diff --git a/analysis/current/analysis-report-hybrid.json b/analysis/current/analysis-report-hybrid.json index a71e41bceb9766d26abfaa15d97f60f4329642b3..8e57fc703e7061cf74bf1cd3ca4b6b6a29925e9d 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": "20260523T000023Z", - "generated_at": "2026-05-23T00:08:49Z", + "snapshot_id": "20260523T060016Z", + "generated_at": "2026-05-23T06:07:39Z", "evidence_quality": "full", "llm_enrichment": true, "meta_bugs": [ @@ -1202,7 +1202,7 @@ "cluster_id": "cluster-46032-3", "summary": "Cluster of 1 issues and 2 PRs centered on issue #46032.", "status": "open", - "confidence": 0.55, + "confidence": 0.75, "canonical_issue_number": 46032, "canonical_pr_number": 46084, "issue_numbers": [ @@ -1213,7 +1213,8 @@ 46098 ], "evidence_types": [ - "closing_reference" + "closing_reference", + "shared_issue_target" ], "pr_comparisons": [ { @@ -2128,7 +2129,7 @@ 46098 ], "target_issue_number": 46032, - "reason": "PRs in cluster-46032-3 are treated as duplicates because they converge on issue #46032 with closing_reference evidence." + "reason": "PRs in cluster-46032-3 are treated as duplicates because they converge on issue #46032 with closing_reference, shared_issue_target evidence." }, { "cluster_id": "cluster-46100-2", diff --git a/analysis/current/analysis-report-hybrid.llm-reviews.json b/analysis/current/analysis-report-hybrid.llm-reviews.json index caed6c59430883cc642458d11dc4e3b52b6ccd74..b8b2133a3d711f8b82f0c66d9dd0c1ece8705089 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": "20260523T000023Z", - "generated_at": "2026-05-23T00:08:49Z", + "snapshot_id": "20260523T060016Z", + "generated_at": "2026-05-23T06:07:39Z", "model": "gpt-5.4-mini?service_tier=flex", "reviews": [ { @@ -138,20 +138,20 @@ "issue:43866|issue:44863", "issue:44292|issue:46015" ], - "prepared_review_unit_hash": "1f14cd5a67f9ddba1457c983382516bacae8e407936c9bdfc71db9cf260f8495", + "prepared_review_unit_hash": "212bc5f396c731663f9121863d11e41123380f1b65f665edef2960fb1b7f20f9", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 69, - "serialized_chars": 48404, + "serialized_chars": 48401, "estimated_input_tokens": 12101, "estimated_eval_tokens": 24458 }, @@ -160,234 +160,426 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most pairs are clearly unrelated; only two soft edges look like true duplicates: the two AutoTokenizer v5 token-id regression reports, and the two DeepSpeed ZeRO-3/BERT index-out-of-bounds reports.", - "confidence": 0.94, - "canonical_issue_reason": "issue:45920 is the best canonical issue: it describes the broader AutoTokenizer v5 regression across multiple model families, which subsumes the narrower Granite-specific report.", + "summary": "This cluster is mostly noisy and spans many unrelated bugs/docs issues. Only a couple of soft edges look like true duplicates: one DeepSpeed ZeRO-3 Bert init crash pair and one AutoTokenizer token-ID regression pair.", + "confidence": 0.61, + "canonical_issue_reason": "issue:45920 is the broadest and most central issue in the tokenizer-regression subcluster; it covers the same AutoTokenizer wrong-token-ID problem across multiple model families, while issue:45812 is a narrower Granite-only report.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45920 is the strongest overall issue because it is broader, more recent, and still open, making it the best representative of the tokenizer regression cluster.", + "best_issue_reason": "issue:45920 best represents the strongest duplicate subcluster because it describes the v5 AutoTokenizer regression in the most general terms and is the likely umbrella for the narrower Granite report.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44322", "right": "issue:45464", "accept": false, - "reason": "Different failures: missing `num_attention_heads` in Qwen3.5 config vs streaming chat/completions error; no shared code path." + "reason": "Different models and failures: missing config attribute vs streaming chat/completions error." }, { "left": "issue:45237", "right": "issue:45685", "accept": false, - "reason": "Unrelated platforms and errors: AMD GPU support for GPT-OSS vs an MPS histogram op failure in MoE." + "reason": "Different platforms and bugs: AMD GPU failure vs MPS histogram op limitation." }, { "left": "issue:43716", "right": "issue:45237", "accept": false, - "reason": "Different model/system mismatch: Mistral-3 image dtype issue vs GPT-OSS on AMD GPUs." + "reason": "Different code paths: image preprocessor dtype mismatch vs GPT-OSS AMD GPU issue." }, { "left": "issue:43828", "right": "issue:45237", "accept": false, - "reason": "Different model and symptom: autocast dtype mismatch in Phi-tiny-MoE vs AMD GPU failure in GPT-OSS." + "reason": "Both are dtype/device problems, but they affect different models and execution paths." }, { "left": "issue:45237", "right": "issue:45507", "accept": false, - "reason": "Different issue class: AMD GPU runtime failure vs GraniteMoEHybrid calling an invalid method." + "reason": "Different model families and error modes; not the same bug." }, { "left": "issue:44292", "right": "issue:45237", "accept": false, - "reason": "Different model/hardware combination: Qwen-3-8B-NVFP4 error vs GPT-OSS on AMD GPUs." + "reason": "Separate inference/runtime failures with no clear shared concrete cause." }, { "left": "issue:30277", "right": "issue:35545", "accept": false, - "reason": "Different underlying problems: DeepSpeed collective mismatch during loading vs ONNX export failure." + "reason": "Deepspeed collective mismatch and ONNX export are unrelated." }, { "left": "issue:35545", "right": "issue:42915", "accept": false, - "reason": "Different models and causes: ModernBERT ONNX export vs Qwen3Moe FP8 config failure." + "reason": "Different model/export issue vs FP8 config failure." }, { "left": "issue:35545", "right": "issue:41720", "accept": false, - "reason": "Different code paths: ONNX export issue vs auto device mapping cuda assert." + "reason": "ONNX export error and auto device-map cuda assert are unrelated." }, { "left": "issue:45901", "right": "issue:46015", "accept": false, - "reason": "One is a task runtime crash, the other is outdated RAG documentation; not the same bug." + "reason": "Table QA crash vs outdated RAG docs; different artifacts and causes." }, { "left": "issue:44509", "right": "issue:44512", "accept": false, - "reason": "Both are docs regressions after v5 removals, but they concern different removed content and are not the same underlying change." + "reason": "Both are docs cleanup, but they reference different removed features and are not the same issue." }, { "left": "issue:35545", "right": "issue:36010", "accept": false, - "reason": "Different failures: export-to-ONNX vs import path breakage for GenerationMixin." + "reason": "Export-to-ONNX and GenerationMixin import failure are separate regressions." }, { "left": "issue:44805", "right": "issue:45823", "accept": false, - "reason": "Different symptoms and subsystems: mask-shape indexing error vs Gemma4 device mismatch with auto device mapping." + "reason": "Different symptoms: mask-shape IndexError vs Gemma4 device mismatch." }, { "left": "issue:44918", "right": "issue:46165", "accept": false, - "reason": "Different training flow failures: Qwen3.5 embedding unpacking in TRL SFT vs resuming from a local model with trust_remote_code." + "reason": "TRL SFT embedding unpacking and resume/trust-remote-code training are unrelated." }, { "left": "issue:43638", "right": "issue:44805", "accept": false, - "reason": "Both are index errors, but the reported failures come from different paths and shapes; not enough to treat as the same bug." + "reason": "Both mention training/ZeRO-3 style failures, but the concrete errors and model paths differ." }, { "left": "issue:43638", "right": "issue:46165", "accept": false, - "reason": "Different scenarios: ZeRO-3/BERT init index error vs training-resume with trust_remote_code." + "reason": "Different training scenarios and error mechanisms." }, { "left": "issue:43596", "right": "issue:44805", "accept": false, - "reason": "Superficially similar index errors, but one is ZeRO-3 BERT loading and the other is a mask/tensor shape mismatch." + "reason": "Both are IndexError reports, but one is ZeRO-3 Bert init and the other is a different shape mismatch." }, { "left": "issue:44805", "right": "issue:45161", "accept": false, - "reason": "Different model and failure mode: mask-shape index error vs GPT-OSS tensor-parallel issue." + "reason": "Generic tensor-shape failure vs GPT-OSS MoE TP issue are not the same bug." }, { "left": "issue:44560", "right": "issue:46165", "accept": false, - "reason": "Different areas entirely: Qwen3-VL video StopIteration vs model-resume/trust_remote_code training failure." + "reason": "Video StopIteration and model-resume/trust-remote-code are unrelated." }, { "left": "issue:43596", "right": "issue:46165", "accept": false, - "reason": "Different code paths and symptoms; no evidence they share the same root cause." + "reason": "Different training/loading failures with no shared concrete code path." }, { "left": "issue:44805", "right": "issue:44918", "accept": false, - "reason": "Different model/training stack problems: tensor mask shape mismatch vs Qwen3.5 embedding unpacking in TRL." + "reason": "Shape mismatch error and Qwen3.5 embedding unpacking are different issues." }, { "left": "issue:43572", "right": "issue:46165", "accept": false, - "reason": "Different failure modes: missing StableLmConfig field vs resume-training issue." + "reason": "StableLm pad_token_idx regression is unrelated to training resume problems." }, { "left": "issue:34634", "right": "issue:35545", "accept": false, - "reason": "Unrelated: BarkProcessor voice_preset vs ModernBERT ONNX export." + "reason": "BarkProcessor voice preset and ModernBERT ONNX export are unrelated." }, { "left": "issue:30277", "right": "issue:43638", "accept": false, - "reason": "Both involve DeepSpeed, but the concrete failures are different models and different errors." + "reason": "Deepspeed collective mismatch and Bert ZeRO-3 index error are different failures." }, { "left": "issue:43596", "right": "issue:43638", "accept": true, - "reason": "Both report the same ZeRO-3 loading bug: `index 0 is out of bounds for dimension 0 with size 0` while initializing/loading BertModel variants." + "reason": "Same DeepSpeed ZeRO-3/BertModel loading crash with index-0-out-of-bounds on zero-sized tensors; the second report is the same bug on a narrower Bert variant." }, { "left": "issue:43039", "right": "issue:45161", "accept": false, - "reason": "Liger Kernel cross-entropy invocation is unrelated to GPT-OSS tensor-parallel failure." + "reason": "Liger kernel cross-entropy call and GPT-OSS MoE TP issue are unrelated." }, { "left": "issue:43975", "right": "issue:44457", "accept": false, - "reason": "Different problems: incorrect detokenization vs LoRA merge/save/reload output mismatch." + "reason": "Detokenization regression and LoRA merge/save mismatch are different problems." }, { "left": "issue:44292", "right": "issue:44457", "accept": false, - "reason": "Different model/runtime issue vs LoRA merge consistency bug." + "reason": "NVFP4 runtime error and LoRA save/reload mismatch are unrelated." }, { "left": "issue:34928", "right": "issue:44805", "accept": false, - "reason": "Different distributed-training bugs: activation-checkpoint/FSDP recomputation mismatch vs tensor mask shape mismatch." + "reason": "Activation-checkpointing/FSDP recompute mismatch is not the same as the mask-shape IndexError." }, { "left": "issue:24643", "right": "issue:30277", "accept": false, - "reason": "Different DeepSpeed failures with different error signatures and code paths." + "reason": "DeepSpeed weight-dimension error and collective mismatch are different bugs." }, { "left": "issue:45859", "right": "issue:45902", "accept": false, - "reason": "Both touch Qwen3_5MoeTextRotaryEmbedding, but one is CPU-offload incompatibility and the other is uninitialized-memory after `to_empty(cuda)`; not the same concrete bug." + "reason": "Same class, but one is CPU-offload incompatibility and the other is uninitialized-memory on materialization; not clearly the same fix." }, { - "left": "issue:43975", + "left": "issue:43638", + "right": "issue:45161", + "accept": false, + "reason": "Bert ZeRO-3 init crash and GPT-OSS TP issue are unrelated." + }, + { + "left": "issue:34928", + "right": "issue:36331", + "accept": false, + "reason": "FSDP checkpointing tensor-size mismatch vs custom trainer compute_loss signature breakage are different issues." + }, + { + "left": "issue:43596", + "right": "issue:45161", + "accept": false, + "reason": "Different models, different failure modes." + }, + { + "left": "issue:43716", + "right": "issue:45507", + "accept": false, + "reason": "Dtype mismatch in Mistral-3 image preprocessing vs GraniteMoEHybrid invalid method." + }, + { + "left": "issue:39290", + "right": "issue:45081", + "accept": false, + "reason": "Gemma3/vLLM config attribute error and Mistral tokenizer regex patch crash are unrelated." + }, + { + "left": "issue:45823", + "right": "issue:45874", + "accept": false, + "reason": "Both are Gemma4-related, but one is device placement and the other is expensive reverse-embedding lookup; not the same bug." + }, + { + "left": "issue:43425", + "right": "issue:44292", + "accept": false, + "reason": "Torch version compatibility issue vs Qwen-3 runtime error are unrelated." + }, + { + "left": "issue:34928", + "right": "issue:41720", + "accept": false, + "reason": "FSDP activation checkpointing mismatch and Qwen3 auto device-map cuda assert are unrelated." + }, + { + "left": "issue:43828", + "right": "issue:45161", + "accept": false, + "reason": "Phi-tiny-MoE autocast dtype mismatch is not the same as GPT-OSS TP failure." + }, + { + "left": "issue:44918", + "right": "issue:45161", + "accept": false, + "reason": "TRI/SFT embedding unpacking and GPT-OSS TP failure are different code paths." + }, + { + "left": "issue:43716", + "right": "issue:45685", + "accept": false, + "reason": "Different model/platform bugs." + }, + { + "left": "issue:43828", + "right": "issue:44292", + "accept": false, + "reason": "Both involve runtime errors, but on different models and mechanisms." + }, + { + "left": "issue:43541", + "right": "issue:43716", + "accept": false, + "reason": "Mixtral grouped_mm tracing error and Mistral-3 dtype mismatch are different failures." + }, + { + "left": "issue:43716", + "right": "issue:44292", + "accept": false, + "reason": "Mistral-3 preprocessing dtype mismatch vs Qwen NVFP4 runtime error are unrelated." + }, + { + "left": "issue:44292", + "right": "issue:44918", + "accept": false, + "reason": "Qwen NVFP4 runtime failure and Qwen3.5 embedding unpacking are different bugs." + }, + { + "left": "issue:45507", + "right": "issue:45685", + "accept": false, + "reason": "Invalid method call in GraniteMoEHybrid vs MPS histogram int limitation are unrelated." + }, + { + "left": "issue:43039", + "right": "issue:44918", + "accept": false, + "reason": "Liger kernel cross-entropy call and TRL embedding unpacking are unrelated." + }, + { + "left": "issue:41762", + "right": "issue:43638", + "accept": false, + "reason": "Both mention ZeRO-3 loading issues, but Gemma3 and Bert failures are different concrete bugs." + }, + { + "left": "issue:43827", + "right": "issue:44512", + "accept": false, + "reason": "Both are docs issues after v5 removals, but they concern different removed features and are separate cleanups." + }, + { + "left": "issue:43716", + "right": "issue:44560", + "accept": false, + "reason": "Different models and error surfaces: image dtype mismatch vs video StopIteration." + }, + { + "left": "issue:43828", + "right": "issue:45685", + "accept": false, + "reason": "Dtype mismatch in autocast and MPS histogram int failure are different issues." + }, + { + "left": "issue:44918", + "right": "issue:45685", + "accept": false, + "reason": "TRl SFT embedding unpacking and MPS histogram limitation are unrelated." + }, + { + "left": "issue:43039", + "right": "issue:43828", + "accept": false, + "reason": "Liger-kernel cross-entropy usage and Phi-tiny-MoE autocast mismatch are different bugs." + }, + { + "left": "issue:30277", + "right": "issue:36331", + "accept": false, + "reason": "Collective mismatch and compute_loss signature breakage are unrelated." + }, + { + "left": "issue:34928", + "right": "issue:36010", + "accept": false, + "reason": "Checkpointing/FSDP mismatch and GenerationMixin import error are unrelated." + }, + { + "left": "issue:45237", + "right": "issue:46015", + "accept": false, + "reason": "GPT-OSS AMD GPU failure and outdated RAG docs are unrelated." + }, + { + "left": "issue:43039", + "right": "issue:44560", + "accept": false, + "reason": "Liger kernel cross-entropy call and Qwen3-vl video StopIteration are unrelated." + }, + { + "left": "issue:41762", + "right": "issue:43596", + "accept": false, + "reason": "Both are ZeRO-3-related, but Gemma3 and Bert loading crashes are distinct bugs." + }, + { + "left": "issue:39290", + "right": "issue:41720", + "accept": false, + "reason": "vLLM Gemma3 config attribute error and Qwen3 device-map cuda assert are unrelated." + }, + { + "left": "issue:43901", "right": "issue:44512", "accept": false, - "reason": "Unrelated: detokenization bug vs docs mentioning removed run command." + "reason": "TextClassificationPipeline docs and transformers run-command docs are separate documentation issues." + }, + { + "left": "issue:45237", + "right": "issue:45901", + "accept": false, + "reason": "Different model/runtime issue vs table-question-answering crash." + }, + { + "left": "issue:44841", + "right": "issue:45823", + "accept": false, + "reason": "Voxtral processor failure and Gemma4 device mismatch are unrelated." + }, + { + "left": "issue:43975", + "right": "issue:46015", + "accept": false, + "reason": "Detokenization regression and outdated RAG docs are unrelated." }, { "left": "issue:45812", "right": "issue:45920", "accept": true, - "reason": "Same underlying regression: AutoTokenizer returns wrong token IDs in v5; 45920 is just the broader multi-model report." + "reason": "Both report the same v5 AutoTokenizer wrong-token-ID regression across model families; 45920 is the broader umbrella report." }, { "left": "issue:44863", "right": "issue:46015", "accept": false, - "reason": "NemotronH checkpoint-loading failure is unrelated to outdated RAG examples." + "reason": "NemotronH checkpoint loading failure and RAG docs are unrelated." }, { "left": "issue:43866", "right": "issue:44863", "accept": false, - "reason": "Both are model-specific load complaints, but for different models and likely different root causes." + "reason": "Corrupted checkpoint report and NemotronH loading implementation bug are different issues." }, { "left": "issue:44292", "right": "issue:46015", "accept": false, - "reason": "Different areas entirely: quantized model runtime error vs documentation issue." + "reason": "Qwen NVFP4 runtime error and RAG docs cleanup are unrelated." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet: issue:45920 is indeed the broader AutoTokenizer v5 regression and a reasonable canonical issue over the narrower Granite-specific report. The two accepted soft edges are also conservative and well supported by the titles/body clues: 45812\u219445920 looks like the same tokenizer regression family, and 43596\u219443638 are very closely matched ZeRO-3/BertModel index-out-of-bounds reports. I don\u2019t see an overstatement that would require rejection." + "feedback": "The analysis is grounded in the packet and stays conservative. The two accepted duplicate pairs are well supported by the issue titles and similarity, and the broader clustering summary does not overstate the evidence. The canonical issue choice (45920 over 45812) is also reasonable given the broader v5 AutoTokenizer regression wording." }, "evaluator_used": true, "retried": false, @@ -498,277 +690,277 @@ "issue:43645|issue:45399", "issue:43404|issue:45005" ], - "prepared_review_unit_hash": "14695ae8e0d6423d97858cb564c0f40b06c94d6bc758e7c78b93f588aefb1ea5", + "prepared_review_unit_hash": "0e5958b90eb8bb04775c57b53e4ef61bb8800dade8c226df8ee9c2a37c10fb87", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 39, - "serialized_chars": 39126, - "estimated_input_tokens": 9782, - "estimated_eval_tokens": 19820 + "serialized_chars": 39124, + "estimated_input_tokens": 9781, + "estimated_eval_tokens": 19818 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a mixed bag of unrelated Transformers issues rather than one coherent duplicate cluster. Only two soft pairs look like true duplicates: SigLIP2 output/embedding regressions and custom remote-code module loading/path collisions. The rest differ by model, API surface, or failure mode and should stay separate.", - "confidence": 0.88, - "canonical_issue_reason": "issue:45698 is the broadest and most active open bug here, centered on custom-module loading after save/load, which also best captures the only plausible duplicate-adjacent path in the set.", + "summary": "This cluster is mostly a loose collection of unrelated model-loading, config, dtype, and runtime issues. Only the SigLIP2 pair looks like a credible duplicate; the custom-module cache-path pair is adjacent but not the same bug.", + "confidence": 0.73, + "canonical_issue_reason": "issue:45698 is the best canonical issue: it is open, user-facing, and has the clearest actionable save_pretrained/from_pretrained custom-code loading regression with broader discussion than the other nearby loading reports.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45698 has the strongest triage value overall: open, discussed, and framed as a general correctness bug in from_pretrained/save_pretrained custom module resolution rather than a narrow model-specific symptom.", + "best_issue_reason": "issue:45698 is the strongest standalone triage target because it has a concrete reproduction, current open status, and a focused failure mode that is easier to act on than the more diffuse runtime reports in the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:39290", "right": "issue:43531", "accept": false, - "reason": "Both mention sliding_window, but they are different model families and different concrete failures (Gemma3 config attribute vs Qwen3-MoE behavior)." + "reason": "Same broad sliding-window theme, but different models and different concrete failures (Gemma3/vLLM attribute access vs Qwen3-MoE behavior)." }, { "left": "issue:43854", "right": "issue:43866", "accept": false, - "reason": "Unrelated models and failure modes: GLM-4.7-Flash loading in tests vs an Ovis2 checkpoint corruption report." + "reason": "A model-loading test failure and a corrupted checkpoint report are not the same underlying bug." }, { "left": "issue:43792", "right": "issue:45901", "accept": false, - "reason": "Whisper loading/runtime failure and a table-question-answering crash are different code paths and different tasks." + "reason": "Whisper inference failure and table-question-answering crash are unrelated code paths." }, { "left": "issue:41093", "right": "issue:43638", "accept": false, - "reason": "Both involve index/shape errors, but one is a mask-length mismatch and the other is a zero-size tensor indexing error during different Bert/DeepSpeed flows." + "reason": "Both mention index errors under ZeRO-3/BERT, but the concrete symptoms and setups differ too much to merge confidently." }, { "left": "issue:43425", "right": "issue:45070", "accept": false, - "reason": "Torch version incompatibility and a PretrainedConfig/pydantic regression are distinct bugs with different root causes." + "reason": "Torch compatibility and a PretrainedConfig/pydantic regression are separate issues." }, { "left": "issue:43596", "right": "issue:45800", "accept": false, - "reason": "DeepSpeed ZeRO-3 index error on BertModel is not the same as a torch/transformers version compatibility problem." + "reason": "A ZeRO-3 BERT index error and a Torch/Transformers incompatibility are different bugs." }, { "left": "issue:43975", "right": "issue:45901", "accept": false, - "reason": "Incorrect detokenization for DeepSeek Coder is unrelated to table QA crashes." + "reason": "Incorrect detokenization and table QA crashes are different failure modes." }, { "left": "issue:43792", "right": "issue:45070", "accept": false, - "reason": "Whisper runtime failure and pydantic field breakage are unrelated." + "reason": "Whisper runtime failure is unrelated to the PretrainedConfig/pydantic regression." }, { "left": "issue:45362", "right": "issue:45464", "accept": false, - "reason": "Both are Qwen3.5 chat-related, but one is a generic chat crash and the other is a streaming-inference API failure on a different model size/code path." + "reason": "Both involve Qwen3.5, but one is a chat-model crash and the other is a streaming API failure; not the same code path." }, { "left": "issue:44292", "right": "issue:45901", "accept": false, - "reason": "init_empty_weights/_is_hf_initialized is a model-loading bug, not the same as the table QA crash." + "reason": "Qwen-3-8B-NVFP4 runtime error and table-question-answering crash are unrelated." }, { "left": "issue:39290", "right": "issue:45823", "accept": false, - "reason": "Both mention model/config oddities, but Gemma3 sliding-window parsing and Gemma4 device-mapping mismatch are different bugs." + "reason": "Both are Gemma-family issues, but missing sliding_window_pattern and a device mismatch are distinct bugs." }, { "left": "issue:45464", "right": "issue:45685", "accept": false, - "reason": "Streaming chat failure on Qwen3.5 and an MPS histogram/int issue in MoE are unrelated." + "reason": "Qwen3.5 streaming inference and an MPS histogram kernel error are unrelated." }, { "left": "issue:43404", "right": "issue:43866", "accept": false, - "reason": "Tied-weight bug in Mistral3 is unrelated to an Ovis2 checkpoint corruption report." + "reason": "Tied lm_head weights in Mistral3 and a corrupted Ovis2 checkpoint are different bugs." }, { "left": "issue:43541", "right": "issue:45685", "accept": false, - "reason": "Different failure modes and code paths: torch dynamo grouped_mm tracing vs MPS int histogram support." + "reason": "Grouped_mm tracing failure and MPS int histogram limitation are separate runtime issues." }, { "left": "issue:43541", "right": "issue:43828", "accept": false, - "reason": "Both are dtype/mixed-precision flavored, but one is dynamo tracing on Mixtral and the other is an autocast mismatch on Phi-tiny-MoE; not the same bug." + "reason": "Both are MoE-related runtime errors, but grouped_mm tracing and autocast dtype mismatch are different concrete bugs." }, { "left": "issue:44841", "right": "issue:45084", "accept": false, - "reason": "Processor failure for Voxtral and a TypeError about template compilation are different subsystems." + "reason": "Processor failure for Voxtral and a compile/template-node TypeError are unrelated." }, { "left": "issue:36010", "right": "issue:39290", "accept": false, - "reason": "Importing GenerationMixin and Gemma3 sliding_window_pattern are completely unrelated issues." + "reason": "ImportError for GenerationMixin and a Gemma3 sliding_window_pattern issue are unrelated." }, { "left": "issue:45800", "right": "issue:45823", "accept": false, - "reason": "Torch/transformers version incompatibility is not the same as a Gemma4 device-map forward mismatch." + "reason": "Torch/Transformers incompatibility and Gemma4 device mismatch are distinct problems." }, { "left": "issue:41093", "right": "issue:41762", "accept": false, - "reason": "Both are zero-size/index errors, but one is a mask shape mismatch during training and the other is a Gemma3 ZeRO-3 load-time failure." + "reason": "Both are ZeRO-3 index failures, but the error signatures and affected models differ enough to avoid merging." }, { "left": "issue:43572", "right": "issue:45070", "accept": false, - "reason": "A missing StableLmConfig field and a pydantic PretrainedConfig regression are separate config issues." + "reason": "Missing pad_token_idx in StableLmConfig is unrelated to the PretrainedConfig/pydantic regression." }, { "left": "issue:43296", "right": "issue:43366", "accept": false, - "reason": "PaddleOCR-VL load failure in vLLM and GGUF gpt-oss support are different model-support problems." + "reason": "PaddleOCR-VL loading in vLLM and GGUF gpt-oss support are different requests/bugs." }, { "left": "issue:43541", "right": "issue:45161", "accept": false, - "reason": "Mixtral torch-dynamo tracing and GPT-OSS tensor-parallel support are different problems, even though both touch MoE models." + "reason": "Both touch MoE/GPT-OSS, but one is a grouped_mm tracing bug and the other is a tensor-parallel support problem." }, { "left": "issue:44387", "right": "issue:45005", "accept": false, - "reason": "Int4 memory/OOM regression and tied-weights translation issues are unrelated v5 regressions." + "reason": "int4 memory growth/OOM and tied-weight translation-model issues are unrelated." }, { "left": "issue:43828", "right": "issue:45464", "accept": false, - "reason": "Autocast dtype mismatch and Qwen3.5 streaming chat failure are not the same code path." + "reason": "dtype mismatch under autocast and streaming chat/completions failure are not the same bug." }, { "left": "issue:44291", "right": "issue:45070", "accept": false, - "reason": "Unexpected _is_hf_initialized argument during init_empty_weights is unrelated to the PretrainedConfig pydantic field break." + "reason": "Both are v5 init/config regressions, but unexpected _is_hf_initialized and pydantic field handling are different code paths." }, { "left": "issue:43366", "right": "issue:45464", "accept": false, - "reason": "GPT-OSS GGUF support and Qwen3.5 streaming chat failures are different features and failure modes." + "reason": "GPT-OSS GGUF support and Qwen3.5 streaming inference are unrelated." }, { "left": "issue:45020", "right": "issue:45800", "accept": false, - "reason": "Remote-code compatibility breakage and torch version incompatibility are different classes of problem." + "reason": "A broad remote_code regression and a Torch/Transformers version incompatibility are different issues." }, { "left": "issue:43531", "right": "issue:45070", "accept": false, - "reason": "Qwen3-MoE sliding_window behavior and pydantic config-field breakage do not share a concrete root cause." + "reason": "Sliding_window handling in Qwen3-MoE is unrelated to the PretrainedConfig/pydantic change." }, { "left": "issue:43541", "right": "issue:44589", "accept": false, - "reason": "Grouped_mm tracing on Mixtral and missing Float8 storage support are distinct low-level failures." + "reason": "Grouped_mm tracing failure and missing Float8 storage are different runtime/toolchain bugs." }, { "left": "issue:41093", "right": "issue:43596", "accept": false, - "reason": "Both mention indexing errors, but the tensor shapes and affected flows differ; not the same bug." + "reason": "Both are index-out-of-bounds reports under ZeRO-3, but they affect different setups and error signatures." }, { "left": "issue:43054", "right": "issue:43994", "accept": true, - "reason": "Both report SigLIP2 producing incorrect embeddings/results from the same base model; one is the embedding-quality symptom and the other is the broader nonsensical-output symptom." + "reason": "Both report broken SigLIP2 behavior on the same model family; bad text embeddings and nonsensical AutoModel/pipeline outputs plausibly share one underlying regression." }, { "left": "issue:45542", "right": "issue:45800", "accept": false, - "reason": "TensorBoard-only-without-TF backend error and a torch/transformers incompatibility are unrelated." + "reason": "TensorBoard-without-TF backend handling is unrelated to Torch/Transformers compatibility." }, { "left": "issue:45759", "right": "issue:45902", "accept": false, - "reason": "Both involve Qwen3.5 internals, but one is config unwrapping in from_config and the other is uninitialized memory after materialization; different fixes." + "reason": "Unwrapping text_config for composite Qwen configs and uninitialized inv_freq after to_empty are different bugs." }, { "left": "issue:39186", "right": "issue:44918", "accept": false, - "reason": "FSDP 2-D weight runtime error and a TRL SFT embedding-unpacking failure are unrelated training bugs." + "reason": "FSDP 2-D weight error and Qwen3.5 embedding unpacking/trl issue are unrelated." }, { "left": "issue:43716", "right": "issue:44918", "accept": false, - "reason": "Image-preprocessor dtype mismatch in Mistral-3 and Qwen3.5 embedding unpacking in TRL are different problems." + "reason": "Gemma3 image-preprocessor dtype mismatch and Qwen3.5 embedding unpacking failure are different issues." }, { "left": "issue:40990", "right": "issue:44960", "accept": false, - "reason": "High perplexity on GPT-OSS and a GLM5 issue are not the same bug." + "reason": "GPT-OSS perplexity regression and a GLM5 issue are unrelated." }, { "left": "issue:45632", "right": "issue:45698", - "accept": true, - "reason": "Both point to custom remote-code module resolution/loading problems; the cache path collision and wrong-module-after-save/load symptoms are plausibly the same underlying path-handling bug." + "accept": false, + "reason": "These are adjacent custom-code loading bugs, but one is a cache-path collision for local models sharing a leaf directory name and the other is wrong module resolution after save_pretrained; not clearly one mergeable fix." }, { "left": "issue:43645", "right": "issue:45399", "accept": false, - "reason": "Jupyter custom-model initialization breakage and flash-attn2 fallback gating are unrelated." + "reason": "Jupyter custom-model initialization breakage and flash-attn fallback blocking are separate regressions." }, { "left": "issue:43404", "right": "issue:45005", "accept": false, - "reason": "Untied lm_head weights in Mistral3 and translation-model tied-weights regressions are both weight-related but not the same concrete bug." + "reason": "Tied-weight handling in a specific Mistral3 multimodal model and broader translation-model tied-weight issues are related in theme but not the same concrete bug." } ] }, "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet and stays appropriately conservative. The two accepted soft pairs are plausibly duplicate-adjacent from the titles alone, and the broader triage choice of issue:45698 is supported by it being open, discussed, and the most general correctness report in the set. No overstatement stands out." + "accept": false, + "feedback": "The cluster summary is mostly reasonable, but it overstates the evidence for at least one soft-edge merge. In particular, accepting issue:45632 vs issue:45698 as the same bug is not well grounded from the packet: one is a trust_remote_code cache-path collision for local models sharing a leaf directory name, while the other is from_pretrained loading the wrong custom module after save_pretrained. Those look adjacent, but not clearly one mergeable fix. Also, phrases like \"broad impact\" and \"likely share one underlying regression\" are a bit stronger than the packet supports." }, "evaluator_used": true, - "retried": false, + "retried": true, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -871,14 +1063,14 @@ "issue:43782|issue:44960", "issue:43881|issue:43883" ], - "prepared_review_unit_hash": "c0d55d43006a7a8f4a0e0e1a0cdaa0ed21c7426748ef135d57d1a2d5c5e7a22d", + "prepared_review_unit_hash": "1b6c635565c820ed5d373d99e857494b922033147f4e4d92fd3d4adab909d236", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, @@ -893,204 +1085,222 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a mix of unrelated Transformers issues; the only clear duplicate group is the v5 non-persistent-buffer regression, with a second likely duplicate in SigLIP2 HF-vs-reference discrepancy reports.", - "confidence": 0.86, - "canonical_issue_reason": "issue:43950 is the clearest root-cause report for the non-persistent-buffer regression and is directly echoed by the duplicate-style title in issue:44534.", + "summary": "Mostly a grab-bag of unrelated Transformers issues. The only clear duplicate-style match is the v5 non-persistent buffer regression pair; the rest are different bugs, models, or docs problems. No PRs were provided.", + "confidence": 0.93, + "canonical_issue_reason": "Issue #43950 is the best canonical because it most clearly states the non-persistent-buffer regression in v5 and is the most specific, root-cause-oriented report among the duplicate-like pair.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43950 is the most concrete and actionable issue in the cluster, with a precise regression description and strong duplicate signals.", + "best_issue_reason": "Issue #43950 is the strongest overall representative of the only true duplicate cluster here: it is precise, broad enough to subsume the related report, and directly describes the underlying regression.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44530", "right": "issue:45005", "accept": false, - "reason": "Different bugs: Qwen3.5 cache/packing crash vs tied-weights regression on translation models." + "reason": "Different bugs: Qwen3.5 cache crash vs. tied-weights behavior in translation models." }, { "left": "issue:42915", "right": "issue:45399", "accept": false, - "reason": "Unrelated failures: FineGrainedFP8Config on Qwen3Moe vs flash-attn fallback gating." + "reason": "FP8 config failure and flash-attn2 fallback gating are separate issues." }, { "left": "issue:43873", "right": "issue:45005", "accept": false, - "reason": "Quantization/offloading behavior is not the same issue as tied-weights handling." + "reason": "Quantization/offload behavior is unrelated to tied-weight handling." }, { "left": "issue:44863", "right": "issue:45005", "accept": false, - "reason": "Checkpoint loading failure for NemotronH is unrelated to translation-model tied weights." + "reason": "NemotronH checkpoint loading failure is not the same as translation-model tied weights." }, { "left": "issue:44360", "right": "issue:44512", "accept": false, - "reason": "DSA indexer algorithm complaint is unrelated to a docs/run-command removal issue." + "reason": "DSA indexer/ReLU discussion and a docs typo are not the same issue." }, { "left": "issue:44918", "right": "issue:45507", "accept": false, - "reason": "Different code paths: Qwen3.5 SFT embedding unpacking vs GraniteMoEHybrid invalid method call." + "reason": "Qwen3.5 SFT embedding unpacking and GraniteMoEHybrid invalid method are different code paths." }, { "left": "issue:44492", "right": "issue:44512", "accept": false, - "reason": "A cache-strategy typo is not the same as outdated CLI documentation." + "reason": "A cache-strategy typo is not the same as stale run-command docs." }, { "left": "issue:43975", "right": "issue:44292", "accept": false, - "reason": "Different model-specific runtime failures with no shared concrete bug." + "reason": "Detokenization regression and NVFP4 runtime error are unrelated." }, { "left": "issue:44292", "right": "issue:45507", "accept": false, - "reason": "NVFP4 runtime error and GraniteMoEHybrid invalid method are unrelated." + "reason": "Qwen-3-8B-NVFP4 failure and GraniteMoEHybrid invalid method are different bugs." }, { "left": "issue:43950", "right": "issue:44534", "accept": true, - "reason": "Same underlying regression: non-persistent buffers get junk/corrupted values in v5." + "reason": "Both describe the same v5 regression where non-persistent buffers are mishandled/corrupted with junk values." }, { "left": "issue:43644", "right": "issue:43950", "accept": false, - "reason": "Missing fields from trust_remote_code config loading is a different bug from buffer corruption." + "reason": "AutoConfig trust_remote_code field loss is a different bug from non-persistent buffer corruption." }, { "left": "issue:39692", "right": "issue:43054", "accept": false, - "reason": "SigLIP2 doc-example errors are not the same as model-quality/embedding regression." + "reason": "SigLIP2 doc/example errors and poor text embeddings are separate problems." }, { "left": "issue:43828", "right": "issue:44928", "accept": false, - "reason": "Autocast dtype mismatch is distinct from the 3D position_ids / SDPA fallback issue." + "reason": "Autocast dtype mismatch is not the same as the Qwen3.5 SDPA/NaN issue." }, { "left": "issue:36331", "right": "issue:41762", "accept": false, - "reason": "Trainer signature mismatch and ZeRO-3 loading IndexError are different failures." + "reason": "Trainer kwarg TypeError and DeepSpeed ZeRO-3 loading failure are unrelated." }, { "left": "issue:41628", "right": "issue:45800", "accept": false, - "reason": "Import error for AutoImageProcessor is not the same as torch/Transformers compatibility issues." + "reason": "Import error for AutoImageProcessor is different from torch/transformers version incompatibility." }, { "left": "issue:43950", "right": "issue:45902", - "accept": true, - "reason": "Same non-persistent-buffer bug family: 45902 is a model-specific manifestation after meta->to_empty materialization." + "accept": false, + "reason": "Both involve non-persistent buffers, but this looks like a different Qwen3_5Moe materialization path rather than the same concrete bug." }, { "left": "issue:43872", "right": "issue:44589", "accept": false, - "reason": "bitsandbytes constructor kwarg mismatch differs from missing Float8 storage type." + "reason": "bitsandbytes Int8Params incompatibility and missing Float8 storage are different failures." }, { "left": "issue:43866", "right": "issue:44403", "accept": false, - "reason": "Corrupted checkpoint and loading noise are not the same underlying problem." + "reason": "Corrupted checkpoint vs. generic loading noise are not the same underlying issue." }, { "left": "issue:43054", "right": "issue:45701", "accept": false, - "reason": "SigLIP2 embedding quality issue is unrelated to version-dependent tokenization changes." + "reason": "SigLIP2 embedding quality regression and tokenization changes are distinct." }, { "left": "issue:41553", "right": "issue:43883", "accept": false, - "reason": "Bad AutoTokenizer error message and missing all_tied_weights_keys are unrelated." + "reason": "Bad AutoTokenizer error for Voxtral and Molmo tied-weights attribute error are unrelated." }, { "left": "issue:38175", "right": "issue:43054", "accept": false, - "reason": "Both mention SigLIP2, but one is zero probabilities and the other is embedding quality; too broad to merge." + "reason": "Unexpected zero probabilities and worse text embeddings in SigLIP2 are different symptoms." }, { "left": "issue:44910", "right": "issue:45859", "accept": false, - "reason": "Flash Attention illegal-memory-access crash is different from CPU-offload incompatibility." + "reason": "Flash-attn illegal memory access on Qwen3.5 and CPU-offload incompatibility are different code paths." }, { "left": "issue:45663", "right": "issue:45902", "accept": false, - "reason": "Gemma4 shared_kv_state/FSDP2 failure is unrelated to Qwen rotary-buffer initialization." + "reason": "Gemma4 FSDP2 shared_kv_states failure is unrelated to Qwen rotary embedding uninitialized memory." }, { "left": "issue:41553", "right": "issue:43881", "accept": false, - "reason": "Tokenizer error-message issue and glm-4v loading failure are separate bugs." + "reason": "Bad Voxtral tokenizer loading error and glm-4v-9b load failure are not the same bug." }, { "left": "issue:42947", "right": "issue:44387", "accept": false, - "reason": "LoRA gradient checkpointing ineffectiveness is not the same as int4 reserved-memory/OOM regression." + "reason": "LoRA gradient checkpointing ineffectiveness and int4 quantization OOM are different issues." }, { "left": "issue:43054", "right": "issue:43493", - "accept": true, - "reason": "Both describe SigLIP2 HF-vs-reference discrepancies that degrade outputs; 43493 is the clearer root-cause report." + "accept": false, + "reason": "SigLIP2 embedding degradation and HF-vs-JAX discrepancy are related only by model family, not the same bug." }, { "left": "issue:44403", "right": "issue:44451", "accept": false, - "reason": "Generic loading noise and inability to load a specific model are not the same bug." + "reason": "Generic loading noise and ScandiBERT load failure are not the same underlying issue." }, { "left": "issue:43975", "right": "issue:45237", "accept": false, - "reason": "Detokenization regression and AMD GPU support/loading failure are unrelated." + "reason": "Detokenization regression and AMD GPU loading failure are unrelated." }, { "left": "issue:43643", "right": "issue:45020", "accept": false, - "reason": "Missing config fields under trust_remote_code is too specific to merge with a broad remote_code breakage report." + "reason": "Missing fields with trust_remote_code and broad remote_code breakage are too broad to treat as the same concrete bug." }, { "left": "issue:44403", "right": "issue:44488", "accept": false, - "reason": "Two model-loading reports without evidence of the same root cause." + "reason": "Generic loading noise and sleng-bert load failure are different reports." + }, + { + "left": "issue:42947", + "right": "issue:43856", + "accept": false, + "reason": "LoRA checkpointing inefficiency and Qwen3 MoE memory usage are different performance issues." + }, + { + "left": "issue:41762", + "right": "issue:42915", + "accept": false, + "reason": "Gemma3 ZeRO-3 loading failure and Qwen3Moe FP8 failure are unrelated." + }, + { + "left": "issue:43782", + "right": "issue:44960", + "accept": false, + "reason": "Qwen3VL weight_only error and GLM5 are different model/load failures." }, { "left": "issue:43881", "right": "issue:43883", "accept": false, - "reason": "glm-4v loading failure and Molmo tied-weights attribute error are different code paths." + "reason": "These are separate model-loading failures with different error signatures and root causes." } ] }, "evaluator_result": { "accept": true, - "feedback": "Grounded and conservative overall. The canonical issue choice is well supported by the titles, and the accepted soft edges are plausibly same-bug families rather than overclaims. The summary stays within the evidence available in the packet." + "feedback": "The summary is grounded in the packet and stays conservative on the soft-edge calls. The only accepted duplicate pair, issue:43950 with issue:44534, is well supported by the titles and packet text. No PR-pair mergeability concerns apply here because no PRs were provided." }, "evaluator_used": true, "retried": false, @@ -1195,238 +1405,238 @@ "issue:43366|issue:43828", "issue:45325|issue:45902" ], - "prepared_review_unit_hash": "9e80fac1cc6813814dd4b141ecaa1da345819fb06b6320ac97ada02c57fb4b5c", + "prepared_review_unit_hash": "d21fa50cb4c4875fa8c27c43d17c5d9f56c54250eb70039a046a247ba6b7f8b6", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 33, - "serialized_chars": 38073, - "estimated_input_tokens": 9519, - "estimated_eval_tokens": 19294 + "serialized_chars": 38072, + "estimated_input_tokens": 9518, + "estimated_eval_tokens": 19292 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most items are distinct bugs spanning different models and subsystems. I found only a few likely duplicate pairs: the docs cleanup for removed pipeline tasks, the repeated `_is_hf_initialized` loading regression, and the repeated mask/tensor shape-mismatch IndexError. No PRs were present.", - "confidence": 0.73, - "canonical_issue_reason": "issue:44291 is the strongest canonical anchor among the true duplicates because it states the regression, version boundary, and underlying `_is_hf_initialized` cause more clearly than the narrower bitsandbytes symptom report.", + "summary": "The cluster is mostly heterogeneous transformer/model-loading regressions with no single broad duplicate. Only one soft edge looks like a true duplicate: the two docs issues about stale v5 pipeline-task references. All other candidate pairs are different bugs or different subsystems.", + "confidence": 0.84, + "canonical_issue_reason": "Issue 44509 is the broader, more explicit version of the docs regression: it covers multiple removed pipeline tasks and subsumes 43827's narrower summary.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44291 is the clearest overall issue to keep: it is specific, reproducible, and broad enough to subsume the related `_is_hf_initialized` failure report.", + "best_issue_reason": "44509 is the best issue candidate because it is the clearest, most comprehensive statement of the docs cleanup needed after v5 pipeline removals.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45685", "right": "issue:45823", "accept": false, - "reason": "Different failures: MPS Int histogram support vs a Gemma4 device placement mismatch." + "reason": "Different failures: MPS histogram/int typing vs Gemma4 PLE device-mapping mismatch." }, { "left": "issue:43824", "right": "issue:43872", "accept": false, - "reason": "An import error for Qwen2.5-VL is not the same bug as the bitsandbytes `_is_hf_initialized` incompatibility." + "reason": "Different bugs: missing Qwen2.5-VL import vs bitsandbytes __new__ kwarg incompatibility." }, { "left": "issue:44991", "right": "issue:45399", "accept": false, - "reason": "Tokenizer loading failure and flash-attn fallback gating are unrelated code paths." + "reason": "Tokenizer load failure for a specific model vs flash-attn fallback being blocked; not the same code path." }, { "left": "issue:45020", "right": "issue:45823", "accept": false, - "reason": "Broad remote_code regressions do not match a specific Gemma4 forward device mismatch." + "reason": "Remote-code regressions and Gemma4 device mismatch are unrelated." }, { "left": "issue:42491", "right": "issue:44368", "accept": false, - "reason": "Both involve Qwen LoRA, but one is an hf4.x/hf5.x compatibility issue and the other is a config warning; not the same bug." + "reason": "LoRA/Qwen3 MoE compatibility issue vs a warning about tie_word_embeddings; related theme but not the same bug." }, { "left": "issue:44661", "right": "issue:45020", "accept": false, - "reason": "Tokenizer-mapping edge case and remote_code regression are different problems." + "reason": "Tokenizer mapping-name edge case vs broader remote_code breakage; distinct failure modes." }, { "left": "issue:43828", "right": "issue:45823", "accept": false, - "reason": "Autocast dtype mismatch is distinct from device_map=\"auto\" device placement failure." + "reason": "Autocast dtype mismatch is unrelated to Gemma4 PLE device placement." }, { "left": "issue:45081", "right": "issue:45399", "accept": false, - "reason": "Mistral regex patch crash and flash-attn fallback checks are separate failures." + "reason": "Mistral regex patch crash is a tokenizer-specific bug, not the flash-attn fallback issue." }, { "left": "issue:45020", "right": "issue:45356", "accept": false, - "reason": "Remote_code compatibility regressions are not the same as the Kimi-K2.5 tokenizer codec-handling bug." + "reason": "General remote_code regression vs Kimi-K2.5 tokenizer codec/regression; different underlying problems." }, { "left": "issue:43742", "right": "issue:45399", "accept": false, - "reason": "Model key lookup failure for MobileLLM is unrelated to flash-attn fallback logic." + "reason": "Model-specific key error vs flash-attn fallback gating; no shared bug." }, { "left": "issue:43827", "right": "issue:44509", "accept": true, - "reason": "Same docs cleanup: both report stale references to removed pipeline tasks after v5." + "reason": "These are the same docs regression: both report stale references to removed v5 pipeline tasks, with 44509 being the broader wording." }, { "left": "issue:43404", "right": "issue:43950", "accept": false, - "reason": "lm_head tying in Mistral3 is a different bug from silent corruption of non-persistent buffers." + "reason": "Weight-tying bug in Mistral3 vs non-persistent buffer corruption on from_pretrained; different mechanisms." }, { "left": "issue:43638", "right": "issue:45823", "accept": false, - "reason": "Zero3/BERT index-out-of-bounds failure is unrelated to Gemma4 device mismatch." + "reason": "Zero-3/Bert index error is unrelated to Gemma4 device-map mismatch." }, { "left": "issue:42915", "right": "issue:45020", "accept": false, - "reason": "Qwen3 MoE FP8 config failure is not the same as a remote_code loading regression." + "reason": "Qwen3 MoE FP8 loading failure vs remote_code breakage; not the same bug." }, { "left": "issue:43257", "right": "issue:43866", "accept": false, - "reason": "Qwen3 MoE weight conversion under accelerate+deepspeed is unrelated to an Ovis2 checkpoint corruption report." + "reason": "Qwen3 MoE weight conversion/loading issue vs Ovis2 checkpoint corruption." }, { "left": "issue:43792", "right": "issue:44220", "accept": false, - "reason": "Whisper failing to run is too broad to merge with a specific `_torch_extract_fbank_features()` issue." + "reason": "Whisper loading/runtime problem vs _torch_extract_fbank_features issue; different code paths." }, { "left": "issue:43872", "right": "issue:44291", - "accept": true, - "reason": "Same underlying `_is_hf_initialized` incompatibility during model loading; 44291 generalizes the regression more clearly." + "accept": false, + "reason": "Different bitsandbytes vs init_empty_weights / _is_hf_initialized incompatibilities." }, { "left": "issue:42915", "right": "issue:43872", "accept": false, - "reason": "Different root causes: Qwen3 MoE FP8 config vs bitsandbytes/init_empty_weights kwarg handling." + "reason": "Qwen3 MoE loading/FP8 issue is unrelated to bitsandbytes constructor incompatibility." }, { "left": "issue:41553", "right": "issue:44991", "accept": false, - "reason": "A bad Voxtral load error message is unrelated to tokenizer loading failures for EMBEDDIA/est-roberta." + "reason": "Voxtral tokenizer error message vs est-roberta tokenizer loading regression; distinct models and failures." }, { "left": "issue:43940", "right": "issue:45313", "accept": false, - "reason": "Both are DeepSpeed weight-loading failures, but they target different models and likely different code paths." + "reason": "Both are DeepSpeed weight-loading issues, but one is Qwen3-Next all-missing params and the other is Qwen3.5 language_model loading; too different to merge." }, { "left": "issue:36331", "right": "issue:41093", "accept": false, - "reason": "Unexpected `num_items_in_batch` in compute_loss is not the same as a mask/tensor shape mismatch." + "reason": "Trainer compute_loss signature mismatch vs tensor/mask shape mismatch; unrelated bugs." }, { "left": "issue:42947", "right": "issue:45923", "accept": false, - "reason": "Gradient checkpointing effectiveness in PEFT LoRA is a different issue from a missing supports_gradient_checkpointing flag on a trust_remote_code model." + "reason": "Gradient checkpointing ineffective in PEFT LoRA vs missing supports_gradient_checkpointing flag; different issues." }, { "left": "issue:43257", "right": "issue:45305", "accept": false, - "reason": "Qwen3 MoE weight conversion under DeepSpeed is unrelated to gradient averaging under GAS." + "reason": "Qwen3 MoE weight conversion under accelerate+deepspeed vs gradient averaging with GAS and DeepSpeed; distinct training bugs." }, { "left": "issue:43854", "right": "issue:44863", "accept": false, - "reason": "GLM-4.7-Flash test loading and NemotronH checkpoint loading are different model-specific failures." + "reason": "Unit-test loading failure for GLM-4.7-Flash vs NemotronH checkpoint loading failure; separate model implementations." }, { "left": "issue:44928", "right": "issue:45588", "accept": false, - "reason": "Gradient explosion/NaN in RLHF is not the same bug as a flash_attention.py crash on `s_aux=None`." + "reason": "RLHF gradient explosion from 3D position_ids/SDPA fallback vs flash_attention AttributeError on s_aux=None." }, { "left": "issue:43872", "right": "issue:45538", "accept": false, - "reason": "bitsandbytes `_is_hf_initialized` incompatibility is unrelated to CLIPTokenizer max-length behavior." + "reason": "bitsandbytes incompatibility is unrelated to CLIPTokenizer model_max_length behavior." }, { "left": "issue:41093", "right": "issue:44805", - "accept": true, - "reason": "Same shape-mismatch IndexError; the later report appears to be the same underlying mask/tensor length bug." + "accept": false, + "reason": "Both are mask/tensor index errors, but the reported shapes and contexts differ enough that they do not look like the same bug." }, { "left": "issue:44479", "right": "issue:44560", "accept": false, - "reason": "A broad video-input regression across several Qwen models is not the same as a StopIteration error in Qwen3-vl-embedding." + "reason": "Video input regression across multiple Qwen variants vs a specific Qwen3-vl-embedding StopIteration; related area but not the same failure." }, { "left": "issue:41553", "right": "issue:44488", "accept": false, - "reason": "Voxtral error-message quality and sleng-bert loading are different model-load problems." + "reason": "Bad AutoTokenizer error for Voxtral vs a model-loading failure for cjvt/sleng-bert; different problems." }, { "left": "issue:44530", "right": "issue:44863", "accept": false, - "reason": "PagedAttentionCache linear_attention crash and NemotronH checkpoint loading are unrelated." + "reason": "PagedAttentionCache linear_attention crash vs NemotronH checkpoint loading failure; unrelated." }, { "left": "issue:24643", "right": "issue:30064", "accept": false, - "reason": "DeepSpeed training weight-shape error is unrelated to void segmentation map processing." + "reason": "DeepSpeed training RuntimeError vs image processor void segmentation-map failure; no shared bug." }, { "left": "issue:43366", "right": "issue:43828", "accept": false, - "reason": "GGUF gpt-oss support and Phi-tiny-MoE autocast dtype mismatch are different issues." + "reason": "GGUF gpt-oss support request vs autocast dtype mismatch in Phi-tiny-MoE; not duplicates." }, { "left": "issue:45325", "right": "issue:45902", "accept": false, - "reason": "Qwen2.5-VL rope scaling bug and Qwen3_5Moe uninitialized-memory issue are distinct." + "reason": "Qwen2.5-VL rope-index scaling bug vs uninitialized-memory read in Qwen3_5MoeTextRotaryEmbedding; different implementations and symptoms." } ] }, "evaluator_result": { "accept": true, - "feedback": "The analysis is conservative and grounded in the packet. The three accepted duplicate pairs are well supported by the issue titles/body summaries, and the summary does not overclaim beyond the evidence. No PRs were present, so there is no mergeability concern." + "feedback": "Grounded and conservative overall. The only accepted soft edge, issue:43827 vs issue:44509, is well supported by the titles: both describe stale docs references to removed v5 pipeline tasks, with 44509 clearly broader. The remaining rejections stay cautious and do not overclaim duplication from the packet evidence." }, "evaluator_used": true, "retried": false, @@ -1529,20 +1739,20 @@ "issue:42491|issue:45276", "issue:43708|issue:44075" ], - "prepared_review_unit_hash": "f63ca155e47280861a3dc868989d5bd2630170c08ecc4ba9e40a6c885f77ed14", + "prepared_review_unit_hash": "b0b3fe14205ef052d89f7b0d611c333ed599226ea3ec36e7a5bfc9d432787219", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 32, - "serialized_chars": 36628, + "serialized_chars": 36625, "estimated_input_tokens": 9157, "estimated_eval_tokens": 18570 }, @@ -1551,222 +1761,210 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Two true duplicate pairs stand out: the non-persistent-buffer regression and the custom-model-init regression. Most other pairs are only loosely similar (same family/model area) but describe different concrete failures.", - "confidence": 0.82, - "canonical_issue_reason": "issue:43644 is the strongest canonical for the non-persistent-buffer regression: it is an earlier, more specific description of the same bug that issue:44534 restates almost verbatim.", + "summary": "This cluster is mostly a mix of unrelated Transformers issues with a few true duplicate pairs. The clearest duplicate groups are SigLIP2 loading/output regressions, non-persistent-buffer corruption in v5, custom model initialization failures, and a BERT model-loading regression.", + "confidence": 0.89, + "canonical_issue_reason": "I\u2019d anchor the cluster on issue 43644 because it is a clear, regression-focused description of the non-persistent-buffer bug and has an exact duplicate in 44534, making it a strong representative issue.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43644 is the best overall representative because it is precise, clearly phrased, and has a near-verbatim duplicate in the set.", + "best_issue_reason": "Issue 43644 is the best overall representative: it is specific enough to describe the core regression cleanly, yet generic enough to subsume the matching duplicate 44534.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45663", "right": "issue:45823", "accept": false, - "reason": "Both are Gemma4 problems, but one is an FSDP2 KeyError in attention while the other is a device_map forward mismatch; different code paths." + "reason": "Same model family, but different symptoms and code paths: FSDP2 KeyError in attention vs device_map forward mismatch." }, { "left": "issue:43366", "right": "issue:43531", "accept": false, - "reason": "Different bugs: GGUF/gpt-oss support versus Qwen3-MoE sliding_window behavior." + "reason": "Different problems: GGUF architecture support vs Qwen3-MoE sliding_window behavior." }, { "left": "issue:44155", "right": "issue:45584", "accept": false, - "reason": "AudioFlamingo3 batched-track leakage and Whisper empty-transcription generation are unrelated failures." + "reason": "Unrelated models and failures: AudioFlamingo3 track leakage vs Whisper empty-transcription generation." }, { "left": "issue:43950", "right": "issue:45834", "accept": false, - "reason": "Non-persistent buffer corruption and Kosmos2.5 long-OCR index error are not the same underlying bug." + "reason": "Both are loading/runtime bugs, but the buffer-corruption regression and Kosmos2.5 OCR index error are distinct." }, { "left": "issue:44560", "right": "issue:45685", "accept": false, - "reason": "Qwen3-vl video StopIteration and MoE MPS histogram Int support are unrelated." + "reason": "Different model-specific failures: Qwen3-vl video StopIteration vs MPS histogram support for int." }, { "left": "issue:43824", "right": "issue:43957", "accept": false, - "reason": "Missing exported Qwen2_5_VL class and meta-device loading issues are different loader regressions." + "reason": "ImportError for a missing class and meta-device loading breakage are different failure modes." }, { "left": "issue:43550", "right": "issue:44910", "accept": false, - "reason": "torch.compile+SDPA failure on Bamba-9B-v2 is distinct from Flash Attention crashing on Qwen3.5 position_ids." + "reason": "Different code paths: torch.compile+SDPA in Bamba vs Flash Attention on Qwen3.5 with 3D position_ids." }, { "left": "issue:42915", "right": "issue:43957", "accept": false, - "reason": "FineGrainedFP8Config on Qwen3Moe and meta-device loading bugs are separate problems." + "reason": "Qwen3-MoE FP8 failure and meta-device loading regression are not the same bug." }, { "left": "issue:43366", "right": "issue:44589", "accept": false, - "reason": "GGUF gpt-oss support and Float8 storage lookup failures are not the same bug." + "reason": "Both involve low-level model loading, but GGUF gpt-oss support and Float8 storage lookup are separate issues." }, { "left": "issue:43645", "right": "issue:43646", "accept": true, - "reason": "These describe the same custom-model initialization regression; the notebook wording in 43645 is just a more specific repro of 43646." + "reason": "Same underlying custom-model initialization regression; the notebook-specific report is just a variant of the generic failure." }, { "left": "issue:44560", "right": "issue:44918", "accept": false, - "reason": "Both mention Qwen3-vl, but one is video StopIteration and the other is SFT embedding unpacking; different failures." + "reason": "Different Qwen3-family failures: video StopIteration vs TRL SFT unpacking of input embeddings." }, { "left": "issue:42617", "right": "issue:43366", "accept": false, - "reason": "3d_parallel.py runtime failure and GGUF architecture support are unrelated." + "reason": "3d_parallel.py runtime failure and GGUF gpt-oss support are unrelated." }, { "left": "issue:36010", "right": "issue:42915", "accept": false, - "reason": "GenerationMixin import failure and Qwen3Moe FP8 failure are unrelated." + "reason": "GenerationMixin import error and Qwen3-MoE FP8 failure do not share a concrete code-path problem." }, { "left": "issue:44661", "right": "issue:45823", "accept": false, - "reason": "Tokenizer-mapping registration bug and Gemma4 PLE device mismatch are different issues." + "reason": "Tokenizer-mapping/model-addition bug and Gemma4 device-mismatch forward bug are unrelated." }, { "left": "issue:44560", "right": "issue:44805", "accept": false, - "reason": "Qwen3-vl video StopIteration and a generic mask/tensor shape IndexError are not the same code path." + "reason": "Both are runtime errors, but StopIteration in video handling and mask shape mismatch are different bugs." }, { "left": "issue:43632", "right": "issue:45834", "accept": false, - "reason": "_is_hf_initialized regression and Kosmos2.5 long-OCR index error are unrelated." + "reason": "_is_hf_initialized flag regression and Kosmos2.5 long-OCR index error are unrelated." }, { "left": "issue:38175", "right": "issue:43994", - "accept": false, - "reason": "Both involve siglip2-base-patch16-224, but one is zero probabilities and the other is nonsensical AutoModel/pipeline output; not enough evidence of one bug." + "accept": true, + "reason": "Same SigLIP2 issue: incorrect loading/processing leads to zero or nonsensical outputs for the same model." }, { - "left": "issue:43824", - "right": "issue:43957", + "left": "issue:43638", + "right": "issue:44661", "accept": false, - "reason": "ImportError for a missing class and meta-device loading breakage are separate regressions." + "reason": "Deepspeed ZeRO-3 Bert index error and add-new-model-like tokenizer-mapping failure are different failures." }, { "left": "issue:43596", "right": "issue:44661", "accept": false, - "reason": "DeepSpeed ZeRO-3 BertModel index error and tokenizer-mapping registration failure are different issues." + "reason": "Both mention index/init problems, but the Bert ZeRO-3 error and tokenizer-mapping bug are not the same code path." }, { "left": "issue:44661", "right": "issue:44805", "accept": false, - "reason": "Tokenizer mapping registration and generic mask-shape IndexError do not describe the same defect." + "reason": "Tokenizer-mapping bug and mask/tensor shape mismatch are unrelated." }, { "left": "issue:44877", "right": "issue:45030", "accept": false, - "reason": "Both involve config loading, but one is strict config rejection for granite_speech and the other is glm4v validation changes; different models and symptoms." + "reason": "Strict config validation on granite_speech and glm4v config validation are separate model/config issues." }, { "left": "issue:43317", "right": "issue:44756", "accept": false, - "reason": "device_map auto dequantized-loading failure and Strix Halo mmap OOM are both memory-related, but not the same underlying bug." + "reason": "Device-map dequantized-model loading and Strix Halo mmap OOM are different concerns." }, { "left": "issue:43644", "right": "issue:44534", "accept": true, - "reason": "Near-verbatim duplicate: both report Transformers v5 filling non-persistent buffers with junk; 44534 is the same regression with slightly different wording." + "reason": "Exact same non-persistent-buffer corruption regression; 44534 is essentially a re-report of 43644." }, { "left": "issue:38175", "right": "issue:39692", "accept": false, - "reason": "SigLIP2 zero probabilities and SigLIP2 docs example errors are related to the model family but not the same concrete bug." + "reason": "SigLIP2 zero-probability output and a documentation example with multiple errors are not the same underlying bug." }, { "left": "issue:42491", "right": "issue:43720", "accept": false, - "reason": "Qwen3_moe LoRA incompatibility after hf5.x and BitNet packed-weight loading are different mechanisms." + "reason": "LoRA qwen3_moe compatibility across hf versions and BitNet packed-weight loading are unrelated." }, { "left": "issue:41720", "right": "issue:42915", "accept": false, - "reason": "Qwen3 auto device mapping cudaErrorAssert and Qwen3Moe FP8 failure are different runtime issues." + "reason": "Qwen3 auto device-mapping cuda assert and Qwen3-MoE FineGrainedFP8 failure are different issues." }, { "left": "issue:44960", "right": "issue:44977", "accept": false, - "reason": "GLM5 and Qwen3.5 flash-attention generation failures are unrelated model-specific bugs." + "reason": "GLM5 and Qwen3.5 flash-attention generation failures are different models and different failures." }, { "left": "issue:45313", "right": "issue:45357", "accept": false, - "reason": "ZeRO-3 failing to load language_model weights and save_pretrained writing wrong visual encoder keys are opposite directions of different bugs." + "reason": "ZeRO-3 weight loading for Qwen3.5 language_model and incorrect visual-encoder keys in save_pretrained are separate bugs." }, { "left": "issue:44451", "right": "issue:44488", - "accept": false, - "reason": "Both are load failures for different BERT models, but the titles do not indicate the same concrete root cause." - }, - { - "left": "issue:43645", - "right": "issue:43646", "accept": true, - "reason": "Same custom-model initialization regression; 43645 is the notebook-specific manifestation of 43646." - }, - { - "left": "issue:44560", - "right": "issue:44918", - "accept": false, - "reason": "The Qwen3-vl video StopIteration and TRL SFT embedding unpacking failures are different." + "reason": "Same loading failure class for the same BERT-style models; 44488 is a duplicate report of the model-loading regression." }, { "left": "issue:42915", "right": "issue:44756", "accept": false, - "reason": "Qwen3Moe FP8 failure and Strix Halo mmap/OOM are not the same bug." + "reason": "Qwen3-MoE FP8 failure and Strix Halo mmap OOM are unrelated." }, { "left": "issue:42491", "right": "issue:45276", "accept": false, - "reason": "Qwen3_moe LoRA incompatibility and Gemma4 resize_token_embeddings mismatch are unrelated." + "reason": "Qwen3 MoE LoRA compatibility and Gemma4 resize_token_embeddings behavior are different problems." }, { "left": "issue:43708", "right": "issue:44075", "accept": false, - "reason": "Trainer resume/max_steps calculation and SGD arguments not being used are different training bugs." + "reason": "Trainer resume/max_steps calculation and SGD args being ignored are different Trainer/optimizer issues." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is well-grounded in the packet and stays conservative on the soft edges. The two accepted duplicates (43644\u219444534 and 43645\u219443646) are plausibly the same regressions, and the other pairs are appropriately rejected without overstating evidence." + "feedback": "The summary is grounded in the packet and stays conservative overall. The accepted duplicate pairs are plausible from the titles/body metadata: 43644/44534 is effectively the same non-persistent-buffer regression, 43645/43646 are clearly the same custom-model initialization bug, 44451/44488 are the same BERT loading regression, and 38175/43994 are reasonably treated as a SigLIP2 duplicate pair. The false soft-edge verdicts are mostly cautious and the cluster-level summary correctly reflects that this set is largely unrelated issues with a few duplicate groups." }, "evaluator_used": true, "retried": false, @@ -1874,256 +2072,256 @@ "issue:44164|issue:44756", "issue:43381|issue:44928" ], - "prepared_review_unit_hash": "2834d5b86e2d4784e23156998c966a46516b6c484f8512576a39e6bbb2ef7d53", + "prepared_review_unit_hash": "0ce7477a3c19eb43ffa84d6f03d00715bae1123b1ae74b5fd0a48f7a84fa59e0", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 36, - "serialized_chars": 38642, - "estimated_input_tokens": 9661, - "estimated_eval_tokens": 19578 + "serialized_chars": 38638, + "estimated_input_tokens": 9660, + "estimated_eval_tokens": 19576 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly heterogeneous issue set with one likely duplicate pair around resize_token_embeddings not updating output embeddings; the rest are related only at a broad subsystem level and should stay separate.", - "confidence": 0.91, - "canonical_issue_reason": "issue:45292 is the broader, model-agnostic report of the resize_token_embeddings/output_embeddings bug; issue:45276 looks like a Gemma4-specific manifestation of the same underlying problem.", + "summary": "Most items are unrelated model-specific regressions or broad theme overlaps. The only clear duplicate family is the resize_token_embeddings failure affecting output_embeddings / model-specific embedding modules across related architectures.", + "confidence": 0.73, + "canonical_issue_reason": "Issue 45292 is the cleanest, most general statement of the resize_token_embeddings regression and best serves as the duplicate anchor for the embedding-resize subgroup.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45292 is the cleanest representative because it states the core bug generically and can subsume the model-specific variant.", + "best_issue_reason": "Issue 45292 is the strongest overall issue representative in this cluster because it is concise, model-agnostic, and directly captures the shared failure mode behind the related resize_token_embeddings reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43854", "right": "issue:43931", "accept": false, - "reason": "Both are model-loading failures, but they involve different models and different failure modes (SigLIP2 example mismatch vs Qwen3-VL weight-shape mismatch)." + "reason": "Different model/loading failures: GLM-4.7 unit-test loading vs Qwen3-VL weight-shape mismatch." }, { "left": "issue:43257", "right": "issue:43854", "accept": false, - "reason": "Different subsystems and symptoms: Qwen3 MoE conversion/offload with accelerate+deepspeed versus a SigLIP2 loading/example issue." + "reason": "Qwen3 MoE conversion/offload problem is unrelated to GLM-4.7 test loading." }, { "left": "issue:45488", "right": "issue:45812", "accept": false, - "reason": "Both are tokenizer regressions, but the affected models and concrete failure modes differ; this is too broad to treat as one bug." + "reason": "Both are tokenizer regressions, but they target different model families and likely different code paths." }, { "left": "issue:44488", "right": "issue:44991", "accept": false, - "reason": "Both are tokenizer/model-loading complaints, but they concern different models and likely different fixes." + "reason": "Both concern tokenizer loading on transformers 5.x, but the model-specific failures are distinct." }, { "left": "issue:39692", "right": "issue:43994", "accept": false, - "reason": "Same model family, but one is a broken documentation example and the other is incorrect inference behavior; not the same underlying bug." + "reason": "SigLIP2 docs/example errors and bad AutoModel/pipeline results are related only at a very broad model-family level." }, { "left": "issue:43644", "right": "issue:43645", "accept": false, - "reason": "Both mention Transformers 5.0.0, but one is buffer initialization junk and the other is custom-model/Jupyter initialization behavior." + "reason": "Different regressions: junk non-persistent buffers vs Jupyter custom-model initialization." }, { "left": "issue:43856", "right": "issue:43873", "accept": false, - "reason": "Both involve memory/quantization themes, but one is Qwen3 MoE training memory usage and the other is offloading under quantization." + "reason": "Training memory usage and quantized offloading behavior are not the same underlying bug." }, { "left": "issue:42491", "right": "issue:43931", "accept": false, - "reason": "Different models and different breakages: qwen3_moe LoRA compatibility versus Qwen3-VL loading shape mismatch." + "reason": "Unrelated issues: Qwen3 MoE LoRA compatibility vs Qwen3-VL load mismatch." }, { "left": "issue:43856", "right": "issue:44387", "accept": false, - "reason": "Both mention memory pressure, but the causes differ materially (Qwen3 MoE training vs int4 quantization reserved memory/OOM)." + "reason": "Both involve Qwen3/quantization memory, but one is training efficiency and the other is reserved-memory OOM under int4." }, { "left": "issue:45362", "right": "issue:45406", "accept": false, - "reason": "Unrelated failures: Qwen3.5 chat crashes versus Gemma4Processor missing _tokenizer in transformers serve." + "reason": "Different crashes in different subsystems/models: chat loading vs serve processor attribute error." }, { "left": "issue:44960", "right": "issue:46050", "accept": false, - "reason": "No shared underlying bug: a GLM5 issue versus save/load corruption after sinq quantization." + "reason": "GLM5 loading and sinq save/load corruption are different bugs." }, { "left": "issue:42491", "right": "issue:45292", "accept": false, - "reason": "Different areas and root causes; one is LoRA portability across branches, the other is resizing embeddings not updating outputs." + "reason": "Qwen3 MoE LoRA compatibility is unrelated to the embedding resize regression." }, { "left": "issue:44910", "right": "issue:45588", "accept": false, - "reason": "Both are runtime crashes, but the code paths differ completely (Qwen3.5 flash-attn position_ids vs sink-less flash_attention.py s_aux handling)." + "reason": "Flash-Attention position_ids crash and flash_attention.py s_aux=None crash are different code-path failures." }, { "left": "issue:44164", "right": "issue:44291", "accept": false, - "reason": "Both are model-loading/save issues, but one is extra_state serialization and the other is init_empty_weights/_is_hf_initialized argument handling." + "reason": "save/from_pretrained extra_state handling and init_empty_weights argument handling are separate loading issues." }, { "left": "issue:36010", "right": "issue:39404", "accept": false, - "reason": "Different API regressions: GenerationMixin import path versus Whisper pipeline return_language behavior." + "reason": "ImportError for GenerationMixin and Whisper return_language pipeline break are unrelated." }, { "left": "issue:45276", "right": "issue:45335", - "accept": false, - "reason": "These are similar only at a very broad level; Gemma4 and t5gemma hit different embedding paths and are not clearly the same concrete bug." + "accept": true, + "reason": "Both report the same resize_token_embeddings failure class: resizing does not propagate to model-specific embedding/output modules." }, { "left": "issue:43404", "right": "issue:43454", "accept": false, - "reason": "Both are weight-tying bugs, but they affect different model classes and likely require different fixes." + "reason": "Similar symptom (lm_head not tied), but different model implementations and not enough evidence of a shared code path." }, { "left": "issue:43425", "right": "issue:43957", "accept": false, - "reason": "Torch version compatibility and meta-device loading are separate issues with different causes." + "reason": "Torch version incompatibility and meta-device loading failures are different issues." }, { "left": "issue:41720", "right": "issue:44155", "accept": false, - "reason": "Completely different problems: Qwen3 auto device mapping CUDA assert versus AudioFlamingo3 batch token leakage." + "reason": "Qwen3 auto device-map crash and AudioFlamingo3 batched inference leakage are unrelated." }, { "left": "issue:39692", "right": "issue:44960", "accept": false, - "reason": "Different model families and failure modes; a doc/example bug versus a GLM5 loading issue." + "reason": "SigLIP2 example problems and GLM5 loading are different model regressions." }, { "left": "issue:42947", "right": "issue:45704", "accept": false, - "reason": "Both concern training/inference behavior, but gradient checkpointing with LoRA is unrelated to T5 using apex FusedRMSNorm." + "reason": "Gradient checkpointing/LoRA behavior is unrelated to T5 apex RMSNorm memory leak." }, { "left": "issue:44402", "right": "issue:44488", "accept": false, - "reason": "Both are tokenizer/vocab loading problems, but they concern different models and specific failure modes." + "reason": "Different tokenizer-loading/model-vocab issues with no clear shared underlying bug." }, { "left": "issue:43873", "right": "issue:44387", "accept": false, - "reason": "Different quantization-related performance problems; one is offloading behavior and the other is increased reserved memory causing OOM." + "reason": "Quantization offloading misbehavior and int4 reserved-memory OOM are related only at a broad theme level." }, { "left": "issue:45663", "right": "issue:45759", "accept": false, - "reason": "Different model architecture bugs: Gemma4 FSDP2 shared_kv_states handling versus Qwen multimodal text_config unwrapping." + "reason": "Gemma4 FSDP2 shared_kv_states bug and Qwen multimodal text_config unwrapping are different model/config issues." }, { "left": "issue:44291", "right": "issue:44756", "accept": false, - "reason": "A meta-initialization TypeError and a Strix Halo mmap OOM are unrelated." + "reason": "Meta init TypeError and Strix Halo mmap OOM are unrelated." }, { "left": "issue:45663", "right": "issue:45859", "accept": false, - "reason": "Different code paths: Gemma4 FSDP2 attention state rebuilding versus Qwen3.5 MoE rotary embedding offload compatibility." + "reason": "Different models and failures: Gemma4 FSDP2 KeyError vs Qwen3_5Moe CPU offload incompatibility." }, { "left": "issue:44155", "right": "issue:45823", "accept": false, - "reason": "No shared bug; AudioFlamingo3 batched inference leakage is unrelated to Gemma4 PLE device mismatch." + "reason": "AudioFlamingo3 embedding/token leak and Gemma4 device mismatch are unrelated." }, { "left": "issue:43335", "right": "issue:45663", "accept": false, - "reason": "SwitchTransformers sparse-layer construction and Gemma4 FSDP2 attention state rebuilding are unrelated." + "reason": "Sparse-layer config bug and Gemma4 attention KeyError are unrelated." }, { "left": "issue:44292", "right": "issue:44912", "accept": false, - "reason": "Both are quantization issues, but they involve different quantization formats and different loading failures." + "reason": "Qwen NVFP4 loading failure and git-oss MXFP4 fallback are different quantization bugs." }, { "left": "issue:44387", "right": "issue:44756", "accept": false, - "reason": "Both involve OOMs, but one is quantization-related memory reservation and the other is mmap on Strix Halo." + "reason": "Reserved-memory OOM under int4 and Strix Halo mmap OOM are different failure modes." }, { "left": "issue:43957", "right": "issue:44756", "accept": false, - "reason": "Meta-device loading regressions are unrelated to the Strix Halo mmap OOM report." + "reason": "Meta-device loading regressions and mmap/OOM on Strix Halo are unrelated." }, { "left": "issue:43854", "right": "issue:45216", "accept": false, - "reason": "Different failures: model-test loading mismatch versus a Qwen3.5 save_pretrained regression." + "reason": "GLM-4.7 load/test failure and Qwen3.5 save_pretrained regression are different issues." }, { "left": "issue:45276", "right": "issue:45292", "accept": true, - "reason": "These describe the same resize_token_embeddings bug; 45276 is the Gemma4-specific manifestation and 45292 is the generic output_embeddings regression." + "reason": "Same underlying bug: resize_token_embeddings does not update output/embedded weights correctly." }, { "left": "issue:42947", "right": "issue:43381", "accept": false, - "reason": "Both mention gradient checkpointing, but one is ineffective with PEFT LoRA and the other is disallowed in eval mode." + "reason": "Gradient checkpointing in PEFT LoRA and eval-mode checkpointing are distinct behaviors." }, { "left": "issue:44164", "right": "issue:44756", "accept": false, - "reason": "Serialization of extra_state and mmap-related OOM are unrelated." + "reason": "Extra_state save/load failure and mmap/OOM on Strix Halo are unrelated." }, { "left": "issue:43381", "right": "issue:44928", "accept": false, - "reason": "Gradient checkpointing in eval mode is unrelated to Qwen3.5 3D position_ids causing SDPA fallback and NaNs." + "reason": "Eval-mode gradient checkpointing and RLHF NaN/SDPA fallback are different bugs." } ] }, "evaluator_result": { "accept": true, - "feedback": "Grounded summary. The only accepted duplicate pair (issue:45276 / issue:45292) is well supported by the titles and the shared resize_token_embeddings/output_embeddings symptom, and the other pairwise rejections stay conservative given the differing models, subsystems, or failure modes." + "feedback": "The clustering is grounded in the packet and the soft-edge calls are conservative overall. The resize_token_embeddings group is the only clearly mergeable duplicate family, and using issue 45292 as the canonical anchor is reasonable given its model-agnostic wording. No obvious overstatement beyond the evidence shown." }, "evaluator_used": true, "retried": false, @@ -2227,233 +2425,236 @@ "issue:44671|issue:44991", "issue:43856|issue:44756" ], - "prepared_review_unit_hash": "e2c4276ca2ccb69fe173d2fde5330234bc9919f97fd280aa771ca8b3b1e29124", + "prepared_review_unit_hash": "4547e36f0048699a8b0d6ddcc47820d6d9ebbf8d929ee6bacf4938de29314f68", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 32, - "serialized_chars": 37359, - "estimated_input_tokens": 9340, - "estimated_eval_tokens": 18936 + "serialized_chars": 37355, + "estimated_input_tokens": 9339, + "estimated_eval_tokens": 18934 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a family of Transformers v5 regression reports around model/tokenizer loading and custom-model initialization, but the soft pairs are mostly different manifestations rather than true duplicates.", - "confidence": 0.67, - "canonical_issue_reason": "issue:43611 is the broadest umbrella for the v5 custom-model loading regression; it captures the general failure mode that several nearby issues describe in narrower ways.", + "summary": "This cluster is mostly a loose theme of Transformers v5 regressions across unrelated models/code paths. Only one soft pair looks like a true duplicate; the rest are just similarly worded issues and should stay separate.", + "confidence": 0.57, + "canonical_issue_reason": "#45292 is the broadest, most central report: it describes the underlying resize_token_embeddings propagation bug, and #45335 looks like a model-specific instance of the same failure.