + }, + { + "left": "issue:46172", + "right": "issue:46710", + "accept": false, + "reason": "Different models and failure modes: tensor transpose bug vs post-upgrade accuracy regression." + }, + { + "left": "issue:43097", + "right": "issue:43873", + "accept": false, + "reason": "Both mention v5-era regressions, but one is a removed config API and the other is quantization/offloading." + }, + { + "left": "issue:44062", + "right": "issue:44117", + "accept": false, + "reason": "Tokenizer AddedToken constructor error is unrelated to TOKENIZER_MAPPING_NAMES returning None." + }, + { + "left": "issue:44117", + "right": "issue:46701", + "accept": false, + "reason": "Tokenizer mapping lookup failure is unrelated to a diffusion/image-text pipeline crash." + }, + { + "left": "issue:44936", + "right": "issue:46365", + "accept": false, + "reason": "trainer.evaluate() after train() is a runtime state bug; importing Trainer depends on a package conflict." + }, + { + "left": "issue:43502", + "right": "issue:43992", + "accept": false, + "reason": "local_files_only network leakage is unrelated to missing embed_tokens.weight in UMT5Encoder loading." + }, + { + "left": "issue:44704", + "right": "issue:45103", + "accept": false, + "reason": "Both involve loading/doc plumbing, but cached_file kwarg forwarding and auto_docstring annotation parsing are separate." + }, + { + "left": "issue:44380", + "right": "issue:46421", + "accept": false, + "reason": "Attention scaling being ignored is unrelated to cache lazy initialization causing recompilation." + }, + { + "left": "issue:43986", + "right": "issue:44117", + "accept": false, + "reason": "Video-model torchvision import failure is unrelated to tokenizer mapping None handling." + }, + { + "left": "issue:44938", + "right": "issue:46178", + "accept": false, + "reason": "Python 3.14 load failure is a compatibility issue; tensorflow_text backend validation is a different backend-mapping bug." + }, + { + "left": "issue:42907", + "right": "issue:44655", + "accept": false, + "reason": "Both touch saving, but one is dequantized model persistence and the other is Pipeline.save_pretrained support." + }, + { + "left": "issue:42907", + "right": "issue:45753", + "accept": false, + "reason": "Dequantized save failure and Qwen3_5 image-triggered infinite loop are unrelated." + }, + { + "left": "issue:43825", + "right": "issue:44655", + "accept": false, + "reason": "Pipeline translation error messaging is unrelated to Pipeline.save_pretrained serialization support." + }, + { + "left": "issue:43208", + "right": "issue:43550", + "accept": false, + "reason": "xLSTM training blockers and Bamba torch.compile/SDPA failure are different model-specific issues." + }, + { + "left": "issue:43012", + "right": "issue:43550", + "accept": false, + "reason": "PyTorch bfloat16 warning is not the same bug as torch.compile failing with SDPA." + }, + { + "left": "issue:45712", + "right": "issue:46298", + "accept": false, + "reason": "Leftover dummy classes/check_repo failures are unrelated to a noisy logging message." + }, + { + "left": "issue:43012", + "right": "issue:44368", + "accept": false, + "reason": "Both are warnings, but they concern different configs and different warning sources." + }, + { + "left": "issue:43992", + "right": "issue:44484", + "accept": false, + "reason": "Missing checkpoint weights and max_shard_size default are not the same problem." + }, + { + "left": "issue:43066", + "right": "issue:43906", + "accept": false, + "reason": "Wrong tokenizer decoder type is unrelated to reproducing issue #38071." + }, + { + "left": "issue:43352", + "right": "issue:43976", + "accept": false, + "reason": "Flash Attention support gating is unrelated to a Python version compatibility failure." + }, + { + "left": "issue:43352", + "right": "issue:44380", + "accept": false, + "reason": "One is missing Flash Attention support; the other is GPT2 attention scaling being ignored under SDPA/FlashAttention." + }, + { + "left": "issue:43097", + "right": "issue:46298", + "accept": false, + "reason": "A removed config API and a noisy logging message do not share an underlying bug." + }, + { + "left": "issue:44162", + "right": "issue:46172", + "accept": false, + "reason": "ESM2 breakage is unrelated to the DeepSeek transpose bug." + }, + { + "left": "issue:44380", + "right": "issue:46132", + "accept": false, + "reason": "Ignored attention scaling and changed AttentionInterface.register behavior are different code-path problems." + }, + { + "left": "issue:44077", + "right": "issue:46365", + "accept": false, + "reason": "PatchTSMixer post_init policy and importing Trainer with datasets installed are unrelated." + }, + { + "left": "issue:45584", + "right": "issue:46365", + "accept": false, + "reason": "Whisper empty-transcription generation bug is unrelated to the Trainer import conflict." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 53, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:29870", + "issue:36010", + "issue:36296", + "issue:36683", + "issue:38617", + "issue:42175", + "issue:42490", + "issue:42757", + "issue:42994", + "issue:43012", + "issue:43064", + "issue:43208", + "issue:43299", + "issue:43352", + "issue:43475", + "issue:43502", + "issue:43525", + "issue:43526", + "issue:43582", + "issue:43611", + "issue:43618", + "issue:43756", + "issue:43761", + "issue:43824", + "issue:43867", + "issue:43901", + "issue:43937", + "issue:43976", + "issue:44038", + "issue:44077", + "issue:44222", + "issue:44230", + "issue:44380", + "issue:44483", + "issue:44485", + "issue:44514", + "issue:44625", + "issue:44655", + "issue:44683", + "issue:44938", + "issue:45216", + "issue:45276", + "issue:45478", + "issue:45479", + "issue:46077", + "issue:46082", + "issue:46365", + "issue:46369" + ], + "soft_pairs": [ + "issue:44222|issue:45478", + "issue:36683|issue:43582", + "issue:44380|issue:45479", + "issue:42175|issue:46365", + "issue:43064|issue:44380", + "issue:42757|issue:43824", + "issue:42490|issue:43012", + "issue:43012|issue:44683", + "issue:43208|issue:43526", + "issue:44483|issue:44514", + "issue:44038|issue:44230", + "issue:43618|issue:43761", + "issue:43618|issue:44380", + "issue:38617|issue:43502", + "issue:43756|issue:44485", + "issue:42994|issue:44655", + "issue:43299|issue:43867", + "issue:44625|issue:46365", + "issue:45276|issue:46077", + "issue:36683|issue:43475", + "issue:36010|issue:42757", + "issue:43352|issue:43937", + "issue:44483|issue:46369", + "issue:43611|issue:44655", + "issue:43901|issue:44655", + "issue:36683|issue:44077", + "issue:43867|issue:45216", + "issue:43976|issue:44938", + "issue:44625|issue:46077", + "issue:46077|issue:46082", + "issue:43618|issue:45216", + "issue:36683|issue:43525", + "issue:36683|issue:44625", + "issue:29870|issue:36296" + ], + "prepared_review_unit_hash": "038dbae25c8610cd449fffeb3b3f8d3f32131a97d615466b486d07d6d01abb5c", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 34, + "serialized_chars": 37644, + "estimated_input_tokens": 9411, + "estimated_eval_tokens": 19078 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most items are unrelated Transformers bug reports. The only soft edge that looks like the same underlying regression is the CLIP v5 output bug; the rest should stay separate.", + "confidence": 0.58, + "canonical_issue_reason": "issue:43761 is the best representative of the only clear duplicate cluster: it names the v5 CLIPVisionModel.forward regression directly and is broader than the attentions-only symptom.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43761 is the most actionable and umbrella-like report for the CLIP regression, with a concrete model/flag and a clear failure mode.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44222", + "right": "issue:45478", + "accept": false, + "reason": "Same Qwen3.5 MoE family, but one is FP8 save_pretrained and the other is from_pretrained loading; different code paths." + }, + { + "left": "issue:36683", + "right": "issue:43582", + "accept": false, + "reason": "Gemma3 config attribute bug vs Apple Silicon allocator warmup TypeError; unrelated subsystems." + }, + { + "left": "issue:44380", + "right": "issue:45479", + "accept": false, + "reason": "Attention-scaling backend bug and degenerate classification loss are different concrete failures." + }, + { + "left": "issue:42175", + "right": "issue:46365", + "accept": false, + "reason": "Packaging/backend install issue vs Trainer import shadowing; not the same bug." + }, + { + "left": "issue:43064", + "right": "issue:44380", + "accept": false, + "reason": "FSDP/PEFT optimizer-state corruption is unrelated to GPT2 attention scaling." + }, + { + "left": "issue:42757", + "right": "issue:43824", + "accept": false, + "reason": "Both are import errors, but for different symbols in different packages." + }, + { + "left": "issue:42490", + "right": "issue:43012", + "accept": false, + "reason": "Local path import warning and bfloat16 compile warning are distinct issues." + }, + { + "left": "issue:43012", + "right": "issue:44683", + "accept": false, + "reason": "One is a warning, the other a flex_attention compile failure; not the same bug." + }, + { + "left": "issue:43208", + "right": "issue:43526", + "accept": false, + "reason": "xLSTM training blockers and BeitImageProcessor label reduction are unrelated." + }, + { + "left": "issue:44483", + "right": "issue:44514", + "accept": false, + "reason": "Chat/completions request rejection and batched processor crash are different code paths." + }, + { + "left": "issue:44038", + "right": "issue:44230", + "accept": false, + "reason": "Qwen3-VL-Moe loading regression vs FP8 inference support request; not one underlying bug." + }, + { + "left": "issue:43618", + "right": "issue:43761", + "accept": true, + "reason": "Both point to the same CLIP v5 forward/output regression where expected model outputs are no longer populated." + }, + { + "left": "issue:38617", + "right": "issue:43502", + "accept": false, + "reason": "Missing import symbol and unwanted network access with local_files_only=True are unrelated." + }, + { + "left": "issue:43756", + "right": "issue:44485", + "accept": false, + "reason": "Different RoPE/model-implementation issues in different model families." + }, + { + "left": "issue:42994", + "right": "issue:44655", + "accept": false, + "reason": "Quantized model serialization failure vs Pipeline.save_pretrained support are different save paths." + }, + { + "left": "issue:43299", + "right": "issue:43867", + "accept": false, + "reason": "Different model-loading regressions: base_model_prefix handling vs sorted state_dict loading." + }, + { + "left": "issue:44625", + "right": "issue:46365", + "accept": false, + "reason": "Config propagation bug vs Trainer import failure; no shared underlying change." + }, + { + "left": "issue:45276", + "right": "issue:46077", + "accept": false, + "reason": "Gemma4 embedding resize bug and Gemma4 attention-mask bug affect different mechanisms." + }, + { + "left": "issue:36683", + "right": "issue:43475", + "accept": false, + "reason": "Gemma3 vocab_size AttributeError and SAM3 fpn_position_embeddings AttributeError are model-specific and unrelated." + }, + { + "left": "issue:36010", + "right": "issue:42757", + "accept": false, + "reason": "Different import regressions from different modules/symbols." + }, + { + "left": "issue:43352", + "right": "issue:43937", + "accept": false, + "reason": "Flash Attention support limitation and invalid GenerationConfig are unrelated." + }, + { + "left": "issue:44483", + "right": "issue:46369", + "accept": false, + "reason": "API request handling and SAM3 memory overflow are different problems." + }, + { + "left": "issue:43611", + "right": "issue:44655", + "accept": false, + "reason": "base_model_prefix loading regression is unrelated to saving Pipeline objects." + }, + { + "left": "issue:43901", + "right": "issue:44655", + "accept": false, + "reason": "Documentation mismatch vs pipeline serialization support; not a duplicate bug." + }, + { + "left": "issue:36683", + "right": "issue:44077", + "accept": false, + "reason": "Gemma3 config attribute bug and patchtsmixer post_init policy are unrelated." + }, + { + "left": "issue:43867", + "right": "issue:45216", + "accept": false, + "reason": "Both involve checkpoint loading/saving, but the symptoms and code paths differ too much to treat as one bug." + }, + { + "left": "issue:43976", + "right": "issue:44938", + "accept": false, + "reason": "Separate Python-version compatibility regressions." + }, + { + "left": "issue:44625", + "right": "issue:46077", + "accept": false, + "reason": "Qwen3.5 num_labels propagation and Gemma4 bidirectional masking are unrelated." + }, + { + "left": "issue:46077", + "right": "issue:46082", + "accept": false, + "reason": "Gemma4 attention-mask behavior and LlamaConfig head_dim validation are different issues." + }, + { + "left": "issue:43618", + "right": "issue:45216", + "accept": false, + "reason": "CLIP output regression and Qwen3.5 checkpoint regression are unrelated." + }, + { + "left": "issue:36683", + "right": "issue:43525", + "accept": false, + "reason": "Different config attribute errors on different model families." + }, + { + "left": "issue:36683", + "right": "issue:44625", + "accept": false, + "reason": "Different Qwen3.5/Gemma3 config propagation bugs." + }, + { + "left": "issue:29870", + "right": "issue:36296", + "accept": false, + "reason": "BART generation mask error and tensor-parallel training bug are unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The clustering is grounded in the packet and stays conservative overall. The single accepted edge, issue:43618 \u2194 issue:43761, is well-supported as the same CLIP v5 output-regression family, while the other pairs are reasonably kept separate based on distinct titles/code paths." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 54, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:28282", + "issue:29870", + "issue:33453", + "issue:36296", + "issue:36683", + "issue:41669", + "issue:42890", + "issue:42898", + "issue:43010", + "issue:43064", + "issue:43502", + "issue:43575", + "issue:43618", + "issue:43723", + "issue:43867", + "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:43010|issue:45278", + "issue:44075|issue:46365", + "issue:29870|issue:33453", + "issue:44355|issue:44855", + "issue:44938|issue:45071", + "issue:43867|issue:46202", + "issue:36683|issue:43064", + "issue:43867|issue:45375", + "issue:42890|issue:45561", + "issue:41669|issue:44246", + "issue:43010|issue:45850", + "issue:36296|issue:43575", + "issue:43618|issue:44077", + "issue:43010|issue:43976", + "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:43502|issue:44062", + "issue:44222|issue:45125", + "issue:44734|issue:46617", + "issue:43723|issue:44556" + ], + "prepared_review_unit_hash": "f650e39021f15998b3dacf93c207f3f6935a8e823270a9d8eb52edf56d2c61fc", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36272, + "estimated_input_tokens": 9068, + "estimated_eval_tokens": 18392 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a set of unrelated Transformers issues that only share broad themes like loading/import regressions, model-specific bugs, and test failures. I did not find any soft pair that clearly describes the same underlying bug or change, so all soft edges are rejected.", + "confidence": 0.84, + "canonical_issue_reason": "issue:45278 is the best umbrella candidate because it is the most general and directly captures the broad v5 upgrade/regression theme affecting multiple import/loading paths.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45278 is the most suitable representative issue for the cluster: it is broad enough to summarize the common v5 regression pattern without being overly model-specific.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43992", + "right": "issue:45310", + "accept": false, + "reason": "Both concern model-loading failures, but for different models and different failure modes; not the same underlying bug." + }, + { + "left": "issue:43064", + "right": "issue:44492", + "accept": false, + "reason": "Training optimizer-state corruption and a cache-strategy typo are unrelated." + }, + { + "left": "issue:43010", + "right": "issue:45278", + "accept": false, + "reason": "One is about adding no_grad to cache update methods; the other is a broad import-error complaint after upgrade." + }, + { + "left": "issue:44075", + "right": "issue:46365", + "accept": false, + "reason": "SGD args not being used and Trainer import failing due to datasets are different code paths and symptoms." + }, + { + "left": "issue:29870", + "right": "issue:33453", + "accept": false, + "reason": "Bart generate decoder mask initialization and tokenizer loading regression are distinct bugs." + }, + { + "left": "issue:44355", + "right": "issue:44855", + "accept": false, + "reason": "Both are import-related errors, but one is about compiled Python files and the other is a torch.jit decorator parsing issue." + }, + { + "left": "issue:44938", + "right": "issue:45071", + "accept": false, + "reason": "Python version import failure and PretrainedConfig type checking are not the same defect." + }, + { + "left": "issue:43867", + "right": "issue:46202", + "accept": false, + "reason": "A state_dict sorting load error and a Gemma pipeline roundtrip issue are unrelated." + }, + { + "left": "issue:36683", + "right": "issue:43064", + "accept": false, + "reason": "Gemma3Config missing vocab_size and optimizer-state corruption are unrelated." + }, + { + "left": "issue:43867", + "right": "issue:45375", + "accept": false, + "reason": "Different model-loading problems with different missing fields; not the same bug." + }, + { + "left": "issue:42890", + "right": "issue:45561", + "accept": false, + "reason": "Seed-related flaky integration tests and xdist file-race failures are different test infrastructure issues." + }, + { + "left": "issue:41669", + "right": "issue:44246", + "accept": false, + "reason": "Import-time performance complaints overlap broadly, but one is a proposal to remove import * and the other is a generic slow-import report; insufficient evidence of the same underlying issue." + }, + { + "left": "issue:43010", + "right": "issue:45850", + "accept": false, + "reason": "A no_grad cache-update fix and a generic 'breaks too many models' complaint are not the same concrete bug." + }, + { + "left": "issue:36296", + "right": "issue:43575", + "accept": false, + "reason": "Tensor-parallel training bug and tensor-parallel OOM on a specific model are different issues." + }, + { + "left": "issue:43618", + "right": "issue:44077", + "accept": false, + "reason": "CLIP attentions assignment and patchtsmixer post_init validation are unrelated model bugs." + }, + { + "left": "issue:43010", + "right": "issue:43976", + "accept": false, + "reason": "Cache no_grad decoration and a Python version support problem are unrelated." + }, + { + "left": "issue:43867", + "right": "issue:45357", + "accept": false, + "reason": "Different save/load regressions affecting different models and keys; not the same defect." + }, + { + "left": "issue:43992", + "right": "issue:45478", + "accept": false, + "reason": "Same general family of load errors, but distinct models and failures; not mergeable as one bug fix." + }, + { + "left": "issue:28282", + "right": "issue:33453", + "accept": false, + "reason": "Missing PyTorch import and tokenizer loading regression have no direct overlap." + }, + { + "left": "issue:43618", + "right": "issue:46285", + "accept": false, + "reason": "CLIP attentions assignment and CLIP text-model structure inconsistency are different changes." + }, + { + "left": "issue:43618", + "right": "issue:44857", + "accept": false, + "reason": "CLIP output fields and LwDetr AMP crash are unrelated." + }, + { + "left": "issue:44360", + "right": "issue:45440", + "accept": false, + "reason": "DSA indexer discussion and DeepSeekV3 implementation divergence are different topics." + }, + { + "left": "issue:42890", + "right": "issue:44964", + "accept": false, + "reason": "A flaky test missing seeds and a multimodal model load failure are unrelated." + }, + { + "left": "issue:42890", + "right": "issue:44112", + "accept": false, + "reason": "These are different CI/test failures with different causes." + }, + { + "left": "issue:42890", + "right": "issue:45600", + "accept": false, + "reason": "A seed-related flaky test and broken auto_mappings references are unrelated CI issues." + }, + { + "left": "issue:42898", + "right": "issue:46664", + "accept": false, + "reason": "Tokenization cleanup behavior and video processor frame semantics are unrelated." + }, + { + "left": "issue:43502", + "right": "issue:44062", + "accept": false, + "reason": "Offline-loading API leakage and AddedToken constructor duplication are unrelated tokenizer/model bugs." + }, + { + "left": "issue:44222", + "right": "issue:45125", + "accept": false, + "reason": "FP8 save_pretrained and missing tensor-parallel plan are separate Qwen3.5 MoE issues." + }, + { + "left": "issue:44734", + "right": "issue:46617", + "accept": false, + "reason": "KV-cache continuation indexing and trust_remote_code cache loading failure are different code paths." + }, + { + "left": "issue:43723", + "right": "issue:44556", + "accept": false, + "reason": "Tokenizer-from-pretrained regression and checkpoint reload compatibility after upgrade are related only broadly to loading, not the same specific bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 55, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:29870", + "issue:30333", + "issue:30990", + "issue:31795", + "issue:33453", + "issue:34689", + "issue:36296", + "issue:38617", + "issue:41628", + "issue:42548", + "issue:42757", + "issue:43010", + "issue:43065", + "issue:43232", + "issue:43299", + "issue:43475", + "issue:43502", + "issue:43576", + "issue:43824", + "issue:43827", + "issue:43874", + "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:44961", + "issue:45071", + "issue:45125", + "issue:45375", + "issue:45468", + "issue:45478", + "issue:46132", + "issue:46143", + "issue:46144", + "issue:46209", + "issue:46617", + "issue:46664", + "issue:46726" + ], + "soft_pairs": [ + "issue:44387|issue:44556", + "issue:43824|issue:46617", + "issue:43010|issue:43065", + "issue:43874|issue:46617", + "issue:43502|issue:44485", + "issue:43576|issue:46132", + "issue:45125|issue:45478", + "issue:30333|issue:33453", + "issue:44355|issue:44749", + "issue:42548|issue:44351", + "issue:45125|issue:45375", + "issue:46664|issue:46726", + "issue:43010|issue:43232", + "issue:30990|issue:31795", + "issue:43502|issue:44961", + "issue:43502|issue:44222", + "issue:43827|issue:44961", + "issue:30333|issue:36296", + "issue:29870|issue:31795", + "issue:43299|issue:44556", + "issue:44355|issue:45071", + "issue:33453|issue:34689", + "issue:44355|issue:44655", + "issue:42757|issue:43576", + "issue:43576|issue:45071", + "issue:44315|issue:44360", + "issue:31795|issue:38617", + "issue:43576|issue:43824", + "issue:44961|issue:45468", + "issue:44351|issue:46617", + "issue:44485|issue:44961", + "issue:44908|issue:44961", + "issue:38617|issue:43576", + "issue:36296|issue:43901", + "issue:46143|issue:46144", + "issue:43475|issue:46617", + "issue:44222|issue:44230", + "issue:41628|issue:42548", + "issue:44315|issue:46209", + "issue:44485|issue:46664" + ], + "prepared_review_unit_hash": "8eab4fca79db1345bc8e625ed7ba8837c657e8849e73cdeca1a4bdea137debba", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 40, + "serialized_chars": 38595, + "estimated_input_tokens": 9649, + "estimated_eval_tokens": 19554 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Loose cluster of largely unrelated Transformers issues, mostly around v5 regressions, loading/import failures, and a few model-specific bugs. No soft-edge pair looks like the same underlying bug/change, so all soft links are rejected.", + "confidence": 0.62, + "canonical_issue_reason": "issue:44556 is the clearest representative of the cluster\u2019s recurring theme: a concrete v4.57\u2192v5.x reload regression affecting users upgrading checkpoints.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44556 is the best global representative because it is broad, user-facing, and directly captures the common upgrade-breakage pattern seen across the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44387", + "right": "issue:44556", + "accept": false, + "reason": "Both mention v5 breakage, but one is int4 OOM/reserved-memory behavior and the other is checkpoint reload failure; different bugs and code paths." + }, + { + "left": "issue:43824", + "right": "issue:46617", + "accept": false, + "reason": "One is a missing exported model class import error, the other is a cache FileNotFoundError for trust_remote_code; unrelated failure modes." + }, + { + "left": "issue:43010", + "right": "issue:43065", + "accept": false, + "reason": "Cache no_grad decoration and Sam3PixelDecoder dummy Conv2d are unrelated implementation issues." + }, + { + "left": "issue:43874", + "right": "issue:46617", + "accept": false, + "reason": "One is a missing image-processor method causing token counting to fail; the other is a cache loading FileNotFoundError." + }, + { + "left": "issue:43502", + "right": "issue:44485", + "accept": false, + "reason": "Local-files-only network leakage and GLM-5 RoPE implementation are different subsystems and bugs." + }, + { + "left": "issue:43576", + "right": "issue:46132", + "accept": false, + "reason": "A broken CLI command and a changed attention registration behavior are not the same underlying issue." + }, + { + "left": "issue:45125", + "right": "issue:45478", + "accept": false, + "reason": "Both involve Qwen3.5 MoE, but one is missing tensor-parallel plan metadata and the other is a from_pretrained load error; not enough evidence they are the same bug." + }, + { + "left": "issue:30333", + "right": "issue:33453", + "accept": false, + "reason": "MLFlow reporting on failed jobs is unrelated to tokenizer-loading regression." + }, + { + "left": "issue:44355", + "right": "issue:44749", + "accept": false, + "reason": "Compiled Python file errors and slower data filtering after upgrade are unrelated." + }, + { + "left": "issue:42548", + "right": "issue:44351", + "accept": false, + "reason": "Both are import errors, but for different symbols and likely different API breakages." + }, + { + "left": "issue:45125", + "right": "issue:45375", + "accept": false, + "reason": "Tensor-parallel plan metadata and missing config field in a different model are separate issues." + }, + { + "left": "issue:46664", + "right": "issue:46726", + "accept": false, + "reason": "Video num_frames semantics and missing-vision-backend AutoProcessor handling are different problems." + }, + { + "left": "issue:43010", + "right": "issue:43232", + "accept": false, + "reason": "Decorator placement for cache updates is unrelated to generation kwargs handling after sync_gpus." + }, + { + "left": "issue:30990", + "right": "issue:31795", + "accept": false, + "reason": "SentenceTransformers loading stall and forward-doc confusion are not the same bug." + }, + { + "left": "issue:43502", + "right": "issue:44961", + "accept": false, + "reason": "Remote API calls despite local_files_only and an unrelated placeholder issue are not the same." + }, + { + "left": "issue:43502", + "right": "issue:44222", + "accept": false, + "reason": "Network access with local_files_only and FP8 save_pretrained MoE are unrelated." + }, + { + "left": "issue:43827", + "right": "issue:44961", + "accept": false, + "reason": "Docs still referencing pipeline() and an unrelated placeholder issue do not match." + }, + { + "left": "issue:30333", + "right": "issue:36296", + "accept": false, + "reason": "MLFlow failure reporting and tensor-parallel training bug are distinct." + }, + { + "left": "issue:29870", + "right": "issue:31795", + "accept": false, + "reason": "A generation-time UnboundLocalError in BART and a docs confusion issue are different." + }, + { + "left": "issue:43299", + "right": "issue:44556", + "accept": false, + "reason": "Qwen3VL/MoE loading breakage and generic checkpoint reload regression are both v5-related but not the same concrete bug." + }, + { + "left": "issue:44355", + "right": "issue:45071", + "accept": false, + "reason": "Compiled-file errors and PretrainedConfig type-checking regression are unrelated." + }, + { + "left": "issue:33453", + "right": "issue:34689", + "accept": false, + "reason": "Tokenizer-loading regression and Llama 3.2 Vision model-loading breakage are separate loading problems." + }, + { + "left": "issue:44355", + "right": "issue:44655", + "accept": false, + "reason": "Compiled Python file errors and Pipeline.save_pretrained failure are different code paths." + }, + { + "left": "issue:42757", + "right": "issue:43576", + "accept": false, + "reason": "Hugging Face Hub offline-mode import error and broken env command are unrelated." + }, + { + "left": "issue:43576", + "right": "issue:45071", + "accept": false, + "reason": "CLI env command breakage and config type-checking regression are different issues." + }, + { + "left": "issue:44315", + "right": "issue:44360", + "accept": false, + "reason": "Liger kernel not applied with model_init and a DSA indexer ReLU discussion are unrelated." + }, + { + "left": "issue:31795", + "right": "issue:38617", + "accept": false, + "reason": "Forward-argument docs confusion and a missing configuration_utils import are different concerns." + }, + { + "left": "issue:43576", + "right": "issue:43824", + "accept": false, + "reason": "Broken env command and missing model import are unrelated." + }, + { + "left": "issue:44961", + "right": "issue:45468", + "accept": false, + "reason": "Placeholder issue and Gemma-4 audio positional encoding bug have no evident relation." + }, + { + "left": "issue:44351", + "right": "issue:46617", + "accept": false, + "reason": "Importing HybridCache and trust_remote_code cache FileNotFoundError are different issues." + }, + { + "left": "issue:44485", + "right": "issue:44961", + "accept": false, + "reason": "GLM-5 RoPE implementation and placeholder issue are unrelated." + }, + { + "left": "issue:44908", + "right": "issue:44961", + "accept": false, + "reason": "Scheduler kwargs handling and placeholder issue are unrelated." + }, + { + "left": "issue:38617", + "right": "issue:43576", + "accept": false, + "reason": "Missing internal import and broken env command are different bug reports." + }, + { + "left": "issue:36296", + "right": "issue:43901", + "accept": false, + "reason": "Tensor-parallel training bug and docs mentioning return_all_scores are unrelated." + }, + { + "left": "issue:46143", + "right": "issue:46144", + "accept": false, + "reason": "Dropped kwargs in RoFormer methods and attention-interface integration are related area but not the same concrete bug." + }, + { + "left": "issue:43475", + "right": "issue:46617", + "accept": false, + "reason": "SAM3 video attribute error and trust_remote_code cache FileNotFoundError are unrelated." + }, + { + "left": "issue:44222", + "right": "issue:44230", + "accept": false, + "reason": "FP8 save_pretrained bug and FP8 inference support request are adjacent but distinct issues." + }, + { + "left": "issue:41628", + "right": "issue:42548", + "accept": false, + "reason": "AutoImageProcessor import failure and PreTrainedModel import failure are different symbols and likely different regressions." + }, + { + "left": "issue:44315", + "right": "issue:46209", + "accept": false, + "reason": "Liger Kernel/model_init behavior and FP8 inference on B200 are unrelated." + }, + { + "left": "issue:44485", + "right": "issue:46664", + "accept": false, + "reason": "GLM-5 RoPE implementation and SmolVLM num_frames semantics are different model-specific bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 56, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "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:45468", + "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:44273|issue:45542", + "issue:44360|issue:46664", + "issue:44623|issue:45230", + "issue:44038|issue:45125", + "issue:44273|issue:45003", + "issue:44393|issue:45071", + "issue:44393|issue:45753", + "issue:41669|issue:43673", + "issue:43576|issue:44556", + "issue:43525|issue:45850", + "issue:44556|issue:45092", + "issue:43572|issue:45850", + "issue:44492|issue:45850", + "issue:45278|issue:45850", + "issue:42371|issue:44492", + "issue:44230|issue:46095", + "issue:41669|issue:46218", + "issue:43408|issue:44315", + "issue:42890|issue:44908", + "issue:44246|issue:44273", + "issue:43576|issue:44908", + "issue:43352|issue:44393", + "issue:43976|issue:44393", + "issue:44704|issue:44908", + "issue:44908|issue:45071", + "issue:44297|issue:45230", + "issue:44295|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:43576|issue:44273", + "issue:44315|issue:44829", + "issue:44360|issue:45468" + ], + "prepared_review_unit_hash": "45c0c2225e3bdc12de609ef0632aa821cff595f23e2e86777dcf72026108bb6c", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 41, + "serialized_chars": 39404, + "estimated_input_tokens": 9851, + "estimated_eval_tokens": 19958 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly a loose collection of v5-regression and model-specific bug reports, not a set of true duplicates. I found no soft pairs that look safe to merge.", + "confidence": 0.42, + "canonical_issue_reason": "issue:45850 is the broadest umbrella report in the set, but it is still only a loose theme match rather than a true duplicate nucleus.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45850 is the best global representative because it captures the widest version-regression complaint, even though the surrounding issues are too heterogeneous to collapse into one duplicate.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45362", + "right": "issue:45850", + "accept": false, + "reason": "Specific Qwen3.5 chat crash vs a broad \"many models broken\" complaint; not the same code-path." + }, + { + "left": "issue:43673", + "right": "issue:44492", + "accept": false, + "reason": "Generation chunked_prefill cache regression vs a typo in cache-strategy naming; different bugs." + }, + { + "left": "issue:44273", + "right": "issue:45542", + "accept": false, + "reason": "Lazy loading failure vs TensorFlow backend error from tensorboard-only install; unrelated symptoms." + }, + { + "left": "issue:44360", + "right": "issue:46664", + "accept": false, + "reason": "DSA indexer/ReLU discussion is unrelated to SmolVLM video num_frames semantics." + }, + { + "left": "issue:44623", + "right": "issue:45230", + "accept": false, + "reason": "Missing processor.save_pretrained files is too specific to merge with a generic bug report." + }, + { + "left": "issue:44038", + "right": "issue:45125", + "accept": false, + "reason": "Qwen3-VL-Moe support/inference issue vs missing _tp_plan for tensor parallelism; related model family but different fixes." + }, + { + "left": "issue:44273", + "right": "issue:45003", + "accept": false, + "reason": "Lazy loading problem vs unsafe sys.modules access in modeling_utils; different underlying failures." + }, + { + "left": "issue:44393", + "right": "issue:45071", + "accept": false, + "reason": "Qwen3-VL bbox hallucination/error vs PretrainedConfig type-checking regression; not the same bug." + }, + { + "left": "issue:44393", + "right": "issue:45753", + "accept": false, + "reason": "Both touch Qwen3-VL-ish models, but one is bbox output errors and the other is an infinite-loop image case." + }, + { + "left": "issue:41669", + "right": "issue:43673", + "accept": false, + "reason": "Import-time slowdown from import * removal is unrelated to the missing GenerationMixin cache issue." + }, + { + "left": "issue:43576", + "right": "issue:44556", + "accept": false, + "reason": "Broken transformers env command vs checkpoint reload incompatibility after upgrade; different command paths." + }, + { + "left": "issue:43525", + "right": "issue:45850", + "accept": false, + "reason": "Missing pad_token_id on Llama4Config is a specific config bug, not the broad regression theme." + }, + { + "left": "issue:44556", + "right": "issue:45092", + "accept": false, + "reason": "Both are upgrade-related, but one breaks checkpoint reload and the other breaks meta initialization for remote-code checkpoints." + }, + { + "left": "issue:43572", + "right": "issue:45850", + "accept": false, + "reason": "StableLmConfig pad_token_idx regression is model-specific and distinct from the broad \"many models\" complaint." + }, + { + "left": "issue:44492", + "right": "issue:45850", + "accept": false, + "reason": "Cache-strategy typo is documentation/label-level, not the same as general v5 breakage." + }, + { + "left": "issue:45278", + "right": "issue:45850", + "accept": false, + "reason": "General import errors after upgrade are broader than the specific model-breakage report, but still not the same underlying bug." + }, + { + "left": "issue:42371", + "right": "issue:44492", + "accept": false, + "reason": "TF32 behavior API guidance is unrelated to cache-strategy typo." + }, + { + "left": "issue:44230", + "right": "issue:46095", + "accept": false, + "reason": "FP8 inference support request vs weight-conversion-script question; different goals and code paths." + }, + { + "left": "issue:41669", + "right": "issue:46218", + "accept": false, + "reason": "Import slowdown from lazy-loading changes vs an actual missing module import error." + }, + { + "left": "issue:43408", + "right": "issue:44315", + "accept": false, + "reason": "Model-type warning during instantiation vs Liger Kernel not applied with model_init; unrelated." + }, + { + "left": "issue:42890", + "right": "issue:44908", + "accept": false, + "reason": "Test flakiness from missing set_seed calls vs scheduler kwargs not being passed." + }, + { + "left": "issue:44246", + "right": "issue:44273", + "accept": false, + "reason": "Slow import and lazy loading issues are related broadly, but the reported failures and likely fixes differ." + }, + { + "left": "issue:43576", + "right": "issue:44908", + "accept": false, + "reason": "Broken env command and inverse_sqrt scheduler kwargs are unrelated." + }, + { + "left": "issue:43352", + "right": "issue:44393", + "accept": false, + "reason": "Flash Attention 2 support check for Nemotron vs Qwen3-VL bbox output bug; different models and code paths." + }, + { + "left": "issue:43976", + "right": "issue:44393", + "accept": false, + "reason": "Python version compatibility regression vs model output hallucination; no shared bug." + }, + { + "left": "issue:44704", + "right": "issue:44908", + "accept": false, + "reason": "AutoProcessor cached_file kwargs passthrough vs scheduler kwargs handling; different subsystems." + }, + { + "left": "issue:44908", + "right": "issue:45071", + "accept": false, + "reason": "Scheduler kwargs bug vs PretrainedConfig type-checking regression." + }, + { + "left": "issue:44297", + "right": "issue:45230", + "accept": false, + "reason": "Tokenizer save_pretrained mismatch is specific and not comparable to a generic bug report." + }, + { + "left": "issue:44295", + "right": "issue:45230", + "accept": false, + "reason": "position_ids buffer access error is a concrete model bug, not a generic placeholder issue." + }, + { + "left": "issue:41669", + "right": "issue:44351", + "accept": false, + "reason": "Import performance regression vs cannot-import HybridCache; distinct failure modes." + }, + { + "left": "issue:44492", + "right": "issue:46328", + "accept": false, + "reason": "Cache-strategy typo and SequenceFeatureExtractor.pad performance issue are unrelated." + }, + { + "left": "issue:41669", + "right": "issue:45341", + "accept": false, + "reason": "Import-* performance regression vs a testing_utils bug; not the same underlying change." + }, + { + "left": "issue:43976", + "right": "issue:45850", + "accept": false, + "reason": "Python support regression is different from the broad v5 breakage complaint." + }, + { + "left": "issue:44393", + "right": "issue:45478", + "accept": false, + "reason": "Qwen3-VL bbox output bug vs Qwen3.5-Moe from_pretrained error; different model/features." + }, + { + "left": "issue:43976", + "right": "issue:46218", + "accept": false, + "reason": "Python version compatibility issue vs module import failure for Qwen3_5ForCausalLM." + }, + { + "left": "issue:44077", + "right": "issue:44861", + "accept": false, + "reason": "post_init allowance change vs tied-weight-key AttributeError; different fixes." + }, + { + "left": "issue:43827", + "right": "issue:44908", + "accept": false, + "reason": "Docs still referencing pipeline() vs scheduler kwargs bug; unrelated." + }, + { + "left": "issue:31795", + "right": "issue:33453", + "accept": false, + "reason": "Forward-API documentation confusion vs tokenizer loading regression; not the same issue." + }, + { + "left": "issue:43576", + "right": "issue:44273", + "accept": false, + "reason": "Broken env command vs lazy loading not working properly; only loosely related to v5 behavior." + }, + { + "left": "issue:44315", + "right": "issue:44829", + "accept": false, + "reason": "Liger Kernel not applied with model_init vs degenerate training under flash_attention_3; different code paths and fixes." + }, + { + "left": "issue:44360", + "right": "issue:45468", + "accept": false, + "reason": "DSA/ReLU discussion is unrelated to Gemma4 audio positional encoding." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 57, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:28348", + "issue:29870", + "issue:30333", + "issue:31515", + "issue:31795", + "issue:34689", + "issue:36296", + "issue:38617", + "issue:41669", + "issue:42371", + "issue:42913", + "issue:43066", + "issue:43298", + "issue:43352", + "issue:43502", + "issue:43519", + "issue:43525", + "issue:43572", + "issue:43576", + "issue:43606", + "issue:43618", + "issue:43704", + "issue:43717", + "issue:44230", + "issue:44261", + "issue:44263", + "issue:44273", + "issue:44351", + "issue:44485", + "issue:44492", + "issue:44623", + "issue:44655", + "issue:44861", + "issue:44908", + "issue:44938", + "issue:44998", + "issue:45230", + "issue:45250", + "issue:45290", + "issue:45310", + "issue:45468", + "issue:45850", + "issue:46018", + "issue:46095", + "issue:46144", + "issue:46172", + "issue:46218", + "issue:46655" + ], + "soft_pairs": [ + "issue:43519|issue:46655", + "issue:44273|issue:44908", + "issue:44655|issue:44861", + "issue:43576|issue:46218", + "issue:45230|issue:45310", + "issue:43519|issue:44485", + "issue:43519|issue:44908", + "issue:41669|issue:43704", + "issue:31795|issue:34689", + "issue:46018|issue:46172", + "issue:43704|issue:44485", + "issue:44485|issue:45468", + "issue:43502|issue:43519", + "issue:42371|issue:43704", + "issue:45850|issue:46218", + "issue:43519|issue:44623", + "issue:43606|issue:45468", + "issue:44263|issue:45468", + "issue:44261|issue:45468", + "issue:43298|issue:46095", + "issue:43519|issue:43704", + "issue:44351|issue:46218", + "issue:43352|issue:46218", + "issue:42913|issue:43066", + "issue:28348|issue:45250", + "issue:44908|issue:44998", + "issue:44492|issue:44998", + "issue:43618|issue:44998", + "issue:44998|issue:45850", + "issue:44938|issue:44998", + "issue:44861|issue:45290", + "issue:31515|issue:38617", + "issue:43525|issue:44861", + "issue:43572|issue:44861", + "issue:44230|issue:44623", + "issue:28348|issue:43352", + "issue:29870|issue:31515", + "issue:30333|issue:31515", + "issue:31515|issue:36296", + "issue:31515|issue:31795", + "issue:44230|issue:45310", + "issue:43298|issue:44230", + "issue:43717|issue:46144" + ], + "prepared_review_unit_hash": "22a49405135aebb24679d7c3bfc6daab5c0edbe0497faad8608658da68fccecd", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 43, + "serialized_chars": 39684, + "estimated_input_tokens": 9921, + "estimated_eval_tokens": 20098 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is heterogeneous and mostly contains unrelated model-loading, tokenizer, processor, and training/runtime bugs. There is no strong duplicate pair; the closest umbrella issue is the broad v5-regression complaint, but it is still not a true canonical duplicate set.", + "confidence": 0.82, + "canonical_issue_reason": "issue:45850 is the broadest umbrella topic in the set (\u201cSince 5.0 version it breaks too many models\u201d) and loosely matches several v5-regression reports, though the cluster is still not a real duplicate group.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45850 is the most representative/high-level issue for this noisy cluster, but only as a loose umbrella rather than a true duplicate anchor.