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43611 is the most representative single issue for the cluster because it names the general loading breakage directly and is broad enough to subsume related follow-ups.", + "best_issue_reason": "#45292 is the best representative issue because it is the most general, actionable statement of the bug and can subsume the narrower t5gemma-specific report.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45684", "right": "issue:45698", "accept": false, - "reason": "Both involve save/from_pretrained and custom modules, but one is about permissions on saved source files and the other is about loading the wrong module; different bugs." + "reason": "Permissions propagation and wrong custom-module loading are different save/load problems, not the same bug." }, { "left": "issue:44479", "right": "issue:44877", "accept": false, - "reason": "Both are v5/model-loading regressions, but they target different models and failure modes (video input regression vs strict config loading)." + "reason": "A video-input regression in Qwen VL models is unrelated to strict config blocking granite_speech loading." }, { "left": "issue:45588", "right": "issue:45663", "accept": false, - "reason": "Different code paths: flash_attention sink-less crash vs Gemma4 FSDP2 shared_kv_states rebuild/KeyError." + "reason": "Flash-attention crashing on s_aux=None is a different code path from Gemma4 FSDP2 shared_kv_states rebuilding." }, { "left": "issue:43646", "right": "issue:43708", "accept": false, - "reason": "Custom-model initialization in notebooks is unrelated to Trainer resume_from_checkpoint max_steps calculation." + "reason": "Custom model initialization breakage and Trainer resume max_steps miscalculation are unrelated issues." }, { "left": "issue:43645", "right": "issue:44164", "accept": false, - "reason": "One is notebook-based custom model initialization; the other is save/from_pretrained extra_state handling. Same area, but not the same bug." + "reason": "Notebook custom-model initialization and extra_state save/load failures are separate lifecycle bugs." }, { "left": "issue:43828", "right": "issue:43957", "accept": false, - "reason": "Both are dtype/loading problems, but one is an autocast mismatch for Phi-tiny-MoE and the other is meta-device loading on Transformers 5." + "reason": "These are both loading/runtime mismatches, but one is autocast dtype behavior and the other is meta-device loading." }, { "left": "issue:45292", "right": "issue:45335", - "accept": false, - "reason": "Both mention resize_token_embeddings, but one reports output_embeddings not updating and the other is a t5gemma-specific decoder.embed_tokens symptom; not clearly the same concrete bug." + "accept": true, + "reason": "Both report resize_token_embeddings not updating downstream embeddings/output layers; one is a general report and the other a t5gemma-specific manifestation." }, { "left": "issue:44402", "right": "issue:44451", "accept": false, - "reason": "Both are tokenizer-load failures, but one is a vocab-size/special-token mismatch and the other is a specific model load failure; too little evidence they share one root cause." + "reason": "Tokenizer vocab-size mismatch and inability to load a specific model are not clearly the same underlying bug." }, { "left": "issue:43425", "right": "issue:43828", "accept": false, - "reason": "Unrelated incompatibility reports: Torch 2.10 support vs an autocast dtype mismatch on a specific MoE model." + "reason": "Torch 2.10 compatibility and autocast dtype mismatch are different failures." }, { "left": "issue:43257", "right": "issue:44863", "accept": false, - "reason": "Different model families and different load-time failures; not the same underlying bug." + "reason": "Qwen3 MoE conversion under accelerate+deepspeed and NemotronH checkpoint loading are distinct model-loading problems." }, { "left": "issue:44683", "right": "issue:45823", "accept": false, - "reason": "Compiled flex_attention on torch>=2.9 and Gemma4 device_map device mismatch are distinct problems." + "reason": "Compiled flex_attention on torch>=2.9 and Gemma4 PLE device-map mismatch do not share the same concrete code-path." }, { "left": "issue:44589", "right": "issue:44756", "accept": false, - "reason": "Storage lookup failure for float8 tensors is unrelated to the Strix Halo mmap OOM report." + "reason": "Float8 storage lookup failure and mmap-induced OOM are unrelated issues." }, { "left": "issue:39404", "right": "issue:45701", "accept": false, - "reason": "Whisper pipeline return_language regression and tokenization changes across versions are different issues." + "reason": "Whisper return_language pipeline regression is unrelated to tokenization changes in general." }, { "left": "issue:43425", "right": "issue:45823", "accept": false, - "reason": "Torch version incompatibility is not the same as Gemma4 PLE device mismatch during forward." + "reason": "Torch version incompatibility is not the same as Gemma4 PLE forward device mismatch." }, { "left": "issue:43425", "right": "issue:43716", "accept": false, - "reason": "Generic Torch compatibility report and Mistral-3 image-preprocessor dtype mismatch are different code paths." + "reason": "Torch 2.10 incompatibility and Mistral-3 image preprocessor dtype mismatch are different regressions." }, { "left": "issue:43611", "right": "issue:43645", "accept": false, - "reason": "Both relate to custom model loading in v5, but one is specifically about base_model_prefix and the other about notebook initialization; not the same concrete regression." + "reason": "Base_model_prefix loading and Jupyter custom-model initialization are separate Transformers 5 breakages." }, { "left": "issue:43906", "right": "issue:44530", "accept": false, - "reason": "These are unrelated: an isolated reproduction of an older issue versus a Qwen3.5 PagedAttentionCache crash." + "reason": "A reproduction of another issue and a Qwen3.5 PagedAttentionCache crash are not the same bug." }, { "left": "issue:44530", "right": "issue:45084", "accept": false, - "reason": "Both touch attention/compilation features, but one is a linear_attention group-type crash and the other is a non-template-node compilation error." + "reason": "PagedAttentionCache linear_attention failure and non-template-node compile errors are different compiler/runtime problems." }, { "left": "issue:39404", "right": "issue:44912", "accept": false, - "reason": "Whisper return_language regression and MXFP4 quantization fallback are unrelated." + "reason": "Whisper pipeline behavior and MXFP4 quantization fallback are unrelated regressions." }, { "left": "issue:44265", "right": "issue:44898", "accept": false, - "reason": "torch.export/torch_compilable_check failure is unrelated to Perceiver interpolation failure." + "reason": "torch.export compatibility with torch_compilable_check is a different failure from Perceiver resolution handling." }, { "left": "issue:43901", "right": "issue:44509", "accept": false, - "reason": "Both are docs issues, but they cover different removed pipeline/task references and do not look like the same document bug." + "reason": "Both are docs-related, but they document different removed/deprecated pipeline behaviors and should not be merged." }, { "left": "issue:44479", "right": "issue:44805", "accept": false, - "reason": "Different regressions: video input support for Qwen VL models versus a mask-shape IndexError." + "reason": "A Qwen video-input regression and a generic mask/index shape error are not clearly the same issue." }, { "left": "issue:43994", "right": "issue:44079", "accept": false, - "reason": "A wrong-result model/pipeline report and a ModelOutput key assignment bug are different problems." + "reason": "Incorrect inference results and ModelOutput key assignment with None are different problems." }, { "left": "issue:43425", "right": "issue:44918", "accept": false, - "reason": "Torch incompatibility and TRL unpacking for Qwen3.5 embeddings are different issues." + "reason": "Torch compatibility and TRL embedding unpacking are unrelated." }, { "left": "issue:44589", "right": "issue:45084", "accept": false, - "reason": "Float8 storage lookup failure and non-template-node compilation error do not share the same underlying bug." + "reason": "Storage-object lookup failure is unrelated to compile-time non-template-node errors." }, { "left": "issue:43299", "right": "issue:43931", "accept": false, - "reason": "Both are Qwen3-VL load issues, but one is a Transformers 5.0 model-conversion regression and the other is a weight-shape mismatch for a specific checkpoint." + "reason": "Qwen3VL MoE loading regression and a specific weight-shape mismatch on Qwen3-VL-30B-A3B-Instruct are not the same concrete bug." }, { "left": "issue:43596", "right": "issue:44683", "accept": false, - "reason": "Different failures: zero3 init/BertModel indexing versus compiled flex_attention on newer torch." + "reason": "DeepSpeed zero3/BertModel index errors and flex_attention compilation failures are separate code paths." }, { "left": "issue:44164", "right": "issue:45759", "accept": false, - "reason": "save/from_pretrained extra_state handling is unrelated to from_config failing to unwrap text_config for composite Qwen configs." + "reason": "extra_state save/load handling and unwrapping text_config in from_config are different model-serialization issues." }, { "left": "issue:39404", "right": "issue:44671", "accept": false, - "reason": "Whisper pipeline behavior and CamemBERT masked-LM prediction regression are unrelated." + "reason": "Whisper pipeline regression and CamemBERT MLM prediction correctness are unrelated." }, { "left": "issue:36032", "right": "issue:43452", "accept": false, - "reason": "Tokenizer special-token method conflicts and gguf_file from_pretrained failures are different loading bugs." + "reason": "Tokenizer add_special_tokens conflict and gguf_file from_pretrained breakage are different tokenizer/model-loading bugs." }, { "left": "issue:44671", "right": "issue:44991", "accept": false, - "reason": "Both are tokenizer/model loading regressions, but one is CamemBERT masked-LM output and the other is a tokenizer load failure for a specific model; not the same bug." + "reason": "CamemBERT prediction correctness and tokenizer loading for EMBEDDIA/est-roberta are not the same underlying bug." }, { "left": "issue:43856", "right": "issue:44756", "accept": false, - "reason": "Qwen3 MoE training memory usage and disabling mmap for Strix Halo OOM are different performance/memory issues." + "reason": "Qwen3 MoE training memory usage and Strix Halo mmap OOM are different performance/memory issues." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative about the soft pairs. The only accepted duplicate pair, issue:45292 vs issue:45335, is well supported by the titles, and the other pairwise rejections do not overclaim equivalence." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -2553,20 +2754,20 @@ "issue:43645|issue:44534", "issue:43646|issue:44534" ], - "prepared_review_unit_hash": "e9ad1932cdc3468b4ad9937f89a8960fcb512496cdabf93a459b74203307b1a7", + "prepared_review_unit_hash": "94b2982e355b2a834df6cb1dbd3513ef648ba574813a190d787b268117342a0a", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 33, - "serialized_chars": 37027, + "serialized_chars": 37025, "estimated_input_tokens": 9257, "estimated_eval_tokens": 18770 }, @@ -2575,180 +2776,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose mix of unrelated Transformers bug reports and regressions; the soft-similarity links are mostly title-level overlaps, not the same underlying defect. No PRs are present, and I don't see a true duplicate pair worth merging.", - "confidence": 0.82, - "canonical_issue_reason": "No single issue cleanly represents a duplicate set here; the cluster spans many distinct subsystems (trainer, tokenizers, multimodal, generation, loading regressions).", + "summary": "This cluster looks mostly heterogeneous. There is a loose Transformers v5 subgroup around loading/initialization/buffer regressions, but the visible evidence does not support treating it as a confirmed duplicate set. I would avoid a strong canonical anchor and treat 44534 only as a lightweight representative of that weak subtheme, not the whole cluster.", + "confidence": 0.62, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "If one issue must stand in as the closest cluster anchor, issue 43646 is the broadest/general regression report, but it still does not unify the rest into one concrete bug.", + "best_issue_reason": "44534 is the clearest representative of the loose Transformers v5 buffer/init regression subtheme, but only as a provisional topic anchor. It is not strong enough evidence to collapse the cluster into one duplicate issue.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:32090", "right": "issue:39404", "accept": false, - "reason": "Different failures: Trainer GPU broadcast TypeError vs Whisper pipeline return_language regression." + "reason": "Different subsystems and symptoms: Trainer GPU broadcast TypeError vs Whisper pipeline return_language regression." }, { "left": "issue:44991", "right": "issue:45081", "accept": false, - "reason": "Both involve tokenizers/loading, but one is a tokenizer load regression and the other is a Mistral regex patch crash with a different root cause." + "reason": "Both are loading-related regressions, but they affect different paths and models; not the same concrete bug." }, { "left": "issue:44164", "right": "issue:44479", "accept": false, - "reason": "save/from_pretrained extra_state handling is unrelated to the Qwen2.5-VL video-input regression." + "reason": "save/from_pretrained extra_state handling is unrelated to the Qwen video-input regression." }, { "left": "issue:44737", "right": "issue:45859", "accept": false, - "reason": "Both mention CPU/device behavior, but XLNet CPU arange overhead and Qwen3_5Moe CPU-offload incompatibility are different code paths." + "reason": "Both involve device behavior, but the model-specific failures and code paths differ." }, { "left": "issue:43582", "right": "issue:43632", "accept": false, - "reason": "Apple Silicon allocator warmup TypeError is unrelated to the _is_hf_initialized flag regression." + "reason": "Apple Silicon warmup TypeError and _is_hf_initialized flag regression are different defects." }, { "left": "issue:44661", "right": "issue:44683", "accept": false, - "reason": "Tokenizers_MAPPING_NAME add-new-model-like failure is unrelated to compiled flex_attention on torch >= 2.9." + "reason": "Tokenizer mapping logic and compiled flex_attention failures are unrelated." }, { "left": "issue:44488", "right": "issue:45081", "accept": false, - "reason": "Both are model/tokenizer loading issues, but the affected code paths and symptoms differ." + "reason": "Generic model loading failure vs Mistral tokenizer regex patch crash; same area, different bug." }, { "left": "issue:44488", "right": "issue:44987", "accept": false, - "reason": "These are separate loading regressions for different models and versions, not one concrete bug." + "reason": "Both are loading complaints, but the models and failure modes differ too much to treat as duplicates." }, { "left": "issue:43611", "right": "issue:43644", "accept": false, - "reason": "Base_model_prefix loading breakage and junk-filled non-persistent buffers are distinct v5 regressions." + "reason": "The titles suggest a related Transformers v5 theme, but the concrete symptoms differ and the packet does not support a confirmed duplicate." }, { "left": "issue:40990", "right": "issue:43421", "accept": false, - "reason": "Perplexity anomaly on gpt-oss-20b is unrelated to runtime post-processor updates in tokenizers backend." + "reason": "Perplexity regression is unrelated to runtime special-token/post-processor update behavior." }, { "left": "issue:43716", "right": "issue:45084", "accept": false, - "reason": "Image preprocessor dtype mismatch is unrelated to template compilation failures." + "reason": "Different bug classes: image preprocessor dtype mismatch vs template-node compilation error." }, { "left": "issue:44206", "right": "issue:44220", "accept": false, - "reason": "Unsupported center arg in LasrFeatureExtractor is not the same as a low-level _torch_extract_fbank_features issue." + "reason": "Both are audio-feature related, but not clearly the same underlying defect or code path." }, { "left": "issue:36032", "right": "issue:42915", "accept": false, - "reason": "Tokenizer add_special_tokens name conflict is unrelated to Qwen3Moe FineGrainedFP8Config failure." + "reason": "T5 tokenizer conflict and Qwen3Moe FP8 failure are unrelated." }, { "left": "issue:43643", "right": "issue:44530", "accept": false, - "reason": "Missing fields with trust_remote_code=True is unrelated to PagedAttentionCache linear_attention crashes." + "reason": "AutoConfig trust_remote_code field loss is unrelated to the Qwen3.5 PagedAttention crash." }, { "left": "issue:34567", "right": "issue:40444", "accept": false, - "reason": "TrainerState num_input_tokens_seen not updating is unrelated to Qwen2.5-VL iterable multimodal finetuning failures." + "reason": "TrainerState token counting and multi-image IterableDataset finetuning failures are unrelated." }, { "left": "issue:43883", "right": "issue:44488", "accept": false, - "reason": "Molmo all_tied_weights_keys missing and model-loading failures for cjvt/sleng-bert are different problems." + "reason": "Missing tied-weights metadata and BERT loading failure are different issues." }, { "left": "issue:44811", "right": "issue:45584", "accept": false, - "reason": "Whisper batch_decode skip_special_tokens bug is unrelated to empty-transcription generation after align_special_tokens." + "reason": "Both are Whisper issues, but the specific symptoms and code paths differ." }, { "left": "issue:43541", "right": "issue:45084", "accept": false, - "reason": "Mixtral grouped_mm torch-dynamo tracing error is a different failure than non-template node compilation." + "reason": "Torch dynamo grouped_mm tracing failure is unrelated to non-template-node compilation errors." }, { "left": "issue:42915", "right": "issue:44530", "accept": false, - "reason": "Qwen3Moe FP8 config failure and Qwen3.5 PagedAttentionCache linear_attention crash are separate model/runtime bugs." + "reason": "Both are Qwen-family regressions, but the FP8 config failure and linear_attention cache crash are different bugs." }, { "left": "issue:44451", "right": "issue:44987", "accept": false, - "reason": "ScandiBERT loading regression is unrelated to physical-intelligence/fast loading on transformers>=5.1.0." + "reason": "Both are model-loading failures, but there is no evidence of the same root cause." }, { "left": "issue:44464", "right": "issue:45987", "accept": false, - "reason": "Chunked generation inconsistency with compiled forward is not the same as StaticCache.get_seq_length() returning the wrong type." + "reason": "Chunked generation inconsistency and seq_length return-type contract issue are unrelated." }, { "left": "issue:43644", "right": "issue:43646", "accept": false, - "reason": "Both are v5 regressions, but one is non-persistent buffers and the other is custom model initialization; not the same bug." + "reason": "Custom model initialization breakage is only superficially related to the non-persistent-buffer junk regression." }, { "left": "issue:43452", "right": "issue:43482", "accept": false, - "reason": "gguf_file breaks auto loading and Qwen2.5-GGUF loading on v5 are related in area but not clearly the same concrete defect." + "reason": "Both touch GGUF loading, but the visible descriptions point to different concrete failures." }, { "left": "issue:43317", "right": "issue:43482", "accept": false, - "reason": "gpu+cpu offload dequantized-model loading and Qwen2.5-GGUF v5 loading failure are separate load-path regressions." + "reason": "GPU+CPU offload loading and Qwen2.5-GGUF loading are different failure modes." }, { "left": "issue:36032", "right": "issue:43645", "accept": false, - "reason": "Tokenizer special-token method conflict is unrelated to Jupyter custom model initialization breakage." + "reason": "Tokenizer method conflict and notebook custom model initialization are unrelated." }, { "left": "issue:44849", "right": "issue:45325", "accept": false, - "reason": "Qwen3.5 output_hidden_states bug and Qwen2.5-VL rope-index scaling issue affect different model internals." + "reason": "Different Qwen model regressions with different symptoms and code paths." }, { "left": "issue:44987", "right": "issue:44991", "accept": false, - "reason": "Different failing models and symptoms; both are loading regressions but not one shared bug." + "reason": "Both are transformers>=5 loading failures, but for different models and no shared concrete bug is visible." }, { "left": "issue:45084", "right": "issue:45507", "accept": false, - "reason": "Template-node compilation error is unrelated to GraniteMoEHybrid invalid method calls." + "reason": "Non-template-node compilation error and GraniteMoE invalid-method call are unrelated." }, { "left": "issue:43883", @@ -2760,31 +2961,34 @@ "left": "issue:43262", "right": "issue:43377", "accept": false, - "reason": "Audio apply_chat_template sampling-rate default is unrelated to MIMI encoder padding-mask behavior." + "reason": "Audio chat-template sample-rate defaulting and MIMI padding-mask mismatch are different bugs." }, { "left": "issue:43089", "right": "issue:45987", "accept": false, - "reason": "Generation GPU sync/dispatch overhead is a performance issue, not the same as StaticCache return-type regression." + "reason": "Generation performance overhead and seq_length contract regression are not the same issue." }, { "left": "issue:43645", "right": "issue:44534", "accept": false, - "reason": "Custom model initialization in notebooks and junk-filled non-persistent buffers are separate v5 breakages." + "reason": "Custom model initialization in notebooks is distinct from the non-persistent-buffer junk regression." }, { "left": "issue:43646", "right": "issue:44534", "accept": false, - "reason": "Custom model initialization and non-persistent buffer corruption are different bugs despite both being v5 regressions." + "reason": "Custom model initialization breakage is only loosely related to the buffer-junk regression; different concrete defects." } ] }, - "evaluator_result": null, - "evaluator_used": false, - "retried": false, + "evaluator_result": { + "accept": false, + "feedback": "The summary is a bit too strong for the packet. The only pair the analyst marked as a true duplicate, issue:43611 vs issue:43644, is not clearly supported by the visible titles, and the packet\u2019s own soft-pair signal marks it as deterministic_accept=false. Because of that, calling this a clear duplicate theme and using 44534 as the canonical anchor overstates the evidence. The conservative read is that the cluster is mostly heterogeneous, with at most a loose Transformers v5 similarity around buffer/init regressions, not a confirmed duplicate set." + }, + "evaluator_used": true, + "retried": true, "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null @@ -2885,233 +3089,236 @@ "issue:42175|issue:45542", "issue:33666|issue:40444" ], - "prepared_review_unit_hash": "7095dbcc2d236cc59a85373a3857cfb180f31ef7ff1ccc977d1263a5c0ccb1fe", + "prepared_review_unit_hash": "a836a7a71932389f811ad07c23e8015dac0b2582c6a6143c6b918461a768d631", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 32, - "serialized_chars": 37175, - "estimated_input_tokens": 9294, - "estimated_eval_tokens": 18844 + "serialized_chars": 37185, + "estimated_input_tokens": 9297, + "estimated_eval_tokens": 18850 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Heterogeneous set of transformer bugs; the soft-similarity pairs mostly connect superficially similar but distinct issues across different models, features, and runtime paths. No true duplicate pair stands out, so this is not a cohesive duplicate cluster.", - "confidence": 0.95, - "canonical_issue_reason": "No single canonical issue is appropriate: the items span unrelated bugs, so any one issue would be a misleading umbrella.", + "summary": "This cluster is thematically related but not a true duplicate set: it mixes many independent Transformers v5/model-loading regressions, tokenizer errors, and multimodal preprocessing bugs. Only a couple of soft pairs look like the same underlying bug; most are merely similar by subsystem or model family.", + "confidence": 0.82, + "canonical_issue_reason": "issue:45020 is the broadest umbrella item in the set: it describes recent Transformers versions breaking remote_code model loading, which best matches the cluster\u2019s repeated pattern of version-regression breakages across models.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 44403 is the broadest/openest candidate, but it still does not actually cover the other failures well enough to serve as a real duplicate target.", + "best_issue_reason": "issue:45020 is the strongest representative to surface because it is broad, regression-focused, and most likely to anchor the shared version-breakage theme across the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43404", "right": "issue:44466", "accept": false, - "reason": "Different bugs: Mistral3 weight-tying serialization vs a broader save_pretrained inconsistency in Qwen3.5." + "reason": "Both concern tied weights in related models, but one is a missing tie in Mistral3ForConditionalGeneration and the other is device-dependent save_pretrained serialization; different bugs." }, { "left": "issue:43322", "right": "issue:43540", "accept": false, - "reason": "Both are multimodal loading failures, but one is a Llava Next segfault and the other is a Qwen3OmniMoe video-input ValueError." + "reason": "Different model families and failure modes: Llava Next segfault vs Qwen3OmniMoe video-input ValueError." }, { "left": "issue:44403", "right": "issue:45081", "accept": false, - "reason": "Generic loading-noise issue vs a specific tokenizer regex patch crash in Mistral." + "reason": "Generic loading noise vs a specific Mistral tokenizer crash; not the same underlying issue." }, { "left": "issue:43994", "right": "issue:44610", "accept": false, - "reason": "Unrelated failures: SigLIP2 produces bad outputs, while OmDet-Turbo has a processor/model image-size mismatch." + "reason": "Unrelated vision bugs: SigLIP2 bad outputs vs OmDet-Turbo processor/model input-size mismatch." }, { "left": "issue:42915", "right": "issue:45084", "accept": false, - "reason": "Different code paths: Qwen3Moe FineGrainedFP8Config vs template compilation failing on non-template nodes." + "reason": "Qwen3Moe FP8 config failure and a template compilation TypeError are different code paths." }, { "left": "issue:43377", "right": "issue:43819", "accept": false, - "reason": "Different models and bugs: MIMI padding-mask batching issue vs DAC from_latents missing-STE mismatch." + "reason": "Both are audio/codec related, but one is batch padding-mask behavior in MIMI and the other is DAC STE mismatch; not duplicates." }, { "left": "issue:44220", "right": "issue:44683", "accept": false, - "reason": "Audio feature extraction bug vs compiled flex_attention failure on torch >= 2.9." + "reason": "Feature extraction helper failure and compiled flex_attention regression are unrelated." }, { "left": "issue:44265", "right": "issue:44610", "accept": false, - "reason": "torch.export with torch_compilable_check is unrelated to the OmDet processor producing the wrong input size." + "reason": "Different subsystems: torch.export with torch_compilable_check vs OmDet-Turbo preprocessing size bug." }, { "left": "issue:44451", "right": "issue:45081", "accept": false, - "reason": "ScandiBERT loading failure is a different model-loading problem than the Mistral tokenizer backend crash." + "reason": "ScandiBERT loading failure and a Mistral regex/tokenizer crash are unrelated model-specific bugs." }, { "left": "issue:44987", "right": "issue:45081", "accept": false, - "reason": "Different regressions: loading physical-intelligence/fast vs tokenizer regex patching for Mistral." + "reason": "Physical-intelligence/fast loading regression is not the same as the Mistral tokenizer backend crash." }, { "left": "issue:42915", "right": "issue:44291", "accept": false, - "reason": "FP8 config failure and init_empty_weights/_is_hf_initialized argument handling are distinct issues." + "reason": "Both mention newer Transformers behavior, but FP8 config failure and init_empty_weights/_is_hf_initialized are distinct bugs." }, { "left": "issue:43632", "right": "issue:44756", "accept": false, - "reason": "Unrelated problems: _is_hf_initialized flag handling vs mmap-related OOM on Strix Halo." + "reason": "_is_hf_initialized breakage and Strix Halo mmap OOM are unrelated." }, { "left": "issue:45216", "right": "issue:45357", - "accept": false, - "reason": "Same model family but different regressions: checkpoint save correctness vs incorrect visual encoder keys." + "accept": true, + "reason": "Same Qwen3.5 save_pretrained regression family: both report incorrect serialized checkpoint contents/keys after v5.4+." }, { "left": "issue:36032", "right": "issue:43742", "accept": false, - "reason": "Tokenizer special-tokens conflict is unrelated to the MobileLLM key error." + "reason": "T5 tokenizer add_special_tokens conflict and MobileLLM key error are unrelated loading issues." }, { "left": "issue:44062", "right": "issue:44987", "accept": false, - "reason": "AddedToken keyword collision is a tokenizer API bug, not the physical-intelligence/fast loading regression." + "reason": "AddedToken special-argument conflict is a tokenizer construction bug, not the same as loading physical-intelligence/fast." }, { "left": "issue:42915", "right": "issue:44589", "accept": false, - "reason": "Qwen3Moe FP8 config failure is unrelated to the Float8 storage object error." + "reason": "FP8 config failure and missing Float8 storage type are different failure points and likely different fixes." }, { "left": "issue:44479", "right": "issue:45325", - "accept": false, - "reason": "Both touch Qwen VL, but one is a broad video-input regression and the other is a specific rope-index scaling bug." + "accept": true, + "reason": "Same Qwen2.5-VL temporal-position/rope-index regression in v5.3.0; 45325 is a narrower symptom of the broader video/input regression." }, { "left": "issue:43404", "right": "issue:45127", "accept": false, - "reason": "Weight-tying serialization bug is different from LoRA merge collapse with extended vocabulary." + "reason": "Both involve tied embeddings, but one is a model implementation bug and the other is LoRA merge collapse with extended vocab; not the same issue." }, { "left": "issue:43296", "right": "issue:43531", "accept": false, - "reason": "PaddleOCR-VL loading failure and Qwen3-MoE sliding_window behavior are different issues." + "reason": "PaddleOCR-VL loading and Qwen3-MoE sliding_window behavior are unrelated." }, { "left": "issue:40444", "right": "issue:45325", "accept": false, - "reason": "IterableDataset multi-image finetuning failure is not the same as the Qwen2.5-VL rope-index temporal scaling bug." + "reason": "IterableDataset multi-image training failure is not the same as the Qwen2.5-VL rope-index temporal-position bug." }, { "left": "issue:43531", "right": "issue:44877", "accept": false, - "reason": "Qwen3-MoE sliding_window bug vs strict config blocking granite_speech loading." + "reason": "Qwen3-MoE sliding_window bug and granite_speech strict-config loading are separate problems." }, { "left": "issue:36032", "right": "issue:41720", "accept": false, - "reason": "Tokenizer load conflict is unrelated to the Qwen3 auto-device-mapping cuda assert." + "reason": "Tokenizer construction conflict vs Qwen3 auto device mapping cudaErrorAssert are unrelated." }, { "left": "issue:43611", "right": "issue:44534", "accept": false, - "reason": "base_model_prefix loading incompatibility does not match the non-persistent-buffer junk-fill regression." + "reason": "Both are v5 regressions, but base_model_prefix loading and non-persistent buffer junk are different code paths." }, { "left": "issue:43425", "right": "issue:44403", "accept": false, - "reason": "Torch 2.10 compatibility is a versioning issue, not the generic loading-noise bug." + "reason": "Torch 2.10 compatibility and generic loading noise are not the same issue." }, { "left": "issue:43632", "right": "issue:44534", "accept": false, - "reason": "Unexpected _is_hf_initialized handling is distinct from non-persistent buffers being initialized with junk." + "reason": "_is_hf_initialized handling and non-persistent buffer initialization are different regression classes." }, { "left": "issue:40990", "right": "issue:44811", "accept": false, - "reason": "High perplexity on WikiText is an evaluation/model issue, not Whisper batch_decode skip_special_tokens behavior." + "reason": "High perplexity on gpt-oss-20b and Whisper batch_decode skip_special_tokens are unrelated." }, { "left": "issue:43792", "right": "issue:44589", "accept": false, - "reason": "Whisper run failure and missing Float8 storage are different runtime problems." + "reason": "Whisper runtime failure and missing Float8 storage are unrelated errors." }, { "left": "issue:44079", "right": "issue:45663", "accept": false, - "reason": "ModelOutput key assignment bug is unrelated to Gemma4 FSDP2 shared_kv_states reconstruction." + "reason": "Different KeyError mechanisms in different models; no shared underlying bug is evident." }, { "left": "issue:44683", "right": "issue:45020", "accept": false, - "reason": "Compiled flex_attention on torch >= 2.9 is unrelated to remote_code model-loading regressions." + "reason": "Compiled flex_attention on torch>=2.9 and remote_code loading regressions are separate concerns." }, { "left": "issue:41553", "right": "issue:43582", "accept": false, - "reason": "Bad Voxtral error messaging is not the same as the Apple Silicon caching_allocator_warmup TypeError." + "reason": "AutoTokenizer Voxtral error-message issue and Apple Silicon TypeError are unrelated." }, { "left": "issue:42175", "right": "issue:45542", "accept": false, - "reason": "Missing TensorFlow in the torch extra and undefined tf backend from tensorboard-only install are different packaging bugs." + "reason": "Both touch TensorFlow backend packaging, but one is missing tf in the torch extra and the other is an undefined backend error when only tensorboard is installed; not clearly the same bug." }, { "left": "issue:33666", "right": "issue:40444", "accept": false, - "reason": "Qwen2-VL multi-GPU training and Qwen2.5-VL iterable-dataset multi-image finetuning are separate training issues." + "reason": "Both are Qwen VL training issues, but multi-GPU training and multi-image iterable finetuning failures are different problems." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: it correctly characterizes the cluster as mostly unrelated Transformers v5/model-loading regressions and multimodal edge cases, with only a small number of plausible duplicate pairs. The canonical issue choice is defensible as a broad umbrella regression, and the soft-edge verdicts are generally conservative rather than overclaiming duplicates." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -3208,20 +3415,20 @@ "issue:42915|issue:43482", "issue:43296|issue:45800" ], - "prepared_review_unit_hash": "ff26653abec839ba4334c18cfb6d0384edbfddc25f47c116aaa0b865ca0dc967", + "prepared_review_unit_hash": "4f123c347033a7bd5895d969630551ed9e845ee306f620d21555b6a0f237250c", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36642, + "serialized_chars": 36641, "estimated_input_tokens": 9161, "estimated_eval_tokens": 18578 }, @@ -3230,193 +3437,196 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly heterogeneous transformer-regression reports. No soft pair is strong enough to treat as the same underlying bug or change, so I reject all soft edges and treat the set as unrelated issues with similar symptom wording.", - "confidence": 0.91, - "canonical_issue_reason": "issue:43646 is the broadest and most representative issue in the set, but only as a loose umbrella for v5 initialization regressions; the cluster overall does not form a true duplicate group.", + "summary": "The cluster is mostly heterogeneous; the only convincing duplicate-like edge is the pair of Transformers v5 model-initialization/load regressions around custom models/base_model_prefix.", + "confidence": 0.46, + "canonical_issue_reason": "issue:43646 is the broader, more representative statement of the v5 model-initialization regression and can absorb the narrower base_model_prefix-loading report.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43646 is the best single issue to anchor this noisy cluster because its title is broad enough to represent the general class of v5 initialization breakages seen here.", + "best_issue_reason": "Best anchor for the only close duplicate subcluster; broad enough to represent the v5 init/load breakage.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44403", "right": "issue:44991", "accept": false, - "reason": "Both involve loading behavior, but one is generic load noise and the other is a specific tokenizer-resolution regression for a particular repo." + "reason": "Both involve tokenization/loading, but the symptoms and target objects differ substantially." }, { "left": "issue:43421", "right": "issue:44568", "accept": false, - "reason": "One is about runtime special-token post-processing, the other about add_special_tokens failing to add BOS/EOS; related area, different bug." + "reason": "Related tokenizer behavior, but one is runtime post-processor updates and the other is missing BOS/EOS insertion." }, { "left": "issue:35707", "right": "issue:40444", "accept": false, - "reason": "Progressive generation with inputs_embeds/past_key_values is unrelated to Qwen2.5-VL finetuning with multiple images per prompt." + "reason": "Different workflows and models: generation with embeds/past kv vs multimodal finetuning with multiple images." }, { "left": "issue:43818", "right": "issue:45357", "accept": false, - "reason": "Video tower temporal attention/weight sharing is a different code path from incorrect visual encoder keys on save_pretrained." + "reason": "Both are multimodal model issues, but the concrete bugs are different (video tower weights/temporal attention vs save_pretrained visual keys)." }, { "left": "issue:43317", "right": "issue:43856", "accept": false, - "reason": "Device-map loading/offload failure is unrelated to Qwen3 MoE training memory efficiency." + "reason": "Device offload/loading bug vs MoE training memory usage; same broad area only." }, { "left": "issue:44462", "right": "issue:45812", "accept": false, - "reason": "Tokenizer.json being ignored and Granite token IDs being wrong are different tokenizer regressions with different symptoms and scope." + "reason": "Both are AutoTokenizer-related, but one ignores tokenizer.json while the other returns wrong Granite token IDs." }, { "left": "issue:43381", "right": "issue:45923", "accept": false, - "reason": "Gradient checkpointing in eval mode is not the same as a missing supports_gradient_checkpointing flag on a remote-code model." + "reason": "One is eval-mode checkpointing behavior; the other is a missing capability flag on a trust_remote_code model." }, { "left": "issue:43482", "right": "issue:45538", "accept": false, - "reason": "GGUF model loading failure and CLIPTokenizer model_max_length behavior are unrelated issues." + "reason": "Different tokenizer/model_max_length problems with no concrete shared failure mode." }, { "left": "issue:44291", "right": "issue:44589", "accept": false, - "reason": "Unexpected _is_hf_initialized init arg and missing Float8 storage registration are different v5 breakages." + "reason": "Both are init/load regressions, but one is an unexpected argument and the other is a storage type lookup failure." }, { "left": "issue:44829", "right": "issue:45479", "accept": false, - "reason": "flash_attention_3 training collapse and single-label num_labels=1 zero-loss are distinct classification/training bugs." + "reason": "Training collapse from flash attention vs degenerate zero loss in single-label classification; distinct bugs." }, { "left": "issue:43482", "right": "issue:43525", "accept": false, - "reason": "Qwen2.5-GGUF loading failure is unrelated to Llama4Config missing pad_token_id." + "reason": "Separate loading failures: GGUF compatibility vs a missing pad_token_id on Llama4Config." }, { "left": "issue:44079", "right": "issue:44534", "accept": false, - "reason": "ModelOutput key assignment and non-persistent buffer serialization are different internals and different symptoms." + "reason": "ModelOutput key assignment and non-persistent buffer initialization are unrelated code paths." }, { "left": "issue:43377", "right": "issue:45491", "accept": false, - "reason": "Missing padding-mask support in MIMI encoder is unrelated to Gemma3 NaNs from sliding-window/all-padding windows." + "reason": "Padding-mask batching discrepancy vs NaNs from mixed-length sliding-window attention; different failures." }, { "left": "issue:43531", "right": "issue:44291", "accept": false, - "reason": "Qwen3-MoE sliding_window behavior and init_empty_weights/_is_hf_initialized are separate regressions." + "reason": "Sliding-window MoE issue vs init_empty_weights argument regression; not the same bug." }, { "left": "issue:41720", "right": "issue:45081", "accept": false, - "reason": "Qwen3 device-mapping cuda assert and Mistral regex patch crash are entirely different failures." + "reason": "CUDA assert during device mapping is unrelated to the Mistral tokenizer regex crash." }, { "left": "issue:40444", "right": "issue:45381", "accept": false, - "reason": "Multiple-images-per-prompt finetuning and video vision_position_ids issues are both Qwen2.5-VL, but not the same concrete bug." + "reason": "Multi-image finetuning failure vs Qwen2.5-VL video position-id regression." }, { "left": "issue:43749", "right": "issue:44079", "accept": false, - "reason": "FSDP CPU RAM efficient loading and ModelOutput key assignment do not share a concrete code-path problem." + "reason": "FSDP CPU RAM-efficient loading and ModelOutput key handling are unrelated." }, { "left": "issue:44265", "right": "issue:45072", "accept": false, - "reason": "torch.export/torch_compilable_check failure is unrelated to dtype mismatches in SwitchTransformers/TimmWrapperModel." + "reason": "torch.export compatibility issue vs bfloat16 dtype mismatch; different concrete problems." }, { "left": "issue:43644", "right": "issue:44079", "accept": false, - "reason": "Non-persistent buffers being filled with junk is a different regression from ModelOutput key bookkeeping." + "reason": "Non-persistent buffer junk and ModelOutput key assignment are separate regressions." }, { "left": "issue:42175", "right": "issue:43296", "accept": false, - "reason": "Packaging TensorFlow into the torch extra and vLLM loading PaddleOCR-VL are unrelated." + "reason": "Packaging/backend dependency issue vs a specific model load failure in vLLM." }, { "left": "issue:43632", "right": "issue:45072", "accept": false, - "reason": "_is_hf_initialized flag handling and bfloat16 dtype mismatches are different failure modes." + "reason": "_is_hf_initialized regression and dtype mismatch are not the same underlying bug." }, { "left": "issue:45325", "right": "issue:45381", "accept": false, - "reason": "Both are Qwen2.5-VL position-id bugs, but one is still-image temporal scaling and the other is video-input vision_position_ids." + "reason": "Same model family and positional-id area, but still-image temporal scaling and video vision_position_ids look like distinct symptoms." }, { "left": "issue:43611", "right": "issue:43646", - "accept": false, - "reason": "base_model_prefix loading failure and generic custom-model initialization breakage overlap thematically, but the concrete bugs are not clearly the same." + "accept": true, + "reason": "Both describe the same Transformers v5 model-init/load regression; the broader issue covers the base_model_prefix-specific report." }, { "left": "issue:43792", "right": "issue:43957", "accept": false, - "reason": "Whisper-large-v2 runtime failure and meta-device loading breakage are distinct model-loading problems." + "reason": "Whisper runtime failure vs meta-device loading regression; unrelated." }, { "left": "issue:43994", "right": "issue:44898", "accept": false, - "reason": "Siglip2 nonsensical outputs and Perceiver non-default-resolution failure are different vision-model issues." + "reason": "Different vision-model failures with different repros and failure modes." }, { "left": "issue:45325", "right": "issue:45834", "accept": false, - "reason": "Qwen2.5-VL position-id scaling and Kosmos2.5 OCR index errors are unrelated even though both involve long-context vision inputs." + "reason": "Qwen2.5-VL positional-id bug vs Kosmos2.5 OCR index error; no shared code path." }, { "left": "issue:43454", "right": "issue:44466", "accept": false, - "reason": "Missing lm_head weight tying causing garbage text and device-dependent serialization of lm_head.weight are separate tied-weight problems." + "reason": "Tied-weight generation bug vs device-dependent serialization inconsistency; related theme, but not the same concrete bug." }, { "left": "issue:42915", "right": "issue:43482", "accept": false, - "reason": "Qwen3Moe FP8 configuration failure and Qwen2.5-GGUF loading failure are different model/setup combinations." + "reason": "Qwen3Moe FP8 failure and GGUF loading failure are unrelated." }, { "left": "issue:43296", "right": "issue:45800", "accept": false, - "reason": "PaddleOCR-VL loading in vLLM and a torch/transformers version incompatibility are not the same underlying defect." + "reason": "Different incompatibility reports with different stacks and failure surfaces." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The assessment is grounded in the packet and stays conservative. The only accepted soft edge, issue:43611 \u2194 issue:43646, is well supported by the titles and the high similarity score as the same Transformers v5 model-init/load regression. The rejected edges are also phrased cautiously and do not overclaim equivalence. The summary matches the evidence that the cluster is mostly heterogeneous with one clear duplicate-like subcluster." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -3515,196 +3725,196 @@ "issue:43881|issue:44991", "issue:44466|issue:45357" ], - "prepared_review_unit_hash": "0a9010904b02096dbf04380dbe27c503ac8fb39eb8753532b596f78fd251b665", + "prepared_review_unit_hash": "a949bc1f42f415472518c95ee6f1009aad0a03c99068433d3eb4327bb73d439f", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 37222, - "estimated_input_tokens": 9306, - "estimated_eval_tokens": 18868 + "serialized_chars": 37220, + "estimated_input_tokens": 9305, + "estimated_eval_tokens": 18866 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is a heterogeneous set of unrelated closed issues. None of the soft-similarity pairs look like the same underlying bug/change, so all candidate duplicates are rejected.", - "confidence": 0.91, - "canonical_issue_reason": "No canonical issue: the items span many unrelated subsystems and failure modes, so there is no single duplicate anchor.", + "summary": "These items are mostly a loose topical cluster around loading/offloading, quantization, tokenization, and multimodal model regressions, but the soft pairs do not look like true duplicates of the same concrete bug. I\u2019d treat this as a broad, mixed issue cluster rather than a duplicate set.", + "confidence": 0.67, + "canonical_issue_reason": "issue:43873 is the broadest, most central representative of the cluster because it is an open, discussion-heavy report about offloading + quantization, which is the most recurring theme among the nearby issues.", "canonical_pr_reason": null, - "best_issue_reason": "No globally best issue: none of the issues is a strong representative of the others, since the cluster is not a true duplicate group.", + "best_issue_reason": "issue:43873 is the best single representative issue because it captures the widest shared theme in this cluster (offloading/quantization) and is more general than the narrower model-specific failures.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44479", "right": "issue:45823", "accept": false, - "reason": "Different models and failure modes: video-input regression vs device-mismatch in Gemma4 PLE." + "reason": "Both are multimodal regressions, but they concern different models and different failure modes (video-input regression vs device mismatch in Gemma4). Not the same bug." }, { "left": "issue:43701", "right": "issue:45325", "accept": false, - "reason": "Checkpoint key mismatch and Qwen2.5-VL rope-index scaling are unrelated bugs." + "reason": "Resume-from-checkpoint key mismatch and Qwen2.5-VL rope-index scaling are unrelated code paths." }, { "left": "issue:44291", "right": "issue:44877", "accept": false, - "reason": "Both are loading/config issues, but one is an init_empty_weights arg mismatch and the other is strict config loading." + "reason": "`init_empty_weights` argument handling and strict config loading for granite_speech are different loading/config issues." }, { "left": "issue:43606", "right": "issue:44265", "accept": false, - "reason": "CPU offload device mismatch and torch.export/torch_compilable_check failures are different code paths." + "reason": "CPU-offload device mismatch in Bark and `torch.export` failure with `torch_compilable_check` are unrelated." }, { "left": "issue:43881", "right": "issue:44488", "accept": false, - "reason": "Both are model-load failures, but for different models and likely different root causes." + "reason": "Both are loading failures, but for different models and likely different root causes; too broad to merge." }, { "left": "issue:43873", "right": "issue:45701", "accept": false, - "reason": "Offloading/quantization trouble is unrelated to tokenization behavior changes." + "reason": "Quantization/offloading behavior and tokenizer-version changes are different problems." }, { "left": "issue:43262", "right": "issue:43746", "accept": false, - "reason": "Audio chat-template sampling-rate defaulting and PEFT checkpoint loading are separate problems." + "reason": "Audio chat-template sample-rate handling and GraniteSpeech PEFT checkpoint loading are unrelated." }, { "left": "issue:40990", "right": "issue:45584", "accept": false, - "reason": "High perplexity on a language model and Whisper empty-transcription generation failure are unrelated." + "reason": "High perplexity on a benchmark and Whisper empty-transcription generation failure are different bugs." }, { "left": "issue:34634", "right": "issue:35707", "accept": false, - "reason": "Bark voice_preset handling and generation with inputs_embeds/past_key_values do not share the same bug." + "reason": "Bark voice preset support and progressive generation with `inputs_embeds`/`past_key_values` are unrelated." }, { "left": "issue:44568", "right": "issue:45920", "accept": false, - "reason": "Both concern tokenization, but one is a single tokenizer special-token bug while the other is a broad wrong-token-ID regression." + "reason": "Both involve tokenization, but one is a single tokenizer special-token bug while the other is a broader regression across many models; not clearly the same concrete defect." }, { "left": "issue:35707", "right": "issue:41720", "accept": false, - "reason": "Progressive generation logic and auto device-mapping CUDA assert are different failures." + "reason": "Generation with `inputs_embeds`/cache and Qwen3 auto device-mapping CUDA asserts are unrelated." }, { "left": "issue:44186", "right": "issue:44265", "accept": false, - "reason": "Tokenizer padding/truncation crash and torch.export/compilable-check failure are not the same issue." + "reason": "Tokenizer padding/truncation crash and `torch.export` failures from `torch_compilable_check` are different subsystems." }, { "left": "issue:43606", "right": "issue:43819", "accept": false, - "reason": "Device-mismatch offload bug vs DAC from_latents/STE mismatch are unrelated." + "reason": "Bark CPU-offload device mismatch and DAC missing-STE behavior are unrelated model bugs." }, { "left": "issue:41720", "right": "issue:45823", "accept": false, - "reason": "Both mention device mapping, but they affect different models and different concrete failures." + "reason": "Both mention device mapping, but the models and observed failures differ enough that this is not the same bug." }, { "left": "issue:43454", "right": "issue:43550", "accept": false, - "reason": "Weight-tying serialization and torch.compile SDPA failure are separate problems." + "reason": "Weight tying in AyaVision and `torch.compile`/SDPA failure in Bamba are unrelated." }, { "left": "issue:43377", "right": "issue:44610", "accept": false, - "reason": "Padding-mask correctness in MIMI is unrelated to an OmDet-Turbo processor/model resolution mismatch." + "reason": "MIMI padding-mask inconsistency and OmDet-Turbo input-size mismatch are unrelated preprocessing/model issues." }, { "left": "issue:39692", "right": "issue:40444", "accept": false, - "reason": "Doc-example/model mismatch and multi-image iterable-dataset finetuning failure are different bugs." + "reason": "Documentation/example errors for SigLIP2 and Qwen2.5-VL dataset multi-image finetuning failure are different issues." }, { "left": "issue:43317", "right": "issue:43873", "accept": false, - "reason": "Both touch offloading/quantization, but one is a specific dequantized load failure and the other is a broad offloading complaint." + "reason": "Both relate to offloading and quantization, but one is about loading a dequantized model with `device_map=auto` and the other is a broader offloading regression; not enough evidence they are the same concrete bug." }, { "left": "issue:43454", "right": "issue:45072", "accept": false, - "reason": "Tied-weight serialization and bfloat16 dtype mismatch are different concrete issues." + "reason": "Garbage text from untied `lm_head` weights and dtype mismatches in bfloat16 inference are unrelated." }, { "left": "issue:43377", "right": "issue:44898", "accept": false, - "reason": "Missing padding-mask support and Perceiver positional-encoding handling are unrelated." + "reason": "Padding-mask output drift in MIMI and Perceiver non-default-resolution failure are different model-specific bugs." }, { "left": "issue:43257", "right": "issue:45923", "accept": false, - "reason": "Qwen3 MoE accelerate/deepspeed loading and missing gradient-checkpointing flag are different code changes." + "reason": "Qwen3 MoE weight conversion/loading and missing `supports_gradient_checkpointing` flag are unrelated." }, { "left": "issue:43572", "right": "issue:45759", "accept": false, - "reason": "Missing pad_token_idx and failing to unwrap text_config are separate configuration bugs." + "reason": "A missing config field in StableLM and `from_config` unwrapping for composite Qwen multimodal configs are different problems." }, { "left": "issue:40990", "right": "issue:42947", "accept": false, - "reason": "Perplexity regression and gradient-checkpointing ineffectiveness are not the same problem." + "reason": "Perplexity regression and gradient checkpointing with LoRA are unrelated." }, { "left": "issue:43322", "right": "issue:46050", "accept": false, - "reason": "Llava Next load segfault and sinq save/load regression are unrelated model-loading failures." + "reason": "Llava-next loading segfault and sinq save/load quantization breakage are different failure modes." }, { "left": "issue:43526", "right": "issue:43883", "accept": false, - "reason": "BeitImageProcessorFast label reduction and missing all_tied_weights_keys are different subsystems." + "reason": "BeitImageProcessorFast label reduction and Molmo tied-weights attribute error are unrelated." }, { "left": "issue:43454", "right": "issue:43720", "accept": false, - "reason": "Weight tying in a model head and BitNet packed-weight unpacking are unrelated." + "reason": "Weight tying serialization and packed-weight unpacking during accelerate loading are different bugs." }, { "left": "issue:41720", "right": "issue:43322", "accept": false, - "reason": "CUDA assert during auto device mapping and segmentation fault on Llava Next loading are different failures." + "reason": "Qwen3 auto device-map CUDA assert and Llava Next segfault are different model/load failures." }, { "left": "issue:43296", @@ -3716,19 +3926,19 @@ "left": "issue:43749", "right": "issue:44534", "accept": false, - "reason": "FSDP RAM-efficient loading and non-persistent buffer junk are not the same bug." + "reason": "FSDP CPU RAM-efficient loading and non-persistent buffer junk values are different v5 regressions." }, { "left": "issue:43881", "right": "issue:44991", "accept": false, - "reason": "Different models and different tokenizer/model-load failures; no shared code-path evidence." + "reason": "A model loading failure and a tokenizer-loading regression are both loading-related but not the same bug." }, { "left": "issue:44466", "right": "issue:45357", "accept": false, - "reason": "Both are serialization regressions, but they affect different objects and mechanisms, so they are not the same fix." + "reason": "Serialization of tied weights and incorrect visual-encoder keys in save_pretrained are different save/export problems." } ] }, @@ -3832,20 +4042,20 @@ "issue:43531|issue:43572", "issue:43316|issue:44913" ], - "prepared_review_unit_hash": "18a82cbed4ba91055d4aa3037bc91f50158cb34f38b4b7183309fb73c32e016a", + "prepared_review_unit_hash": "dde36e2275f8963edbe32d8ef0117a805df94b2b7c7e0c6d82a85323b590caf8", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 30, - "serialized_chars": 36283, + "serialized_chars": 36282, "estimated_input_tokens": 9071, "estimated_eval_tokens": 18398 }, @@ -3854,192 +4064,186 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is very heterogeneous: the soft pairs share broad language (Qwen, tokenization, offload, vision, dtype, etc.) but not the same concrete bug or change. I found no duplicate-level matches to merge, and there are no PRs here.", - "confidence": 0.8, - "canonical_issue_reason": "No issue is a clear canonical duplicate for the cluster. If one issue must represent it, issue:43873 is the closest fit because it is the broadest active bug with multiple inbound references around quantization/offloading, but it is not actually the same problem as the other items.", + "summary": "The cluster is heterogeneous and looks like a loose set of Transformers regressions around tokenization, model loading, offloading/quantization, and processor/model-specific bugs. None of the soft pairs appear to be the same underlying bug or change, so I would not merge any of them as duplicates.", + "confidence": 0.84, + "canonical_issue_reason": "issue:43873 is the best representative because it is the broadest open issue in the offloading/quantization family and has the strongest connectivity to several other memory/offload-related reports, even though the cluster is not a true duplicate set.", "canonical_pr_reason": null, - "best_issue_reason": "Issue:43873 is the best representative only in a loose sense: it is open, well-described, and has the most surrounding activity/inbound references. The rest of the items are mostly unrelated regressions or model-specific failures.", + "best_issue_reason": "issue:43873 is the best single issue to anchor this cluster: it is open, broad, and closest to the recurring offloading/quantization theme that shows up in several nearby reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:41720", "right": "issue:44977", "accept": false, - "reason": "Different failures: CUDA assert during auto device mapping vs flash-attention generation bug." + "reason": "Different failure modes: auto device-mapping cuda assert vs flash-attention generation behavior on Qwen3.5. Same family, not the same bug." }, { "left": "issue:42617", "right": "issue:42915", "accept": false, - "reason": "Unrelated: 3d_parallel.py execution failure vs Qwen3Moe FineGrainedFP8Config loading failure." + "reason": "Unrelated problems: running 3d_parallel.py versus FineGrainedFP8Config model failure. No shared concrete code path." }, { "left": "issue:44008", "right": "issue:45823", "accept": false, - "reason": "Different models and code paths: Gemma3n forward variable collision vs Gemma4 PLE device mismatch." + "reason": "Different Gemma models and different bugs: a variable-name collision in Gemma3n forward vs a Gemma4 PLE device mismatch." }, { "left": "issue:40990", "right": "issue:45701", "accept": false, - "reason": "Perplexity anomaly on one model vs a broad tokenization-version regression; not the same bug." + "reason": "One is an evaluation/perplexity anomaly, the other is a tokenization regression across versions. Not the same underlying issue." }, { "left": "issue:44387", "right": "issue:45704", "accept": false, - "reason": "Both are memory-related, but one is int4 quantization OOM and the other is a T5 RMSNorm leak." + "reason": "Different mechanisms and subsystems: int4 quantization memory growth vs an apex FusedRMSNorm memory leak in T5." }, { "left": "issue:43404", "right": "issue:45357", "accept": false, - "reason": "Different regression classes: untied lm_head in Mistral3 vs wrong visual encoder keys in Qwen3.5 save_pretrained." + "reason": "Different models and code paths: Mistral3 lm_head tying bug vs Qwen3.5 visual-encoder key saving regression." }, { "left": "issue:44186", "right": "issue:45072", "accept": false, - "reason": "Tokenizer crash vs dtype mismatch; no shared code-path bug." + "reason": "Tokenizer crash in LayoutLMv2 is unrelated to dtype mismatch issues in SwitchTransformers/TimmWrapperModel." }, { "left": "issue:36331", "right": "issue:42617", "accept": false, - "reason": "Trainer API kwarg mismatch is unrelated to a 3d_parallel.py runtime failure." + "reason": "Trainer compute_loss signature regression and 3d_parallel.py runtime failure are unrelated." }, { "left": "issue:43577", "right": "issue:43708", "accept": false, - "reason": "Different problems: BLIP2 dtype not propagating vs resume/max_steps calculation." + "reason": "BLIP2 dtype propagation and Trainer resume/max_steps calculation are separate bugs in different layers." }, { "left": "issue:44877", "right": "issue:44991", "accept": false, - "reason": "Strict config loading and tokenizer loading are separate issues." + "reason": "Both are loading regressions, but one is strict config validation and the other is tokenizer loading for a different model; not the same fix." }, { "left": "issue:43873", "right": "issue:44756", "accept": false, - "reason": "Both touch memory/offload, but one is quantization offloading behavior and the other is mmap/OOM on Strix Halo." + "reason": "Both mention offloading/memory, but one is quantization offloading behavior and the other is mmap-related OOM on Strix Halo; different root causes." }, { "left": "issue:36331", "right": "issue:42915", "accept": false, - "reason": "Trainer compute_loss API break is unrelated to Qwen3Moe FP8 config loading." + "reason": "Trainer loss API regression and Qwen3Moe FP8 loading failure do not share the same code path." }, { "left": "issue:43450", "right": "issue:45701", "accept": false, - "reason": "Video processor batching shape bug is unrelated to tokenization changes." + "reason": "Video processor batch-shape bug is unrelated to a tokenizer version regression." }, { "left": "issue:43550", "right": "issue:43720", "accept": false, - "reason": "torch.compile/SDPA failure and packed-weight loading failure are different code paths." + "reason": "Different models and failures: torch.compile/SDPA on Bamba vs packed-weight unpacking during accelerate loading for BitNet." }, { "left": "issue:44423", "right": "issue:45406", "accept": false, - "reason": "Both are serve crashes, but the failing attributes and processor code paths differ." + "reason": "Continuous-batching multimodal crash and Gemma4Processor missing _tokenizer are different processor/serve failures." }, { "left": "issue:45632", "right": "issue:45684", "accept": false, - "reason": "Cache-path collision and read-only permission propagation are unrelated filesystem behaviors." + "reason": "Cache-path collision for local models is unrelated to permissions propagation in save_pretrained." }, { "left": "issue:41720", "right": "issue:42617", "accept": false, - "reason": "Different subsystems and symptoms; no shared underlying bug." + "reason": "Qwen3 auto device-mapping cuda assert and 3d_parallel.py execution failure are not the same bug." }, { "left": "issue:44479", "right": "issue:44918", "accept": false, - "reason": "Video-input regression in Qwen models vs TRL SFT embedding unpacking failure." + "reason": "Video-input regression and trl SFT embedding unpacking are distinct bugs despite both involving multimodal/training workflows." }, { "left": "issue:44610", "right": "issue:44898", "accept": false, - "reason": "Both involve image sizing, but they affect different models and different failure mechanisms." + "reason": "Both are image-shape related, but one is an OmDet-Turbo processor resolution mismatch and the other is a Perceiver interpolation issue." }, { "left": "issue:43550", "right": "issue:43606", "accept": false, - "reason": "torch.compile SDPA issue is unrelated to CPU-offload device mismatch." + "reason": "Different models and different failures: Bamba torch.compile SDPA issue vs Bark CPU-offload device mismatch." }, { "left": "issue:44186", "right": "issue:44610", "accept": false, - "reason": "Tokenizer/NER crash is unrelated to OmDet-Turbo image-size mismatch." + "reason": "Tokenizer crash on NER inputs is not the same as an image-processor resolution mismatch." }, { "left": "issue:42915", "right": "issue:43317", "accept": false, - "reason": "FineGrainedFP8 loading failure and dequantized device_map offload failure are not the same bug." + "reason": "Both touch quantized/model loading, but one is Qwen3Moe FP8 config failure and the other is dequantized model offload loading. Not a single bug." }, { "left": "issue:43295", "right": "issue:45701", "accept": false, - "reason": "Processor.tokenizer regression is different from a generic tokenization-change report." - }, - { - "left": "issue:41720", - "right": "issue:45161", - "accept": false, - "reason": "Different parallelism/device-mapping failures on different model families." + "reason": "Custom processor/tokenizer access regression is different from the broader tokenization-version behavior change." }, { "left": "issue:44671", "right": "issue:44743", "accept": false, - "reason": "CamemBERT masked-LM predictions and Qwen recurrent-state caching are unrelated." + "reason": "CamemBERT masked-LM prediction regression and Qwen3.5 recurrent-state reset are unrelated." }, { "left": "issue:43526", "right": "issue:43881", "accept": false, - "reason": "BeitImageProcessorFast label reduction bug is unrelated to glm-4v-9b loading failure." + "reason": "BEiT label reduction bug and glm-4v-9b load failure are separate issues." }, { "left": "issue:36010", "right": "issue:42617", "accept": false, - "reason": "ImportError for GenerationMixin is unrelated to 3d_parallel.py." + "reason": "GenerationMixin import failure is unrelated to the 3d_parallel.py runtime problem." }, { "left": "issue:45081", "right": "issue:45812", "accept": false, - "reason": "Different tokenizer bugs: Mistral regex patch crash vs Granite wrong token IDs." + "reason": "Both are tokenizer-related, but one is a regex-patch crash during loading and the other is silent token-ID corruption in Granite. Different code paths and fixes." }, { "left": "issue:43531", "right": "issue:43572", "accept": false, - "reason": "Qwen3-MoE sliding_window behavior and StableLmConfig pad_token_idx are unrelated." + "reason": "Sliding-window behavior in Qwen3-MoE and missing pad_token_idx in StableLmConfig are different issues." }, { "left": "issue:43316", "right": "issue:44913", "accept": false, - "reason": "Different config fields and models; no shared underlying defect." + "reason": "Gemma3TextConfig API mismatch and GPTNeoX rotary_pct reload regression are unrelated config bugs." } ] }, @@ -4142,217 +4346,217 @@ "issue:43782|issue:45381", "issue:45759|issue:45859" ], - "prepared_review_unit_hash": "530937aaa4bd4e91d1d3be77b257b2e71ff07517f79e2d2291414ff4b3f529e6", + "prepared_review_unit_hash": "fd833ca7908358b643b58713e063661e220a3ad8d86db753b2eb043c1fedbd01", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 29, - "serialized_chars": 35966, - "estimated_input_tokens": 8992, - "estimated_eval_tokens": 18240 + "serialized_chars": 35961, + "estimated_input_tokens": 8991, + "estimated_eval_tokens": 18238 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly heterogeneous and does not look like one duplicate family. The only clear duplicate pair is the two Qwen2_5_VLProcessor.apply_chat_template crash reports (44514/44545).", - "confidence": 0.76, - "canonical_issue_reason": "issue:44514 is the earlier, more detailed report of the same Qwen2_5_VLProcessor.apply_chat_template batched padding=False crash as 44545, so it is the best duplicate target.", + "summary": "The cluster is mostly heterogeneous. The only clear duplicate is the Qwen2_5_VLProcessor batched-input crash pair (44514/44545). The EMBEDDIA/est-roberta and v5 tokenizer-load reports are only softly similar from this packet and should not be treated as a duplicate match.", + "confidence": 0.82, + "canonical_issue_reason": "44545 is the best canonical anchor because it is an exact match to 44514 on the same `Qwen2_5_VLProcessor.apply_chat_template` crash with `padding=False`, with slightly stronger reference signals.", "canonical_pr_reason": null, - "best_issue_reason": "44514 is the clearest true duplicate hub in this set; the rest are largely unrelated bugs, so it is the strongest representative issue overall.", + "best_issue_reason": "44545 is the strongest overall issue candidate because it is the clearest confirmed duplicate target in the set and has the strongest duplicate evidence.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43540", "right": "issue:43782", "accept": false, - "reason": "Both are Qwen-family issues, but one is video processing at runtime and the other is a from_pretrained weight_only load error." + "reason": "Related Qwen multimodal loading failures, but the concrete code paths and symptoms differ." }, { "left": "issue:42915", "right": "issue:43322", "accept": false, - "reason": "Different models and failure modes: FP8 config handling vs a Llava Next segfault on load." + "reason": "Different models and failure modes: FP8 config loading vs Llava Next segfault." }, { "left": "issue:43421", "right": "issue:45812", "accept": false, - "reason": "Both concern tokenization, but one is runtime post-processor refresh and the other is a Granite token-ID regression." + "reason": "Both are tokenizer-related, but one is runtime post-processor behavior and the other is a token-id regression for Granite models." }, { "left": "issue:44186", "right": "issue:44361", "accept": false, - "reason": "Different tokenizer-specific crashes on different models and code paths." + "reason": "Both involve tokenizers, but they affect different tokenizers and different crash conditions." }, { "left": "issue:43388", "right": "issue:45059", "accept": false, - "reason": "Metric gathering label truncation and SAM3 text/box behavior are unrelated subsystems." + "reason": "Unrelated bugs: metrics label gathering vs SAM3 text/bounding-box behavior." }, { "left": "issue:29159", "right": "issue:36032", "accept": false, - "reason": "One is slow/fast tokenizer inconsistency; the other is a T5 tokenizer load-time attribute conflict." + "reason": "Both mention tokenizers, but the first is slow/fast inconsistency and the second is a load-time name collision." }, { "left": "issue:45406", "right": "issue:45464", "accept": false, - "reason": "Processor attribute initialization vs streaming chat/completions failure are not the same bug." + "reason": "Different products and failure modes: serve processor attribute error vs streaming inference failure." }, { "left": "issue:30064", "right": "issue:34634", "accept": false, - "reason": "Image segmentation-map processing and Bark voice_preset handling are unrelated processor issues." + "reason": "Different subsystems and bugs: image segmentation-map processing vs Bark voice preset handling." }, { "left": "issue:33357", "right": "issue:35707", "accept": false, - "reason": "MacOS CLIP bus error and progressive generation with inputs_embeds/past_key_values are different problems." + "reason": "MacOS CLIP bus error and generation with `inputs_embeds`/`past_key_values` are unrelated." }, { "left": "issue:41720", "right": "issue:43746", "accept": false, - "reason": "Auto device mapping cuda assert and PEFT local checkpoint loading are different code paths." + "reason": "Different model-specific failures: auto device mapping assert vs local PEFT checkpoint loading." }, { "left": "issue:43606", "right": "issue:43720", "accept": false, - "reason": "CPU offload device mismatch and packed-weight unpacking during accelerate loading are distinct issues." + "reason": "Both touch loading/offload, but the concrete bugs are device mismatch vs packed-weight unpacking." }, { "left": "issue:43550", "right": "issue:43653", "accept": false, - "reason": "torch.compile/SDPA failure is unrelated to BigBirdTokenizer special-token registration." + "reason": "Different areas entirely: torch.compile SDPA failure vs tokenizer special-token registration." }, { "left": "issue:43421", "right": "issue:45701", "accept": false, - "reason": "Both touch tokenization, but runtime special-token updates are not the same as a version-wide tokenization regression." + "reason": "A generic tokenization-change report is not the same as the runtime post-processor update bug." }, { "left": "issue:43873", "right": "issue:45704", "accept": false, - "reason": "Quantization/offloading behavior and T5 RMSNorm memory leak are unrelated." + "reason": "Quantization/offloading behavior and a T5 RMSNorm memory leak are distinct issues." }, { "left": "issue:44492", "right": "issue:44509", "accept": false, - "reason": "A cache-strategy typo and removed-pipeline docs are not the same underlying change." + "reason": "A cache-strategy typo and docs mentioning removed pipelines are unrelated documentation issues." }, { "left": "issue:43606", "right": "issue:43653", "accept": false, - "reason": "Different bugs: Bark CPU offload device mismatch vs tokenizer special-token registration." + "reason": "Bark CPU-offload device mismatch is unrelated to BigBirdTokenizer mask-token registration." }, { "left": "issue:34634", "right": "issue:41720", "accept": false, - "reason": "BarkProcessor voice_preset and Qwen3 auto device mapping failures are unrelated." + "reason": "Different models and bugs: BarkProcessor voice preset vs Qwen3 auto device mapping assert." }, { "left": "issue:42490", "right": "issue:43329", "accept": false, - "reason": "A local-path import warning and an undefined video-token helper bug are not the same issue." + "reason": "Local-path import warning is unrelated to multimodal video token counting." }, { "left": "issue:43531", "right": "issue:45759", "accept": false, - "reason": "Both are Qwen3-related, but one is sliding_window behavior and the other is composite config unwrapping." + "reason": "Both mention Qwen configs, but one is sliding-window behavior and the other is config unwrapping in `from_config`." }, { "left": "issue:34567", "right": "issue:35707", "accept": false, - "reason": "TrainerState token counting and progressive generation are unrelated." + "reason": "Trainer token counting and progressive generation are unrelated." }, { "left": "issue:44514", "right": "issue:44545", "accept": true, - "reason": "Same Qwen2_5_VLProcessor.apply_chat_template crash on batched input with padding=False; titles and failure mode match exactly." + "reason": "Same underlying bug and effectively the same report: identical Qwen2_5_VLProcessor batched-input crash with `padding=False`." }, { "left": "issue:43421", "right": "issue:43927", "accept": false, - "reason": "Runtime special-token updates and DiaConfig save/load token-ID loss are different bugs." + "reason": "Runtime special-token/post-processor handling is not the same as DiaConfig losing custom token IDs on save/load." }, { "left": "issue:44625", "right": "issue:45479", "accept": false, - "reason": "Qwen3.5 num_labels propagation and a zero-loss classification regression are different sequence-classification issues." + "reason": "Both involve classification configs, but one is Qwen3.5 `num_labels` propagation and the other is a broader loss-behavior bug." }, { "left": "issue:44898", "right": "issue:45072", "accept": false, - "reason": "Perceiver non-default resolution failure and dtype mismatches in other models are unrelated." + "reason": "Different model families and failure modes: Perceiver image resolution handling vs bfloat16 dtype mismatch." }, { "left": "issue:44451", "right": "issue:44991", "accept": false, - "reason": "Both are tokenizer-loading regressions, but they affect different models and likely different root causes." + "reason": "Both are tokenizer-load failures, but the packet only shows different models and different concrete failures; that is not enough to conclude the same regression." }, { "left": "issue:43531", "right": "issue:44589", "accept": false, - "reason": "Sliding-window Qwen3-MoE behavior and Float8 storage lookup failure are unrelated." + "reason": "Qwen3-MoE sliding-window behavior and a Float8 storage lookup error are unrelated." }, { "left": "issue:44077", "right": "issue:44625", "accept": false, - "reason": "PatchTSMixer post_init validation and Qwen3.5 num_labels propagation are different configuration problems." + "reason": "PatchTSMixer post-init validation is unrelated to Qwen3.5 label propagation." }, { "left": "issue:43782", "right": "issue:45381", "accept": false, - "reason": "A from_pretrained weight_only load error and a video vision_position_ids bug are different Qwen-VL paths." + "reason": "Both are Qwen vision-related, but one is a `from_pretrained` weight-only load error and the other is a video-position-id correctness bug." }, { "left": "issue:45759", "right": "issue:45859", "accept": false, - "reason": "Composite config unwrapping and CPU-offload rotary embedding incompatibility are not the same bug." + "reason": "Config unwrapping for multimodal models and CPU-offload rotary-embedding compatibility are different bugs." } ] }, "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet, and the only accepted soft edge (44514/44545) is well supported by the near-exact title match and matching failure mode. The remaining soft-edge rejections are conservative and avoid overclaiming duplicates." + "accept": false, + "feedback": "The 44514/44545 duplicate call is well-grounded, but the 44451/44991 pair is overstated. From the packet we only have two different tokenizer-load failures on different models; that is not enough to conclude the same underlying regression. The summary should be more conservative there, e.g. treat it as a soft similarity rather than a clear duplicate." }, "evaluator_used": true, - "retried": false, + "retried": true, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -4451,20 +4655,20 @@ "issue:43618|issue:43697", "issue:45183|issue:45479" ], - "prepared_review_unit_hash": "3f8311f7399a42b277824c701c0540023bc63a268e397155367b7f2fb830c6c1", + "prepared_review_unit_hash": "3716efbfc479daffc4a880ce53dd45b88411f72025544cc3b53434f1628b463e", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 37219, + "serialized_chars": 37217, "estimated_input_tokens": 9305, "estimated_eval_tokens": 18866 }, @@ -4473,202 +4677,199 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous and does not look like one duplicate set. The only strong duplicate-like pair is the offline tokenizer/cache regression around `_patch_mistral_regex` and HF_HUB_OFFLINE.", - "confidence": 0.59, - "canonical_issue_reason": "issue:44843 is the clearest root-cause report: it explicitly identifies `_patch_mistral_regex` calling `model_info()` unconditionally and breaking offline tokenizer loading.", + "summary": "This cluster is not a duplicate set: the items span many unrelated bugs/features across tokenization, multimodal models, training, offloading, and documentation. No single issue or PR is a good canonical representative.", + "confidence": 0.98, + "canonical_issue_reason": "No canonical issue: the artifacts describe distinct failures in different subsystems and models, not one underlying bug.