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43519", + "right": "issue:46655", + "accept": false, + "reason": "Different bugs: Qwen3VL timestamp calculation vs video padding logic in SmolVLM." + }, + { + "left": "issue:44273", + "right": "issue:44908", + "accept": false, + "reason": "Lazy loading failure is unrelated to inverse_sqrt scheduler kwargs handling." + }, + { + "left": "issue:44655", + "right": "issue:44861", + "accept": false, + "reason": "Saving Pipeline objects and tied-weight-key crashes are different code paths." + }, + { + "left": "issue:43576", + "right": "issue:46218", + "accept": false, + "reason": "Broken env command is unrelated to importing Qwen3_5ForCausalLM." + }, + { + "left": "issue:45230", + "right": "issue:45310", + "accept": false, + "reason": "Generic bug report vs a specific Qwen3.5 MoE from_pretrained error; not the same issue." + }, + { + "left": "issue:43519", + "right": "issue:44485", + "accept": false, + "reason": "Timestamp calculation in Qwen3VL is unrelated to GLM-5 RoPE implementation." + }, + { + "left": "issue:43519", + "right": "issue:44908", + "accept": false, + "reason": "Video timestamp bug and scheduler kwargs bug are different subsystems." + }, + { + "left": "issue:41669", + "right": "issue:43704", + "accept": false, + "reason": "Import-star cleanup/perf issue is unrelated to a Qwen3ForCausalLM VRAM leak." + }, + { + "left": "issue:31795", + "right": "issue:34689", + "accept": false, + "reason": "Documentation confusion about inputs/past_key_values is not the same as a model-loading regression." + }, + { + "left": "issue:46018", + "right": "issue:46172", + "accept": false, + "reason": "DeepSeek-V4 gating logic and a tensor transpose in weight conversion are different defects." + }, + { + "left": "issue:43704", + "right": "issue:44485", + "accept": false, + "reason": "VRAM leak in Qwen3ForCausalLM is unrelated to GLM-5 RoPE math." + }, + { + "left": "issue:44485", + "right": "issue:45468", + "accept": false, + "reason": "Two separate model families and two different positional-encoding bugs." + }, + { + "left": "issue:43502", + "right": "issue:43519", + "accept": false, + "reason": "local_files_only network requests do not match timestamp miscalculation." + }, + { + "left": "issue:42371", + "right": "issue:43704", + "accept": false, + "reason": "TF32 API guidance is unrelated to the Qwen3ForCausalLM memory leak." + }, + { + "left": "issue:45850", + "right": "issue:46218", + "accept": false, + "reason": "Broad v5 breakage complaint is not the same as the Qwen3_5 import error." + }, + { + "left": "issue:43519", + "right": "issue:44623", + "accept": false, + "reason": "Processor save_pretrained file omission is unrelated to Qwen3VL timestamp math." + }, + { + "left": "issue:43606", + "right": "issue:45468", + "accept": false, + "reason": "CPU offload device mismatch and Gemma-4 audio positional encoding are different failures." + }, + { + "left": "issue:44263", + "right": "issue:45468", + "accept": false, + "reason": "GlmMoeDsaIndexer split return values is unrelated to Gemma-4 audio positional encoding." + }, + { + "left": "issue:44261", + "right": "issue:45468", + "accept": false, + "reason": "Missing rms_norm_eps in MLA q_a_layernorm is not the same as Gemma-4 audio RoPE/positional encoding." + }, + { + "left": "issue:43298", + "right": "issue:46095", + "accept": false, + "reason": "Continuous batching audio-model support and DeepSeek-V4 weight conversion are unrelated." + }, + { + "left": "issue:43519", + "right": "issue:43704", + "accept": false, + "reason": "Timestamp calculation bug and VRAM leak are distinct issues." + }, + { + "left": "issue:44351", + "right": "issue:46218", + "accept": false, + "reason": "HybridCache import failure is a different missing-symbol problem than Qwen3_5ForCausalLM import failure." + }, + { + "left": "issue:43352", + "right": "issue:46218", + "accept": false, + "reason": "Flash Attention support for NemotronH and a Qwen3_5 import error are unrelated." + }, + { + "left": "issue:42913", + "right": "issue:43066", + "accept": false, + "reason": "Both involve v5 tokenizer behavior, but one is a broad behavior regression and the other a specific decoder-type bug; not enough to merge." + }, + { + "left": "issue:28348", + "right": "issue:45250", + "accept": false, + "reason": "GPT2LMHeadModel FlashAttention-2 support request is not the same as a generic Flash Attention 2.0 issue." + }, + { + "left": "issue:44908", + "right": "issue:44998", + "accept": false, + "reason": "Scheduler kwargs bug is unrelated to the placeholder/garbage issue title." + }, + { + "left": "issue:44492", + "right": "issue:44998", + "accept": false, + "reason": "Cache-strategy typo is unrelated to the unrelated title noise." + }, + { + "left": "issue:43618", + "right": "issue:44998", + "accept": false, + "reason": "CLIPOutput attentions regression is unrelated to the unrelated title noise." + }, + { + "left": "issue:44998", + "right": "issue:45850", + "accept": false, + "reason": "Placeholder/unemployment issue is not the same as the broad v5 breakage report." + }, + { + "left": "issue:44938", + "right": "issue:44998", + "accept": false, + "reason": "Python 3.14 load failure is unrelated to the unrelated title noise." + }, + { + "left": "issue:44861", + "right": "issue:45290", + "accept": false, + "reason": "Tied-weight key crash and apply_chat_template tool-call crash are different code paths." + }, + { + "left": "issue:31515", + "right": "issue:38617", + "accept": false, + "reason": "Slow from_pretrained checkpoint loading is unrelated to the layer_type_validation import error." + }, + { + "left": "issue:43525", + "right": "issue:44861", + "accept": false, + "reason": "Missing pad_token_id on Llama4Config is unrelated to tied-weight-key handling." + }, + { + "left": "issue:43572", + "right": "issue:44861", + "accept": false, + "reason": "Missing pad_token_idx in StableLmConfig is unrelated to tied-weight-key handling." + }, + { + "left": "issue:44230", + "right": "issue:44623", + "accept": false, + "reason": "FP8 inference support request is unrelated to processor.save_pretrained missing files." + }, + { + "left": "issue:28348", + "right": "issue:43352", + "accept": false, + "reason": "Both mention Flash Attention 2.0, but they target different models and changes; not one fix." + }, + { + "left": "issue:29870", + "right": "issue:31515", + "accept": false, + "reason": "Bart generate() decoder_attention_mask bug is unrelated to slow checkpoint loading." + }, + { + "left": "issue:30333", + "right": "issue:31515", + "accept": false, + "reason": "MLflow reporting on failed jobs is unrelated to from_pretrained performance." + }, + { + "left": "issue:31515", + "right": "issue:36296", + "accept": false, + "reason": "Slow checkpoint loading and tensor-parallel training bug are unrelated." + }, + { + "left": "issue:31515", + "right": "issue:31795", + "accept": false, + "reason": "Performance loading issue and documentation confusion are unrelated." + }, + { + "left": "issue:44230", + "right": "issue:45310", + "accept": false, + "reason": "FP8 support request and Qwen3.5 MoE from_pretrained error are not the same concrete bug." + }, + { + "left": "issue:43298", + "right": "issue:44230", + "accept": false, + "reason": "Audio continuous-batching support and Qwen3-VL/Qwen3.5 FP8 support are different feature areas." + }, + { + "left": "issue:43717", + "right": "issue:46144", + "accept": false, + "reason": "Mamba init_weights discussion is unrelated to RoFormer attention interface usage." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 23, + "review_unit_index": 58, + "review_unit_count": 58, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-483", + "nodes": [ + "issue:37934", + "issue:42936", + "issue:43240", + "issue:43668", + "issue:43813", + "issue:44031", + "issue:46018", + "issue:46144" + ], + "soft_pairs": [ + "issue:42936|issue:46018", + "issue:43668|issue:44031", + "issue:42936|issue:43240", + "issue:43240|issue:46144", + "issue:42936|issue:44031", + "issue:42936|issue:46144", + "issue:42936|issue:43813", + "issue:43240|issue:43668", + "issue:43668|issue:43813", + "issue:37934|issue:43668", + "issue:42936|issue:43668" + ], + "prepared_review_unit_hash": "3276d63282510663825ca4b10e39198d94989a497e03e7d48de477b4667b1a87", + "component_budget": { + "node_count": 483, + "item_count": 483, + "soft_pair_count": 1779, + "serialized_chars": 922164, + "estimated_input_tokens": 230541, + "estimated_eval_tokens": 461338 + }, + "budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 11, + "serialized_chars": 7776, + "estimated_input_tokens": 1944, + "estimated_eval_tokens": 4144 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These issues are largely unrelated across different models and subsystems (norms, loss wiring, config typing, tokenizer warnings, MoE gating, attention implementation). No duplicate cluster emerges, so no soft edge should be merged.", + "confidence": 0.92, + "canonical_issue_reason": "No single canonical issue stands out: the items describe distinct bugs and questions in separate code paths and models.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43240 is the best general anchor only because it has the strongest discussion/inbound reference signal, but it is not a duplicate of the others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:42936", + "right": "issue:46018", + "accept": false, + "reason": "Different models and failures: Mask2Former ignore_value handling vs DeepSeek-V4 MoE gating." + }, + { + "left": "issue:43668", + "right": "issue:44031", + "accept": false, + "reason": "Unrelated bug types: ModernBERT config type hint vs tokenizer regex warning regression." + }, + { + "left": "issue:42936", + "right": "issue:43240", + "accept": false, + "reason": "Mask2Former loss ignore_value bug is unrelated to kwargs not being forwarded into a loss function." + }, + { + "left": "issue:43240", + "right": "issue:46144", + "accept": false, + "reason": "Loss-kwargs forwarding and RoFormer attention-interface usage are different code paths and fixes." + }, + { + "left": "issue:42936", + "right": "issue:44031", + "accept": false, + "reason": "Different subsystem and symptom: segmentation ignore_value vs tokenizer warning message." + }, + { + "left": "issue:42936", + "right": "issue:46144", + "accept": false, + "reason": "No shared underlying bug; one is a Mask2Former training target issue, the other an attention API usage issue." + }, + { + "left": "issue:42936", + "right": "issue:43813", + "accept": false, + "reason": "Mask2Former loss masking and a typo in a quantization attribute are unrelated." + }, + { + "left": "issue:43240", + "right": "issue:43668", + "accept": false, + "reason": "Passing kwargs to loss calculation is unrelated to an incorrect type hint in ModernBERTConfig." + }, + { + "left": "issue:43668", + "right": "issue:43813", + "accept": false, + "reason": "A type-hint correction and a string-attribute typo are not the same bug." + }, + { + "left": "issue:37934", + "right": "issue:43668", + "accept": false, + "reason": "Llama4TextL2Norm behavior question is unrelated to ModernBERTConfig typing." + }, + { + "left": "issue:42936", + "right": "issue:43668", + "accept": false, + "reason": "These concern different models and different failure modes; no shared code-path bug 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": 2, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10519-3", + "nodes": [ + "issue:10519", + "issue:11005", + "issue:11307" + ], + "soft_pairs": [ + "issue:10519|issue:11307", + "issue:10519|issue:11005", + "issue:11005|issue:11307" + ], + "prepared_review_unit_hash": "99a860ef127888c5a487f5a4acca4190bbcc7d515aaa97277fa10c3127883d70", + "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": "All three artifacts are unrelated feature requests: truncation behavior in tokenization, ReduceLROnPlateau-like scheduler functionality, and Wav2Vec2 word boundary offsets. No duplicate cluster emerges from the titles or available metadata.", + "confidence": 0.99, + "canonical_issue_reason": "No canonical issue: the items describe different features in different parts of the library, so none represents the others.", + "canonical_pr_reason": null, + "best_issue_reason": "None of the issues are suitable as a global representative because they are not about the same underlying problem.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10519", + "right": "issue:11307", + "accept": false, + "reason": "Truncation-from-beginning vs Wav2Vec2 word offset extraction are unrelated requests in different code paths." + }, + { + "left": "issue:10519", + "right": "issue:11005", + "accept": false, + "reason": "Token truncation behavior and ReduceLROnPlateau-like scheduling functionality are distinct, unrelated features." + }, + { + "left": "issue:11005", + "right": "issue:11307", + "accept": false, + "reason": "Learning-rate scheduler enhancements and Wav2Vec2 alignment offsets do not share 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": 3, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-13244-3", + "nodes": [ + "issue:13244", + "issue:15980", + "issue:16618" + ], + "soft_pairs": [ + "issue:13244|issue:16618", + "issue:15980|issue:16618" + ], + "prepared_review_unit_hash": "afa98300a0847fa8ea8a628caf9411f8daf6668e4ebdbbb49717e001bcbaddce", + "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 looks mixed rather than duplicate: one tokenizer discrepancy issue, one private-model/auth error issue, and one generic model-loading failure. I would not merge any of the soft pairs.", + "confidence": 0.91, + "canonical_issue_reason": "No clear canonical duplicate exists; 13244 is about Tapas tokenization, 15980 is about a bad login/private-model error message, and 16618 is about failing to load bert-base-uncased.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 16618 is the best standalone representative because it is the broadest and most directly about a model-loading failure, but it still does not match the other issues as the same bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:13244", + "right": "issue:16618", + "accept": false, + "reason": "Different problem domains: tokenizer output mismatch vs model loading failure; not the same underlying bug." + }, + { + "left": "issue:15980", + "right": "issue:16618", + "accept": false, + "reason": "Both mention model access/loading, but 15980 is specifically about unauthenticated private-model download messaging while 16618 is a generic load failure for bert-base-uncased." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 4, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-26350-2", + "nodes": [ + "issue:26350", + "issue:33260" + ], + "soft_pairs": [ + "issue:26350|issue:33260" + ], + "prepared_review_unit_hash": "f9e609c556f9741817e8def7e24d755d16ea5d6dfce432507aa93ed4c8f60667", + "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": "These are both broad community-contribution issues, but they target different features and code paths: Flash Attention 2 support vs GGUF support. They are not 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:26350", + "right": "issue:33260", + "accept": false, + "reason": "Different underlying changes: one requests Flash Attention 2 support, the other requests GGUF support. Same general theme of adding architecture support, but not 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": 5, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-27088-8", + "nodes": [ + "issue:27088", + "issue:28936", + "issue:38435", + "issue:43834", + "issue:44365", + "issue:44370", + "issue:46222", + "issue:46226" + ], + "soft_pairs": [ + "issue:27088|issue:44370", + "issue:27088|issue:44365", + "issue:27088|issue:43834", + "issue:27088|issue:38435", + "issue:27088|issue:28936", + "issue:28936|issue:43834", + "issue:28936|issue:44365", + "issue:28936|issue:38435", + "issue:28936|issue:44370", + "issue:38435|issue:44370", + "issue:38435|issue:44365", + "issue:38435|issue:43834", + "issue:38435|issue:46222", + "issue:38435|issue:46226", + "issue:43834|issue:44370", + "issue:43834|issue:44365", + "issue:43834|issue:46222", + "issue:43834|issue:46226", + "issue:44365|issue:44370", + "issue:44365|issue:46222", + "issue:44365|issue:46226", + "issue:44370|issue:46222", + "issue:44370|issue:46226", + "issue:46222|issue:46226" + ], + "prepared_review_unit_hash": "b0a5fc370acb74d29fbd3f92b4aa477b5ac98e45f8909f638317aa715eabca17", + "component_budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 24, + "serialized_chars": 13524, + "estimated_input_tokens": 3381, + "estimated_eval_tokens": 7018 + }, + "budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 24, + "serialized_chars": 13524, + "estimated_input_tokens": 3381, + "estimated_eval_tokens": 7018 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is mostly a set of translation-request issues for different languages, so it is not a true duplicate cluster overall. The only pair that looks like a real duplicate is the two identical Thai issues; the rest are either different languages or generic template-like requests without enough evidence of the same underlying change.", + "confidence": 0.92, + "canonical_issue_reason": "issue:27088 is the most concrete, language-specific request and has the clearest discussion trail; the other items are either different languages or generic placeholder-style issues, so there is no stronger cluster anchor.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:27088 is the best standalone representative because it is explicit, specific, and active, whereas most other items are either template-like or clearly different translation targets.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:27088", + "right": "issue:44370", + "accept": false, + "reason": "Different translation-request artifacts; no evidence they target the same language or underlying task." + }, + { + "left": "issue:27088", + "right": "issue:44365", + "accept": false, + "reason": "Different translation-request artifacts; the shared theme is only broad i18n/docs translation." + }, + { + "left": "issue:27088", + "right": "issue:43834", + "accept": false, + "reason": "Different translation-request artifacts; similar wording is not enough to show the same request." + }, + { + "left": "issue:27088", + "right": "issue:38435", + "accept": false, + "reason": "Different languages (Turkish vs Romanian), so these are distinct issues." + }, + { + "left": "issue:27088", + "right": "issue:28936", + "accept": false, + "reason": "Different languages (Turkish vs Spanish), so these are distinct issues." + }, + { + "left": "issue:28936", + "right": "issue:43834", + "accept": false, + "reason": "Spanish translation request vs a generic placeholder-style issue; no same-language evidence." + }, + { + "left": "issue:28936", + "right": "issue:44365", + "accept": false, + "reason": "Spanish request and generic placeholder request are not the same concrete issue." + }, + { + "left": "issue:28936", + "right": "issue:38435", + "accept": false, + "reason": "Different languages (Spanish vs Romanian), so these are distinct issues." + }, + { + "left": "issue:28936", + "right": "issue:44370", + "accept": false, + "reason": "Spanish request and generic placeholder request are not the same concrete issue." + }, + { + "left": "issue:38435", + "right": "issue:44370", + "accept": false, + "reason": "Romanian request vs generic placeholder-style issue; not enough to treat as the same request." + }, + { + "left": "issue:38435", + "right": "issue:44365", + "accept": false, + "reason": "Romanian request vs generic placeholder-style issue; similar template text is insufficient." + }, + { + "left": "issue:38435", + "right": "issue:43834", + "accept": false, + "reason": "Romanian request vs generic placeholder-style issue; not the same concrete translation target." + }, + { + "left": "issue:38435", + "right": "issue:46222", + "accept": false, + "reason": "Different visible request text and no evidence of the same underlying translation target." + }, + { + "left": "issue:38435", + "right": "issue:46226", + "accept": false, + "reason": "Different visible request text and no evidence of the same underlying translation target." + }, + { + "left": "issue:43834", + "right": "issue:44370", + "accept": false, + "reason": "These look like generic template variants, but there is no concrete evidence they are the same exact request." + }, + { + "left": "issue:43834", + "right": "issue:44365", + "accept": false, + "reason": "Template-like similarity alone is too weak to merge as duplicates." + }, + { + "left": "issue:43834", + "right": "issue:46222", + "accept": false, + "reason": "Generic placeholder-style issue vs a different visible request; no same-target evidence." + }, + { + "left": "issue:43834", + "right": "issue:46226", + "accept": false, + "reason": "Generic placeholder-style issue vs a different visible request; no same-target evidence." + }, + { + "left": "issue:44365", + "right": "issue:44370", + "accept": false, + "reason": "Very similar template wording, but not enough evidence that they are the same underlying request." + }, + { + "left": "issue:44365", + "right": "issue:46222", + "accept": false, + "reason": "Generic placeholder-style issue vs a different visible request; not enough to merge." + }, + { + "left": "issue:44365", + "right": "issue:46226", + "accept": false, + "reason": "Generic placeholder-style issue vs a different visible request; not enough to merge." + }, + { + "left": "issue:44370", + "right": "issue:46222", + "accept": false, + "reason": "Generic placeholder-style issue vs a different visible request; no concrete duplicate signal." + }, + { + "left": "issue:44370", + "right": "issue:46226", + "accept": false, + "reason": "Generic placeholder-style issue vs a different visible request; no concrete duplicate signal." + }, + { + "left": "issue:46222", + "right": "issue:46226", + "accept": true, + "reason": "Same visible Thai title and nearly identical metadata/body length strongly indicate the same underlying issue." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The assessment is grounded in the packet and stays conservative. It correctly treats the Turkish/Spanish/Romanian issues as distinct language-specific requests, treats the template-title issues as similar but not proven duplicates, and only accepts the Thai pair where the titles and metadata are effectively identical. No overstatement of evidence detected." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 6, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-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": "60726fb48adde51efb445779676315b1850526e7d7d40e67acec1e5b207babd5", + "component_budget": { + "node_count": 18, + "item_count": 18, + "soft_pair_count": 40, + "serialized_chars": 24146, + "estimated_input_tokens": 6037, + "estimated_eval_tokens": 12330 + }, + "budget": { + "node_count": 18, + "item_count": 18, + "soft_pair_count": 40, + "serialized_chars": 24146, + "estimated_input_tokens": 6037, + "estimated_eval_tokens": 12330 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These items are mostly distinct model-support requests that share a generic 'add model' template. The apparent similarity is broad subsystem overlap, not the same underlying bug or change, so the soft duplicate links should be rejected.", + "confidence": 0.95, + "canonical_issue_reason": "No canonical duplicate issue: the cluster spans many different model requests and only shares superficial wording.