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44843 is the most actionable and specific issue in the cluster, with a concrete code path and symptom match to its near-duplicate.", + "best_issue_reason": "No issue in this set is broadly representative of the others; choosing one would be arbitrary and misleading.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44987", "right": "issue:45042", "accept": false, - "reason": "Different failures in different areas: model loading/version compatibility vs PIL image processor torchvision dependency." + "reason": "One is a physical-intelligence model loading failure; the other is a PIL backend image-processor torchvision dependency regression. Different code paths and symptoms." }, { "left": "issue:44368", "right": "issue:45357", "accept": false, - "reason": "Both mention Qwen3.5, but one is a tie-word-embeddings warning and the other is incorrect visual-encoder keys on save_pretrained; different bugs." + "reason": "Qwen3.5 warning about tie_word_embeddings is unrelated to Qwen3.5 save_pretrained writing incorrect visual encoder keys." }, { "left": "issue:44062", "right": "issue:44451", "accept": false, - "reason": "Tokenizer constructor argument error vs inability to load a specific model; no shared code-path bug is evident." + "reason": "Different tokenizer bugs: AddedToken special-flag argument collision vs inability to load a specific model. Not the same underlying defect." }, { "left": "issue:45588", "right": "issue:45759", "accept": false, - "reason": "Flash-attention sink handling and composite Qwen config unwrapping are unrelated problems." + "reason": "Flash-attention sink-less model crash is unrelated to AutoModelForCausalLM.from_config not unwrapping text_config for multimodal Qwen configs." }, { "left": "issue:45704", "right": "issue:45923", "accept": false, - "reason": "A T5 RMSNorm memory leak and a missing gradient-checkpointing flag are separate issues." + "reason": "T5 apex.FusedRMSNorm memory leak and missing supports_gradient_checkpointing on a trust_remote_code model are unrelated." }, { "left": "issue:39692", "right": "issue:44610", "accept": false, - "reason": "Documentation/example errors for SigLIP2 do not match an OmDet-Turbo processor size mismatch." + "reason": "SigLIP2 example/model-processor mismatch and quantization errors are unrelated to OmDet-Turbo processor output size mismatch." }, { "left": "issue:43746", "right": "issue:46050", "accept": false, - "reason": "PEFT adapter local-checkpoint loading and sinq quantized save/load breakage are different code paths." + "reason": "PEFT adapter local-checkpoint loading failure is unrelated to sinq quantized model save/load breakage." }, { "left": "issue:43421", "right": "issue:43653", "accept": false, - "reason": "Runtime special-token post-processing update and BigBird mask-token registration are related only broadly, not the same bug." + "reason": "Runtime post-processor update for special tokens and BigBirdTokenizer mask-token registration are separate tokenizer behaviors, not one bug." }, { "left": "issue:44918", "right": "issue:44936", "accept": false, - "reason": "TRl SFT input unpacking failure and trainer evaluate-after-train failure are distinct trainer behaviors." + "reason": "TRL SFT input-embedding unpacking and trainer.evaluate() after trainer.train() are different trainer workflows with different failure modes." }, { "left": "issue:45663", "right": "issue:45923", "accept": false, - "reason": "Gemma-4 FSDP2 shared_kv_states rebuilding and missing gradient-checkpointing support are unrelated." + "reason": "Gemma-4 FSDP2 shared_kv_states KeyError is unrelated to missing gradient-checkpointing support on a different model." }, { "left": "issue:42886", "right": "issue:44843", - "accept": true, - "reason": "These both describe offline tokenizer loading failing because the Mistral regex patch path triggers network/model-info access instead of using cache." + "accept": false, + "reason": "Offline tokenizer cache loading failure is distinct from unconditional model_info() calls in _patch_mistral_regex." }, { "left": "issue:43381", "right": "issue:43844", "accept": false, - "reason": "Gradient checkpointing in eval mode and abnormal gradient growth under ZeRO-3 are different training issues." + "reason": "Gradient checkpointing failing in eval mode is not the same as abnormal gradient increase with HfDeepSpeedConfig + ZeRO-3." }, { "left": "issue:43421", "right": "issue:44186", "accept": false, - "reason": "Both are tokenizer-related, but one is a runtime post-processor update feature gap and the other is a LayoutLMv2 NER padding/truncation crash." + "reason": "TokenizerBackend special-token post-processor update and LayoutLMv2Tokenizer NER/batched padding crashes are different tokenizer bugs." }, { "left": "issue:43531", "right": "issue:43645", "accept": false, - "reason": "Qwen3-MoE sliding-window behavior and Jupyter custom-model initialization are unrelated." + "reason": "Qwen3-MoE sliding_window issue and Jupyter custom-model initialization breakage do not share a concrete code path." }, { "left": "issue:45704", "right": "issue:45859", "accept": false, - "reason": "T5 RMSNorm memory leak vs Qwen3.5 MoE CPU offload incompatibility are separate bugs." + "reason": "T5 Apex RMSNorm leak and Qwen3.5MoETextRotaryEmbedding CPU-offload incompatibility are unrelated implementation issues." }, { "left": "issue:41720", "right": "issue:45704", "accept": false, - "reason": "CUDA assert during Qwen3 auto device mapping is not the same as the T5 RMSNorm memory leak." + "reason": "Qwen3 auto device mapping cudaErrorAssert and T5 apex RMSNorm memory leak are unrelated failures." }, { "left": "issue:43994", "right": "issue:44671", "accept": false, - "reason": "Different model regressions with different symptoms and code paths." + "reason": "SigLIP2 nonsensical outputs with AutoModel/pipeline and CamemBERT masked LM regression are separate model-specific bugs." }, { "left": "issue:43596", "right": "issue:44220", "accept": false, - "reason": "DeepSpeed ZeRO-3/BertModel indexing failure is unrelated to audio feature extraction." + "reason": "DeepSpeed ZeRO-3 init with BertModel index error is unrelated to _torch_extract_fbank_features()." }, { "left": "issue:43746", "right": "issue:43873", "accept": false, - "reason": "Local checkpoint loading with adapters and quantization/offloading behavior are not the same underlying bug." + "reason": "PEFT adapter loading from local checkpoints and quantization/offloading behavior are different problems, not one mergeable fix." }, { "left": "issue:39692", "right": "issue:43873", "accept": false, - "reason": "SigLIP2 docs/model mismatch and quantization offloading are unrelated." + "reason": "SigLIP2 docs/model example errors and quantization offloading behavior are too broad and unrelated to be the same bug." }, { "left": "issue:45081", "right": "issue:45356", "accept": false, - "reason": "Both involve tokenizer regex fixes, but one is a crash in `_patch_mistral_regex` and the other is a Kimi tokenizer regression with codec handling; not clearly the same bug." + "reason": "Mistral regex AttributeError when fixing tokenizer regex and Kimi-K2.5 codec/regression warning are different tokenizer paths." }, { "left": "issue:40444", "right": "issue:43746", "accept": false, - "reason": "Multimodal finetuning with iterable datasets and PEFT checkpoint loading are unrelated." + "reason": "IterableDataset multi-image Qwen2.5-VL finetuning failure is unrelated to GraniteSpeech PEFT local checkpoint loading." }, { "left": "issue:45356", "right": "issue:45907", "accept": false, - "reason": "Tokenizer codec/regex regression and `list_repo_files` revision forwarding are unrelated." + "reason": "Kimi tokenizer codec/regression issue and list_repo_files failing to forward revision are unrelated areas." }, { "left": "issue:30064", "right": "issue:36010", "accept": false, - "reason": "Image processor segmentation-map handling and a GenerationMixin import error have no common code path." + "reason": "Image processor void segmentation map failure and GenerationMixin import error are completely unrelated." }, { "left": "issue:43720", "right": "issue:43819", "accept": false, - "reason": "Packed-weight loading for BitNet and DAC latent conversion mismatch are different bugs." + "reason": "BitNet packed-weight unpacking during accelerate loading and DAC.from_latents forward mismatch are different code paths and defects." }, { "left": "issue:44220", "right": "issue:44661", "accept": false, - "reason": "Audio fbank extraction and tokenizer-mapping model registration are unrelated." + "reason": "Audio fbank feature extraction issue and add-new-model-like TOKENIZER_MAPPING_NAMES failure are unrelated." }, { "left": "issue:43653", "right": "issue:43720", "accept": false, - "reason": "BigBird special-token decode behavior and BitNet accelerate-loading unpacking are not the same issue." + "reason": "BigBirdTokenizer special-token registration and BitNet packed-weight loading are unrelated; shared theme is only model loading/tokenization broadly." }, { "left": "issue:43540", "right": "issue:44008", "accept": false, - "reason": "Video-input processing in Qwen3OmniMoe and Gemma3n audio-mask variable collision are separate multimodal bugs." + "reason": "Qwen3OmniMoe video-input ValueError and Gemma 3n audio_mel_mask variable collision are different multimodal model bugs." }, { "left": "issue:43618", "right": "issue:43697", "accept": false, - "reason": "CLIP attentions assignment regression and RTDetrV2 output differences are distinct model regressions." + "reason": "CLIPOutput attentions assignment regression and RTDetrV2 output differences in v5 are not the same underlying issue." }, { "left": "issue:45183", "right": "issue:45479", "accept": false, - "reason": "Input validation messaging and degenerate single-label classification loss are different problems." + "reason": "XOR validation error messaging for input_ids/inputs_embeds and single-label num_labels=1 zero-loss behavior are unrelated classification/model-input bugs." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet: the only clearly matching pair is 42886/44843 around offline tokenizer loading and `_patch_mistral_regex`/`HF_HUB_OFFLINE`, and the rest of the soft-edge pairs are rejected conservatively based on distinct titles/symptoms. No pair is overstated as a duplicate beyond what the issue text supports." - }, - "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 }, @@ -4763,196 +4964,196 @@ "issue:43644|issue:43749", "issue:44561|issue:45020" ], - "prepared_review_unit_hash": "8012a91bb8a394b45190d3f855636848a5de9bb396491ad4bb2f40e1ac9e83b0", + "prepared_review_unit_hash": "64b9ac42c5f7521af769ecd4e8d58ef9131d34af2218bfc71b80548826a966bb", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 27, - "serialized_chars": 36046, - "estimated_input_tokens": 9012, - "estimated_eval_tokens": 18280 + "serialized_chars": 36044, + "estimated_input_tokens": 9011, + "estimated_eval_tokens": 18278 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly false-positive similarity matches: the issues span unrelated model, tokenizer, config, cache, and multimodal regressions rather than one duplicate bug cluster.", - "confidence": 0.86, - "canonical_issue_reason": "If forced to pick an anchor, issue 45020 is the broadest umbrella ('recent transformers versions break models using remote_code'), but the set is too heterogeneous for it to be a true canonical duplicate.", + "summary": "Loose cluster around Transformers loading/tokenizer/config regressions, but the items mostly describe different model-specific failures or different code paths. None of the soft pairs look like true duplicates or the same fixable change.", + "confidence": 0.74, + "canonical_issue_reason": "issue:43643 is the best anchor: it\u2019s a concrete, reproducible loading regression in a core API path (`AutoConfig.from_pretrained` + `trust_remote_code`) and sits near several other loading/config-related reports.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45020 is the least specific and most representative of the broad regression theme, but no issue cleanly captures the whole cluster as one underlying defect.", + "best_issue_reason": "issue:43643 is the strongest global representative because it is specific enough to be actionable yet broad enough to cover a family of nearby loading regressions without being model-specific.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43550", "right": "issue:45072", "accept": false, - "reason": "Different failures: SDPA/torch.compile on Bamba vs dtype mismatches in SwitchTransformers/TimmWrapperModel." + "reason": "Both mention Bamba/SDPA-style failures, but one is a `torch.compile` + SDPA bug and the other is a bf16 dtype mismatch; different failure modes and fixes." }, { "left": "issue:43550", "right": "issue:43819", "accept": false, - "reason": "Different code paths: Bamba SDPA compile issue vs DAC latent/STE forward mismatch." + "reason": "`torch.compile`/SDPA on Bamba is unrelated to DAC `from_latents` vs forward-pass mismatch; different model families and code paths." }, { "left": "issue:43388", "right": "issue:46032", "accept": false, - "reason": "One is metrics label truncation; the other is Mamba2 cache correctness with seq_len > 1." + "reason": "Both are cache/batch-behavior bugs, but one is metrics label truncation and the other is Mamba2 `use_cache` corruption; not the same underlying bug." }, { "left": "issue:43716", "right": "issue:44008", "accept": false, - "reason": "Different multimodal bugs: image preprocessor dtype mismatch vs Gemma3n variable-name collision in forward()." + "reason": "Both are dtype/model-input mismatches, but one is Mistral image preprocessor vs model dtype and the other is a Gemma3n variable-name collision; unrelated fixes." }, { "left": "issue:43295", "right": "issue:45834", "accept": false, - "reason": "Custom model processor/tokenizer regression vs Kosmos2.5 long-OCR index error; not the same bug." + "reason": "Both involve loading/processing regressions, but one is a processor/tokenizer API break and the other is a Kosmos2.5 OCR index error; no shared code-path." }, { "left": "issue:44062", "right": "issue:44488", "accept": false, - "reason": "Tokenizer constructor argument collision vs model loading failure for a specific tokenizer." + "reason": "Both are tokenizer-load failures, but one is an `AddedToken` construction error and the other is a model loading failure for `cjvt/sleng-bert`; different root causes." }, { "left": "issue:43322", "right": "issue:45685", "accept": false, - "reason": "Segfault on Llava Next loading is unrelated to the MPS histogram int-typed operator error." + "reason": "Segmentation fault loading Llava Next is unrelated to the MPS histogram dtype error in moe; different subsystems and bug classes." }, { "left": "issue:43335", "right": "issue:43756", "accept": false, - "reason": "Single-layer sparse config bug vs Smollm3 RoPE-layer dropping; different model setup defects." + "reason": "SwitchTransformers sparse-layer construction and Smollm3 RoPE-layer omission are both model-architecture bugs, but they are not the same defect or fix." }, { "left": "issue:44186", "right": "issue:44898", "accept": false, - "reason": "Tokenizer padding/truncation crash vs Perceiver image-size/interpolation issue." + "reason": "LayoutLMv2 tokenizer NER/padding crash and Perceiver interpolation failure are distinct tokenizer vs vision-model issues." }, { "left": "issue:43950", "right": "issue:45987", "accept": false, - "reason": "Non-persistent buffer corruption is unrelated to StaticCache.get_seq_length() returning a tensor." + "reason": "Both touch cache/buffer contracts, but one is a `from_pretrained` non-persistent-buffer corruption regression and the other is `StaticCache.get_seq_length()` returning the wrong type; not duplicates." }, { "left": "issue:43653", "right": "issue:43819", "accept": false, - "reason": "BigBirdTokenizer special-token registration bug vs DAC from_latents/forward mismatch." + "reason": "BigBirdTokenizer special-token registration and DAC `from_latents` mismatch are unrelated tokenizer vs audio-model bugs." }, { "left": "issue:43531", "right": "issue:43643", "accept": false, - "reason": "Qwen3-MoE sliding_window problem and AutoConfig trust_remote_code missing-fields regression are different defects." + "reason": "Qwen3-MoE sliding_window behavior and `AutoConfig.from_pretrained` missing fields are different model/config issues with no common fix." }, { "left": "issue:44843", "right": "issue:45081", "accept": false, - "reason": "Both involve _patch_mistral_regex, but one is an offline hub call and the other is a backend_tokenizer AttributeError; not the same concrete bug." + "reason": "Offline-mode `model_info()` call in `_patch_mistral_regex` and the `backend_tokenizer` AttributeError are both Mistral-related but distinct failures at different points." }, { "left": "issue:43643", "right": "issue:44877", "accept": false, - "reason": "Missing fields from trust_remote_code AutoConfig and strict-config rejection of granite_speech are distinct config-loading issues." + "reason": "Both concern config/loading, but one is missing fields from `trust_remote_code` configs and the other is strict config blocking `granite_speech`; different symptoms and likely patches." }, { "left": "issue:43540", "right": "issue:45381", "accept": false, - "reason": "Video input processing ValueError in Qwen3OmniMoe is unrelated to qwen2.5-vl vision_position_ids being wrong." + "reason": "Qwen3OmniMoe video-input validation and qwen2.5-vl video vision-position ids are separate multimodal bugs." }, { "left": "issue:30064", "right": "issue:36331", "accept": false, - "reason": "ImageProcessor void segmentation maps and CustomTrainer.compute_loss signature mismatch are unrelated." + "reason": "Image segmentation-map processing and `CustomTrainer.compute_loss()` keyword mismatch are unrelated." }, { "left": "issue:44448", "right": "issue:44671", "accept": false, - "reason": "Different model regressions: Pegasus v4/v5 output drift vs CamemBERT masked-LM prediction errors." + "reason": "Pegasus v4/v5 output difference and CamemBERT masked-LM predictions are different model behavior regressions, not the same bug." }, { "left": "issue:43577", "right": "issue:43818", "accept": false, - "reason": "Blip2 dtype loading bug is unrelated to Video-LLaVA missing temporal attention / weight sharing." + "reason": "Blip2 dtype handling and Video-LLaVA temporal-attention/weight-sharing issues are unrelated model-loading defects." }, { "left": "issue:43421", "right": "issue:44811", "accept": false, - "reason": "Post-processor not updating special tokens and Whisper batch_decode skip_special_tokens behavior are different tokenizer issues." + "reason": "TokenizerBackend post-processor update on special-token changes and Whisper `batch_decode` skipping special tokens are different tokenizer behaviors." }, { "left": "issue:43782", "right": "issue:44977", "accept": false, - "reason": "Qwen3VL weight_only load failure vs Qwen3.5 flash-attention generation issue; different code paths." + "reason": "Qwen3VL `weight_only=True` load failure and Qwen3.5 flash-attention generation issue are separate loading vs inference-path bugs." }, { "left": "issue:41720", "right": "issue:44991", "accept": false, - "reason": "Qwen3 auto device-map assert on A800 is unrelated to loading EMBEDDIA/est-roberta tokenizer." + "reason": "Qwen3 auto device mapping CUDA assert and tokenizer loading for EMBEDDIA/est-roberta are unrelated failures." }, { "left": "issue:44821", "right": "issue:45084", "accept": false, - "reason": "AutoImageProcessor URL loading and template-node compilation failure are unrelated." + "reason": "AutoImageProcessor URL loading and template-node compilation errors are unrelated API failures." }, { "left": "issue:39692", "right": "issue:45834", "accept": false, - "reason": "SigLIP2 docs/example errors and Kosmos2.5 OCR index error are not the same underlying bug." + "reason": "SigLIP2 doc-example errors and Kosmos2.5 OCR index error are not the same bug or change." }, { "left": "issue:45584", "right": "issue:45834", "accept": false, - "reason": "Whisper empty-transcription generation after align_special_tokens is unrelated to Kosmos2.5 long OCR input indexing." + "reason": "Whisper empty-transcription generation and Kosmos2.5 long-OCR index error are different multimodal inference bugs." }, { "left": "issue:44186", "right": "issue:44442", "accept": false, - "reason": "LayoutLMv2 NER/padding crash vs AutoTokenizer failing to load FastSpeech2ConformerTokenizer." + "reason": "LayoutLMv2 tokenizer NER crash and FastSpeech2ConformerTokenizer auto-loading failure are both tokenizer-related but not the same concrete problem." }, { "left": "issue:43644", "right": "issue:43749", "accept": false, - "reason": "Non-persistent buffer junk in from_pretrained and FSDP_CPU_RAM_EFFICIENT_LOADING are different regressions." + "reason": "Non-persistent buffer corruption and FSDP CPU RAM-efficient loading are both loading-related, but they affect different mechanisms and fixes." }, { "left": "issue:44561", "right": "issue:45020", "accept": false, - "reason": "Both mention remote_code breakage, but one is the specific is_torch_fx_available removal and the other is a broad umbrella report; not one concrete duplicate bug." + "reason": "Both are broad remote-code breakages, but one is `is_torch_fx_available` removal and the other is a general recent-version complaint; too vague to treat as the same underlying bug." } ] }, @@ -5053,20 +5254,20 @@ "issue:43742|issue:45020", "issue:43329|issue:45600" ], - "prepared_review_unit_hash": "34b05f6ec57d0b4ddef7e4549a9d68b592669a45077820a4061e02698be8265b", + "prepared_review_unit_hash": "9d801691627dc0d817433e5eac028c5a2783eb613d93fbb39754c2c6f0f7eb5e", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 27, - "serialized_chars": 35509, + "serialized_chars": 35511, "estimated_input_tokens": 8878, "estimated_eval_tokens": 18012 }, @@ -5075,184 +5276,181 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These items are mostly unrelated issues across different models/components. Only one soft pair looks like a true duplicate: the tokenizer added-token spacing/inconsistency reports. No PRs were present.", - "confidence": 0.73, - "canonical_issue_reason": "issue:28218 is the clearest concrete report of the tokenizer added-token spacing bug; issue:29159 appears to describe the same behavior more broadly as a slow/fast tokenizer inconsistency.", + "summary": "This cluster is not a true duplicate set: it spans many unrelated tokenizer, model-loading, processor, and training bugs. A few pairs are superficially similar, but none look like the same concrete underlying defect or a single PR-worthy fix.", + "confidence": 0.93, + "canonical_issue_reason": "No strong canonical issue exists because the items are heterogeneous. If forced, issue 45020 is the broadest umbrella report, but it is still not a duplicate target for the rest of the cluster.", "canonical_pr_reason": null, - "best_issue_reason": "issue:28218 is the best canonical issue candidate because it states the bug precisely and is the older, more specific report of the same tokenizer problem.", + "best_issue_reason": "Issue 45020 is the most globally representative only in the sense that it is broad and user-facing, but it does not capture the other reports as one bug, so it is a weak canonical choice.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43606", "right": "issue:45072", "accept": false, - "reason": "Different bugs: CPU offload device mismatch for Bark vs bfloat16 dtype mismatch in SwitchTransformers/TimmWrapperModel." + "reason": "Both mention inference mismatches, but they affect different models and likely different code paths/root causes." }, { "left": "issue:44479", "right": "issue:45834", "accept": false, - "reason": "Different model families and failures: v5 video regression in Qwen models vs long-OCR index error in Kosmos2.5." + "reason": "One is a video regression in Qwen VL models; the other is a Kosmos2.5 OCR length/index error. Different bugs." }, { "left": "issue:43577", "right": "issue:43582", "accept": false, - "reason": "Unrelated symptoms and code paths: Blip2 dtype loading issue vs Apple Silicon caching_allocator_warmup TypeError." + "reason": "Blip2 dtype loading and Apple Silicon allocator TypeError are unrelated." }, { "left": "issue:43825", "right": "issue:43994", "accept": false, - "reason": "One is a pipeline v5 error-message issue; the other is a SigLIP2 bad output issue with AutoModel/pipeline, not the same underlying bug." + "reason": "A pipeline error message about removed translation support is unrelated to SigLIP2 producing bad outputs." }, { "left": "issue:34634", "right": "issue:36010", "accept": false, - "reason": "Different components: BarkProcessor voice_preset handling vs GenerationMixin import error." + "reason": "BarkProcessor voice preset handling and GenerationMixin import failure are different subsystems and defects." }, { "left": "issue:43825", "right": "issue:43827", "accept": false, - "reason": "Both relate to v5 pipeline removals, but one is an error message problem and the other is docs fallout; not the same bug/change." + "reason": "These are related follow-ups around v5 pipeline removals, but one is an error message issue and the other is docs cleanup, not the same bug." }, { "left": "issue:44485", "right": "issue:44509", "accept": false, - "reason": "GLM-5 RoPE implementation discussion is unrelated to stale pipeline task docs." + "reason": "GLM-5 RoPE implementation discussion is unrelated to documentation still referencing removed pipeline tasks." }, { "left": "issue:44805", "right": "issue:44936", "accept": false, - "reason": "Different failures: mask/tensor shape IndexError vs trainer.evaluate() after train state bug." + "reason": "Different failures: a mask/index shape mismatch versus trainer.evaluate() breaking after train()." }, { "left": "issue:44671", "right": "issue:45701", "accept": false, - "reason": "CamemBERT masked-LM regression is not the same as a broad tokenization-change report." + "reason": "CamemBERT masked-LM regression and a broad tokenization-change report are not the same concrete issue." }, { "left": "issue:28218", "right": "issue:29159", - "accept": true, - "reason": "Both describe the same tokenizer added-token spacing problem; 29159 is the broader slow-vs-fast inconsistency report of that behavior." + "accept": false, + "reason": "Both are tokenizer-related, but one is about added-token spacing and the other about slow/fast tokenizer inconsistency; too broad to merge." }, { "left": "issue:44610", "right": "issue:45072", "accept": false, - "reason": "Processor image-size mismatch and dtype mismatch are different code paths and bugs." + "reason": "Different models and different dtype/input-size problems; not the same bug." }, { "left": "issue:43720", "right": "issue:45072", "accept": false, - "reason": "Packed-weight loading failure is unrelated to the dtype mismatch reported in issue:45072." + "reason": "BitNet packed-weight loading and generic bfloat16 dtype mismatches are unrelated." }, { "left": "issue:43782", "right": "issue:45325", "accept": false, - "reason": "Different Qwen VL bugs: weight_only loading error vs rope index scaling for still images." + "reason": "Qwen3VL weight_only loading failure and Qwen2.5-VL rope-index scaling are separate bugs." }, { "left": "issue:42175", "right": "issue:45081", "accept": false, - "reason": "Backend install dependency issue is unrelated to Mistral regex patch crash." + "reason": "Package/backend installation issue versus a tokenizer patch crash; no shared underlying defect." }, { "left": "issue:43749", "right": "issue:44291", "accept": false, - "reason": "Both concern loading, but the failing mechanisms differ: FSDP CPU RAM efficient loading vs unexpected _is_hf_initialized argument." + "reason": "Both touch model loading, but one is FSDP CPU RAM-efficient loading and the other is init_empty_weights argument handling." }, { "left": "issue:43582", "right": "issue:44857", "accept": false, - "reason": "Apple Silicon warmup TypeError and float16 AMP CUDA crash are separate issues." + "reason": "Apple Silicon warmup TypeError and CUDA AMP loss crash are unrelated runtime failures." }, { "left": "issue:45081", "right": "issue:45406", "accept": false, - "reason": "Different processor/tokenizer failures: Mistral regex patch crash vs Gemma4Processor missing _tokenizer in serve." + "reason": "Different objects and different missing-attribute failures; not the same code path." }, { "left": "issue:43650", "right": "issue:43906", "accept": false, - "reason": "A vague placeholder issue is not the same as an isolated reproduction of another bug." + "reason": "One is a nonspecific request, the other a reproduction of another issue; not duplicates." }, { "left": "issue:43262", "right": "issue:43819", "accept": false, - "reason": "Audio chat-template sample-rate defaulting and DAC.from_latents STE mismatch are unrelated." + "reason": "Audio chat-template sampling-rate defaulting and DAC latent reconstruction mismatch are unrelated." }, { "left": "issue:43526", "right": "issue:43697", "accept": false, - "reason": "BeitImageProcessorFast label reduction bug is unrelated to RTDetrV2 output drift in v5." + "reason": "BeitImageProcessorFast label reduction and RTDetrV2 output drift are separate bugs." }, { "left": "issue:44060", "right": "issue:44368", "accept": false, - "reason": "Both mention weight tying/warnings, but they involve different models and root causes." + "reason": "Both involve tied-weights warnings, but the concrete problems differ and do not look like one mergeable fix." }, { "left": "issue:34634", "right": "issue:36331", "accept": false, - "reason": "BarkProcessor voice preset failure is unrelated to CustomTrainer.compute_loss signature mismatch." + "reason": "BarkProcessor voice preset and CustomTrainer.compute_loss signature mismatch are unrelated." }, { "left": "issue:43653", "right": "issue:44912", "accept": false, - "reason": "Tokenizer special-token registration bug is unrelated to MXFP4 quantization fallback behavior." + "reason": "Tokenizer special-token registration and MXFP4 quantization fallback are different issues." }, { "left": "issue:45084", "right": "issue:45538", "accept": false, - "reason": "Template compilation error and CLIPTokenizer max_length bug are different issues." + "reason": "Template compilation error and CLIPTokenizer max-length bug are unrelated." }, { "left": "issue:43526", "right": "issue:44554", "accept": false, - "reason": "Beit label reduction and MPS attention correctness are unrelated." + "reason": "A label-reduction bug is unrelated to an MPS attention correctness issue." }, { "left": "issue:43742", "right": "issue:45020", "accept": false, - "reason": "MobileLLM key error and broad remote_code breakage are not the same concrete bug." + "reason": "A specific MobileLLM key error is not the same underlying problem as the broad remote_code regression." }, { "left": "issue:43329", "right": "issue:45600", "accept": false, - "reason": "Undefined video-branch helpers in multimodal token counting is unrelated to stale auto_mappings config references." + "reason": "Multimodal token counting bug and removed auto-mapping references are unrelated." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative on the soft pairs. The 28218/29159 tokenizer pair is the only plausible duplicate-style match from the titles, and the rest are reasonably treated as unrelated or only loosely similar. No PRs 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 }, @@ -5349,217 +5547,217 @@ "issue:44568|issue:45356", "issue:44849|issue:45701" ], - "prepared_review_unit_hash": "c08fa81563f9e2094e7b6727ab6f98a14c20b38391e3ba416af9fa16946bdf00", + "prepared_review_unit_hash": "99f221b6c374693eaccfdc53c5cd1769c27b79d6a95988430adc737dd47c335a", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36622, - "estimated_input_tokens": 9156, - "estimated_eval_tokens": 18568 + "serialized_chars": 36617, + "estimated_input_tokens": 9155, + "estimated_eval_tokens": 18566 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly heterogeneous transformers bugs; only one soft duplicate pair is convincing: the Qwen video regression reports. The rest differ by model, subsystem, or failure mode and should not be merged as duplicates.", - "confidence": 0.91, - "canonical_issue_reason": "issue:44479 is the broadest, most user-facing report of the video-input regression across multiple Qwen VL/MoE models; issue:43329 looks like a narrower root-cause note in the same area.", + "summary": "Mostly unrelated Transformer bugs with one plausible duplicate pair around the Qwen video-input regression. I would keep only issue:43329 \u2194 issue:44479 as a duplicate/soft-duplicate candidate; the rest differ by model family, failure mode, or code path.", + "confidence": 0.77, + "canonical_issue_reason": "issue:44479 is the best canonical issue in this packet because its title describes a broader video-input regression across several Qwen VL variants, which makes it a reasonable anchor for the accepted soft pair.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44479 best represents the cluster\u2019s only real duplicate theme because it describes the regression across several affected model families and the symptom users hit directly.", + "best_issue_reason": "issue:44479 is the strongest single issue to represent the cluster because it has the broadest affected-model scope among the paired video-regression reports and reads like the more general symptom-level entry.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44589", "right": "issue:45042", "accept": false, - "reason": "Different subsystems: float8 storage/loading vs PIL image processor dependency handling." + "reason": "Different bug classes: float8 storage lookup vs PIL/torchvision dependency handling." }, { "left": "issue:43746", "right": "issue:44811", "accept": false, - "reason": "Unrelated models and bugs: GraniteSpeech local checkpoint loading vs Whisper batch_decode token handling." + "reason": "Both involve loading/processing, but they target different components and failure modes." }, { "left": "issue:45459", "right": "issue:45803", "accept": false, - "reason": "Both involve bad exception handling, but they are different code paths and different failure modes." + "reason": "Both mention exception handling, but in different code paths and with different symptoms." }, { "left": "issue:43638", "right": "issue:44936", "accept": false, - "reason": "Different trainer failures: ZeRO-3/Non-pretrained BERT indexing vs evaluate-after-train lifecycle issue." + "reason": "Separate training/evaluation failures without a shared concrete root cause." }, { "left": "issue:43452", "right": "issue:44991", "accept": false, - "reason": "Both are loading regressions, but one is gguf_file support and the other is a tokenizer-specific regression; not the same bug." + "reason": "Both are loading regressions, but for different assets and likely different code paths." }, { "left": "issue:44387", "right": "issue:44617", "accept": false, - "reason": "Both mention OOM, but they are different causes, models, and execution paths." + "reason": "OOM reports for different models and memory paths." }, { "left": "issue:36010", "right": "issue:43645", "accept": false, - "reason": "ImportError for GenerationMixin is unrelated to custom model initialization in notebooks." + "reason": "ImportError for GenerationMixin is unrelated to notebook custom-model initialization." }, { "left": "issue:43550", "right": "issue:44912", "accept": false, - "reason": "Different model-specific loading/compile problems; no shared code-path bug is evident." + "reason": "Different model/runtime failures: torch.compile+SDPA vs MXFP4 loading behavior." }, { "left": "issue:43819", "right": "issue:43927", "accept": false, - "reason": "One is a DAC forward/from_latents mismatch; the other is config save/load losing token IDs." + "reason": "Distinct bugs: DAC latent mismatch vs DiaConfig token-ID persistence." }, { "left": "issue:43854", "right": "issue:44977", "accept": false, - "reason": "Different model-specific generation bugs with different likely causes." + "reason": "Both are generation/loading issues at a broad level, but the models and symptoms differ." }, { "left": "issue:43257", "right": "issue:43329", "accept": false, - "reason": "Qwen3 MoE weight conversion/loading vs multimodal token counting in the video branch are unrelated." + "reason": "Qwen3 MoE loading/conversion is unrelated to multimodal video token counting." }, { "left": "issue:39692", "right": "issue:44912", "accept": false, - "reason": "SigLIP2 doc example issues are unrelated to Qwen3.5/OSS quantization loading." + "reason": "Documentation/example failures are not the same as quantized model loading behavior." }, { "left": "issue:43720", "right": "issue:44610", "accept": false, - "reason": "Packed-weight loading bug vs image processor output-size mismatch; different code paths." + "reason": "Packed-weight loading and image-size mismatch are unrelated problems." }, { "left": "issue:38175", "right": "issue:44610", "accept": false, - "reason": "Model output anomaly vs processor size mismatch; no common underlying bug." + "reason": "Different model output/shape problems with no clear shared code path." }, { "left": "issue:43454", "right": "issue:43577", "accept": false, - "reason": "Weight tying bug and dtype propagation bug are different issues." + "reason": "Weight tying failure and dtype propagation are separate loading defects." }, { "left": "issue:36010", "right": "issue:36331", "accept": false, - "reason": "Import path breakage vs trainer API keyword mismatch; not duplicates." + "reason": "Importing GenerationMixin and custom compute_loss signature handling are unrelated API issues." }, { "left": "issue:43531", "right": "issue:45042", "accept": false, - "reason": "Qwen3-MoE sliding_window behavior is unrelated to PIL/torchvision backend requirements." + "reason": "Qwen3-MoE sliding_window behavior is unrelated to PIL backend requirements." }, { "left": "issue:43257", "right": "issue:43696", "accept": false, - "reason": "Different problems: weight conversion during accelerated loading vs plain CUDA OOM." + "reason": "A loading/conversion bug and a plain CUDA OOM are not the same issue." }, { "left": "issue:43262", "right": "issue:44811", "accept": false, - "reason": "Both are audio processor issues, but one is chat template sample rate handling and the other is Whisper decode token skipping." + "reason": "Audio chat-template sampling-rate defaults and Whisper batch_decode token skipping are different behaviors." }, { "left": "issue:43329", "right": "issue:44479", "accept": true, - "reason": "Both describe the same video-path regression in Qwen VL/MoE models; 43329 identifies the likely root cause in _get_num_multimodal_tokens, and 44479 reports the broader user-facing regression." + "reason": "These look like the same underlying Qwen video-input regression: 43329 points to the broken video-token counting branch, and 44479 reports the broader downstream failure across the same model family." }, { "left": "issue:42490", "right": "issue:43257", "accept": false, - "reason": "Local-path slash import warning is unrelated to Qwen3 MoE accelerate/deepspeed weight conversion." + "reason": "Local-path import warnings and Qwen3 MoE loading/conversion are unrelated." }, { "left": "issue:43749", "right": "issue:44756", "accept": false, - "reason": "Both are memory/loading related, but one is an FSDP loading regression and the other is a Strix Halo mmap OOM workaround." + "reason": "Different loading/memory issues: FSDP CPU RAM efficient loading vs mmap on Strix Halo OOM." }, { "left": "issue:43643", "right": "issue:45759", "accept": false, - "reason": "AutoConfig missing fields with trust_remote_code is not the same as AutoModelForCausalLM.from_config failing to unwrap text_config." + "reason": "Missing config fields with trust_remote_code is different from unwrapping text_config in composite multimodal configs." }, { "left": "issue:43329", "right": "issue:43450", "accept": false, - "reason": "Video token counting bug vs batched video processor shape bug; related area but different underlying defects." + "reason": "Video token counting and batched video processor shape are different code paths." }, { "left": "issue:34567", "right": "issue:40990", "accept": false, - "reason": "TrainerState token accounting is unrelated to GPT-oss perplexity on WikiText-2." + "reason": "Trainer token accounting and GPT-oss perplexity are unrelated." }, { "left": "issue:43606", "right": "issue:45859", "accept": false, - "reason": "Both involve CPU offload, but they affect different models and code paths." + "reason": "CPU offload device mismatch is unrelated to Qwen3_5Moe rotary embedding compatibility." }, { "left": "issue:43257", "right": "issue:43931", "accept": false, - "reason": "Qwen3 MoE loading/convert issue is distinct from Qwen3-VL weight shape mismatch." + "reason": "Different Qwen loading failures with different symptoms and likely different root causes." }, { "left": "issue:44568", "right": "issue:45356", "accept": false, - "reason": "Both are tokenizer regressions, but on different models and different root causes." + "reason": "Tokenizer special-token insertion and codec handling regressions are separate bugs." }, { "left": "issue:44849", "right": "issue:45701", "accept": false, - "reason": "Qwen3.5 hidden-state behavior is unrelated to a broad tokenization-version change." + "reason": "Hidden-state behavior in Qwen3.5 is not the same as broader tokenization changes across versions." } ] }, "evaluator_result": { - "accept": true, - "feedback": "The analysis is well-grounded in the packet and stays conservative. The only accepted soft duplicate pair, issue:43329 vs issue:44479, is plausibly the same Qwen video-path regression, while the other pairs are reasonably separated by model, subsystem, or failure mode based on the titles/body summaries provided." + "accept": false, + "feedback": "The cluster-level conclusion is mostly conservative, and the one accepted soft pair (issue:43329 vs issue:44479) is plausible. However, the canonical/best-issue rationale overstates what is supported by the packet: the metadata only shows titles and discussion counts, so claims like \"clearly reproducible,\" \"user-facing,\" and \"umbrella symptom\" are not directly grounded. Please soften those statements to match the evidence available from the packet alone." }, "evaluator_used": true, - "retried": false, + "retried": true, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -5661,232 +5859,214 @@ "issue:28282|issue:29127", "issue:43582|issue:43782" ], - "prepared_review_unit_hash": "ad0d9aade35d05b812c810d30dbdfccc18e55af6c672c4e17c6059f4b2bc7a3c", + "prepared_review_unit_hash": "64efa54ddb92119ebb1d85079f07cb2a53af388dfb2e103738475ba9717b6f3b", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 33, - "serialized_chars": 37679, - "estimated_input_tokens": 9420, - "estimated_eval_tokens": 19096 + "serialized_chars": 37685, + "estimated_input_tokens": 9422, + "estimated_eval_tokens": 19100 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a mixed bag of unrelated issue reports, with one credible duplicate pair around the Qwen VL video-input regression. No PRs are present.", - "confidence": 0.45, - "canonical_issue_reason": "issue:44479 is the broadest and most representative report of the Qwen video-input regression, and issue:45381 is a narrower symptom report that fits under it.", + "summary": "This cluster is mostly a grab-bag of unrelated Transformers bugs. The only clear duplicate-like pair is the Qwen video-input regression pair; the rest are only superficially similar by subsystem or model family and should stay separate.", + "confidence": 0.89, + "canonical_issue_reason": "issue:44479 is the best representative of the only coherent subgroup: a broad v5.3.0 Qwen video-input regression affecting multiple models, with a narrower companion issue (45381) pointing at the same area.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44479 has the clearest scope for the only true duplicate relation here, covering the wider v5.3.0 video-input regression across Qwen VL variants.", + "best_issue_reason": "issue:44479 is the strongest cluster anchor because it describes a concrete regression, spans multiple affected Qwen variants, and has the clearest overlap with the one acceptable soft pair.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44568", "right": "issue:45488", "accept": false, - "reason": "Both are tokenizer regressions, but they describe different failures and different code paths/models." + "reason": "Both are tokenizer regressions, but one is about add_special_tokens/BOS-EOS behavior and the other about LlamaTokenizer overriding tokenizer.json pre-tokenization; different bugs." }, { "left": "issue:29159", "right": "issue:34567", "accept": false, - "reason": "Slow/fast tokenizer inconsistency is unrelated to TrainerState token counting." + "reason": "Tokenizer slow/fast inconsistency and TrainerState token counting are unrelated code paths and symptoms." }, { "left": "issue:42915", "right": "issue:43742", "accept": false, - "reason": "Different models and different failures: FP8 config loading versus a key error on MobileLLM." + "reason": "Qwen3Moe FP8 loading failure and MobileLLM key error are different model-loading problems." }, { "left": "issue:44291", "right": "issue:45042", "accept": false, - "reason": "One is an init_empty_weights argument bug; the other is an image processor torchvision dependency regression." + "reason": "Unexpected _is_hf_initialized during init_empty_weights is unrelated to the PIL backend's torchvision dependency regression." }, { "left": "issue:42915", "right": "issue:44821", "accept": false, - "reason": "Qwen3Moe FP8 loading and AutoImageProcessor URL loading are unrelated issues." + "reason": "Quantization/model-loading failure and AutoImageProcessor URL loading are not the same bug." }, { "left": "issue:43452", "right": "issue:43582", "accept": false, - "reason": "GGUF tokenizer/model loading and an Apple Silicon caching_allocator_warmup TypeError are unrelated." + "reason": "gguf_file tokenizer/model loading and AppleSilicon caching_allocator_warmup TypeError are unrelated." }, { "left": "issue:44987", "right": "issue:45538", "accept": false, - "reason": "Model loading failure for one repo versus CLIPTokenizer max_length behavior are different bugs." + "reason": "A model-loading regression for a specific repo and CLIPTokenizer's absurd model_max_length are different issues." }, { "left": "issue:41720", "right": "issue:43540", "accept": false, - "reason": "Both involve Qwen models, but one is a device-mapping CUDA assert and the other is a video-input processing error." + "reason": "Both mention Qwen and failures, but one is device-mapping/cuda assert and the other is video-input processing; not the same bug." }, { "left": "issue:43697", "right": "issue:45357", "accept": false, - "reason": "Different problems: RTDetrV2 output drift versus incorrect save_pretrained visual encoder keys." + "reason": "RTDetrV2 output divergence and Qwen3.5 save_pretrained key corruption are unrelated." }, { "left": "issue:42915", "right": "issue:45538", "accept": false, - "reason": "Different tokenizer/model behaviors with no shared code-path evidence." + "reason": "Qwen3Moe FP8 loading and CLIP tokenizer max_length are unrelated." }, { "left": "issue:43927", "right": "issue:44912", "accept": false, - "reason": "DiaConfig save/load token-id loss is unrelated to MXFP4 quantization loading." + "reason": "DiaConfig token ID persistence and MXFP4 quantization fallback are different failure modes." }, { "left": "issue:44479", "right": "issue:45381", "accept": true, - "reason": "Both report the same Qwen VL video-input regression in v5.3.0; 45381 is a narrower manifestation of the broader issue." + "reason": "45381 is a narrower manifestation of the same Qwen video-input regression covered by 44479, both centered on broken video position handling in v5.3.0." }, { "left": "issue:33357", "right": "issue:34567", "accept": false, - "reason": "MacOS bus error on CLIP loading is unrelated to TrainerState token accounting." + "reason": "MacOS CLIP bus error and TrainerState token counter stalling are unrelated." }, { "left": "issue:42175", "right": "issue:43824", "accept": false, - "reason": "Backend dependency packaging and an import error for a missing model class are unrelated." + "reason": "Missing TensorFlow in the torch extra and missing Qwen2_5_VLForConditionalGeneration export are separate packaging/import issues." }, { "left": "issue:43540", "right": "issue:44479", "accept": false, - "reason": "Both are video-input related, but they target different Qwen model families and the symptoms differ." + "reason": "Both are video-related, but they affect different Qwen model families and likely different code paths." + }, + { + "left": "issue:43317", + "right": "issue:45538", + "accept": false, + "reason": "Device-map/offload dequantization failure is unrelated to CLIPTokenizer's model_max_length bug." }, { "left": "issue:43257", "right": "issue:43756", "accept": false, - "reason": "Qwen3 MOE weight conversion with accelerate/deepspeed is unrelated to Smollm3 RoPE layer dropping." + "reason": "Qwen3 MoE weight conversion under accelerate+deepspeed and Smollm3 RoPE layer loss are unrelated." }, { "left": "issue:43577", "right": "issue:45072", "accept": false, - "reason": "Both mention dtype mismatches, but the affected models and failure modes are different." + "reason": "Blip2 dtype persistence and SwitchTransformers/TimmWrapper bfloat16 mismatches are distinct dtype issues with different models." }, { "left": "issue:44987", "right": "issue:45907", "accept": false, - "reason": "Model loading regression and forwarding a revision argument are unrelated." + "reason": "Model loading failure and list_repo_files not forwarding revision are unrelated." }, { "left": "issue:44062", "right": "issue:44991", "accept": false, - "reason": "AddedToken special-flag duplication is not the same as loading EMBEDDIA/est-roberta." + "reason": "AddedToken special-kwarg TypeError and tokenizer loading failure for est-roberta are not the same defect." }, { "left": "issue:45137", "right": "issue:45405", "accept": false, - "reason": "DeepSpeed ZeRO3 deque failure and a peft version bump are unrelated." + "reason": "DeepSpeed ZeRO3 deque underflow is unrelated to the PEFT minimum-version pin issue." }, { "left": "issue:24643", "right": "issue:29159", "accept": false, - "reason": "DeepSpeed training weight-dimension error is unrelated to tokenizer inconsistency." + "reason": "DeepSpeed weight dimensionality error and tokenizer slow/fast inconsistency are unrelated." }, { "left": "issue:28282", "right": "issue:34689", "accept": false, - "reason": "Missing PyTorch import and Llama 3.2 vision model loading are different issues." + "reason": "Missing PyTorch import and Llama 3.2 Vision model-loading regression are different issues." }, { "left": "issue:43742", "right": "issue:44291", "accept": false, - "reason": "KeyError loading a model and an init_empty_weights argument mismatch are unrelated." + "reason": "MobileLLM key error and unexpected _is_hf_initialized during init_empty_weights are different loading bugs." }, { "left": "issue:42175", "right": "issue:44488", "accept": false, - "reason": "Packaging/backends issue and a specific model load failure are unrelated." + "reason": "Torch extra backend packaging and inability to load cjvt/sleng-bert are unrelated." }, { "left": "issue:42915", "right": "issue:43749", "accept": false, - "reason": "Qwen3Moe FP8Config failure and FSDP CPU RAM efficient loading breakage are unrelated." + "reason": "Qwen3Moe FP8 config failure and FSDP CPU RAM efficient loading broken are different issues." }, { "left": "issue:44737", "right": "issue:45588", "accept": false, - "reason": "XLNet CPU placement bug and flash_attention sink handling are unrelated." + "reason": "XLNet CPU-side arange and flash_attention s_aux=None crash are unrelated code paths." }, { "left": "issue:43404", "right": "issue:45987", "accept": false, - "reason": "Tied lm_head weights in Mistral3 and StaticCache return type are unrelated." - }, - { - "left": "issue:43906", - "right": "issue:45381", - "accept": false, - "reason": "A reproduction of an older issue is not the same bug as the Qwen2.5-VL video-input regression." - }, - { - "left": "issue:42175", - "right": "issue:44938", - "accept": false, - "reason": "TensorFlow backend packaging and Python 3.14 import/load failure are different problems." - }, - { - "left": "issue:43906", - "right": "issue:45381", - "accept": false, - "reason": "An isolated reproduction of a prior issue does not match the Qwen2.5-VL video-input regression." - }, - { - "left": "issue:28282", - "right": "issue:29127", - "accept": false, - "reason": "AutoModel missing PyTorch and LayoutLMv3 box-validation messaging are unrelated." + "reason": "lm_head weight tying in Mistral3 and StaticCache returning a tensor instead of int are unrelated." }, { - "left": "issue:43582", - "right": "issue:43782", + "left": "issue:43782", + "right": "issue:43582", "accept": false, - "reason": "Apple Silicon caching_allocator_warmup TypeError and Qwen3VL weight_only loading are unrelated." + "reason": "Qwen3VL weight_only loading error and AppleSilicon caching_allocator_warmup TypeError are unrelated." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet: all items are issues, so noting no PRs is correct. The lone accepted duplicate pair (44479/45381) is reasonably supported by the titles as the broader Qwen VL v5.3.0 video-input regression and a narrower qwen2.5-vl symptom. The rejected soft pairs are phrased conservatively and do not overclaim duplication." + "feedback": "The clustering is grounded in the packet: the Qwen v5.3.0 video-input regression pair (44479/45381) is the only clearly related subgroup, and the other soft-edge rejections are conservative and model/code-path specific. No overstatement stands out from the provided titles and descriptions." }, "evaluator_used": true, "retried": false, @@ -5985,203 +6165,206 @@ "issue:43452|issue:44843", "issue:44514|issue:45290" ], - "prepared_review_unit_hash": "847b8b5b37d5c8a8d6dd555bdd6a9d43394141872c9972ff6de6b1968bc3a99a", + "prepared_review_unit_hash": "bdf7406f496e1b7144f52f893e469ccc011a692d222abdaf7b07ad534eb2bdfa", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 27, - "serialized_chars": 36213, - "estimated_input_tokens": 9054, - "estimated_eval_tokens": 18364 + "serialized_chars": 36219, + "estimated_input_tokens": 9055, + "estimated_eval_tokens": 18366 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is highly heterogeneous: these are mostly unrelated loader, config, trainer, and model-specific bugs. I would not merge any of the soft pairs as duplicates.", - "confidence": 0.19, - "canonical_issue_reason": "Issue 43452 is the broadest, most user-facing loading regression in the set, but there is no strong evidence that any items here are true duplicates.", + "summary": "Mostly unrelated issue reports; only the two SigLIP2 reports look like the same underlying model-quality bug. No PRs are present in this cluster.", + "confidence": 0.66, + "canonical_issue_reason": "Issue 45920 is the broadest, most representative open report: it describes a cross-model tokenizer regression with multiple affected checkpoints and clear reproduction potential.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 43452 best represents the cluster only in a loose sense because it covers a general from_pretrained loading failure; the rest are mostly narrower, distinct bugs.", + "best_issue_reason": "Issue 45920 is the best issue candidate because it is open, spans several models, and is phrased as a higher-level regression rather than a narrow single-model symptom.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:38175", "right": "issue:44898", "accept": false, - "reason": "Different models and different failure modes; one is a zero-probability output issue, the other is an image-resolution/interpolation failure." + "reason": "Both are vision-model bugs, but they concern different models and different failure modes; not the same underlying defect." }, { "left": "issue:43540", "right": "issue:45823", "accept": false, - "reason": "Completely different subsystems: Qwen3OmniMoe video processing vs Gemma4 PLE device placement." + "reason": "Different models and different runtime problems: video-input handling versus device-mismatch in forward pass." }, { "left": "issue:43493", "right": "issue:43994", - "accept": false, - "reason": "Same model family, but the reports describe different symptoms and code paths; not enough evidence of one underlying bug." + "accept": true, + "reason": "Both report SigLIP2 producing incorrect outputs; one frames it as HF-vs-JAX discrepancy and the other as nonsensical AutoModel/pipeline results, which plausibly stem from the same model implementation bug." }, { "left": "issue:43526", "right": "issue:45479", "accept": false, - "reason": "Beit image label reduction bug is unrelated to single-label classification zero-loss behavior." + "reason": "Unrelated bugs in different components: image processor label reduction vs sequence-classification zero-loss behavior." }, { "left": "issue:44291", "right": "issue:44821", "accept": false, - "reason": "Init-empty-weights loading error and AutoImageProcessor URL loading are unrelated." + "reason": "Different loading paths and failures; init_empty_weights/_is_hf_initialized is not the same issue as loading AutoImageProcessor from URL." }, { "left": "issue:43756", "right": "issue:45325", "accept": false, - "reason": "Both involve RoPE/position logic, but on different models and with different mechanisms; too weak to treat as the same bug." + "reason": "Both involve RoPE, but they are distinct model-specific bugs with different code paths and symptoms." }, { "left": "issue:34567", "right": "issue:36331", "accept": false, - "reason": "Both are trainer-related, but one is a state counter update bug and the other is an unexpected compute_loss API argument." + "reason": "TrainerState token counting and custom compute_loss signature regression are separate trainer issues." }, { "left": "issue:44913", "right": "issue:46121", "accept": false, - "reason": "Both mention rope config handling, but one is reload persistence and the other is JSON list validation in a different helper." + "reason": "Both touch rope-config handling, but one is a reload defaulting bug and the other is a type error during validation; different failures." }, { "left": "issue:44062", "right": "issue:45081", "accept": false, - "reason": "Tokenizer-related, but one is AddedToken construction and the other is mistral regex patching; different concrete failures." + "reason": "Different tokenizer-related crashes in different helpers; no shared concrete bug." }, { "left": "issue:44792", "right": "issue:44964", "accept": false, - "reason": "Different multimodal models and distinct loading/test failures." + "reason": "Both are multimodal-model reports, but they concern different models and failures." }, { "left": "issue:45357", "right": "issue:45920", "accept": false, - "reason": "One is Qwen3.5 save_pretrained visual key serialization, the other is tokenizer ID regression across multiple models." + "reason": "Qwen3.5 save_pretrained visual-key corruption is unrelated to the tokenizer ID regression in 45920." }, { "left": "issue:45042", "right": "issue:45081", "accept": false, - "reason": "Image processor torchvision dependency issue is unrelated to Mistral regex patch crashes." + "reason": "Image-processor torchvision dependency regression is unrelated to the Mistral regex/backend_tokenizer crash." }, { "left": "issue:43452", "right": "issue:44488", "accept": false, - "reason": "Both are loading-related, but 43452 is a gguf_file path regression while 44488 is a model-specific load failure." + "reason": "Both are load failures, but one is about gguf_file/AutoModel loading and the other about a specific model not loading; not enough evidence of the same bug." }, { "left": "issue:43742", "right": "issue:44589", "accept": false, - "reason": "KeyError on MobileLLM loading is unrelated to the float8 storage object error." + "reason": "KeyError while loading a model and missing Float8 storage are different deserialization issues." }, { "left": "issue:43531", "right": "issue:44821", "accept": false, - "reason": "Qwen3-MoE sliding window behavior and AutoImageProcessor URL loading are unrelated." + "reason": "Qwen3-MoE sliding-window behavior and AutoImageProcessor URL loading are unrelated." }, { "left": "issue:43643", "right": "issue:45375", "accept": false, - "reason": "Both involve config fields, but one is generic trust_remote_code missing fields and the other is a specific strict-field drop in Qwen3_5MoeVisionConfig." + "reason": "Strict config/missing fields with trust_remote_code is not the same as a missing deepstack_visual_indexes field in Qwen3_5MoeVisionConfig." }, { "left": "issue:45405", "right": "issue:45600", "accept": false, - "reason": "Release-version pinning and removed auto_mappings references are different CI/package issues." + "reason": "Release/versioning mismatch for MIN_PEFT_VERSION is unrelated to removed auto_mappings references." }, { "left": "issue:43720", "right": "issue:44898", "accept": false, - "reason": "BitNet packed-weight loading is unrelated to Perceiver image classification interpolation failure." + "reason": "BitNet packed-weight loading and Perceiver positional-encoding failure are different issues." }, { "left": "issue:43697", "right": "issue:45216", "accept": false, - "reason": "Both are regressions, but they affect different models and distinct code paths." + "reason": "RTDetrV2 output drift and Qwen3.5 save_pretrained regression are different code paths and symptoms." }, { "left": "issue:44625", "right": "issue:44877", "accept": false, - "reason": "Num-label propagation bug and strict config loading failure are not the same issue." + "reason": "Qwen3.5 num_labels propagation and granite_speech strict-config loading are separate config bugs." }, { "left": "issue:24643", "right": "issue:29127", "accept": false, - "reason": "Training-time DeepSpeed runtime error is unrelated to LayoutLMv3 box-information error messaging." + "reason": "DeepSpeed training weight-shape error and LayoutLMv3 error-message clarity are unrelated." }, { "left": "issue:44514", "right": "issue:45381", "accept": false, - "reason": "Both are Qwen2.5-VL related, but one is apply_chat_template batching and the other is video position-id computation." + "reason": "Both concern Qwen2.5-VL, but one is chat-template batching/padding and the other is video vision_position_ids; different bugs." }, { "left": "issue:41762", "right": "issue:44936", "accept": false, - "reason": "DeepSpeed ZeRO-3 loading crash and trainer.evaluate-after-train failure are unrelated." + "reason": "Gemma3 DeepSpeed ZeRO-3 loading failure and trainer.evaluate() regression are unrelated." }, { "left": "issue:43688", "right": "issue:43749", "accept": false, - "reason": "Aux-loss normalization bug and FSDP CPU RAM-efficient loading breakage are different problems." + "reason": "Auxiliary-loss normalization and FSDP CPU RAM efficient loading are different issues." }, { "left": "issue:43452", "right": "issue:43577", "accept": false, - "reason": "GGUF loader breakage and BLIP2 dtype propagation are unrelated." + "reason": "gguf_file loading breakage and Blip2 dtype propagation are unrelated loading problems." }, { "left": "issue:43452", "right": "issue:44843", "accept": false, - "reason": "GGUF loading regression and offline mistral_regex model_info call are different tokenizer/model-loading bugs." + "reason": "gguf_file/AutoModel loading failure and offline model_info call in _patch_mistral_regex are different bugs." }, { "left": "issue:44514", "right": "issue:45290", "accept": false, - "reason": "Both touch apply_chat_template, but the failures are different: batched padding=False crash vs tool-call assistant message handling." + "reason": "Same helper function, but the failing conditions differ materially: batched padding=False vs assistant tool-call messages with no content." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "Grounded overall: the selection of issue 45920 is supported by the packet (open, multi-model, v5 regression), and the lone accepted soft pair between the two SigLIP2 reports is a conservative match. The other pairwise rejections are appropriately cautious, and there are no PRs in this cluster." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -6277,20 +6460,20 @@ "issue:44464|issue:44977", "issue:43550|issue:43927" ], - "prepared_review_unit_hash": "eaee31f93710a13ed1f682e156ea7dbac6f78d3107bf10cb13d803b50b49ee1c", + "prepared_review_unit_hash": "ac81d24d938d679b1d12a3af3a5b61f79311008003b27770de18da239fe5481f", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 29, - "serialized_chars": 35810, + "serialized_chars": 35812, "estimated_input_tokens": 8953, "estimated_eval_tokens": 18162 }, @@ -6299,196 +6482,190 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is mostly unrelated one-off bug reports. The only plausible duplicate pair is the v5 tied-weights serialization regression (44466/45005); everything else is too model- or symptom-specific to merge safely.", - "confidence": 0.72, - "canonical_issue_reason": "issue:44466 is the clearest, most concrete report in the only plausible duplicate cluster: it names the v5 tied-weights/serialization path and gives a specific device-dependent symptom.", + "summary": "Most pairs are clearly unrelated. The tied-weights pair is the only close match, but with just titles and no shared targets or code evidence, it is safer to treat it as related rather than a confirmed duplicate.", + "confidence": 0.76, + "canonical_issue_reason": "issue:45005 is the best umbrella issue by title because it is broader and more general, but the evidence is only enough to justify cautious relatedness, not a definitive subsumption of issue:44466.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44466 is the best representative issue overall because it is specific, actionable, and points at a concrete serialization bug rather than a broad symptom report.", + "best_issue_reason": "issue:45005 is the strongest single issue to keep because it names the v5 tied-weights problem at a broader level and is the likeliest representative of the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44493", "right": "issue:46050", "accept": false, - "reason": "Both mention unexpected keys/serialization, but one is a general position-id key issue and the other is quantized save/load breakage; not the same bug." + "reason": "Different failure modes: position-id key handling vs quantized save/load breakage." }, { "left": "issue:44466", "right": "issue:45005", - "accept": true, - "reason": "Both describe the v5 tied-weights problem around translation models / lm_head serialization; 44466 is the concrete device-dependent manifestation of the same underlying regression." + "accept": false, + "reason": "The titles are closely related, but the packet does not provide enough evidence to confirm the same underlying bug or a clear duplicate relationship." }, { "left": "issue:43819", "right": "issue:44811", "accept": false, - "reason": "DAC latent reconstruction and Whisper batch decoding are unrelated code paths and symptoms." + "reason": "Different code paths: DAC latent decoding vs Whisper batch decoding behavior." }, { "left": "issue:30064", "right": "issue:33666", "accept": false, - "reason": "Void segmentation-map processing and Qwen2-VL multi-GPU training are unrelated." + "reason": "Segmentation-map preprocessing and multi-GPU training are unrelated problems." }, { "left": "issue:44936", "right": "issue:45823", "accept": false, - "reason": "Trainer evaluate-after-train and Gemma4 device-map mismatch are different failures in different subsystems." + "reason": "Trainer lifecycle failure vs device-mapping mismatch in Gemma4." }, { "left": "issue:44484", "right": "issue:44704", "accept": false, - "reason": "save_pretrained shard-size behavior and AutoProcessor kwargs forwarding are unrelated." + "reason": "A parameter/documentation question vs a cached_file kwargs-forwarding bug." }, { "left": "issue:43295", "right": "issue:45356", "accept": false, - "reason": "Both are regressions in tokenizer-related flows, but they affect different models and different failure modes." + "reason": "Both are tokenizer-related regressions, but they affect different models and mechanisms." }, { "left": "issue:44464", "right": "issue:44743", "accept": false, - "reason": "Both involve generation/cache behavior, but one is a broad compiled-forward inconsistency and the other is a Qwen3.5 recurrent-state reset bug; not the same defect." + "reason": "Both involve generation/state behavior, but the concrete bugs differ." }, { "left": "issue:43653", "right": "issue:44361", "accept": false, - "reason": "Different tokenizer bugs on different models; no clear shared root cause." + "reason": "Different tokenizer failures: special-token registration vs task-time AttributeError." }, { "left": "issue:44361", "right": "issue:44442", "accept": false, - "reason": "MLukeTokenizer AttributeError and AutoTokenizer failing to load FastSpeech2ConformerTokenizer are separate load/task failures." + "reason": "MLukeTokenizer task failure and FastSpeech2ConformerTokenizer loading are unrelated." }, { "left": "issue:28282", "right": "issue:32090", "accept": false, - "reason": "PyTorch-not-found import error and Trainer GPU broadcast NoneType error are unrelated." + "reason": "Import-time PyTorch missing error is unrelated to a Trainer broadcast TypeError." }, { "left": "issue:28282", "right": "issue:33357", "accept": false, - "reason": "ImportError for missing PyTorch and macOS bus error on CLIP are unrelated." + "reason": "Dependency import error and MacOS bus error are different failure modes." }, { "left": "issue:41720", "right": "issue:43906", "accept": false, - "reason": "The second item is only an isolated reproduction reference; there is no evidence it is the same Qwen3 auto-device-map crash." + "reason": "The reproduction issue does not clearly match the specific CUDA assert / device-mapping bug." }, { "left": "issue:38175", "right": "issue:43295", "accept": false, - "reason": "Zero probabilities in SigLIP2 and a processor/tokenizer regression are unrelated." + "reason": "Zero probabilities in SigLIP2 are unrelated to the processor/tokenizer regression." }, { "left": "issue:28282", "right": "issue:35141", "accept": false, - "reason": "Missing PyTorch and embedding reinitialization after resize are unrelated issues." + "reason": "PyTorch import failure and tied-embedding reinitialization are unrelated." }, { "left": "issue:45491", "right": "issue:46002", "accept": false, - "reason": "Both mention NaNs, but in different components and with different triggers; not the same bug." + "reason": "Different NaN sources: sliding-window attention vs classification-head init." }, { "left": "issue:43577", "right": "issue:43697", "accept": false, - "reason": "BLIP2 dtype handling and RTDetrV2 output divergence are unrelated." + "reason": "Blip2 dtype-casting and RTDetr output divergence are different bugs." }, { "left": "issue:43653", "right": "issue:44568", "accept": false, - "reason": "Both are tokenizer special-token complaints, but they concern different models and different behaviors; too broad to merge." + "reason": "Both mention special tokens, but they concern different token-handling bugs." }, { "left": "issue:43688", "right": "issue:45704", "accept": false, - "reason": "Auxiliary-loss normalization in MoE models and an apex RMSNorm memory leak are unrelated." + "reason": "Aux-loss normalization and apex RMSNorm memory leak are unrelated." }, { "left": "issue:43688", "right": "issue:45440", "accept": false, - "reason": "Both mention MoE models, but one is an auxiliary-loss bug and the other is a native-vs-remote implementation divergence; different underlying problems." + "reason": "MoE auxiliary-loss scaling and DeepSeekV3 implementation divergence are not the same issue." }, { "left": "issue:44843", "right": "issue:44991", "accept": false, - "reason": "Offline-mode model_info misuse and tokenizer load failure for a specific model are not clearly the same regression." + "reason": "Offline hub lookup bug and tokenizer-load regression for a specific model are different code paths." }, { "left": "issue:43646", "right": "issue:43749", "accept": false, - "reason": "Custom model initialization breakage and FSDP CPU RAM efficient loading are separate initialization/loading issues." - }, - { - "left": "issue:43688", - "right": "issue:45701", - "accept": false, - "reason": "MoE auxiliary-loss normalization and a broad tokenization-change report are unrelated." + "reason": "Custom model initialization breakage and FSDP CPU RAM-efficient loading are unrelated." }, { "left": "issue:29159", "right": "issue:33357", "accept": false, - "reason": "Slow-vs-fast tokenizer inconsistency and a macOS bus error on CLIP are different failures." + "reason": "Tokenizer inconsistency and a MacOS bus error do not share the same root cause." }, { "left": "issue:43531", "right": "issue:43742", "accept": false, - "reason": "Qwen3-MoE sliding-window behavior and MobileLLM key error are unrelated." + "reason": "Qwen3 sliding-window behavior and MobileLLM loading key errors are different problems." }, { "left": "issue:43653", "right": "issue:43927", "accept": false, - "reason": "BigBird special-token registration and DiaConfig custom token-ID loss are different bugs." + "reason": "Special-token registration and saved custom token ID persistence are related broadly, but not the same concrete bug." + }, + { + "left": "issue:43550", + "right": "issue:43927", + "accept": false, + "reason": "Torch.compile/SDPA failure is unrelated to DiaConfig token-ID persistence." }, { "left": "issue:43643", "right": "issue:44561", "accept": false, - "reason": "Both involve trust_remote_code, but one is missing fields from AutoConfig and the other is a v5 helper-removal regression; not enough evidence of the same root cause." + "reason": "Missing fields with trust_remote_code and removal of is_torch_fx_available are different regressions." }, { "left": "issue:44464", "right": "issue:44977", "accept": false, - "reason": "Compiled-forward chunked generation inconsistency and Qwen3.5 flash-attention generation failure are different symptoms and likely different code paths." - }, - { - "left": "issue:43550", - "right": "issue:43927", - "accept": false, - "reason": "SDPA/torch.compile failure in Bamba and DiaConfig token-ID persistence are unrelated." + "reason": "Chunked generation with compiled forward is not the same as flash-attention generation on Qwen3.5." } ] }, "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative overall. The only accepted soft edge, 44466/45005, is supported by closely aligned titles about tied weights and v5 serialization behavior. The other rejects are also reasonably cautious given the available evidence." + "accept": false, + "feedback": "The tied-weights match (issue:44466 vs issue:45005) is plausible from the titles, but the packet only supports cautious relatedness, not the stronger claim that 45005 clearly subsumes 44466 or that they are definitively the same underlying bug. The rest of the soft-edge false verdicts are conservative enough, but this accepted pair overstates the evidence, so I can\u2019t accept the summary as written." }, "evaluator_used": true, - "retried": false, - "accepted_nontrivial_soft_edge": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -6587,232 +6764,220 @@ "issue:44964|issue:45405", "issue:39692|issue:43493" ], - "prepared_review_unit_hash": "392b5b50b8cc4a3eacd6f33efa7df5bd67be943be86b10de56feaf6f4a3a9224", + "prepared_review_unit_hash": "2bb8e27306b5c7264f2984a01db02689089e9b690e8bc298dac00dd2fa3f33c4", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 37105, - "estimated_input_tokens": 9277, - "estimated_eval_tokens": 18810 + "serialized_chars": 37103, + "estimated_input_tokens": 9276, + "estimated_eval_tokens": 18808 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly a grab bag of unrelated Transformers bugs across tokenizers, configs, training, loading, vision/audio, and device-mapping. None of the soft pairs look like the same underlying bug/change, so all soft edges are rejected. The most representative issue is the broad silent tokenizer regression in Granite (45812).", - "confidence": 0.84, - "canonical_issue_reason": "issue:45812 is the strongest representative: it describes a broad, user-facing silent regression in core tokenization behavior affecting all Granite models, with clear impact and a concrete code path.", + "summary": "These items are a grab-bag of unrelated Transformers bugs and regressions. None of the soft pairs look like the same underlying issue or change, so all candidate duplicate links should be rejected.", + "confidence": 0.97, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:45812 is the best cluster representative because it is high-impact, well-scoped, and directly about a concrete regression rather than an implementation-specific edge case.