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43671 is the clearest representative request in this set because it is explicit, model-specific, and has discussion activity, but it is still not a duplicate of the others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:36817", + "right": "issue:41211", + "accept": false, + "reason": "Different models and different requested support; shared template only." + }, + { + "left": "issue:36817", + "right": "issue:43671", + "accept": false, + "reason": "EuroBert config vs Qwen3-TTS are unrelated model additions." + }, + { + "left": "issue:36817", + "right": "issue:36879", + "accept": false, + "reason": "EuroBert and RF-DETR are distinct model requests." + }, + { + "left": "issue:36879", + "right": "issue:41211", + "accept": false, + "reason": "RF-DETR vs DEIMv2 are different model architectures." + }, + { + "left": "issue:36879", + "right": "issue:43909", + "accept": false, + "reason": "Different targets: object detection vs audio model support." + }, + { + "left": "issue:36879", + "right": "issue:44121", + "accept": false, + "reason": "RF-DETR and sparse transformer request are unrelated." + }, + { + "left": "issue:36879", + "right": "issue:43837", + "accept": false, + "reason": "Different models and modalities; not the same change." + }, + { + "left": "issue:36879", + "right": "issue:43671", + "accept": false, + "reason": "RF-DETR vs Qwen3-TTS are separate support requests." + }, + { + "left": "issue:37015", + "right": "issue:43837", + "accept": false, + "reason": "NeoBERT and Qwen3-ASR are different models and code paths." + }, + { + "left": "issue:37015", + "right": "issue:44121", + "accept": false, + "reason": "NeoBERT vs weight-sparse transformer is not a duplicate request." + }, + { + "left": "issue:37015", + "right": "issue:43909", + "accept": false, + "reason": "Different model families; only generic add-support wording overlaps." + }, + { + "left": "issue:36879", + "right": "issue:37015", + "accept": false, + "reason": "RF-DETR and NeoBERT are distinct model additions." + }, + { + "left": "issue:37015", + "right": "issue:42503", + "accept": false, + "reason": "NeoBERT vs ModernVBERT are separate model requests." + }, + { + "left": "issue:41115", + "right": "issue:43837", + "accept": false, + "reason": "MiniCPM3 and Qwen3-ASR target different models and functionality." + }, + { + "left": "issue:41115", + "right": "issue:44121", + "accept": false, + "reason": "MiniCPM3 support vs sparse-transformer support are unrelated." + }, + { + "left": "issue:41115", + "right": "issue:42503", + "accept": false, + "reason": "MiniCPM3 and ModernVBERT are different model requests." + }, + { + "left": "issue:41115", + "right": "issue:45027", + "accept": false, + "reason": "MiniCPM3 vs Voxtral TTS are different models and modalities." + }, + { + "left": "issue:41211", + "right": "issue:43671", + "accept": false, + "reason": "DEIMv2 and Qwen3-TTS are unrelated model additions." + }, + { + "left": "issue:42503", + "right": "issue:43837", + "accept": false, + "reason": "ModernVBERT vs Qwen3-ASR are different model-specific requests." + }, + { + "left": "issue:42503", + "right": "issue:43671", + "accept": false, + "reason": "ModernVBERT and Qwen3-TTS are unrelated support requests." + }, + { + "left": "issue:42503", + "right": "issue:42738", + "accept": false, + "reason": "Specific ModernVBERT support vs broad BERT-like RoPE feature request." + }, + { + "left": "issue:42738", + "right": "issue:45027", + "accept": false, + "reason": "BERT-like RoPE support and Voxtral TTS are unrelated changes." + }, + { + "left": "issue:42738", + "right": "issue:46653", + "accept": false, + "reason": "The latter is not a technical model request; no duplicate relation." + }, + { + "left": "issue:43671", + "right": "issue:43837", + "accept": false, + "reason": "Qwen3-TTS and Qwen3-ASR share a family but are different modalities and fixes." + }, + { + "left": "issue:43671", + "right": "issue:43909", + "accept": false, + "reason": "Different audio models; not the same code change." + }, + { + "left": "issue:43671", + "right": "issue:44121", + "accept": false, + "reason": "Qwen3-TTS vs sparse transformer are unrelated." + }, + { + "left": "issue:43671", + "right": "issue:45027", + "accept": false, + "reason": "Both are speech-related, but TTS model support is model-specific and not mergeable into one PR here." + }, + { + "left": "issue:43671", + "right": "issue:45177", + "accept": false, + "reason": "TTS support request vs OCR model request are unrelated." + }, + { + "left": "issue:43837", + "right": "issue:43909", + "accept": false, + "reason": "Qwen3-ASR and LFM2.5 Audio are different model support tasks." + }, + { + "left": "issue:43837", + "right": "issue:45322", + "accept": false, + "reason": "Speech ASR request vs EUPE vision-language encoder request are unrelated." + }, + { + "left": "issue:43909", + "right": "issue:44205", + "accept": false, + "reason": "Audio model support vs SAM3-LiteText are different model requests." + }, + { + "left": "issue:43909", + "right": "issue:46645", + "accept": false, + "reason": "Model-add request vs Granite SWA+sink option change are different scopes." + }, + { + "left": "issue:44121", + "right": "issue:45322", + "accept": false, + "reason": "Sparse transformer request and EUPE are unrelated." + }, + { + "left": "issue:43837", + "right": "issue:45027", + "accept": false, + "reason": "ASR and TTS are different speech code paths; not the same fix." + }, + { + "left": "issue:37015", + "right": "issue:45027", + "accept": false, + "reason": "NeoBERT support and Voxtral TTS support are unrelated." + }, + { + "left": "issue:43837", + "right": "issue:45177", + "accept": false, + "reason": "ASR model support vs OCR model support are different changes." + }, + { + "left": "issue:45322", + "right": "issue:46133", + "accept": false, + "reason": "Both are vision-language models, but they are distinct model requests with different implementation needs." + }, + { + "left": "issue:42503", + "right": "issue:46653", + "accept": false, + "reason": "ModernVBERT support is unrelated to the non-technical personal-name issue." + }, + { + "left": "issue:45027", + "right": "issue:46653", + "accept": false, + "reason": "Voxtral TTS support and the personal-name issue are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 7, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-38740-2", + "nodes": [ + "issue:38740", + "issue:46443" + ], + "soft_pairs": [ + "issue:38740|issue:46443" + ], + "prepared_review_unit_hash": "c987aaf8f1024c3a532031bfa5303d231438909bacba21b7ed0f273fdeb63500", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1504, + "estimated_input_tokens": 376, + "estimated_eval_tokens": 1008 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1504, + "estimated_input_tokens": 376, + "estimated_eval_tokens": 1008 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are not duplicates: one is a docs request to add Pruna as an optimization framework, the other is a Vision RoPE refactor. They concern different parts of the project and different kinds of work.", + "confidence": 0.96, + "canonical_issue_reason": "Issue #46443 is the better canonical representative because it is a concrete technical refactor topic; #38740 is a separate documentation-related request.", + "canonical_pr_reason": null, + "best_issue_reason": "#46443 is the most suitable issue to anchor the cluster since it is the more substantive software-change topic and not just a docs addition request.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:38740", + "right": "issue:46443", + "accept": false, + "reason": "Different subjects and artifact types: a docs enhancement for Pruna versus a Vision RoPE refactor. No evidence they 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": 8, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-41084-2", + "nodes": [ + "issue:41084", + "issue:43037" + ], + "soft_pairs": [ + "issue:41084|issue:43037" + ], + "prepared_review_unit_hash": "4da79feca56257751fb8425c42bf1254ffdd96a3aa0040f48168807ec9efc2c7", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1479, + "estimated_input_tokens": 370, + "estimated_eval_tokens": 996 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1479, + "estimated_input_tokens": 370, + "estimated_eval_tokens": 996 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues do not look like duplicates: one is about enabling block decoding, while the other asks for DeepSeek v3.2 model support. They are different feature requests with no shared targets or code context.", + "confidence": 0.94, + "canonical_issue_reason": "Issue #41084 is the better canonical issue only in the sense that it is more specific and likely to map to a concrete implementation task; however, it is not a duplicate of #43037.", + "canonical_pr_reason": null, + "best_issue_reason": "#41084 is the most actionable issue in the pair because it states a narrower technical change. #43037 is a broader model-support request and does not subsume the block-decoding request.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:41084", + "right": "issue:43037", + "accept": false, + "reason": "Reject: the titles describe different underlying requests. 'Set Block Decoding' concerns a decoding behavior/feature, while 'DeepSeek v3.2 support' concerns adding support for a specific model version. No explicit linkage, shared targets, or evidence they refer to the same bug/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 9, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-41306-2", + "nodes": [ + "issue:41306", + "issue:41826" + ], + "soft_pairs": [ + "issue:41306|issue:41826" + ], + "prepared_review_unit_hash": "e3150bad90c7054af57d7420885ade23fad31d0f78bb88ba12471ce854fd5b3a", + "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": "The two issues are both about memory-efficient model components, but they describe different feature requests and different code paths: FLCE/TiledFusedLogitsLoss support in `from_pretrained()` versus adding TiledMLP integration. They are not duplicates.", + "confidence": 0.96, + "canonical_issue_reason": "No canonical duplicate here; 41306 is a concrete `from_pretrained()` integration request for FLCE/TiledFusedLogitsLoss, while 41826 is a separate TiledMLP feature request.", + "canonical_pr_reason": null, + "best_issue_reason": "41306 is the more specific and actionable tracking issue, but it should still be kept separate from 41826 because the requested change is different.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:41306", + "right": "issue:41826", + "accept": false, + "reason": "Not the same underlying bug/change: one asks for a `from_pretrained()` flag for FLCE/TiledFusedLogitsLoss, the other for integrating TiledMLP to reduce memory footprint." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 10, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-43598-2", + "nodes": [ + "issue:43598", + "issue:43600" + ], + "soft_pairs": [ + "issue:43598|issue:43600" + ], + "prepared_review_unit_hash": "0d1bb28870fffd4a041fc72e874888271b07439aca6913b4e8f0935e5e900be0", + "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": "Both issues mention Trainer customization, but they target different problems: one is about `remove_unused_column` behavior, the other about FLOPS estimation for custom trainers. They are related in subsystem, not duplicates.", + "confidence": 0.9, + "canonical_issue_reason": "Issue #43598 is the broader, more central Trainer customization request and better represents the cluster theme.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #43598 is the best issue representative because it focuses on the underlying Trainer customization behavior rather than a narrower metrics-estimation concern.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43598", + "right": "issue:43600", + "accept": false, + "reason": "They concern different code paths and outcomes: `remove_unused_column` customizability vs FLOPS estimation for custom trainers. Similar theme, but not the same underlying bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 11, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-43684-2", + "nodes": [ + "issue:43684", + "issue:44405" + ], + "soft_pairs": [ + "issue:43684|issue:44405" + ], + "prepared_review_unit_hash": "603739588dcb4328e1aabd0637d9727dba7ace564cadc1126774670d475d7010", + "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 are related Qwen model-registration requests, but they target different API surfaces and model variants: one asks for AutoModel/AutoModelForConditionalGeneration registration for Qwen3-Omni, while the other asks for AutoModelForSequenceClassification support for Qwen3.5. They are not the same underlying bug/change.", + "confidence": 0.92, + "canonical_issue_reason": "Issue #43684 is the broader infrastructure request (model registration into AutoModel/AutoModelForConditionalGeneration) and is phrased as a general missing-registration problem, so it is the better canonical issue among the two.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #43684 is the best representative of the cluster because it describes a core registration gap rather than a narrower task-specific class addition.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43684", + "right": "issue:44405", + "accept": false, + "reason": "Different model families and different heads/tasks: Qwen3-Omni registration for base/generative auto classes vs Qwen3.5 sequence classification support. Related, 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": 12, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-43835-4", + "nodes": [ + "issue:43835", + "issue:43845", + "issue:43846", + "issue:43859" + ], + "soft_pairs": [ + "issue:43835|issue:43845", + "issue:43835|issue:43846", + "issue:43835|issue:43859", + "issue:43845|issue:43846", + "issue:43845|issue:43859", + "issue:43846|issue:43859" + ], + "prepared_review_unit_hash": "6d367fd5ad4c7516e300c170a38d142b939052024ed20cec5accefa9127fa4bf", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 6, + "serialized_chars": 4267, + "estimated_input_tokens": 1067, + "estimated_eval_tokens": 2390 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 6, + "serialized_chars": 4267, + "estimated_input_tokens": 1067, + "estimated_eval_tokens": 2390 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Four closed issues with extremely sparse content; three share the generic title \"huggingface\" and one is unrelated-looking \"Fett.\" There is not enough concrete signal to confirm any duplicate cluster beyond superficial title similarity, and there are no PRs.", + "confidence": 0.9, + "canonical_issue_reason": "Issue 43845 is the best of a weak set: it is one of the repeated \"huggingface\" reports and has the longest body among those near-duplicates, so it is the least arbitrary anchor. Still, the evidence is very thin.", + "canonical_pr_reason": null, + "best_issue_reason": "43845 is the most plausible representative issue because it matches the repeated title pattern and appears marginally more descriptive than the others. However, the cluster lacks actionable details, so it is only a weak canonical choice.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43835", + "right": "issue:43845", + "accept": false, + "reason": "Only superficial text similarity; \"Fett\" and \"huggingface\" do not clearly describe the same bug or request." + }, + { + "left": "issue:43835", + "right": "issue:43846", + "accept": false, + "reason": "No concrete shared problem statement or target; title similarity is insufficient to treat these as duplicates." + }, + { + "left": "issue:43835", + "right": "issue:43859", + "accept": false, + "reason": "Different titles and no supporting detail linking them to the same underlying issue." + }, + { + "left": "issue:43845", + "right": "issue:43846", + "accept": false, + "reason": "Both are generically titled \"huggingface,\" but there is no substantive body, target, or repro detail proving the same underlying bug." + }, + { + "left": "issue:43845", + "right": "issue:43859", + "accept": false, + "reason": "Shared generic wording alone is not enough; the sparse reports do not establish the same issue." + }, + { + "left": "issue:43846", + "right": "issue:43859", + "accept": false, + "reason": "The reports are too underspecified to confirm they refer to the same problem rather than unrelated mention of the same keyword." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 13, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-44371-5", + "nodes": [ + "issue:44371", + "issue:44418", + "issue:44572", + "issue:44573", + "issue:44937" + ], + "soft_pairs": [ + "issue:44371|issue:44572", + "issue:44371|issue:44573", + "issue:44371|issue:44418", + "issue:44371|issue:44937", + "issue:44418|issue:44572", + "issue:44418|issue:44573", + "issue:44572|issue:44573", + "issue:44572|issue:44937", + "issue:44573|issue:44937" + ], + "prepared_review_unit_hash": "6ad56872d2c285fe966664734a39ebc406449f86954ad43416371314d1c88f9a", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 9, + "serialized_chars": 5794, + "estimated_input_tokens": 1449, + "estimated_eval_tokens": 3154 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 9, + "serialized_chars": 5794, + "estimated_input_tokens": 1449, + "estimated_eval_tokens": 3154 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster looks like low-signal closed noise/spam rather than a real duplicate set. Three issues share the exact title \"\", but there is no substantive body or code-context evidence tying them to the same underlying bug. The remaining two issues are generic/unrelated titles and do not match the spam subgroup.", + "confidence": 0.88, + "canonical_issue_reason": "issue:44371 is the earliest of the repeated exact-title spam issues and is the cleanest representative of that subgroup.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44371 is the best overall representative because it matches two other items exactly by title and is earlier than the rest.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44371", + "right": "issue:44572", + "accept": false, + "reason": "Same placeholder title, but no body or context showing a shared underlying bug/change; looks like duplicate noise, not a real duplicate issue." + }, + { + "left": "issue:44371", + "right": "issue:44573", + "accept": false, + "reason": "Exact same spam title, but there is no substantive evidence that these describe the same real problem." + }, + { + "left": "issue:44371", + "right": "issue:44418", + "accept": false, + "reason": "Titles are unrelated and generic; no indication they refer to the same issue." + }, + { + "left": "issue:44371", + "right": "issue:44937", + "accept": false, + "reason": "\"Check out \\\"Google LLC\\\"\" is unrelated to the spam placeholder issue." + }, + { + "left": "issue:44418", + "right": "issue:44572", + "accept": false, + "reason": "Generic documentation suggestion versus spam placeholder; no shared underlying bug/change." + }, + { + "left": "issue:44418", + "right": "issue:44573", + "accept": false, + "reason": "Generic documentation suggestion versus spam placeholder; no shared underlying bug/change." + }, + { + "left": "issue:44572", + "right": "issue:44573", + "accept": false, + "reason": "Identical placeholder titles, but still only spam/noise with no real bug or change description to confirm a true duplicate." + }, + { + "left": "issue:44572", + "right": "issue:44937", + "accept": false, + "reason": "Unrelated titles with no evidence of the same problem." + }, + { + "left": "issue:44573", + "right": "issue:44937", + "accept": false, + "reason": "Unrelated titles with no evidence of the same problem." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 14, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45145-2", + "nodes": [ + "issue:45145", + "issue:45151" + ], + "soft_pairs": [ + "issue:45145|issue:45151" + ], + "prepared_review_unit_hash": "491798b717d6a702e48ffb85074d3741d5e1af961e175a3413016eee02a2eaa2", + "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 issues are highly similar on title and general shape, but the packet only gives soft similarity and does not provide enough evidence to conclusively treat them as duplicates. #45145 is still the better canonical representative because it has an inbound reference and is the more established thread.", + "confidence": 0.79, + "canonical_issue_reason": "#45145 is the better canonical issue because it has the same title as #45151 and already has an inbound reference, so it is the likelier existing discussion to retain.", + "canonical_pr_reason": null, + "best_issue_reason": "#45145 is the best overall issue representative for this cluster because it has the stronger reference signal while still matching the same topic/title as #45151.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45145", + "right": "issue:45151", + "accept": false, + "reason": "The only explicit pair evidence is soft_similarity, and deterministic_accept is false. The issues share the same title and similar size, but the packet does not provide enough concrete evidence to confidently mark them as the same underlying issue." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The canonical-choice reasoning is mostly grounded, but the soft-edge verdict is too strong for the packet: the only pair evidence is `soft_similarity`, and `deterministic_accept` is false. The summary overstates certainty by saying the issues are duplicates with \u201chighly similar body content\u201d and no distinct scope, when the packet only directly supports same title, similar length, and soft similarity. Please keep this more conservative or mark the edge as unaccepted." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 15, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45295-2", + "nodes": [ + "issue:45295", + "issue:45373" + ], + "soft_pairs": [], + "prepared_review_unit_hash": "f5fd95fada2f530ab0999e343c17c16266cde0c7182bb22509972ca44827cd49", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 0, + "serialized_chars": 1286, + "estimated_input_tokens": 322, + "estimated_eval_tokens": 900 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 0, + "serialized_chars": 1286, + "estimated_input_tokens": 322, + "estimated_eval_tokens": 900 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues describe the same missing Gemma 4 sequence-classification support; the later, more specific open issue is the better cluster representative.", + "confidence": 0.94, + "canonical_issue_reason": "issue #45373 is the clearest canonical tracker: it is open, explicitly names the missing class `Gemma4ForSequenceClassification`, and states the module-level gap relative to Gemma 2/3.", + "canonical_pr_reason": null, + "best_issue_reason": "issue #45373 best captures the concrete request and likely remaining work, while #45295 is a shorter earlier duplicate with less detail and is already closed.", + "best_pr_reason": null, + "soft_edge_verdicts": [] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 16, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45672-2", + "nodes": [ + "issue:45672", + "issue:45674" + ], + "soft_pairs": [ + "issue:45672|issue:45674" + ], + "prepared_review_unit_hash": "90f51230189b38d1f482db70426d7adca9f8885f589e8c7e14880d4338fcd8f0", + "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 related to 4-bit inference but describe two different bugs in different code paths, so they should not be deduplicated.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45674 is the better cluster representative because it describes a clearer, more general BitsAndBytesConfig regression with a concrete failure mode.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45672", + "right": "issue:45674", + "accept": false, + "reason": "Different underlying bugs: one is a torch.finfo() TypeError on uint8 weights in audio modules, the other is llm_int8_skip_modules clearing the default lm_head exclusion and triggering an AssertionError in 4-bit inference." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 17, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45762-2", + "nodes": [ + "issue:45762", + "issue:45763" + ], + "soft_pairs": [ + "issue:45762|issue:45763" + ], + "prepared_review_unit_hash": "02f7cc330a8a635bdc66d0c9e95a9c3d868a80f146905aacbe29e9adc53dafa0", + "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 appear to describe the same underlying bug or request; they are just unrelated name/date strings with no body or shared target.", + "confidence": 0.96, + "canonical_issue_reason": "No canonical issue: the titles are unrelated and there is no evidence of a shared problem, repro, or target.", + "canonical_pr_reason": null, + "best_issue_reason": "Neither issue is a good duplicate anchor; if forced, 45762 is no more representative than 45763, but there is no real basis to choose one.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45762", + "right": "issue:45763", + "accept": false, + "reason": "Do not merge: the titles are different personal-name/date entries with no body, links, files, or discussion indicating the same underlying issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 18, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45764-2", + "nodes": [ + "issue:45764", + "issue:45765" + ], + "soft_pairs": [ + "issue:45764|issue:45765" + ], + "prepared_review_unit_hash": "94d4fb9f8d8ee93318d18cc91698b3752c6381080ba77b90b2001150921d9272", + "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 just distinct name/date placeholders and do not appear to describe the same underlying bug or request.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45764", + "right": "issue:45765", + "accept": false, + "reason": "The titles reference different people and different dates, with no shared technical context, files, or issue targets; this does not look like the same underlying issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 19, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45810-2", + "nodes": [ + "issue:45810", + "issue:46295" + ], + "soft_pairs": [ + "issue:45810|issue:46295" + ], + "prepared_review_unit_hash": "84c955ba51257653795b5ba0362cfc9a7c55d3e1ccf6a305ed5dee884a142557", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1541, + "estimated_input_tokens": 386, + "estimated_eval_tokens": 1028 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1541, + "estimated_input_tokens": 386, + "estimated_eval_tokens": 1028 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues are unrelated feature requests: one asks for a new token-classification head for Qwen3_5, the other asks for FP8 training support. They should not be clustered as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45810", + "right": "issue:46295", + "accept": false, + "reason": "Different requests in different areas: model-head addition for token classification vs FP8 training support. No shared code-path or concrete duplicate bug/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 20, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-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": "d53ca198c00e3878831813bdc61ab71d1a232805dfc7d039db65933ce50c542b", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 8, + "serialized_chars": 5729, + "estimated_input_tokens": 1433, + "estimated_eval_tokens": 3122 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 8, + "serialized_chars": 5729, + "estimated_input_tokens": 1433, + "estimated_eval_tokens": 3122 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster contains two separate exact-duplicate issue pairs: 45870/45871 and 45872/45873. Issue 45869 is unrelated promotional dataset content and should not be merged with the DataForge requests.", + "confidence": 0.95, + "canonical_issue_reason": "45872 is the strongest representative issue: it is substantive, clearly scoped, and has an exact duplicate (45873). The 45870/45871 pair is also duplicated, but it is a more generic example request.", + "canonical_pr_reason": null, + "best_issue_reason": "45872 is the best single issue to keep because it is the clearest, most specific request among the set and stands in an exact duplicate pair.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45869", + "right": "issue:45870", + "accept": false, + "reason": "Different underlying content: a dataset-bundle promotion versus an example request for financial time-series prediction." + }, + { + "left": "issue:45869", + "right": "issue:45871", + "accept": false, + "reason": "Different underlying content: a dataset-bundle promotion versus an example request for financial time-series prediction." + }, + { + "left": "issue:45870", + "right": "issue:45871", + "accept": true, + "reason": "Exact duplicate issue pair with the same title and apparent same request." + }, + { + "left": "issue:45870", + "right": "issue:45872", + "accept": false, + "reason": "Related domain, but different request: example prediction workflow versus time series plus scraping service." + }, + { + "left": "issue:45870", + "right": "issue:45873", + "accept": false, + "reason": "Related domain, but different request: example prediction workflow versus time series plus scraping service." + }, + { + "left": "issue:45871", + "right": "issue:45872", + "accept": false, + "reason": "Related domain, but different request: example prediction workflow versus time series plus scraping service." + }, + { + "left": "issue:45871", + "right": "issue:45873", + "accept": false, + "reason": "Related domain, but different request: example prediction workflow versus time series plus scraping service." + }, + { + "left": "issue:45872", + "right": "issue:45873", + "accept": true, + "reason": "Exact duplicate issue pair with the same title and apparent same request." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and the soft-edge decisions are conservative. The two exact-title, same-length pairs (45870/45871 and 45872/45873) are reasonable duplicate calls, and the 45869 rejections are well-supported by the titles. No PR-pair merge issue applies here." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 21, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46439-2", + "nodes": [ + "issue:46439", + "issue:46441" + ], + "soft_pairs": [ + "issue:46439|issue:46441" + ], + "prepared_review_unit_hash": "fca36374313af495e35f992bc69a0dc1adc93c8867bc8bbf3ac92119c71cd34d", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1637, + "estimated_input_tokens": 410, + "estimated_eval_tokens": 1076 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1637, + "estimated_input_tokens": 410, + "estimated_eval_tokens": 1076 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both issues involve `cache_implementation=\"static\"` and `linear_attention`, but they describe different failure modes: one is a shape-corruption `RuntimeError` during early initialization, the other is a `KeyError: 'linear_attention'` in Qwen3.5. Related, but not clearly the same duplicate.", + "confidence": 0.81, + "canonical_issue_reason": "Issue 46439 is the broader, more general report of the static-cache/linear-attention initialization bug and better captures the core code-path problem.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 46439 is the better representative because it points at the underlying static-cache corruption path rather than a model-specific `KeyError` symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:46439", + "right": "issue:46441", + "accept": false, + "reason": "Same subsystem, but the concrete failures differ: 46439 reports wrong-shape corruption and `RuntimeError`, while 46441 reports a missing `linear_attention` entry causing `KeyError`. Not enough to treat as the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 22, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46551-2", + "nodes": [ + "issue:46551", + "issue:46552" + ], + "soft_pairs": [ + "issue:46551|issue:46552" + ], + "prepared_review_unit_hash": "512215a01f854c372a69e7eb38f497297d21e1a5b6928c769d1fdcf630f75080", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1707, + "estimated_input_tokens": 427, + "estimated_eval_tokens": 1110 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1707, + "estimated_input_tokens": 427, + "estimated_eval_tokens": 1110 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues are effectively identical reports of the same Wav2Vec2CTCTokenizer v5 regression, so they should be treated as duplicates.", + "confidence": 0.99, + "canonical_issue_reason": "issue:46552 is the better surviving report because it is open and already has inbound references, making it the more useful hub for the duplicate cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:46552 best represents the cluster for ongoing tracking and discussion; it matches the other issue exactly and is the active report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:46551", + "right": "issue:46552", + "accept": true, + "reason": "Exact same title and same underlying bug description; these are duplicate reports of the same tokenizer regression." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The conclusion is grounded in the packet: both issues have the same title, identical body length, and a perfect soft-similarity score, with 46552 open and carrying inbound references. Treating 46552 as the canonical duplicate hub is a conservative, well-supported choice." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 23, + "component_count": 23, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46657-2", + "nodes": [ + "issue:46657", + "issue:46692" + ], + "soft_pairs": [ + "issue:46657|issue:46692" + ], + "prepared_review_unit_hash": "d447cb0095c500d27c06ca4c483e3c9f07832b33607d8f8e98b69c86a32bcc3e", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1568, + "estimated_input_tokens": 392, + "estimated_eval_tokens": 1040 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1568, + "estimated_input_tokens": 392, + "estimated_eval_tokens": 1040 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are not the same underlying bug: one reports a tensor-size mismatch during object detection training, while the other reports an out-of-memory failure after several epochs in RT-DETR v2 training.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "Neither issue is a clean canonical duplicate target; if one must be surfaced, #46657 is the more concrete, immediately reproducible failure report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:46657", + "right": "issue:46692", + "accept": false, + "reason": "Different failure modes and likely different code paths: tensor shape mismatch vs memory exhaustion after epochs. Shared object-detection training context is too broad to treat as the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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-39895-15", + "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", + "pull_request:46744" + ], + "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:46744", + "pull_request:44827|pull_request:46048", + "pull_request:44827|pull_request:45213", + "pull_request:43823|pull_request:45213", + "pull_request:44827|pull_request:45655", + "pull_request:43251|pull_request:43254" + ], + "prepared_review_unit_hash": "67e4cc8c91cbfd41cd58dee8c75a04dd390f9afb69fb563482fdcdb6aa38fcd9", + "component_budget": { + "node_count": 15, + "item_count": 15, + "soft_pair_count": 20, + "serialized_chars": 40203, + "estimated_input_tokens": 10051, + "estimated_eval_tokens": 20358 + }, + "budget": { + "node_count": 15, + "item_count": 15, + "soft_pair_count": 20, + "serialized_chars": 40203, + "estimated_input_tokens": 10051, + "estimated_eval_tokens": 20358 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is mostly a loose set of unrelated PRs grouped by broad soft similarity. The only clear duplicate-like pair is the two `fixed_cross_entropy` PRs, which both target the same issue and same code path; the rest are too different in model, subsystem, or intent to merge together.", + "confidence": 0.41, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 43251 is the clearest focused bug-fix PR in the cluster: it has an explicit issue target, a single-file change, and directly addresses the `cross_entropy` kwargs forwarding problem.", + "best_issue_reason": null, + "best_pr_reason": "PR 43251 is the best representative for triage because it is the most concrete and scoped fix, while the other PRs are either unrelated feature additions, docs/tests, or broad refactors.", + "soft_edge_verdicts": [ + { + "left": "pull_request:39895", + "right": "pull_request:43424", + "accept": false, + "reason": "Different features: VideoPrism model addition vs. executorch dynamic-shape test. Same broad ML stack, but not the same underlying change." + }, + { + "left": "pull_request:41855", + "right": "pull_request:43823", + "accept": false, + "reason": "Tokenizer method additions vs. adding a new MobileLLM model. Both are model-related, but they fix unrelated gaps." + }, + { + "left": "pull_request:41855", + "right": "pull_request:43424", + "accept": false, + "reason": "A tokenizer API patch and an executorch export test are unrelated changes." + }, + { + "left": "pull_request:42668", + "right": "pull_request:43823", + "accept": false, + "reason": "Processor auto-loading robustness and adding MobileLLM are different changes with different code paths." + }, + { + "left": "pull_request:43251", + "right": "pull_request:43823", + "accept": false, + "reason": "`fixed_cross_entropy` kwargs forwarding is unrelated to adding a new model implementation." + }, + { + "left": "pull_request:43251", + "right": "pull_request:44827", + "accept": false, + "reason": "Both touch training/model code, but one is a loss utility fix and the other is Mistral4 test/model wiring; not the same bug or change." + }, + { + "left": "pull_request:43424", + "right": "pull_request:44827", + "accept": false, + "reason": "Executorch export dynamic-shape testing and Mistral4 test fixes are separate test concerns." + }, + { + "left": "pull_request:43424", + "right": "pull_request:43823", + "accept": false, + "reason": "Dynamic-shape export testing does not match a new model addition." + }, + { + "left": "pull_request:43823", + "right": "pull_request:45655", + "accept": false, + "reason": "Model addition vs. `cls.config` resolution order fix; unrelated code paths." + }, + { + "left": "pull_request:43823", + "right": "pull_request:43995", + "accept": false, + "reason": "New MobileLLM model and Falcon output-interface refactor are different model implementations with no shared bug." + }, + { + "left": "pull_request:43995", + "right": "pull_request:45213", + "accept": false, + "reason": "Falcon refactor and model-creation skill tooling are not the same change." + }, + { + "left": "pull_request:43995", + "right": "pull_request:44827", + "accept": false, + "reason": "Falcon standardized-output refactor and Mistral4 test fixes are unrelated model internals." + }, + { + "left": "pull_request:44827", + "right": "pull_request:45170", + "accept": false, + "reason": "Mistral4 test/model fixes and a CLIP-like weight-renaming conversion patch address different problems." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46267", + "accept": false, + "reason": "Mistral4 test fixes and chat response parsing updates are unrelated subsystems." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46744", + "accept": false, + "reason": "Mistral4 test/model changes are unrelated to Apple Silicon safetensors docs." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46048", + "accept": false, + "reason": "Mistral4 model/test work and adding a TDT loss kernel are distinct changes." + }, + { + "left": "pull_request:44827", + "right": "pull_request:45213", + "accept": false, + "reason": "A model/test fix and a repository skill/documentation PR are not the same underlying change." + }, + { + "left": "pull_request:43823", + "right": "pull_request:45213", + "accept": false, + "reason": "Adding MobileLLM and adding a model-creation skill are different deliverables." + }, + { + "left": "pull_request:44827", + "right": "pull_request:45655", + "accept": false, + "reason": "Mistral4 test work and class config resolution ordering are unrelated fixes." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46048", + "accept": false, + "reason": "A Mistral4 test/model PR and a kernel-backed TDT loss addition are different features." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46267", + "accept": false, + "reason": "Mistral4 fixes do not duplicate chat parsing updates." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46744", + "accept": false, + "reason": "Mistral4 fixes and MPS loading docs are unrelated." + }, + { + "left": "pull_request:43251", + "right": "pull_request:43254", + "accept": true, + "reason": "Both PRs fix the same `fixed_cross_entropy` gap in `loss_utils.py` for issue 43240, and the changes are alternative variants of the same code-path fix." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: it stays conservative, reflects that most pairwise links are only soft similarity, and correctly identifies the two `fixed_cross_entropy` PRs as the only mergeable duplicate-like pair with the same issue/code path. The other verdicts are appropriately rejected as unrelated without overstating evidence." + }, + "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": "0d3a8637fcbc3aa5c54dbb4f54acc73a25a69d4b37f2271a98158ec0ffe8c9f1", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 7224, + "estimated_input_tokens": 1806, + "estimated_eval_tokens": 3868 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 7224, + "estimated_input_tokens": 1806, + "estimated_eval_tokens": 3868 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs are about VibeVoice, but they are not the same change: #40546 is the base VibeVoice integration, while #42781 adds a separate realtime extension with acoustic-tokenizer and generation/auto-mapping updates.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "#40546 is the foundational VibeVoice implementation and broader model integration; it reads as the main base change rather than the later realtime extension.", + "best_issue_reason": null, + "best_pr_reason": "#40546 is the best representative of the cluster because it introduces the core VibeVoice package/docs and anchors the model integration, while #42781 is a follow-on realtime-specific expansion.", + "soft_edge_verdicts": [ + { + "left": "pull_request:40546", + "right": "pull_request:42781", + "accept": false, + "reason": "Reject: they share the VibeVoice namespace and some files, but #40546 implements the base model, whereas #42781 adds realtime-specific acoustic tokenizer and generation changes. The concrete code paths differ enough that these would not plausibly be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "f016eb055b7c9dc9c597b02966551d99e05ce70c90b714a5c4799e7646eae9da", + "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": "The two pull requests are unrelated feature additions in different model areas (TimesFM covariates vs. DINOv3 image classification), so 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": "pull_request:40755", + "right": "pull_request:41224", + "accept": false, + "reason": "Different underlying changes and code paths: one adds TimesFM covariate forecasting support, the other adds DINOv3 image classification support. They do not appear mergeable into a single PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "c270010a5e1d275e35aeba58bc60513d91e0aa90a98d3cf60ae3562d8e76045e", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5935, + "estimated_input_tokens": 1484, + "estimated_eval_tokens": 3224 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5935, + "estimated_input_tokens": 1484, + "estimated_eval_tokens": 3224 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two workflow-maintenance PRs in the same area, but they update different GitHub Actions families: one is a broad Node 24 compatibility sweep, the other is a docker/build-push-action version bump. Related, but not the same underlying change.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "pull_request:42975 is the broader, more representative workflow-actions upgrade: it updates multiple core actions across many workflows for Node 24 compatibility, making it the better cluster anchor.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:42975 is the best single PR to represent the cluster because it covers the main cross-workflow action-compatibility update rather than a narrower docker-action bump.", + "soft_edge_verdicts": [ + { + "left": "pull_request:42975", + "right": "pull_request:42976", + "accept": false, + "reason": "Reject: both are GitHub Actions upgrade PRs, but they target different concrete dependencies and code paths. 42975 updates checkout/cache/upload-artifact for Node 24 compatibility; 42976 updates docker/build-push-action. They could be merged in principle as a broad maintenance sweep, but they are not the same underlying fix/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-42978-7", + "nodes": [ + "pull_request:42978", + "pull_request:43448", + "pull_request:43451", + "pull_request:45490", + "pull_request:46562", + "pull_request:46658", + "pull_request:46705" + ], + "soft_pairs": [ + "pull_request:42978|pull_request:43448", + "pull_request:42978|pull_request:43451", + "pull_request:42978|pull_request:45490", + "pull_request:42978|pull_request:46705", + "pull_request:43448|pull_request:43451", + "pull_request:43448|pull_request:45490", + "pull_request:43448|pull_request:46705", + "pull_request:43451|pull_request:45490", + "pull_request:43451|pull_request:46705", + "pull_request:45490|pull_request:46562", + "pull_request:45490|pull_request:46705", + "pull_request:46562|pull_request:46658" + ], + "prepared_review_unit_hash": "490abf623a6ceed6a81a816f766bc3c943c5ce828368b77a4bcc49a3046c9a1a", + "component_budget": { + "node_count": 7, + "item_count": 7, + "soft_pair_count": 12, + "serialized_chars": 24621, + "estimated_input_tokens": 6156, + "estimated_eval_tokens": 12568 + }, + "budget": { + "node_count": 7, + "item_count": 7, + "soft_pair_count": 12, + "serialized_chars": 24621, + "estimated_input_tokens": 6156, + "estimated_eval_tokens": 12568 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is mostly a set of unrelated model-addition PRs plus one closely related Step 3.7 Flash pair. The only strong duplicate-like match is the WIP Step 3 7 Flash PR versus the main Step 3.7 Flash model PR; the rest share broad auto-registration/docs patterns but are distinct features/models.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "46562 is the best representative PR: it contains the actual Step 3.7 Flash model implementation, docs, and tests. The nearby 46658 PR is a separate WIP/mini-model conversion harness rather than a distinct user-facing change.", + "best_issue_reason": null, + "best_pr_reason": "46562 is the strongest global representative because it is the complete model-addition PR with the broadest user-facing scope; the other PRs are separate model additions or infrastructure changes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:42978", + "right": "pull_request:43448", + "accept": false, + "reason": "Both add different models (ViT NEPA vs Molmo). Shared docs/auto-registration files are incidental, not the same change." + }, + { + "left": "pull_request:42978", + "right": "pull_request:43451", + "accept": false, + "reason": "ViT NEPA and Molmo2 are unrelated model additions; overlap is only in shared registration/docs plumbing." + }, + { + "left": "pull_request:42978", + "right": "pull_request:45490", + "accept": false, + "reason": "Different model families and use cases; no evidence they fix the same underlying code-path or feature." + }, + { + "left": "pull_request:42978", + "right": "pull_request:46705", + "accept": false, + "reason": "Model-addition PR vs distributed-config infrastructure PR; unrelated scope." + }, + { + "left": "pull_request:43448", + "right": "pull_request:43451", + "accept": false, + "reason": "Molmo and Molmo2 are related by name, but they are separate model implementations with different files and behavior." + }, + { + "left": "pull_request:43448", + "right": "pull_request:45490", + "accept": false, + "reason": "Molmo and CTSM are different model additions; shared auto/docs changes do not make them duplicates." + }, + { + "left": "pull_request:43448", + "right": "pull_request:46705", + "accept": false, + "reason": "Molmo model addition is unrelated to adding distributed config support." + }, + { + "left": "pull_request:43451", + "right": "pull_request:45490", + "accept": false, + "reason": "Molmo2 and CTSM are distinct model ports; no shared concrete bug or feature implementation." + }, + { + "left": "pull_request:43451", + "right": "pull_request:46705", + "accept": false, + "reason": "Molmo2 model support and distributed config are different areas of the codebase." + }, + { + "left": "pull_request:45490", + "right": "pull_request:46562", + "accept": false, + "reason": "CTSM and Step 3.7 Flash are different model additions with no shared code-path problem." + }, + { + "left": "pull_request:45490", + "right": "pull_request:46705", + "accept": false, + "reason": "CTSM model support is unrelated to distributed configuration plumbing." + }, + { + "left": "pull_request:46562", + "right": "pull_request:46658", + "accept": true, + "reason": "Both are about Step 3.7 Flash and the same model code-path; the WIP mini-model/conversion harness plausibly belongs in the same PR as the model addition." + } + ] + }, + "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-43291-9", + "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", + "pull_request:46719" + ], + "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:43498|pull_request:46719", + "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", + "pull_request:46579|pull_request:46719", + "pull_request:43665|pull_request:46719" + ], + "prepared_review_unit_hash": "432ad07f793cb9600e16cca94126cde3f78308fe5c5e1522b1d3cef600ea5473", + "component_budget": { + "node_count": 9, + "item_count": 9, + "soft_pair_count": 17, + "serialized_chars": 26393, + "estimated_input_tokens": 6599, + "estimated_eval_tokens": 13454 + }, + "budget": { + "node_count": 9, + "item_count": 9, + "soft_pair_count": 17, + "serialized_chars": 26393, + "estimated_input_tokens": 6599, + "estimated_eval_tokens": 13454 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is not duplicate-coherent: it mixes unrelated PRs on Whisper tests, backward-compat shims, model test stubs, FP8 dequantization, doctest cleanup, generation validation, CI memory limits, typo fixes, and sparse attention.