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44479", "right": "issue:44625", "accept": false, - "reason": "Different subsystems and failures: video-input regression for Qwen VL models vs missing num_labels propagation in Qwen3.5 config." + "reason": "Different Qwen3/VL regression vs config propagation bug; no shared code-path or fix." }, { "left": "issue:43550", "right": "issue:44361", "accept": false, - "reason": "Unrelated bugs: torch.compile/SDPA failure in Bamba vs tokenizer task AttributeError in MLukeTokenizer." + "reason": "Bamba torch.compile/SDPA failure and MLukeTokenizer task AttributeError are unrelated." }, { "left": "issue:43577", "right": "issue:45588", "accept": false, - "reason": "Different code paths: BLIP2 dtype handling vs flash_attention crash on s_aux=None." + "reason": "Blip2 dtype loading bug vs flash_attention s_aux=None crash; different subsystems." }, { "left": "issue:41720", "right": "issue:44464", "accept": false, - "reason": "A800 auto-device-map cuda assert and chunked generation with compiled forward are separate model-runtime issues." + "reason": "Qwen3 device-mapping CUDA assert and chunked-generation compiled-forward inconsistency are distinct bugs." }, { "left": "issue:43577", "right": "issue:44554", "accept": false, - "reason": "Both touch attention/model numerics, but one is BLIP2 dtype loading and the other is an MPS correctness bug with mismatched head dims." + "reason": "dtype loading issue is unrelated to an MPS attention correctness bug." }, { "left": "issue:32090", "right": "issue:34567", "accept": false, - "reason": "Trainer broadcast NoneType error and TrainerState token counter not updating are different trainer bugs." + "reason": "Trainer broadcast TypeError and num_input_tokens_seen bookkeeping are separate trainer issues." }, { "left": "issue:43122", "right": "issue:43906", "accept": false, - "reason": "Tokenization regression across versions vs an isolated reproduction of another issue; no clear same bug evidence." + "reason": "Tokenization-version regression and an isolated reproduction are not enough to confirm the same bug." }, { "left": "issue:43504", "right": "issue:43720", "accept": false, - "reason": "Both are loading-related, but Beit legacy-field loading and BitNet packed-weight unpacking are different mechanisms." + "reason": "BEiT legacy-field loading failure is unrelated to BitNet packed-weight loading." }, { "left": "issue:43504", "right": "issue:43550", "accept": false, - "reason": "Different models and failures: Beit pretrained preset loading vs Bamba torch.compile/SDPA behavior." + "reason": "Pretrained preset load failure and Bamba SDPA compile failure affect different paths." }, { "left": "issue:43504", "right": "issue:44610", "accept": false, - "reason": "Beit loading bug and OmDet-Turbo image-size mismatch are unrelated." + "reason": "Legacy-field load bug vs OmDet-Turbo input-size mismatch are unrelated." }, { "left": "issue:32090", "right": "issue:33666", "accept": false, - "reason": "Trainer broadcast error and Qwen2-VL multi-GPU training request are not the same bug." + "reason": "Trainer broadcast error and Qwen2-VL multi-GPU training are different problems." }, { "left": "issue:44466", "right": "issue:46050", "accept": false, - "reason": "Different serialization bugs: tied lm_head.weight behavior vs quantized sinq save/load corruption." + "reason": "Serialization of tied lm_head weights is not the same as sinq quantized save/load breakage." }, { "left": "issue:33666", "right": "issue:36010", "accept": false, - "reason": "Multi-GPU training issue for Qwen2-VL and GenerationMixin import error are unrelated." + "reason": "Multi-GPU training issue and GenerationMixin import error do not share a concrete cause." }, { "left": "issue:43742", "right": "issue:44821", "accept": false, - "reason": "MobileLLM loading key error and AutoImageProcessor-from-URL failure affect different loaders and code paths." + "reason": "MobileLLM key error and AutoImageProcessor URL loading failure are unrelated loading bugs." }, { "left": "issue:45823", "right": "issue:46050", "accept": false, - "reason": "Gemma4 device-mapping forward mismatch and sinq quantized save/load failure are unrelated." + "reason": "Gemma4 device-map mismatch and quantized model save/load corruption are different failure modes." }, { "left": "issue:43452", "right": "issue:44451", "accept": false, - "reason": "gguf_file breaks specific tokenizer/model loading, while ScandiBERT cannot load for an unspecified separate reason." + "reason": "gguf_file loader breakage is not the same as ScandiBERT loading failure." }, { "left": "issue:43606", "right": "issue:43927", "accept": false, - "reason": "CPU offload device mismatch in bark-small and DiaConfig custom-token-ID loss after save/load are different bugs." + "reason": "CPU offload device mismatch and DiaConfig token-ID loss on save/load are distinct." }, { "left": "issue:43720", - "right": "issue:44361", + "right": "issue:43927", "accept": false, - "reason": "BitNet accelerate-loading unpacking bug is unrelated to MLukeTokenizer task AttributeError." + "reason": "Packed-weight unpacking during accelerate loading is unrelated to DiaConfig serialization of custom token IDs." }, { "left": "issue:43278", "right": "issue:43688", "accept": false, - "reason": "Embedding dtype drift during eval and auxiliary-loss normalization are different training/eval correctness issues." + "reason": "Embedding dtype drift and auxiliary-loss normalization are different training behaviors." }, { "left": "issue:43262", "right": "issue:43295", "accept": false, - "reason": "Audio chat-template sampling-rate default and processor.tokenizer/image-passing regression are separate processor API bugs." + "reason": "Audio chat-template sampling-rate default and processor.tokenizer/image handling regression are different processor bugs." }, { "left": "issue:44849", "right": "issue:45357", "accept": false, - "reason": "Qwen3.5 hidden-states bug and incorrect visual-encoder save_pretrained keys are unrelated." + "reason": "Qwen3.5 hidden-states bug and visual-encoder save_pretrained regression are not the same issue." }, { "left": "issue:43525", "right": "issue:43572", "accept": false, - "reason": "Both mention missing pad token config fields, but they affect different model configs and are not clearly the same underlying defect." + "reason": "Missing pad_token_id in Llama4Config and missing pad_token_idx in StableLmConfig are similar in theme but different concrete bugs." }, { "left": "issue:33666", "right": "issue:36331", "accept": false, - "reason": "Qwen2-VL multi-GPU training and CustomTrainer.compute_loss signature mismatch are separate trainer/model issues." + "reason": "Qwen2-VL multi-GPU training and compute_loss signature mismatch are unrelated trainer/model issues." }, { "left": "issue:43493", "right": "issue:43825", "accept": false, - "reason": "SigLIP2 implementation discrepancy and pipeline translation-task error messaging are unrelated." + "reason": "SigLIP2 implementation discrepancy and pipeline translation-message wording are unrelated." }, { "left": "issue:43720", - "right": "issue:43927", - "accept": false, - "reason": "BitNet packed-weight loading and DiaConfig token-ID persistence are different loading/serialization bugs." - }, - { - "left": "issue:43278", - "right": "issue:43688", - "accept": false, - "reason": "No shared bug: dtype mismatch in evaluation vs aux-loss normalization." - }, - { - "left": "issue:43262", - "right": "issue:43295", - "accept": false, - "reason": "Different processor regressions affecting different modalities and APIs." - }, - { - "left": "issue:44849", - "right": "issue:45357", + "right": "issue:44361", "accept": false, - "reason": "Hidden-states output bug is not the same as incorrect visual encoder key serialization." + "reason": "BitNet loading/unpacking and MLukeTokenizer task AttributeError do not share a code path." }, { - "left": "issue:44964", - "right": "issue:45405", + "left": "issue:44625", + "right": "issue:46002", "accept": false, - "reason": "Model-loading failure for Phi-4 multimodal and a dependency version bump are unrelated." + "reason": "num_labels propagation and classification-head NaNs are different bugs." }, { - "left": "issue:39692", - "right": "issue:43493", + "left": "issue:43756", + "right": "issue:44913", "accept": false, - "reason": "Doc example errors for SigLIP2 are not the same as the HF-vs-JAX model discrepancy." + "reason": "Smollm3 RoPE-layer count issue and GPTNeoX rotary_pct reload bug are not the same defect." }, { "left": "issue:44568", "right": "issue:45812", "accept": false, - "reason": "Both are tokenizer regressions in v5, but they affect different tokenizers/models and different observed failures." + "reason": "mdeberta special-token regression and Granite wrong token IDs are separate tokenizer bugs." }, { "left": "issue:28218", "right": "issue:33357", "accept": false, - "reason": "Tokenizer added-space behavior and a MacOS bus error in CLIP are unrelated." + "reason": "Tokenizer added-space bug and MacOS CLIP bus error are unrelated." }, { - "left": "issue:44913", - "right": "issue:44964", + "left": "issue:44964", + "right": "issue:45405", + "accept": false, + "reason": "Phi-4 multimodal load failure and unreleased MIN_PEFT_VERSION bump are unrelated." + }, + { + "left": "issue:39692", + "right": "issue:43493", "accept": false, - "reason": "GPTNeoX config reload bug and Phi-4 multimodal load failure are separate issues." + "reason": "SigLIP2 docs example errors and SigLIP2 implementation discrepancy are different issues: documentation vs model behavior." } ] }, @@ -6914,14 +7079,14 @@ "issue:44423|issue:44734", "issue:29942|issue:30333" ], - "prepared_review_unit_hash": "15d0ab926b753a3fbd8684b711309f1da923163fb8221d2327fa4451d31e1377", + "prepared_review_unit_hash": "2879e682be0b243af5041e7f813092ad7e27a6cdd0b4b03ed26ea5167b74024a", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 47, @@ -6936,180 +7101,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: the issues cover unrelated bugs across multimodal loading, tokenizers/configs, training/inference, CI, and model-specific regressions. None of the soft pairs look like the same underlying bug or change, so no duplicate consolidation is warranted.", - "confidence": 0.95, - "canonical_issue_reason": null, + "summary": "No true duplicates here: the soft pairs are mostly superficial title similarity across unrelated regressions, model-loading issues, tokenizer/config bugs, and CI/test failures. Treat the items as separate issues rather than one dedup cluster.", + "confidence": 0.89, + "canonical_issue_reason": "No single canonical issue stands out; the cluster is too heterogeneous to safely collapse into one duplicate representative.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "If forced to pick one representative, 'Transformers 5.0.0 breaks custom model initialization' is the broadest version-5 loading regression, but it still does not cover the rest of the items as duplicates.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43540", "right": "issue:44977", "accept": false, - "reason": "Different failure modes and code paths: Qwen3OmniMoe video input processing vs Qwen3.5 flash-attention generation." + "reason": "Different models and failure modes: video-input processing vs flash-attention generation." }, { "left": "issue:43493", "right": "issue:43844", "accept": false, - "reason": "Different model families and bugs: SigLIP2 implementation discrepancy vs ZeRO-3 gradient growth during DeepSpeed training." + "reason": "Both are training-related, but SigLIP2 implementation discrepancy is not the same bug as ZeRO-3 gradient growth." }, { "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. Not the same bug." + "reason": "Same model family, but missing text-only token defaults is unrelated to multi-GPU inference OOM." }, { "left": "issue:43278", "right": "issue:43844", "accept": false, - "reason": "Embedding dtype regression in eval vs training instability under HfDeepSpeedConfig/ZeRO-3 are unrelated." + "reason": "Embedding dtype drift during evaluate is a different code-path from DeepSpeed gradient instability." }, { "left": "issue:43334", "right": "issue:43572", "accept": false, - "reason": "Both are config attribute regressions, but for different models and different missing fields (pad_token_id vs pad_token_idx)." + "reason": "Both are config-field regressions, but they affect different models and different missing attributes." }, { "left": "issue:45803", "right": "issue:45987", "accept": false, - "reason": "Fuyu batch tensor conversion exception handling vs StaticCache return type contract are unrelated." + "reason": "Unrelated components: Fuyu tensor conversion vs StaticCache API return-type contract." }, { "left": "issue:44488", "right": "issue:44843", "accept": false, - "reason": "Different loading failures: a specific model load regression vs unconditional hub metadata lookup breaking offline mode." + "reason": "Both are loading bugs, but offline hub access and the mistral-regex patch are different root causes." }, { "left": "issue:43257", "right": "issue:43299", "accept": false, - "reason": "Both involve Qwen MoE, but one is accelerate+deepspeed conversion and the other is model loading in dev; not the same concrete defect." + "reason": "Related Qwen3 MoE area, but accelerate+deepspeed weight conversion is not the same as Qwen3VL loading failure." }, { "left": "issue:43756", "right": "issue:45083", "accept": false, - "reason": "RoPE layer mismatch in Smollm3 is unrelated to feature-extraction length behavior in qwen3_omni_moe." + "reason": "Different subsystems entirely: RoPE-layer handling vs feature-extraction length helper behavior." }, { "left": "issue:43334", "right": "issue:45759", "accept": false, - "reason": "Different multimodal config issues: missing pad_token_id vs from_config not unwrapping text_config." + "reason": "Both touch Qwen multimodal config loading, but missing pad_token_id and not unwrapping text_config are distinct bugs." }, { "left": "issue:39692", "right": "issue:43295", "accept": false, - "reason": "SigLIP2 doc/example errors are not the same as processor.tokenizer regression in custom model code." + "reason": "Both mention processor/tokenizer mismatch, but one is a docs/quantization example problem and the other is a runtime regression." }, { "left": "issue:45229", "right": "issue:45823", "accept": false, - "reason": "OOM from multi-GPU inference and device mismatch during forward are distinct issues." + "reason": "Same Gemma4 family, but CUDA OOM and device mismatch are different failure modes." }, { "left": "issue:43646", "right": "issue:44466", "accept": false, - "reason": "Custom model initialization regression vs lm_head.weight serialization inconsistency are different code paths." + "reason": "Both are v5 regressions, but custom model initialization is unrelated to lm_head serialization/tied-weight behavior." }, { "left": "issue:43824", "right": "issue:44589", "accept": false, - "reason": "ImportError for a missing model class vs missing Float8 storage type are unrelated." + "reason": "Different loading/storage issues with different modules and symptoms." }, { "left": "issue:33357", "right": "issue:33666", "accept": false, - "reason": "MacOS bus error on a CLIP model vs Qwen2-VL multi-GPU training is not the same bug." + "reason": "Both are multimodal issues, but a Mac bus error in CLIP is not the same as Qwen2-VL multi-GPU training." }, { "left": "issue:43504", "right": "issue:43606", "accept": false, - "reason": "Legacy field blocking Beit semantic segmentation load vs CPU offload device mismatch are different problems." + "reason": "Legacy-field load failure and CPU-offload device mismatch are separate bugs." }, { "left": "issue:44589", "right": "issue:45520", "accept": false, - "reason": "Float8 storage type resolution issue vs flash_attn import_utils KeyError on Python 3.13 are unrelated." + "reason": "Both are low-level import/storage errors, but they are different code-paths and causes." }, { "left": "issue:43572", "right": "issue:44291", "accept": false, - "reason": "Missing StableLmConfig field after 5.0 vs init_empty_weights argument mismatch are different regressions." + "reason": "Both are 5.0-era init regressions, but missing pad_token_idx and unexpected _is_hf_initialized argument are different issues." }, { "left": "issue:43927", "right": "issue:44361", "accept": false, - "reason": "DiaConfig save/load token ID loss and MLukeTokenizer AttributeError are separate tokenizer/config bugs." + "reason": "Token-ID persistence across save/load is unrelated to MLLukeTokenizer task-time AttributeError." }, { "left": "issue:44464", "right": "issue:45584", "accept": false, - "reason": "Compiled-forward chunked generation inconsistency vs whisper empty transcription failure after align_special_tokens are unrelated." + "reason": "Compiled-forward chunked generation and empty-transcription token alignment are different bugs." }, { "left": "issue:43606", "right": "issue:44112", "accept": false, - "reason": "One is a runtime device mismatch with CPU offload; the other is a stale CI test failure in GraniteSpeech." + "reason": "A real CPU-offload mismatch is not the same as a stale GraniteSpeech CI test." }, { "left": "issue:43504", "right": "issue:43653", "accept": false, - "reason": "Beit legacy-field load failure and BigBirdTokenizer special-token decode bug do not share the same underlying cause." + "reason": "Pretrained-model legacy field handling and BigBird special-token registration are unrelated." }, { "left": "issue:43575", "right": "issue:45405", "accept": false, - "reason": "TP OOM on a large Qwen model is unrelated to an unreleased PEFT version bump." + "reason": "Tensor-parallel OOM and unreleased dependency version bump are not duplicates." }, { "left": "issue:43278", "right": "issue:43701", "accept": false, - "reason": "Evaluation dtype change and resume_from_checkpoint key mismatch are different issues." + "reason": "Training/eval dtype drift is unrelated to resume_from_checkpoint key mismatch." }, { "left": "issue:44792", "right": "issue:45405", "accept": false, - "reason": "Janus image-generation test failure is unrelated to the PEFT version pin problem." + "reason": "janus image-generation test failure has nothing to do with the PEFT version bump." }, { "left": "issue:43994", "right": "issue:44493", "accept": false, - "reason": "SigLIP2 output quality issue vs generic unexpected position-id key warnings are not the same defect." + "reason": "SigLIP2 AutoModel/pipeline correctness and position-id warnings are different bugs." }, { "left": "issue:44423", "right": "issue:44734", "accept": false, - "reason": "Continuous-batching multimodal crash from string/.to misuse vs KV-cache continuation tensor indexing bug are different code paths." + "reason": "Both are serve crashes, but continuous-batching string handling and KV-cache indexing are distinct code-path failures." }, { "left": "issue:29942", "right": "issue:30333", "accept": false, - "reason": "Flash Attention 2 test failures and MLflow reporting after failed training jobs are unrelated." + "reason": "Completely unrelated: Flash Attention 2 tests vs MLflow reporting after training failure." } ] }, @@ -7214,214 +7379,214 @@ "issue:44466|issue:45127", "issue:43550|issue:43844" ], - "prepared_review_unit_hash": "0960e151b877f9be133a4d150f63cd99069d87f94443ec686253b8fa26962e46", + "prepared_review_unit_hash": "ad0c694f0fd3c8cee93ccce42744ad069c9acbfe1ca4d66f14ea4a43a03e11c2", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36892, - "estimated_input_tokens": 9223, - "estimated_eval_tokens": 18702 + "serialized_chars": 36893, + "estimated_input_tokens": 9224, + "estimated_eval_tokens": 18704 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The items are mostly unrelated across tokenization, loading, training, quantization, multimodal processing, and attention internals. None of the soft pairs look like true duplicate-level matches; they share broad subsystem names but not the same concrete failure path.", - "confidence": 0.88, - "canonical_issue_reason": "issue:43873 is the strongest representative in this set: it has the most discussion/inbound refs, a clear user-facing quantization/offloading failure, and is still open; the other items are mostly orthogonal bugs.", + "summary": "This is a heterogeneous false-positive cluster: the items span unrelated Transformers bugs across tokenization, generation, loading, multimodal processors, quantization, and training. I do not see any soft pair that looks like the same underlying bug/change.", + "confidence": 0.96, + "canonical_issue_reason": "No single canonical issue is safe to choose here; the cluster is too mixed and the soft links do not describe true duplicates.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43873 is the best overall issue candidate because it is active, impactful, and well-scoped, while the rest of the list is heterogeneous and less suitable as a canonical representative.", + "best_issue_reason": "No best issue stands out as a representative duplicate target because the items cover several unrelated failure modes.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43873", "right": "issue:46050", "accept": false, - "reason": "Both involve quantization, but one is about offloading behavior and the other about save/load breakage after sinq quantization; different bug paths." + "reason": "Both involve quantization/offloading, but one is an offloading behavior issue and the other is broken save/load after sinq quantization." }, { "left": "issue:45406", "right": "issue:46121", "accept": false, - "reason": "Gemma4Processor missing _tokenizer is unrelated to rope-parameter JSON list handling; different components and failure modes." + "reason": "Gemma4 processor `_tokenizer` access and rope-validation list handling are unrelated bugs in different code paths." }, { "left": "issue:45588", "right": "issue:46050", "accept": false, - "reason": "Flash-attention crash with s_aux=None is a runtime integration bug, not the same as quantized model serialization/loading corruption." + "reason": "A flash-attention `s_aux=None` crash is unrelated to quantized model persistence after save/load." }, { "left": "issue:42490", "right": "issue:43696", "accept": false, - "reason": "Local-path import warning and CUDA OOM are unrelated symptoms with no shared underlying code path." + "reason": "A trailing-slash import warning and a CUDA OOM are clearly unrelated." }, { "left": "issue:44521", "right": "issue:45290", "accept": false, - "reason": "Both touch apply_chat_template, but one is about multimodal assistant masks and the other is a crash on tool-call messages with no content; distinct bugs." + "reason": "Both touch chat templating, but one is multimodal assistant-mask generation and the other is tool-call/no-content tokenization handling." }, { "left": "issue:43844", "right": "issue:44964", "accept": false, - "reason": "ZeRO-3 gradient growth and Phi-4 multimodal load failure are different training/loading issues." + "reason": "ZeRO-3 gradient growth and a Phi-4 multimodal loading failure are unrelated." }, { "left": "issue:28218", "right": "issue:34567", "accept": false, - "reason": "Tokenizer spacing after added tokens is unrelated to TrainerState.num_input_tokens_seen not updating." + "reason": "Tokenizer spacing after added tokens does not match the TrainerState token-count update bug." }, { "left": "issue:29159", "right": "issue:32090", "accept": false, - "reason": "Tokenizer slow/fast inconsistency and a NoneType broadcast error in Trainer are different subsystems and failure signatures." + "reason": "Slow/fast tokenizer inconsistency is unrelated to the NoneType broadcast error in Trainer." }, { "left": "issue:45084", "right": "issue:45520", "accept": false, - "reason": "Template-node compilation TypeError is unrelated to the flash_attn import KeyError on Python 3.13." + "reason": "Template-node compilation failure and a Python 3.13 flash_attn import KeyError are unrelated." }, { "left": "issue:44554", "right": "issue:45446", "accept": false, - "reason": "MPS attention correctness and a PyTorch version gate for flex_attention imports are unrelated." + "reason": "An MPS attention correctness issue is unrelated to the flex_attention PyTorch version check." }, { "left": "issue:43582", "right": "issue:44977", "accept": false, - "reason": "Apple Silicon warmup TypeError and Qwen3.5 flash-attention generation bug are different runtime failures." + "reason": "Apple Silicon allocator warmup TypeError and Qwen3.5 flash-attention generation bugs are unrelated." }, { "left": "issue:43540", "right": "issue:45200", "accept": false, - "reason": "Video-input processing failure in Qwen3OmniMoe is unrelated to Gemma4 text-only mm_token_type_ids handling." + "reason": "Video-input handling in Qwen3OmniMoe is unrelated to Gemma 4 text-only `mm_token_type_ids` defaults." }, { "left": "issue:15354", "right": "issue:38175", "accept": false, - "reason": "TorchScript GeneratorExp export failure and SigLIP2 zero-probability outputs are clearly different problems." + "reason": "TorchScript GeneratorExp export failure and SigLIP2 zero probabilities are unrelated." }, { "left": "issue:45200", "right": "issue:45823", "accept": false, - "reason": "Gemma4 text-only token-type defaults and PLE device mismatch during forward are different model-behavior bugs." + "reason": "Gemma4 text-only defaults and Gemma4 device-map forward mismatch are different bugs in different code paths." }, { "left": "issue:43232", "right": "issue:44849", "accept": false, - "reason": "Generation kwarg update after sync_gpus and Qwen3.5 hidden-state output bug do not share the same code path." + "reason": "Generation kwargs after `sync_gpus` and Qwen3.5 hidden-state output are unrelated generation issues." }, { "left": "issue:45072", "right": "issue:45198", "accept": false, - "reason": "Dtype mismatches in CI and Wav2Vec2 save_pretrained/tokenization failure are unrelated." + "reason": "dtype mismatches in bfloat16 inference are unrelated to Wav2Vec2 save/tokenization failures." }, { "left": "issue:37428", "right": "issue:43957", "accept": false, - "reason": "Flash-attention import error and meta-device model loading breakage are both loading-related but stem from different mechanisms and fixes." + "reason": "A missing flash-attention import and meta-device model loading failures are different problems." }, { "left": "issue:43493", "right": "issue:43701", "accept": false, - "reason": "SigLIP2 HF/JAX discrepancy and resume_from_checkpoint key mismatch are unrelated issues." + "reason": "SigLIP2 implementation discrepancy is unrelated to checkpoint key mismatches on resume." }, { "left": "issue:44315", "right": "issue:44849", "accept": false, - "reason": "Liger Kernel not applied with model_init is unrelated to the Qwen3.5 output_hidden_states regression." + "reason": "Liger Kernel not applied in `model_init` and Qwen3.5 hidden-state output are unrelated." }, { "left": "issue:43295", "right": "issue:44610", "accept": false, - "reason": "Processor.tokenizer regression is unrelated to OmDet-Turbo producing the wrong image size." + "reason": "Processor/tokenizer regression and OmDet-Turbo image-size mismatch are different processor bugs." }, { "left": "issue:43572", "right": "issue:44987", "accept": false, - "reason": "StableLmConfig missing pad_token_idx and physical-intelligence/fast load failure are both loading regressions but not the same bug." + "reason": "Missing `pad_token_idx` in StableLmConfig and a transformers>=5.1 loading failure are unrelated." }, { "left": "issue:43295", "right": "issue:43450", "accept": false, - "reason": "Processor.tokenizer regression is not the same as batched video processor shape output." + "reason": "Processor/tokenizer regression is unrelated to video processor batch-shape errors." }, { "left": "issue:44554", "right": "issue:45356", "accept": false, - "reason": "MPS attention correctness issue and Kimi-K2.5 tokenizer codec regression are unrelated." + "reason": "An MPS attention correctness bug is unrelated to a Kimi-K2.5 tokenizer regression." }, { "left": "issue:43572", "right": "issue:43957", "accept": false, - "reason": "Config missing pad_token_idx and meta-device model loading failures are different regressions." + "reason": "A config default bug and a model-loading meta-device regression are not the same issue." }, { "left": "issue:37428", "right": "issue:43824", "accept": false, - "reason": "Flash-attention symbol import failure is unrelated to missing Qwen2_5_VLForConditionalGeneration export." + "reason": "Both are import errors, but they involve different missing symbols and different root causes." }, { "left": "issue:45200", "right": "issue:45325", "accept": false, - "reason": "Gemma4 mm_token_type_ids defaults and Qwen2.5-VL rope-index scaling are different multimodal modeling bugs." + "reason": "Gemma4 `mm_token_type_ids` defaults and Qwen2.5-VL rope-index scaling are unrelated multimodal bugs." }, { "left": "issue:43295", "right": "issue:43873", "accept": false, - "reason": "Processor.tokenizer regression and quantization offloading failure are unrelated despite both touching model setup." + "reason": "Processor/tokenizer regression is unrelated to quantization offloading behavior." }, { "left": "issue:44898", "right": "issue:45412", "accept": false, - "reason": "Perceiver non-default resolution failure and RT-DETR memory leak are different issues." + "reason": "Perceiver resolution handling and RT-DETR memory release are unrelated." }, { "left": "issue:44466", "right": "issue:45127", "accept": false, - "reason": "Both mention tied weights, but one is about inconsistent serialization by device and the other is LoRA merge collapse with extended vocab; not the same concrete bug." + "reason": "Tied-weight serialization differences and LoRA merge collapse are different bugs despite both involving weights." }, { "left": "issue:43550", "right": "issue:43844", "accept": false, - "reason": "Bamba torch.compile SDPA failure and ZeRO-3 gradient growth are distinct training/runtime problems." + "reason": "torch.compile SDPA failure and ZeRO-3 gradient growth are unrelated training/runtime issues." } ] }, @@ -7525,208 +7690,214 @@ "issue:44448|issue:44743", "issue:43278|issue:43825" ], - "prepared_review_unit_hash": "f4e31586fa3c91b790c9b3947ed428cbf09715e8cae626e8e96f445c8d76bde6", + "prepared_review_unit_hash": "f7b3011ff890841b68050d1c2ca734e59954fecac20c9293dcc17e0263afb650", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36539, - "estimated_input_tokens": 9135, - "estimated_eval_tokens": 18526 + "serialized_chars": 36535, + "estimated_input_tokens": 9134, + "estimated_eval_tokens": 18524 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "All soft pairs look like false positives: they share broad vocabulary or a subsystem, but not the same concrete bug/change. The set is heterogeneous, so there is no safe canonical issue or PR here.", - "confidence": 0.94, - "canonical_issue_reason": null, + "summary": "The set is mostly a grab bag of unrelated Transformer bugs and regressions. None of the soft pairs are strong duplicates under a strict same-root-cause standard. No PRs are present; issue 44521 is the best single issue to anchor on because it has the broadest, most active bug report among the items.", + "confidence": 0.89, + "canonical_issue_reason": "Issue 44521 has the richest discussion/inbound references and a broad, clearly stated bug in a widely used API, making it the best anchor despite the cluster being heterogeneous.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "44521 is the most representative and actionable issue in the set; it is active, well-described, and general enough to serve as the closest canonical issue.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44792", "right": "issue:45563", "accept": false, - "reason": "Different failures: a janus model test case vs a paged generate warning about num_return_sequences." + "reason": "Both involve generation-related behavior, but one is a Janus image-generation test and the other is a stale paging warning; different code paths and symptoms." }, { "left": "issue:43334", "right": "issue:43525", "accept": false, - "reason": "Both are missing-pad-token_id errors, but for different models/config classes (Qwen3-VL vs Llama4) and likely different code paths." + "reason": "Same missing-attribute symptom, but for different config classes/models; no evidence they share the same underlying bug." }, { "left": "issue:43824", "right": "issue:43975", "accept": false, - "reason": "ImportError for Qwen2.5-VL class vs a detokenization bug on deepseek-coder; unrelated symptoms and components." + "reason": "An import error for a Qwen VL class and a detokenization regression in DeepSeek are unrelated." }, { "left": "issue:42915", "right": "issue:43824", "accept": false, - "reason": "Qwen3MoE FP8 config failure is unrelated to the Qwen2.5-VL import failure." + "reason": "FineGrainedFP8Config failure for Qwen3Moe is unrelated to a missing import for a Qwen VL model class." }, { "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 behavior." + "reason": "Both concern labels, but one is evaluation batch gathering and the other is BEiT image preprocessing; different bug surfaces." }, { "left": "issue:43450", "right": "issue:43575", "accept": false, - "reason": "Video batched-shape bug vs Qwen2-57B-A14B-Instruct tensor-parallel OOM; not the same issue." + "reason": "Batched video processor shape handling and tensor-parallel OOM are different failure modes in different areas." }, { "left": "issue:42490", "right": "issue:43493", "accept": false, - "reason": "Local-path trailing-slash import warning is unrelated to SigLIP2 implementation discrepancy." + "reason": "Local-path import warning and SigLIP2 parity discrepancy are unrelated." }, { "left": "issue:43278", "right": "issue:43856", "accept": false, - "reason": "Embedding dtype regression in eval vs Qwen3 MoE memory usage; different model/runtime problems." + "reason": "Dtype drift in evaluate and MoE memory usage are different issues." }, { "left": "issue:43975", "right": "issue:45520", "accept": false, - "reason": "Detokenization bug vs flash_attn import KeyError on Python 3.13; unrelated." + "reason": "Detokenization regression and a Python 3.13 flash_attn import KeyError are unrelated." }, { "left": "issue:43653", "right": "issue:44112", "accept": false, - "reason": "Tokenizer special-token registration bug vs a stale GraniteSpeech CI test; not the same code-path." + "reason": "Tokenizer special-token registration and a stale CI device-override test do not describe the same bug." }, { "left": "issue:28218", "right": "issue:35141", "accept": false, - "reason": "Tokenizer added-space behavior is unrelated to embedding reinitialization during post_init." + "reason": "Tokenizer spacing after added tokens and embedding reinitialization after resize are distinct tokenizer/model-state issues." }, { "left": "issue:38175", "right": "issue:43493", "accept": false, - "reason": "Same model family, but zero probabilities vs HF/JAX discrepancy are not the same concrete bug." + "reason": "Both are about SigLIP2, but one is zero probabilities and the other is a general HF-vs-JAX discrepancy; not enough to treat as the same defect." }, { "left": "issue:44442", "right": "issue:45198", "accept": false, - "reason": "Different tokenizer/loading failures for different model families; no shared root cause shown." + "reason": "Different audio-tokenizer/model workflows with different failures; no shared code path is evident." }, { "left": "issue:44514", "right": "issue:44521", "accept": false, - "reason": "Both touch apply_chat_template, but one crashes on batched padding=False input while the other returns all-zero assistant masks for multimodal inputs." + "reason": "Both touch apply_chat_template, but one crashes on batched padding=False input while the other produces all-zero assistant masks for multimodal inputs; not the same bug." }, { "left": "issue:38175", "right": "issue:45412", "accept": false, - "reason": "SigLIP2 output issue and RT-DETR memory leak are unrelated." + "reason": "SigLIP2 probability issues and RT-DETR memory-release behavior are unrelated." }, { "left": "issue:44206", "right": "issue:45584", "accept": false, - "reason": "FeatureExtractor center-arg crash vs Whisper empty-transcription failure; different components and failures." + "reason": "A feature extractor unsupported-arg crash and Whisper empty-transcription generation failure are separate problems." }, { "left": "issue:43827", "right": "issue:43901", "accept": false, - "reason": "Both are docs-related pipeline issues, but they concern different APIs and different broken guidance." + "reason": "Both are doc mismatches, but they cover different pipeline APIs and different stale behavior; not one underlying change." }, { "left": "issue:43262", "right": "issue:43450", "accept": false, - "reason": "Audio chat-template sampling-rate default vs video processor batch-shape bug; unrelated." + "reason": "Audio chat-template sampling-rate defaulting and video processor batched-shape handling are unrelated." }, { "left": "issue:43819", "right": "issue:44112", "accept": false, - "reason": "DAC latent/STE mismatch vs GraniteSpeech CI test staleness; no common bug." + "reason": "DAC latent mismatch and a GraniteSpeech CI test issue are different code paths." }, { "left": "issue:44610", "right": "issue:44857", "accept": false, - "reason": "Processor output resolution mismatch vs AMP/CUDA loss crash; different defects." + "reason": "Processor image-size mismatch and AMP crash in a different loss are distinct." }, { "left": "issue:43819", "right": "issue:45083", "accept": false, - "reason": "DAC forward/from_latents mismatch is unrelated to qwen3_omni_moe feature-length helper behavior." + "reason": "DAC latent conversion and qwen3_omni_moe feature-extraction length behavior are unrelated." }, { "left": "issue:37428", "right": "issue:43425", "accept": false, - "reason": "Flash-attention import error and Torch 2.10 incompatibility are separate compatibility problems." + "reason": "An import error for one helper is not the same as broad Torch 2.10 incompatibility." }, { "left": "issue:43335", "right": "issue:43550", "accept": false, - "reason": "SwitchTransformers sparse-layer creation bug vs Bamba SDPA torch.compile failure; different subsystems." + "reason": "A sparse-layer config bug and a torch.compile/SDPA failure in Bamba are different issues." }, { "left": "issue:43606", "right": "issue:45083", "accept": false, - "reason": "CPU offload device mismatch on bark-small is unrelated to qwen3_omni_moe length calculation." + "reason": "CPU-offload device mismatch and qwen3_omni_moe feature-length behavior are unrelated." }, { "left": "issue:44857", "right": "issue:45593", "accept": false, - "reason": "Both are detection-related, but one is an AMP/CUDA crash and the other is missing auxiliary losses when denoising is off." + "reason": "A float16 AMP crash and missing auxiliary losses are both detection-related but clearly different defects." }, { "left": "issue:45127", "right": "issue:45245", "accept": false, - "reason": "LoRA+extended-vocab collapse with tied embeddings has nothing to do with the 2^24 category limit error." + "reason": "LoRA/extended-vocab collapse and a category-count runtime limit are unrelated." }, { "left": "issue:45290", "right": "issue:45381", "accept": false, - "reason": "apply_chat_template tool-call crash is unrelated to qwen2.5-vl video vision_position_ids." + "reason": "Assistant tool-call chat-template crash and VL video position-id bug are different multimodal issues." }, { "left": "issue:44448", "right": "issue:44743", "accept": false, - "reason": "Pegasus output regression vs qwen3_5 cache/recurrent-state reset bug; unrelated model paths." + "reason": "Pegasus v4/v5 output drift and Qwen3_5 recurrent-state reset are unrelated." + }, + { + "left": "issue:43824", + "right": "issue:45083", + "accept": false, + "reason": "A missing Qwen2.5-VL import and qwen3_omni_moe feature-length behavior are unrelated." }, { "left": "issue:43278", "right": "issue:43825", "accept": false, - "reason": "Embedding dtype regression and a pipeline error-message/docs issue are unrelated." + "reason": "Embedding dtype drift and a pipeline error-message regression are unrelated." } ] }, @@ -7837,256 +8008,256 @@ "issue:44462|issue:44561", "issue:43550|issue:45083" ], - "prepared_review_unit_hash": "a9c692ec333d2d7412ec1248c10c6e079ba29b8909970df463db0148cecbf1a3", + "prepared_review_unit_hash": "ff18e0df2cba74b745b002ce2a94aea40dabb7d261ee2789bdc40e923860a87d", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 36, - "serialized_chars": 38451, - "estimated_input_tokens": 9613, - "estimated_eval_tokens": 19482 + "serialized_chars": 38447, + "estimated_input_tokens": 9612, + "estimated_eval_tokens": 19480 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is almost entirely a grab-bag of unrelated Transformers bug reports. The only clear duplicate pair is the two identical Qwen3.5 MoE `from_pretrained` regression reports.", - "confidence": 0.95, - "canonical_issue_reason": "issue:45310 is the best canonical issue because it is the earlier report of the exact Qwen3.5 MoE `from_pretrained` regression and matches the duplicate candidate 45478 nearly verbatim.", + "summary": "This set is mostly unrelated issues; the only clear duplicate pair is the Qwen3.5 MoE from_pretrained regression (45310/45478). No PRs are present.", + "confidence": 0.86, + "canonical_issue_reason": "issue:45310 is the earliest and slightly more detailed report for the only true duplicate pair in the set, so it is the best canonical issue to anchor that regression.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45310 is the strongest representative issue in the cluster because it is a clean, specific duplicate target with the clearest title and earliest filing.", + "best_issue_reason": "issue:45310 is the clearest global anchor here: it matches its duplicate exactly, names the concrete regression, and is the earliest report among the pair.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:29942", "right": "issue:32090", "accept": false, - "reason": "Different failures: Flash Attention 2 test regressions vs a Trainer `_gpu_broadcast_one` NoneType error." + "reason": "Different bugs: Flash Attention 2 test failures vs Trainer/_gpu_broadcast_one NoneType broadcast error." }, { "left": "issue:45200", "right": "issue:45588", "accept": false, - "reason": "Different subsystems and symptoms: Gemma 4 token type IDs vs a flash_attention `s_aux=None` crash." + "reason": "Different code paths and failures: Gemma4 token-type IDs vs flash_attention.py AttributeError on s_aux=None." }, { "left": "issue:37428", "right": "issue:44351", "accept": false, - "reason": "Both are import errors, but they name different missing symbols and different code paths." + "reason": "Both are import errors, but for different missing symbols and likely different underlying breakages." }, { "left": "issue:43611", "right": "issue:43931", "accept": false, - "reason": "Both are model-loading issues, but one is about `base_model_prefix` handling and the other is a Qwen3-VL weight-shape mismatch." + "reason": "Model-loading regression with base_model_prefix is not the same as a Qwen3-VL weight shape mismatch." }, { "left": "issue:44206", "right": "issue:45823", "accept": false, - "reason": "Different bugs: an unsupported `center` argument in a feature extractor versus a device mismatch in Gemma4 PLE forward." + "reason": "Unsupported center arg in a feature extractor is unrelated to Gemma4 PLE device mismatch during forward." }, { "left": "issue:43335", "right": "issue:43653", "accept": false, - "reason": "Unrelated model/config bug vs tokenizer special-token registration/empty decode issue." + "reason": "SwitchTransformers sparse-layer config bug and BigBirdTokenizer special-token bug are unrelated." }, { "left": "issue:43550", "right": "issue:44112", "accept": false, - "reason": "Different areas: torch.compile/SDPA behavior in Bamba vs a stale CI device override test." + "reason": "torch.compile/SDPA failure in Bamba is unrelated to a GraniteSpeech CI device-override test issue." }, { "left": "issue:43819", "right": "issue:44493", "accept": false, - "reason": "Different underlying problems: DAC `from_latents` mismatch vs widespread unexpected key/position-id behavior." + "reason": "DAC.from_latents/STE mismatch is a different bug than unexpected position-id keys in many models." }, { "left": "issue:44855", "right": "issue:45083", "accept": false, - "reason": "Python 3.13 import parsing issue in DebertaV2 vs a qwen3_omni_moe helper behavior regression." + "reason": "Python 3.13 IndentationError in DebertaV2Model is unrelated to qwen3_omni_moe length helper behavior." }, { "left": "issue:43335", "right": "issue:43504", "accept": false, - "reason": "Both involve Switch/Beit family issues, but the concrete failures and affected code paths differ." + "reason": "Different BEIT/SwitchTransformers model-loading/configuration problems." }, { "left": "issue:45588", "right": "issue:45859", "accept": false, - "reason": "Both mention Qwen3.5/MoE and attention-related code, but one is a sink-less flash_attention crash and the other is CPU-offload incompatibility in rotary embedding." + "reason": "Both involve Qwen-related attention code, but one is a flash_attention s_aux=None crash and the other is CPU-offload rotary embedding incompatibility." }, { "left": "issue:29127", "right": "issue:32090", "accept": false, - "reason": "Completely different bugs: LayoutLMv3 error-message clarity vs Trainer broadcast failure." + "reason": "LayoutLMv3 error messaging and Trainer GPU broadcast TypeError are unrelated." }, { "left": "issue:43335", "right": "issue:45083", "accept": false, - "reason": "Different components and failure modes: SwitchTransformers config creation vs qwen3_omni_moe feature-length helper behavior." + "reason": "SwitchTransformers sparse-layer creation and qwen3_omni_moe feature-length helper are different bugs." }, { "left": "issue:39692", "right": "issue:43526", "accept": false, - "reason": "SigLIP2 doc-example problems are unrelated to BeitImageProcessorFast `reduce_labels` returning the wrong shape/value." + "reason": "SigLIP2 docs/example issues are unrelated to BeitImageProcessorFast reduce_labels behavior." }, { "left": "issue:45310", "right": "issue:45478", "accept": true, - "reason": "Same issue: identical title, same Qwen3.5 MoE `from_pretrained` error, and clearly the same regression report." + "reason": "Same concrete regression, same title, same version range, and 45478 is a later duplicate report of 45310." }, { "left": "issue:39692", "right": "issue:44493", "accept": false, - "reason": "Documentation/example failures are unrelated to the unexpected position-id key behavior." + "reason": "Documentation/example errors in SigLIP2 are unrelated to position-id key warnings." }, { "left": "issue:44360", "right": "issue:44561", "accept": false, - "reason": "Different changes: DSA indexer activation bug vs v5 removal of `is_torch_fx_available` breaking `trust_remote_code` models." + "reason": "DSA indexer ReLU discussion is unrelated to removal of is_torch_fx_available breaking trust_remote_code models." }, { "left": "issue:43720", "right": "issue:44112", "accept": false, - "reason": "Different code paths: BitNet packed-weight loading vs GraniteSpeech CI device override staleness." + "reason": "Packed-weights loading failure in BitNet is unrelated to a stale CI device-override test." }, { "left": "issue:44292", "right": "issue:45520", "accept": false, - "reason": "Different issues: Qwen-3-8B-NVFP4 runtime error vs Python 3.13 `flash_attn` import key error." + "reason": "Qwen-3-8B-NVFP4 runtime error is different from the Python 3.13 flash_attn import KeyError." }, { "left": "issue:43335", "right": "issue:44112", "accept": false, - "reason": "Unrelated: SwitchTransformers sparse-layer config bug versus a CI test/device override failure." + "reason": "Different subsystems and failure modes: SwitchTransformers config bug vs GraniteSpeech CI device mismatch." }, { "left": "issue:43335", "right": "issue:45072", "accept": false, - "reason": "Different bugs: sparse-layer creation in SwitchTransformers vs dtype mismatches in inference." + "reason": "Sparse-layer config and bfloat16 dtype mismatch are unrelated." }, { "left": "issue:44448", "right": "issue:44779", "accept": false, - "reason": "Different models and failure modes: Pegasus output drift vs Deepseek tokenizer correctness regression." + "reason": "Pegasus output drift across versions is not the same as Deepseek tokenizer regression." }, { "left": "issue:42490", "right": "issue:43854", "accept": false, - "reason": "One is a local-path import warning; the other is a GLM-4.7-Flash loading failure in tests." + "reason": "Local-path import warning and GLM-4.7-Flash loading failure are distinct model-loading issues." }, { "left": "issue:44977", "right": "issue:45229", "accept": false, - "reason": "Different symptoms: flash-attention generation breakage vs Gemma4 multi-GPU CUDA OOM." + "reason": "Qwen3.5 flash-attention generation bug and Gemma4 multi-GPU CUDA OOM are different problems." }, { "left": "issue:38175", "right": "issue:44493", "accept": false, - "reason": "SigLIP2 zero-probability output issue is unrelated to the position-id key regression." + "reason": "SigLIP2 zero probabilities and unexpected position-id keys are unrelated." }, { "left": "issue:43606", "right": "issue:44857", "accept": false, - "reason": "Different runtime problems: CPU offload device mismatch vs float16 AMP crash in LwDetrImageLoss." + "reason": "CPU-offload device mismatch in bark-small is unrelated to LwDetrImageLoss AMP crash." }, { "left": "issue:34689", "right": "issue:35141", "accept": false, - "reason": "Both are loading/model-init issues, but one is a Llama 3.2 Vision breakage and the other is embedding reinitialization after resize." + "reason": "Llama 3.2 Vision model-loading regression and embedding reinitialization after resize are different bugs." }, { "left": "issue:44857", "right": "issue:45072", "accept": false, - "reason": "Different failures: AMP/CUDA loss crash vs bfloat16 dtype mismatch in inference." + "reason": "float16 AMP crash in LwDetrImageLoss is unrelated to bfloat16 dtype mismatches." }, { "left": "issue:44448", "right": "issue:44977", "accept": false, - "reason": "Different models and root causes: Pegasus v4/v5 output differences vs Qwen3.5 flash-attention generation failure." + "reason": "Pegasus v4/v5 output differences and Qwen3.5 flash-attention generation failures are unrelated." }, { "left": "issue:40444", "right": "issue:43526", "accept": false, - "reason": "Different areas: multi-image IterableDataset finetuning failure vs BeitImageProcessorFast label reduction bug." + "reason": "IterableDataset multi-image finetuning failure is unrelated to BeitImageProcessorFast label reduction." }, { "left": "issue:28218", "right": "issue:32090", "accept": false, - "reason": "Tokenizer added-space behavior is unrelated to the Trainer `_gpu_broadcast_one` error." + "reason": "Tokenizer added-token spacing bug is unrelated to the Trainer broadcast NoneType error." }, { "left": "issue:43701", "right": "issue:45083", "accept": false, - "reason": "Resume-from-checkpoint key mismatch is unrelated to the qwen3_omni_moe helper behavior issue." + "reason": "Checkpoint key mismatch is unrelated to qwen3_omni_moe helper output lengths." }, { "left": "issue:44561", "right": "issue:44779", "accept": false, - "reason": "Different changes: v5 removal of a helper breaking remote-code models vs Deepseek tokenizer output regression." + "reason": "Both mention version 5 regressions, but one is an API removal breakage and the other is tokenizer output drift." }, { "left": "issue:43723", "right": "issue:44554", "accept": false, - "reason": "Tokenizer loading in v5 is unrelated to the MPS attention correctness issue." + "reason": "AutoTokenizer loading issue in v5 is unrelated to an MPS attention correctness problem." }, { "left": "issue:44462", "right": "issue:44561", "accept": false, - "reason": "AutoTokenizer ignoring `tokenizer.json` is a different bug from the removed-helper regression." + "reason": "Ignoring tokenizer.json and removal of is_torch_fx_available are different regression classes." }, { "left": "issue:43550", "right": "issue:45083", "accept": false, - "reason": "Unrelated: Bamba torch.compile/SDPA failure versus qwen3_omni_moe feature-length helper behavior." + "reason": "Bamba SDPA compile failure is unrelated to qwen3_omni_moe feature-length helper behavior." } ] }, "evaluator_result": { "accept": true, - "feedback": "Grounded overall: the packet supports one clear duplicate pair (45310/45478) and the rest look like unrelated issues. The soft-edge rejections are conservative and avoid overstating similarity. The canonical issue choice and earliest-filing rationale are well supported by the identical titles." + "feedback": "The summary is grounded in the packet: all items are issues, and 45310/45478 is the only exact-title duplicate pair with the earlier report correctly identified as canonical. The soft-edge verdicts are conservative and the distinctions are generally supported by the titles/body metadata." }, "evaluator_used": true, "retried": false, @@ -8189,226 +8360,220 @@ "issue:44871|issue:44977", "issue:44361|issue:45198" ], - "prepared_review_unit_hash": "351d6d13d3451184654cff5da0b4eb78232dcfcc53bd7cf8b87c7ba300e01efc", + "prepared_review_unit_hash": "acc87e52515044ec5c2a808ca31bef2fbe61d739d9ce258a2d6b9b267a555248", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 37307, - "estimated_input_tokens": 9327, - "estimated_eval_tokens": 18910 + "serialized_chars": 37309, + "estimated_input_tokens": 9328, + "estimated_eval_tokens": 18912 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is highly heterogeneous: the soft pairs share broad keywords or subsystems, but they do not describe the same concrete bug or fix. No PRs are present. I\u2019d treat this as a loose topical cluster rather than duplicate issues.", - "confidence": 0.94, - "canonical_issue_reason": "issue:43873 is the broadest, most central report in the set by activity and impact, but the cluster is not truly duplicate-heavy, so it is only a pragmatic representative.", + "summary": "The cluster is largely heterogeneous: these are distinct bug reports across different models, components, and failure modes. I rejected all soft pairs as true duplicates; none look like the same underlying bug or change.", + "confidence": 0.95, + "canonical_issue_reason": "If one issue must represent the cluster, issue #43873 is the most active and broadly framed (open, multiple references, discussion-heavy). That said, it is not a true semantic canonical for the whole set because the cluster is not actually cohesive.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43873 has the strongest discussion/inbound references and a clearly reproducible, user-visible failure, making it the best representative issue among these unrelated reports.", + "best_issue_reason": "Issue #43873 is the best global representative only by breadth and activity, not because it matches the other issues semantically.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:32090", "right": "issue:33357", "accept": false, - "reason": "Different failures: Trainer broadcast TypeError vs MacOS bus error on a CLIP model; no shared code-path bug." + "reason": "Different failures: Trainer/_gpu_broadcast_one TypeError vs MacOS bus error on a CLIP model." }, { "left": "issue:29942", "right": "issue:35141", "accept": false, - "reason": "Flash Attention test failures and embedding reinitialization are unrelated issues in different model paths." + "reason": "Flash Attention 2 test failures are unrelated to token embedding reinitialization in post_init." }, { "left": "issue:37428", "right": "issue:44987", "accept": false, - "reason": "Both are load/import failures, but one is a missing flash-attention helper import and the other is a model-loading regression; not the same bug." + "reason": "Import error for a flash-attention utility is unrelated to a model-loading failure for physical-intelligence/fast." }, { "left": "issue:43873", "right": "issue:45229", "accept": false, - "reason": "Both mention memory/resource problems, but offloading+quantization and multi-GPU OOM are different concrete failures and fixes." + "reason": "Quantization/offloading behavior and Gemma4 multi-GPU OOM are different code paths and symptoms." }, { "left": "issue:43927", "right": "issue:45698", "accept": false, - "reason": "Both involve save/load behavior, but one loses custom token IDs and the other loads the wrong custom module; distinct root causes." + "reason": "Saving/loading custom token IDs is a different bug from loading the wrong custom module after save_pretrained." }, { "left": "issue:41720", "right": "issue:45229", "accept": false, - "reason": "Different model-specific runtime issues: cudaErrorAssert during device mapping vs plain CUDA OOM." + "reason": "Auto device mapping cudaErrorAssert on Qwen3 is distinct from Gemma4 CUDA OOM." }, { "left": "issue:43278", "right": "issue:43329", "accept": false, - "reason": "Embedding dtype drift and an undefined-variable bug in multimodal token counting are unrelated." + "reason": "Embedding dtype drift in eval is unrelated to a multimodal token counting bug in the video branch." }, { "left": "issue:44977", "right": "issue:45563", "accept": false, - "reason": "Flash-attention generation regression and a stale warning in paged generate() are not the same code-path problem." + "reason": "Qwen3.5 flash-attention generation regression is not the same as a stale num_return_sequences warning." }, { "left": "issue:44568", "right": "issue:45491", "accept": false, - "reason": "Tokenizer special-token handling and Gemma3 NaN embeddings under sliding-window attention are different bugs." + "reason": "Tokenizer BOS/EOS handling is unrelated to Gemma3 mixed-length NaN embeddings from attention masking." }, { "left": "issue:10000", "right": "issue:15354", "accept": false, - "reason": "German DistilBertModel error and a torch.jit.script GeneratorExp export problem are unrelated." + "reason": "German DistilBertModel issue and torch.jit.script GeneratorExp export failure are different model/runtime problems." }, { "left": "issue:45372", "right": "issue:45588", "accept": false, - "reason": "Gemma4 processor import failure and flash_attention.py crashing on s_aux=None are separate integration bugs." + "reason": "Missing ReasoningEffort import for Gemma4 processor loading is unrelated to an AttributeError in flash_attention.py on s_aux=None." }, { "left": "issue:42490", "right": "issue:43575", "accept": false, - "reason": "A trailing-slash path warning and a Qwen2 tensor-parallel OOM are not the same underlying issue." + "reason": "Trailing-slash local path import warning and Qwen2-57B-A14B-Instruct TP OOM are unrelated." }, { "left": "issue:43412", "right": "issue:45588", "accept": false, - "reason": "Executorch export guards/erfinv support and sink-less flash-attention AttributeError are unrelated." + "reason": "Gemma3n executorch export failures and flash_attention.py s_aux=None crashes affect different pipelines." }, { "left": "issue:45198", "right": "issue:45803", "accept": false, - "reason": "Wav2Vec2 save/tokenization failure and a bare-except swallowing errors in Fuyu conversion are different bugs." + "reason": "Wav2Vec2 save/tokenization failures are not the same as FuyuBatchFeature swallowing errors with a bare except." }, { "left": "issue:44857", "right": "issue:45588", "accept": false, - "reason": "AMP loss crash in LwDetrImageLoss and flash_attention.py s_aux=None are unrelated failures." + "reason": "LwDetrImageLoss AMP/CUDA crash is a different failure mode than a flash_attention attribute error." }, { "left": "issue:43976", "right": "issue:45440", "accept": false, - "reason": "Python version compatibility and DeepseekV3MoE implementation divergence are not the same bug." + "reason": "Python version compatibility in Transformers 5.1.0 is unrelated to DeepseekV3MoE divergence." }, { "left": "issue:45372", "right": "issue:45406", "accept": false, - "reason": "Both are Gemma4-related, but one fails on missing mistral_common import and the other crashes in serve due to missing _tokenizer; distinct causes." + "reason": "Both involve Gemma4-related loading, but one is an import dependency issue and the other is a missing _tokenizer attribute in transformers serve." }, { "left": "issue:45459", "right": "issue:45712", "accept": false, - "reason": "Tokenizer exception masking and leftover dummy classes are unrelated maintenance/bug issues." + "reason": "Tokenizer error masking without protobuf and leftover dummy classes failing repo checks are unrelated." }, { "left": "issue:45588", "right": "issue:45823", "accept": false, - "reason": "Flash-attention sink handling and Gemma4 PLE device mismatch are different forward-pass problems." + "reason": "flash_attention.py s_aux=None crash is different from Gemma4 PLE device mismatch under device_map=auto." }, { "left": "issue:43278", "right": "issue:43381", "accept": false, - "reason": "Training/eval dtype mismatch and gradient checkpointing in eval mode are separate state-management bugs." + "reason": "Eval-time dtype change for embeddings is not the same as gradient checkpointing being disallowed in eval mode." }, { "left": "issue:44448", "right": "issue:44466", "accept": false, - "reason": "Model output regression and serialization inconsistency of lm_head.weight are different concrete issues." + "reason": "Different output across Transformers versions for Pegasus is unrelated to device-dependent lm_head.weight serialization." }, { "left": "issue:43653", "right": "issue:44743", "accept": false, - "reason": "Tokenizer special-token registration and recurrent-state resetting in Qwen3.5 are unrelated." + "reason": "BigBirdTokenizer special-token registration and Qwen3.5 recurrent-state reset are separate tokenizer/model bugs." }, { "left": "issue:45245", "right": "issue:45563", "accept": false, - "reason": "Category-cardinality runtime limit and a stale warning in generate() do not share a code-path." + "reason": "Category-count limit error has no relation to a stale generate() warning." }, { "left": "issue:25251", "right": "issue:32090", "accept": false, - "reason": "Pipeline top_k nesting behavior and distributed broadcast TypeError are unrelated." + "reason": "Pipeline top_k output nesting and Trainer broadcast TypeError are unrelated." }, { "left": "issue:45372", "right": "issue:45459", "accept": false, - "reason": "Gemma4 processor import breakage and protobuf-related tokenizer error masking are different failure modes." + "reason": "Gemma4 processor import dependency failure is different from protobuf exception masking in tokenizers." }, { "left": "issue:44206", "right": "issue:44479", "accept": false, - "reason": "Unsupported center argument in feature extraction and video input regression in Qwen VL models are not the same bug." + "reason": "Unsupported center arg in LasrFeatureExtractor is unrelated to the Qwen2.5/Qwen3.5 video input regression." }, { "left": "issue:43065", "right": "issue:43994", "accept": false, - "reason": "A dummy Conv2d placeholder and nonsensical SigLIP2 outputs are unrelated issues." + "reason": "Dummy Conv2d in Sam3PixelDecoder and SigLIP2 nonsensical outputs are different model-specific issues." }, { "left": "issue:44186", "right": "issue:45198", "accept": false, - "reason": "LayoutLMv2 tokenizer crashes on NER/padding and Wav2Vec2 save/tokenization failure are different tokenizer paths." + "reason": "LayoutLMv2 tokenizer crashes on NER/padding are unrelated to Wav2Vec2 save_pretrained/tokenization failures." + }, + { + "left": "issue:43065", + "right": "issue:45412", + "accept": false, + "reason": "A dummy Conv2d placeholder bug is unrelated to RT-DETR memory not being released." }, { "left": "issue:44871", "right": "issue:44977", "accept": false, - "reason": "Gemma eos_token_id inconsistency and Qwen3.5 flash-attention generation failure are separate configuration/runtime issues." + "reason": "Gemma3 eos_token_id configuration mismatch is not the same as Qwen3.5 flash-attention generation failure." }, { "left": "issue:44361", "right": "issue:45198", "accept": false, - "reason": "MLukeTokenizer AttributeError and Wav2Vec2 save/tokenization failure do not indicate the same underlying defect." - }, - { - "left": "issue:45491", - "right": "issue:45588", - "accept": false, - "reason": "NaN embeddings from mixed-length sliding-window attention and a flash_attention.py AttributeError are different bugs." - }, - { - "left": "issue:45698", - "right": "issue:45823", - "accept": false, - "reason": "Wrong custom module loading after save_pretrained and Gemma4 PLE device mismatch are unrelated." + "reason": "MLukeTokenizer AttributeError and Wav2Vec2 tokenization/save failures are different tokenizer bugs." } ] }, @@ -8509,14 +8674,14 @@ "issue:44625|issue:45584", "issue:29127|issue:35141" ], - "prepared_review_unit_hash": "2c8420a337d059860e2d9d76306631a5e9aca63188e223b8996499cb6964b8c1", + "prepared_review_unit_hash": "a8285994e2aeb8a5e2df7d57323787947952aae91c50865ab165f01aaaf65da6", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 47, @@ -8531,150 +8696,150 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: the soft pairs span unrelated tokenizer, model-loading, training, quantization, and image-processing bugs. No pair looks like the same underlying issue, so none should be merged as duplicates.", - "confidence": 0.95, - "canonical_issue_reason": "issue:43873 is the best anchor only because it has the strongest discussion/inbound activity and is a broad, active bug report; however, it is not a true duplicate hub for the other items.", + "summary": "This cluster is mostly a heterogeneous mix of unrelated Transformers issues; a few pairs are superficially similar by subsystem, but none look like the same underlying bug/change. No soft-edge duplicate should be merged.", + "confidence": 0.87, + "canonical_issue_reason": "issue:43873 is the broadest and most discussion-heavy unresolved bug here, so it is the best anchor for the cluster even though the rest of the items do not form a true duplicate set.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43873 is the most suitable representative among otherwise unrelated issues due to breadth and engagement, but it should be treated as its own distinct bug.", + "best_issue_reason": "issue:43873 is the strongest overall representative by scope, activity, and inbound references; it is the best issue to anchor triage for this mixed cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44451", "right": "issue:45520", "accept": false, - "reason": "Both are import/load failures, but one is a local-path warning and the other is a Python 3.13 flash_attn KeyError; different code paths and symptoms." + "reason": "Different failures: model loading for a specific checkpoint vs a Python 3.13 flash_attn import-path KeyError." }, { "left": "issue:43611", "right": "issue:46121", "accept": false, - "reason": "One is a Transformers v5 base_model_prefix loading regression; the other is a rope-validation TypeError on JSON-loaded lists. Not the same bug." + "reason": "Unrelated code paths: base_model_prefix loading regression vs rope-param validation on a JSON-loaded list." }, { "left": "issue:42915", "right": "issue:42994", "accept": false, - "reason": "Both mention quantization, but one fails with FineGrainedFP8Config during Qwen3Moe loading and the other fails when saving quantized models. Different operations." + "reason": "Both touch quantization, but one is a Qwen3Moe FP8 runtime failure and the other is quantized model saving; not the same bug." }, { "left": "issue:43066", "right": "issue:44462", "accept": false, - "reason": "Tokenizer-related, but one is a decoder-type change in v5 and the other is AutoTokenizer ignoring tokenizer.json. Different underlying problems." + "reason": "Both involve tokenizers/loading, but one is a v5 decoder-type warning and the other is AutoTokenizer ignoring tokenizer.json." }, { "left": "issue:32090", "right": "issue:33290", "accept": false, - "reason": "Trainer NoneType broadcast error versus Deepspeed Adafactor OOM; unrelated failures." + "reason": "Trainer broadcast TypeError and DeepSpeed Adafactor OOM are unrelated failure modes." }, { "left": "issue:42490", "right": "issue:43873", "accept": false, - "reason": "A trailing-slash local model import warning is unrelated to quantization/offloading behavior." + "reason": "Local-path trailing-slash import warning is unrelated to quantization/offloading behavior." }, { "left": "issue:29942", "right": "issue:30064", "accept": false, - "reason": "Flash Attention 2 test failures and void segmentation-map processing are unrelated." + "reason": "Flash Attention 2 test failures and void segmentation-map processing are different subsystems and bugs." }, { "left": "issue:33357", "right": "issue:39692", "accept": false, - "reason": "MacOS bus error for a CLIP model is not the same as documentation example errors for SigLIP2." + "reason": "MacOS bus error on a CLIP model is not the same as SigLIP2 documentation/example errors." }, { "left": "issue:29127", "right": "issue:34567", "accept": false, - "reason": "LayoutLMv3 error messaging and TrainerState token-count updating are separate issues." + "reason": "LayoutLMv3 error-message clarity and TrainerState token counting are unrelated issues." }, { "left": "issue:43994", "right": "issue:45412", "accept": false, - "reason": "SigLIP2 bad outputs and RT-DETR memory release are different bugs in different subsystems." + "reason": "SigLIP2 inference correctness and RT-DETR memory release are different code paths." }, { "left": "issue:44265", "right": "issue:45198", "accept": false, - "reason": "torch.export compatibility with torch_compilable_check is unrelated to wav2vec2 save/tokenization failures." + "reason": "torch.export support for torch_compilable_check is unrelated to Wav2Vec2 save_pretrained/tokenization failures." }, { "left": "issue:43257", "right": "issue:43278", "accept": false, - "reason": "Qwen3 MoE weight-conversion problems and BF16-to-FP32 dtype drift are distinct issues." + "reason": "Qwen3 MoE weight conversion under accelerate+deepspeed is separate from embedding dtype drift between train/eval." }, { "left": "issue:43756", "right": "issue:45440", "accept": false, - "reason": "Different models and different defects: RoPE-layer dropping versus DeepSeekV3 implementation divergence." + "reason": "Smollm3 RoPE-layer drop and DeepseekV3MoE divergence are distinct model-implementation issues." }, { "left": "issue:43421", "right": "issue:45491", "accept": false, - "reason": "Runtime special-token/post-processor updating is unrelated to Gemma3 NaN embeddings with sliding-window attention." + "reason": "Runtime special-token post-processing and Gemma3 NaN embeddings on GPU are not the same failure." }, { "left": "issue:43653", "right": "issue:44442", "accept": false, - "reason": "BigBird special-token registration and FastSpeech2Conformer AutoTokenizer loading are both tokenizer issues but not the same failure." + "reason": "BigBirdTokenizer special-token registration and AutoTokenizer failing to load FastSpeech2ConformerTokenizer are different tokenizer bugs." }, { "left": "issue:43381", "right": "issue:43701", "accept": false, - "reason": "Eval-mode gradient checkpointing limitation is unrelated to resume_from_checkpoint key mismatch." + "reason": "Gradient checkpointing in eval mode and resume_from_checkpoint key mismatch are unrelated training-path issues." }, { "left": "issue:44610", "right": "issue:45412", "accept": false, - "reason": "Processor image-size mismatch and RT-DETR memory cleanup are separate bugs." + "reason": "Processor image-size mismatch is unrelated to RT-DETR garbage-collection memory release." }, { "left": "issue:43824", "right": "issue:44279", "accept": false, - "reason": "ImportError for a specific Qwen2.5-VL class is not the same as a generic dependency issue." + "reason": "Importing a missing Qwen2.5-VL class and a generic dependency issue are not the same underlying bug." }, { "left": "issue:45440", "right": "issue:45593", "accept": false, - "reason": "DeepSeekV3 implementation divergence and D-FINE auxiliary-loss behavior are unrelated." + "reason": "DeepseekV3MoE implementation divergence and D-FINE auxiliary-loss behavior are different model bugs." }, { "left": "issue:29127", "right": "issue:33357", "accept": false, - "reason": "LayoutLMv3 box-validation clarity and CLIP bus error on MacOS do not share a code path." + "reason": "LayoutLMv3 box-information messaging and CLIP bus errors on MacOS are unrelated." }, { "left": "issue:29942", "right": "issue:36010", "accept": false, - "reason": "Flash Attention 2 test failures are unrelated to the GenerationMixin import error." + "reason": "Flash Attention 2 test failures are not the same as the GenerationMixin import breakage." }, { "left": "issue:43526", "right": "issue:45834", "accept": false, - "reason": "A BeitImageProcessorFast label-reduction bug is unrelated to Kosmos2.5 OCR index errors." + "reason": "BeitImageProcessorFast label reduction and Kosmos2.5 long-OCR index errors are unrelated processing bugs." }, { "left": "issue:43761", "right": "issue:45859", "accept": false, - "reason": "CLIPVisionModel hidden_states regression and Qwen3_5Moe CPU-offload incompatibility are different model-path bugs." + "reason": "CLIPVisionModel missing hidden_states and Qwen3.5 MoE CPU-offload incompatibility are different regressions." }, { "left": "issue:43526", @@ -8686,19 +8851,19 @@ "left": "issue:45103", "right": "issue:45372", "accept": false, - "reason": "auto_docstring annotation handling and Gemma 4 processor import failure have no shared underlying bug." + "reason": "auto_docstring annotation handling and Mistral ReasoningEffort import failure affect different components and code paths." }, { "left": "issue:44625", "right": "issue:45584", "accept": false, - "reason": "Qwen3.5 config propagation and Whisper empty-transcription generation failure are unrelated." + "reason": "Qwen3.5 num_labels propagation and Whisper empty-transcription generation are unrelated bugs." }, { "left": "issue:29127", "right": "issue:35141", "accept": false, - "reason": "LayoutLMv3 error messaging and embedding reinitialization on resize are separate defects." + "reason": "LayoutLMv3 error-message clarity and embedding reinitialization after resize are different issues." } ] }, @@ -8802,14 +8967,14 @@ "issue:43761|issue:44188", "issue:34689|issue:43611" ], - "prepared_review_unit_hash": "8f5205f3cd04fbd03d9d0d293e9ee3dde2398426ef7e0f07caf72da168f215b5", + "prepared_review_unit_hash": "a9955991463e0d523388434a13b24b0b94aa9afb12bf109835dccac9e7217a91", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, @@ -8824,192 +8989,186 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is highly heterogeneous, with most items describing unrelated regressions across loading, generation, audio, vision, quantization, and training utilities. No soft pair looks like the same underlying bug/change under strict triage.", - "confidence": 0.91, - "canonical_issue_reason": "No strong canonical issue: the items are too diverse to share one underlying bug. If a representative issue is required, issue 43611 is the broadest v5 model-loading regression.", + "summary": "This cluster is highly heterogeneous: most items are unrelated regressions or bugs across loading, generation, audio, quantization, configs, and trainer utilities. None of the proposed soft pairs look like true duplicates or mergeable PR-equivalents. The broadest central issue is the quantization/offloading memory report.", + "confidence": 0.41, + "canonical_issue_reason": "issue:43873 is the broadest and most central report, with the highest discussion/inbound-reference activity, and it captures a recurring quantization/offloading failure more generally than the other issues.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 43611 is the most broadly phrased and cluster-representative report, centered on a general Transformers 5.0.0 model-loading breakage.", + "best_issue_reason": "issue:43873 is the best cluster anchor because it is broad, actively discussed, and most likely to subsume related quantization/offloading follow-ups.