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46579 is the broadest substantive code-fix PR here; the others are narrower, test/doc-only, infra-only, or clearly unrelated.", + "best_issue_reason": null, + "best_pr_reason": "PR 46579 is the strongest representative because it makes concrete code-path changes across multiple core modules, while the other PRs are mostly isolated cleanup or unrelated fixes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43291", + "right": "pull_request:43665", + "accept": false, + "reason": "Different areas and failure modes: Whisper tokenizer/test fix vs. CLIP/ViT test stubs; no shared underlying bug." + }, + { + "left": "pull_request:43291", + "right": "pull_request:44815", + "accept": false, + "reason": "Whisper decoding/tests and FP8 dequantization are unrelated code paths." + }, + { + "left": "pull_request:43291", + "right": "pull_request:46380", + "accept": false, + "reason": "Whisper test fix vs. generation watermarking validation; different subsystems and bugs." + }, + { + "left": "pull_request:43498", + "right": "pull_request:46579", + "accept": false, + "reason": "Backward-compat alias for tie_weights is unrelated to broad typo cleanup across many files." + }, + { + "left": "pull_request:43498", + "right": "pull_request:46719", + "accept": false, + "reason": "tie_weights compatibility shim and sparse-attention/layout fix are unrelated." + }, + { + "left": "pull_request:43665", + "right": "pull_request:46539", + "accept": false, + "reason": "Bogus model test edits are not the same as CI memory-limit/test-environment changes." + }, + { + "left": "pull_request:43665", + "right": "pull_request:44815", + "accept": false, + "reason": "Test-only edits for CLIP/ViT do not match FP8 loading/dequant logic." + }, + { + "left": "pull_request:43665", + "right": "pull_request:46380", + "accept": false, + "reason": "Different targets: model test stubs versus generation config validation." + }, + { + "left": "pull_request:43665", + "right": "pull_request:46579", + "accept": false, + "reason": "Model test changes and multi-file typo cleanup are not the same underlying change." + }, + { + "left": "pull_request:43665", + "right": "pull_request:46719", + "accept": false, + "reason": "Test stubs for CLIP/ViT are unrelated to sparse attention / Minimax-DeepSeek code." + }, + { + "left": "pull_request:44815", + "right": "pull_request:46539", + "accept": false, + "reason": "FP8 dequantization fixes and Phimoe CI harness changes address different problems." + }, + { + "left": "pull_request:44815", + "right": "pull_request:46380", + "accept": false, + "reason": "Different code paths: FP8 quantization/loading versus generation watermarking validation." + }, + { + "left": "pull_request:44815", + "right": "pull_request:46579", + "accept": false, + "reason": "Only broad subsystem overlap in core/model-loading; the concrete fixes are different (FP8 dequant vs typo cleanup)." + }, + { + "left": "pull_request:45114", + "right": "pull_request:46579", + "accept": false, + "reason": "Doctest documentation cleanup is not the same as code typo fixes across core modules." + }, + { + "left": "pull_request:46380", + "right": "pull_request:46539", + "accept": false, + "reason": "Generation config validation and CI memory patching are unrelated." + }, + { + "left": "pull_request:46579", + "right": "pull_request:46719", + "accept": false, + "reason": "Typo cleanup across many modules is not the same concrete bug as the sparse-attention reshape/layout 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": 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-43488-3", + "nodes": [ + "pull_request:43488", + "pull_request:43532", + "pull_request:46747" + ], + "soft_pairs": [ + "pull_request:43488|pull_request:46747", + "pull_request:43488|pull_request:43532", + "pull_request:43532|pull_request:46747" + ], + "prepared_review_unit_hash": "c4a36a22a4341ff21320797c3aa9e463a7109eb67013e425f9e696dea838cbf6", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 6350, + "estimated_input_tokens": 1588, + "estimated_eval_tokens": 3432 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 6350, + "estimated_input_tokens": 1588, + "estimated_eval_tokens": 3432 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Three unrelated \"[don't merge]\" PRs: one tweaks setup/versioning and a test, one rewires several CI/comment/testing utilities, and one changes the PR security gate workflow. They share superficial repo/CI similarity but do not address the same concrete bug or change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 43532 is the most representative cluster item because it has the broadest and most explicit CI-related diff plus the most discussion, but it is still not a duplicate of the others.", + "best_issue_reason": null, + "best_pr_reason": "PR 43532 is the best single representative of the cluster due to its larger surface area and clearer workflow/testing context, making it the most informative anchor for review triage.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43488", + "right": "pull_request:46747", + "accept": false, + "reason": "Different changes: setup/version/dependency and VIT test tweaks versus a PR security gate workflow switch. Same broad repo, not the same underlying bug or fix." + }, + { + "left": "pull_request:43488", + "right": "pull_request:43532", + "accept": false, + "reason": "The diffs target different areas: packaging/versioning and a test file versus CI workflows, testing utilities, and notifications. No shared concrete code-path problem." + }, + { + "left": "pull_request:43532", + "right": "pull_request:46747", + "accept": false, + "reason": "Both touch CI-related files, but one rewrites several comment/test workflows while the other only changes the security gate trigger. 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": 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": "b4e1634d6da8cd4030bc5e95f307f712ad9acf0761fc304187d1364439f43da1", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6083, + "estimated_input_tokens": 1521, + "estimated_eval_tokens": 3298 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6083, + "estimated_input_tokens": 1521, + "estimated_eval_tokens": 3298 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related but distinct PRs: one adds a Promptable Concept Segmentation pipeline, the other adds a Promptable Visual Segmentation pipeline. They share docs/registry plumbing, but the actual pipeline code, model coverage, and tests target different tasks and code paths.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No clear duplicate canonical PR; the two PRs implement different segmentation tasks with different pipeline modules and model integrations.", + "best_issue_reason": null, + "best_pr_reason": "PR #43613 is the broader/more general feature addition, covering multiple segmentation backends and corresponding tests/docs, so it is the better representative of the cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43612", + "right": "pull_request:43613", + "accept": false, + "reason": "Different underlying features: concept segmentation vs visual segmentation. Shared docs/registration changes are incidental; the pipeline implementations and model/test coverage are not the same concrete 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": 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": "8aa0d836135130b29b3942456726bbdb4e9e312a3a178889339493aed123aa64", + "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": "Both pull requests modify `src/transformers/trainer.py`, but they address different trainer extension points: 43651 changes gradient-accumulation loss scaling, while 43663 refactors signature-column filtering. They are related by file and subsystem, not by the same underlying bug or code path.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:43651", + "right": "pull_request:43663", + "accept": false, + "reason": "Different concrete changes in the same file: one fixes loss scaling during training, the other adds an overridable method for column filtering. They do not look mergeable into one PR or the same 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": 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": "9a2682decb04f4e94521f25b18a23bb1c3324814599b004252d30acc3e92185c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3663, + "estimated_input_tokens": 916, + "estimated_eval_tokens": 2088 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3663, + "estimated_input_tokens": 916, + "estimated_eval_tokens": 2088 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two open PRs touch the same CLI file, but they fix different annotation/runtime typing problems in separate code paths. They are related by symptom class, not duplicates of the same bug.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "No clear canonical PR: 43656 and 43842 address different underlying NameError causes in the same module, so they should stay separate rather than be collapsed into one canonical change.", + "best_issue_reason": null, + "best_pr_reason": "43842 is the tighter representative for the cluster because it directly targets the explicit TypeAdapter/pydantic NameError with a minimal scope and an issue link, whereas 43656 is a broader typing cleanup/future-annotations change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43656", + "right": "pull_request:43842", + "accept": false, + "reason": "Reject: these are distinct fixes. 43656 changes module annotations/import typing behavior broadly; 43842 specifically guards TypeAdapter behind TYPE_CHECKING/quotes the annotation to avoid a pydantic runtime import/name error. Same file, different underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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:44086|pull_request:44722", + "pull_request:44019|pull_request:44085", + "pull_request:44018|pull_request:44019", + "pull_request:44019|pull_request:44071", + "pull_request:44068|pull_request:44722", + "pull_request:44068|pull_request:44116", + "pull_request:44018|pull_request:44116", + "pull_request:44071|pull_request:44722", + "pull_request:44018|pull_request:44085", + "pull_request:44018|pull_request:44086", + "pull_request:44068|pull_request:44086", + "pull_request:44071|pull_request:44085", + "pull_request:44059|pull_request:44086", + "pull_request:44085|pull_request:44086", + "pull_request:44098|pull_request:44101", + "pull_request:44059|pull_request:44722", + "pull_request:44018|pull_request:44071", + "pull_request:44017|pull_request:44161", + "pull_request:44017|pull_request:44722", + "pull_request:44017|pull_request:44073", + "pull_request:44068|pull_request:44071", + "pull_request:43997|pull_request:44073", + "pull_request:44071|pull_request:44086", + "pull_request:44017|pull_request:44071", + "pull_request:44073|pull_request:44101", + "pull_request:44073|pull_request:44161", + "pull_request:44059|pull_request:44071", + "pull_request:44056|pull_request:44101", + "pull_request:44018|pull_request:44059", + "pull_request:44059|pull_request:44068", + "pull_request:44101|pull_request:44161", + "pull_request:43997|pull_request:44017", + "pull_request:43997|pull_request:44056", + "pull_request:44071|pull_request:44076", + "pull_request:44074|pull_request:44101", + "pull_request:43997|pull_request:44018", + "pull_request:44056|pull_request:44161", + "pull_request:43997|pull_request:44161", + "pull_request:44074|pull_request:44161", + "pull_request:44018|pull_request:44161", + "pull_request: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:44010|pull_request:44018", + "pull_request:43997|pull_request:44129", + "pull_request:44071|pull_request:44129", + "pull_request:43996|pull_request:44161", + "pull_request:43996|pull_request:44017", + "pull_request:44010|pull_request:44044", + "pull_request:44044|pull_request:44074", + "pull_request:44066|pull_request:44722", + "pull_request:44085|pull_request:44722", + "pull_request:44085|pull_request:44129", + "pull_request:43996|pull_request:44007", + "pull_request:43996|pull_request:44129", + "pull_request:44044|pull_request:44129", + "pull_request:44044|pull_request:44154" + ], + "prepared_review_unit_hash": "f4f0e9ae733ab8c27da6a967662a33d887a6f18c95035a5dc6ca2dde728f754f", + "component_budget": { + "node_count": 28, + "item_count": 28, + "soft_pair_count": 78, + "serialized_chars": 84896, + "estimated_input_tokens": 21224, + "estimated_eval_tokens": 42704 + }, + "budget": { + "node_count": 27, + "item_count": 27, + "soft_pair_count": 72, + "serialized_chars": 75093, + "estimated_input_tokens": 18774, + "estimated_eval_tokens": 37804 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a batch of model-by-model refactors for standardized output tracing. Most pairs are only thematically related and should not be merged as duplicates; the clear duplicate groups are the GPT-J/CodeGen PRs and the GPT-Neo PRs.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 44066 is the best canonical PR because it explicitly targets the tracking issue, makes a concrete GPT-J/CodeGen output-tracing refactor, and is the most central duplicate candidate among the GPT-family changes.", + "best_issue_reason": null, + "best_pr_reason": "PR 44066 is the strongest overall representative of the cluster: explicit issue linkage, coherent code-path changes, and broad enough scope to stand in for the shared GPT-J/CodeGen tracing work.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43996", + "right": "pull_request:44085", + "accept": false, + "reason": "Different model families and code paths (CVT/FNet vs GPT-J); same broad refactor theme is not enough." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44044", + "accept": false, + "reason": "Different models and implementations (CVT/FNet vs DeBERTa); not the same underlying change." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44085", + "accept": true, + "reason": "Both target GPT-J output tracing in the same file/code path; 44085 looks like a narrower variant of the same refactor." + }, + { + "left": "pull_request:44007", + "right": "pull_request:44072", + "accept": false, + "reason": "RegNet and EfficientNet are separate model implementations; only the refactor style overlaps." + }, + { + "left": "pull_request:44072", + "right": "pull_request:44722", + "accept": false, + "reason": "EfficientNet vs GPT-J/CodeGen are unrelated concrete code paths." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44072", + "accept": false, + "reason": "GPT-J/CodeGen output tracing and EfficientNet output tracing are different model-specific changes." + }, + { + "left": "pull_request:44013", + "right": "pull_request:44044", + "accept": false, + "reason": "MobileNetV2 and DeBERTa touch unrelated modules; not the same bug/change." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-J/CodeGen vs MGP-STR are different model refactors." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44071", + "accept": false, + "reason": "Different model families and different forward paths; same standardization effort only." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44068", + "accept": true, + "reason": "Both are GPT-Neo tracing refactors on the same model file/test surface, so they appear to be the same change." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44068", + "accept": false, + "reason": "GPT-J/CodeGen vs GPT-Neo are distinct model changes." + }, + { + "left": "pull_request:44019", + "right": "pull_request:44722", + "accept": false, + "reason": "ResNet and GPT-J/CodeGen are unrelated code paths." + }, + { + "left": "pull_request:44086", + "right": "pull_request:44722", + "accept": false, + "reason": "MGP-STR and GPT-J/CodeGen are separate model refactors." + }, + { + "left": "pull_request:44019", + "right": "pull_request:44085", + "accept": false, + "reason": "ResNet and GPT-J target different implementations and behaviors." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44019", + "accept": false, + "reason": "GPT-Neo and ResNet are unrelated models; not mergeable as one PR." + }, + { + "left": "pull_request:44019", + "right": "pull_request:44071", + "accept": false, + "reason": "ResNet and MPT do not share the same concrete bug/change." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44722", + "accept": false, + "reason": "GPT-Neo vs GPT-J/CodeGen are different model paths." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44116", + "accept": false, + "reason": "GPT-Neo and Flaubert/XLM are unrelated model implementations." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44116", + "accept": false, + "reason": "Different models and different output-tracing implementations." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44722", + "accept": false, + "reason": "MPT and GPT-J/CodeGen are unrelated concrete changes." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44085", + "accept": false, + "reason": "GPT-Neo and GPT-J are separate model-specific refactors." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-Neo and MGP-STR are unrelated modules." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-Neo vs MGP-STR are different underlying changes." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44086", + "accept": false, + "reason": "MPT and MGP-STR are different model code paths." + }, + { + "left": "pull_request:44059", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-2 and MGP-STR are unrelated implementations." + }, + { + "left": "pull_request:44085", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-J and MGP-STR are unrelated model changes." + }, + { + "left": "pull_request:44098", + "right": "pull_request:44101", + "accept": false, + "reason": "ViLT and Flaubert/XLM are different model families; same refactor theme only." + }, + { + "left": "pull_request:44059", + "right": "pull_request:44071", + "accept": false, + "reason": "GPT-2 and MPT are different model-specific output changes." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44101", + "accept": false, + "reason": "MPNet and Flaubert/XLM do not share the same concrete bug/change." + }, + { + "left": "pull_request:44074", + "right": "pull_request:44101", + "accept": false, + "reason": "TextNet and Flaubert/XLM are unrelated model implementations." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44017", + "accept": false, + "reason": "RegNet and SegFormer are separate model code paths." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44056", + "accept": false, + "reason": "RegNet and MPNet are unrelated changes." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44076", + "accept": false, + "reason": "MPT and ImageGPT are different models; not the same fix." + }, + { + "left": "pull_request:44074", + "right": "pull_request:44161", + "accept": false, + "reason": "TextNet and LongT5 are unrelated concrete changes." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44018", + "accept": false, + "reason": "RegNet and GPT-Neo are separate model refactors." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44161", + "accept": false, + "reason": "MPNet and LongT5 are different model implementations." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44161", + "accept": false, + "reason": "RegNet and LongT5 are unrelated code paths." + }, + { + "left": "pull_request:44074", + "right": "pull_request:44161", + "accept": false, + "reason": "TextNet and LongT5 are separate model-specific changes." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44017", + "accept": false, + "reason": "CVT/FNet and SegFormer are different models; only the tracing pattern overlaps." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44074", + "accept": false, + "reason": "CVT/FNet and TextNet are unrelated concrete changes." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44071", + "accept": false, + "reason": "SqueezeBERT and MPT are unrelated model implementations." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44073", + "accept": false, + "reason": "SqueezeBERT and VisualBert are distinct model code paths." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44722", + "accept": true, + "reason": "Both are GPT-J/CodeGen standardized output-tracing refactors on the same underlying code paths, so they are plausible duplicates." + }, + { + "left": "pull_request:44085", + "right": "pull_request:44722", + "accept": true, + "reason": "Both touch GPT-J output tracing in the same file and are close enough to be one combined PR; 44722 just adds the CodeGen companion change." + }, + { + "left": "pull_request:44085", + "right": "pull_request:44129", + "accept": false, + "reason": "GPT-J and SpeechT5 are unrelated models; not the same underlying change." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44161", + "accept": false, + "reason": "CVT/FNet and LongT5 are different model-specific refactors." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44007", + "accept": false, + "reason": "CVT/FNet and RegNet are separate model code paths." + }, + { + "left": "pull_request:44044", + "right": "pull_request:44129", + "accept": false, + "reason": "DeBERTa and SpeechT5 are unrelated model implementations." + }, + { + "left": "pull_request:44044", + "right": "pull_request:44154", + "accept": false, + "reason": "DeBERTa and VITS are different concrete changes despite similar refactor wording." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The result is grounded in the packet: the only clearly mergeable duplicate groups are GPT-J/CodeGen (44066, 44085, 44722) and GPT-Neo (44018, 44068), and the other false verdicts stay conservative by requiring the same concrete model/code path rather than just the same refactor theme. The canonical/best-PR choice of 44066 is also supported by its explicit issue link and broader GPT-J/CodeGen scope." + }, + "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": "d9516d39a30b89f919c27cc64cfbbe51d7aa234509ac72d0dfb5b9844d22e884", + "component_budget": { + "node_count": 28, + "item_count": 28, + "soft_pair_count": 78, + "serialized_chars": 84896, + "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": "These PRs are all architecture-specific refactors toward standardized output tracing/capture, but they affect different model code paths and are not duplicate fixes/changes. They should remain separate.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #44073 is a clean representative of the cluster: it applies the standardized capture_outputs/can_return_tuple pattern in a single model with a clear, focused refactor and explicit output-tracing setup.", + "best_issue_reason": null, + "best_pr_reason": "PR #44073 is the best representative PR for this cluster because it is well-scoped, clearly titled, and directly demonstrates the standardized output-tracing pattern without being broader than the others.", + "soft_edge_verdicts": [ + { + "left": "pull_request:44024", + "right": "pull_request:44076", + "accept": false, + "reason": "Both are output-tracing refactors, but FocalNet and ImageGPT are different models with different forward paths and output wiring; not the same underlying change." + }, + { + "left": "pull_request:44073", + "right": "pull_request:44074", + "accept": false, + "reason": "VisualBert vs TextNet are separate model implementations; these are similar refactors, not one concrete shared bug or patch." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44073", + "accept": false, + "reason": "MPNet and VisualBert each add model-specific _can_record_outputs/capture_outputs plumbing in different files and code paths; not duplicates." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44074", + "accept": false, + "reason": "MPNet and TextNet are distinct architectures with separate refactors; similarity is only at the pattern level." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44010", + "accept": false, + "reason": "RegNet and SqueezeBert touch unrelated model implementations; both are standardization work, but not the same bug or change." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44129", + "accept": false, + "reason": "SqueezeBert and SpeechT5 are different model stacks and the PRs address model-specific output tracing changes, not one mergeable fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "774f3bde0e2832123e9fd7d70575a3bfe3f717b1d5139aef6afb22892cdc2cfc", + "component_budget": { + "node_count": 13, + "item_count": 13, + "soft_pair_count": 40, + "serialized_chars": 40177, + "estimated_input_tokens": 10045, + "estimated_eval_tokens": 20346 + }, + "budget": { + "node_count": 13, + "item_count": 13, + "soft_pair_count": 40, + "serialized_chars": 40177, + "estimated_input_tokens": 10045, + "estimated_eval_tokens": 20346 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are all related output-tracing refactors, but each PR targets a different model family and code path, so they are not duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the cluster is a set of similar-but-separate model-specific refactors, not one underlying fix/change.", + "best_issue_reason": null, + "best_pr_reason": "PR 44025 is the best single representative because it shows the refactor pattern clearly across two closely related models, but it still isn't a duplicate of the others.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43998", + "right": "pull_request:44002", + "accept": false, + "reason": "Different model families and code paths; same refactor theme only." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44000", + "accept": false, + "reason": "Unrelated model implementations; not the same underlying bug or change." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44001", + "accept": false, + "reason": "Different architectures; only shares the output-tracing refactor pattern." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44003", + "accept": false, + "reason": "Separate model code paths; too broad to be one duplicate PR." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44029", + "accept": false, + "reason": "Different model implementation; not a mergeable duplicate." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44001", + "accept": false, + "reason": "Different model families; same refactoring style is not enough." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44002", + "accept": false, + "reason": "Different code paths and model goals; not the same fix." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44025", + "accept": false, + "reason": "Different model families; no shared concrete bug/change." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44026", + "accept": false, + "reason": "Separate wrappers/models; not the same underlying change." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44027", + "accept": false, + "reason": "Different encoder-decoder implementation; only superficially similar." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44001", + "accept": false, + "reason": "Distinct model implementations; both are refactors, not duplicates." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44002", + "accept": false, + "reason": "Different models and files; same modernization pattern only." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44003", + "accept": false, + "reason": "Different architecture and code path; not one concrete change." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44025", + "accept": false, + "reason": "Different model families; no shared bug to collapse." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44002", + "accept": false, + "reason": "Separate model-specific refactors; not the same PR-worthy fix." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44003", + "accept": false, + "reason": "Different sequence-model implementations; too broad for duplication." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44025", + "accept": false, + "reason": "Different models; only the tracing refactor theme matches." + }, + { + "left": "pull_request:44002", + "right": "pull_request:44029", + "accept": false, + "reason": "Unrelated models and code paths; not the same underlying issue." + }, + { + "left": "pull_request:44002", + "right": "pull_request:44003", + "accept": false, + "reason": "Different architectures; same refactor label is insufficient." + }, + { + "left": "pull_request:44003", + "right": "pull_request:44028", + "accept": false, + "reason": "Falcon/Mamba refactor vs SuperPoint refactor are unrelated changes." + }, + { + "left": "pull_request:44003", + "right": "pull_request:44004", + "accept": false, + "reason": "Different model code and behavior; not a duplicate fix." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44004", + "accept": false, + "reason": "Different model families and implementation details; only broad similarity." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44004", + "accept": false, + "reason": "Unrelated model code paths; not mergeable as one PR." + }, + { + "left": "pull_request:44002", + "right": "pull_request:44004", + "accept": false, + "reason": "Different models; same general cleanup pattern only." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44004", + "accept": false, + "reason": "Distinct model implementations; no shared concrete bug." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44026", + "accept": false, + "reason": "Depth Anything vs Vision-Encoder-Decoder are separate model families." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44027", + "accept": false, + "reason": "Different model wrappers and code paths; not the same fix." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44028", + "accept": false, + "reason": "Separate vision model implementations; same pattern only." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44029", + "accept": false, + "reason": "Different architecture and module behavior; not a duplicate." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44030", + "accept": false, + "reason": "Different model code paths; same refactor theme is too broad." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44029", + "accept": false, + "reason": "Different models and runtime behavior; not the same change." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44027", + "accept": false, + "reason": "Encoder-decoder wrappers for different modalities; not duplicates." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44028", + "accept": false, + "reason": "Different model families; no shared concrete bug." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44030", + "accept": false, + "reason": "Different model implementations; only broad similarity." + }, + { + "left": "pull_request:44027", + "right": "pull_request:44028", + "accept": false, + "reason": "Speech encoder-decoder vs SuperPoint are unrelated code paths." + }, + { + "left": "pull_request:44027", + "right": "pull_request:44029", + "accept": false, + "reason": "Different model families; not the same underlying issue." + }, + { + "left": "pull_request:44027", + "right": "pull_request:44030", + "accept": false, + "reason": "Separate implementations with different behavior; not a duplicate." + }, + { + "left": "pull_request:44028", + "right": "pull_request:44029", + "accept": false, + "reason": "Different model types; same tracing pattern only." + }, + { + "left": "pull_request:44028", + "right": "pull_request:44030", + "accept": false, + "reason": "Unrelated model code paths; too broad to accept." + }, + { + "left": "pull_request:44029", + "right": "pull_request:44030", + "accept": false, + "reason": "Different models and outputs; not the same fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "f99ab1f0a5e666ba59b13557a9adb7941322d07507e545a1a9c8981a088481f0", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6551, + "estimated_input_tokens": 1638, + "estimated_eval_tokens": 3532 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6551, + "estimated_input_tokens": 1638, + "estimated_eval_tokens": 3532 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two different model-integration PRs: one adds PenguinVL, the other adds Voxtral TTS. They overlap in shared Transformers plumbing, but they do not appear to fix the same underlying code path or change set.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR for a duplicate cluster: the two PRs target different models and different integration work.", + "best_issue_reason": null, + "best_pr_reason": "PR #45401 is the better representative only because it is more narrowly scoped and has an explicit issue target, but it is still unrelated to #44662.", + "soft_edge_verdicts": [ + { + "left": "pull_request:44662", + "right": "pull_request:45401", + "accept": false, + "reason": "Different features for different models (PenguinVL vs Voxtral TTS); shared auto/registry files are framework boilerplate, not evidence of the same bug or mergeable 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": "ae042fdf44ec47ea55765f264ebfe72fad97c1eea3204eb4b1b26b834e4bf00e", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4408, + "estimated_input_tokens": 1102, + "estimated_eval_tokens": 2460 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4408, + "estimated_input_tokens": 1102, + "estimated_eval_tokens": 2460 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are related thematically around fusion, but they implement different layers and use cases: one introduces a generic Module Fusion API, while the other adds a specific n-to-1 kernel fusion path via KernelConfig and integration changes. They are not duplicates.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #44979 is the cleaner foundational change: it adds a new module fusion API and tests without the extra integration surface of the draft kernel-config PR.", + "best_issue_reason": null, + "best_pr_reason": "PR #44979 is the best representative artifact in this cluster because it is more self-contained, non-draft, and establishes the core abstraction.", + "soft_edge_verdicts": [ + { + "left": "pull_request:44979", + "right": "pull_request:45363", + "accept": false, + "reason": "Different underlying changes: #44979 defines a generic module fusion API, while #45363 implements specific n-to-1 kernel fusion through KernelConfig and model-loading integrations. They would not plausibly be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "bc4452be8c2e014f0625fabf3ac559c57d3eda8bb09d993d9b14454f37e138fd", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4414, + "estimated_input_tokens": 1104, + "estimated_eval_tokens": 2464 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4414, + "estimated_input_tokens": 1104, + "estimated_eval_tokens": 2464 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs fix the same AttributeError in `_patch_mistral_regex` by switching from `backend_tokenizer.pre_tokenizer` to `tokenizer.pre_tokenizer`; `45317` is the more complete version because it adds test coverage.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 45317 is the stronger canonical candidate: it addresses the same concrete bug as 45086 and also includes a test change, making it the more complete fix.", + "best_issue_reason": null, + "best_pr_reason": "PR 45317 is the best PR candidate because it fixes the same code-path bug and adds validation in tests, whereas 45086 is the minimal patch only.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45086", + "right": "pull_request:45317", + "accept": true, + "reason": "Same underlying bug and same fix in the same file, both tied to issue 45081; 45317 is essentially a superset with test coverage." + } + ] + }, + "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-45115-2", + "nodes": [ + "pull_request:45115", + "pull_request:45569" + ], + "soft_pairs": [ + "pull_request:45115|pull_request:45569" + ], + "prepared_review_unit_hash": "bcefe9c0f546358a88049052d477e1672f70b8fa6bb5f26dcad260958eea0a8a", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6066, + "estimated_input_tokens": 1517, + "estimated_eval_tokens": 3290 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6066, + "estimated_input_tokens": 1517, + "estimated_eval_tokens": 3290 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are related NemotronH architecture changes, but they are not the same fix/change: #45115 is a refactor toward GraniteMoeHybrid-compatible conversion and inheritance, while #45569 is a broader implementation split introducing NemotronHDense/NemotronHSparse plus docs, auto-mapping, and tests.", + "confidence": 0.82, + "canonical_issue_reason": null, + "canonical_pr_reason": "#45569 is the more complete and representative NemotronH code change: it adds the dense/sparse model packages, updates auto mappings and docs, and includes test coverage, so it better captures the final architecture work than the narrower refactor in #45115.", + "best_issue_reason": null, + "best_pr_reason": "#45569 is the best candidate to anchor this cluster because it is the broader end-state implementation and subsumes the model-structure direction that #45115 is working toward.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45115", + "right": "pull_request:45569", + "accept": false, + "reason": "They share the NemotronH area and some files, but the concrete changes differ: #45115 is a conversion/inheritance refactor, while #45569 introduces separate dense and sparse model variants with docs and tests. They do not look like one mergeable PR for 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": "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": "e5cfa185df3e02d4bd4128c514a77e9b18815c4cc633eaa7f65a0890eda62ea0", + "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": "The two PRs are closely related parts of a larger heterogeneous-model effort, but they address different layers of the stack: one adds per-layer config support, the other adds per-layer modeling/cache behavior. They are not the same underlying change and would not be merged as a single duplicate PR.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 45333 is the more foundational change because it introduces per-layer configuration support, which the modeling PR builds on.", + "best_issue_reason": null, + "best_pr_reason": "PR 45333 is the best representative of this cluster because it defines the core heterogeneous-config API and is the enabling piece for the broader feature.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45332", + "right": "pull_request:45333", + "accept": false, + "reason": "Related feature work, but not the same concrete change: 45333 adds heterogeneous per-layer config, while 45332 implements heterogeneous model/cache behavior on top of that. They are complementary, 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": 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": "7dd0e54a547a87e57bab0076f600577ceb8c84997b0964575750743cb2c519bd", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5513, + "estimated_input_tokens": 1379, + "estimated_eval_tokens": 3014 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5513, + "estimated_input_tokens": 1379, + "estimated_eval_tokens": 3014 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests are typing-related and overlap in a few files, but they address different changesets: #45415 is a broad repository-wide type-checking sweep, while #45425 adds shared typing protocol/helpers for specific integrations. They are not the same underlying bug/change and should not be deduplicated.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "No strong duplicate canonical PR here; if forced, #45425 is the narrower, more concrete typing helper change, while #45415 is a broader type-checking sweep.", + "best_issue_reason": null, + "best_pr_reason": "#45425 is the better representative of the cluster\u2019s more specific typing work because it introduces shared protocol/type aliases tied to particular code paths, whereas #45415 is a wide mechanical type-checking pass across many files.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45415", + "right": "pull_request:45425", + "accept": false, + "reason": "Both are typing-related and share some touched files, but #45415 is a broad mypy/pyright cleanup across many modules, while #45425 adds specific typing protocols/aliases for modeling_utils and PEFT-related code. Different scope and not mergeable as one duplicate change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "78ee2bf4b42f4af3d46f543da3f62dfc83b517bba01f1db5e89f0e2dc04f4326", + "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 are both cache/seq_len>1 bug-fix PRs, but they target different model families and different execution paths. The similarity is only at a broad symptom level, not the same underlying change.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #45513 is the broader representative: it fixes multi-token cached forward across several Qwen3-family model files and includes the underlying control-flow change for cached continuation.", + "best_issue_reason": null, + "best_pr_reason": "PR #45513 is the best representative of the cluster because it spans multiple related model implementations and addresses the shared cached-forward logic directly.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45513", + "right": "pull_request:46084", + "accept": false, + "reason": "Different concrete code paths and model families: Qwen3/OLMo hybrid linear attention cached forward vs Mamba2 CUDA kernel cache handling. They share the general theme of seq_len>1 cache bugs, but not the same underlying fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "159284bfeb8921cfbd03edfdd74e67a8aa897bf17eb18e7c53e01887fcdf7b2a", + "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 PRs are related by expert-parallel loading, but they fix different failure modes on different code paths: one addresses EP + DeepSpeed ZeRO-3/accelerate launch config handling, the other addresses EP + FSDP2 expert parameters being overwritten during sharded loading. They are not duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #45662 is the stronger representative of the cluster because it targets a concrete loading corruption bug and includes the broader set of loader-side fixes around sharded parameter handling; PR #45548 is narrower and focused on config serialization plus an EP capability flag for a different launch path.", + "best_issue_reason": null, + "best_pr_reason": "PR #45662 is the better standalone PR for a canonical bug-fix anchor since it fixes a specific expert-parameter overwrite path with direct loading consequences, whereas #45548 is a separate accelerate/ZeRO-3 compatibility fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45548", + "right": "pull_request:45662", + "accept": false, + "reason": "Both mention EP and touch modeling_utils, but they fix different bugs: #45548 is about serializing distributed_config and exposing has_ep for DeepSpeed ZeRO-3/accelerate launch, while #45662 fixes post-shard wrapping to prevent experts from being overwritten under FSDP2. Different concrete code paths and not safely mergeable into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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": "8822e831a7c5a45aa89bb32e2b69f5aca680fa2d296f394f2b3f00a94c8cff3e", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4590, + "estimated_input_tokens": 1148, + "estimated_eval_tokens": 2552 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4590, + "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 in `testing_utils.py`; PR #45645 is the broader, more complete version that also updates CI log collection.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #45645 is the better canonical PR because it covers the core xdist-safe `captured_info` fix plus the downstream CI/workflow changes needed to preserve and collect the split log files.", + "best_issue_reason": null, + "best_pr_reason": "PR #45645 is the strongest representative of the cluster: same underlying bug as #45639, but with the full end-to-end handling across test utilities and CI artifact collection.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45639", + "right": "pull_request:45645", + "accept": true, + "reason": "Same underlying xdist `captured_info` collision fix in `testing_utils.py`; #45645 appears to extend and supersede the narrower implementation in #45639." + } + ] + }, + "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-45720-2", + "nodes": [ + "pull_request:45720", + "pull_request:46259" + ], + "soft_pairs": [ + "pull_request:45720|pull_request:46259" + ], + "prepared_review_unit_hash": "899ac2bbbbea35d3e3eb4dec737526106082b63916ca4bee3c24b866322ce783", + "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 are both transformers modeling lint/chore PRs, but they address different rules and different code paths: PR #45720 adds `can_return_tuple`/return-dict handling across model forwards, while PR #46259 changes `_init_weights` behavior for a broad set of models. The overlap in filenames looks incidental, not duplicate work.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #45720 is the better anchor because it is a narrower, more concrete code change with a clearly scoped rule implementation (TRF003) and a consistent forward-path modification pattern.", + "best_issue_reason": null, + "best_pr_reason": "PR #45720 is the best representative of the cluster because it is more focused and easier to triage as a single change, whereas PR #46259 targets a different lint rule and a different initialization path.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45720", + "right": "pull_request:46259", + "accept": false, + "reason": "Reject: they implement different linter rules on different code paths (`forward` return handling vs `_init_weights` initialization). Shared model filenames do not indicate the same underlying fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 25, + "component_count": 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": 26, + "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": "6a0648c14b70b48510cfeb02f3611e3ba8fd1ef446db9ba1e57fff3455b72f78", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5161, + "estimated_input_tokens": 1291, + "estimated_eval_tokens": 2838 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5161, + "estimated_input_tokens": 1291, + "estimated_eval_tokens": 2838 + }, + "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 Grafana/test CI and model-configuration miscellany PR, while the other is a test migration focused on kernel and vision-model tests. They touch different changesets and are not plausibly one merged fix/change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "pull_request:46100 is the more coherent, self-contained change set (test migration and kernel test additions), making it the better cluster representative.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:46100 is the best representative PR because it has a single, clear purpose and a tighter scope than pull_request:45932, which mixes CI/workflow and unrelated model edits.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45932", + "right": "pull_request:46100", + "accept": false, + "reason": "They are different kinds of changes affecting different areas: 45932 is a mixed workflow/model tweak PR, while 46100 is a test migration with new kernel and vision tests. They do not look like the same underlying bug or change and would not sensibly 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": 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-46126-2", + "nodes": [ + "pull_request:46126", + "pull_request:46269" + ], + "soft_pairs": [ + "pull_request:46126|pull_request:46269" + ], + "prepared_review_unit_hash": "0d8835e117242dab959ba7b91e349ae861aedb2433b6d18efcd95132845634ce", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5522, + "estimated_input_tokens": 1381, + "estimated_eval_tokens": 3018 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5522, + "estimated_input_tokens": 1381, + "estimated_eval_tokens": 3018 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are not duplicates: one PR is a focused EP/MoE + Trainer integration change, while the other is a broad distributed-branch base refactor touching many unrelated distributed components. The shared filenames are too generic to imply the same underlying fix.