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43334", "right": "issue:43531", "accept": false, - "reason": "Different model families and failures: missing pad_token_id for Qwen3-VL vs sliding_window behavior in Qwen3-MoE." + "reason": "Different model families and failures: Qwen3-VL missing pad_token_id vs Qwen3-MoE sliding_window behavior." }, { "left": "issue:44442", "right": "issue:45072", "accept": false, - "reason": "Tokenizer loading failure vs bfloat16 dtype mismatch in inference; unrelated code paths and symptoms." + "reason": "Tokenizer loading failure vs dtype mismatch in inference; different code paths and symptoms." }, { "left": "issue:42915", "right": "issue:43931", "accept": false, - "reason": "Different problems: FP8 config failure for Qwen3-MoE vs weight-shape mismatch loading a Qwen3-VL model." + "reason": "Qwen3Moe FP8 config failure and Qwen3-VL weight-shape mismatch are distinct model-loading bugs." }, { "left": "issue:44977", "right": "issue:45245", "accept": false, - "reason": "Flash-attention generation regression and category-count limit error are unrelated." + "reason": "Flash-attention generation regression is unrelated to the category-count RuntimeError." }, { "left": "issue:43526", "right": "issue:44857", "accept": false, - "reason": "Label reduction bug in an image processor vs AMP/CUDA loss crash; not the same bug." + "reason": "BEiT label reduction bug and LwDetr AMP crash are different image-processing/loss issues." }, { "left": "issue:42490", "right": "issue:43381", "accept": false, - "reason": "Local-path import warning and gradient-checkpointing eval-mode restriction are different issues." + "reason": "Trailing-slash import warning is unrelated to the eval-mode gradient-checkpointing restriction." }, { "left": "issue:44464", "right": "issue:45491", "accept": false, - "reason": "Compiled-forward chunked generation inconsistency vs Gemma3 NaN embeddings from sliding-window padding; different root causes." + "reason": "Both involve generation, but one is compiled forward chunked generation and the other is Gemma3 mixed-length sliding-window NaNs." }, { "left": "issue:43575", "right": "issue:43701", "accept": false, - "reason": "OOM during tensor-parallel model load is unrelated to checkpoint resume key mismatch." + "reason": "Tensor-parallel OOM during load is unrelated to checkpoint key mismatches." }, { "left": "issue:43262", "right": "issue:44206", "accept": false, - "reason": "Audio chat-template sampling-rate default mismatch vs unsupported center argument in a feature extractor." + "reason": "Audio chat-template sampling-rate default and unsupported `center` arg are different audio-processing bugs." }, { "left": "issue:43575", "right": "issue:44387", "accept": false, - "reason": "Both mention OOM, but one is TP load OOM and the other is increased reserved memory under int4 quantization; not the same concrete bug." + "reason": "Both mention OOM, but one is TP model loading and the other is int4 reserved-memory growth; different causes." }, { "left": "issue:44743", "right": "issue:44811", "accept": false, - "reason": "Recurrent-state reset in Qwen3.5-MoE is unrelated to Whisper batch_decode skip_special_tokens handling." + "reason": "Qwen3.5 recurrent-state reset and Whisper batch_decode token handling are unrelated." }, { "left": "issue:43646", "right": "issue:44077", "accept": false, - "reason": "Both touch initialization, but one is a broad custom-model-init regression and the other is a specific optional post_init policy for patchtsmixer." + "reason": "Generic v5 custom-init breakage and a PatchTSMixer-specific `post_init` policy bug are not the same issue." }, { "left": "issue:43441", "right": "issue:45588", "accept": false, - "reason": "Both involve FlashAttention, but one is a Ministral-3 regression and the other is an AttributeError on sink-less models; different failures." + "reason": "Different flash-attention failures with different models and root causes." }, { "left": "issue:44062", "right": "issue:45907", "accept": false, - "reason": "AddedToken keyword collision and list_repo_files revision forwarding are unrelated API bugs." + "reason": "Tokenizer AddedToken construction error is unrelated to hub file-list revision forwarding." }, { "left": "issue:43761", "right": "issue:44079", "accept": false, - "reason": "Hidden-states not returned from CLIPVisionModel forward is not the same as ModelOutput key assignment when a key was previously None." + "reason": "CLIPVisionModel hidden_states regression and ModelOutput key assignment are distinct output-handling bugs." }, { "left": "issue:45003", "right": "issue:45440", "accept": false, - "reason": "Unsafe sys.modules access and DeepseekV3MoE divergence from remote implementation are unrelated." + "reason": "sys.modules access safety and DeepSeekV3 implementation divergence are unrelated problems." }, { "left": "issue:43723", "right": "issue:45479", "accept": false, - "reason": "Tokenizer-loading regression and zero-loss classification behavior are unrelated." + "reason": "AutoTokenizer loading in v5 is unrelated to the single-label-classification zero-loss bug." }, { "left": "issue:43232", "right": "issue:43388", "accept": false, - "reason": "Generation kwargs update after sync_gpus and metric-gathering label truncation are different subsystems and bugs." + "reason": "Generation kwargs syncing and gather_for_metrics label dropping are separate distributed/runtime issues." }, { "left": "issue:44479", "right": "issue:45290", "accept": false, - "reason": "Video-input regression for Qwen models and chat-template tokenization crash with tool calls are distinct issues." + "reason": "Video-input regression and chat-template tool-call crash affect different multimodal paths." }, { "left": "issue:44442", - "right": "issue:45803", - "accept": false, - "reason": "FastSpeech2ConformerTokenizer loading failure and Fuyu tensor conversion swallowing errors are unrelated." - }, - { - "left": "issue:43381", - "right": "issue:43873", + "right": "issue:44898", "accept": false, - "reason": "Eval-mode gradient checkpointing restriction and quantization offloading behavior are not the same bug." + "reason": "Tokenizer loading and Perceiver image-size handling are different subsystems." }, { "left": "issue:33290", "right": "issue:34567", "accept": false, - "reason": "Deepspeed Adafactor OOM is unrelated to TrainerState token counting not updating." + "reason": "Optimizer OOM in deepspeed is unrelated to TrainerState token-count tracking." }, { "left": "issue:33290", "right": "issue:34689", "accept": false, - "reason": "OOM with Adafactor/Deepspeed and Llama 3.2 Vision model loading breakage are different failures." + "reason": "AdaFactor/deepspeed OOM and Llama 3.2 vision model-loading regression are unrelated." }, { "left": "issue:43316", "right": "issue:43525", "accept": false, - "reason": "Gemma3TextConfig API discrepancy and Llama4Config missing pad_token_id are similar in theme only, but not the same concrete bug." + "reason": "Both are config discrepancies, but they concern different model families and different missing fields." }, { "left": "issue:42490", "right": "issue:44743", "accept": false, - "reason": "Local-path warning and Qwen3.5 recurrent-state reset are unrelated." + "reason": "Local-path import warning and Qwen3.5 recurrent-state reset do not share a code path." }, { "left": "issue:33290", "right": "issue:35141", "accept": false, - "reason": "OOM in optimizer/Deepspeed and embedding reinitialization after resize are unrelated." + "reason": "Optimizer OOM is unrelated to embedding reinitialization during post_init." }, { "left": "issue:44442", "right": "issue:45803", "accept": false, - "reason": "Tokenizer loading failure and broad exception handling in FuyuBatchFeature are different code paths." + "reason": "Tokenizer loading failure is unrelated to Fuyu's broad exception handling bug." }, { "left": "issue:43381", "right": "issue:43873", "accept": false, - "reason": "Gradient checkpointing in eval mode and quantization offloading issues are distinct." + "reason": "Gradient checkpointing in eval mode and quantization offloading failures are different behaviors." }, { "left": "issue:43761", "right": "issue:44188", "accept": false, - "reason": "CLIPVision hidden_states regression and attention-kernel divergence under torch.compile are unrelated." + "reason": "Hidden-states regression in CLIPVisionModel is unrelated to torch.compile attention-kernel divergence." }, { "left": "issue:34689", "right": "issue:43611", "accept": false, - "reason": "Both concern loading regressions, but one is a specific model failure and the other is a broad base_model_prefix change; not the same bug." + "reason": "Both are model-loading regressions, but they affect different versions, model families, and root causes." } ] }, @@ -9111,14 +9270,14 @@ "issue:44077|issue:44534", "issue:44871|issue:45584" ], - "prepared_review_unit_hash": "10a81ea272afa4b1754a7251eac80a4ad896afcef214f3febcc25c7ec8dc6712", + "prepared_review_unit_hash": "918271073fec3cb53d44c6a9094f8ace771c596c0132d651603145d51a13661d", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, @@ -9133,174 +9292,174 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is noisy and mostly contains unrelated Transformers issues around loading, configs, processors, and runtime regressions. None of the soft pairs look like the same underlying bug/change, so all soft edges are rejected.", - "confidence": 0.38, - "canonical_issue_reason": "issue:43749 is the broadest, highest-engagement report in the set and best matches the cluster\u2019s recurring loading/offloading/runtime-regression theme.", + "summary": "This cluster is very loose: most items are separate Transformers bug reports spanning config, loading, processors, training, and multimodal edge cases. None of the soft pairs look like true duplicates of the same concrete bug or fix.", + "confidence": 0.34, + "canonical_issue_reason": "issue:43873 is the best single issue to anchor the cluster because it is open, has the most inbound references, and is the broadest/most discussion-heavy of the loading/offload-related reports.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43749 is the most representative single issue here because it is broad, active, and close to the dominant loading/offloading failures represented across the cluster.", + "best_issue_reason": "issue:43873 is the strongest representative issue: it is still open, relatively central to the cluster\u2019s recurring runtime/loading failures, and more likely than the others to absorb related follow-up.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43720", "right": "issue:44442", "accept": false, - "reason": "Different failures: BitNet packed-weight loading vs tokenizer class loading." + "reason": "Different failures in different areas: packed-weight loading for BitNet vs tokenizer loading for a FastSpeech2Conformer tokenizer." }, { "left": "issue:44077", "right": "issue:44479", "accept": false, - "reason": "Unrelated models/code paths: patchtsmixer post_init policy vs Qwen video regression." + "reason": "Unrelated code paths: PatchTSMixer post_init policy vs Qwen video-input regression." }, { "left": "issue:43493", "right": "issue:43827", "accept": false, - "reason": "SigLIP2 implementation mismatch is unrelated to pipeline v5 docs/removal references." + "reason": "Different problems: SigLIP2 implementation discrepancy vs docs still mentioning pipeline() after v5 removals." }, { "left": "issue:43931", "right": "issue:45216", "accept": false, - "reason": "Both are Qwen model-loading issues, but one is weight-shape mismatch and the other is save_pretrained checkpoint correctness; not the same bug." + "reason": "Both involve Qwen models, but one is a weight-shape load failure and the other is a save_pretrained checkpoint regression." }, { "left": "issue:42907", "right": "issue:45698", "accept": false, - "reason": "Save/dequantization failure is a different problem from loading the wrong custom module after save_pretrained." + "reason": "Related to save/load behavior in a broad sense, but the concrete bugs differ: dequantized model saving vs wrong custom module restoration." }, { "left": "issue:34689", "right": "issue:36010", "accept": false, - "reason": "Llama 3.2 model loading regression is unrelated to the GenerationMixin import error." + "reason": "Different import/load failures affecting different APIs and model families." }, { "left": "issue:43316", "right": "issue:45759", "accept": false, - "reason": "Both involve config handling, but Gemma3TextConfig API mismatch and Qwen composite-config unwrapping are distinct issues." + "reason": "Both concern config/model API mismatches, but they are different composite-model behaviors and not the same bug." }, { "left": "issue:41628", "right": "issue:45372", "accept": false, - "reason": "Different import failures for different symbols and packages." + "reason": "Both are ImportErrors, but for unrelated symbols and packages with different fixes." }, { "left": "issue:33290", "right": "issue:36331", "accept": false, - "reason": "DeepSpeed/Adafactor OOM is unrelated to CustomTrainer.compute_loss signature breakage." + "reason": "Completely different issues: optimizer memory usage vs trainer signature mismatch." }, { "left": "issue:43749", "right": "issue:43931", "accept": false, - "reason": "FSDP CPU RAM-efficient loading failure is a different code path from Qwen3-VL shape mismatches." + "reason": "Different loading subsystems and symptoms: FSDP CPU RAM efficient loading vs Qwen3-VL shape mismatch." }, { "left": "issue:43873", "right": "issue:44871", "accept": false, - "reason": "Quantized offloading behavior and Gemma eos_token_id inconsistency are unrelated." + "reason": "Offloading/quantization behavior is unrelated to Gemma-3 eos_token_id configuration." }, { "left": "issue:43643", "right": "issue:44913", "accept": false, - "reason": "trust_remote_code field loss is unrelated to GPTNeoX rotary_pct not persisting on reload." + "reason": "Both involve config persistence, but one is missing fields from trust_remote_code and the other is a specific attribute not round-tripping." }, { "left": "issue:44190", "right": "issue:44496", "accept": false, - "reason": "Dataset-loading example failure is unrelated to an unrecognized custom model/config error." + "reason": "Dataset loading for an example script is unrelated to an unrecognized model/config.json error." }, { "left": "issue:44855", "right": "issue:45137", "accept": false, - "reason": "Python 3.13 JIT parsing/import error is unrelated to DeepSpeed ZeRO3 deque underflow." + "reason": "No shared underlying bug: Python 3.13 decorator parsing vs DeepSpeed ZeRO3 deque underflow." }, { "left": "issue:25251", "right": "issue:28282", "accept": false, - "reason": "Pipeline output shape change is unrelated to missing PyTorch import." + "reason": "Pipeline output nesting and missing PyTorch are unrelated issues." }, { "left": "issue:43540", "right": "issue:44514", "accept": false, - "reason": "Different multimodal/video processor failures in different models and call sites." + "reason": "Both are Qwen multimodal processor bugs, but the concrete failures differ: video processing ValueError vs batched chat-template crash." }, { "left": "issue:43316", "right": "issue:44877", "accept": false, - "reason": "Both are config-related, but Gemma3TextConfig API inconsistency and granite_speech strict-config loading are different problems." + "reason": "Different config issues: API discrepancy in Gemma3TextConfig vs strict config blocking granite_speech loading." }, { "left": "issue:43295", "right": "issue:43825", "accept": false, - "reason": "Processor.tokenizer regression is unrelated to pipeline translation error messaging." + "reason": "Different regressions: processor/tokenizer handling vs a misleading pipeline() error message about translation tasks." }, { "left": "issue:44162", "right": "issue:44792", "accept": false, - "reason": "ESM2 model breakage is unrelated to a janus image-generation test failure." + "reason": "Different models and code paths: ESM2 breakage vs janus image-generation test failure." }, { "left": "issue:43065", "right": "issue:44617", "accept": false, - "reason": "Sam3PixelDecoder dummy layer issue is unrelated to Sam3Video CUDA OOM." + "reason": "Both mention Sam3, but one is a dummy Conv2d implementation detail and the other is a CUDA OOM symptom." }, { "left": "issue:43316", "right": "issue:43716", "accept": false, - "reason": "Gemma3TextConfig API discrepancy is different from a dtype mismatch between image preprocessor and model." + "reason": "Different multimodal config issues with no evidence they stem from the same bug." }, { "left": "issue:43404", "right": "issue:43441", "accept": false, - "reason": "Weight tying bug and FlashAttention regression affect related models but not the same concrete failure." + "reason": "Same model family, but different failures: untied lm_head weights vs FlashAttention compatibility." }, { "left": "issue:43550", "right": "issue:43701", "accept": false, - "reason": "torch.compile/SDPA failure is unrelated to resume_from_checkpoint key mismatch." + "reason": "Completely unrelated: torch.compile/SDPA failure vs resume_from_checkpoint key mismatch." }, { "left": "issue:43606", "right": "issue:43701", "accept": false, - "reason": "CPU offload device mismatch is unrelated to resume_from_checkpoint key mismatch." + "reason": "CPU offload device mismatch is unrelated to checkpoint key mismatch." }, { "left": "issue:16998", "right": "issue:22355", "accept": false, - "reason": "Model max length question is unrelated to a transformers.onnx import error." + "reason": "Different import/config questions with no shared underlying bug." }, { "left": "issue:44077", "right": "issue:44534", "accept": false, - "reason": "patchtsmixer post_init policy is unrelated to non-persistent buffer initialization in v5." + "reason": "PatchTSMixer post_init policy is unrelated to non-persistent buffer initialization in v5." }, { "left": "issue:44871", "right": "issue:45584", "accept": false, - "reason": "Gemma eos_token_id inconsistency is unrelated to Whisper empty-transcription generation failure." + "reason": "Gemma eos_token_id inconsistency and Whisper empty-transcription failure are unrelated." } ] }, @@ -9402,20 +9561,20 @@ "issue:44792|issue:45600", "issue:44521|issue:45381" ], - "prepared_review_unit_hash": "7cb26f52c7bab59ca09bdadc1c7672ad76fa8e08e8a622aa2dd750a1e14b531f", + "prepared_review_unit_hash": "d6bf92b2a553e26dfa7135d05a19239505367461fa6d32d7bd3a1210245c26c3", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 28, - "serialized_chars": 35676, + "serialized_chars": 35673, "estimated_input_tokens": 8919, "estimated_eval_tokens": 18094 }, @@ -9424,180 +9583,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is thematically loose and mostly contains unrelated issue reports; none of the proposed soft pairs look like true duplicates or the same fix. All soft edges should be rejected.", - "confidence": 0.93, - "canonical_issue_reason": "There is no single clear canonical duplicate target because the items span many different subsystems and root causes. If one representative issue is needed, issue:45698 is the broadest high-signal loader/save-load regression in the set.", + "summary": "This cluster is mostly a loose semantic group of unrelated Transformers bugs and regressions; the soft pairs share broad subsystem similarity but not the same concrete underlying defect, so none should be merged as duplicates. There are no PRs in the cluster.", + "confidence": 0.61, + "canonical_issue_reason": "issue:43761 is the cleanest single anchor: a concrete v5 regression in multimodal/vision forward behavior with a clear symptom and repro. It\u2019s the best representative issue if this cluster needs one, though the overall set is too heterogeneous to treat as one duplicate group.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45698 is the best representative issue because it has broad impact, strong discussion activity, and captures a common save/load custom-module failure mode.", + "best_issue_reason": "issue:43761 is the most representative issue because it is specific, user-facing, and fits the common v5 regression/multimodal theme better than the more isolated loading/config or script failures. It is still only a cluster anchor, not evidence of duplication with the others.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:42617", "right": "issue:43575", "accept": false, - "reason": "Both are runtime failures, but one is a 3D parallel script failure and the other is an OOM when loading a large Qwen model with tensor parallelism; different code paths and causes." + "reason": "Different failures: 3d_parallel.py execution vs tensor-parallel OOM while loading a specific Qwen2 model." }, { "left": "issue:43673", "right": "issue:44743", "accept": false, - "reason": "Both involve cache behavior, but one is a GenerationMixin cache regression in chunked prefill while the other is recurrent-state reset logic in modular Qwen3.5; not the same bug." + "reason": "Both involve cache/state, but one is a GenerationMixin chunked_prefill cache regression and the other is recurrent-state reset in qwen3_5; different code paths and symptoms." }, { "left": "issue:43575", "right": "issue:43742", "accept": false, - "reason": "Tensor-parallel OOM on a large model is unrelated to a KeyError while loading MobileLLM-125M; different failure modes." + "reason": "OOM during tensor-parallel model load is not the same bug as a key error loading MobileLLM-125M config." }, { "left": "issue:41950", "right": "issue:43475", "accept": false, - "reason": "Both are vision/video-adjacent, but one is a processor-selection bug in a pipeline and the other is a missing attribute on a SAM3 output object; not the same underlying issue." + "reason": "Pipeline processor lookup for video classification is unrelated to Sam3 vision encoder output missing an attribute." }, { "left": "issue:43388", "right": "issue:43688", "accept": false, - "reason": "One drops label elements during gather_for_metrics; the other is auxiliary-loss normalization in MoE models. Different training bugs." + "reason": "Gathering metrics dropping labels and auxiliary-loss normalization are distinct training bugs." }, { "left": "issue:43295", "right": "issue:44871", "accept": false, - "reason": "A regression in custom model code accessing processor.tokenizer is unrelated to an eos_token_id config inconsistency in Gemma-3." + "reason": "Custom processor/tokenizer API regression is unrelated to an eos_token_id configuration mismatch." }, { "left": "issue:43761", "right": "issue:44479", "accept": false, - "reason": "Both are model forward regressions, but they affect different models and different multimodal/video code paths; not mergeable as one fix." + "reason": "Both are multimodal regressions, but one is CLIPVision hidden_states handling and the other is Qwen video-input position-id behavior; not the same defect." }, { "left": "issue:43452", "right": "issue:44496", "accept": false, - "reason": "Both concern model loading, but one is gguf_file handling for AutoTokenizer/AutoModel and the other is missing model_type in config.json; different root causes." + "reason": "Broken gguf loading for tokenizer/model classes is different from an unrecognized-model config.json missing model_type." }, { "left": "issue:45127", "right": "issue:45200", "accept": false, - "reason": "LoRA merge collapse with tied embeddings is unrelated to Gemma 4\u2019s mm_token_type_ids defaulting problem." + "reason": "LoRA merge with extended vocab and Gemma 4 mm_token_type_ids defaulting are separate model-behavior issues." }, { "left": "issue:44871", "right": "issue:46050", "accept": false, - "reason": "An eos_token_id config mismatch is not the same as broken sinq quantization serialization across save_pretrained/from_pretrained." + "reason": "EOS-token config inconsistency is unrelated to save/load breakage for sinq-quantized models." }, { "left": "issue:43646", "right": "issue:44913", "accept": false, - "reason": "A broad Transformers 5.0 custom-init breakage is not the same as GPTNeoXConfig rotary_pct not persisting on reload." + "reason": "Broad custom-model initialization breakage is not the same as GPTNeoX rotary_pct failing to round-trip on reload." }, { "left": "issue:43504", "right": "issue:45698", "accept": false, - "reason": "Both are loading-related, but one fails on a legacy field in a preset while the other loads the wrong custom module after save/load; distinct bugs." + "reason": "Legacy-field pretrained loading failure and wrong custom module resolution after save_pretrained are different load-path bugs." }, { "left": "issue:32090", "right": "issue:35141", "accept": false, - "reason": "Trainer GPU broadcast NoneType error and post_init reinitializing output embeddings are unrelated." + "reason": "Trainer GPU broadcast TypeError is unrelated to output-embedding reinitialization during post_init after resizing embeddings." }, { "left": "issue:44734", "right": "issue:45406", "accept": false, - "reason": "One is KV-cache continuation indexing in serve; the other is Gemma4Processor missing _tokenizer in serve. Same surface area, different defects." + "reason": "KV-cache continuation tensor indexing in serve is a different failure from Gemma4Processor missing _tokenizer." }, { "left": "issue:44617", "right": "issue:45584", "accept": false, - "reason": "CUDA OOM in Sam3Video is unrelated to Whisper failing on empty transcription after align_special_tokens." + "reason": "SAM3 video OOM and Whisper empty-transcription generation failure are unrelated model-specific bugs." }, { "left": "issue:45563", "right": "issue:45584", "accept": false, - "reason": "A stale warning for num_return_sequences is not the same as a generation crash on empty transcription." + "reason": "A stale warning for num_return_sequences is not the same as Whisper failing on empty transcription." }, { "left": "issue:43761", "right": "issue:44242", "accept": false, - "reason": "CLIPVision hidden_states regression and missing load-balancing loss are different model behaviors and code paths." + "reason": "CLIPVision hidden_states regression and missing load-balancing loss when output_router_logits=False are different issues." }, { "left": "issue:43525", "right": "issue:43531", "accept": false, - "reason": "Missing pad_token_id in Llama4Config is unrelated to a sliding_window issue in Qwen3-MoE." + "reason": "Missing pad_token_id in Llama4Config and Qwen3-MoE sliding_window behavior are distinct model/config bugs." }, { "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 producing bad outputs." + "reason": "Janus image-generation test failure and Qwen3.5 flash-attention generation issue are not the same underlying bug." }, { "left": "issue:43931", "right": "issue:44060", "accept": false, - "reason": "Shape-mismatch loading Qwen3-VL weights and incorrect tied-weights warning in Qwen3-Next are different issues." + "reason": "Weight-shape mismatch while loading Qwen3-VL is unrelated to the tied-weights warning that binds embed_tokens.weight to linear_attn.dt_bias." }, { "left": "issue:44479", "right": "issue:44734", "accept": false, - "reason": "Video-input regression in Qwen2.5-VL/Qwen3-VL is unrelated to serve-side KV-cache continuation indexing." + "reason": "Qwen video-input regression and serve KV-cache continuation crash are separate subsystems and code paths." }, { "left": "issue:43906", "right": "issue:45290", "accept": false, - "reason": "An isolated reproduction of another issue is not the same as a crash in apply_chat_template with tool-call assistant messages." + "reason": "An isolated reproduction of another issue is not the same as an apply_chat_template crash on tool-call assistant messages." }, { "left": "issue:44464", "right": "issue:45563", "accept": false, - "reason": "Chunked generation inconsistency with compiled forward is unrelated to a stale warning in paged generate()." + "reason": "Compiled-forward chunked-generation inconsistency and a stale warning for paged generate are different generation bugs." }, { "left": "issue:44190", "right": "issue:44964", "accept": false, - "reason": "Local dataset loading in a training script is unrelated to loading the Phi-4 multimodal model." + "reason": "Local dataset loading in an image-classification script is unrelated to loading a multimodal model checkpoint." }, { "left": "issue:43827", "right": "issue:43994", "accept": false, - "reason": "Docs still referencing pipeline() and a model producing nonsensical outputs under AutoModel/pipeline are not the same bug." + "reason": "Docs still referencing pipeline() and SigLIP2 producing nonsensical outputs are not the same defect." }, { "left": "issue:44534", "right": "issue:44913", "accept": false, - "reason": "Non-persistent buffers being filled with junk and a config value reverting on reload are both persistence-related but not the same defect." + "reason": "Non-persistent buffer junk in v5 is not the same as GPTNeoXConfig rotary_pct resetting on reload." }, { "left": "issue:44792", "right": "issue:45600", "accept": false, - "reason": "A janus test failure and auto_mappings.py referencing removed Sam3LiteText configs are unrelated." + "reason": "A Janus generate-images test failure is unrelated to auto_mappings referencing removed Sam3LiteText configs." }, { "left": "issue:44521", "right": "issue:45381", "accept": false, - "reason": "All-zero assistant masks in apply_chat_template and wrong vision_position_ids for qwen2.5-vl video input are distinct multimodal bugs." + "reason": "All-zero assistant masks in apply_chat_template and wrong vision_position_ids for Qwen2.5-VL video input are different multimodal bugs." } ] }, @@ -9701,20 +9860,20 @@ "issue:41950|issue:44016", "issue:42175|issue:44279" ], - "prepared_review_unit_hash": "85e8fdcc6b186b9c3e01b0fee8ca1585cba8aa85012b4d9b2ff36cf7ec145854", + "prepared_review_unit_hash": "ef442f064f25dd664cac2ac4adb1a28e7acdc321a290854c47d8712ab74353dc", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36544, + "serialized_chars": 36542, "estimated_input_tokens": 9136, "estimated_eval_tokens": 18528 }, @@ -9723,180 +9882,186 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab bag of unrelated Transformer issues: import/export regressions, model-specific config bugs, docs problems, and runtime failures. None of the soft pairs look like true duplicates; the only close thematic area is token-embedding resizing, but even there the bugs differ in model and failure mode.", - "confidence": 0.94, - "canonical_issue_reason": "If one issue must serve as the representative, issue 35141 is the clearest broadly-scoped bug report: it describes a concrete resize_token_embeddings/post_init regression with a well-defined failure mode. It is the most reusable anchor in the set, even though the overall cluster is not a true duplicate group.", + "summary": "This cluster is highly heterogeneous; the soft-similarity edges are mostly superficial and do not describe the same underlying bug or change. No PRs are present. The only plausible mini-cluster is around token-embedding resize behavior, centered on issue 35141.", + "confidence": 0.84, + "canonical_issue_reason": "Issue 35141 is the clearest root-cause report for the only related subtopic here: resize_token_embeddings / post_init reinitialization when embeddings are untied.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 35141 is the best overall issue to represent the cluster because it is specific, technically actionable, and closest to another embedding-resize report without being over-specific to one model family.", + "best_issue_reason": "35141 is the best anchor because it states the underlying code-path problem more generally than the model-specific symptom in 45276.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:41720", "right": "issue:45245", "accept": false, - "reason": "Different subsystems and failures: Qwen3 device-mapping CUDA assert vs an unrelated category-cardinality runtime error." + "reason": "Different failures: cudaErrorAssert during Qwen3 auto device mapping vs a category-count limit runtime error." }, { "left": "issue:44855", "right": "issue:45588", "accept": false, - "reason": "One is a Python 3.13 import/parsing problem, the other is a flash-attention runtime AttributeError; no shared code path." + "reason": "Unrelated issues: Python 3.13 import parsing failure vs flash_attention crashing on s_aux=None." }, { "left": "issue:43065", "right": "issue:44610", "accept": false, - "reason": "Sam3 pixel decoder dummy layer vs OmDet-Turbo input-size mismatch are unrelated model-specific issues." + "reason": "Different models and different bugs: Sam3PixelDecoder dummy conv vs OmDet-Turbo image-size mismatch." }, { "left": "issue:44933", "right": "issue:45362", "accept": false, - "reason": "Import/export cleanup in image_utils is unrelated to a Qwen3.5 chat crash." + "reason": "Import alias problem vs Qwen3.5 chat crash; no shared code path." }, { "left": "issue:35141", "right": "issue:42490", "accept": false, - "reason": "Token-embedding reinitialization bug and trailing-slash local-model import warning are unrelated." + "reason": "Resize-token-embedding/post_init bug is unrelated to import warning on trailing slash paths." }, { "left": "issue:43335", "right": "issue:44855", "accept": false, - "reason": "SwitchTransformers sparse-layer config bug is unrelated to the DebertaV2 import parse failure." + "reason": "SwitchTransformers sparse-layer construction bug is unrelated to DebertaV2 import parsing on Python 3.13." }, { "left": "issue:44077", "right": "issue:44079", "accept": false, - "reason": "Optional post_init handling in patchtsmixer and ModelOutput key assignment are different bugs in different areas." + "reason": "Optional post_init enforcement and ModelOutput None-key assignment are distinct behaviors in different code paths." }, { "left": "issue:43550", "right": "issue:44190", "accept": false, - "reason": "torch.compile/SDPA failure in Bamba and dataset-loading failure in an example script are not the same underlying issue." + "reason": "torch.compile + SDPA issue is unrelated to local dataset loading in a training script." }, { "left": "issue:43976", "right": "issue:44855", "accept": false, - "reason": "Python version support regression and a JIT parsing indentation error are distinct problems." + "reason": "Python-version support regression is unrelated to the DebertaV2 JIT parsing bug." }, { "left": "issue:44737", "right": "issue:45803", "accept": false, - "reason": "XLNet device placement bug and a bare-except error handler are unrelated." + "reason": "CPU device placement bug in XLNet is unrelated to FuyuBatchFeature swallowing exceptions." }, { "left": "issue:43644", "right": "issue:44913", "accept": false, - "reason": "Non-persistent buffer initialization and GPTNeoX rotary_pct reload behavior are different serialization/config issues." + "reason": "Both involve config/state persistence, but one is non-persistent buffers and the other is GPTNeoX rotary_pct reload; different bugs." }, { "left": "issue:45127", "right": "issue:45357", "accept": false, - "reason": "LoRA vocabulary-merging collapse and incorrect visual-encoder keys on save_pretrained are not the same bug." + "reason": "LoRA merge with extended vocab and incorrect visual encoder keys are different Qwen-related issues." }, { "left": "issue:43540", "right": "issue:44734", "accept": false, - "reason": "Video input processing failure in Qwen3OmniMoe and KV-cache tensor-indexing crash in transformers serve are unrelated." + "reason": "Video-input validation error in Qwen3OmniMoe is unrelated to serve KV-cache tensor indexing." }, { "left": "issue:44462", "right": "issue:45907", "accept": false, - "reason": "Tokenizer.json loading behavior and list_repo_files revision forwarding are separate API bugs." + "reason": "AutoTokenizer repository-file selection and list_repo_files revision forwarding are separate hub API behaviors." }, { "left": "issue:44112", "right": "issue:45600", "accept": false, - "reason": "A stale CI test and removed auto_mappings references are different CI breakages." + "reason": "CI test flake in GraniteSpeech is unrelated to stale Sam3LiteText auto_mappings references." }, { "left": "issue:44938", "right": "issue:45542", "accept": false, - "reason": "Python 3.14 import failure and missing TensorFlow backend wiring are unrelated environment/dependency issues." + "reason": "Python 3.14 import/load failure and missing TensorFlow backend are different environment/setup problems." + }, + { + "left": "issue:43335", + "right": "issue:44190", + "accept": false, + "reason": "SwitchTransformers sparse-layer bug is unrelated to local dataset loading in run_image_classification_no_trainer.py." }, { "left": "issue:46082", "right": "issue:46121", "accept": false, - "reason": "LlamaConfig head_dim validation and rope-parameter JSON-list handling are different validation bugs." + "reason": "LlamaConfig head_dim validation and rope-parameter JSON list handling are distinct config-validation bugs." }, { "left": "issue:39692", "right": "issue:44190", "accept": false, - "reason": "SigLIP2 documentation mistakes and local dataset loading failure are not the same change or bug." + "reason": "SigLIP2 doc example errors are unrelated to loading a local dataset in the image-classification script." }, { "left": "issue:44496", "right": "issue:45803", "accept": false, - "reason": "Unrecognized model/config loading and the Fuyu batch-feature exception handler bug are unrelated." + "reason": "Model config recognition failure is unrelated to exception swallowing in FuyuBatchFeature." }, { "left": "issue:45276", "right": "issue:45584", "accept": false, - "reason": "Gemma4 resize_token_embeddings behavior and Whisper empty-transcription handling do not share a code path." + "reason": "Both involve model/input handling, but Gemma4 resize-token-embedding behavior is unrelated to Whisper empty-transcription failure." }, { "left": "issue:36246", "right": "issue:41628", "accept": false, - "reason": "Different missing exports from different modules; both are import errors but not the same underlying breakage." + "reason": "Two different missing imports for different symbols; same symptom class, not the same bug." }, { "left": "issue:43295", "right": "issue:45276", "accept": false, - "reason": "Processor/tokenizer regression and Gemma4 embedding-resize behavior are separate API regressions." + "reason": "Processor/tokenizer regression in v4.57.5 is unrelated to Gemma4 resize_token_embeddings not updating embeddings." }, { "left": "issue:43479", "right": "issue:45375", "accept": false, - "reason": "Both are multimodal config issues, but they concern different fields and different model families." + "reason": "Default config initialization and strict-dropping of a vision field are different multimodal config bugs." }, { "left": "issue:43650", "right": "issue:44220", "accept": false, - "reason": "Placeholder issue text and a feature-extraction bug are unrelated." + "reason": "One report is unusable/title-only; the other is a specific _torch_extract_fbank_features issue." }, { "left": "issue:45245", "right": "issue:45584", "accept": false, - "reason": "Category-count limit error and Whisper transcription failure are unrelated." + "reason": "Category-count overflow and Whisper empty-transcription are unrelated." }, { "left": "issue:36246", "right": "issue:38617", "accept": false, - "reason": "Two different ImportErrors from different missing symbols and modules; not the same bug." + "reason": "Both are import errors, but they concern different missing names in different modules." }, { "left": "issue:41950", "right": "issue:44016", "accept": false, - "reason": "Video pipeline processor mismatch and a notebook syntax error are entirely different problems." + "reason": "Video-classification pipeline image-processor lookup and a notebook syntax error are unrelated." }, { "left": "issue:42175", "right": "issue:44279", "accept": false, - "reason": "Missing TensorFlow in the torch extra and a generic dependency issue are not the same concrete bug." + "reason": "TensorFlow-not-in-backend packaging issue is unrelated to a generic dependency problem." } ] }, @@ -10003,214 +10168,214 @@ "issue:43299|issue:45357", "issue:43479|issue:43525" ], - "prepared_review_unit_hash": "2f9793b2a88d43a07c6557c3b7366fd9a2926a66e37035a9e5b963cb7c9e9633", + "prepared_review_unit_hash": "758c0c9199ab992b3762f684657e5724eba506a842df1ca831230c7a8c923ad1", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 32, - "serialized_chars": 37246, - "estimated_input_tokens": 9312, - "estimated_eval_tokens": 18880 + "serialized_chars": 37249, + "estimated_input_tokens": 9313, + "estimated_eval_tokens": 18882 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These look like a collection of unrelated Transformers bugs, mostly centered on model-loading, config, tokenizer, and backend regressions; none of the soft pairs appear to be the same concrete issue, so I rejected them all.", - "confidence": 0.91, - "canonical_issue_reason": "issue:43824 is the strongest representative issue here: it is a concrete user-facing API import failure with multiple inbound references and broad impact, while still being narrowly scoped and easy to triage.", + "summary": "These issues are a heterogeneous mix of unrelated model-loading, config, import, and docs/test regressions. None of the soft pairs look like true duplicates or the same concrete fix, so no soft edge is accepted.", + "confidence": 0.93, + "canonical_issue_reason": "Issue 45632 is the cleanest single library-level bug in the set: a precise `trust_remote_code` cache-path collision for local models with the same leaf directory name. The rest are mostly unrelated one-off model or tooling regressions, so there is no strong cluster-wide duplicate target.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43824 is the best overall issue candidate because it is specific, reproducible, and already shown to affect downstream users via references; it is a cleaner canonical target than broader or more ambiguous regressions in the set.", + "best_issue_reason": "45632 is the best representative issue to keep because it has a narrow, actionable root cause and a fixable central code path affecting multiple local-model setups.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44186", "right": "issue:44734", "accept": false, - "reason": "Different subsystems and failures: tokenizer NER/padding crash vs server KV-cache continuation indexing bug." + "reason": "Different failures: tokenizer/NER padding crash vs serving KV-cache continuation tensor indexing." }, { "left": "issue:44683", "right": "issue:45446", "accept": false, - "reason": "Both mention flex_attention, but one is a torch>=2.9 compiled-kernel failure and the other is an import/version-check bug; not the same fix." + "reason": "Both touch flex_attention, but one is a torch-version check bug and the other is a compiled flex_attention failure on torch>=2.9." }, { "left": "issue:44246", "right": "issue:45003", "accept": false, - "reason": "Possible loose relation to import behavior, but one is slow imports and the other is unsafe sys.modules access; too indirect to treat as the same bug." + "reason": "Import slowness from package loading is not the same as unsafe `sys.modules` access in `modeling_utils`." }, { "left": "issue:44016", "right": "issue:44220", "accept": false, - "reason": "Notebook syntax error and audio feature extraction issue are unrelated." + "reason": "Notebook syntax error and `_torch_extract_fbank_features()` are unrelated bugs." }, { "left": "issue:43404", "right": "issue:45216", "accept": false, - "reason": "Both involve saving or weight tying, but they concern different models and different failure modes." + "reason": "Different code paths: weight tying in Mistral3 vs incorrect `save_pretrained` output for Qwen3.5." }, { "left": "issue:38175", "right": "issue:43065", "accept": false, - "reason": "Completely different models and code paths." + "reason": "SigLIP zero-probabilities and a dummy `Conv2d` in Sam3PixelDecoder are distinct model bugs." }, { "left": "issue:44734", "right": "issue:45632", "accept": false, - "reason": "Both use cache terminology, but one is KV-cache continuation indexing and the other is local trust_remote_code cache-path collision." + "reason": "Serving KV-cache indexing and local trust_remote_code cache-path collisions are unrelated." }, { "left": "issue:43643", "right": "issue:44496", "accept": false, - "reason": "Both are config/loading issues, but one is missing fields with trust_remote_code and the other is an unrecognized model/config.json problem." + "reason": "Missing fields in `AutoConfig.from_pretrained(..., trust_remote_code=True)` is different from an unrecognized-model/config.json failure." }, { "left": "issue:44561", "right": "issue:45632", "accept": false, - "reason": "Different trust_remote_code regressions affecting different mechanisms." + "reason": "Removal of `is_torch_fx_available` breaking remote-code models is a different regression than cache-path collision." }, { "left": "issue:44964", "right": "issue:45600", "accept": false, - "reason": "Both are model-loading/CI failures, but they target different models and different broken references." + "reason": "Phi-4 multimodal load failure and removed Sam3LiteText config references are different model-specific regressions." }, { "left": "issue:44112", "right": "issue:44964", "accept": false, - "reason": "CI test flake in GraniteSpeech is unrelated to Phi-4 model loading." + "reason": "A stale GraniteSpeech CI test is not the same bug as loading Phi-4 multimodal." }, { "left": "issue:43854", "right": "issue:44964", "accept": false, - "reason": "Separate models and separate loading failures." + "reason": "GLM-4.7-Flash loading issues are unrelated to the Phi-4 multimodal loading failure." }, { "left": "issue:43976", "right": "issue:44987", "accept": false, - "reason": "Version-support breakage is not the same as a specific model-loading failure." + "reason": "Python-version compatibility and a specific model-loading regression are not the same underlying bug." }, { "left": "issue:42907", "right": "issue:43441", "accept": false, - "reason": "Same broad Ministral family, but saving dequantized weights and FlashAttention loading are different code paths." + "reason": "Saving dequantized Ministral/Devstral models and FlashAttention load failures are distinct problems." }, { "left": "issue:42490", "right": "issue:45632", "accept": false, - "reason": "Both involve local paths, but one is a trailing-slash import warning and the other is a cache collision between same leaf names." + "reason": "A warning for local paths ending with a slash is a different path-handling issue than cache-key collisions across leaf directories." }, { "left": "issue:43531", "right": "issue:45600", "accept": false, - "reason": "Different issues in different models; no shared concrete bug." + "reason": "Qwen3-MoE sliding-window behavior and Sam3LiteText auto-mapping cleanup are unrelated." }, { "left": "issue:43295", "right": "issue:45335", "accept": false, - "reason": "Processor/tokenizer regression and embedding resize propagation bug are unrelated." + "reason": "Processor/tokenizer regression and `resize_token_embeddings` not updating decoder embeddings are different bugs." }, { "left": "issue:43526", "right": "issue:43901", "accept": false, - "reason": "Image-processor label reduction bug vs pipeline docs mismatch; not the same issue." + "reason": "BEiT `reduce_labels` output corruption is unrelated to a docs mismatch about `return_all_scores`." }, { "left": "issue:43673", "right": "issue:44336", "accept": false, - "reason": "Generation cache regression and ANSI logging cleanup are unrelated." + "reason": "Chunked-prefill cache regression and ANSI escape output in loading_report are unrelated." }, { "left": "issue:44496", "right": "issue:45042", "accept": false, - "reason": "Model config loading failure vs image-processor torchvision dependency regression." + "reason": "Unrecognized model/config handling and PIL backend requiring torchvision are different code paths." }, { "left": "issue:44016", "right": "issue:44062", "accept": false, - "reason": "Notebook syntax error and tokenizer keyword-argument error are unrelated." + "reason": "Notebook syntax error is unrelated to the `AddedToken(..., special=...)` argument conflict." }, { "left": "issue:43335", "right": "issue:45593", "accept": false, - "reason": "Different models and different training/configuration bugs." + "reason": "Sparse-layer creation in SwitchTransformersConfig and missing auxiliary losses in D-FINE are unrelated." }, { "left": "issue:43824", "right": "issue:44933", "accept": false, - "reason": "Both are import-related, but they concern different missing symbols and different modules." + "reason": "Missing `Qwen2_5_VLForConditionalGeneration` export and a nonexistent `image_utils` import are different import bugs." }, { "left": "issue:43299", "right": "issue:45216", "accept": false, - "reason": "Different Qwen model families and different regressions: loading vs save_pretrained output." + "reason": "Qwen3VL loading and Qwen3.5 checkpoint saving are related only by family name, not the same fix." }, { "left": "issue:44075", "right": "issue:44368", "accept": false, - "reason": "Optimizer-arg handling and tie_word_embeddings warning are not the same code-path problem." + "reason": "Unused SGD args and a `tie_word_embeddings` warning are unrelated." }, { "left": "issue:43381", "right": "issue:45593", "accept": false, - "reason": "Gradient checkpointing eval-mode restriction and D-FINE auxiliary-loss behavior are unrelated." + "reason": "Gradient checkpointing in eval mode and D-FINE auxiliary-loss behavior do not share a code path." }, { "left": "issue:43901", "right": "issue:44496", "accept": false, - "reason": "Docs/behavior mismatch for a pipeline is unrelated to an unrecognized-model config error." + "reason": "Docs mentioning `return_all_scores` is unrelated to model/config loading failure." }, { "left": "issue:45335", "right": "issue:45584", "accept": false, - "reason": "Embedding resize bug and Whisper empty-transcription generation bug are different." + "reason": "Embedding resize propagation and Whisper empty-transcription generation are unrelated." }, { "left": "issue:44496", "right": "issue:45759", "accept": false, - "reason": "Both touch config loading, but one is missing model_type and the other is composite text_config unwrapping; distinct fixes." + "reason": "Unrecognized model/config parsing and `AutoModelForCausalLM.from_config` not unwrapping `text_config` are different issues." }, { "left": "issue:43479", "right": "issue:43525", "accept": false, - "reason": "Different configs and different attributes; no shared underlying bug." + "reason": "Phi4Multimodal default config initialization and missing `pad_token_id` on Llama4Config are separate config bugs." } ] }, @@ -10313,14 +10478,14 @@ "issue:43653|issue:45593", "issue:44623|issue:44749" ], - "prepared_review_unit_hash": "5728cbb67da3903004e81f4beb3adf42dad7a5c0266fe912eb191419490fda8b", + "prepared_review_unit_hash": "c5755cbebd0bbc5b68e9ddc55dc2096b9645978acc2737819b6a49c4f8de0723", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, @@ -10335,180 +10500,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly a grab bag of unrelated Transformers issues across tokenizers, multimodal processors, model loading, CI, and docs. None of the soft pairs look like true duplicate bugs or the same concrete fix path.", - "confidence": 0.87, - "canonical_issue_reason": "If one issue must anchor the cluster, issue 44521 is the best representative: it is an active, concrete multimodal bug with multiple inbound references. The rest of the set is too heterogeneous to collapse into a single canonical duplicate.", + "summary": "This cluster is heterogeneous: mostly unrelated Transformers bugs spanning tokenizers, multimodal processors, model loading/saving, quantization/offload, deepspeed, CI, and docs. I do not see any true duplicate pair that should be merged; the soft links are mostly broad topical similarity rather than the same underlying defect.", + "confidence": 0.96, + "canonical_issue_reason": "No single canonical issue cleanly represents the cluster because the items are about different code paths and regressions, not one duplicate bug.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 44521 is the most triage-worthy anchor because it is open, clearly scoped, and has the strongest discussion/reference signal among the listed issues.", + "best_issue_reason": "issue:44521 is the strongest single representative only because it is open, has the most inbound references, and is a concrete user-facing multimodal bug; it still does not subsume the rest of the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44521", "right": "issue:45325", "accept": false, - "reason": "Both are multimodal, but one is assistant-mask generation in apply_chat_template and the other is Qwen2.5-VL rope/position scaling; different code paths and symptoms." + "reason": "Different bugs: assistant mask generation in chat templates vs Qwen2.5-VL RoPE/temporal position scaling." }, { "left": "issue:44242", "right": "issue:45200", "accept": false, - "reason": "MoE load-balancing loss handling and Gemma 4 mm_token_type_ids defaults are separate bugs in different components." + "reason": "Different models and failure modes: MoE load-balancing loss logic vs Gemma4 mm_token_type_ids defaults." }, { "left": "issue:44351", "right": "issue:45278", "accept": false, - "reason": "A single missing import/export is not the same as a broad import-regression report across many symbols." + "reason": "A single HybridCache import error is not the same as broad import failures across many modules." }, { "left": "issue:41950", "right": "issue:45020", "accept": false, - "reason": "Video-classification pipeline processor lookup and remote_code loading regressions are different underlying problems." + "reason": "Video-classification pipeline processor lookup is unrelated to remote_code model loading regressions." }, { "left": "issue:44779", "right": "issue:45812", "accept": false, - "reason": "Both involve tokenizers, but they are model-specific correctness bugs in different tokenizer code paths." + "reason": "Both involve tokenizers, but they affect different model families and likely different code paths." }, { "left": "issue:41950", "right": "issue:45823", "accept": false, - "reason": "Pipeline processor selection and Gemma4 device placement mismatch are unrelated failures." + "reason": "Processor lookup bug in a pipeline is not the same as Gemma4 device-mapping mismatch in forward." }, { "left": "issue:43577", "right": "issue:45216", "accept": false, - "reason": "Blip2 dtype propagation and Qwen3.5 save_pretrained serialization are separate regressions." + "reason": "Loading dtype propagation and save_pretrained checkpoint serialization are distinct issues." }, { "left": "issue:43976", "right": "issue:45405", "accept": false, - "reason": "Python-version compatibility and an unreleased PEFT version bump are not the same issue." + "reason": "Python version compatibility and an unreleased dependency minimum are different release blockers." }, { "left": "issue:44683", "right": "issue:45542", "accept": false, - "reason": "Compiled flex_attention on new torch versions and missing TensorFlow backend handling are unrelated." + "reason": "Torch flex_attention compile failure and TensorFlow backend detection are unrelated." }, { "left": "issue:38175", "right": "issue:44617", "accept": false, - "reason": "SigLIP zero-probabilities and Sam3Video CUDA OOM are different models and failure modes." + "reason": "Unexpected zero probabilities and CUDA OOM are different symptoms with different likely causes." }, { "left": "issue:44279", "right": "issue:45405", "accept": false, - "reason": "A generic dependency issue is not the same as a specific minimum-version pin problem." + "reason": "Generic dependency issue is not the same as a specific PEFT version bump problem." }, { "left": "issue:39692", "right": "issue:43901", "accept": false, - "reason": "Both touch docs, but one is a broken SigLIP2 example and the other is stale TextClassificationPipeline parameter docs." + "reason": "Doc example errors in SigLIP2 are not the same as stale return_all_scores documentation." }, { "left": "issue:43299", "right": "issue:43854", "accept": false, - "reason": "Different model-loading failures affecting different models and test paths." + "reason": "Both are model-loading regressions, but for different models and different underlying failures." }, { "left": "issue:43295", "right": "issue:44779", "accept": false, - "reason": "Processor.tokenizer API regression and Deepseek token-ID corruption are different bugs." + "reason": "Processor.tokenizer regression and Deepseek tokenizer output regression are separate code paths." }, { "left": "issue:43650", "right": "issue:46132", "accept": false, - "reason": "A placeholder issue and an AttentionInterface.register behavior change are not duplicates." + "reason": "Placeholder issue title versus AttentionInterface.register behavior change; no shared bug." }, { "left": "issue:44462", "right": "issue:45081", "accept": false, - "reason": "Ignoring tokenizer.json in AutoTokenizer and a Mistral regex patch crash are different tokenizer-loading bugs." + "reason": "AutoTokenizer ignoring tokenizer.json and Mistral regex patch crashing are distinct tokenizer-loading problems." }, { "left": "issue:44297", "right": "issue:44462", "accept": false, - "reason": "Tokenizer class serialization mismatch and ignoring tokenizer.json share a theme but not the same concrete defect." + "reason": "Tokenizer save_pretrained metadata mismatch is not the same as AutoTokenizer preferring the wrong source file." }, { "left": "issue:44016", "right": "issue:44291", "accept": false, - "reason": "Notebook syntax error and init_empty_weights TypeError are unrelated." + "reason": "Notebook syntax error and init_empty_weights TypeError are unrelated failures." }, { "left": "issue:43450", "right": "issue:44279", "accept": false, - "reason": "Video batch shape handling and dependency resolution are unrelated." + "reason": "Batched video processor shape bug is unrelated to a dependency issue." }, { "left": "issue:45081", "right": "issue:45907", "accept": false, - "reason": "Tokenizer regex failure and list_repo_files revision forwarding are different code paths." + "reason": "Tokenizer regex crash and list_repo_files revision forwarding are different subsystems entirely." }, { "left": "issue:41950", "right": "issue:43650", "accept": false, - "reason": "These have no meaningful overlap in behavior or subsystem." + "reason": "A real video-classification bug is not related to a meaningless placeholder issue." }, { "left": "issue:43441", "right": "issue:45834", "accept": false, - "reason": "FlashAttention failures for Ministral and OCR index errors in Kosmos2.5 are distinct bugs." + "reason": "Ministral FlashAttention failure and Kosmos2.5 OCR index error are unrelated model-specific bugs." }, { "left": "issue:45600", "right": "issue:45712", "accept": false, - "reason": "Both are CI/cleanup-adjacent, but one is stale auto_mappings references and the other is leftover dummy class leakage." + "reason": "Removed config references in auto_mappings and leftover dummy class cleanup are separate CI issues." }, { "left": "issue:44466", "right": "issue:45276", "accept": false, - "reason": "lm_head serialization inconsistency and Gemma4 resize_token_embeddings behavior are separate model-specific issues." + "reason": "Inconsistent lm_head serialization and resize_token_embeddings behavior are different model-state bugs." }, { "left": "issue:42994", "right": "issue:43317", "accept": false, - "reason": "Quantized-model saving and dequantized-model offload loading are different stages with different failures." + "reason": "Quantized save failure and dequantized load/offload failure are different phases and code paths." }, { "left": "issue:33290", "right": "issue:45656", "accept": false, - "reason": "Deepspeed OOM with Adafactor and double optimizer-step behavior are not the same bug." + "reason": "Adafactor+deepspeed OOM and optimizer.step being called twice with deepspeed are not the same defect." }, { "left": "issue:43653", "right": "issue:45593", "accept": false, - "reason": "BigBirdTokenizer special-token registration and D-FINE auxiliary-loss handling are unrelated." + "reason": "BigBirdTokenizer special-token registration and D-FINE auxiliary-loss omission are unrelated." }, { "left": "issue:44623", "right": "issue:44749", "accept": false, - "reason": "Missing processor save files and a post-upgrade performance regression are different problems." + "reason": "Missing processor save files and a data-filtering slowdown after upgrade are different issues." } ] }, @@ -10611,202 +10776,202 @@ "issue:44336|issue:44373", "issue:44242|issue:45593" ], - "prepared_review_unit_hash": "ddffb35b9051c3cace4959efad5a074e8fd89707c41b97112e8c96e3e14af771", + "prepared_review_unit_hash": "164c246caf3163616da69b149c913d33bee3979c83af8728e1915825540f3ac6", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 35690, - "estimated_input_tokens": 8923, - "estimated_eval_tokens": 18102 + "serialized_chars": 35687, + "estimated_input_tokens": 8922, + "estimated_eval_tokens": 18100 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a loose collection of unrelated Transformers bugs/regressions; none of the soft pairs look like true duplicates. The best canonical issue is the open multimodal chat-template regression because it has the strongest activity and user impact signal in the set.", - "confidence": 0.88, - "canonical_issue_reason": "issue:44521 is the most suitable canonical issue: it is open, has strong discussion/inbound activity, and describes a concrete user-facing regression in multimodal `apply_chat_template` behavior.", + "summary": "The set is largely a grab-bag of distinct regressions across model loading, chat templates, processors, attention kernels, and docs. I did not find any soft pair that is clearly the same underlying bug/change closely enough to merge.", + "confidence": 0.61, + "canonical_issue_reason": "Issue #44521 is the clearest representative: it is a concrete, actively discussed regression in a core multimodal API, with multiple inbound refs and broad user impact.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44521 is the best overall issue to anchor the cluster because it is active, well-discussed, and narrowly framed around a reproducible bug rather than a doc or packaging complaint.", + "best_issue_reason": "#44521 is the best anchor because it is open, specific, and central to a widely used code path; it is more representative than niche model-specific or docs-only reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43650", "right": "issue:45290", "accept": false, - "reason": "Completely different bugs: a vague data request vs a crash in `apply_chat_template` with tool-call assistant messages." + "reason": "Completely different problems: a generic placeholder issue versus a chat-template crash on tool-call assistant messages." }, { "left": "issue:44623", "right": "issue:45397", "accept": false, - "reason": "Different code paths and symptoms: missing files from `processor.save_pretrained(...)` vs Zero3 `from_pretrained` loading." + "reason": "Different code paths: missing files on processor save versus ZeRO3/from_pretrained loading failure for Gemma4." }, { "left": "issue:43065", "right": "issue:44898", "accept": false, - "reason": "Different models and failures: `Sam3PixelDecoder` dummy conv vs Perceiver non-default resolution/interpolation." + "reason": "Both touch vision models, but one is a dummy Conv2d in Sam3PixelDecoder and the other is Perceiver interpolation failure; not the same bug." }, { "left": "issue:44261", "right": "issue:44263", "accept": false, - "reason": "Unrelated model bugs: missing `rms_norm_eps` precision issue vs `torch.split()` return-value handling." + "reason": "Unrelated model internals: MLA q_a_layernorm precision/config issue versus GlmMoeDsaIndexer torch.split return handling." }, { "left": "issue:43441", "right": "issue:44683", "accept": false, - "reason": "Both touch attention, but one is Ministral FlashAttention breakage and the other is compiled flex_attention on newer torch; not the same fix." + "reason": "Both mention attention, but one is a Ministral FlashAttention regression and the other is a torch>=2.9 compiled flex_attention failure." }, { "left": "issue:43723", "right": "issue:44162", "accept": false, - "reason": "Tokenizer loading regression vs an ESM2 model breakage; too different to merge." + "reason": "Tokenizer loading regression and ESM2 model breakage are different issues and likely different subsystems." }, { "left": "issue:42175", "right": "issue:43976", "accept": false, - "reason": "Packaging/backend dependency issue vs Python 3.9/3.10 compatibility problem." + "reason": "Package/backend dependency installation issue versus Python version compatibility breakage; not the same defect." }, { "left": "issue:44361", "right": "issue:45593", "accept": false, - "reason": "Different subsystems and symptoms: MLukeTokenizer task AttributeError vs D-FINE auxiliary-loss logic." + "reason": "Different behaviors in different components: MLukeTokenizer task attribute error versus D-FINE auxiliary-loss logic." }, { "left": "issue:44016", "right": "issue:45823", "accept": false, - "reason": "Notebook syntax error is unrelated to Gemma4 device placement at forward time." + "reason": "Notebook syntax error is unrelated to Gemma4 device-map forward mismatch." }, { "left": "issue:44671", "right": "issue:44779", "accept": false, - "reason": "Different failures: CamemBERT masked-LM predictions vs Deepseek tokenizer output regression." + "reason": "Different model families and failure modes: CamemBERT masked-LM predictions versus Deepseek tokenizer output regression." }, { "left": "issue:43262", "right": "issue:44521", "accept": false, - "reason": "Audio sampling-rate defaulting bug is separate from multimodal assistant-mask generation." + "reason": "Both involve chat-template-related APIs, but one is audio sampling-rate defaulting and the other is multimodal assistant-mask generation; distinct bugs." }, { "left": "issue:43673", "right": "issue:43761", "accept": false, - "reason": "Generation cache absence during chunked prefill is unrelated to CLIP vision hidden-state regression." + "reason": "Generation cache missing in chunked prefill is unrelated to CLIPVisionModel hidden_states handling." }, { "left": "issue:43901", "right": "issue:43994", "accept": false, - "reason": "One is a docs mismatch; the other is a model/pipeline correctness bug." + "reason": "Docs mismatch for return_all_scores is unrelated to SigLIP2 model/pipeline output quality." }, { "left": "issue:44987", "right": "issue:45542", "accept": false, - "reason": "Different underlying problems: loading a package/model under v5.1.0 vs missing TensorFlow backend from install extras." + "reason": "Model loading failure for physical-intelligence/fast is unrelated to the TensorFlow backend install problem." }, { "left": "issue:43784", "right": "issue:45542", "accept": false, - "reason": "Import-time `nn` NameError in sentence-transformers is unrelated to the TensorFlow backend installation issue." + "reason": "Sentence-transformers import NameError and missing TensorFlow backend are different dependency/import regressions." }, { "left": "issue:43824", "right": "issue:43976", "accept": false, - "reason": "Missing Qwen2.5-VL export/import symbol vs Python version support; not the same bug." + "reason": "ImportError for a specific Qwen VL class is unrelated to the Python version compatibility report." }, { "left": "issue:43450", "right": "issue:44617", "accept": false, - "reason": "Incorrect batched video shape is not the same as a CUDA out-of-memory issue." + "reason": "Video processor batched-shape bug and Sam3Video CUDA OOM are different problems, not one code-path fix." }, { "left": "issue:44297", "right": "issue:45397", "accept": false, - "reason": "Tokenizer metadata mismatch on save is unrelated to Gemma4 Zero3 loading." + "reason": "Tokenizer save_pretrained metadata mismatch is unrelated to Gemma4 zero3 loading." }, { "left": "issue:43525", "right": "issue:43881", "accept": false, - "reason": "Llama4 `pad_token_id` attribute error and GLM-4V loading failure are different concrete issues." + "reason": "Missing pad_token_id on Llama4Config and glm-4v-9b loading failure are not the same underlying bug." }, { "left": "issue:43901", "right": "issue:44246", "accept": false, - "reason": "Docs for `return_all_scores` are unrelated to intermittent slow imports." + "reason": "A docs discrepancy and an occasional slow import are unrelated." }, { "left": "issue:44188", "right": "issue:44336", "accept": false, - "reason": "Attention-kernel divergence on `torch.compile` is unrelated to ANSI code emission in `loading_report`." + "reason": "Attention-kernel divergence under torch.compile is unrelated to ANSI codes in loading_report." }, { "left": "issue:43525", "right": "issue:43643", "accept": false, - "reason": "Missing `pad_token_id` on Llama4Config is not the same as `trust_remote_code=True` dropping config fields." + "reason": "Both involve config-ish loading behavior, but one is a missing config attribute and the other is trust_remote_code returning incomplete fields; different defects." }, { "left": "issue:41628", "right": "issue:43976", "accept": false, - "reason": "Importing `AutoImageProcessor` and Python 3.9/3.10 support are unrelated failures." + "reason": "ImportError for AutoImageProcessor is unrelated to the Python version support issue." }, { "left": "issue:44589", "right": "issue:45362", "accept": false, - "reason": "Missing Float8 storage support vs Qwen3.5 chat loading crash; different code paths." + "reason": "Float8 storage lookup failure and Qwen3.5 chat crash are different runtime issues." }, { "left": "issue:45561", "right": "issue:45563", "accept": false, - "reason": "A pytest-xdist file race is unrelated to a stale warning in paged `generate()`." + "reason": "pytest-xdist file race and stale num_return_sequences warning are unrelated test/runtime behaviors." }, { "left": "issue:42491", "right": "issue:43299", "accept": false, - "reason": "Both mention Qwen3 MoE, but one is LoRA compatibility across hf4.x/hf5.x and the other is v5 dev loading failure for Qwen3VL; not the same concrete bug." + "reason": "Both mention Qwen3 MoE loading, but one is a LoRA model compatibility issue and the other is a Qwen3VL MoE loading regression; not clearly the same concrete bug." }, { "left": "issue:44336", "right": "issue:44373", "accept": false, - "reason": "ANSI-code output in a utility and a wrong docstring are clearly unrelated." + "reason": "Terminal ANSI-code leakage is unrelated to the wrong docstring for position_ids." }, { "left": "issue:44242", "right": "issue:45593", "accept": false, - "reason": "Load-balancing loss omission in MoE routing is unrelated to D-FINE auxiliary losses when denoising is off." + "reason": "Router-logits load balancing loss logic is unrelated to D-FINE auxiliary losses when denoising is off." } ] }, @@ -10912,220 +11077,220 @@ "issue:43232|issue:44336", "issue:44995|issue:45200" ], - "prepared_review_unit_hash": "876c0456505167bf7b4a9cd6b892faacc39e543c75b2f12e7740e33b5943ef52", + "prepared_review_unit_hash": "9d6ca5a83bd0ea7b7d318152f201b88a6c37e80aa720ec68ecfcd91a79f25b40", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 36862, - "estimated_input_tokens": 9216, - "estimated_eval_tokens": 18688 + "serialized_chars": 36859, + "estimated_input_tokens": 9215, + "estimated_eval_tokens": 18686 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "No soft-edge pair looks like a true duplicate; most are distinct model-/pipeline-specific bugs, config regressions, or unrelated import failures. The best issue anchor is the broad, open tokenization regression report with the strongest cross-user signals.", - "confidence": 0.84, - "canonical_issue_reason": "issue:45701 is the most broadly impactful and best-documented open report here: it describes a cross-version tokenization regression, has multiple inbound references, and is likely the most useful anchor for related follow-up reports.", + "summary": "This cluster is highly heterogeneous: most items are unrelated model-, tokenizer-, pipeline-, config-, or Deepspeed-specific bugs. I do not see any soft edge that safely represents the same underlying issue/change, so no merges are justified.", + "confidence": 0.43, + "canonical_issue_reason": "issue:45701 is the broadest open regression report with multiple inbound references; it is the best representative placeholder in an otherwise mixed cluster.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45701 is the strongest single issue to keep as the representative anchor because it is open, broadly scoped, and already referenced by multiple other threads.", + "best_issue_reason": "issue:45701 is the most globally useful issue here because it is broad, user-facing, and has the strongest evidence of attracting related reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44295", "right": "issue:45397", "accept": false, - "reason": "Different failures: buffer/position_ids access versus Gemma-4 ZeRO3 loading." + "reason": "Different failures: position_ids buffer access vs Gemma-4 Zero3 from_pretrained loading." }, { "left": "issue:44246", "right": "issue:44496", "accept": false, - "reason": "Unrelated symptoms: import slowdown versus unrecognized model/config loading." + "reason": "Generic import slowdown versus unrecognized-model config error; unrelated code paths." }, { "left": "issue:41950", "right": "issue:44220", "accept": false, - "reason": "Different pipeline/code paths: video classification processor lookup versus audio feature extraction." + "reason": "Both involve processors, but one is video-classification pipeline lookup and the other is fbank feature extraction." }, { "left": "issue:43531", "right": "issue:45362", "accept": false, - "reason": "Both involve Qwen, but one is sliding-window MoE behavior and the other is a chat crash; not the same bug." + "reason": "Both mention Qwen, but one is a sliding_window bug and the other is a chat crash; different underlying problems." }, { "left": "issue:42886", "right": "issue:46132", "accept": false, - "reason": "Offline tokenizer cache loading and AttentionInterface registration are unrelated." + "reason": "Offline tokenizer cache loading and AttentionInterface.register behavior are unrelated." }, { "left": "issue:43526", "right": "issue:45003", "accept": false, - "reason": "Image processor label reduction bug versus unsafe sys.modules access; different code paths." + "reason": "BeitImageProcessorFast label reduction bug is unrelated to unsafe sys.modules access." }, { "left": "issue:36246", "right": "issue:42757", "accept": false, - "reason": "Both are import errors, but for different symbols and packages." + "reason": "Different missing imports from different packages: Qwen image processor vs huggingface_hub helper." }, { "left": "issue:36246", "right": "issue:44351", "accept": false, - "reason": "Missing Qwen2_5_VLImageProcessor export and missing HybridCache export are distinct issues." + "reason": "Both are import errors, but they concern different symbols and different fixes." }, { "left": "issue:44336", "right": "issue:44485", "accept": false, - "reason": "ANSI terminal output bug is unrelated to GLM-5 RoPE implementation." + "reason": "ANSI-color output bug is unrelated to GLM-5 RoPE implementation." }, { "left": "issue:45003", "right": "issue:45561", "accept": false, - "reason": "Unsafe sys.modules access and pytest-xdist file races are different bugs." + "reason": "Unsafe sys.modules access and pytest-xdist file races are unrelated." }, { "left": "issue:44246", "right": "issue:44871", "accept": false, - "reason": "Import performance regression versus Gemma eos_token_id mismatch; no shared code path." + "reason": "Import performance regression vs Gemma eos_token_id config inconsistency." }, { "left": "issue:43352", "right": "issue:45250", "accept": false, - "reason": "Generic Flash Attention 2 topic is not the same as a specific Nemotron support gap." + "reason": "Generic Flash Attention topic, but one is a model-specific unsupported warning and the other is a broader feature request." }, { "left": "issue:43673", "right": "issue:44561", "accept": false, - "reason": "Generation cache/chunked_prefill bug and trust_remote_code FX helper removal are unrelated." + "reason": "Chunked-prefill cache bug and trust_remote_code breakage from removed helper are different issues." }, { "left": "issue:36246", "right": "issue:43596", "accept": false, - "reason": "ImportError for a model class is unrelated to a ZeRO3/BertModel index error." + "reason": "Qwen import failure and DeepSpeed/BertModel index error are unrelated." }, { "left": "issue:44242", "right": "issue:44857", "accept": false, - "reason": "MoE router loss behavior and DETR AMP crash are different failures." + "reason": "Load-balancing loss omission is unrelated to float16 AMP crash in LwDetrImageLoss." }, { "left": "issue:44779", "right": "issue:45292", "accept": false, - "reason": "Tokenizer regression and embedding-resize mismatch are separate issues." + "reason": "Tokenizer output regression and resize_token_embeddings not updating output embeddings are different bugs." }, { "left": "issue:44162", "right": "issue:45200", "accept": false, - "reason": "ESM2 breakage and Gemma 4 token-type defaults do not match." + "reason": "ESM2 breakage and Gemma-4 mm_token_type_ids defaulting are unrelated model issues." }, { "left": "issue:44295", "right": "issue:44749", "accept": false, - "reason": "Position_ids buffer handling is unrelated to the performance regression report." + "reason": "Position_ids buffer access and slower filtering after upgrade are not the same bug." }, { "left": "issue:43901", "right": "issue:45003", "accept": false, - "reason": "Docs mismatch in text classification pipeline versus sys.modules access bug." + "reason": "Documentation mismatch in TextClassificationPipeline is unrelated to sys.modules access." }, { "left": "issue:45412", "right": "issue:46123", "accept": false, - "reason": "RT-DETR memory release and MaskGenerationPipeline batching are unrelated." + "reason": "RT-DETR memory cleanup and MaskGenerationPipeline final-batch dropping are unrelated pipeline bugs." }, { "left": "issue:44496", "right": "issue:45003", "accept": false, - "reason": "Model config recognition error and sys.modules access issue are different code paths." + "reason": "Model config recognition failure is unrelated to sys.modules lookup safety." }, { "left": "issue:45292", "right": "issue:45584", "accept": false, - "reason": "Embedding resize bug and Whisper empty-transcription failure are unrelated." + "reason": "Embedding resize bug and Whisper empty-transcription failure are different code paths." }, { "left": "issue:45987", "right": "issue:46082", "accept": false, - "reason": "Cache length return type bug and LlamaConfig head_dim validation are distinct." + "reason": "Cache length return type and LlamaConfig head_dim validation are unrelated." }, { "left": "issue:44877", "right": "issue:45375", "accept": false, - "reason": "Different config-model problems in different models; not the same underlying bug." + "reason": "Both are config-validation issues, but they affect different models and different missing fields." }, { "left": "issue:45292", "right": "issue:45701", "accept": false, - "reason": "Embedding resize/output mismatch is unrelated to tokenization regression." + "reason": "Embedding resize behavior and version-dependent tokenization regression are not the same underlying issue." }, { "left": "issue:43257", "right": "issue:45137", "accept": false, - "reason": "Different DeepSpeed issues: weight conversion versus deque underflow." + "reason": "Both touch DeepSpeed, but one is Qwen3 MoE weight conversion and the other is a deque underflow during ZeRO3 init." }, { "left": "issue:43262", "right": "issue:44246", "accept": false, - "reason": "Audio chat-template sampling-rate bug is unrelated to import slowdown." + "reason": "Audio chat-template sampling-rate defaulting and import slowdown are unrelated." }, { "left": "issue:45375", "right": "issue:45759", "accept": false, - "reason": "One is strict config field handling, the other is composite multimodal config unwrapping." + "reason": "Missing vision-config field and text_config unwrapping for composite Qwen configs are distinct composite-model bugs." }, { "left": "issue:44987", "right": "issue:45459", "accept": false, - "reason": "Loading failure for a model repo and protobuf error masking are different issues." + "reason": "Loading a specific model under v5.1.0 and protobuf-masked tokenizer errors are unrelated." }, { "left": "issue:43232", "right": "issue:44336", "accept": false, - "reason": "Generation kwargs update bug is unrelated to ANSI code output." + "reason": "Generation kwargs update after sync_gpus and ANSI code emission are unrelated." }, { "left": "issue:44995", "right": "issue:45200", "accept": false, - "reason": "MoE stale cache crash and Gemma 4 token-type defaults are different bugs." + "reason": "GlmMoeDsa second-pass cache crash and Gemma-4 token type defaults are unrelated." } ] }, @@ -11232,20 +11397,20 @@ "issue:44038|issue:44623", "issue:46077|issue:46123" ], - "prepared_review_unit_hash": "00001a2116f63d2c52676439804911b6760733012c3394e271d6d62d02e6317f", + "prepared_review_unit_hash": "e14707550622f4978da4b5b86376520a0d0547b4195608b8ae980a76fdd464ff", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 32, - "serialized_chars": 37141, + "serialized_chars": 37143, "estimated_input_tokens": 9286, "estimated_eval_tokens": 18828 }, @@ -11254,204 +11419,198 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These are mostly thematically similar Transformers issues, but the soft pairs do not describe the same concrete bug or change. I would not merge any of the candidate pairs; the most representative issue is the broad v5 custom-model initialization regression.", - "confidence": 0.74, - "canonical_issue_reason": "issue:43646 is the broadest, most representative regression here: it describes a concrete Transformers v5 initialization break that could plausibly underlie several adjacent loading/config issues, even though the listed soft pairs are not true duplicates.", + "summary": "These items are largely unrelated issue reports across different models, tokenizer paths, generation flows, and serialization behaviors. I do not see any soft pair that is safely mergeable as the same underlying bug/change, so all soft edges should be rejected. If a single representative issue is needed, 45397 is the clearest standalone bug report with a concrete failing path.", + "confidence": 0.93, + "canonical_issue_reason": "issue:45397 is the most focused, actionable bug report here: a concrete `gemma-4 zero3 from_pretrained` failure with a narrow code path and likely direct fix.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43646 has the clearest, highest-level problem statement among the set and is the best umbrella representative for the cluster\u2019s v5 initialization/load regressions.", + "best_issue_reason": "issue:45397 is the best standalone issue candidate because it is specific, reproducible, and not just a broad symptom or regression description.