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #46126 is the more canonical candidate because it is narrowly scoped to EP and Trainer integration on top of DistributedConfig, with a concrete, coherent feature intent.", + "best_issue_reason": null, + "best_pr_reason": "PR #46126 is the best standalone PR because it targets a specific code path and feature set, whereas PR #46269 is a broad integration/base branch with many disparate changes.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46126", + "right": "pull_request:46269", + "accept": false, + "reason": "Reject: despite overlapping distributed filenames, the PRs have different goals and scopes. #46126 adds EP/Trainer integration and router/tensor-parallel behavior, while #46269 is a large distributed branch base touching FSDP, generation, integrations, and model loading. They do not look like the same concrete bug or mergeable single 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": 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-46138-10", + "nodes": [ + "pull_request:46138", + "pull_request:46432", + "pull_request:46501", + "pull_request:46502", + "pull_request:46503", + "pull_request:46504", + "pull_request:46554", + "pull_request:46558", + "pull_request:46563", + "pull_request:46716" + ], + "soft_pairs": [ + "pull_request:46138|pull_request:46554", + "pull_request:46138|pull_request:46558", + "pull_request:46138|pull_request:46563", + "pull_request:46138|pull_request:46716", + "pull_request:46138|pull_request:46504", + "pull_request:46432|pull_request:46503", + "pull_request:46432|pull_request:46501", + "pull_request:46432|pull_request:46502", + "pull_request:46432|pull_request:46554", + "pull_request:46432|pull_request:46558", + "pull_request:46501|pull_request:46503", + "pull_request:46501|pull_request:46502", + "pull_request:46501|pull_request:46504", + "pull_request:46501|pull_request:46554", + "pull_request:46502|pull_request:46503", + "pull_request:46502|pull_request:46504", + "pull_request:46502|pull_request:46554", + "pull_request:46503|pull_request:46504", + "pull_request:46503|pull_request:46554", + "pull_request:46504|pull_request:46554", + "pull_request:46504|pull_request:46558", + "pull_request:46554|pull_request:46558", + "pull_request:46554|pull_request:46563", + "pull_request:46554|pull_request:46716", + "pull_request:46558|pull_request:46563", + "pull_request:46558|pull_request:46716", + "pull_request:46563|pull_request:46716", + "pull_request:46504|pull_request:46563", + "pull_request:46504|pull_request:46716" + ], + "prepared_review_unit_hash": "ee65201344f36eea5359dae4ec2081b14ec88c60fe3e93ec67946de5030cb12d", + "component_budget": { + "node_count": 10, + "item_count": 10, + "soft_pair_count": 29, + "serialized_chars": 25926, + "estimated_input_tokens": 6482, + "estimated_eval_tokens": 13220 + }, + "budget": { + "node_count": 10, + "item_count": 10, + "soft_pair_count": 29, + "serialized_chars": 25926, + "estimated_input_tokens": 6482, + "estimated_eval_tokens": 13220 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are all workflow-maintenance changes, but they split into unrelated subtopics: several distinct CI workflow tweaks and a separate set of different security-gate edits. The shared filenames/title pattern creates superficial similarity, but the concrete changes do not match well enough to treat any pair as duplicates.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "No true duplicate PR stands out; PR 46554 is the best representative of the cluster because it makes the most substantive security-gate workflow change.", + "best_issue_reason": null, + "best_pr_reason": "PR 46554 is the strongest representative of the set: it has concrete runtime behavior changes in the security-gate workflow, whereas most others are smaller, unrelated workflow tweaks.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46138", + "right": "pull_request:46554", + "accept": false, + "reason": "Different workflows and unrelated changes; this is only a broad CI-maintenance similarity." + }, + { + "left": "pull_request:46138", + "right": "pull_request:46558", + "accept": false, + "reason": "Different workflows and different fixes; no shared underlying bug or change." + }, + { + "left": "pull_request:46138", + "right": "pull_request:46563", + "accept": false, + "reason": "Different workflows and unrelated edits; both are workflow chores, not duplicates." + }, + { + "left": "pull_request:46138", + "right": "pull_request:46716", + "accept": false, + "reason": "Different workflows and different behavior changes; not the same PR." + }, + { + "left": "pull_request:46138", + "right": "pull_request:46504", + "accept": false, + "reason": "Different workflow files and different shell-handling changes; only the maintenance style matches." + }, + { + "left": "pull_request:46432", + "right": "pull_request:46503", + "accept": false, + "reason": "Different workflow files and unrelated CI behavior changes." + }, + { + "left": "pull_request:46432", + "right": "pull_request:46501", + "accept": false, + "reason": "Different workflow files and different operational concerns; not duplicates." + }, + { + "left": "pull_request:46432", + "right": "pull_request:46502", + "accept": false, + "reason": "Different workflow files and different changes; same subsystem is too broad." + }, + { + "left": "pull_request:46432", + "right": "pull_request:46554", + "accept": false, + "reason": "SSH runner hardening vs PR CI security-gate logic; unrelated code paths." + }, + { + "left": "pull_request:46432", + "right": "pull_request:46558", + "accept": false, + "reason": "Different workflow areas and different fixes; no single underlying change." + }, + { + "left": "pull_request:46501", + "right": "pull_request:46503", + "accept": false, + "reason": "Both touch workflows, but they change different jobs and behaviors." + }, + { + "left": "pull_request:46501", + "right": "pull_request:46502", + "accept": false, + "reason": "Same broad workflow-maintenance theme, but unrelated workflow files and changes." + }, + { + "left": "pull_request:46501", + "right": "pull_request:46504", + "accept": false, + "reason": "Different workflow files and different fixes; not the same bug/change." + }, + { + "left": "pull_request:46501", + "right": "pull_request:46554", + "accept": false, + "reason": "Unrelated workflow updates; one is anti-slop, the other is PR CI security gating." + }, + { + "left": "pull_request:46502", + "right": "pull_request:46503", + "accept": false, + "reason": "Different workflow files and different trigger/behavior changes." + }, + { + "left": "pull_request:46502", + "right": "pull_request:46504", + "accept": false, + "reason": "Different workflow files and different shell/input handling changes." + }, + { + "left": "pull_request:46502", + "right": "pull_request:46554", + "accept": false, + "reason": "Assign-reviewers vs PR CI security-gate are unrelated workflow changes." + }, + { + "left": "pull_request:46503", + "right": "pull_request:46504", + "accept": false, + "reason": "Different workflow files and different edits; same CI domain is too broad." + }, + { + "left": "pull_request:46503", + "right": "pull_request:46554", + "accept": false, + "reason": "Benchmark trigger logic vs PR CI security checks are unrelated." + }, + { + "left": "pull_request:46504", + "right": "pull_request:46554", + "accept": false, + "reason": "Build-image trigger handling vs PR CI security-gate logic are different changes." + }, + { + "left": "pull_request:46504", + "right": "pull_request:46558", + "accept": false, + "reason": "Different workflow files and different behavior changes; not duplicates." + }, + { + "left": "pull_request:46554", + "right": "pull_request:46558", + "accept": false, + "reason": "Same file, but different concerns: workflow-file checks vs permissions/approval gating." + }, + { + "left": "pull_request:46554", + "right": "pull_request:46563", + "accept": false, + "reason": "Same file, but one changes gate logic and the other adds explanatory comments; not the same fix." + }, + { + "left": "pull_request:46554", + "right": "pull_request:46716", + "accept": false, + "reason": "Same file, but different security-gate adjustments: checks vs secrets handling." + }, + { + "left": "pull_request:46558", + "right": "pull_request:46563", + "accept": false, + "reason": "Both in the same workflow file, but they address different aspects of the gate." + }, + { + "left": "pull_request:46558", + "right": "pull_request:46716", + "accept": false, + "reason": "Permissions/approval gating vs secrets scoping are different changes." + }, + { + "left": "pull_request:46563", + "right": "pull_request:46716", + "accept": false, + "reason": "Comment-only clarification vs secrets handling; not the same underlying change." + }, + { + "left": "pull_request:46504", + "right": "pull_request:46563", + "accept": false, + "reason": "Different workflow files and unrelated CI maintenance changes." + }, + { + "left": "pull_request:46504", + "right": "pull_request:46716", + "accept": false, + "reason": "Different workflow files and different security/CI concerns; 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-46152-2", + "nodes": [ + "pull_request:46152", + "pull_request:46207" + ], + "soft_pairs": [ + "pull_request:46152|pull_request:46207" + ], + "prepared_review_unit_hash": "0b07ab9c2abd60b54bd9d9c0ae14502e8aa1b79f7618f0aa131e30b1f3b103ff", + "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": "Both pull requests are about eliminating brittle source-file open() access, but they fix different code paths in different modules, so they are not duplicates and should not be merged together.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46207 is the stronger standalone change: it targets a specific source-reading heuristic in `modeling_utils.py`, includes a focused test update, and is more concrete than the import-utils-only fix in 46152.", + "best_issue_reason": null, + "best_pr_reason": "PR 46207 is the best representative of this cluster because it has a clearly scoped runtime behavior change and test coverage, whereas 46152 addresses a separate utility-path implementation.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46152", + "right": "pull_request:46207", + "accept": false, + "reason": "Both avoid reading source via `open()`, but 46152 changes `import_utils._resolve_traversable` while 46207 changes `modeling_utils._can_set_attn_implementation`; these are different concrete bugs/code paths and would not plausibly be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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-46154-2", + "nodes": [ + "pull_request:46154", + "pull_request:46196" + ], + "soft_pairs": [ + "pull_request:46154|pull_request:46196" + ], + "prepared_review_unit_hash": "bc902838473adbcaf4340697a008df6eba12291f1f94ef38d48ebd8842bb990a", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5700, + "estimated_input_tokens": 1425, + "estimated_eval_tokens": 3106 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5700, + "estimated_input_tokens": 1425, + "estimated_eval_tokens": 3106 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests add SDPA attention dispatch, but they target different model families and code paths: T5 versus wav2vec2/audio models. They are thematically similar but not the same underlying change, so they should not be merged as duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: 46154 is a T5-specific SDPA dispatch change, while 46196 is a separate, broader audio-model SDPA dispatch effort across multiple wav2vec2-related architectures.", + "best_issue_reason": null, + "best_pr_reason": "If one PR must be highlighted as the broader representative of this SDPA rollout, 46196 is the more comprehensive change set; however, it is still not a duplicate of 46154.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46154", + "right": "pull_request:46196", + "accept": false, + "reason": "Different concrete implementations and model families. T5 attention dispatch and wav2vec2/audio attention dispatch are separate code paths, so they are not mergeable as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 31, + "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-46173-2", + "nodes": [ + "pull_request:46173", + "pull_request:46474" + ], + "soft_pairs": [ + "pull_request:46173|pull_request:46474" + ], + "prepared_review_unit_hash": "8ac88de891156efc041ae6eac2ff2b6f957f37c7ee42f44803b224f91f9b44ec", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3631, + "estimated_input_tokens": 908, + "estimated_eval_tokens": 2072 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3631, + "estimated_input_tokens": 908, + "estimated_eval_tokens": 2072 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests make the same code change in `src/transformers/cache_utils.py` to widen `StaticLayer.get_seq_length` from `int` to `int | torch.Tensor`; they are near-duplicate implementations of the same fix for issue #45987.", + "confidence": 0.99, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #46173 is the better canonical representative because it is the earlier, more explicitly titled version of the same fix and already points to the tracking issue #45987.", + "best_issue_reason": null, + "best_pr_reason": "PR #46173 is the strongest standalone representative of the change: it contains the same minimal patch, the same target issue, and a clearer description of the intended fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46173", + "right": "pull_request:46474", + "accept": true, + "reason": "Same underlying bug fix, same file, same diff hunk, and same explicit issue target (#45987); these are functionally duplicate PRs." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: both PRs touch the same file, make the same type-annotation change for `StaticLayer.get_seq_length`, and both explicitly target issue #45987. The canonical-pr rationale is also conservative and supported by the titles/metadata. No overstatement beyond the visible diff evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "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-46310-2", + "nodes": [ + "pull_request:46310", + "pull_request:46546" + ], + "soft_pairs": [ + "pull_request:46310|pull_request:46546" + ], + "prepared_review_unit_hash": "0580782e39049fb5fc7bcb9dbb2b9172024f2b945066699300fa26ef2b88dd1e", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4851, + "estimated_input_tokens": 1213, + "estimated_eval_tokens": 2682 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4851, + "estimated_input_tokens": 1213, + "estimated_eval_tokens": 2682 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs fix the same off-by-one bug in the sliding-window right boundary (`-dist <= right_window_size` instead of `<`) for Moonshine streaming. PR 46546 is the more complete change set because it also adds the parallel Gemma4 fix and tests, while 46310 is the narrower Moonshine-only patch.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46546 is the better canonical PR: it contains the Moonshine streaming boundary fix plus the related Gemma4 update and tests, making it the more complete implementation of the same bug fix.", + "best_issue_reason": null, + "best_pr_reason": "PR 46546 is the strongest representative of the fix because it addresses the boundary bug across all affected code paths and includes test coverage.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46310", + "right": "pull_request:46546", + "accept": true, + "reason": "Both PRs correct the same right-window off-by-one boundary in the same inner_mask logic; 46546 is just a broader, more complete version of the same fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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-46376-2", + "nodes": [ + "pull_request:46376", + "pull_request:46417" + ], + "soft_pairs": [ + "pull_request:46376|pull_request:46417" + ], + "prepared_review_unit_hash": "cf8c6d7e2c3963198e26621f078a33a72c38829a191779487ffc6a613a30c922", + "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 internals, while the other adds the HunYuan VL model, docs, processors, and tests. They do not look like the same underlying change or bug.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "PR #46417 is the more complete, self-contained feature addition with docs and tests, so it is the better representative PR for the cluster, even though it is not a duplicate of #46376.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46376", + "right": "pull_request:46417", + "accept": false, + "reason": "Different subsystems and goals: continuous batching multimodal runtime changes vs a new model implementation/docs/tests. No shared filenames, targets, or concrete code-path overlap suggesting a single 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": 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-46574-2", + "nodes": [ + "pull_request:46574", + "pull_request:46580" + ], + "soft_pairs": [ + "pull_request:46574|pull_request:46580" + ], + "prepared_review_unit_hash": "fb7049813207caa05053baee268914f7423f6aeb7188471c4972a6475caa8327", + "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 are two unrelated tokenizer-loading fixes: one changes CodeLlama decode behavior to preserve leading spaces, the other fixes DeepSeek-Coder/DeepSeek-LLM auto-tokenizer selection for certain checkpoints. They are not the same bug or mergeable change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: the two pull requests fix different code paths and different underlying failures.", + "best_issue_reason": null, + "best_pr_reason": "PR #46580 is the better representative of the cluster only in the sense that it addresses a broader auto-tokenizer misclassification problem; however, it is not a duplicate of #46574.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46574", + "right": "pull_request:46580", + "accept": false, + "reason": "Different failures and fixes: CodeLlama decoder stripping spaces on decode vs. DeepSeek checkpoint-to-tokenizer-class selection in auto tokenization." + } + ] + }, + "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-46603-2", + "nodes": [ + "pull_request:46603", + "pull_request:46604" + ], + "soft_pairs": [ + "pull_request:46603|pull_request:46604" + ], + "prepared_review_unit_hash": "b0dee0e19f87d47edd1c689847fa8d395df21f1b801f3e141a86bd42b11975a0", + "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": "Both PRs touch the Mistral tokenizer integration, but they address different changes: #46603 is a refactor that moves `MistralConverter` into the new integration package, while #46604 adds tekken tokenizer detection, conversion, and save utilities. They are related codebase areas, not the same fix/change.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #46604 is the more substantive and representative Mistral change: it introduces new tokenizer behavior and utilities, whereas #46603 is mainly a code relocation/refactor.", + "best_issue_reason": null, + "best_pr_reason": "PR #46604 best represents the cluster because it covers the broader functional work in the Mistral integration and is more likely to be the primary user-facing change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46603", + "right": "pull_request:46604", + "accept": false, + "reason": "Reject: they share the Mistral integration area and files, but #46603 is a package move/refactor while #46604 implements tekken tokenizer detection/conversion/save utilities. Different concrete changes, not one underlying bug or mergeable single PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 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-46741-2", + "nodes": [ + "pull_request:46741", + "pull_request:46748" + ], + "soft_pairs": [ + "pull_request:46741|pull_request:46748" + ], + "prepared_review_unit_hash": "10614e98502db49184d921c7bf8bb3ac912ff1424cf806e251a5d4e898bed70a", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4434, + "estimated_input_tokens": 1109, + "estimated_eval_tokens": 2474 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4434, + "estimated_input_tokens": 1109, + "estimated_eval_tokens": 2474 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two PRs address different bugs in different code paths: one fixes Mamba2-style chunked-prefill/speculative decoding in specific model implementations, the other fixes cache prefetch/device-mismatch behavior for offloaded hybrid models. They are not duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #46748 is the better canonical representative because it has a single, focused fix in cache.prefetch plus tests, making it more self-contained than the model-specific chunked-prefill changes in #46741.", + "best_issue_reason": null, + "best_pr_reason": "PR #46748 is the strongest standalone PR in the cluster: it targets one concrete cache/offload bug with a narrow implementation surface and validation coverage.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46741", + "right": "pull_request:46748", + "accept": false, + "reason": "Different underlying problems: #46741 fixes chunked-prefill/speculative decoding behavior in Nemotron-H/Zamba2 Mamba2 mixer code, while #46748 fixes offloaded cache prefetch/device mismatch in cache_utils for hybrid models." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + } + ] +} diff --git a/snapshots/20260619T120025Z/analysis-runs/hybrid-model-20260619t120025z/manifest.json b/snapshots/20260619T120025Z/analysis-runs/hybrid-model-20260619t120025z/manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..f79a3a222cfc4fafe8086600494817a6a52cf6af --- /dev/null +++ b/snapshots/20260619T120025Z/analysis-runs/hybrid-model-20260619t120025z/manifest.json @@ -0,0 +1,14 @@ +{ + "analysis_id": "hybrid-model-20260619t120025z", + "artifacts": { + "hybrid": "snapshots/20260619T120025Z/analysis-runs/hybrid-model-20260619t120025z/analysis-report-hybrid.json", + "hybrid_reviews": "snapshots/20260619T120025Z/analysis-runs/hybrid-model-20260619t120025z/analysis-report-hybrid.llm-reviews.json" + }, + "channel": "canonical", + "model": null, + "published_at": "2026-06-19T12:09:57Z", + "repo": "huggingface/transformers", + "schema_version": 1, + "snapshot_id": "20260619T120025Z", + "variant": "hybrid" +} diff --git a/snapshots/20260619T120025Z/manifest.json b/snapshots/20260619T120025Z/manifest.json index 98acefae039f830f0a3a512af170016179b921c4..5d25d86636d72e82561074c7792659997a041dd8 100644 --- a/snapshots/20260619T120025Z/manifest.json +++ b/snapshots/20260619T120025Z/manifest.json @@ -32,6 +32,24 @@ "timeline_events": 1544 }, "extracted_at": "2026-06-19T12:00:25Z", + "published_analysis": { + "canonical_analysis_id": "hybrid-model-20260619t120025z", + "runs": { + "hybrid-model-20260619t120025z": { + "analysis_id": "hybrid-model-20260619t120025z", + "artifacts": { + "hybrid": "snapshots/20260619T120025Z/analysis-runs/hybrid-model-20260619t120025z/analysis-report-hybrid.json", + "hybrid_reviews": "snapshots/20260619T120025Z/analysis-runs/hybrid-model-20260619t120025z/analysis-report-hybrid.llm-reviews.json" + }, + "channel": "canonical", + "manifest_path": "snapshots/20260619T120025Z/analysis-runs/hybrid-model-20260619t120025z/manifest.json", + "model": null, + "published_at": "2026-06-19T12:09:57Z", + "variant": "hybrid" + } + }, + "schema_version": 1 + }, "repo": "huggingface/transformers", "snapshot_id": "20260619T120025Z", "watermark": {