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43475", "right": "issue:43540", "accept": false, - "reason": "Different models and failures: SAM3 vision encoder attribute error vs Qwen3OmniMoe video-processing ValueError." + "reason": "Both mention video/SAM/Qwen families, but they describe different model paths and different failures; not the same bug." }, { "left": "issue:43065", "right": "issue:43408", "accept": false, - "reason": "Both mention SAM3, but one is a pixel-decoder implementation detail and the other is a model-type/config mismatch warning." + "reason": "Same broad SAM3 area, but one is a dummy Conv2d concern and the other is a model-type mismatch warning; different underlying issues." }, { "left": "issue:42886", "right": "issue:45229", "accept": false, - "reason": "Unrelated bugs: offline tokenizer cache loading vs Gemma4 multi-GPU inference CUDA OOM." + "reason": "Tokenizer offline-cache loading and multi-GPU CUDA OOM are unrelated failure modes." }, { "left": "issue:44466", "right": "issue:45335", "accept": false, - "reason": "Both touch embeddings/serialization, but one is save-pretrained tied-weight serialization and the other is resize_token_embeddings not updating decoder embeddings." + "reason": "Both touch tied/embedding behavior, but one is about serialization inconsistency and the other about resize_token_embeddings not updating a decoder embedding; different code paths." }, { "left": "issue:43653", "right": "issue:44162", "accept": false, - "reason": "Different components and symptoms: BigBirdTokenizer special-token registration vs ESM2 being broadly broken." + "reason": "BigBirdTokenizer special-token decode bug and ESM2 breakage are unrelated." }, { "left": "issue:45397", "right": "issue:45656", "accept": false, - "reason": "Both involve training/runtime infrastructure, but one is gemma4 Zero3 from_pretrained and the other is a double optimizer-step bug with deepspeed." + "reason": "Gemma-4 from_pretrained on ZeRO3 and optimizer stepping twice under deepspeed are different bugs in different subsystems." }, { "left": "issue:44060", "right": "issue:44077", "accept": false, - "reason": "No shared concrete code path; tied-weights warning in Qwen3-Next is unrelated to patchtsmixer post_init eligibility." + "reason": "One is a tied-weights warning in Qwen3-Next, the other is a PatchTSMixer post_init policy issue; not the same problem." }, { "left": "issue:44995", "right": "issue:45588", "accept": false, - "reason": "Both crash in model internals, but the stale indexer cache bug and the flash_attention s_aux=None bug are distinct failures." + "reason": "Stale indexer cache on second forward pass and flash_attention None-handling crash are distinct runtime failures." }, { "left": "issue:44368", "right": "issue:44977", "accept": false, - "reason": "Similar model family, but one is a tied-embeddings warning during LoRA fine-tuning and the other is a flash-attention generation failure." + "reason": "A tie_word_embeddings warning during fine-tuning is not the same as a flash-attention generation failure." }, { "left": "issue:45137", "right": "issue:45600", "accept": false, - "reason": "Completely different areas: DeepSpeed ZeRO3 deque underflow vs removed Sam3LiteText configs breaking auto_mappings CI." + "reason": "DeepSpeed deque underflow and auto_mappings CI breakage are unrelated." }, { "left": "issue:44368", "right": "issue:45563", "accept": false, - "reason": "One is a tie_word_embeddings warning; the other is a stale num_return_sequences warning in paged generate." + "reason": "A tied-embedding warning and a stale num_return_sequences warning affect different generation logic." }, { "left": "issue:43673", "right": "issue:44734", "accept": false, - "reason": "Both involve generation cache, but one is missing cache in chunked_prefill and the other is a KV-cache continuation indexing crash." + "reason": "Both involve cache/generation, but one is missing cache in chunked prefill and the other is a KV-cache continuation indexing crash; not clearly the same bug." }, { "left": "issue:43931", "right": "issue:44368", "accept": false, - "reason": "Model weight-shape mismatch for Qwen3-VL-30B is unrelated to the tied-embeddings warning in a LoRA fine-tuning setup." + "reason": "Qwen3-VL weight-shape mismatch and a tied-weights warning are different issues." }, { "left": "issue:43065", "right": "issue:46123", "accept": false, - "reason": "SAM3 pixel-decoder dummy layer issue is unrelated to MaskGenerationPipeline dropping final partial batches." + "reason": "SAM3 pixel decoder initialization and MaskGenerationPipeline partial-batch handling are unrelated." }, { "left": "issue:44933", "right": "issue:45459", "accept": false, - "reason": "Both are import-related, but one is a missing import in image_utils and the other is tokenizer error masking when protobuf is absent." + "reason": "Nonexistent image_utils import and protobuf decode-error handling are different tokenizer/import problems." }, { "left": "issue:44869", "right": "issue:45356", "accept": false, - "reason": "Tokenizer/codec symptoms differ: Whisper word-timestamp decode crash vs Kimi-K2.5 codec regression and warning behavior." + "reason": "Whisper timestamp decode crash and Kimi-K2.5 tokenizer regression are unrelated tokenization bugs." }, { "left": "issue:44297", "right": "issue:44623", "accept": false, - "reason": "Both involve save_pretrained, but tokenizer_config mismatch is not the same bug as processor missing files." + "reason": "Tokenizer save_pretrained metadata mismatch and processor.save_pretrained missing files are both serialization-related but not the same concrete failure." }, { "left": "issue:43232", "right": "issue:43673", "accept": false, - "reason": "Both concern generation, but sync_gpus kwarg handling and missing cache in chunked_prefill are different code paths." + "reason": "Generation kwargs update after sync_gpus and missing cache in chunked prefill are separate generation-path issues." }, { "left": "issue:44361", "right": "issue:44749", "accept": false, - "reason": "An AttributeError in MLukeTokenizer tasks is unrelated to a performance slowdown after upgrading Transformers." + "reason": "MLukeTokenizer attribute errors and a performance regression after upgrade are unrelated." }, { "left": "issue:43646", "right": "issue:44162", "accept": false, - "reason": "The former is a custom-model initialization regression in v5; the latter is a broad ESM2 breakage with a different failure surface." + "reason": "Custom model initialization breakage and ESM2 breakage are different root problems." }, { "left": "issue:44871", "right": "issue:46132", "accept": false, - "reason": "Different concern entirely: eos_token_id config inconsistency vs AttentionInterface.register changing callback behavior." + "reason": "Gemma3 eos_token_id inconsistency and AttentionInterface.register behavior change are unrelated." }, { "left": "issue:43673", "right": "issue:45278", "accept": false, - "reason": "Both are v5 regressions, but one is GenerationMixin cache missing in chunked_prefill and the other is broad import errors after upgrade." + "reason": "A specific chunked_prefill cache bug and broad import errors after upgrade are too different to merge." }, { "left": "issue:44623", "right": "issue:45656", "accept": false, - "reason": "Processor save_pretrained missing files is unrelated to deepspeed calling optimizer.step twice." + "reason": "Processor save_pretrained file emission and optimizer double-step under deepspeed are unrelated." }, { "left": "issue:44749", "right": "issue:45656", "accept": false, - "reason": "Performance slowdown after upgrade is not the same bug as duplicate optimizer stepping under deepspeed." + "reason": "Performance slowdown in filtering data and double optimizer stepping are not the same bug." }, { "left": "issue:44749", "right": "issue:45310", "accept": false, - "reason": "One is a general slowdown report; the other is a Qwen3.5 Moe from_pretrained error on newer Transformers versions." + "reason": "Data filtering slowdown and Qwen3.5 Moe from_pretrained failure are unrelated." }, { "left": "issue:42994", "right": "issue:43482", "accept": false, - "reason": "Different model/save issues: quantized model saving failure vs Qwen2.5-GGUF loading failure in v5." + "reason": "Quantized model saving failure and GGUF loading failure are opposite directions and distinct code paths." }, { "left": "issue:43475", "right": "issue:45381", "accept": false, - "reason": "Both are video-related, but the SAM3 vision encoder attribute error and qwen2.5-vl vision_position_ids bug are distinct." + "reason": "Both concern video/VL models, but the reported attributes and failures differ; not a single concrete bug." }, { "left": "issue:28282", "right": "issue:30990", "accept": false, - "reason": "ImportError for missing PyTorch is not the same as a sentence-transformers loading hang." + "reason": "AutoModel missing PyTorch and Sentence Transformers hanging on load are unrelated environment/runtime issues." }, { "left": "issue:45230", "right": "issue:45397", "accept": false, - "reason": "Generic bug report / vague report vs a specific gemma-4 Zero3 from_pretrained failure; no shared concrete bug." - }, - { - "left": "issue:43475", - "right": "issue:43643", - "accept": false, - "reason": "SAM3 video encoder attribute error is unrelated to trust_remote_code AutoConfig missing fields." + "reason": "A placeholder generic bug report cannot be confidently merged with a specific Gemma-4 loading bug." }, { "left": "issue:44038", "right": "issue:44623", "accept": false, - "reason": "Qwen3-VL-Moe v5 bug and processor.save_pretrained missing files are different failures." + "reason": "Qwen3-VL-Moe from_pretrained error and processor.save_pretrained missing files are different failures." }, { "left": "issue:46077", "right": "issue:46123", "accept": false, - "reason": "Gemma4 causal-mask construction and MaskGenerationPipeline final-batch handling are different bugs, even though both affect generation." + "reason": "Bidirectional-attention mask construction and final partial-batch handling in MaskGenerationPipeline are unrelated." } ] }, @@ -11558,233 +11717,236 @@ "issue:43479|issue:44869", "issue:43278|issue:45137" ], - "prepared_review_unit_hash": "171ecf90b2aac0cd1be9bb28cd674db97cd1457fe226c559c6a90c935ff74b08", + "prepared_review_unit_hash": "c1494400716b455290108e593ba195e3d31b190270143f93c739b5153dd17d3b", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 32, - "serialized_chars": 37231, - "estimated_input_tokens": 9308, - "estimated_eval_tokens": 18872 + "serialized_chars": 37234, + "estimated_input_tokens": 9309, + "estimated_eval_tokens": 18874 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a loose collection of unrelated closed bug reports around Transformers models/imports. The only strong duplicate is the Qwen3.5 MoE from_pretrained regression (issues 45310 and 45478). No PRs are present.", - "confidence": 0.71, - "canonical_issue_reason": "Issue 45310 is the cleanest anchor: it is the earlier report for the same Qwen3.5 MoE from_pretrained error, with more discussion/inbound context than its duplicate 45478.", + "summary": "The packet is mostly heterogeneous, but it does contain two separate exact-duplicate issue families: the Qwen3.5 MoE `from_pretrained` error reports (45310\u219445478) and the Qwen3ForCausalLM VRAM leak reports (42673\u219443704). The Qwen3_5MoeVisionConfig config-field report (45375) is a different bug and should not be merged into the Qwen3.5 MoE family.", + "confidence": 0.86, + "canonical_issue_reason": "Issue 45310 is the best canonical issue for the Qwen3.5 MoE `from_pretrained` duplicate family: it clearly states the failure mode and is directly duplicated by 45478. The packet also contains a separate exact-duplicate family at 42673/43704, but 45310 is the stronger single canonical representative.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45310 best represents the only clear duplicate subcluster and is the most complete, discussion-rich report for that exact regression.", + "best_issue_reason": "45310 is the strongest single issue representative because it has an exact-title duplicate and a concrete, well-scoped bug description. It is a better canonical pick than the other Qwen3.5-related reports, while 42673 is best treated as the canonical for a separate duplicate family.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:41628", "right": "issue:44933", "accept": false, - "reason": "Both are import errors, but for different symbols and likely different fixes: AutoImageProcessor vs a missing image_utils export." + "reason": "Both are import-related, but they concern different missing symbols and different modules; not the same bug." }, { "left": "issue:42371", "right": "issue:45092", "accept": false, - "reason": "Unrelated topics: TF32 settings guidance versus a legacy remote-code/meta-init incompatibility for InternVL2." + "reason": "TF32 API guidance and InternVL2 meta-init incompatibility are unrelated failure modes." }, { "left": "issue:43475", "right": "issue:44661", "accept": false, - "reason": "Different bugs in different code paths: missing Sam3 vision output attribute vs add-new-model-like failing around TOKENIZER_MAPPING_NAMES." + "reason": "One is a missing attribute on SAM3 vision output; the other is a model-addition helper failure. Different code paths." }, { "left": "issue:42994", "right": "issue:43525", "accept": false, - "reason": "Quantized model saving failure is not the same issue as Llama4Config missing pad_token_id." + "reason": "Quantized model saving and missing `pad_token_id` in `Llama4Config` are different issues." }, { "left": "issue:45375", "right": "issue:45478", "accept": false, - "reason": "These mention the same model family, but the concrete bugs differ: missing config field vs from_pretrained error." + "reason": "45375 is a config-field report (`deepstack_visual_indexes`), not the same `from_pretrained` failure as 45478." }, { "left": "issue:45081", "right": "issue:45103", "accept": false, - "reason": "Different subsystems and failures: Mistral tokenizer regex patching vs auto_docstring/annotations crash." + "reason": "Both crash with `AttributeError`, but in different utilities and on different objects; not the same underlying bug." }, { "left": "issue:45310", "right": "issue:45375", "accept": false, - "reason": "Same family, but not the same underlying bug; one is a from_pretrained failure, the other is a missing config field silently dropped." + "reason": "45310 is a Qwen3.5 MoE `from_pretrained` error; 45375 reports a different config-field omission." }, { "left": "issue:43873", "right": "issue:44945", "accept": false, - "reason": "VRAM leak with multi-threaded Qwen3ForCausalLM is unrelated to incorrect LLM output under pipeline parallelism." + "reason": "Quantization offloading and pipeline-parallel output corruption are separate problems." }, { "left": "issue:41628", "right": "issue:44351", "accept": false, - "reason": "Both are import failures, but they concern different missing exports and different code paths." + "reason": "Both are import errors, but for different names and different release breakages." }, { "left": "issue:43232", "right": "issue:44188", "accept": false, - "reason": "Generation kwargs sync issue is unrelated to attention-kernel divergence under torch.compile." + "reason": "Generation kwargs propagation and attention-kernel divergence under `torch.compile` are unrelated." }, { "left": "issue:42673", "right": "issue:43704", - "accept": false, - "reason": "These have the same title, but without evidence they are distinct reports of the same VRAM leak; still, this looks like a duplicated issue pair only if the body confirms identical repro. The available metadata alone is not enough, so reject." + "accept": true, + "reason": "Exact same VRAM leak report for Qwen3ForCausalLM in multiple dataloader threads." }, { "left": "issue:42831", "right": "issue:44222", "accept": false, - "reason": "Both involve FP8, but one is an accuracy issue while the other is saving/pretrained serialization for MoE models." + "reason": "FineGrainedFP8 accuracy regression and FP8 `save_pretrained` moe are different symptoms and likely different fixes." }, { "left": "issue:44945", "right": "issue:45753", "accept": false, - "reason": "Pipeline parallelism output corruption is unrelated to a Qwen3.5 image-specific infinite loop." + "reason": "Pipeline parallelism output errors and Qwen3_5 infinite-loop-on-image are distinct bugs." }, { "left": "issue:36010", "right": "issue:41628", "accept": false, - "reason": "Different import targets and failure modes: GenerationMixin vs AutoImageProcessor." + "reason": "GenerationMixin import removal and AutoImageProcessor import failure are separate import breakages." }, { "left": "issue:44222", "right": "issue:45310", "accept": false, - "reason": "FP8 MoE save_pretrained and Qwen3.5 MoE from_pretrained are separate bugs despite both involving Qwen/MoE." + "reason": "FP8 `save_pretrained` moe and Qwen3.5 MoE `from_pretrained` are different issues." }, { "left": "issue:42994", "right": "issue:43824", "accept": false, - "reason": "Quantized model saving failure is unrelated to an import error for Qwen2.5-VLForConditionalGeneration." + "reason": "Quantized saving failure and Qwen2.5-VL import failure do not describe the same bug." }, { "left": "issue:44964", "right": "issue:45137", "accept": false, - "reason": "Model-loading incompatibility for Phi-4 multimodal is unrelated to a DeepSpeed ZeRO3 deque crash." + "reason": "Phi-4 multimodal loading and DeepSpeed ZeRO3 deque underflow are unrelated." }, { "left": "issue:42175", "right": "issue:45712", "accept": false, - "reason": "Packaging/backends dependency issue is unrelated to dummy PT objects leaking into dir(transformers)." + "reason": "Packaging dependency omission and dummy class cleanup are unrelated maintenance issues." }, { "left": "issue:43673", "right": "issue:44188", "accept": false, - "reason": "Generation cache missing during chunked prefill is a different failure than compile-time attention-kernel branching divergence." + "reason": "Generation cache missing in chunked prefill and attention-kernel branching divergence are different defects." }, { "left": "issue:43232", "right": "issue:44792", "accept": false, - "reason": "A generation kwargs update bug is unrelated to a failed janus image-generation test." + "reason": "Generation kwargs handling and janus image-generation test failure are not the same bug." }, { "left": "issue:44038", "right": "issue:45310", "accept": false, - "reason": "Qwen3-VL-Moe compatibility and Qwen3.5 MoE from_pretrained are different model families and different regressions." + "reason": "Both mention Qwen MoE, but they target different model variants and different version breakages; too broad to merge." }, { "left": "issue:44496", "right": "issue:45561", "accept": false, - "reason": "Unrecognized model/config loading error is unrelated to an xdist race in captured_info.txt." + "reason": "Model config recognition and xdist race on `captured_info.txt` are unrelated." }, { "left": "issue:43526", "right": "issue:45561", "accept": false, - "reason": "BeitImageProcessorFast label reduction bug is unrelated to the xdist captured_info.txt race." + "reason": "Fast image label reduction bug and xdist file race are different areas and failures." }, { "left": "issue:43504", "right": "issue:45561", "accept": false, - "reason": "Legacy Beit pretrained preset loading issue is unrelated to the xdist testing-utils race." + "reason": "Beit pretrained-load legacy-field issue is unrelated to the xdist race." }, { "left": "issue:43377", "right": "issue:44242", "accept": false, - "reason": "MIMI padding-mask discrepancy is unrelated to router-logits/load-balancing-loss behavior in MoE." + "reason": "MIMI padding-mask mismatch and router-logit load-balancing loss are different model behaviors." }, { "left": "issue:42222", "right": "issue:45200", "accept": false, - "reason": "VitPose model breakage is unrelated to Gemma 4 token-type ID defaults." + "reason": "VitPose breakage and Gemma 4 token-type defaulting are unrelated." }, { "left": "issue:42175", "right": "issue:44933", "accept": false, - "reason": "Package/backend dependency behavior is unrelated to the missing image_utils import." + "reason": "Backend packaging and missing image_utils export are different problems." }, { "left": "issue:42898", "right": "issue:44568", "accept": false, - "reason": "Tokenization-space cleanup behavior change is unrelated to mdeberta-v3 BOS/EOS special-token handling." + "reason": "Tokenizer cleanup behavior change and BOS/EOS special-token regression are related to tokenization, but not the same concrete failure." }, { "left": "issue:43976", "right": "issue:44933", "accept": false, - "reason": "Python version compatibility issue is unrelated to the image_utils import failure." + "reason": "Python version compatibility and missing image_utils import are unrelated." }, { "left": "issue:30990", "right": "issue:35141", "accept": false, - "reason": "Sentence Transformers loading hangs and embedding resize/reinit are completely different bugs." + "reason": "Sentence-Transformers loading hang and embedding reinitialization are distinct issues." }, { "left": "issue:43479", "right": "issue:44869", "accept": false, - "reason": "Phi4 multimodal config initialization and Whisper timestamp decode crash are unrelated." + "reason": "Multimodal config initialization and Whisper timestamp decode crash are unrelated." }, { "left": "issue:43278", "right": "issue:45137", "accept": false, - "reason": "Embedding dtype drift during evaluate is unrelated to the DeepSpeed ZeRO3 deque underflow." + "reason": "Embedding dtype drift during evaluation and DeepSpeed deque underflow are different bugs." } ] }, - "evaluator_result": null, - "evaluator_used": false, - "retried": false, - "accepted_nontrivial_soft_edge": false, + "evaluator_result": { + "accept": false, + "feedback": "The summary is a bit too strong for the packet. The Qwen3.5 pair is supported for issue:45310 \u2194 issue:45478 (identical titles), but issue:45375 is a different config-field report and is not grounded as the same bug. Also, the Qwen3ForCausalLM VRAM leak pair issue:42673 \u2194 issue:43704 is an exact-title duplicate in the packet, so the writeup should treat it separately and avoid implying there is only one substantial duplicate family. Several accepted soft-edge verdicts therefore look non-conservative." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -11888,250 +12050,244 @@ "issue:43723|issue:44977", "issue:44222|issue:44484" ], - "prepared_review_unit_hash": "34e5e61a2e3c4898e24e38ae6b21a977118f2027b5188edc673197841e50b195", + "prepared_review_unit_hash": "e62a25818d3d5fd749843b59b4b847541142639c998716fced81c7ac2fe86a85", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 36, - "serialized_chars": 38521, - "estimated_input_tokens": 9631, - "estimated_eval_tokens": 19518 + "serialized_chars": 38511, + "estimated_input_tokens": 9628, + "estimated_eval_tokens": 19512 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These issues are mostly unrelated; several cluster around v5/tokenizer regressions or import failures, but none are the same underlying bug. The closest representative for the cluster is the v5 tokenizer regression issue, not a true duplicate merge target.", - "confidence": 0.59, - "canonical_issue_reason": "issue:42898 is the broadest and most representative of the v5 tokenizer-regression subset; it has the clearest thematic overlap with other tokenizer behavior reports in this set.", + "summary": "This cluster is mostly heterogeneous: import errors, tokenizer regressions, training-loop bugs, config issues, and save/load problems. I don\u2019t see any soft-edge pair that clearly describes the same underlying bug/change, so all candidates should be kept separate.", + "confidence": 0.94, + "canonical_issue_reason": "issue:42898 is the broadest, highest-signal regression report in the set (v5 tokenizer behavior), so it is the closest thing to a cluster representative, though the cluster is not actually a duplicate set.", "canonical_pr_reason": null, - "best_issue_reason": "issue:42898 best fits as the cluster anchor because it captures a general v5 tokenization behavior regression, even though the rest of the items are not true duplicates.", + "best_issue_reason": "issue:42898 is the most central and reusable issue title among these mixed reports, but it should not be treated as a duplicate umbrella for the others.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43479", "right": "issue:44079", "accept": false, - "reason": "Different bugs: multimodal config defaults vs ModelOutput key assignment." + "reason": "Both involve None handling, but they affect different classes and code paths (config defaults vs ModelOutput key assignment)." }, { "left": "issue:44623", "right": "issue:45310", "accept": false, - "reason": "Different code paths: processor save_pretrained output files vs Qwen3.5 MoE loading failure." + "reason": "Different features and failures: processor save_pretrained file output vs Qwen3.5 MoE from_pretrained loading." }, { "left": "issue:42898", "right": "issue:45812", "accept": false, - "reason": "Both are tokenization regressions, but different models and failure modes; not the same bug." + "reason": "Both are tokenizer regressions, but they concern different models and distinct behavior changes." }, { "left": "issue:44188", "right": "issue:44373", "accept": false, - "reason": "Attention-kernel branching bug vs docstring typo; unrelated." + "reason": "Attention-kernel divergence and a docstring issue are unrelated." }, { "left": "issue:30333", "right": "issue:33290", "accept": false, - "reason": "MLflow reporting on failed training vs DeepSpeed Adafactor OOM; unrelated." + "reason": "MLFlow reporting on failure and a DeepSpeed OOM in Adafactor are different training bugs." }, { "left": "issue:43784", "right": "issue:43824", "accept": false, - "reason": "Different import failures for different symbols/packages." + "reason": "Both are import errors, but for different symbols, packages, and root causes." }, { "left": "issue:44734", "right": "issue:45103", "accept": false, - "reason": "Serve KV-cache indexing crash vs auto-docstring annotations crash; unrelated." + "reason": "Serving KV-cache indexing and auto_docstring annotation handling are unrelated code paths." }, { "left": "issue:43723", "right": "issue:44568", "accept": false, - "reason": "Both tokenizer-related, but one is loading in v5 and the other is add_special_tokens behavior for a specific model." + "reason": "Both hit tokenizer behavior in v5, but one is a loading issue and the other is special-token insertion semantics." }, { "left": "issue:42898", "right": "issue:44779", "accept": false, - "reason": "Both tokenization regressions, but they affect different models and distinct logic." + "reason": "Different tokenizer regressions for different model families; not the same concrete bug." }, { "left": "issue:42898", "right": "issue:45356", "accept": false, - "reason": "Tokenizer regression in v5, but Kimi-K2.5 codec handling is a separate bug." + "reason": "Both are tokenizer regressions, but the reported codec/warning issue is distinct from clean_up_tokenization_spaces behavior." }, { "left": "issue:43874", "right": "issue:45092", "accept": false, - "reason": "Image processor API missing method vs remote-code checkpoint/meta-init incompatibility." + "reason": "Different multimodal loading failures affecting different architectures and mechanisms." }, { "left": "issue:44060", "right": "issue:45125", "accept": false, - "reason": "Wrong tied-weights warning vs missing tensor-parallel plan; different issues." + "reason": "Qwen model-specific tied-weight warning and tensor-parallel plan omission are separate issues." }, { "left": "issue:45405", "right": "issue:45712", "accept": false, - "reason": "Dependency version bump issue vs leftover dummy class cleanup; unrelated." + "reason": "A dependency version bump and leftover dummy classes are unrelated repository hygiene problems." }, { "left": "issue:42898", "right": "issue:44448", "accept": false, - "reason": "Tokenizer cleanup regression vs model output difference on Pegasus; different failures." + "reason": "Both are v4\u2192v5 behavioral regressions, but they affect different models and output paths." }, { "left": "issue:44295", "right": "issue:45656", "accept": false, - "reason": "Position_ids buffer read error vs DeepSpeed double optimizer step; unrelated." - }, - { - "left": "issue:44448", - "right": "issue:46132", - "accept": false, - "reason": "Pegasus v4/v5 output regression vs AttentionInterface.register behavior change; unrelated." + "reason": "Buffer access/readback and optimizer stepping with DeepSpeed are unrelated bugs." }, { "left": "issue:44933", "right": "issue:45712", "accept": false, - "reason": "Import path issue from image_utils vs dummy_pt_objects cleanup; not the same bug." + "reason": "Public import exposure from image_utils and dummy-class leakage into dir(transformers) are different export problems." }, { "left": "issue:44222", "right": "issue:44297", "accept": false, - "reason": "FP8 save_pretrained issue vs tokenizer metadata mismatch; unrelated save_pretrained problems." + "reason": "Both involve save_pretrained, but one is MoE FP8 saving and the other is tokenizer config metadata." }, { "left": "issue:43116", "right": "issue:44190", "accept": false, - "reason": "Different example scripts and different failure modes." + "reason": "Different example scripts and different failures: multi-label inference output vs local dataset loading." }, { "left": "issue:44297", "right": "issue:45656", "accept": false, - "reason": "Tokenizer save_pretrained metadata vs optimizer stepping bug; unrelated." + "reason": "Tokenizer save metadata and DeepSpeed optimizer stepping are unrelated." }, { "left": "issue:43479", "right": "issue:43644", "accept": false, - "reason": "Different config/buffer initialization bugs; not the same underlying change." + "reason": "Both mention initialization/defaults, but one is multimodal config defaults and the other is non-persistent buffer filling." }, { "left": "issue:44190", "right": "issue:46082", "accept": false, - "reason": "Dataset loading in an example script vs LlamaConfig validation; unrelated." + "reason": "Dataset loading in an example script and LlamaConfig validation are unrelated." }, { "left": "issue:42831", "right": "issue:44623", "accept": false, - "reason": "FP8 accuracy issue vs processor save_pretrained missing files; unrelated." + "reason": "FP8 accuracy and processor save_pretrained file emission are different concerns." }, { "left": "issue:30333", "right": "issue:36331", "accept": false, - "reason": "MLflow job reporting vs CustomTrainer loss signature change; unrelated." + "reason": "MLFlow reporting on failed jobs and Trainer compute_loss signature mismatch are separate training-loop regressions." }, { "left": "issue:44222", "right": "issue:44623", "accept": false, - "reason": "Both involve save_pretrained, but different components and failure modes." + "reason": "Both touch save_pretrained, but the artifacts and failure modes are different." }, { "left": "issue:45245", "right": "issue:46132", "accept": false, - "reason": "Category limit runtime error vs register behavior change; unrelated." + "reason": "Category-count limits and AttentionInterface registration behavior are unrelated." }, { "left": "issue:30333", "right": "issue:34689", "accept": false, - "reason": "Training-failure MLflow reporting vs Llama 3.2 vision loading regression; unrelated." + "reason": "Training-job failure reporting and Llama Vision model loading are unrelated." }, { "left": "issue:41628", "right": "issue:42757", "accept": false, - "reason": "Different missing-import regressions from different packages." + "reason": "Import errors in different dependency namespaces with different missing symbols." }, { "left": "issue:22355", "right": "issue:28282", "accept": false, - "reason": "Both are import errors, but one is missing transformers.onnx and the other is missing PyTorch support." + "reason": "Both are import/setup failures, but one is a missing transformers submodule and the other is missing PyTorch." }, { "left": "issue:44206", "right": "issue:45103", "accept": false, - "reason": "Feature extractor argument bug vs auto-docstring annotations crash; unrelated." + "reason": "Feature extractor argument compatibility and docstring processing are unrelated." }, { "left": "issue:45071", "right": "issue:46082", "accept": false, - "reason": "PretrainedConfig typing regression vs LlamaConfig head_dim validation; different issues." + "reason": "Config type-checking in v5.4 and explicit head_dim validation are different config issues." }, { "left": "issue:28282", "right": "issue:30333", "accept": false, - "reason": "Import-time dependency failure vs MLflow reporting bug; unrelated." + "reason": "Missing PyTorch import and MLFlow reporting on failed training are unrelated." }, { "left": "issue:30333", "right": "issue:35141", "accept": false, - "reason": "Failed-training tracking bug vs embedding resize/post_init reinit bug; unrelated." + "reason": "Training failure reporting and embedding reinitialization after resize are different bugs." }, { "left": "issue:44297", "right": "issue:45310", "accept": false, - "reason": "Tokenizer config metadata bug vs Qwen3.5 MoE loading error; unrelated." + "reason": "Tokenizer save metadata and Qwen3.5 MoE loading are unrelated." }, { "left": "issue:43723", "right": "issue:44977", "accept": false, - "reason": "Tokenizer loading regression vs flash-attention generation bug; unrelated." + "reason": "Both are tokenizer-related regressions, but one is AutoTokenizer loading and the other is DeepSeek tokenization correctness." }, { "left": "issue:44222", "right": "issue:44484", "accept": false, - "reason": "FP8 save_pretrained vs shard-size question; not a duplicate bug." + "reason": "FP8 save_pretrained and max_shard_size documentation/behavior are not the same bug." } ] }, @@ -12234,20 +12390,20 @@ "issue:42886|issue:43066", "issue:22355|issue:28218" ], - "prepared_review_unit_hash": "6697ae0b1dbf9784c8b78b8df4f1319ea58b05d6f37dab41457a259e16930108", + "prepared_review_unit_hash": "41d4c6bf7abd1fd83f5eb2239cf8e08f390bb3602c2ec9018366c9d34090af64", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 29, - "serialized_chars": 35478, + "serialized_chars": 35480, "estimated_input_tokens": 8870, "estimated_eval_tokens": 17996 }, @@ -12256,186 +12412,186 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a set of unrelated Hugging Face/Transformers issues that only look superficially similar by shared keywords (tokenizer, import, save/load, config). I found no soft pairs that clearly describe the same underlying bug or change, so all candidate edges should be rejected. The most representative issue is the broad tokenizer-loading regression report.", - "confidence": 0.74, - "canonical_issue_reason": "issue:43723 is the broadest and most central description of the recurring tokenizer-loading regression theme, so it is the best canonical issue for this cluster.", + "summary": "These items are mostly unrelated bug reports; the soft pairs are only superficially similar by subsystem/topic and do not look like the same concrete underlying bug or change. The most central issue is the v5 tokenizer-loading regression cluster.", + "confidence": 0.79, + "canonical_issue_reason": "issue:43723 is the broadest and most representative tokenizer-loading regression report, and it has the strongest thematic overlap with multiple tokenizer-related soft candidates.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43723 is the best overall representative because it captures the widest recurring user-facing failure mode (tokenizer loading in v5) without being overly model-specific.", + "best_issue_reason": "issue:43723 is the best single issue to anchor this cluster because it captures the recurring v5 tokenizer-loading failure pattern better than the narrower, model-specific reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43232", "right": "issue:45834", "accept": false, - "reason": "Different failures: generation kwargs handling vs Kosmos2.5 OCR length/index error." + "reason": "Different failures: generation kwargs handling after sync_gpus vs Kosmos2.5 OCR index error." }, { "left": "issue:44263", "right": "issue:45092", "accept": false, - "reason": "Unrelated models/code paths: GlmMoeDsaIndexer split behavior vs InternVL2 meta-initialization incompatibility." + "reason": "Unrelated bugs: torch.split return values in a DSA indexer vs old InternVL2 checkpoints failing under v5 meta init." }, { "left": "issue:43761", "right": "issue:46082", "accept": false, - "reason": "Different components: CLIP vision forward hidden states vs LlamaConfig head_dim validation." + "reason": "Different model behaviors: CLIPVisionModel hidden_states regression vs LlamaConfig head_dim validation." }, { "left": "issue:44483", "right": "issue:45406", "accept": false, - "reason": "Different runtime surfaces: chat-completions API rejection vs Gemma4Processor missing _tokenizer." + "reason": "Both are v5-era failures, but one is /v1/chat/completions request handling and the other is a Gemma4Processor attribute error." }, { "left": "issue:44373", "right": "issue:44485", "accept": false, - "reason": "A docstring issue is not the same bug as a GLM-5 RoPE implementation problem." + "reason": "Docstring typo for position_ids is not the same bug as GLM-5 RoPE implementation." }, { "left": "issue:44295", "right": "issue:44623", "accept": false, - "reason": "Buffer/position_ids access error and processor.save_pretrained file omission are distinct issues." + "reason": "Registering position_ids as a buffer causing read errors is distinct from processor.save_pretrained missing files." }, { "left": "issue:44263", "right": "issue:44360", "accept": false, - "reason": "Both mention DSA/GLM components, but one is a split-return issue and the other is a missing ReLU; not the same bug." + "reason": "Same general DSA indexer area, but different concrete issues: split outputs vs missing ReLU." }, { "left": "issue:44514", "right": "issue:45103", "accept": false, - "reason": "Qwen2_5_VL batched chat_template crash is unrelated to auto_docstring failures with future annotations." + "reason": "Both involve processor/template code, but one is batched apply_chat_template padding behavior and the other is auto_docstring crashing on future annotations." }, { "left": "issue:43723", "right": "issue:44871", "accept": false, - "reason": "Both are tokenizer-related, but one is a generic v5 loading regression and the other is a Gemma eos_token_id mismatch." + "reason": "Tokenizer loading regression vs eos_token_id config inconsistency; related area but different underlying bug." }, { "left": "issue:30990", "right": "issue:38175", "accept": false, - "reason": "Sentence Transformers loading stall and SigLIP2 zero probabilities are different failures." + "reason": "Sentence Transformers loading hang is unrelated to SigLIP2 zero-probability outputs." }, { "left": "issue:45092", "right": "issue:45357", "accept": false, - "reason": "Meta-init incompatibility for old InternVL2 checkpoints is not the same as incorrect visual encoder keys on save_pretrained." + "reason": "Meta-initialization incompatibility for old InternVL2 checkpoints is different from incorrect visual encoder keys on save_pretrained." }, { "left": "issue:42907", "right": "issue:46050", "accept": false, - "reason": "Both concern quantized models, but one is dequantized model saving and the other is sinq save/load breakage; too different to merge." + "reason": "Both mention quantization/save issues, but failing to save dequantized Ministrals/Devstrals is not the same as sinq quantized save_pretrained/from_pretrained breakage." }, { "left": "issue:43723", "right": "issue:44991", "accept": false, - "reason": "Both involve tokenizer loading in v5, but the titles suggest a broad framework regression vs a specific model checkpoint failure; not enough to treat as the same bug." + "reason": "Both are tokenizer-load regressions, but the affected models and failure modes are not specific enough to treat as the same bug." }, { "left": "issue:44829", "right": "issue:46002", "accept": false, - "reason": "Degenerate training with flash_attention_3 is unrelated to NaNs after checkout/init." + "reason": "Degenerate training with flash_attention_3 is unrelated to NaNs after checkout + init." }, { "left": "issue:43723", "right": "issue:44442", "accept": false, - "reason": "AutoTokenizer failing for FastSpeech2ConformerTokenizer is a specific model import/load problem, not clearly the same as the generic v5 tokenizer-loading issue." + "reason": "AutoTokenizer failing on FastSpeech2ConformerTokenizer may be related by theme, but it is not clearly the same bug as issue loading tokenizer in v5." }, { "left": "issue:44261", "right": "issue:45092", "accept": false, - "reason": "MLA q_a_layernorm epsilon precision mismatch is unrelated to InternVL2 meta initialization." + "reason": "MLA q_a_layernorm precision error and old InternVL2 meta-init incompatibility are different classes of bugs." }, { "left": "issue:41628", "right": "issue:43824", "accept": false, - "reason": "Both are import errors, but for different symbols and likely different missing-export causes." + "reason": "Both are import errors, but for different symbols/models and likely different causes." }, { "left": "issue:42831", "right": "issue:45310", "accept": false, - "reason": "FineGrainedFP8 accuracy issues and Qwen3.5 Moe from_pretrained errors are different problems." + "reason": "FineGrainedFP8 accuracy issue is not the same as Qwen3.5 Moe from_pretrained error." }, { "left": "issue:43479", "right": "issue:44913", "accept": false, - "reason": "Phi4MultimodalConfig None-handling and GPTNeoXConfig rotary_pct reload behavior do not match." + "reason": "Phi4MultimodalConfig None-handling and GPTNeoXConfig rotary_pct reload behavior are unrelated configuration bugs." }, { "left": "issue:44263", "right": "issue:44373", "accept": false, - "reason": "torch.split return-value handling and a wrong docstring are not the same underlying change." + "reason": "Indexer output handling vs a wrong docstring are not the same issue." }, { "left": "issue:42831", "right": "issue:44038", "accept": false, - "reason": "Different model families and failure modes: FP8 accuracy vs Qwen3-VL-Moe bug." + "reason": "FineGrainedFP8 accuracy and Qwen3-VL-Moe bug are different model-specific problems." }, { "left": "issue:39401", "right": "issue:44568", "accept": false, - "reason": "Both are tokenizer bugs, but one is offset_mapping and the other is add_special_tokens/BOS-EOS behavior; not the same defect." + "reason": "Qwen3 wrong offset_mapping and mdeberta-v3 add_special_tokens BOS/EOS regression are distinct tokenizer bugs." }, { "left": "issue:42757", "right": "issue:45278", "accept": false, - "reason": "A specific huggingface_hub import error is not the same as a broad upgrade-related import breakage report." + "reason": "Both mention imports after version changes, but one is a specific missing huggingface_hub symbol and the other is a broad import-error report." }, { "left": "issue:42886", "right": "issue:42994", "accept": false, - "reason": "Offline tokenizer cache loading and quantized model saving failures are unrelated." + "reason": "Offline cache loading failure is a different problem from quantized model saving failure." }, { "left": "issue:43232", "right": "issue:45593", "accept": false, - "reason": "Generation kwarg updates after sync_gpus and D-FINE auxiliary-loss behavior are unrelated." + "reason": "Generation kwargs handling and D-FINE auxiliary-loss behavior are unrelated." }, { "left": "issue:42831", "right": "issue:45753", "accept": false, - "reason": "FineGrainedFP8 accuracy regression and Qwen3_5 image infinite-loop behavior are different code paths." + "reason": "FineGrainedFP8 accuracy issue and Qwen3_5 infinite loop on a specific image are unrelated." }, { "left": "issue:43756", "right": "issue:44060", "accept": false, - "reason": "Smollm3 RoPE-layer drop and Qwen3-Next tied-weights warning are distinct model/config issues." + "reason": "Smollm3 RoPE layer mismatch is not the same as Qwen3-Next tied-weights warning." }, { "left": "issue:42886", "right": "issue:43066", "accept": false, - "reason": "Offline cache-tokenizer loading and tokenizer decoder-type mismatch are different tokenizer regressions." + "reason": "Tokenizer cache/offline loading and wrong tokenizer decoder type are different bugs." }, { "left": "issue:22355", "right": "issue:28218", "accept": false, - "reason": "No module named transformers.onnx is unrelated to added-token spacing behavior." + "reason": "Missing transformers.onnx import and extra-space after added token are unrelated tokenizer/module issues." } ] }, @@ -12542,14 +12698,14 @@ "issue:36296|issue:43116", "issue:44038|issue:44222" ], - "prepared_review_unit_hash": "d9f79878ca7487ab5e8fcdcffdb91f83ac9140f030411006eba359d4309fb846", + "prepared_review_unit_hash": "239cec6d47de06d6b3bb5aaeb8d04e548c3a9cdec0a7288ea02505445f35b4f2", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, @@ -12564,204 +12720,198 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The candidate pairs are mostly semantically related only at a very broad subsystem level; none look like the same concrete underlying bug or change, so this cluster is not a true duplicate set.", - "confidence": 0.93, - "canonical_issue_reason": "No clear canonical issue: the items span unrelated tokenizer, import, model-loading, training, and generation bugs.", + "summary": "This cluster is heterogeneous: most items are unrelated issue reports spanning tokenization, model loading, generation, import errors, and model-specific regressions. I would not merge any of the soft pairs, and there are no PRs to consider.", + "confidence": 0.86, + "canonical_issue_reason": "issue:43122 is the broadest tokenizer-regression report here and the closest thing to a representative topic anchor, though the cluster is too mixed to treat as a true duplicate set.", "canonical_pr_reason": null, - "best_issue_reason": "No single issue is a good global representative for this cluster because the soft pairs are not duplicates and do not share one concrete code-path problem.", + "best_issue_reason": "issue:43122 is the most reusable issue to keep as a representative for the tokenizer-regression subset; it has the widest apparent topical overlap among the listed issues.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44671", "right": "issue:46132", "accept": false, - "reason": "Different areas: masked LM predictions vs AttentionInterface.register behavior." + "reason": "Different bugs: masked-LM prediction quality vs AttentionInterface.register changing callback behavior." }, { "left": "issue:43576", "right": "issue:43784", "accept": false, - "reason": "CLI/env command issue vs import-time NameError in sentence-transformers." + "reason": "Both are import/loading failures, but one is the v5 env command and the other is a sentence-transformers import NameError." }, { "left": "issue:43122", "right": "issue:44521", "accept": false, - "reason": "Tokenizer output drift across versions vs multimodal assistant_mask bug." + "reason": "Both involve tokenization/chat handling, but one is a versioned tokenizer regression and the other is multimodal assistant mask generation." }, { "left": "issue:43646", "right": "issue:46002", "accept": false, - "reason": "Custom model init breakage vs NaNs in classification heads; not the same failure." + "reason": "Custom model initialization failure is not the same concrete bug as NaNs in classification heads after checkout/init." }, { "left": "issue:44117", "right": "issue:44704", "accept": false, - "reason": "TOKENIZER_MAPPING_NAMES None handling vs AutoProcessor kwargs propagation." + "reason": "TOKENIZER_MAPPING_NAMES handling and AutoProcessor kwargs forwarding are unrelated code paths." }, { "left": "issue:30990", "right": "issue:33357", "accept": false, - "reason": "Loading hang vs Mac bus error; unrelated symptoms and code paths." + "reason": "Sentence-transformers loading stall and a MacOS bus error on a CLIP model are different failures." }, { "left": "issue:42371", "right": "issue:44263", "accept": false, - "reason": "TF32 settings API vs torch.split behavior in a GLM indexer." + "reason": "TF32 settings guidance and a torch.split return-value bug in GlmMoeDsaIndexer are unrelated." }, { "left": "issue:44295", "right": "issue:44297", "accept": false, - "reason": "position_ids buffer/read issue vs tokenizer_class mismatch on save." + "reason": "Reading position_ids after buffer registration is unrelated to tokenizer_class mismatch on save_pretrained." }, { "left": "issue:43232", "right": "issue:44743", "accept": false, - "reason": "Generation kwargs sync issue vs recurrent-state reset bug in a different model file." + "reason": "Both touch generation/cache logic, but they describe different failures in different code paths." }, { "left": "issue:41628", "right": "issue:45278", "accept": false, - "reason": "Specific AutoImageProcessor import failure vs a broader set of import errors after upgrade." + "reason": "A specific missing AutoImageProcessor import is not the same as broad import errors after upgrade." }, { "left": "issue:44704", "right": "issue:45290", "accept": false, - "reason": "cached_file kwargs forwarding vs apply_chat_template crash on tool-call messages." + "reason": "cached_file kwarg forwarding and apply_chat_template crashes are distinct processor issues." }, { "left": "issue:42907", "right": "issue:44945", "accept": false, - "reason": "Dequantized save_pretrained bug vs incorrect output under pipeline parallelism." + "reason": "Saving dequantized models and incorrect output under pipeline parallelism are unrelated." }, { "left": "issue:43122", "right": "issue:44514", "accept": false, - "reason": "General tokenization change vs batched apply_chat_template padding crash." + "reason": "Both mention tokenizer/template behavior, but one is a tokenizer regression across versions and the other is a batched apply_chat_template crash." }, { "left": "issue:43784", "right": "issue:45600", "accept": false, - "reason": "Import NameError in downstream package vs removed config references in auto_mappings." + "reason": "Importing sentence-transformers with latest transformers and broken auto_mappings CI are different import-related regressions." }, { "left": "issue:39401", "right": "issue:44488", "accept": false, - "reason": "Qwen3 offset_mapping bug vs model loading failure for a different checkpoint." + "reason": "Wrong offset_mapping in Qwen3 tokenizer is unrelated to failing to load a specific model." }, { "left": "issue:44261", "right": "issue:44373", "accept": false, - "reason": "RMS norm epsilon precision bug vs incorrect docstring; not the same issue." + "reason": "A precision/config issue in MLA q_a_layernorm is not the same as a wrong docstring for position_ids." }, { "left": "issue:43976", "right": "issue:44945", "accept": false, - "reason": "Python version compatibility problem vs pipeline-parallel output corruption." + "reason": "Python version compatibility breakage and pipeline-parallel output corruption are unrelated." }, { "left": "issue:43408", "right": "issue:43475", "accept": false, - "reason": "sam3_video vs sam3_tracker warning mismatch and missing attribute error are different bugs." + "reason": "Both are SAM3-related, but one is a model-type warning and the other is a missing output attribute." }, { "left": "issue:42907", "right": "issue:43937", "accept": false, - "reason": "Saving dequantized models vs invalid GenerationConfig for GLM-5." + "reason": "Dequantized save failures and invalid GenerationConfig are different bugs." }, { "left": "issue:44360", "right": "issue:44485", "accept": false, - "reason": "DSA indexer activation question vs GLM-5 RoPE implementation bug." + "reason": "A missing ReLU in the DSA indexer and GLM-5 RoPE implementation are not the same issue." }, { "left": "issue:39401", "right": "issue:44521", "accept": false, - "reason": "Qwen3 tokenizer offset mapping vs multimodal assistant masks; different tokenization failures." + "reason": "Wrong offset_mapping and all-zero assistant masks are different tokenizer/chat-template failures." }, { "left": "issue:44625", "right": "issue:44829", "accept": false, - "reason": "num_labels propagation vs flash_attention_3 training degeneration." + "reason": "num_labels propagation and flash_attention_3 degenerate training are unrelated." }, { "left": "issue:43116", "right": "issue:45200", "accept": false, - "reason": "Example script classification bug vs Gemma 4 token-type id defaulting." + "reason": "Empty multi-label classification results and Gemma 4 mm_token_type_ids defaults are different problems." }, { "left": "issue:42907", "right": "issue:44964", "accept": false, - "reason": "Saving dequantized models vs loading a specific multimodal model." + "reason": "Dequantized save_pretrained failure is unrelated to loading Phi-4 multimodal." }, { "left": "issue:44038", "right": "issue:44297", "accept": false, - "reason": "Qwen3-VL-Moe model bug vs tokenizer_class metadata mismatch." + "reason": "Qwen3-VL-Moe bug and tokenizer_class mismatch on save are unrelated." }, { "left": "issue:28218", "right": "issue:30990", "accept": false, - "reason": "Tokenizer added-space bug vs sentence-transformers loading hang." + "reason": "Tokenizer added-space behavior and Sentence Transformers loading hang are unrelated." }, { "left": "issue:44038", "right": "issue:44295", "accept": false, - "reason": "Qwen3-VL-Moe issue vs position_ids buffer access error." + "reason": "Qwen3-VL-Moe behavior and position_ids buffer access are unrelated." }, { "left": "issue:43122", "right": "issue:44779", "accept": false, - "reason": "Different tokenizer regressions affecting different models and outputs." - }, - { - "left": "issue:42907", - "right": "issue:43452", - "accept": false, - "reason": "Dequantized save issue vs gguf_file loading breakage." + "reason": "Both are tokenizer regressions in v5, but they involve different models and concrete failure modes." }, { "left": "issue:43824", "right": "issue:44351", "accept": false, - "reason": "Different missing-import names and likely different export gaps." + "reason": "Missing Qwen2_5_VLForConditionalGeneration and missing HybridCache are separate import errors." }, { "left": "issue:36296", "right": "issue:43116", "accept": false, - "reason": "Tensor parallel training bug vs classification example returning empty results." + "reason": "Tensor parallel training instability and empty classification outputs are unrelated." }, { "left": "issue:44038", "right": "issue:44222", "accept": false, - "reason": "Qwen3-VL-Moe bug vs FP8 save_pretrained issue; unrelated." + "reason": "Qwen3-VL-Moe regressions and FP8 save_pretrained moe are different bugs, despite both mentioning moe." } ] }, @@ -12863,20 +13013,20 @@ "issue:43827|issue:45446", "issue:33453|issue:39401" ], - "prepared_review_unit_hash": "2f5c3590549a29bf0f2bc24263a1bb2e7363b8cda9a77307c3f427bad7392990", + "prepared_review_unit_hash": "66236ff23197e74d8ed3e9c9f33582c64f95fe410e637dd786767c6e98c5fbc5", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 28, - "serialized_chars": 35097, + "serialized_chars": 35098, "estimated_input_tokens": 8775, "estimated_eval_tokens": 17806 }, @@ -12885,180 +13035,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is highly heterogeneous: most pairs are superficially similar version-regression or import/tokenizer issues, but they affect different code paths and should not be deduped. No soft pair looks like the same concrete bug/change.", - "confidence": 0.36, - "canonical_issue_reason": "issue:45278 is the broadest umbrella candidate, covering a general Transformers v5 upgrade/import-regression pattern that subsumes several import-related reports more naturally than the model-specific or feature-specific issues.", + "summary": "This is not a coherent duplicate cluster: the items span many unrelated bugs across tokenization, imports, training, generation, multimodal models, and docs. Under strict triage, every soft pair should be rejected.", + "confidence": 0.86, + "canonical_issue_reason": "No single canonical issue stands out because the reports are heterogeneous and do not share one underlying defect.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45278 is the best representative issue because it is the most general, cross-cutting regression report in the set; the rest are mostly narrow, unrelated bugs.", + "best_issue_reason": "If one representative issue must be chosen, issue 43122 is the cleanest broad regression-style tokenizer report; however, it is only a weak global representative because the set is not a true duplicate cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45216", "right": "issue:45478", "accept": false, - "reason": "Both involve Qwen3.5 and recent v5 regressions, but one is about save_pretrained checkpoint contents and the other about from_pretrained loading; different failure modes and code paths." + "reason": "Both concern Qwen3.5 save/load failures, but one is a bad checkpoint saved by save_pretrained and the other is a from_pretrained error; different concrete code paths." }, { "left": "issue:42757", "right": "issue:44351", "accept": false, - "reason": "Both are import errors, but they concern different missing symbols from different packages; not the same bug." + "reason": "Both are import errors, but they involve different missing symbols from different packages/modules and are not the same bug." }, { "left": "issue:43278", "right": "issue:44945", "accept": false, - "reason": "BF16/FP32 dtype drift in eval is unrelated to incorrect output under pipeline parallelism." + "reason": "One is a dtype mismatch between train/evaluate; the other is incorrect output under pipeline parallelism. Different failure modes and subsystems." }, { "left": "issue:43784", "right": "issue:45446", "accept": false, - "reason": "A missing nn import in sentence-transformers is unrelated to a PyTorch version gate in flex_attention." + "reason": "Unrelated import/version issues: sentence-transformers import failure versus a PyTorch version check in flex_attention." }, { "left": "issue:22355", "right": "issue:29159", "accept": false, - "reason": "One is a missing module import, the other is tokenizer behavior inconsistency." + "reason": "Missing transformers.onnx module import is unrelated to slow/fast tokenizer behavioral inconsistency." }, { "left": "issue:43122", "right": "issue:44977", "accept": false, - "reason": "Different tokenization output vs generation failure; no shared concrete code-path bug." + "reason": "Tokenizer behavior regression across versions is unrelated to flash-attention generation failure on Qwen3.5." }, { "left": "issue:43493", "right": "issue:45753", "accept": false, - "reason": "Both are vision/model correctness issues, but they involve different models and symptoms; too broad to merge." + "reason": "Different models and different symptoms: SigLIP2 implementation discrepancy versus a Qwen3.5 infinite loop on a specific image." }, { "left": "issue:38617", "right": "issue:45278", "accept": false, - "reason": "Specific missing symbol import versus a broad upgrade-wide import breakage report; not the same underlying defect." + "reason": "One is a specific missing import symbol; the other is broad import breakage after an upgrade. Not the same underlying defect." }, { "left": "issue:42673", "right": "issue:45072", "accept": false, - "reason": "VRAM leak in multi-threaded Qwen3ForCausalLM use is unrelated to dtype mismatch in bfloat16 inference." + "reason": "VRAM leak in dataloader threads is unrelated to bfloat16 dtype mismatches in inference." }, { "left": "issue:43066", "right": "issue:44779", "accept": false, - "reason": "Both are tokenizer regressions, but one is decoder type selection and the other is Deepseek tokenization correctness; different bugs." + "reason": "Wrong tokenizer decoder type and Deepseek tokenization regression are different tokenizer bugs with no shared concrete code-path." }, { "left": "issue:22355", "right": "issue:29127", "accept": false, - "reason": "Completely different problems: missing module import vs unclear error messaging in layoutlmv3." + "reason": "Module import failure and a LayoutLMv3 error-message clarity issue are unrelated." }, { "left": "issue:43992", "right": "issue:44704", "accept": false, - "reason": "A missing embed_tokens weight during UMT5Encoder load is unrelated to AutoProcessor not forwarding kwargs." + "reason": "Missing embed_tokens weights during from_pretrained is unrelated to AutoProcessor not forwarding kwargs to cached_file." }, { "left": "issue:43937", "right": "issue:44485", "accept": false, - "reason": "GenerationConfig validation failure and RoPE implementation discussion are distinct issues." + "reason": "GenerationConfig validation failure and GLM-5 RoPE discussion are different model-generation issues." }, { "left": "issue:43611", "right": "issue:46002", "accept": false, - "reason": "Base_model_prefix loading regression is unrelated to NaNs in classification heads after checkout/init." + "reason": "Base_model_prefix loading regression and NaNs after checkout/init are unrelated symptoms and code paths." }, { "left": "issue:43976", "right": "issue:45446", "accept": false, - "reason": "These are different version-compatibility problems in different subsystems; not a duplicate." + "reason": "Python version support mismatch is not the same bug as the flex_attention import/version check." }, { "left": "issue:42222", "right": "issue:43441", "accept": false, - "reason": "Vitpose import/model breakage and Ministral-3 FlashAttention failure are unrelated." + "reason": "VitPose model breakage and Ministral flash-attention failure are different model-specific issues." }, { "left": "issue:43116", "right": "issue:43493", "accept": false, - "reason": "Example script multi-label output issue is unrelated to SigLIP2 HF/JAX discrepancy." + "reason": "Empty multi-label classification results in an example script is unrelated to a SigLIP2 implementation discrepancy." }, { "left": "issue:42371", "right": "issue:43874", "accept": false, - "reason": "TF32 settings guidance and a missing image-patch method are unrelated." + "reason": "TF32 behavior control messaging/API and a missing image-processor method are unrelated." }, { "left": "issue:44075", "right": "issue:44936", "accept": false, - "reason": "SGD argument handling and trainer.evaluate() after train() are separate training bugs." + "reason": "Ignored SGD args and trainer.evaluate() failing after train() are different trainer/optimizer problems." }, { "left": "issue:43064", "right": "issue:43673", "accept": false, - "reason": "Optimizer-state corruption under FSDP2/PEFT is unrelated to missing GenerationMixin cache in chunked prefill." + "reason": "Wrong optimizer state placement under FSDP2/PEFT is unrelated to missing GenerationMixin cache in chunked prefill." }, { "left": "issue:44038", "right": "issue:45478", "accept": false, - "reason": "Both mention Qwen3-family v5 regressions, but one is a Qwen3-VL-Moe bug and the other is a Qwen3.5 from_pretrained error; different model/code paths." + "reason": "These are different Qwen3.5/Qwen3-VL-Moe issues with different failing operations and symptoms." }, { "left": "issue:45357", "right": "issue:45478", "accept": false, - "reason": "Both are Qwen3.5 save/load regressions, but one is incorrect visual encoder key saving and the other is a from_pretrained failure; not the same fix." + "reason": "Incorrect visual encoder keys saved by save_pretrained is not the same as a from_pretrained loading error." }, { "left": "issue:43116", "right": "issue:43825", "accept": false, - "reason": "Example-script classification output and pipeline error messaging are unrelated." + "reason": "Example script multi-label classification failure and pipeline translation-task messaging are unrelated." }, { "left": "issue:45083", "right": "issue:45446", "accept": false, - "reason": "A feature-extraction length helper issue is unrelated to a version-check import problem." + "reason": "Helper-function length behavior in qwen3_omni_moe is unrelated to a PyTorch version check bug." }, { "left": "issue:43937", "right": "issue:45563", "accept": false, - "reason": "GenerationConfig invalidation and stale num_return_sequences warning are different generate()-path issues." + "reason": "Invalid GenerationConfig and a stale warning about num_return_sequences are not the same issue." }, { "left": "issue:44380", "right": "issue:46077", "accept": false, - "reason": "Both touch attention behavior, but they affect different models and different configuration semantics; not mergeable as one bug." + "reason": "GPT2 attention scaling ignored in backends versus Gemma4 causal-mask construction are different code paths and models." }, { "left": "issue:43827", "right": "issue:45446", "accept": false, - "reason": "Docs still referencing pipeline() and a flex_attention import/version check are unrelated." + "reason": "Docs still referencing pipeline() and a version check for AuxRequest are unrelated." }, { "left": "issue:33453", "right": "issue:39401", "accept": false, - "reason": "Tokenizer loading regression and Qwen3 wrong offset_mapping are different tokenizer bugs." + "reason": "Tokenizer loading regression and Qwen3 offset_mapping errors are distinct tokenizer bugs, not one underlying defect." } ] }, @@ -13162,208 +13312,208 @@ "issue:42371|issue:42831", "issue:43012|issue:43577" ], - "prepared_review_unit_hash": "932e99d461fd7296c7594289b261eb7e6dbd8434bc091858d56dba8474eb0abb", + "prepared_review_unit_hash": "d51b4fa44b61f8e535b8745e83c365c516754e4f997592b0864346879b9864d8", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36288, - "estimated_input_tokens": 9072, - "estimated_eval_tokens": 18400 + "serialized_chars": 36289, + "estimated_input_tokens": 9073, + "estimated_eval_tokens": 18402 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is heterogeneous, with a loose concentration of tokenizer/v5 regression reports, but none of the soft pairs look like true duplicates of the same underlying bug or change.", - "confidence": 0.86, - "canonical_issue_reason": "Issue 43122 is the clearest representative of the recurring tokenizer-regression theme: it is version-bounded, user-visible, and broad enough to anchor the cluster better than the narrower import/backend/model-specific reports.", + "summary": "This cluster is mostly a set of unrelated Transformers regressions that only share broad surface areas like tokenizers, imports, or model loading. None of the soft pairs look like safe duplicates, so I would not merge any of them. The most representative issue is the broad tokenizer-regression report in 43122.", + "confidence": 0.34, + "canonical_issue_reason": "issue:43122 is the broadest, most central regression report here: same tokenizer behaving differently across 4.57.3 to 5.0, which best captures the tokenizer-related subset of the cluster.", "canonical_pr_reason": null, - "best_issue_reason": "43122 is the strongest standalone issue here because it describes a concrete cross-version tokenization regression with broad impact and a clear reproduction surface.", + "best_issue_reason": "issue:43122 is the best standalone anchor for this cluster because it is the most general user-facing regression and has the clearest version-bump framing.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44117", "right": "issue:45588", "accept": false, - "reason": "Tokenizer mapping None handling is unrelated to a video-model crash in flash_attention.py." + "reason": "Import-time tokenizer mapping fallback failure vs flash-attention crash on sink-less models; different code paths and symptoms." }, { "left": "issue:39401", "right": "issue:44779", "accept": false, - "reason": "Both are tokenizer bugs, but they affect different models and behaviors: offset mapping vs changed tokenization output." + "reason": "Both are tokenizer bugs, but one is offset mapping and the other is model-specific output regression; not the same underlying defect." }, { "left": "issue:42673", "right": "issue:43906", "accept": false, - "reason": "A Qwen3 VRAM leak is not the same as an isolated reproduction of a different prior issue." + "reason": "VRAM leak in multi-threaded dataloader use vs an isolated reproduction of another issue; no clear shared bug signature." }, { "left": "issue:43937", "right": "issue:44977", "accept": false, - "reason": "GenerationConfig validation failure is a different failure mode from flash-attention generation issues." + "reason": "GenerationConfig validation failure for GLM-5 vs flash-attention generation failure for Qwen3.5; different failure modes." }, { "left": "issue:44466", "right": "issue:46002", "accept": false, - "reason": "Serialization inconsistency and NaNs on checkout/init are different model-state bugs." + "reason": "Serialization/tied-weights inconsistency vs NaNs after checkout+init; these are distinct model-state problems." }, { "left": "issue:44291", "right": "issue:45071", "accept": false, - "reason": "Unexpected init argument handling is distinct from PreTrainedConfig type-checking regressions." + "reason": "Unexpected _is_hf_initialized argument during init_empty_weights loading vs PretrainedConfig type-checking breakage; related version area but not the same bug." }, { "left": "issue:43450", "right": "issue:43986", "accept": false, - "reason": "Batched video-processor shape errors are different from torchvision-missing import crashes." + "reason": "Incorrect batched video-processor shape vs missing-torchvision crash in AutoProcessor; different layers of the video stack." }, { "left": "issue:45405", "right": "issue:45446", "accept": false, - "reason": "A dependency version bump issue is unrelated to a PyTorch version guard bug." + "reason": "Dependency version bump/release issue vs a PyTorch version guard bug in flex_attention; unrelated." }, { "left": "issue:45341", "right": "issue:45561", "accept": false, - "reason": "A small testing-utils bug is not the same as a pytest-xdist file-race condition." + "reason": "General testing_utils bug vs xdist race on captured_info.txt; the latter is a specific concurrency issue, not the same defect." }, { "left": "issue:38617", "right": "issue:41628", "accept": false, - "reason": "Both are import errors, but for different missing symbols and likely different code paths." + "reason": "Different missing imports from different modules; both are import errors, but not the same missing symbol or code path." }, { "left": "issue:44117", "right": "issue:45081", "accept": false, - "reason": "Tokenizer mapping fallback logic is unrelated to the Mistral regex patch crash." + "reason": "TOKENIZER_MAPPING_NAMES returning None vs Mistral regex patch crashing on backend_tokenizer absence; distinct tokenizer internals." }, { "left": "issue:44410", "right": "issue:45310", "accept": false, - "reason": "Missing Qwen3Next projections and Qwen3.5 from_pretrained errors are different model-loading problems." + "reason": "Qwen3Next missing projections in layer 0 vs Qwen3.5 MoE from_pretrained failure; different model families and bugs." }, { "left": "issue:44829", "right": "issue:45216", "accept": false, - "reason": "Flash-attention training degeneration is not the same as incorrect save_pretrained checkpoint serialization." + "reason": "Flash-attention-3 training degeneration vs save_pretrained checkpoint regression; different subsystems and symptoms." }, { "left": "issue:39401", "right": "issue:43421", "accept": false, - "reason": "Offset mapping bugs and runtime special-token post-processor updates are different tokenizer subsystems." + "reason": "Offset mapping bug vs post-processor not updating after runtime special-token changes; both tokenizer-related but not the same issue." }, { "left": "issue:42617", "right": "issue:45341", "accept": false, - "reason": "A script execution failure is unrelated to a testing-utils bug." + "reason": "3d_parallel.py runtime failure vs a small testing_utils bug; no overlap in underlying code path." }, { "left": "issue:42673", "right": "issue:43582", "accept": false, - "reason": "VRAM leakage in dataloader threads is unrelated to Apple Silicon allocator warmup TypeError." + "reason": "Multi-thread VRAM leak vs Apple Silicon TypeError in caching_allocator_warmup; unrelated runtime failures." }, { "left": "issue:44484", "right": "issue:45310", "accept": false, - "reason": "Chat/completions API request rejection and Qwen3.5 MoE from_pretrained failure are different surfaces." + "reason": "chat/completions API request handling failure vs Qwen3.5 MoE loading error; different APIs and failure points." }, { "left": "issue:42371", "right": "issue:44261", "accept": false, - "reason": "TF32 API guidance and MLA RMS norm precision are unrelated issues." + "reason": "TF32 control API warning vs missing rms_norm_eps causing precision drift; not the same bug." }, { "left": "issue:42490", "right": "issue:42673", "accept": false, - "reason": "Local-path import warnings are not the same as a model VRAM leak." + "reason": "Local-path trailing-slash import warning vs VRAM leak in multi-threaded loading; unrelated issues." }, { "left": "issue:44933", "right": "issue:45278", "accept": false, - "reason": "A specific missing import and broad version-upgrade import failures overlap only superficially." + "reason": "Specific missing import from image_utils vs broad import breakage after upgrade; the latter is a wider symptom, not the same concrete defect." }, { "left": "issue:43408", "right": "issue:43742", "accept": false, - "reason": "A wrong model-type warning is not the same as a key error when loading a specific model." + "reason": "Model-type mismatch warning for sam3 variants vs key error loading MobileLLM; different loaders and errors." }, { "left": "issue:38617", "right": "issue:44220", "accept": false, - "reason": "A missing import in configuration_utils is unrelated to _torch_extract_fbank_features behavior." + "reason": "Configuration import error vs audio feature extraction issue; completely different subsystems." }, { "left": "issue:42222", "right": "issue:45542", "accept": false, - "reason": "Broken vitpose models and a TensorBoard/TensorFlow backend packaging issue are different areas." + "reason": "ViTPose model breakage vs TensorFlow backend install ambiguity; only broad backend context overlaps." }, { "left": "issue:43937", "right": "issue:44496", "accept": false, - "reason": "Invalid GenerationConfig and unrecognized model/config.json errors are separate load-time failures." + "reason": "GLM-5 generation config invalid vs unrecognized model type in config.json; both load-time validation issues but with different roots." }, { "left": "issue:44483", "right": "issue:45464", "accept": false, - "reason": "Chat/completions API compatibility and streaming inference failures on Qwen3.5 are not the same bug." + "reason": "General chat/completions request rejection vs streaming inference failure on Qwen3.5-0.8B; not enough evidence for one bug." }, { "left": "issue:42175", "right": "issue:42222", "accept": false, - "reason": "Missing TensorFlow in torch-only install is a packaging issue, not a model breakage in vitpose." + "reason": "Missing TensorFlow in torch-only install vs vitpose model breakage; different packaging/model problems." }, { "left": "issue:43066", "right": "issue:43122", "accept": false, - "reason": "Wrong decoder type and changed tokenization are related only broadly; they are different concrete regressions." + "reason": "Wrong tokenizer decoder type vs changed tokenization output across versions; closely related theme, but not the same concrete defect." }, { "left": "issue:42371", "right": "issue:42831", "accept": false, - "reason": "TF32 control API messaging is unrelated to FineGrainedFP8 accuracy regressions." + "reason": "TF32 API settings warning vs FineGrainedFP8 accuracy regression; unrelated numeric-precision issues." }, { "left": "issue:43012", "right": "issue:43577", "accept": false, - "reason": "A compile-time float32 warning for bfloat16 is distinct from BLIP2 dtype persistence during loading." + "reason": "Compilation warning for bf16 models vs BLIP2 dtype fields staying float32 on load; different dtype-handling problems." } ] }, @@ -13476,262 +13626,262 @@ "issue:44704|issue:45103", "issue:43986|issue:44117" ], - "prepared_review_unit_hash": "c8c735888c572405826505466e2a2cc00ffba410891547895bf64240e34557f7", + "prepared_review_unit_hash": "4d5047fe6b8deb1fe7a596022e02081bb061a56f13cf2845168f96353098a8d0", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 38, - "serialized_chars": 39072, - "estimated_input_tokens": 9768, - "estimated_eval_tokens": 19792 + "serialized_chars": 39066, + "estimated_input_tokens": 9767, + "estimated_eval_tokens": 19790 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is a grab bag of unrelated Transformers issues; none of the soft pairs look like the same underlying bug or change, so all proposed duplicate links should be rejected.", - "confidence": 0.88, - "canonical_issue_reason": "No single canonical issue fits this cluster: the items span unrelated import errors, model-specific regressions, training bugs, and config warnings.", - "canonical_pr_reason": "No PRs are present.", - "best_issue_reason": "issue:43097 is the best standalone anchor because it is a clear, user-facing regression report with some cross-references and a broad compatibility impact.", + "summary": "This cluster is mostly a grab bag of unrelated Transformers bug reports; several pairs share only superficial themes like imports, loading, or v5 regressions, but none look like the same underlying defect closely enough to merge.", + "confidence": 0.68, + "canonical_issue_reason": "issue:43097 is the strongest cluster anchor: it\u2019s a broadly impactful v5 regression report, has inbound references, and sits near several weak similarity links, even though the neighboring issues are not true duplicates.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45372 is the best standalone issue to keep: it is open, concrete, and currently actionable, with a clear processor-loading breakage and active discussion.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44075", "right": "issue:45656", "accept": false, - "reason": "Both involve optimizers, but one is about unused SGD args and the other about double stepping under DeepSpeed; different code paths and fixes." + "reason": "Different optimizer bugs: unused SGD args vs optimizer stepping twice under deepspeed." }, { "left": "issue:43673", "right": "issue:45341", "accept": false, - "reason": "Chunked prefill cache behavior and a testing_utils bug are unrelated." + "reason": "Unrelated failures: generation cache/chunked prefill vs a testing utility bug." }, { "left": "issue:44410", "right": "issue:45478", "accept": false, - "reason": "Both are Qwen-family reports, but they concern different model variants and different failures (missing projections vs from_pretrained error)." + "reason": "Different Qwen model-path issues: missing projections in qwen3next vs a from_pretrained load error for Qwen3.5 MoE." }, { "left": "issue:44484", "right": "issue:45478", "accept": false, - "reason": "Save sharding defaults and a Qwen3.5 MoE load failure are unrelated." + "reason": "No shared underlying bug; one is a save_pretrained shard-size question, the other is a model loading failure." }, { "left": "issue:43097", "right": "issue:44493", "accept": false, - "reason": "One is about a removed config option warning; the other is an unexpected position-id key issue. Not the same bug." + "reason": "Both are v5-related regressions, but one is a removed config flag and the other is unexpected position-id keys." }, { "left": "issue:41628", "right": "issue:45071", "accept": false, - "reason": "Different API breakages with different root causes: missing AutoImageProcessor export vs PretrainedConfig type checking." + "reason": "Both involve import/type issues, but they affect different APIs and code paths." }, { "left": "issue:43874", "right": "issue:44261", "accept": false, - "reason": "Both affect multimodal/GLM-related models, but the failures are different: missing image-patch method vs RMS norm epsilon precision mismatch." + "reason": "Different model components: image processor missing a method vs MLA q_a_layernorm precision/config handling." }, { "left": "issue:44263", "right": "issue:44485", "accept": false, - "reason": "Same broad GLM area, but torch.split return handling and RoPE implementation are distinct code-path bugs." + "reason": "Both are GLM-family issues, but one is a torch.split return handling bug and the other is a RoPE implementation concern." }, { "left": "issue:45071", "right": "issue:45542", "accept": false, - "reason": "Type checking regression and tensorboard/TensorFlow backend initialization are unrelated." + "reason": "Unrelated: PreTrainedConfig type-checking regression vs TensorFlow/tensorboard backend detection." }, { "left": "issue:42222", "right": "issue:45459", "accept": false, - "reason": "VitPose breakage and tokenizer protobuf error masking are different subsystems and symptoms." + "reason": "Different failures: vitpose model breakage vs tokenizer protobuf error masking." }, { "left": "issue:43012", "right": "issue:43408", "accept": false, - "reason": "Bfloat16 compile warnings and sam3_video/sam3_tracker model-type mismatch are unrelated." + "reason": "Both are warnings in different model paths; they do not describe the same bug." }, { "left": "issue:43441", "right": "issue:44829", "accept": false, - "reason": "Both mention flash attention, but they concern different models and different outcomes; too broad to merge." + "reason": "Both mention FlashAttention, but one is a load-time failure and the other is degenerate training behavior." }, { "left": "issue:43931", "right": "issue:44410", "accept": false, - "reason": "Both are Qwen model loading issues, but one is a weight-shape mismatch and the other is missing layer projections; not the same concrete bug." + "reason": "Distinct model-loading issues for different model families and different failure modes." }, { "left": "issue:43673", "right": "issue:46123", "accept": false, - "reason": "Cache missing in generation and final-batch dropping in a mask pipeline are unrelated." + "reason": "Different pipeline behaviors: generation cache during chunked prefill vs final partial batch handling in a mask pipeline." }, { "left": "issue:43931", "right": "issue:44038", "accept": false, - "reason": "Different Qwen3-VL variants and different load-time failures; not enough evidence of a single underlying bug." + "reason": "Both are Qwen model issues, but the concrete failures differ and are not the same code-path bug." }, { "left": "issue:43097", "right": "issue:43452", "accept": false, - "reason": "Config deprecation/warning vs GGUF loading failures are unrelated." + "reason": "Different regressions: removed tie_embeddings flag vs gguf_file loading in tokenizer/model loaders." }, { "left": "issue:43335", "right": "issue:45071", "accept": false, - "reason": "SwitchTransformers sparse-layer construction and PretrainedConfig type checking are unrelated." + "reason": "Unrelated: SwitchTransformers sparse-layer config bug vs PretrainedConfig type-checking regression." }, { "left": "issue:45003", "right": "issue:45341", "accept": false, - "reason": "Unsafe sys.modules access and a testing_utils bug are unrelated." + "reason": "Different subsystems: unsafe sys.modules access vs a testing utility bug." }, { "left": "issue:43986", "right": "issue:45341", "accept": false, - "reason": "AutoProcessor/video-model loading without torchvision and a testing_utils bug are unrelated." + "reason": "Completely different areas: AutoProcessor video-model crash vs testing utilities." }, { "left": "issue:44117", "right": "issue:44987", "accept": false, - "reason": "Both are loading-related, but one is tokenizer mapping None handling and the other is a model-specific load failure; not the same issue." + "reason": "Both are from_pretrained-related, but they target different assumptions and different failures." }, { "left": "issue:43986", "right": "issue:44484", "accept": false, - "reason": "Video processor dependency crash and max_shard_size default question are unrelated." + "reason": "No common underlying bug: video processor dependency handling vs save_pretrained shard sizing." }, { "left": "issue:43408", "right": "issue:44368", "accept": false, - "reason": "Different warnings with different causes: model-type mismatch vs tie_word_embeddings config warning." + "reason": "Both are warnings, but they concern different model/config mismatches." }, { "left": "issue:43874", "right": "issue:44263", "accept": false, - "reason": "Different GLM subsystems and failures; no concrete duplicate signal." + "reason": "Different bugs in different GLM components; not the same defect." }, { "left": "issue:43531", "right": "issue:45341", "accept": false, - "reason": "Qwen3-MoE sliding_window behavior and a testing utility bug are unrelated." + "reason": "Qwen3-MoE sliding_window behavior is unrelated to testing_utils." }, { "left": "issue:43986", "right": "issue:45372", "accept": false, - "reason": "Different missing-dependency import errors in different loading paths; not the same bug." + "reason": "Both involve loading processors, but the missing torchvision path and missing ReasoningEffort import are different causes." }, { "left": "issue:38617", "right": "issue:42757", "accept": false, - "reason": "Both are import errors, but for different symbols and different packages/modules." + "reason": "Both are ImportError reports, but they concern different symbols from different dependencies and different breakages." }, { "left": "issue:44261", "right": "issue:44485", "accept": false, - "reason": "Both mention GLM-family models, but the actual defects differ materially." + "reason": "Different GLM-related correctness issues; no shared code-path evidence." }, { "left": "issue:43064", "right": "issue:44315", "accept": false, - "reason": "Training-state corruption under FSDP2/PEFT and Liger Kernel not applied with model_init are different issues." + "reason": "Training/optimizer-state bug vs Liger kernel application under model_init; not the same problem." }, { "left": "issue:43097", "right": "issue:46050", "accept": false, - "reason": "Removed config warning and quantized save/load breakage are unrelated." + "reason": "Different serialization regressions: config flag removal vs quantized model save/load breakage." }, { "left": "issue:43097", "right": "issue:45588", "accept": false, - "reason": "Config removal warning and flash_attention crash on sink-less models are different code paths." + "reason": "Unrelated: tie_embeddings config removal vs flash_attention AttributeError on sink-less models." }, { "left": "issue:44062", "right": "issue:44117", "accept": false, - "reason": "Tokenizer AddedToken argument collision and tokenizer mapping None handling are unrelated." + "reason": "Tokenizer construction bug vs tokenizer mapping lookup assumption; related area but different defects." }, { "left": "issue:38617", "right": "issue:44351", "accept": false, - "reason": "Both are missing-symbol import errors, but they involve different symbols and different import surfaces." + "reason": "Both are import errors, but they are for different missing names and different modules." }, { "left": "issue:44410", "right": "issue:45375", "accept": false, - "reason": "Qwen-related config/model issues, but one is missing projections and the other is a silently dropped config field; not the same bug." + "reason": "Different vision/config issues: missing attention projections vs a dropped config field under strict handling." }, { "left": "issue:44038", "right": "issue:44410", "accept": false, - "reason": "Different Qwen model families and different failure modes." + "reason": "Both are Qwen variants, but the concrete failures are different model-internal bugs." }, { "left": "issue:43502", "right": "issue:43992", "accept": false, - "reason": "Local-files-only network leakage and missing embed_tokens.weight during UMT5 loading are unrelated." + "reason": "Different from_pretrained failures: local_files_only network leakage vs missing embed_tokens weight." }, { "left": "issue:43097", "right": "issue:43873", "accept": false, - "reason": "Config deprecation warning and quantization offloading behavior are unrelated." + "reason": "Different problem classes: config removal regression vs quantization/offloading behavior." }, { "left": "issue:44704", "right": "issue:45103", "accept": false, - "reason": "AutoProcessor cached_file kwargs forwarding and auto_docstring future-annotations crash are unrelated." + "reason": "Unrelated kwargs-passing bug vs auto_docstring annotation crash." }, { "left": "issue:43986", "right": "issue:44117", "accept": false, - "reason": "Video AutoProcessor dependency crash and tokenizer mapping None are unrelated." + "reason": "Different loading failures in different subsystems; not a duplicate." } ] }, @@ -13840,20 +13990,20 @@ "issue:43901|issue:44655", "issue:45276|issue:46077" ], - "prepared_review_unit_hash": "6a0801adadcd579850de8d28982e17b51538d58a275cd2fc363691793742d630", + "prepared_review_unit_hash": "a15949a66d907da2f680c917de67ae8fd47a64a2da2cd41b422246d4dcedd72f", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 34, - "serialized_chars": 37718, + "serialized_chars": 37720, "estimated_input_tokens": 9430, "estimated_eval_tokens": 19116 }, @@ -13862,216 +14012,210 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a set of unrelated Transformers bugs; the soft-similarity links are mostly superficial wording overlaps (imports, loading, compile, MoE, chat/template), not true duplicates.", - "confidence": 0.87, - "canonical_issue_reason": "issue:44222 is the broadest representative of the only closely related theme here (Qwen3.5/MoE/FP8 model I/O regressions), even though the surrounding cluster is still largely heterogeneous.", + "summary": "No true duplicates stand out in this noisy cluster. The soft-similarity pairs mostly share a subsystem or model family, but they describe different failures and should stay separate. No PRs are present; I\u2019d use issue #43012 as the canonical representative only because it is a clear, broadly phrased regression report.", + "confidence": 0.33, + "canonical_issue_reason": "#43012 is the cleanest, most generally framed bug report in the set and is a better representative than the many model-specific or UX-only reports.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45478 is the most concrete, actionable regression report in the nearest theme, with a specific version boundary and a clear from_pretrained failure.", + "best_issue_reason": "#43012 is a concrete, actionable runtime regression with a clear symptom and wider applicability than the other items.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43824", "right": "issue:43986", "accept": false, - "reason": "Both are load/import failures, but one is a missing model-class export and the other is a torchvision-dependent AutoProcessor crash; different root causes." + "reason": "Both are import/load failures, but one is a missing model class export and the other is a torchvision dependency crash; different root causes." }, { "left": "issue:43066", "right": "issue:43906", "accept": false, - "reason": "Tokenizer decoder-type regression vs an isolated reproduction of another issue; no shared concrete bug." + "reason": "These concern different v5 issues: tokenizer decoder typing vs an isolated reproduction of another bug." }, { "left": "issue:43208", "right": "issue:43550", "accept": false, - "reason": "xLSTM training blockers vs Bamba torch.compile/SDPA failure; different models and failure modes." + "reason": "Different model families and failures: xLSTM training bugs vs Bamba torch.compile/SDPA behavior." }, { "left": "issue:43825", "right": "issue:44655", "accept": false, - "reason": "Pipeline task-warning behavior vs save_pretrained persistence bug; unrelated functionality." + "reason": "Pipeline task-message behavior and pipeline saving are separate code paths and symptoms." }, { "left": "issue:43992", "right": "issue:44484", "accept": false, - "reason": "Missing encoder weight on from_pretrained vs a question about default shard size in save_pretrained; not the same issue." + "reason": "Missing checkpoint weight loading is unrelated to the default shard-size question in save_pretrained()." }, { "left": "issue:43012", "right": "issue:43550", "accept": false, - "reason": "PyTorch precision warning on bfloat16 compile vs a compile/SDPA failure in another model; different code paths." + "reason": "One is a PyTorch precision warning during compile; the other is a model-specific compile failure under SDPA." }, { "left": "issue:43012", "right": "issue:44368", "accept": false, - "reason": "bfloat16 compile warning vs a tie_word_embeddings config warning; similar wording, different behavior." + "reason": "Different symptoms and code paths: compile-time warning vs repeated tie_word_embeddings warning in LoRA fine-tuning." }, { "left": "issue:43352", "right": "issue:44380", "accept": false, - "reason": "FlashAttention 2 unsupported model vs GPT2 attention scaling ignored under backends; distinct attention bugs." + "reason": "Both mention attention backends, but one is unsupported FlashAttention 2 for a model and the other is ignored GPT-2 attention scaling." }, { "left": "issue:42907", "right": "issue:45753", "accept": false, - "reason": "Model save failure for dequantized Ministrals/Devstrals vs Qwen3_5 infinite-loop generation; unrelated symptoms." + "reason": "Saving dequantized models and an image-triggered infinite loop are unrelated problems." }, { "left": "issue:42907", "right": "issue:44655", "accept": false, - "reason": "Dequantized-model save failure vs pipeline save_pretrained; same broad area but different objects and failure modes." + "reason": "Model serialization after dequantization and saving Pipeline objects involve different save paths and failures." }, { "left": "issue:43792", "right": "issue:46123", "accept": false, - "reason": "Whisper runtime/load failure vs partial-batch dropping in MaskGenerationPipeline; unrelated bugs." + "reason": "Whisper model loading failure and MaskGenerationPipeline partial-batch dropping are unrelated." }, { "left": "issue:43352", "right": "issue:43976", "accept": false, - "reason": "FlashAttention support error vs Python version compatibility problem; different causes." + "reason": "A model capability error is unrelated to the Python-version compatibility regression." }, { "left": "issue:44380", "right": "issue:45479", "accept": false, - "reason": "Attention-scaling ignored in SDPA/FlashAttention vs degenerate zero-loss classification; not the same code path." + "reason": "Attention-scaling being ignored is not the same bug as degenerate zero loss in sequence classification." }, { "left": "issue:44380", "right": "issue:46132", "accept": false, - "reason": "GPT2 backend scaling bug vs AttentionInterface.register behavior change; distinct API surfaces." + "reason": "Attention backend scaling and AttentionInterface.register behavior are distinct issues." }, { "left": "issue:43064", "right": "issue:44380", "accept": false, - "reason": "FSDP2/PEFT optimizer-state corruption vs GPT2 attention scaling; training-state bug vs inference attention bug." + "reason": "Distributed optimizer-state corruption is unrelated to GPT-2 attention scaling under SDPA/FlashAttention." }, { "left": "issue:44483", "right": "issue:44514", "accept": false, - "reason": "OpenAI-compatible chat request rejection vs batched apply_chat_template crash; different layers of the stack." + "reason": "Chat-completions request handling and batched chat-template crashes are different interfaces and failures." }, { "left": "issue:44222", "right": "issue:45478", "accept": false, - "reason": "Both touch Qwen3.5/MoE model I/O, but one is FP8 save_pretrained serialization and the other is a from_pretrained loading regression; too different to merge as duplicates." + "reason": "Both involve Qwen3.5 MoE, but one is FP8 save_pretrained and the other is a from_pretrained regression; not the same concrete bug." }, { "left": "issue:36683", "right": "issue:43582", "accept": false, - "reason": "Gemma3Config missing vocab_size vs Apple Silicon allocator TypeError; no overlap in subsystem or symptom." + "reason": "Gemma3Config vocab_size and Apple Silicon allocator warmup errors are unrelated." }, { "left": "issue:43208", "right": "issue:43526", "accept": false, - "reason": "xLSTM training bugs vs BeitImageProcessorFast label reduction bug; unrelated models and code paths." + "reason": "xLSTM training bugs and BeitImageProcessorFast label reduction are different subsystems and defects." }, { "left": "issue:43012", "right": "issue:44683", "accept": false, - "reason": "bfloat16 compile warning vs compiled flex_attention failure on torch 2.9; different compile issues." + "reason": "Both mention compilation, but one is a warning on bfloat16 models and the other is a torch>=2.9 flex_attention failure." }, { "left": "issue:42757", "right": "issue:43824", "accept": false, - "reason": "ImportError from huggingface_hub offline-mode helper vs missing Qwen2_5_VL export from transformers; different modules and failures." + "reason": "Different missing imports from different packages with different fixes and call sites." }, { "left": "issue:42490", "right": "issue:43012", "accept": false, - "reason": "Trailing-slash local model path warning vs bfloat16 compile warning; unrelated behavior." + "reason": "Local-path import warning and bfloat16 compile warning are unrelated." }, { "left": "issue:43618", "right": "issue:43761", "accept": false, - "reason": "CLIP attentions missing vs CLIPVisionModel hidden_states regression; both CLIP, but different outputs and bugs." + "reason": "Both are CLIP regressions, but one drops attentions and the other drops hidden_states; not the same behavior." }, { "left": "issue:43618", "right": "issue:44380", "accept": false, - "reason": "CLIP output-field regression vs GPT2 attention scaling bug; unrelated code paths." + "reason": "CLIP output field regression is unrelated to GPT-2 attention scaling semantics." }, { "left": "issue:44038", "right": "issue:44230", "accept": false, - "reason": "Qwen3-VL-Moe loading bug vs FP8 inference support request; not the same concrete bug." + "reason": "Different Qwen3-VL-Moe problems: general loading bug vs FP8 inference support." }, { "left": "issue:38617", "right": "issue:43502", "accept": false, - "reason": "ImportError for layer_type_validation vs local_files_only still making requests; unrelated import vs networking behavior." + "reason": "A missing import symbol and unwanted network requests under local_files_only are different bugs." }, { "left": "issue:44483", "right": "issue:45290", "accept": false, - "reason": "API request validation failure vs assistant-message/tool-call chat_template crash; different API surfaces." + "reason": "Chat-completions API rejection and apply_chat_template tool-call crashes are different entry points." }, { "left": "issue:44483", "right": "issue:45381", "accept": false, - "reason": "OpenAI chat completion rejection vs Qwen2.5-VL video position-id bug; unrelated issues." + "reason": "API request acceptance and Qwen2.5-VL video vision_position_ids are unrelated failures." }, { "left": "issue:43756", "right": "issue:44485", "accept": false, - "reason": "Smollm3 RoPE-layer discrepancy vs GLM-5 RoPE implementation discussion; same broad topic only, not one bug." + "reason": "Both mention RoPE, but the model/configuration defects are different and not mergeable into one fix." }, { "left": "issue:43299", "right": "issue:43867", "accept": false, - "reason": "Qwen3-VL-Moe loading breakage vs load failure when state_dict is sorted; different loading mechanics." - }, - { - "left": "issue:36683", - "right": "issue:43475", - "accept": false, - "reason": "Gemma3Config missing vocab_size vs Sam3VisionEncoderOutput missing fpn_position_embeddings; different model families and missing fields." + "reason": "Qwen3VL MoE loading and sorted-state_dict load errors are separate loading-path bugs." }, { "left": "issue:43299", "right": "issue:45753", "accept": false, - "reason": "Qwen3-VL-Moe loading failure vs Qwen3_5 image infinite loop; same family but different failure type and code path." + "reason": "Model-loading regression and image-specific infinite loop are unrelated." }, { "left": "issue:43901", "right": "issue:44655", "accept": false, - "reason": "Docs around return_all_scores vs pipeline save_pretrained bug; documentation mismatch vs persistence bug." + "reason": "Documentation mismatch for return_all_scores and pipeline serialization failure are different issues." }, { "left": "issue:45276", "right": "issue:46077", "accept": false, - "reason": "Gemma4 embedding resize propagation bug vs bidirectional attention mask still causal; different model components and symptoms." + "reason": "Embedding resizing and bidirectional attention mask construction are different Gemma4 bugs." } ] }, @@ -14180,20 +14324,20 @@ "issue:42490|issue:44075", "issue:36296|issue:43575" ], - "prepared_review_unit_hash": "ce92a2a9228fbba8417e8788b8fff1ba7e29b1a017132795268f4c4d9793fc92", + "prepared_review_unit_hash": "659923f8cd3f90101a57fd04b16b378368ea41468d70a68be844d52d467e8353", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 34, - "serialized_chars": 37352, + "serialized_chars": 37351, "estimated_input_tokens": 9338, "estimated_eval_tokens": 18932 }, @@ -14202,216 +14346,216 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: most items are distinct model-specific regressions, import breakages, or test/perf issues that only share broad Transformers-v5-era symptoms. I do not see any soft pair that is the same underlying bug, so all soft edges should be rejected. The best representative issue is the broad umbrella report about v5 breaking many models.", - "confidence": 0.86, - "canonical_issue_reason": "issue:45850 is the broadest umbrella report and best summarizes the shared v5 breakage theme across multiple models; it is the most representative issue in an otherwise mixed cluster.", + "summary": "The cluster is dominated by unrelated Transformers v5 regression reports: import failures, model-loading/config issues, save_pretrained regressions, and a few training/runtime bugs. The soft pairs are mostly thematic similarities rather than true duplicates, so I would not merge any of them.", + "confidence": 0.7, + "canonical_issue_reason": "issue:45278 is the broadest concrete umbrella in this set: a v5 upgrade causing many import failures. It best represents the cluster\u2019s recurring version-regression theme without being as vague as the other broad reports.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45850 is the most useful cluster representative because it captures the general regression class rather than a single model- or config-specific failure.", + "best_issue_reason": "issue:45278 is the most suitable issue to anchor this cluster because it is broad, concrete, and closest to the shared upgrade-breakage pattern across several items.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:42994", "right": "issue:44655", "accept": false, - "reason": "Both involve saving, but one is quantized model save failure and the other is Pipeline.save_pretrained; different objects and failure modes." + "reason": "Both involve saving, but one is quantized-model serialization and the other is Pipeline.save_pretrained. Different code paths and failure modes." }, { "left": "issue:36010", "right": "issue:42757", "accept": false, - "reason": "Both are import errors, but for different symbols/modules and different dependency breakages." + "reason": "Both are import errors, but they are missing different symbols from different packages. Same symptom, not the same bug." }, { "left": "issue:42898", "right": "issue:43066", "accept": false, - "reason": "Both touch v5 tokenization behavior, but one is cleanup-space semantics and the other is tokenizer decoder type; not the same bug." + "reason": "One is a v5 tokenization cleanup behavior change, the other is a tokenizer decoder type regression. Related area, different underlying issue." }, { "left": "issue:44077", "right": "issue:44380", "accept": false, - "reason": "Unrelated issues: one is about patchtsmixer post_init policy, the other about GPT2 attention scaling under SDPA/FlashAttention." + "reason": "Unrelated failures: optional post_init handling vs GPT2 attention scaling being ignored. No shared concrete code path." }, { "left": "issue:43352", "right": "issue:43937", "accept": false, - "reason": "Different code paths: Flash Attention support gating for Nemotron vs GenerationConfig validation for GLM-5." + "reason": "Flash Attention support error for a model vs invalid GenerationConfig validation. Different subsystems." }, { "left": "issue:43867", "right": "issue:45216", "accept": false, - "reason": "Both involve loading/saving checkpoints, but one is state_dict ordering and the other is Qwen3.5 save_pretrained regression." + "reason": "Both are loading/saving regressions, but one is a sorted state_dict load failure and the other is an incorrect Qwen3.5 checkpoint save. Not the same defect." }, { "left": "issue:36683", "right": "issue:44077", "accept": false, - "reason": "Different models and different issues; only a superficial config-related similarity." + "reason": "A missing config attribute in Gemma3 is unrelated to patchtsmixer post_init restrictions. Only a superficial model/config similarity." }, { "left": "issue:43611", "right": "issue:44655", "accept": false, - "reason": "Model loading with base_model_prefix vs saving Pipeline objects are separate behaviors." + "reason": "base_model_prefix loading breakage and Pipeline.save_pretrained failure are separate serialization/loading paths." }, { "left": "issue:44625", "right": "issue:46077", "accept": false, - "reason": "Different models and different semantics: num_labels propagation vs Gemma4 attention mask construction." + "reason": "Qwen3.5 num_labels propagation and Gemma4 bidirectional attention masks are unrelated config bugs." }, { "left": "issue:43976", "right": "issue:44938", "accept": false, - "reason": "Both are Python-version compatibility complaints, but for different versions and different breakages." + "reason": "Both mention Python version compatibility, but one is a 5.1.0/Python3.9+ mismatch and the other is a Python 3.14 import/load failure. Not the same bug." }, { "left": "issue:46077", "right": "issue:46082", "accept": false, - "reason": "Gemma4 bidirectional attention masking and LlamaConfig head_dim validation are unrelated." + "reason": "Different config validation problems: attention-mask construction vs rejecting explicit head_dim." }, { "left": "issue:43352", "right": "issue:45593", "accept": false, - "reason": "Different models and different training/inference behaviors; no shared underlying bug." + "reason": "Unsupported Flash Attention on Nemotron and missing auxiliary losses in D-FINE are unrelated model-specific issues." }, { "left": "issue:36683", "right": "issue:43525", "accept": false, - "reason": "Both are missing config attributes, but for different models and different fields." + "reason": "Both are missing config attributes, but for different models and different fields. Too broad to be a duplicate." }, { "left": "issue:43618", "right": "issue:45216", "accept": false, - "reason": "CLIPOutput attentions assignment issue is unrelated to Qwen3.5 checkpoint saving." + "reason": "CLIPOutput attentions assignment and Qwen3.5 checkpoint saving are distinct regressions." }, { "left": "issue:44263", "right": "issue:45440", "accept": false, - "reason": "Different MoE implementations and different tensor computation problems; not the same defect." + "reason": "A torch.split indexing issue in GlmMoeDsaIndexer is unrelated to DeepSeekV3 implementation divergence." }, { "left": "issue:36683", "right": "issue:44242", "accept": false, - "reason": "Gemma3Config vocab_size absence and load balancing loss omission are unrelated." + "reason": "Gemma3Config missing vocab_size and load-balancing loss omission are different bugs in different model families." }, { "left": "issue:43352", "right": "issue:43931", "accept": false, - "reason": "Nemotron Flash Attention support and Qwen3-VL weight-shape mismatch are separate model issues." + "reason": "Flash Attention unsupported for Nemotron is unrelated to Qwen3-VL weight-shape mismatch." }, { "left": "issue:43064", "right": "issue:44492", "accept": false, - "reason": "Trainer/FSDP optimizer-state bug vs a cache-strategy typo; no common underlying change." + "reason": "Trainer/FSDP2 optimizer state corruption and a cache-strategy typo are unrelated issues." }, { "left": "issue:43010", "right": "issue:45278", "accept": false, - "reason": "No shared bug: one is about no_grad decoration for cache updates, the other is broad import failures after upgrade." + "reason": "No-grad decoration for cache updates and broad import failures are different problem areas." }, { "left": "issue:43992", "right": "issue:45310", "accept": false, - "reason": "Both are Qwen3.5 loading-related, but one is missing embed_tokens weights in UMT5Encoder and the other is a Moe from_pretrained error." + "reason": "Missing embed_tokens.weight in UMT5Encoder and Qwen3.5 MoE from_pretrained errors are separate loading bugs." }, { "left": "issue:43867", "right": "issue:45375", "accept": false, - "reason": "Different checkpoint/model-config problems: state_dict sorting vs missing deepstack_visual_indexes field." + "reason": "Sorted state_dict load failure and missing deepstack_visual_indexes in config are not the same defect." }, { "left": "issue:44355", "right": "issue:44855", "accept": false, - "reason": "Compiled Python file execution errors and a torch.jit.script parsing/IndentationError are different root causes." + "reason": "Compiled Python file loading errors and the Python 3.13 @torch.jit.script indentation parsing issue are distinct interpreter/codegen problems." }, { "left": "issue:41669", "right": "issue:44246", "accept": false, - "reason": "Both concern import performance, but one is a specific import-star refactor issue and the other is intermittent slow import; not the same concrete bug." + "reason": "Import-* performance cleanup and intermittent import slowdowns are related only at a very high level; not a duplicate." }, { "left": "issue:42890", "right": "issue:45561", "accept": false, - "reason": "Test flakiness from missing set_seed and xdist capture-file races are different test infrastructure problems." + "reason": "Missing set_seed in tests and xdist race on captured_info.txt are test-infrastructure issues, but different concrete failures." }, { "left": "issue:36683", "right": "issue:43064", "accept": false, - "reason": "Different subsystems entirely: model config attribute error vs distributed training optimizer-state bug." + "reason": "Gemma3Config missing vocab_size is not the same as FSDP2/PEFT optimizer state misalignment." }, { "left": "issue:44938", "right": "issue:45071", "accept": false, - "reason": "Python 3.14 import/load failure and PretrainedConfig type-checking regression are unrelated." + "reason": "Both are version-compatibility regressions, but one is Python 3.14 loading and the other is type-checking breakage in v5.4.0." }, { "left": "issue:43867", "right": "issue:45357", "accept": false, - "reason": "Both mention loading/saving checkpoints, but one is state_dict sorting and the other is incorrect visual encoder keys in Qwen3.5 save_pretrained." + "reason": "Generic load-model state_dict sorting failure and incorrect visual encoder save keys are different serialization bugs." }, { "left": "issue:43010", "right": "issue:45850", "accept": false, - "reason": "One is a specific cache update no_grad request; the other is a broad complaint that v5 breaks many models." + "reason": "The no_grad cache update proposal is a specific fix request, while 45850 is a vague broad complaint about many model breakages." }, { - "left": "issue:43976", + "left": "issue:44938", "right": "issue:45381", "accept": false, - "reason": "Different version-specific regressions and different models; no shared code-path." + "reason": "Python 3.14 load failure and Qwen2.5-VL video vision_position_ids are unrelated issues." }, { "left": "issue:43010", "right": "issue:43976", "accept": false, - "reason": "No common underlying bug: cache update decoration vs Python-version compatibility." + "reason": "Cache update no_grad decoration and Python version support for 5.1.0 are unrelated." }, { "left": "issue:43618", "right": "issue:44077", "accept": false, - "reason": "CLIP attentions assignment and patchtsmixer post_init policy are unrelated." + "reason": "CLIPOutput attentions regression and patchtsmixer post_init restrictions are different model-specific changes." }, { "left": "issue:43992", "right": "issue:46002", "accept": false, - "reason": "Missing encoder weights at load time vs NaNs in classification heads are separate issues." + "reason": "UMT5Encoder missing weights and NaNs in classification heads after init/checkout are not the same bug." }, { "left": "issue:42490", "right": "issue:44075", "accept": false, - "reason": "Import warning for local paths with trailing slash and SGD args not used are unrelated." + "reason": "Local-path trailing-slash import warning and ignored SGD args are unrelated." }, { "left": "issue:36296", "right": "issue:43575", "accept": false, - "reason": "Both mention tensor parallelism, but one is a generic training bug and the other is an OOM during loading a specific model; not the same bug." + "reason": "Tensor parallel training bug and Qwen2-57B-A14B-A14B Instruct OOM are different runtime problems, not a single duplicate." } ] }, @@ -14523,14 +14667,14 @@ "issue:44222|issue:44230", "issue:33453|issue:36296" ], - "prepared_review_unit_hash": "8eb273c8243325067b03138d5715887936274e6c5eac771b0e844deafa4e9f38", + "prepared_review_unit_hash": "736a8e2a852d9fe0084970f99dbbebf6357114db6da7d9835fadbb6ec5408569", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, @@ -14545,234 +14689,234 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly heterogeneous issues around transformers v5 regressions, loading/import failures, FP8/tensor-parallel features, and unrelated test/docs bugs. I found no strong duplicate pairs; the closest ones are still too broad or differ in concrete code path.", - "confidence": 0.88, - "canonical_issue_reason": "issue:44556 is the best representative because it is a clear, high-impact v4-to-v5 loading regression with a concrete failure mode and broad user relevance.", + "summary": "This cluster is very heterogeneous. The soft pairs mostly share only broad wording like \u201cv5\u201d, \u201cloading\u201d, or \u201cregression\u201d, but they point to different code paths, models, or tooling issues. I would not merge any of the soft edges.", + "confidence": 0.76, + "canonical_issue_reason": "issue:43723 is the broadest, most user-facing loading regression in the set and best fits the loose v5 breakage theme, even though the overall cluster is too mixed for a strong duplicate canonical.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44556 is the strongest standalone issue in the set: it describes a reproducible checkpoint reload breakage across versions, which is more general and actionable than the narrower model- or test-specific reports.", + "best_issue_reason": "issue:43723 is the clearest general anchor: it is concrete, actionable, and broadly about v5 tokenizer loading breakage rather than a model-specific or test-only failure.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:28282", "right": "issue:33453", "accept": false, - "reason": "Both are loading-related failures, but one is a missing PyTorch dependency import error and the other is a tokenizer loading regression; different bug classes and code paths." + "reason": "Missing PyTorch import error vs tokenizer-loading regression; same general area, different bug." }, { "left": "issue:43867", "right": "issue:44484", "accept": false, - "reason": "State-dict sorting/load error versus save_pretrained shard-size question; unrelated behavior and no shared failure mode." + "reason": "State-dict load ordering issue vs a save_pretrained shard-size question; unrelated root causes." }, { "left": "issue:43618", "right": "issue:44857", "accept": false, - "reason": "CLIP output attentions assignment and LwDetrImageLoss AMP/CUDA crash are unrelated model issues." + "reason": "CLIP attentions regression vs LwDetr AMP/CUDA crash; different models and failures." }, { "left": "issue:44360", "right": "issue:45440", "accept": false, - "reason": "DSA indexer/ReLU discussion is unrelated to DeepseekV3MoE implementation divergence." + "reason": "DSA indexer/ReLU discussion vs DeepseekV3MoE implementation divergence; not the same bug." }, { "left": "issue:43618", "right": "issue:44977", "accept": false, - "reason": "CLIP attentions bug and Qwen3.5 flash-attention generation bug are distinct model/runtime problems." + "reason": "CLIP attentions assignment bug vs Qwen3.5 flash-attention generation bug; different code paths." }, { "left": "issue:42890", "right": "issue:44964", "accept": false, - "reason": "Test seed flakiness in SAM HQ is unrelated to Phi-4 multimodal model loading." + "reason": "Flaky integration test from missing seed vs Phi-4 multimodal load failure; unrelated." }, { "left": "issue:42890", "right": "issue:44112", "accept": false, - "reason": "Both involve flaky tests, but they are different tests with different root causes; not the same underlying bug." + "reason": "Test flakiness vs CI stale device override failure; both test-related, but not the same issue." }, { "left": "issue:42890", "right": "issue:45600", "accept": false, - "reason": "SAM HQ test seeding issue and auto_mappings CI breakage are unrelated." + "reason": "Missing set_seed flake vs auto_mappings CI break; different causes." }, { "left": "issue:43502", "right": "issue:44062", "accept": false, - "reason": "Local-files-only network leak versus AddedToken argument duplication are different bugs." + "reason": "local_files_only network leakage vs AddedToken kwarg collision; unrelated failures." }, { "left": "issue:43576", "right": "issue:46132", "accept": false, - "reason": "Broken env command and changed AttentionInterface.register behavior are unrelated API issues." + "reason": "v5 env command breakage vs AttentionInterface.register behavior change; different features." }, { "left": "issue:43723", "right": "issue:44556", "accept": false, - "reason": "Tokenizer loading in v5 and checkpoint reload failures after v4.57\u2192v5.2/5.3 are both loading regressions, but not the same concrete code path." + "reason": "Both are loading regressions, but tokenizer.from_pretrained failure and checkpoint reload failure are distinct paths." }, { "left": "issue:44387", "right": "issue:44556", "accept": false, - "reason": "Int4 CUDA reserved-memory/OOM regression is unrelated to checkpoint reload compatibility." + "reason": "Int4 memory regression vs checkpoint reload regression; different symptoms and fixes." }, { "left": "issue:42548", "right": "issue:44351", "accept": false, - "reason": "Import errors for PreTrainedModel and HybridCache are separate symbols with no evidence of a shared root cause." + "reason": "Import errors for different symbols from different packages; not duplicates." }, { "left": "issue:42890", "right": "issue:45341", "accept": false, - "reason": "SAM HQ test seed flakiness and a small testing_utils bug are not duplicates." + "reason": "SamHQ test flake vs a testing_utils bug; different scopes." }, { "left": "issue:43010", "right": "issue:43065", "accept": false, - "reason": "No-grad decoration for cache/layer updates is unrelated to Sam3PixelDecoder's dummy Conv2d." + "reason": "@no_grad cache update fix vs dummy Conv2d in Sam3PixelDecoder; unrelated changes." }, { "left": "issue:44222", "right": "issue:45125", "accept": false, - "reason": "FP8 save_pretrained behavior and missing _tp_plan for tensor parallelism are both Qwen/MoE-adjacent, but they affect different features and code paths." + "reason": "FP8 save_pretrained bug vs missing tensor-parallel plan; same broad area, different concrete problems." }, { "left": "issue:43502", "right": "issue:44485", "accept": false, - "reason": "API requests despite local_files_only and GLM-5 RoPE implementation are unrelated." + "reason": "Network call leakage vs GLM-5 RoPE discussion; no shared bug." }, { "left": "issue:44355", "right": "issue:44749", "accept": false, - "reason": "Compiled Python file errors versus a performance regression after upgrade are not the same bug." + "reason": "Compiled Python file errors vs slower filtering after upgrade; different symptoms." }, { "left": "issue:45125", "right": "issue:45478", "accept": false, - "reason": "Both concern Qwen3.5 MoE loading, but one is specifically tensor-parallel plan support and the other is a broader from_pretrained error; not enough evidence they are the same defect." + "reason": "Missing _tp_plan vs Qwen3.5 Moe from_pretrained error; related family, but not the same fixable bug." }, { "left": "issue:30333", "right": "issue:33453", "accept": false, - "reason": "MLFlow reporting on training failure and tokenizer loading regression are unrelated." + "reason": "MLflow reporting failure vs tokenizer loading regression; unrelated." }, { "left": "issue:43010", "right": "issue:43232", "accept": false, - "reason": "Both are generation-related, but one is about no_grad decoration on cache/layer updates and the other about _update_model_kwargs_for_generation after sync_gpus." + "reason": "Cache update no_grad fix vs generation kwarg-sync issue; different code paths." }, { "left": "issue:45125", "right": "issue:45375", "accept": false, - "reason": "Tensor-parallel _tp_plan and a missing config field dropped by @strict are different issues." + "reason": "Missing tensor-parallel plan vs missing config field; different model metadata bugs." }, { "left": "issue:30990", "right": "issue:31795", "accept": false, - "reason": "Sentence Transformers loading hang and documentation confusion about model.forward args are unrelated." + "reason": "Sentence-Transformers loading stall vs documentation confusion; not duplicates." }, { "left": "issue:43502", "right": "issue:44222", "accept": false, - "reason": "local_files_only network leakage and FP8 save_pretrained moe are separate problems." + "reason": "local_files_only API leakage vs FP8 save_pretrained bug; different features." }, { "left": "issue:43299", "right": "issue:44556", "accept": false, - "reason": "Qwen3VL MoE loading breakage in v5.0.0.dev0 and checkpoint reload failures after v4.57\u2192v5.2/5.3 are different regressions." + "reason": "Both are model-loading regressions, but they affect different model families and likely need different fixes." }, { "left": "issue:42831", "right": "issue:43874", "accept": false, - "reason": "FineGrainedFP8 accuracy and GLM46V image-processor token counting are unrelated." + "reason": "FP8 accuracy issue vs missing image-patch-count method; different underlying defects." }, { "left": "issue:44315", "right": "issue:44360", "accept": false, - "reason": "Liger Kernel application in model_init and DSA indexer/ReLU discussion have no shared bug." + "reason": "Liger Kernel not applied via model_init vs DSA indexer ReLU discussion; unrelated." }, { "left": "issue:43502", "right": "issue:44961", "accept": false, - "reason": "local_files_only network behavior and an unrelated 'racoon' issue are clearly different." + "reason": "Real bug vs placeholder title; no duplicate signal." }, { "left": "issue:43827", "right": "issue:44961", "accept": false, - "reason": "Docs still referencing pipeline() and 'racoon' are unrelated." + "reason": "Docs still mention pipeline() vs placeholder title; not a duplicate pair." }, { "left": "issue:30333", "right": "issue:36296", "accept": false, - "reason": "MLFlow reporting on training failure and tensor-parallel training bug are distinct training concerns." + "reason": "MLflow reporting bug vs tensor-parallel training bug; different subsystems." }, { "left": "issue:44355", "right": "issue:45071", "accept": false, - "reason": "Compiled Python file errors and PretrainedConfig type-checking regression are different failure modes." + "reason": "Compiled-file runtime errors vs PretrainedConfig type-checking regression; different problems." }, { "left": "issue:44355", "right": "issue:44655", "accept": false, - "reason": "Compiled Python file errors and pipeline save_pretrained failures are unrelated." + "reason": "Compiled-file runtime errors vs pipeline save_pretrained problem; unrelated." }, { "left": "issue:33453", "right": "issue:34689", "accept": false, - "reason": "Tokenizer loading regression and Llama 3.2 Vision model-loading breakage are separate loading bugs." + "reason": "Tokenizer-loading regression vs Llama 3.2 model-loading regression; similar wording, different concrete failures." }, { "left": "issue:42757", "right": "issue:43576", "accept": false, - "reason": "Missing is_offline_mode import from huggingface_hub and broken transformers env command are different issues." + "reason": "Missing export from huggingface_hub vs broken transformers env command; distinct bugs." }, { "left": "issue:43576", "right": "issue:45071", "accept": false, - "reason": "Broken env command and PretrainedConfig type checking regression are unrelated." + "reason": "CLI env regression vs config type-checking regression; not the same issue." }, { "left": "issue:44222", "right": "issue:44230", "accept": false, - "reason": "FP8 save_pretrained for MoE and FP8 inference support for Qwen3-VL/Qwen3.5 are related thematically but not the same concrete bug." + "reason": "FP8 save_pretrained bug vs FP8 inference support request/bug; related area but not one fix." }, { "left": "issue:33453", "right": "issue:36296", "accept": false, - "reason": "Tokenizer loading regression and tensor parallel training bug are unrelated." + "reason": "Tokenizer-loading regression vs tensor-parallel training bug; different root causes and code paths." } ] }, @@ -14885,244 +15029,262 @@ "issue:44295|issue:45230", "issue:44908|issue:45071" ], - "prepared_review_unit_hash": "978d8df38abff134ddb7910781a66acd3993774ae13a72ae0e9ef6ecbc74f92b", + "prepared_review_unit_hash": "cbc8b9166ba34a4ffa013b9da34de4989b09fe92509b3e0785279957ea74b557", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 38, - "serialized_chars": 38439, - "estimated_input_tokens": 9610, - "estimated_eval_tokens": 19476 + "serialized_chars": 38435, + "estimated_input_tokens": 9609, + "estimated_eval_tokens": 19474 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is very noisy: most items are unrelated issue reports that only share broad Transformers-v5 regression or import-failure themes. I would not merge any of the soft pairs as duplicates.", - "confidence": 0.91, - "canonical_issue_reason": "issue:45850 is the broadest umbrella in the set and best matches the overall repeated theme of widespread v5 breakage, even though it is still too general to be a precise duplicate target.", + "summary": "This cluster is mostly a grab bag of unrelated Transformers v5 regressions: import failures, loading/lazy-loading issues, model-specific config bugs, and a few docs/test reports. None of the soft pairs looks like the same concrete bug/change, so I would not merge any of them as duplicates. The broadest umbrella issue is the general v5 breakage report.", + "confidence": 0.48, + "canonical_issue_reason": "Issue 45850 is the broadest umbrella in the set: it explicitly frames a general 'v5 breaks too many models' regression, which is the best catch-all for the cluster\u2019s version-5 compatibility complaints.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45850 is the most representative issue for the cluster\u2019s main recurring theme and can serve as the least-bad umbrella among otherwise heterogeneous reports.", + "best_issue_reason": "Issue 45850 is the most suitable global representative because it is the broadest open regression report and can anchor follow-up links to more specific v5 breakages.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43576", "right": "issue:43824", "accept": false, - "reason": "Both are v5-related import/runtime problems, but one is a broken CLI command and the other is a missing model export. Different code paths and fixes." + "reason": "Both are v5-related, but one is a broken `transformers env` command and the other is an ImportError for a specific model class; different bugs." }, { "left": "issue:41628", "right": "issue:42548", "accept": false, - "reason": "These are separate missing-import errors for different symbols (`AutoImageProcessor` vs `PreTrainedModel`)." + "reason": "These are different missing imports (`AutoImageProcessor` vs `PreTrainedModel`) and do not point to the same failure path." }, { "left": "issue:31795", "right": "issue:38617", "accept": false, - "reason": "Documentation confusion in `forward` is unrelated to an ImportError for `layer_type_validation`." + "reason": "Documentation confusion in `model.forward` is unrelated to an ImportError for `layer_type_validation`." }, { "left": "issue:44961", "right": "issue:45468", "accept": false, - "reason": "No common underlying bug; one is a placeholder/generic issue and the other is a Gemma audio positional encoding defect." + "reason": "A meaningless title (`racoon`) does not match the Gemma-4 audio positional encoding bug." }, { "left": "issue:38617", "right": "issue:43576", "accept": false, - "reason": "A missing import from `configuration_utils` is not the same as the `transformers env` command breaking in v5." + "reason": "One is a specific import failure, the other is a broken CLI command; only a broad v5 theme is shared." }, { "left": "issue:44908", "right": "issue:44961", "accept": false, - "reason": "Scheduler kwargs handling and an unrelated placeholder issue are not the same bug." + "reason": "The scheduler kwargs bug has nothing to do with the unrelated `racoon` issue." }, { "left": "issue:43673", "right": "issue:44492", "accept": false, - "reason": "A missing cache object during chunked prefill is unrelated to a typo in cache strategy docs/code." + "reason": "Chunked-prefill cache missing is a runtime bug; the cache-strategies item is just a typo/wording issue." }, { "left": "issue:44485", "right": "issue:44961", "accept": false, - "reason": "RoPE implementation discussion is unrelated to the placeholder issue." + "reason": "GLM-5 RoPE implementation and `racoon` are clearly unrelated." }, { "left": "issue:36296", "right": "issue:43901", "accept": false, - "reason": "Tensor parallel training and pipeline docs mismatch are different subsystems and failures." + "reason": "Tensor-parallel training is unrelated to a documentation mismatch about `return_all_scores`." }, { "left": "issue:41628", "right": "issue:44355", "accept": false, - "reason": "Importing a model class and errors running compiled Python files are distinct problems." + "reason": "A missing `AutoImageProcessor` import and errors in compiled Python files are different problems." }, { "left": "issue:36296", "right": "issue:43526", "accept": false, - "reason": "Tensor parallel training bug is unrelated to BeitImageProcessorFast label reduction behavior." + "reason": "Tensor-parallel training and `BeitImageProcessorFast.reduce_labels` are different code paths." }, { "left": "issue:45362", "right": "issue:45850", "accept": false, - "reason": "The Qwen3.5 chat crash is a specific model/runtime issue; 45850 is a broad complaint about many models breaking, not the same bug." + "reason": "Both are broad regression complaints, but one is a specific Qwen3.5 chat crash and the other is a general v5 breakage meta-report." }, { "left": "issue:44623", "right": "issue:45230", "accept": false, - "reason": "Missing files in `processor.save_pretrained` is not the same as a generic bug report." + "reason": "`processor.save_pretrained(...)` missing files is a concrete bug; `Bug report` is too vague to be the same issue." }, { "left": "issue:44261", "right": "issue:44360", "accept": false, - "reason": "These refer to different internal components and different numerical/activation bugs." + "reason": "MLA layernorm precision/config issue and a DSA indexer ReLU omission are unrelated." }, { "left": "issue:44273", "right": "issue:45542", "accept": false, - "reason": "Lazy loading behavior is unrelated to the tensorboard-without-TF backend error." + "reason": "Lazy loading malfunction and a TensorBoard-without-TensorFlow backend error are different failures." }, { "left": "issue:44273", "right": "issue:45003", "accept": false, - "reason": "Both touch import/lazy-loading concerns, but one is a general lazy-loading failure and the other is unsafe `sys.modules` access; not the same concrete bug." + "reason": "Lazy loading and unsafe `sys.modules` access may be adjacent internals, but they are not the same concrete bug." }, { "left": "issue:44038", "right": "issue:45125", "accept": false, - "reason": "Both involve Qwen/Moe, but one is a Qwen3-VL bug and the other is missing tensor-parallel plan metadata; different fixes." + "reason": "A Qwen3-VL bug report and missing `_tp_plan` for Qwen3_5Moe tensor parallelism are different model-specific issues." }, { "left": "issue:44393", "right": "issue:45071", "accept": false, - "reason": "Qwen3-VL bounding-box generation and `PretrainedConfig` type-checking are unrelated." + "reason": "A Qwen3-VL 2D bbox error is unrelated to `PretrainedConfig` type-checking breakage." }, { "left": "issue:44393", "right": "issue:45753", "accept": false, - "reason": "Both are Qwen vision issues, but one is 2D bbox hallucination/error and the other is an infinite loop on a specific image; not the same underlying defect." + "reason": "Both mention Qwen3-VL/Qwen3_5, but one is bounding-box output corruption and the other is an infinite loop on a specific image; not the same bug." }, { "left": "issue:44230", "right": "issue:46095", "accept": false, - "reason": "FP8 inference support for Qwen models is unrelated to a DeepSeekV4 weight conversion question." + "reason": "FP8 inference support for Qwen models and a DeepSeekV4 weight-conversion question are unrelated." }, { "left": "issue:43576", "right": "issue:44556", "accept": false, - "reason": "A broken env command and checkpoint reload incompatibility after upgrade are different regressions." + "reason": "Broken `transformers env` and inability to reload old checkpoints after upgrading are separate regression classes." }, { "left": "issue:41669", "right": "issue:43673", "accept": false, - "reason": "Import-star cleanup/performance work is unrelated to cache handling in generation during chunked prefill." + "reason": "Import-time `import *` slowdown and missing GenerationMixin cache in chunked prefill are unrelated." }, { "left": "issue:43525", "right": "issue:45850", "accept": false, - "reason": "Missing `pad_token_id` on `Llama4Config` is a specific config bug, not the broad multi-model breakage described in 45850." + "reason": "Missing `pad_token_id` in Llama4Config is a specific model config bug, not the same as the broad v5 breakage complaint." }, { "left": "issue:44556", "right": "issue:45092", "accept": false, - "reason": "Both are upgrade-related loading problems, but one is general checkpoint reload failure and the other is remote-code/meta-init incompatibility for InternVL2; different code paths." + "reason": "Both are checkpoint-compatibility issues after v5, but they affect different model families and likely different load paths." }, { "left": "issue:43572", "right": "issue:45850", "accept": false, - "reason": "Missing `pad_token_idx` in StableLmConfig is a narrow config regression, not the broad umbrella issue." + "reason": "StableLmConfig missing `pad_token_idx` is a specific config regression, not the same as the generic v5 complaint." }, { "left": "issue:44492", "right": "issue:45850", "accept": false, - "reason": "A typo in cache strategies is not the same as broad v5 breakage." + "reason": "A typo in cache-strategy docs/code is not the same as broad model breakage." }, { "left": "issue:42371", "right": "issue:44492", "accept": false, - "reason": "TF32 API guidance and a cache-strategy typo are unrelated." + "reason": "TF32 settings behavior and cache-strategies typo are unrelated." }, { "left": "issue:45278", "right": "issue:45850", "accept": false, - "reason": "Both mention widespread breakage after upgrade, but 45278 is specifically about import errors while 45850 is a broad model-breakage complaint; too coarse to treat as one bug." + "reason": "The import-errors-on-upgrade report is related at a high level, but it is still a different concrete problem than the generic 'many models break' meta-issue." }, { "left": "issue:42890", "right": "issue:44908", "accept": false, - "reason": "A flaky test due to missing `set_seed` has nothing to do with scheduler kwargs handling." + "reason": "A missing `set_seed()` in tests is unrelated to scheduler kwargs handling." }, { "left": "issue:44246", "right": "issue:44273", "accept": false, - "reason": "Slow import timing and lazy loading behavior are related at a high level, but the reports do not describe the same concrete failure." + "reason": "Slow import timing and broken lazy loading are both import-related, but they are not the same underlying bug." }, { "left": "issue:44704", "right": "issue:44908", "accept": false, - "reason": "`AutoProcessor.from_pretrained` not forwarding kwargs and inverse-sqrt scheduler kwargs are different APIs and bugs." + "reason": "Both mention kwargs, but one is `AutoProcessor.from_pretrained` not forwarding them and the other is an LR scheduler ignoring them." }, { "left": "issue:43408", "right": "issue:44315", "accept": false, - "reason": "A model-type warning for SAM3 and Liger Kernel application with `model_init` are unrelated." + "reason": "A model-type mismatch warning and Liger Kernel not applying under `model_init` are different issues." }, { "left": "issue:43976", "right": "issue:44393", "accept": false, - "reason": "Python-version incompatibility and Qwen3-VL bounding-box output errors are not the same issue." + "reason": "Python-version compatibility breakage is unrelated to a Qwen3-VL bounding-box hallucination/error." }, { "left": "issue:43352", "right": "issue:44393", "accept": false, - "reason": "Flash-Attention support missing for NemotronH is unrelated to Qwen3-VL bounding-box hallucination." + "reason": "Flash Attention 2 support for Nemotron and Qwen3-VL bbox output problems are different model features and bugs." }, { "left": "issue:43576", "right": "issue:44908", "accept": false, - "reason": "A broken env command and an inverse-sqrt scheduler kwargs bug are unrelated." + "reason": "Broken `transformers env` and scheduler kwargs ignored do not share the same code path." + }, + { + "left": "issue:44297", + "right": "issue:45230", + "accept": false, + "reason": "Tokenizer config mismatch and a generic bug report are not the same issue." + }, + { + "left": "issue:44295", + "right": "issue:45230", + "accept": false, + "reason": "Position_ids buffer read errors are unrelated to the vague bug report." + }, + { + "left": "issue:44908", + "right": "issue:45071", + "accept": false, + "reason": "Scheduler kwargs handling and `PretrainedConfig` type-checking breakage are unrelated." } ] }, @@ -15239,274 +15401,286 @@ "issue:31515|issue:38617", "issue:44230|issue:44623" ], - "prepared_review_unit_hash": "7bfe29e3ee4e533425bbdb721c3985e674a248aac6b75b6590305f5e8d81ead0", + "prepared_review_unit_hash": "6811cc7eba357048bca7c967009d8d1285b6da2c66f46ffdb0cf161aa4987ff2", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 42, - "serialized_chars": 39663, - "estimated_input_tokens": 9916, - "estimated_eval_tokens": 20088 + "serialized_chars": 39658, + "estimated_input_tokens": 9915, + "estimated_eval_tokens": 20086 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a mix of unrelated Transformers bugs and docs issues with a loose common thread around model/tokenizer loading and v5 breakage; I found no true duplicate pair among the soft edges.", - "confidence": 0.34, - "canonical_issue_reason": "issue:45850 is the broadest umbrella report about v5 breaking many models, which best matches the cluster\u2019s general loading/compatibility theme, even though it is not a duplicate of the more specific bugs.", + "summary": "This cluster is mostly noisy false-positive similarity across unrelated Transformers issues. The clearest representative is the Qwen3.5 MoE from_pretrained regression; the later same-title report looks like the closest duplicate anchor. No PRs are present.", + "confidence": 0.69, + "canonical_issue_reason": "issue:45310 is the earliest concrete report of a specific from_pretrained regression for Qwen3.5 Moe and is the best anchor for the duplicated same-title follow-up.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45850 is the most representative anchor for this cluster because it is broad, central to the shared v5-regression theme, and higher-level than the other narrow edge cases.", + "best_issue_reason": "issue:45310 is the most useful issue to keep as the representative bug report: specific, version-scoped, and directly tied to a reproducible model-loading failure.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:41669", "right": "issue:44351", "accept": false, - "reason": "Both are import/load-related, but one is about removing import * for performance and the other is a missing HybridCache export; different defects and fixes." + "reason": "Both are import-related, but one is about import-star performance/scale and the other is a missing public import; different failure modes." }, { "left": "issue:41669", "right": "issue:45341", "accept": false, - "reason": "Unrelated: import * cleanup/perf vs a testing_utils bug." + "reason": "Unrelated: model import mechanics vs a small testing_utils bug." }, { "left": "issue:44492", "right": "issue:44945", "accept": false, - "reason": "Cache-strategy typo and pipeline-parallelism output corruption are different code paths and symptoms." + "reason": "Cache-strategy typo and pipeline-parallelism output bug are different code paths." }, { "left": "issue:43874", "right": "issue:45850", "accept": false, - "reason": "Specific missing image-patch method vs a generic v5 model-breakage complaint; not the same bug." + "reason": "A missing image-processor method is too specific to merge with a broad '5.0 breaks too many models' complaint." }, { "left": "issue:44393", "right": "issue:45478", "accept": false, - "reason": "Different models and failures: 2D bounding box hallucination vs from_pretrained error in Qwen3.5 MoE." + "reason": "Different model families and different bugs: 2D bbox output error vs Qwen3.5 MoE from_pretrained failure." }, { "left": "issue:44077", "right": "issue:44861", "accept": false, - "reason": "PatchTST post_init policy and tied-weights key handling are unrelated." + "reason": "One is a validation policy issue, the other is an AttributeError in tied-weight key handling." }, { "left": "issue:43827", "right": "issue:44908", "accept": false, - "reason": "Docs still referencing pipeline() and inverse_sqrt scheduler kwargs are separate issues." + "reason": "Docs still referencing pipeline() and scheduler kwargs handling are unrelated." }, { "left": "issue:41669", "right": "issue:44908", "accept": false, - "reason": "Performance/import cleanup vs scheduler configuration bug; no shared code path." + "reason": "Import-star cleanup and scheduler kwargs are unrelated." }, { "left": "issue:31795", "right": "issue:33453", "accept": false, - "reason": "Docs confusion about forward args vs tokenizer loading regression; related domain, different problem." + "reason": "Documentation confusion in forward signatures does not match a tokenizer loading regression." }, { "left": "issue:44315", "right": "issue:44829", "accept": false, - "reason": "Liger kernel application in model_init and flash_attention_3 training degeneration are not the same defect." + "reason": "Both involve training/model creation, but they are different concrete bugs and not one mergeable fix." }, { "left": "issue:43576", "right": "issue:44273", "accept": false, - "reason": "Broken env command and lazy loading failure are separate v5 regressions." + "reason": "Broken env command and lazy loading are separate issues." }, { "left": "issue:44360", "right": "issue:45468", "accept": false, - "reason": "DSA indexer activation issue vs Gemma-4 audio positional encoding bug; different subsystems." + "reason": "DSA indexer activation behavior and Gemma audio positional encoding are unrelated." }, { "left": "issue:44655", "right": "issue:44861", "accept": false, - "reason": "Saving pipeline objects and tied-weight key extraction are unrelated save/load vs model-weight utility bugs." + "reason": "Saving Pipeline objects and tied-weight key crashes do not share the same underlying defect." }, { "left": "issue:45230", "right": "issue:45310", "accept": false, - "reason": "Generic bug report and Qwen3.5 MoE from_pretrained error are not the same concrete failure." + "reason": "'Bug report' is too generic to establish the same concrete from_pretrained regression." }, { "left": "issue:44273", "right": "issue:44908", "accept": false, - "reason": "Lazy loading and scheduler kwargs are unrelated." + "reason": "Lazy loading and inverse_sqrt scheduler kwargs are different subsystems." }, { "left": "issue:43519", "right": "issue:44485", "accept": false, - "reason": "Qwen3VL timestamp calculation and GLM-5 RoPE implementation are different model-specific bugs." + "reason": "Timestamp calculation in Qwen3VL and GLM-5 RoPE implementation are different model-specific bugs." }, { "left": "issue:43519", "right": "issue:44908", "accept": false, - "reason": "Processor timestamp logic and scheduler configuration are unrelated." + "reason": "A processor timestamp bug is unrelated to scheduler kwarg propagation." }, { "left": "issue:31795", "right": "issue:34689", "accept": false, - "reason": "Forward-doc confusion and Llama 3.2 model loading breakage are distinct issues." + "reason": "Forward-doc confusion and Llama 3.2 Vision loading regression are not the same bug." }, { "left": "issue:41669", "right": "issue:43704", "accept": false, - "reason": "Import * cleanup and VRAM leakage across dataloader threads do not share the same root cause." + "reason": "Import-star cleanup and VRAM leak under dataloader threads are unrelated." }, { "left": "issue:31795", "right": "issue:37428", "accept": false, - "reason": "Documentation of model.forward parameters and flash-attention helper import failure are unrelated." + "reason": "Forward docs and a missing flash-attention import are unrelated." }, { "left": "issue:44315", "right": "issue:45092", "accept": false, - "reason": "Different model-init problems affecting different model families and code paths." + "reason": "Both mention model_init/meta-init context, but the concrete failures are different and not one code-path fix." }, { "left": "issue:44485", "right": "issue:45468", "accept": false, - "reason": "RoPE implementation discussion and Gemma-4 audio positional encoding are separate model bugs." + "reason": "Different model implementations and different symptoms." }, { "left": "issue:43704", "right": "issue:44485", "accept": false, - "reason": "Memory leak across threads vs RoPE implementation; no overlap in defect or fix." + "reason": "VRAM leak and RoPE implementation are unrelated." }, { "left": "issue:43298", "right": "issue:46095", "accept": false, - "reason": "Continuous batching support for audio models is unrelated to a deepseekv4 weight conversion question." + "reason": "Continuous batching for audio models and a DeepSeek weight-conversion question are unrelated." }, { "left": "issue:43502", "right": "issue:43519", "accept": false, - "reason": "local_files_only network behavior and Qwen3VL timestamp calculation are different bugs." + "reason": "Local-files-only network leakage and timestamp calculation are unrelated." }, { "left": "issue:42371", "right": "issue:43704", "accept": false, - "reason": "TF32 API settings and VRAM leakage are unrelated." + "reason": "TF32 settings guidance and VRAM leak are unrelated." }, { "left": "issue:43519", "right": "issue:44623", "accept": false, - "reason": "Timestamp math in a processor and missing processor.save_pretrained files are separate issues." + "reason": "Processor timestamp math and missing processor save files are separate bugs." }, { "left": "issue:43519", "right": "issue:43704", "accept": false, - "reason": "Processor timestamp bug vs dataloader-thread VRAM leak; different component and symptom." + "reason": "Timestamp calculation and VRAM leakage do not share a code path." }, { - "left": "issue:31795", - "right": "issue:34689", + "left": "issue:43606", + "right": "issue:45468", + "accept": false, + "reason": "CPU offload device mismatch and Gemma audio positional encoding are unrelated." + }, + { + "left": "issue:44263", + "right": "issue:45468", + "accept": false, + "reason": "torch.split return handling in an indexer and Gemma audio positional encoding are different defects." + }, + { + "left": "issue:44261", + "right": "issue:45468", "accept": false, - "reason": "Documentation confusion and a concrete model-loading regression are not the same bug." + "reason": "RMS norm epsilon precision issue in MLA and Gemma audio positional encoding are unrelated." }, { - "left": "issue:316?", - "right": "issue:37428", + "left": "issue:42913", + "right": "issue:43066", "accept": false, - "reason": "Not enough direct overlap; different missing-import vs docs problems." + "reason": "Both are tokenizer v5 issues, but one is broad behavior drift while the other is a specific decoder-type problem; not enough to treat as the same concrete bug." }, { - "left": "issue:44315", - "right": "issue:45092", + "left": "issue:44908", + "right": "issue:44998", "accept": false, - "reason": "Both mention model initialization, but they affect different architectures and failure modes." + "reason": "Scheduler kwargs and an unrelated placeholder issue are not duplicates." }, { - "left": "issue:44485", - "right": "issue:45468", + "left": "issue:44492", + "right": "issue:44998", "accept": false, - "reason": "Distinct model-specific implementation bugs." + "reason": "Cache-strategy typo and an unrelated placeholder issue are not duplicates." }, { - "left": "issue:43704", - "right": "issue:44485", + "left": "issue:44861", + "right": "issue:45290", "accept": false, - "reason": "Memory leak and RoPE implementation issues are unrelated." + "reason": "Tied-weight key handling and chat-template tool-call crashes are different code paths." }, { - "left": "issue:43298", - "right": "issue:46095", + "left": "issue:43618", + "right": "issue:44998", "accept": false, - "reason": "Audio continuous batching support and deepseekv4 conversion-script request are different topics." + "reason": "CLIPOutput attentions assignment and an unrelated placeholder issue are not duplicates." }, { - "left": "issue:43502", - "right": "issue:43519", + "left": "issue:44998", + "right": "issue:45850", "accept": false, - "reason": "Different components: loading transport vs processor math." + "reason": "A placeholder issue cannot be merged with a broad v5 complaints issue." }, { - "left": "issue:42371", - "right": "issue:43704", + "left": "issue:44998", + "right": "issue:45341", "accept": false, - "reason": "TF32 configuration guidance does not match a VRAM leak bug." + "reason": "An unrelated placeholder issue and a testing_utils bug are not duplicates." }, { - "left": "issue:43519", - "right": "issue:44623", + "left": "issue:43525", + "right": "issue:44861", "accept": false, - "reason": "Processor timestamp bug and save_pretrained file omission are distinct failures." + "reason": "Missing pad_token_id on Llama4Config and tied-weight key AttributeError are different issues." }, { - "left": "issue:43519", - "right": "issue:43704", + "left": "issue:43572", + "right": "issue:44861", "accept": false, - "reason": "Different root causes and code paths." + "reason": "Missing pad_token_idx on StableLmConfig and tied-weight key handling are unrelated." }, { "left": "issue:31515", "right": "issue:38617", "accept": false, - "reason": "Slow from_pretrained checkpoint loading and missing configuration_utils import are not duplicates." + "reason": "Slow checkpoint loading and missing layer_type_validation import are unrelated." }, { "left": "issue:44230", "right": "issue:44623", "accept": false, - "reason": "FP8 support request and missing files in processor.save_pretrained are unrelated." + "reason": "fp8 inference support request and processor save_pretrained missing files are unrelated." } ] }, @@ -15547,20 +15721,20 @@ "issue:31515|issue:33453", "issue:43298|issue:44230" ], - "prepared_review_unit_hash": "dbfe363ab0421ea0a846c7d6942fd104b1fabf93dacfe6b73372f3de1c3f0f2c", + "prepared_review_unit_hash": "b87fac37ae6cd5746a3974ae02fa2b9277272c07e1847aeee05a4ca6f6f70f5e", "component_budget": { "node_count": 414, "item_count": 414, "soft_pair_count": 1552, - "serialized_chars": 800311, - "estimated_input_tokens": 200078, - "estimated_eval_tokens": 400412 + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 }, "budget": { "node_count": 8, "item_count": 8, "soft_pair_count": 6, - "serialized_chars": 6305, + "serialized_chars": 6307, "estimated_input_tokens": 1577, "estimated_eval_tokens": 3410 }, @@ -15569,48 +15743,48 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These issues are largely unrelated; the soft-similarity edges are driven by generic phrases like \"from_pretrained\" but do not describe the same bug. No pair looks like a true duplicate or mergeable same-change candidate.", - "confidence": 0.88, - "canonical_issue_reason": "issue:45310 is the clearest concrete bug report in the set: it names a specific regression, affected model family, and failure mode, whereas the others are unrelated or more generic.", + "summary": "These issues are mostly unrelated and do not form a duplicate cluster. The only apparent thematic overlap is between the two Qwen/MoE loading issues, but they still describe different failure modes and scopes.", + "confidence": 0.93, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:45310 is the best single representative issue because it is specific, actionable, and has the strongest bug-focused framing.", + "best_issue_reason": "issue:45310 is the strongest representative of the cluster\u2019s bug-report style: it is concrete, recent, and describes a specific regression in from_pretrained for a named model family. That said, the set overall is too heterogeneous for a true canonical issue.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:30333", "right": "issue:31515", "accept": false, - "reason": "Different subsystems and failure modes: MLFlow/job status reporting vs slow checkpoint loading." + "reason": "Different subsystems and symptoms: MLflow reporting after failed training vs slow checkpoint loading in from_pretrained." }, { "left": "issue:31515", "right": "issue:36296", "accept": false, - "reason": "One is about slow checkpoint loading, the other about tensor parallel training; no shared code-path bug is evident." + "reason": "One is about slow checkpoint load time; the other is a tensor-parallel training bug. No same code-path evidence." }, { "left": "issue:31515", "right": "issue:31795", "accept": false, - "reason": "Performance/load-time complaint vs documentation confusion; not the same underlying issue." + "reason": "Checkpoint load performance is unrelated to documentation confusion about input_ids and past_key_values." }, { "left": "issue:44230", "right": "issue:45310", "accept": false, - "reason": "Both involve Qwen3/MoE/fp8 wording, but one is inference support for a model class and the other is a from_pretrained regression; too different to treat as one bug." + "reason": "Both mention Qwen3.5/MoE and loading/inference, but one is fp8 inference support and the other is a transformers>=5.4.0 from_pretrained error. Different concrete fixes." }, { "left": "issue:31515", "right": "issue:33453", "accept": false, - "reason": "Checkpoint loading speed and tokenizer loading regression are separate problems affecting different parts of model initialization." + "reason": "Slow checkpoint loading and tokenizer loading regression are both loading-related but not the same bug or code path." }, { "left": "issue:43298", "right": "issue:44230", "accept": false, - "reason": "Continuous batching audio-model support and fp8 Qwen3-VL/Qwen3.5 MoE inference support are distinct feature/compatibility gaps." + "reason": "Continuous batching support for audio models and fp8 inference support are separate feature/compatibility issues." } ] }, @@ -15643,7 +15817,7 @@ "issue:10519|issue:11005", "issue:11005|issue:11307" ], - "prepared_review_unit_hash": "0f9d53557157f515689d295ef95135160ad6e9709afec65a895faef966f0c370", + "prepared_review_unit_hash": "f8550c830647b2cd6150b5eb785f204a3c0e55af4d46d489eb30168238ebaef2", "component_budget": { "node_count": 3, "item_count": 3, @@ -15665,30 +15839,30 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These three issues are not duplicates of one another: one is about truncation behavior in tokenization, one requests ReduceLROnPlateau-style training scheduler behavior, and one asks for word-level time offsets in Wav2Vec2. They touch different features and code paths, so no soft-edge merge is justified.", - "confidence": 0.96, - "canonical_issue_reason": "Issue #10519 is the most concrete and self-contained feature request, but there is no true canonical duplicate in this set because the three issues are unrelated.", + "summary": "The three issues are not duplicates; they cover unrelated topics: tokenizer truncation behavior, a ReduceLROnPlateau-like training utility, and Wav2Vec2 word time offsets. No soft edge should be accepted.", + "confidence": 0.97, + "canonical_issue_reason": "Issue 10519 is the clearest standalone request: it specifies one concrete behavior change for truncation direction and is well-scoped.", "canonical_pr_reason": null, - "best_issue_reason": "#10519 is the clearest representative issue only because it is the most specific and actionable; the other two are separate feature requests rather than duplicates.", + "best_issue_reason": "10519 is the best representative issue only because it is the most concrete and actionable; the other two are separate feature requests in different areas.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:10519", "right": "issue:11307", "accept": false, - "reason": "Different subsystems and asks: truncation strategy vs Wav2Vec2 word-offset extraction. No shared underlying bug." + "reason": "Different subsystems and user goals: tokenizer truncation direction vs Wav2Vec2 word boundary timestamps." }, { "left": "issue:10519", "right": "issue:11005", "accept": false, - "reason": "Unrelated functionality: tokenizer truncation vs learning-rate scheduler behavior." + "reason": "Completely unrelated requests: truncation behavior vs learning-rate scheduler-like functionality." }, { "left": "issue:11005", "right": "issue:11307", "accept": false, - "reason": "Different domains and code paths: scheduler feature request vs speech-model timestamp offsets." + "reason": "No shared bug or change path; one is about optimization scheduling, the other about speech model offsets." } ] }, @@ -15720,7 +15894,7 @@ "issue:13244|issue:16618", "issue:15980|issue:16618" ], - "prepared_review_unit_hash": "ce6a22b4a3ef4deff020d287d7a403476914484eface55b55bd4893cc289716d", + "prepared_review_unit_hash": "be741c67b264bcfbb39dad868fecd1110b0e115a7ff4098c8594216ca60801d3", "component_budget": { "node_count": 3, "item_count": 3, @@ -15742,24 +15916,24 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The three issues are not duplicates of the same bug. Two are about model loading/authentication errors, while one is about Tapas tokenization mismatch with TensorFlow code. The soft-similarity links are too broad to accept.", - "confidence": 0.92, - "canonical_issue_reason": "Issue #16618 is the most central representative among these because it is a concrete model-loading failure with broad user impact, but it is still distinct from the other issues rather than a true duplicate target.", + "summary": "These three issues are only loosely related by transformers/model loading or tokenization themes, but the titles point to different problems: tokenizer mismatch, private-model auth/error messaging, and a generic inability to load a specific model. I would not merge any pair as duplicates.", + "confidence": 0.84, + "canonical_issue_reason": "Issue 16618 is the broadest and most discussion-heavy representative of the cluster, but it still appears to describe a different failure mode than the others.", "canonical_pr_reason": null, - "best_issue_reason": "Issue #16618 is the best single issue to represent the cluster if one must be chosen: it is the most general and actionable of the three, with clearer overlap potential to the authentication/model-loading topic than the Tapas tokenization report.", + "best_issue_reason": "Issue 16618 is the best single representative if one must be chosen, because it is the most general and likely to attract the most follow-up context.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:13244", "right": "issue:16618", "accept": false, - "reason": "Different bug classes: Tapas tokenization/parsing behavior versus loading bert-base-uncased. They do not describe the same underlying code-path failure." + "reason": "Tokenizer output mismatch for TAPAS is a different bug from a generic model-loading failure; no evidence they share the same code path." }, { "left": "issue:15980", "right": "issue:16618", "accept": false, - "reason": "Both mention loading models, but one is specifically a private-model login/error-message issue and the other is a generic load failure for a public model. Too broad to treat as the same bug." + "reason": "One is specifically about a bad auth-related error when downloading a private model, while the other is a generic failure loading bert-base-uncased; titles do not indicate the same underlying issue." } ] }, @@ -15789,7 +15963,7 @@ "soft_pairs": [ "issue:26350|issue:33260" ], - "prepared_review_unit_hash": "97e2558efbb6574eb0612f2050aa4972e563ab37e807329f1633dcf63b911b6c", + "prepared_review_unit_hash": "5864f33bdb333612b76f9a9d25ea04893163a2675ae634a49d79bb0829303bae", "component_budget": { "node_count": 2, "item_count": 2, @@ -15811,18 +15985,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are two separate feature-request/community-contribution issues: one for Flash Attention 2 support across more architectures and one for GGUF support across more architectures. They share a broad theme but not the same underlying change.", + "summary": "These are two separate community-contribution issues for different feature areas: Flash Attention 2 support vs GGUF support. They are not duplicates and should remain distinct.", "confidence": 0.97, - "canonical_issue_reason": null, + "canonical_issue_reason": "Issue 26350 is the older and more discussed thread, so it is the better canonical issue if one must be chosen for the cluster.", "canonical_pr_reason": null, - "best_issue_reason": "Neither issue is a canonical duplicate target for the other; they request distinct feature work in different areas.", + "best_issue_reason": "Issue 26350 is the strongest representative because it has higher discussion activity and is the earlier, more established request.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:26350", "right": "issue:33260", "accept": false, - "reason": "Different requested features and code paths (Flash Attention 2 vs GGUF support); same broad 'more architectures' theme is too general to treat as duplicates." + "reason": "Different underlying requests and code paths: one is about Flash Attention 2 architecture support, the other about GGUF support. They are not the same bug/change." } ] }, @@ -15864,12 +16038,12 @@ "issue:43834|issue:44365", "issue:44365|issue:44370" ], - "prepared_review_unit_hash": "3cc4d7f772880126ed8a4a4bc03d44b31aea8db2c1c347500e29b2fef6d0c860", + "prepared_review_unit_hash": "3ede41fe204c778e76c6b85a718d198f529edc9792a3afdea77c837270da12ed", "component_budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 10, - "serialized_chars": 6444, + "serialized_chars": 6443, "estimated_input_tokens": 1611, "estimated_eval_tokens": 3478 }, @@ -15877,7 +16051,7 @@ "node_count": 5, "item_count": 5, "soft_pair_count": 10, - "serialized_chars": 6444, + "serialized_chars": 6443, "estimated_input_tokens": 1611, "estimated_eval_tokens": 3478 }, @@ -15886,79 +16060,82 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "All items are about documentation translation, but the concrete language targets differ and the generic placeholder issues are not enough to prove a single underlying request. No soft edge is safe to merge as a duplicate.", - "confidence": 0.93, - "canonical_issue_reason": "issue:43834 is the most generic and representative of the repeated placeholder-template translation issues, so it is the best anchor if a single issue must represent this cluster.", + "summary": "This cluster mixes two distinct language-specific translation requests (Turkish, Spanish) with three near-identical generic template issues. Only the three generic template issues look like true duplicates of the same underlying item.", + "confidence": 0.74, + "canonical_issue_reason": "Issue 43834 is the earliest of the three near-identical generic template issues ('[i18n-] Translating docs to '), so it is the best duplicate representative within that subgroup.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43834 is the best issue candidate because it is the earliest generic template entry and closest to the other placeholder-language issues, though the cluster is not a true duplicate set.", + "best_issue_reason": "43834 is the cleanest representative of the duplicate template subgroup and is more suitable as the cluster anchor than the language-specific requests, which are distinct translation targets.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:27088", "right": "issue:44370", "accept": false, - "reason": "Turkish vs generic placeholder translation request; not the same concrete language request." + "reason": "Turkish-specific request vs generic placeholder; not the same concrete issue." }, { "left": "issue:27088", "right": "issue:44365", "accept": false, - "reason": "Turkish vs generic placeholder translation request; clearly different targets." + "reason": "Turkish-specific request vs generic placeholder; not the same concrete issue." }, { "left": "issue:27088", "right": "issue:43834", "accept": false, - "reason": "Specific Turkish translation request does not match the generic template issue." + "reason": "Turkish-specific request vs generic placeholder; not the same concrete issue." }, { "left": "issue:27088", "right": "issue:28936", "accept": false, - "reason": "Turkish and Spanish translation requests are distinct issues." + "reason": "Different target languages (Turkish vs Spanish), so these are distinct translation requests." }, { "left": "issue:28936", "right": "issue:43834", "accept": false, - "reason": "Specific Spanish translation request does not match the generic template issue." + "reason": "Spanish-specific request vs generic placeholder; not the same concrete issue." }, { "left": "issue:28936", "right": "issue:44365", "accept": false, - "reason": "Spanish vs generic placeholder translation request; not the same underlying request." + "reason": "Spanish-specific request vs generic placeholder; not the same concrete issue." }, { "left": "issue:28936", "right": "issue:44370", "accept": false, - "reason": "Spanish vs generic placeholder translation request; not the same underlying request." + "reason": "Spanish-specific request vs generic placeholder; not the same concrete issue." }, { "left": "issue:43834", "right": "issue:44370", - "accept": false, - "reason": "These are near-identical placeholder templates, but the evidence does not prove they are the same concrete request." + "accept": true, + "reason": "Near-identical generic template issue; appears to be the same underlying request duplicated." }, { "left": "issue:43834", "right": "issue:44365", - "accept": false, - "reason": "These are near-identical placeholder templates, but not enough evidence to treat them as the same underlying issue." + "accept": true, + "reason": "Near-identical generic template issue; appears to be the same underlying request duplicated." }, { "left": "issue:44365", "right": "issue:44370", - "accept": false, - "reason": "Both are generic placeholder template issues, but there is no evidence they refer to the same specific translation request." + "accept": true, + "reason": "Near-identical generic template issue; appears to be the same underlying request duplicated." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet: the two language-specific issues are distinct requests, and the three generic-template issues are near-identical and plausibly duplicates of one another. The soft-edge verdicts are conservative, and the representative choice of 43834 among the generic subgroup is reasonable." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -16030,82 +16207,82 @@ "issue:43837|issue:45177", "issue:45322|issue:46133" ], - "prepared_review_unit_hash": "b826ea2e7282b5e71179850acec85ca1a6b85ab2449d94c8c47342db9dbb20e3", + "prepared_review_unit_hash": "490eb4cb209d028eeadc68d474a8dfad4eaedb59a352a5ee6ccb1b9fce042d92", "component_budget": { "node_count": 16, "item_count": 16, "soft_pair_count": 36, - "serialized_chars": 21655, - "estimated_input_tokens": 5414, - "estimated_eval_tokens": 11084 + "serialized_chars": 21658, + "estimated_input_tokens": 5415, + "estimated_eval_tokens": 11086 }, "budget": { "node_count": 16, "item_count": 16, "soft_pair_count": 36, - "serialized_chars": 21655, - "estimated_input_tokens": 5414, - "estimated_eval_tokens": 11084 + "serialized_chars": 21658, + "estimated_input_tokens": 5415, + "estimated_eval_tokens": 11086 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are mostly distinct model-request issues, not duplicates: they share a broad \u201cadd support for model X\u201d pattern, but the requested models and modalities differ (text, audio, vision, OCR, etc.).", - "confidence": 0.95, - "canonical_issue_reason": "issue:42738 is the closest umbrella-like tracking issue, but it is still too narrow to canonically cover the rest of these distinct model requests.", + "summary": "This cluster is a loose set of model-request issues across different architectures and modalities. They share a broad \"add support for model X\" theme, but none of the pairs look like true duplicates of the same underlying bug/change.", + "confidence": 0.67, + "canonical_issue_reason": "issue:42738 is the broadest umbrella topic here ('BERT-like models with RoPE') and best represents the shared model-architecture-support theme, though it is still not a duplicate of the other specific model requests.", "canonical_pr_reason": null, - "best_issue_reason": "issue:42738 is the most general issue in the set, but none of the issues is a strong global duplicate target because they span different model families and capabilities.", + "best_issue_reason": "issue:42738 is the most general and reusable representative issue in the set; the rest are mostly specific requests for unrelated models or modalities.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:36817", "right": "issue:41211", "accept": false, - "reason": "EuroBert and DEIMv2 are different model requests; only the generic \u201cadd model\u201d theme overlaps." + "reason": "Both are model-add requests, but they concern different models and different architectures; not the same code-path change." }, { "left": "issue:36817", "right": "issue:43671", "accept": false, - "reason": "EuroBert text model request and Qwen3-TTS audio request are unrelated model families." + "reason": "EuroBert config support and Qwen3-TTS support are unrelated model additions." }, { "left": "issue:36817", "right": "issue:36879", "accept": false, - "reason": "EuroBert and RF-DETR are different models/tasks, so this is not the same underlying change." + "reason": "Different model families entirely: EuroBert vs RF-DETR." }, { "left": "issue:36879", "right": "issue:41211", "accept": false, - "reason": "RF-DETR and DEIMv2 are separate model additions with no concrete duplicate overlap." + "reason": "RF-DETR and DEIMv2 are separate model integrations with different implementation needs." }, { "left": "issue:36879", "right": "issue:43909", "accept": false, - "reason": "RF-DETR vision detection and LFM2.5 Audio are different modalities and implementation paths." + "reason": "RF-DETR is an object-detection model request; LFM2.5 Audio is an audio model request." }, { "left": "issue:36879", "right": "issue:44121", "accept": false, - "reason": "RF-DETR and weight-sparse transformer support are different model requests." + "reason": "Different model families and likely different adapters/config paths." }, { "left": "issue:36879", "right": "issue:43837", "accept": false, - "reason": "RF-DETR and Qwen3-ASR are unrelated; one is vision detection, the other speech recognition." + "reason": "RF-DETR and Qwen3-ASR are unrelated capabilities and would not merge into one PR." }, { "left": "issue:36879", "right": "issue:43671", "accept": false, - "reason": "RF-DETR and Qwen3-TTS are different tasks and code paths." + "reason": "Different Qwen3 TTS/ASR request vs RF-DETR object-detection support." }, { "left": "issue:37015", @@ -16117,163 +16294,157 @@ "left": "issue:37015", "right": "issue:44121", "accept": false, - "reason": "NeoBERT and OpenAI weight-sparse transformer are distinct architectures, not duplicates." + "reason": "NeoBERT and circuit-sparsity/circuitgpt are different model families with distinct implementation work." }, { "left": "issue:37015", "right": "issue:43909", "accept": false, - "reason": "NeoBERT text model request and LFM2.5 Audio request are different capabilities." - }, - { - "left": "issue:36879", - "right": "issue:37015", - "accept": false, - "reason": "RF-DETR and NeoBERT target different model families and code paths." + "reason": "NeoBERT and LFM2.5 Audio are not the same underlying change." }, { "left": "issue:37015", "right": "issue:42503", "accept": false, - "reason": "NeoBERT and ModernVBERT are different model requests, even if both are BERT-like." + "reason": "NeoBERT and ModernVBERT are different models despite superficial BERT naming overlap." }, { "left": "issue:41115", "right": "issue:43837", "accept": false, - "reason": "MiniCPM3 architecture support and Qwen3-ASR are different model additions." + "reason": "MiniCPM3 architecture support and Qwen3-ASR support are different model requests." }, { "left": "issue:41115", "right": "issue:44121", "accept": false, - "reason": "MiniCPM3 and weight-sparse transformer support are separate requests, not the same bug or change." + "reason": "MiniCPM3 and circuit-sparsity/circuitgpt are unrelated additions." }, { "left": "issue:41115", "right": "issue:42503", "accept": false, - "reason": "MiniCPM3 and ModernVBERT are different models; same broad subsystem is not enough." + "reason": "MiniCPM3 and ModernVBERT target different model families." }, { "left": "issue:41115", "right": "issue:45027", "accept": false, - "reason": "MiniCPM3 and Voxtral-4B-TTS are different model/task requests." + "reason": "MiniCPM3 and Voxtral TTS support are different code paths and modalities." }, { "left": "issue:41211", "right": "issue:43671", "accept": false, - "reason": "DEIMv2 and Qwen3-TTS are unrelated requests for different modalities." + "reason": "DEIMv2 and Qwen3-TTS are distinct model integrations." }, { "left": "issue:42503", "right": "issue:43837", "accept": false, - "reason": "ModernVBERT and Qwen3-ASR are different model families and tasks." + "reason": "ModernVBERT and Qwen3-ASR are different model additions." }, { "left": "issue:42503", "right": "issue:43671", "accept": false, - "reason": "ModernVBERT and Qwen3-TTS are different model additions." + "reason": "ModernVBERT and Qwen3-TTS are unrelated model requests." }, { "left": "issue:42503", "right": "issue:42738", "accept": false, - "reason": "ModernVBERT and generic BERT-like RoPE support are related only at a very high level, not as the same issue." + "reason": "ModernVBERT is a specific model request; the RoPE issue is a broad architectural topic, not the same change." }, { "left": "issue:42738", "right": "issue:45027", "accept": false, - "reason": "BERT-like RoPE support and Voxtral TTS support are different capabilities and code paths." + "reason": "Broad BERT-like/RoPE architecture support is unrelated to Voxtral TTS support." }, { "left": "issue:43671", "right": "issue:43837", "accept": false, - "reason": "Qwen3-TTS and Qwen3-ASR are different speech tasks and would need separate support." + "reason": "Qwen3-TTS and Qwen3-ASR are separate capabilities and likely separate model classes." }, { "left": "issue:43671", "right": "issue:45027", "accept": false, - "reason": "Qwen3-TTS and Voxtral TTS are different model requests, not a single duplicate change." + "reason": "Qwen3-TTS and Voxtral-4B-TTS are different TTS model requests, not the same underlying fix." }, { "left": "issue:43671", "right": "issue:45177", "accept": false, - "reason": "Qwen3-TTS and DeepSeek-OCR2 are different modalities and implementations." + "reason": "Qwen3-TTS support and DeepSeek-OCR2 are unrelated model integrations." }, { "left": "issue:43671", "right": "issue:44121", "accept": false, - "reason": "Qwen3-TTS and weight-sparse transformer support are unrelated model requests." + "reason": "TTS support and weight-sparse transformer support are different changes." }, { "left": "issue:43671", "right": "issue:43909", "accept": false, - "reason": "Qwen3-TTS and LFM2.5 Audio are different audio model requests, not the same concrete change." + "reason": "Qwen3-TTS and LFM2.5 Audio are both audio-related but not the same model or implementation path." }, { "left": "issue:43837", "right": "issue:43909", "accept": false, - "reason": "Qwen3-ASR and LFM2.5 Audio are different speech-model additions." + "reason": "Qwen3-ASR and LFM2.5 Audio are different audio model requests with different support needs." }, { "left": "issue:43837", "right": "issue:45322", "accept": false, - "reason": "Qwen3-ASR and EUPE are different model families and tasks." + "reason": "Qwen3-ASR and EUPE are unrelated model families and modalities." }, { "left": "issue:43837", "right": "issue:44121", "accept": false, - "reason": "Qwen3-ASR support and OpenAI weight-sparse transformer support are distinct requests." + "reason": "ASR support and weight-sparse transformer support are not the same underlying code change." }, { "left": "issue:43909", "right": "issue:44205", "accept": false, - "reason": "LFM2.5 Audio and SAM3-LiteText are unrelated model requests." + "reason": "LFM2.5 Audio and SAM3-LiteText target different modalities and implementations." }, { "left": "issue:44121", "right": "issue:45322", "accept": false, - "reason": "Weight-sparse transformer support and EUPE are different architectures with no shared concrete bug." + "reason": "Circuit-sparsity/circuitgpt and EUPE are different research models, not one duplicate request." }, { "left": "issue:43837", "right": "issue:45027", "accept": false, - "reason": "Qwen3-ASR and Voxtral-4B-TTS are different speech-model features." + "reason": "Qwen3-ASR and Voxtral TTS are different audio models and would not be merged as one PR." }, { "left": "issue:37015", "right": "issue:45027", "accept": false, - "reason": "NeoBERT and Voxtral-4B-TTS are unrelated model requests." + "reason": "NeoBERT and Voxtral-4B-TTS are unrelated model support requests." }, { "left": "issue:43837", "right": "issue:45177", "accept": false, - "reason": "Qwen3-ASR and DeepSeek-OCR2 are different modality/model additions." + "reason": "Qwen3-ASR and DeepSeek-OCR2 are distinct model additions." }, { "left": "issue:45322", "right": "issue:46133", "accept": false, - "reason": "EUPE and TIPSv2 are different model requests, even though both are vision-language related." + "reason": "EUPE and TIPSv2 are both vision-language oriented, but they are different models and different architectural changes." } ] }, @@ -16323,7 +16494,7 @@ "issue:43717|issue:46144", "issue:46143|issue:46144" ], - "prepared_review_unit_hash": "02746b4fd27e2b7d5e029d182fbc3aac18dd9cb12cfd9aaea01d9bed50352594", + "prepared_review_unit_hash": "9a7209a1b83271e9e20d0910d09b1cd2ebfcf19d74d16194f919779c203d524f", "component_budget": { "node_count": 10, "item_count": 10, @@ -16345,90 +16516,90 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Most items are clearly unrelated despite broad soft-similarity signals. The only near-match is the pair about RoFormer, but the titles describe different failure modes and I would not merge them as the same duplicate. No PRs are present.", - "confidence": 0.86, - "canonical_issue_reason": "issue:46144 is the most concrete and code-path-specific RoFormer report, so it is the best representative issue for the only close cluster in this set.", + "summary": "This is a loose similarity set with mostly unrelated issues. The only near-adjacent pair is the RoFormer kwargs/attention-interface reports, but they still look like different defects, so I would not merge any pair as duplicates.", + "confidence": 0.67, + "canonical_issue_reason": "issue:46143 is the closest thing to a cluster center: it describes a concrete, actionable RoFormer forwarding bug. The other items are either unrelated model/config issues or adjacent but distinct RoFormer concerns.", "canonical_pr_reason": null, - "best_issue_reason": "issue:46144 is the strongest standalone issue because it identifies a specific implementation problem rather than a symptom, and it is the most actionable of the set.", + "best_issue_reason": "issue:46143 is the best representative issue overall because it is the clearest and broadest statement of the only recurring theme in the set (missing kwargs propagation).", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:37934", "right": "issue:43668", "accept": false, - "reason": "Different models and different bug types: norm behavior question vs config type-hint issue." + "reason": "Different models and different bug classes: a possible norm-definition question vs an incorrect type hint." }, { "left": "issue:42936", "right": "issue:43240", "accept": false, - "reason": "Both mention missing usage of parameters, but one is Mask2Former ignore_value and the other is kwargs propagation to a loss function; not the same bug." + "reason": "Both mention missing usage of a setting/kwargs, but one is about Mask2Former ignore_value and the other about loss-function kwargs forwarding; not the same bug." }, { "left": "issue:42936", "right": "issue:46018", "accept": false, - "reason": "Unrelated model behavior: unused ignore_value vs DeepSeek-V4 shared expert gating." + "reason": "Mask2Former ignore_value and DeepSeek-V4 shared expert gating are unrelated model behaviors." }, { "left": "issue:42936", "right": "issue:46144", "accept": false, - "reason": "Mask2Former ignore_value and RoFormer attention interface are different subsystems and different concrete problems." + "reason": "Different subsystems and different symptoms; ignore_value handling is unrelated to RoFormer attention interface usage." }, { "left": "issue:42936", "right": "issue:44031", "accept": false, - "reason": "No real overlap beyond generic bug-report language; tokenizer regex warning is unrelated." + "reason": "Mask2Former ignore_value and tokenizer regex warnings are unrelated." }, { "left": "issue:42936", "right": "issue:43813", "accept": false, - "reason": "Unused ignore_value and a typo in quantization naming are unrelated." + "reason": "Unrelated: ignore_value handling vs a typo in quantization-operation naming." }, { "left": "issue:43240", "right": "issue:46144", "accept": false, - "reason": "kwargs not passed to a loss function is not the same as RoFormer attention not using the attention interface." + "reason": "Both involve kwargs, but one is about loss calculation forwarding and the other is about RoFormer attention architecture; not the same concrete bug." }, { "left": "issue:43240", "right": "issue:43668", "accept": false, - "reason": "Generic kwargs issue vs incorrect type hint; different fix and code path." + "reason": "Loss kwargs forwarding is unrelated to a ModernBERT config type-hint issue." }, { "left": "issue:43668", "right": "issue:44031", "accept": false, - "reason": "Config type hint bug and tokenizer regex warning are unrelated." + "reason": "Incorrect type hint in ModernBERTConfig is unrelated to tokenizer regex warnings." }, { "left": "issue:43668", "right": "issue:43813", "accept": false, - "reason": "Incorrect type hint and a typo in a quantization attribute name are different issues." + "reason": "Different modules and bug types: type-hint error vs typo in a conversion attribute name." }, { "left": "issue:42936", "right": "issue:43668", "accept": false, - "reason": "Different model areas and different bug classes; no shared concrete failure." + "reason": "No shared underlying defect; these are distinct correctness/documentation issues." }, { "left": "issue:43717", "right": "issue:46144", "accept": false, - "reason": "Mamba init_weights handling is unrelated to RoFormer attention interface behavior." + "reason": "Mamba init_weights usage and RoFormer attention-interface usage are unrelated architecture issues." }, { "left": "issue:46143", "right": "issue:46144", "accept": false, - "reason": "Related RoFormer area, but one is kwargs forwarding and the other is attention-interface usage; close but not the same underlying bug." + "reason": "Both are RoFormer-related, but one is about missing kwargs propagation through methods and the other about not using the attention interface; overlapping area, different underlying defect." } ] }, @@ -16458,12 +16629,12 @@ "soft_pairs": [ "issue:41084|issue:43037" ], - "prepared_review_unit_hash": "19f76cab2a660a37780aaacc75b92e361b7868117e40f382a2bf0e015b6efbd4", + "prepared_review_unit_hash": "6c8cfeb31812421cbf4ba4606606b49b7c59067aabbc1c1c0fe2924892d780cb", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1479, + "serialized_chars": 1480, "estimated_input_tokens": 370, "estimated_eval_tokens": 996 }, @@ -16471,7 +16642,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1479, + "serialized_chars": 1480, "estimated_input_tokens": 370, "estimated_eval_tokens": 996 }, @@ -16480,18 +16651,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two issues are not duplicates: one is about Set block decoding, the other about DeepSeek v3.2 support, with no shared targets or code-path evidence.", + "summary": "The two issues do not look like duplicates: one is about block decoding, the other is a request for DeepSeek v3.2 support. They appear unrelated, so the soft edge should be rejected.", "confidence": 0.97, "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "No single canonical issue fits this pair; they address different product areas and different requested changes.", + "best_issue_reason": "Issue 43037 is the better standalone representative because it is a specific feature request; issue 41084 is terse and appears to target a different topic entirely.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:41084", "right": "issue:43037", "accept": false, - "reason": "Different underlying requests: decoding behavior vs model support. The similarity is only superficial in that both are feature asks." + "reason": "Different underlying topics: block decoding vs model/version support. No shared issue target or code-path indication." } ] }, @@ -16521,12 +16692,12 @@ "soft_pairs": [ "issue:43598|issue:43600" ], - "prepared_review_unit_hash": "dd1a5beb1043470013e6c03bcc827634b89db7495f736f59ff667f8041208ff6", + "prepared_review_unit_hash": "fd979d2cf4c4cfbbe62b4f3504812e3ac3067435ea07c930ac5c51e8cb4fc855", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1584, + "serialized_chars": 1583, "estimated_input_tokens": 396, "estimated_eval_tokens": 1048 }, @@ -16534,7 +16705,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1584, + "serialized_chars": 1583, "estimated_input_tokens": 396, "estimated_eval_tokens": 1048 }, @@ -16543,18 +16714,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two issues are related only at a broad Trainer/custom-trainer theme, but they describe different behaviors in different code paths: `remove_unused_column` customization vs FLOPS estimation for custom trainers. I would not treat them as duplicates.", - "confidence": 0.93, - "canonical_issue_reason": "Issue #43598 is the better canonical issue because it is the more general request about making Trainer behavior customizable, while the other issue is a narrower estimation problem.", + "summary": "The two issues are only loosely related through custom trainers, but they describe different problems in different Trainer code paths: column pruning customizability vs FLOPS estimation.", + "confidence": 0.95, + "canonical_issue_reason": "issue #43598 is the broader, more central Trainer-behavior concern; it reads like the more general customization problem, while #43600 is a narrower FLOPS-estimation follow-up.", "canonical_pr_reason": null, - "best_issue_reason": "Issue #43598 is the best representative of the cluster since it frames the broader Trainer customizability concern that could subsume related follow-ups.", + "best_issue_reason": "#43598 is the better representative issue because it captures the broader API/customization theme and is more likely to subsume adjacent Trainer customizability concerns.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43598", "right": "issue:43600", "accept": false, - "reason": "Same broad area, but different underlying problems and code paths; one is about unused-column removal customization, the other about FLOPS estimation for custom trainers." + "reason": "Reject: they share the idea of 'custom trainers' but not the same underlying bug. #43598 is about `remove_unused_column` behavior; #43600 is about FLOPS estimation logic for custom trainers. Different code paths and fixes." } ] }, @@ -16584,7 +16755,7 @@ "soft_pairs": [ "issue:43684|issue:44405" ], - "prepared_review_unit_hash": "e990719cf0b3a1a2ce4dfa139120ed60ba7cef935bc4e77e41ab368ddd74830b", + "prepared_review_unit_hash": "ea3b9f4900ebe61695c95359754f127d983f84d7ffbd26a000daa25f88de8276", "component_budget": { "node_count": 2, "item_count": 2, @@ -16606,18 +16777,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are related only at a broad API-registration level, but they request different model families and different auto-class tasks, so they should not be merged as duplicates.", - "confidence": 0.94, - "canonical_issue_reason": "Issue 43684 is the broader, more general API-registration request and is the better representative if grouping by theme.", + "summary": "These are distinct feature requests: one asks for Qwen3-Omni auto-class registration, while the other asks for AutoModelForSequenceClassification support for Qwen3.5. They target different model families and different auto classes, so they are not duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Issue 44405 is the more specific user-facing gap, but it is a different task/model combination and not a better duplicate-canonical for 43684.", + "best_issue_reason": "issue:43684 is the broader and cleaner representative of the cluster, but it is still a separate request from the Qwen3.5 sequence-classification support issue.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43684", "right": "issue:44405", "accept": false, - "reason": "Different underlying requests: Qwen3-Omni AutoModel/AutoModelForConditionalGeneration registration vs Qwen3.5 AutoModelForSequenceClassification support. Same broad subsystem, but not the same code-path fix." + "reason": "Different underlying changes: Qwen3-Omni registration for AutoModel/AutoModelForConditionalGeneration versus Qwen3.5 sequence-classification support. They do not describe the same bug or fix." } ] }, @@ -16654,77 +16825,80 @@ "issue:43845|issue:43859", "issue:43846|issue:43859" ], - "prepared_review_unit_hash": "9e5466be22e367883007fb46bfd4763f6bdc134ae55aa9b550ee4948bcb2e1a0", + "prepared_review_unit_hash": "2ad89ded5360b1c2d8d0eb735b7e66209ea9bf17d2dcf8c6d4e0fca5e65e2697", "component_budget": { "node_count": 4, "item_count": 4, "soft_pair_count": 6, - "serialized_chars": 4263, - "estimated_input_tokens": 1066, - "estimated_eval_tokens": 2388 + "serialized_chars": 4267, + "estimated_input_tokens": 1067, + "estimated_eval_tokens": 2390 }, "budget": { "node_count": 4, "item_count": 4, "soft_pair_count": 6, - "serialized_chars": 4263, - "estimated_input_tokens": 1066, - "estimated_eval_tokens": 2388 + "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": "The cluster is too sparse to confidently deduplicate beyond noting that the three \"huggingface\" issues are likely related by title only; \"Fett\" appears unrelated. No PRs are present.", - "confidence": 0.39, - "canonical_issue_reason": "Issue 43845 is the earliest of the three near-identical \"huggingface\" issues and is the most reasonable representative of that weak cluster.", + "summary": "Issues 43845, 43846, and 43859 share the exact same title text and have similar low-activity metadata, so they form the strongest duplicate cluster visible in the packet. Issue 43835 has a different title and weaker similarity signals, so it does not match the cluster as well.", + "confidence": 0.92, + "canonical_issue_reason": "43845 is the earliest of the three identical-title issues and is the cleanest representative based on title match, similar body length/activity, and timestamps.", "canonical_pr_reason": null, - "best_issue_reason": "43845 is the best available issue candidate because it is earliest and shares the same title as the other plausible duplicates, though the evidence is very thin.", + "best_issue_reason": "43845 is the best global issue candidate because it anchors the only clearly repeated title group and predates the other matching issues.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43835", "right": "issue:43845", "accept": false, - "reason": "Unrelated titles; \"Fett\" has no clear connection to the \"huggingface\" issues." + "reason": "Different titles and no stronger shared signals beyond soft similarity." }, { "left": "issue:43835", "right": "issue:43846", "accept": false, - "reason": "Unrelated titles; no evidence these describe the same bug or request." + "reason": "Different titles and no shared targets, filenames, or other concrete match signals." }, { "left": "issue:43835", "right": "issue:43859", "accept": false, - "reason": "Unrelated titles; no concrete shared problem is shown." + "reason": "Different titles; the visible metadata does not show a same-issue match." }, { "left": "issue:43845", "right": "issue:43846", - "accept": false, - "reason": "Same vague title, but there is no description or explicit evidence proving the same underlying issue." + "accept": true, + "reason": "Exact same title text with close body length and activity/timestamp patterns; this is a plausible duplicate pair." }, { "left": "issue:43845", "right": "issue:43859", - "accept": false, - "reason": "Same vague title, but the artifacts are too sparse to confirm they are the same underlying bug." + "accept": true, + "reason": "Exact same title text and similar lightweight metadata; they look like the same report cluster." }, { "left": "issue:43846", "right": "issue:43859", - "accept": false, - "reason": "Same vague title, but no concrete issue details are available to justify merging them." + "accept": true, + "reason": "Exact same title text with similar metadata signals, supporting a duplicate relationship." } ] }, - "evaluator_result": null, - "evaluator_used": false, - "retried": false, - "accepted_nontrivial_soft_edge": false, + "evaluator_result": { + "accept": false, + "feedback": "The cluster call is a bit too strong for the packet. The data only shows identical titles for 43845/43846/43859 plus soft-similarity scores; it does not support calling them \"near-identical placeholders\" or asserting any shared underlying content beyond title similarity. Likewise, 43835 can be treated as a weaker/non-matching candidate because its title differs, but \"likely unrelated content\" is not grounded in the packet. Please keep the wording limited to the visible signals (title similarity, body length, activity, timestamps) and avoid content claims." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -16758,88 +16932,88 @@ "issue:44572|issue:44937", "issue:44573|issue:44937" ], - "prepared_review_unit_hash": "14af7d6d5a4f6d4ccc8c7d4c0bd3c787069007f6356a39098b19307d31b5fc34", + "prepared_review_unit_hash": "701141eb8dc963e2a11d463685997aafff1f540690157089b490c12463667893", "component_budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 9, - "serialized_chars": 5791, - "estimated_input_tokens": 1448, - "estimated_eval_tokens": 3152 + "serialized_chars": 5788, + "estimated_input_tokens": 1447, + "estimated_eval_tokens": 3150 }, "budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 9, - "serialized_chars": 5791, - "estimated_input_tokens": 1448, - "estimated_eval_tokens": 3152 + "serialized_chars": 5788, + "estimated_input_tokens": 1447, + "estimated_eval_tokens": 3150 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are low-information, likely unrelated or spammy issues; none of the soft pairs show enough concrete overlap to treat as duplicates. Issue #44418 is the most usable representative because it is the only one with a non-spam, topic-bearing title.", - "confidence": 0.88, - "canonical_issue_reason": "#44418 is the clearest and most descriptive issue in the cluster; the others are generic spam-like titles, so this is the best stand-in if one issue must represent the group.", + "summary": "All items look like unrelated low-signal or spammy issues; there is no evidence they describe the same underlying bug. If one must be kept as the most plausible real issue, #44418 is the least obviously spam-like.", + "confidence": 0.9, + "canonical_issue_reason": "#44418 has the only non-spam, issue-shaped title (\u201cDocumentation Enhancement Suggestion\u201d), so it is the best candidate to anchor the cluster even though the body is still too sparse to confirm legitimacy.", "canonical_pr_reason": null, - "best_issue_reason": "#44418 is the best overall issue candidate because it at least names a document-related enhancement, while the rest provide no actionable subject matter.", + "best_issue_reason": "#44418 is the most suitable global representative because it is the only item that reads like a plausible product issue rather than obvious spam or filler text.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44371", "right": "issue:44572", "accept": false, - "reason": "Both titles are just \"\" with no body details; that is not enough to show the same underlying issue." + "reason": "Both are titled only \"\" with no body, targets, or links; identical spam labels do not indicate the same underlying issue." }, { "left": "issue:44371", "right": "issue:44573", "accept": false, - "reason": "Both are generic \"\" issues, but there is no concrete bug or shared problem description tying them together." + "reason": "Both are titled only \"\" and contain no actionable evidence of a shared bug or change." }, { "left": "issue:44371", "right": "issue:44418", "accept": false, - "reason": "One is spam-like and the other is a documentation enhancement request; they do not describe the same issue." + "reason": "One is generic spam, the other is a vague documentation request; they do not describe the same concrete issue." }, { "left": "issue:44371", "right": "issue:44937", "accept": false, - "reason": "\"\" and \"Check out \\\"Google LLC\\\"\" are unrelated generic titles with no evidence of a common underlying bug." + "reason": "\"\" and \"Check out \\\"Google LLC\\\"\" are both spam-like, but there is no sign they refer to the same underlying problem." }, { "left": "issue:44418", "right": "issue:44572", "accept": false, - "reason": "A documentation enhancement request is not the same as a spam-titled issue; no shared concrete problem is shown." + "reason": "A generic documentation enhancement and a spam placeholder are not the same issue." }, { "left": "issue:44418", "right": "issue:44573", "accept": false, - "reason": "The titles point to different things: documentation suggestion versus generic spam, so they should not be merged as duplicates." + "reason": "The titles and available metadata do not point to the same bug/change; one is a vague doc suggestion, the other is spam." }, { "left": "issue:44572", "right": "issue:44573", "accept": false, - "reason": "Identical spam-like titles alone are insufficient; there is no description or code-path evidence that they refer to the same real issue." + "reason": "Both are spam placeholders, but there is no concrete shared bug, change, or target to merge them as duplicates." }, { "left": "issue:44572", "right": "issue:44937", "accept": false, - "reason": "Both look spammy, but the titles are different and there is no concrete shared issue to unify." + "reason": "Spam placeholder vs. spammy brand mention; similar low quality is not enough to treat them as the same issue." }, { "left": "issue:44573", "right": "issue:44937", "accept": false, - "reason": "These are different generic/spam titles with no shared technical substance or explicit linkage." + "reason": "No shared code-path, target, or substantive description is present; these are just separate spam-like issues." } ] }, @@ -16869,12 +17043,12 @@ "soft_pairs": [ "issue:45145|issue:45151" ], - "prepared_review_unit_hash": "cda5b6dbc761ca783ca0b56ddcf9b0079324f9002036b878ab1a41ba5b81af6e", + "prepared_review_unit_hash": "7dc1f8a861ea4de3f0b55ef1ca251eee7fc0c8a8260a01f567af86056fa8e4dc", "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 }, @@ -16882,7 +17056,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1645, + "serialized_chars": 1646, "estimated_input_tokens": 412, "estimated_eval_tokens": 1080 }, @@ -16891,27 +17065,27 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two closed issues appear to be duplicates of the same energy-optimization proposal: identical titles, very similar body sizes, and no differentiating metadata or linked targets.", - "confidence": 0.99, - "canonical_issue_reason": "Issue #45145 is the better canonical issue because it was created first and has an inbound reference, while #45151 appears to be the later duplicate.", + "summary": "The pair looks like a likely duplicate/near-duplicate issue set: the titles are identical and the body lengths are very close, but the evidence is sparse, so confidence should stay moderate. Issue 45145 is the better canonical representative because it was created first and has the only inbound reference in the pair.", + "confidence": 0.78, + "canonical_issue_reason": "Issue 45145 is the best canonical representative because it was created first and has the only inbound reference among these two near-duplicate issues.", "canonical_pr_reason": null, - "best_issue_reason": "Issue #45145 is the best representative issue: earliest submission, same title/topic, and slightly more central due to existing inbound reference.", + "best_issue_reason": "Issue 45145 is the strongest overall issue candidate for the same reason: earliest creation time and slightly more external reference activity.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45145", "right": "issue:45151", "accept": true, - "reason": "Same title and evidently the same proposed energy-reduction writeup; no evidence of distinct scope, so this is a duplicate pair." + "reason": "Same exact title and very similar body size make these likely duplicate/near-duplicate reports, though the available evidence is limited and does not prove identical content." } ] }, "evaluator_result": { - "accept": true, - "feedback": "Grounded and conservative overall: the packet supports same title, close body sizes, no explicit links/targets, and a soft-similarity signal, so treating them as likely duplicates is reasonable. The canonical choice is also supported by earlier creation time and the inbound reference on #45145." + "accept": false, + "feedback": "The comparison is plausible, but the verdict is a bit too confident for the evidence shown. The packet only provides identical titles, similar body lengths, and a soft_similarity signal; it does not give direct body-content evidence or explicit duplicate links. Please soften the language from \"same underlying issue\" to \"likely duplicate/near-duplicate,\" and reduce the confidence accordingly. The canonical choice is grounded in earlier creation time and one inbound reference, but \"more established in discussion/links\" is slightly overstated given the sparse activity." }, "evaluator_used": true, - "retried": false, + "retried": true, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -16989,12 +17163,12 @@ "soft_pairs": [ "issue:45672|issue:45674" ], - "prepared_review_unit_hash": "59d927e62707aa7fb902a8f9f67330581f9cfeb8cee483ecc0518d574d5bff2b", + "prepared_review_unit_hash": "b9e7be76507fc1b84bb5e45f31bca4829ea9965b0a67e1bdaeb27dc5e9f8e23d", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1667, + "serialized_chars": 1668, "estimated_input_tokens": 417, "estimated_eval_tokens": 1090 }, @@ -17002,7 +17176,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1667, + "serialized_chars": 1668, "estimated_input_tokens": 417, "estimated_eval_tokens": 1090 }, @@ -17011,18 +17185,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are two separate 4-bit inference bugs in the same general BitsAndBytes area, but they have different failure modes and repro paths: one is a Gemma4/audio-module torch.finfo() TypeError on uint8 weights, the other is an llm_int8_skip_modules regression that drops the default lm_head exclusion and triggers an AssertionError.", - "confidence": 0.95, - "canonical_issue_reason": "Issue #45674 is the cleaner representative report: it states the configuration regression more directly, has a clearer causal chain, and is broader than the model-specific Gemma4 symptom in #45672.", + "summary": "These are two separate 4-bit inference bugs: one is a Gemma4 audio-module uint8/torch.finfo TypeError, the other is a BitsAndBytesConfig lm_head exclusion regression causing an AssertionError. They are related only at a broad quantization-inference level, not the same underlying defect.", + "confidence": 0.97, + "canonical_issue_reason": "issue #45674 is the better representative of the cluster because it is framed as a concrete configuration/regression bug in BitsAndBytesConfig with a clear failure mode and broader applicability than the Gemma4-specific report.", "canonical_pr_reason": null, - "best_issue_reason": "#45674 is the best issue to anchor triage because it describes a concrete, reproducible BitsAndBytesConfig bug with a distinct code-path and likely wider applicability.", + "best_issue_reason": "issue #45674 is the strongest standalone issue to keep because it describes a specific, actionable regression with an explicit API trigger and a clear assertion failure path.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45672", "right": "issue:45674", "accept": false, - "reason": "Reject: same subsystem and both affect 4-bit inference, but they are different bugs. #45672 is a torch.finfo() TypeError on uint8 audio-module weights for Gemma4; #45674 is an AssertionError caused by llm_int8_skip_modules clearing the default lm_head exclusion." + "reason": "Different underlying problems: Gemma4 audio modules hit a torch.finfo TypeError on uint8 weights, while the BitsAndBytesConfig issue is about llm_int8_skip_modules clearing the default lm_head exclusion and triggering an AssertionError. Same broad area, but not the same bug." } ] }, @@ -17052,7 +17226,7 @@ "soft_pairs": [ "issue:45762|issue:45763" ], - "prepared_review_unit_hash": "a78e14f2779e230273c91076eb9ece2057b5929b4f2865a3510f262dd6095f4b", + "prepared_review_unit_hash": "630b139043b751d835d2c24ebb7e29ca8e6af3ecd21166c2503654b065d68c82", "component_budget": { "node_count": 2, "item_count": 2, @@ -17074,18 +17248,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two closed issue titles that look like unrelated personal-name/date entries, with no body, links, or technical context. The soft similarity signal is too weak to treat them as duplicates.", - "confidence": 0.97, - "canonical_issue_reason": "Issue 45762 is slightly cleaner/earlier and can serve as the cluster representative, but there is no meaningful technical signal in either issue.", + "summary": "The two closed issues are not duplicates: their titles are different personal-name/date entries with no shared problem statement, code path, or references. The soft similarity appears superficial only.", + "confidence": 0.98, + "canonical_issue_reason": "No canonical issue here; the items look unrelated rather than like two reports of the same underlying bug/request.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45762 is the best available representative only by chronology and formatting; neither issue appears to describe an actual bug or change.", + "best_issue_reason": "Neither issue is a good global representative for a duplicate cluster because there is no real cluster to collapse.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45762", "right": "issue:45763", "accept": false, - "reason": "Different named individuals and dates; no shared problem description or evidence of the same underlying issue." + "reason": "Different names and dates, no body, links, or shared context indicating the same underlying issue." } ] }, @@ -17115,30 +17289,30 @@ "soft_pairs": [ "issue:45764|issue:45765" ], - "prepared_review_unit_hash": "d8e7d1b0bd7e42a5a7fafe4f5ad2e6aeee3a49a608f24fa6a963ec7b33b5ac52", + "prepared_review_unit_hash": "90279dcfb8e03d81bff1fb84dcac3dbf0c35aafc2c63ab347f1377d072d3d6b6", "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 are just distinct name/date entries with no body or shared technical context, so they do not look like duplicates.", - "confidence": 0.96, + "summary": "The two issues look like unrelated low-signal entries (personal names plus dates) with no shared technical context, so they should not be clustered as duplicates.", + "confidence": 0.98, "canonical_issue_reason": null, "canonical_pr_reason": null, "best_issue_reason": null, @@ -17148,7 +17322,7 @@ "left": "issue:45764", "right": "issue:45765", "accept": false, - "reason": "Different personal-name/date titles, no bodies, links, or shared technical context; no sign of the same underlying issue." + "reason": "Titles do not describe the same bug or request; they are distinct name/date strings with no body, links, or code context indicating a shared underlying issue." } ] }, @@ -17188,7 +17362,7 @@ "issue:45871|issue:45873", "issue:45872|issue:45873" ], - "prepared_review_unit_hash": "ada7cbc42300b03d58a9df249d56f7f120886cabccaf317df934721d9b61ca86", + "prepared_review_unit_hash": "3a29482d2f409117b5b66dca55d74aaef52b17948db4a666e59a4dfb5bfb7ced", "component_budget": { "node_count": 5, "item_count": 5, @@ -17210,66 +17384,66 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The cluster is not one single duplicate set. It contains two exact-duplicate issue pairs: 45870/45871 and 45872/45873. Issue 45869 is only thematically related and should remain separate.", - "confidence": 0.95, - "canonical_issue_reason": "45870 is the best canonical representative for the main duplicate pair because it is the earliest exact duplicate of 45871 and has the clearest, most general title; 45872/45873 form a separate duplicate pair, and 45869 is not the same request.", + "summary": "There are two exact duplicate issue pairs in the cluster: 45870/45871 and 45872/45873. Issue 45869 is related only at a broad financial-data level, but it is a distinct request. The two DataForge issues describe different underlying asks, so they should not be merged together.", + "confidence": 0.84, + "canonical_issue_reason": "45872 is the strongest representative of the cluster: it is the earlier copy of an exact duplicate pair (45872/45873) and has the most descriptive body/title among the clearly duplicated issues.", "canonical_pr_reason": null, - "best_issue_reason": "45870 is the best overall issue to keep as the representative cluster item: it is an exact duplicate target, appears earliest, and is the broadest phrasing of the shared DataForge financial prediction example.", + "best_issue_reason": "45872 is the best global representative because it is one of the exact-duplicate heads and appears more fully specified than the other duplicate pair.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45869", "right": "issue:45870", "accept": false, - "reason": "Dataset bundle offer vs. an example request; only loose financial-domain overlap, not the same issue." + "reason": "Related finance/DataForge theme, but the dataset bundle request and the prediction example are different underlying issues." }, { "left": "issue:45869", "right": "issue:45871", "accept": false, - "reason": "Same as above: unrelated dataset promotion vs. example issue." + "reason": "Broad topical overlap only; this is a dataset bundle request, not the same change/request as the prediction example." }, { "left": "issue:45870", "right": "issue:45871", "accept": true, - "reason": "Exact duplicate title and matching body size; same DataForge financial time-series example request." + "reason": "Same title and apparently the same issue text; this looks like an exact duplicate pair." }, { "left": "issue:45870", "right": "issue:45872", "accept": false, - "reason": "Related topic, but 45872 adds a scraping service and is a distinct request rather than the same underlying issue." + "reason": "Both mention financial time series in DataForge, but one is an example/prediction request and the other is a scraping-service request." }, { "left": "issue:45870", "right": "issue:45873", "accept": false, - "reason": "Same as 45870/45872: not the same concrete request." + "reason": "Same broad domain, but the concrete request differs from the prediction-example issue, so they are not the same underlying issue." }, { "left": "issue:45871", "right": "issue:45872", "accept": false, - "reason": "Different request scope; one is a prediction example, the other includes scraping service." + "reason": "The titles point to different concrete asks: prediction example versus scraping service." }, { "left": "issue:45871", "right": "issue:45873", "accept": false, - "reason": "Different request scope; not the same underlying issue." + "reason": "Related only at a high level; these are not the same underlying issue." }, { "left": "issue:45872", "right": "issue:45873", "accept": true, - "reason": "Exact duplicate title and matching body size; same DataForge financial time-series plus scraping service request." + "reason": "Same title and matching issue shape; this appears to be an exact duplicate pair." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet: 45870/45871 and 45872/45873 are exact title/body-length duplicates, while 45869 is only loosely related by theme. The soft-edge rejections are conservative, and the representative choice is supported by the evidence provided." + "feedback": "The cluster summary is grounded in the packet: 45870/45871 and 45872/45873 are exact duplicate pairs by identical title/body length and perfect Jaccard, while 45869 is only loosely related and distinct. The representative choice is conservative and supported by the longer body on 45872. No PR mergeability issue applies here because this is an issue-only cluster." }, "evaluator_used": true, "retried": false, @@ -17329,172 +17503,178 @@ "pull_request:43823|pull_request:45213", "pull_request:43251|pull_request:43254" ], - "prepared_review_unit_hash": "63b3f800ba1aafaa4723ff006644e4749e5e805c416b696b57d00b2a93cb2ba9", + "prepared_review_unit_hash": "b4a06a54dfcb41d8099445a4a68063d95fd60a1ec466262d7219639c15ad8285", "component_budget": { "node_count": 14, "item_count": 14, "soft_pair_count": 22, - "serialized_chars": 38374, - "estimated_input_tokens": 9594, - "estimated_eval_tokens": 19444 + "serialized_chars": 38370, + "estimated_input_tokens": 9593, + "estimated_eval_tokens": 19442 }, "budget": { "node_count": 14, "item_count": 14, "soft_pair_count": 22, - "serialized_chars": 38374, - "estimated_input_tokens": 9594, - "estimated_eval_tokens": 19444 + "serialized_chars": 38370, + "estimated_input_tokens": 9593, + "estimated_eval_tokens": 19442 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Mostly unrelated PRs grouped by loose similarity; the only true duplicate-like pair is the two fixed_cross_entropy patches for the same cross-entropy kwargs change.", - "confidence": 0.93, + "summary": "Mostly heterogeneous PRs across unrelated models, docs, and refactors. The only clear duplicate/overlap is the pair of loss_utils.py changes for fixed_cross_entropy kwargs; everything else is unrelated and should stay separate.", + "confidence": 0.86, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #43251 is the better canonical among the only real duplicate pair: it directly targets issue #43240 and cleanly forwards the new cross-entropy kwargs in the shared code path.", + "canonical_pr_reason": "PR 43251 is the clearest representative of the fixed_cross_entropy kwargs fix: it has the explicit issue target, broader review activity, and the same code-path change as 43254.", "best_issue_reason": null, - "best_pr_reason": "PR #43251 is the best representative of the cluster\u2019s only genuine overlap; the rest are unrelated feature/fix PRs and should not be merged into a single duplicate set.", + "best_pr_reason": "PR 43251 is the strongest standalone PR in the cluster because it directly addresses a concrete bug report with explicit tracking, while the rest are mostly unrelated feature additions or refactors.", "soft_edge_verdicts": [ { "left": "pull_request:39895", "right": "pull_request:43424", "accept": false, - "reason": "Unrelated changes: VideoPrism model/docs vs ExecuTorch dynamic-shape export test." + "reason": "Different features in unrelated areas: VideoPrism model addition vs Executorch dynamic-shape test." }, { "left": "pull_request:41855", "right": "pull_request:43823", "accept": false, - "reason": "Unrelated: Mistral tokenizer methods vs adding MobileLLM model implementation." + "reason": "Mistral tokenizer method additions and MobileLLM model addition are unrelated changes." }, { "left": "pull_request:41855", "right": "pull_request:43424", "accept": false, - "reason": "Unrelated tokenizer API fix vs ExecuTorch export test." + "reason": "Tokenizer API patch vs Executorch export test; no shared bug or code path." }, { "left": "pull_request:42668", "right": "pull_request:43823", "accept": false, - "reason": "Unrelated processor-pretrained robustness work vs MobileLLM model addition." + "reason": "Processor robustness changes for existing models vs adding MobileLLM; not the same underlying fix." }, { "left": "pull_request:43251", "right": "pull_request:46067", "accept": false, - "reason": "Different code paths: cross-entropy helper kwargs vs model conversion loading/base_model_prefix handling." + "reason": "fixed_cross_entropy kwargs bug fix is unrelated to base_model_prefix conversion-loading changes." }, { "left": "pull_request:43251", "right": "pull_request:43823", "accept": false, - "reason": "Cross-entropy loss helper change is unrelated to MobileLLM model addition." + "reason": "Loss-function fix and new model implementation are unrelated." }, { "left": "pull_request:43251", "right": "pull_request:44827", "accept": false, - "reason": "Loss utility kwargs fix is unrelated to Mistral4 test/model changes." + "reason": "Loss-function kwargs support and Mistral4 test fixes do not share the same underlying bug." + }, + { + "left": "pull_request:43251", + "right": "pull_request:46048", + "accept": false, + "reason": "Cross-entropy kwargs fix and TDT loss kernel integration are different changes." }, { "left": "pull_request:43424", "right": "pull_request:46067", "accept": false, - "reason": "ExecuTorch export tests are unrelated to conversion/base_model_prefix loading fixes." + "reason": "Executorch dynamic-shape testing is unrelated to conversion mapping/base_model_prefix handling." }, { "left": "pull_request:43424", "right": "pull_request:44827", "accept": false, - "reason": "ExecuTorch export test vs Mistral4 test/fix work; not the same bug." + "reason": "Executorch export tests and Mistral4 test fixes are unrelated." }, { "left": "pull_request:43424", "right": "pull_request:43823", "accept": false, - "reason": "Different models and purposes: export test vs new model implementation." + "reason": "Exportability test additions and MobileLLM model addition are unrelated." }, { "left": "pull_request:43823", "right": "pull_request:45655", "accept": false, - "reason": "MobileLLM addition is unrelated to config resolution ordering in modeling_utils." + "reason": "New MobileLLM model code and config-resolution ordering fix address different problems." }, { "left": "pull_request:43823", "right": "pull_request:43995", "accept": false, - "reason": "New model implementation vs Falcon output-interface refactor; not the same concrete issue." + "reason": "Adding MobileLLM and refactoring Falcon output collection are unrelated model changes." }, { "left": "pull_request:43823", "right": "pull_request:46067", "accept": false, - "reason": "MobileLLM model addition is unrelated to loading/conversion prefix handling." + "reason": "New model addition and conversion scope/base_model_prefix fix are unrelated." }, { "left": "pull_request:43995", "right": "pull_request:44827", "accept": false, - "reason": "Falcon output-capture refactor and Mistral4 test fixes are different model-specific changes." + "reason": "Falcon output-interface refactor and Mistral4 test fixes are different code paths." }, { "left": "pull_request:43995", "right": "pull_request:45213", "accept": false, - "reason": "Falcon model refactor is unrelated to the model-creation skill PR." + "reason": "Falcon refactor and model-creation skill/docs PR are unrelated." }, { "left": "pull_request:44827", "right": "pull_request:46067", "accept": false, - "reason": "Mistral4 test fixes are unrelated to conversion loading/base_model_prefix logic." + "reason": "Mistral4 test fixes and conversion scoping/base_model_prefix handling are unrelated." }, { "left": "pull_request:44827", "right": "pull_request:45170", "accept": false, - "reason": "Mistral4 test fixes vs layernorm renaming/conversion mapping; different underlying issues." + "reason": "Mistral4 test fixes and layrnorm-to-layernorm rename are different fixes." }, { "left": "pull_request:44827", "right": "pull_request:46048", "accept": false, - "reason": "Mistral4 test fixes are unrelated to adding the TDT loss kernel." + "reason": "Mistral4 tests and TDT loss kernel integration do not target the same issue." }, { "left": "pull_request:44827", "right": "pull_request:45655", "accept": false, - "reason": "Mistral4 test changes do not match config-resolution ordering fixes." + "reason": "Mistral4 test changes and cls.config resolution ordering are unrelated." }, { "left": "pull_request:44827", "right": "pull_request:45213", "accept": false, - "reason": "Mistral4 test fixes are unrelated to the model-creation skill PR." + "reason": "Mistral4 test fixes and model-creation skill/docs are unrelated." }, { "left": "pull_request:43823", "right": "pull_request:45213", "accept": false, - "reason": "MobileLLM model addition and the model-creation skill are not the same concrete change." + "reason": "MobileLLM model addition and model-creation skill/docs are not the same change." }, { "left": "pull_request:43251", "right": "pull_request:43254", "accept": true, - "reason": "Both PRs patch the same helper, fixed_cross_entropy, to forward new nn.functional.cross_entropy kwargs; they address the same underlying bug/change and could be merged together." + "reason": "Both PRs change the same loss_utils.py code path to add cross_entropy kwargs support, and they look like overlapping implementations of the same fix." } ] }, "evaluator_result": { "accept": true, - "feedback": "The evaluation is grounded in the packet and stays conservative. The only mergeable same-bug pair is PR #43251 / #43254: both modify src/transformers/loss/loss_utils.py, both target issue #43240, and both forward new cross-entropy kwargs. The other soft-edge rejections are reasonable based on the provided diffs and titles, without overstating unrelatedness." + "feedback": "The summary is grounded in the packet and stays conservative: the cluster is mostly unrelated model additions, docs, and refactors, and the only clear overlap is the 43251/43254 fixed_cross_entropy pair, which share the same file, function, and issue target. The mergeability judgment for that pair is reasonable. No evident overstatement." }, "evaluator_used": true, "retried": false, @@ -17521,12 +17701,12 @@ "soft_pairs": [ "pull_request:40546|pull_request:42781" ], - "prepared_review_unit_hash": "0c5d348500307fdd83910d876844138a083b72b4de44fdfafcebab57e8eb7858", + "prepared_review_unit_hash": "6d9c1f19b2a27c051b36b6d8be08195a358cd53f156bd124d653f952561b0026", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 7226, + "serialized_chars": 7225, "estimated_input_tokens": 1807, "estimated_eval_tokens": 3870 }, @@ -17534,7 +17714,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 7226, + "serialized_chars": 7225, "estimated_input_tokens": 1807, "estimated_eval_tokens": 3870 }, @@ -17543,18 +17723,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both pull requests are about VibeVoice, but they are not the same change: PR 40546 implements the core VibeVoice model/docs integration, while PR 42781 adds a separate VibeVoice Realtime path plus an acoustic tokenizer and generation updates. They share model-family files, but they target different functionality and would not plausibly be merged as one duplicate PR.", - "confidence": 0.92, + "summary": "Two PRs for the same model family, but they target different deliverables: initial VibeVoice support vs. a later VibeVoice Realtime extension. Shared files mostly reflect the same subsystem and not the same concrete change.", + "confidence": 0.89, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 40546 is the better canonical PR because it is the original broad implementation of VibeVoice and covers the base model integration that the later realtime work builds on.", + "canonical_pr_reason": "PR #40546 is the more complete, non-draft baseline implementation of VibeVoice and is better suited as the representative PR for this cluster.", "best_issue_reason": null, - "best_pr_reason": "PR 40546 is the best representative artifact for the cluster because it is the main VibeVoice implementation; PR 42781 is a distinct extension for realtime support rather than a duplicate of the base model PR.", + "best_pr_reason": "PR #40546 is the strongest standalone representative: it introduces the core model support, has substantial review activity, and is broader/more central than the draft realtime follow-up.", "soft_edge_verdicts": [ { "left": "pull_request:40546", "right": "pull_request:42781", "accept": false, - "reason": "Reject: despite overlapping filenames, 40546 implements the core VibeVoice model, while 42781 adds VibeVoice Realtime plus acoustic tokenizer and generation changes. These are related but distinct changes, not the same concrete fix or mergeable duplicate PR." + "reason": "Reject: #40546 adds core VibeVoice support, while #42781 adds VibeVoice Realtime and related acoustic-tokenizer/generation work. They share model infrastructure but are not the same concrete fix/change and would not plausibly be merged as one PR." } ] }, @@ -17584,7 +17764,7 @@ "soft_pairs": [ "pull_request:40755|pull_request:41224" ], - "prepared_review_unit_hash": "77be279bd61915214e96a0ea0f78776ea9fa2a2bcf3d5f9670afebb815efa125", + "prepared_review_unit_hash": "5a44b4713bf3951383f74da078c4fd7113480be3b331c9c972d1ac9ea0ad9225", "component_budget": { "node_count": 2, "item_count": 2, @@ -17606,18 +17786,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "False-positive cluster: one PR adds TimesFM covariate forecasting, the other adds DINOv3 image-classification support. They touch unrelated models and code paths.", - "confidence": 0.99, + "summary": "The two pull requests are on unrelated model families and implement different features, so they should not be clustered as duplicates.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: these changes target different models, filenames, and code paths, so they are not duplicates of the same change.", + "canonical_pr_reason": "No canonical PR: one adds TimesFM covariate forecasting, the other adds DINOv3 image classification support; they do not fix the same underlying problem.", "best_issue_reason": null, - "best_pr_reason": "No best PR for deduplication: neither PR is a suitable representative of the other because they solve unrelated tasks.", + "best_pr_reason": "PR #41224 is the cleaner representative of its own change because it is a focused, self-contained model support addition with clear model/code/test updates.", "soft_edge_verdicts": [ { "left": "pull_request:40755", "right": "pull_request:41224", "accept": false, - "reason": "Reject: TimesFM forecasting with covariates and DINOv3 image-classification support are unrelated features with no shared target code path." + "reason": "Different model domains and different code paths: TimesFM forecasting with covariates versus DINOv3 image classification support. They are not the same underlying bug or change and would not be merged into one PR." } ] }, @@ -17712,12 +17892,12 @@ "pull_request:41797|pull_request:45186", "pull_request:45186|pull_request:45630" ], - "prepared_review_unit_hash": "c92b3d2c659d022ab8be79301e31565e1826a6e3f508e1c8ea566e59d1576fdd", + "prepared_review_unit_hash": "b6c7cec47605b04117cdc75d5581e0ed27ff81dcaf43c9c8d35076d2e69a1df9", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 2, - "serialized_chars": 10433, + "serialized_chars": 10436, "estimated_input_tokens": 2609, "estimated_eval_tokens": 5474 }, @@ -17725,7 +17905,7 @@ "node_count": 3, "item_count": 3, "soft_pair_count": 2, - "serialized_chars": 10433, + "serialized_chars": 10436, "estimated_input_tokens": 2609, "estimated_eval_tokens": 5474 }, @@ -17734,24 +17914,24 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are three unrelated model-addition PRs (DeepSeek OCR, Isaac, Kimi2-6) that share transformer scaffolding but do not fix the same code-path problem. They should not be deduplicated or merged as one change.", - "confidence": 0.98, + "summary": "These are three separate model-integration PRs (DeepSeek OCR, Isaac, and Kimi2-6). They share Transformers boilerplate and auto-mapping updates, but they target different model families and code paths, so they are not duplicates.", + "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45630 is the strongest representative of the cluster by activity and scope, but it is still an independent model-addition PR rather than a duplicate of the others.", + "canonical_pr_reason": "Kimi2-6 is the broadest and most actively discussed PR in the set, touching the most integration surfaces and likely serving best as the representative item for this cluster.", "best_issue_reason": null, - "best_pr_reason": "PR 45630 has the broadest and most reviewed integration work in this set, making it the best single representative if one must be chosen.", + "best_pr_reason": "Kimi2-6 is the strongest representative PR overall because it has the highest discussion activity and the widest integration footprint; the other two are narrower, separate model additions.", "soft_edge_verdicts": [ { "left": "pull_request:41797", "right": "pull_request:45186", "accept": false, - "reason": "Different model integrations (DeepSeek OCR vs Isaac). Shared docs/auto-registration files are just standard scaffold updates, not evidence of the same underlying fix." + "reason": "DeepSeek OCR and Isaac are different model additions with different filenames, configs, and model code; the overlap is only shared Transformers scaffolding and docs, not the same underlying change." }, { "left": "pull_request:45186", "right": "pull_request:45630", "accept": false, - "reason": "Different model integrations (Isaac vs Kimi2-6). They only overlap on generic model-registration/docs plumbing, so they are not the same change." + "reason": "Isaac and Kimi2-6 integrate different models and touch different concrete implementations; shared auto-mapping and docs updates are routine for new-model PRs, not evidence of a duplicate fix." } ] }, @@ -17781,7 +17961,7 @@ "soft_pairs": [ "pull_request:42975|pull_request:42976" ], - "prepared_review_unit_hash": "790d50588d3fed99ec733c840c8449a83b62a2910cedac326b8be9f521798340", + "prepared_review_unit_hash": "0bbf1410d99b72a178a35dbc74ca234e253f82ebc8b8d4b6ce6e6dc1223ce5f4", "component_budget": { "node_count": 2, "item_count": 2, @@ -17803,18 +17983,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs are broad GitHub Actions maintenance changes, but they upgrade different actions and fix different compatibility tracks, so they should not be merged as duplicates.", - "confidence": 0.94, + "summary": "Two separate GitHub Actions maintenance PRs: one is a broad Node 24 compatibility upgrade for several workflow actions, the other is a docker/build-push-action version bump. They overlap in workflows but do not fix the same underlying problem.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 42975 is the better representative because it is the broader, more systematic workflow-action upgrade across many files; PR 42976 is a separate Docker action bump with a different change set.", + "canonical_pr_reason": "PR 42975 is the better canonical representative: it has a clearer, broader goal (Node 24 compatibility) and touches the major GitHub Actions version updates across workflows.", "best_issue_reason": null, - "best_pr_reason": "PR 42975 is the strongest standalone candidate for the cluster because it covers the wider Node 24 compatibility effort and touches more of the affected workflow surface area.", + "best_pr_reason": "PR 42975 is the best overall representative of the cluster because its title and diff describe a single, concrete maintenance objective, while PR 42976 is a narrower, separate action-version bump.", "soft_edge_verdicts": [ { "left": "pull_request:42975", "right": "pull_request:42976", "accept": false, - "reason": "They are not the same underlying change: 42975 updates general GitHub Actions/runtime compatibility (checkout/cache/upload-artifact and other workflow actions), while 42976 specifically bumps docker/build-push-action in Docker image workflows. Shared workflow filenames alone are not enough to treat them as one PR." + "reason": "Do not merge: 42975 upgrades several GitHub Actions to Node 24-compatible versions; 42976 only bumps docker/build-push-action from v5 to v6. They are related maintenance changes but not the same underlying fix and would not naturally be one PR." } ] }, @@ -17853,7 +18033,7 @@ "pull_request:43451|pull_request:45490", "pull_request:45490|pull_request:46162" ], - "prepared_review_unit_hash": "b8fcab7199b4abe7f14d4a62734117ebec12b1f4c84ffbed82601bf243334077", + "prepared_review_unit_hash": "ad136c739266ae3b104edcfbf2daa78607ed70c2efb480ce79c5cfca94be0d06", "component_budget": { "node_count": 5, "item_count": 5, @@ -17875,54 +18055,54 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are not duplicates: they are separate model-addition PRs for different models (ViT NEPA, Molmo, Molmo2, CTSM) plus one small test-only XPU change. They share some framework/doc/auto-mapping files, but each targets a different code path and model family.", - "confidence": 0.92, + "summary": "These are separate model-addition or test-only PRs with only superficial overlap in docs and auto-mapping files. The shared filenames do not indicate the same underlying bug or change, so this does not form a duplicate cluster.", + "confidence": 0.93, "canonical_issue_reason": null, - "canonical_pr_reason": "No PR here appears to be a duplicate of another; if one must be chosen as the representative, #43451 (Molmo2) is the broadest and most complete model-addition PR, with substantial model, processor, auto-mapping, docs, and test coverage plus the highest review activity.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "#43451 is the strongest single representative of the cluster because it is a full-featured model integration, not just a doc/test tweak, and it has the most extensive implementation footprint and discussion.", + "best_pr_reason": "No PR here is a true duplicate of the others; if one must be picked as the strongest standalone representative, PR 43451 is the most substantial and complete model-addition PR, but it is still distinct from the rest.", "soft_edge_verdicts": [ { "left": "pull_request:42978", "right": "pull_request:43448", "accept": false, - "reason": "Different model families and different integrations (ViT NEPA vs Molmo); shared docs/auto files are generic framework plumbing, not the same change." + "reason": "ViT NEPA vs Molmo are different model additions; overlap is limited to shared docs/auto-registration files, not the same change." }, { "left": "pull_request:42978", "right": "pull_request:43451", "accept": false, - "reason": "ViT NEPA and Molmo2 are unrelated model additions touching different architectures and code paths." + "reason": "ViT NEPA and Molmo2 are unrelated model integrations with different codepaths and artifacts." }, { "left": "pull_request:42978", "right": "pull_request:45490", "accept": false, - "reason": "ViT NEPA and CTSM are distinct model additions; overlap is only in shared registry/documentation files." + "reason": "ViT NEPA and CTSM are distinct model additions; the common files are generic registry/docs updates only." }, { "left": "pull_request:43448", "right": "pull_request:43451", "accept": false, - "reason": "Molmo and Molmo2 are separate model releases with different model code, processors, and tests; they are not one concrete bug fix." + "reason": "Molmo and Molmo2 are related by name but are separate model families with different implementations and files." }, { "left": "pull_request:43448", "right": "pull_request:45490", "accept": false, - "reason": "Molmo and CTSM are different model integrations, not the same underlying change." + "reason": "Molmo and CTSM are unrelated model additions; shared auto/docs files are insufficient for duplicate triage." }, { "left": "pull_request:43451", "right": "pull_request:45490", "accept": false, - "reason": "Molmo2 and CTSM target different model families and implementation paths; shared auto-registry edits are incidental." + "reason": "Molmo2 and CTSM target different models and codepaths; this is not the same concrete change." }, { "left": "pull_request:45490", "right": "pull_request:46162", "accept": false, - "reason": "CTSM is a new model integration, while #46162 is a tiny test-only XPU expectation update; they do not fix the same problem and should not be merged together." + "reason": "CTSM model addition vs a small XPU test annotation tweak are entirely different changes." } ] }, @@ -17960,70 +18140,70 @@ "pull_request:43665|pull_request:46161", "pull_request:44815|pull_request:46161" ], - "prepared_review_unit_hash": "6a633d677e23ca93abfcfa5633787af07261d539712ed4ca09dcbcf2965b08ac", + "prepared_review_unit_hash": "1db9eaad1d51f9ce6fc5a5661285912de85f924c3e37f3c5ddfc7ee6a07c639f", "component_budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 6, - "serialized_chars": 11991, - "estimated_input_tokens": 2998, - "estimated_eval_tokens": 6252 + "serialized_chars": 12045, + "estimated_input_tokens": 3012, + "estimated_eval_tokens": 6280 }, "budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 6, - "serialized_chars": 11991, - "estimated_input_tokens": 2998, - "estimated_eval_tokens": 6252 + "serialized_chars": 12045, + "estimated_input_tokens": 3012, + "estimated_eval_tokens": 6280 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These PRs do not look like duplicates: they span unrelated areas (Whisper tokenization/tests, model test cleanup, FP8 dequant/loading, FSDP tied weights, and tokenizer subfolder fallback).", - "confidence": 0.92, + "summary": "These PRs do not form a duplicate set: they span unrelated areas (Whisper tokenization, CLIP/ViT tests, FP8 dequant/loading, FSDP tied weights, and processor subfolder fallback). Even the one soft-similar pair is not the same concrete fix.", + "confidence": 0.93, "canonical_issue_reason": null, - "canonical_pr_reason": "No clear duplicate PR cluster here; if one must be used as an anchor, 44815 is the strongest standalone product fix because it targets a specific FP8 dequant/loading path with concrete code changes.", + "canonical_pr_reason": "PR 46161 is the clearest standalone bug fix: it targets one concrete processor/tokenizer fallback path and explicitly links to an issue.", "best_issue_reason": null, - "best_pr_reason": "44815 is the best representative of the set because it is the most substantive and clearly defined code-path fix; the others are test-only or touch unrelated subsystems.", + "best_pr_reason": "PR 46161 is the best representative candidate because it is narrowly scoped, traceable to an explicit issue, and describes a single code-path fix.", "soft_edge_verdicts": [ { "left": "pull_request:43291", "right": "pull_request:43665", "accept": false, - "reason": "Both are test-related, but they target different models and failure modes (Whisper tokenizer behavior vs CLIP/ViT test cleanup). Not the same underlying bug/change." + "reason": "Whisper tokenizer decode logic vs unrelated CLIP/ViT test edits; different subsystems and different change types." }, { "left": "pull_request:43291", "right": "pull_request:44815", "accept": false, - "reason": "Whisper tokenization/tests and FP8 dequant/loading are unrelated code paths; no plausible single merged fix." + "reason": "Whisper decode/batch handling is unrelated to FP8 dequant/model-loading changes in a different subsystem." }, { "left": "pull_request:43665", "right": "pull_request:46102", "accept": false, - "reason": "Test cleanup assertions in model tests are unrelated to the FSDP tied-weights change." + "reason": "Test-only CLIP/ViT edits are not the same bug as FSDP tied-weights cleanup/fix." }, { "left": "pull_request:43665", "right": "pull_request:44815", "accept": false, - "reason": "A test-only PR and an FP8 kernel/loading fix do not address the same concrete problem." + "reason": "One is test scaffolding in vision models; the other is FP8/loading infrastructure. No shared underlying defect." }, { "left": "pull_request:43665", "right": "pull_request:46161", "accept": false, - "reason": "Model test edits are unrelated to tokenizer subfolder fallback logic in processing_utils." + "reason": "Placeholder test changes are unrelated to tokenizer subfolder fallback behavior." }, { "left": "pull_request:44815", "right": "pull_request:46161", "accept": false, - "reason": "FP8 dequant padding/loading changes and tokenizer fallback behavior are distinct subsystems with different bugs." + "reason": "Both touch tokenizer-adjacent code, but one is FP8/dequant plus loading internals while the other is processor/tokenizer subfolder fallback; not the same fix." } ] }, @@ -18060,12 +18240,12 @@ "pull_request:45932|pull_request:46100", "pull_request:46100|pull_request:46114" ], - "prepared_review_unit_hash": "6e78a23b3229ba4adaeab1b46484990cbe94677985c2cc2cf8785990a0a536af", + "prepared_review_unit_hash": "472b5bce35f57f33724ff8ec20035891324f8aefe14203eaca33e079d751ee46", "component_budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 5, - "serialized_chars": 11342, + "serialized_chars": 11343, "estimated_input_tokens": 2836, "estimated_eval_tokens": 5928 }, @@ -18073,7 +18253,7 @@ "node_count": 5, "item_count": 5, "soft_pair_count": 5, - "serialized_chars": 11342, + "serialized_chars": 11343, "estimated_input_tokens": 2836, "estimated_eval_tokens": 5928 }, @@ -18082,48 +18262,48 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Only one clear duplicate pair exists: the two identical ViT test-migration PRs. The other PRs touch different subsystems and/or contain unrelated CI/config/model changes, so they should not be merged together as duplicates.", - "confidence": 0.96, + "summary": "This cluster is mostly a set of unrelated \"do not merge\" test PRs; the only clear duplicate pair is the two identical ViT test-migration PRs.", + "confidence": 0.88, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46100 is the best canonical representative of the exact duplicate pair: it has the same diff as 46114, a cleaner/generic title, and no extra fork-specific framing.", + "canonical_pr_reason": "PR 46100 is the cleanest canonical representative of the exact duplicate pair: it has the simpler title and the same minimal ViT test-file diff as 46114.", "best_issue_reason": null, - "best_pr_reason": "PR 46100 is the best global PR choice because it exactly matches its duplicate partner while being the cleaner canonical instance of that change; the other PRs are unrelated experiments in setup, CI workflows, or broader model/testing changes.", + "best_pr_reason": "PR 46100 is the best overall representative because it is the clearest self-contained duplicate target; 46114 is just a forked copy of the same change.", "soft_edge_verdicts": [ { "left": "pull_request:43488", "right": "pull_request:46114", "accept": false, - "reason": "Different changes: 43488 is setup/dependency/version-bump plumbing, while 46114 only edits ViT tests for migration. Shared test filename is incidental." + "reason": "Different changes: 43488 is a setup/dependency/version-bump repo-bot check, while 46114 is a ViT test migration clone. Shared test-file presence is not enough." }, { "left": "pull_request:43488", "right": "pull_request:43532", "accept": false, - "reason": "They target different code paths and goals: package/setup changes versus CI workflow and notification/testing utilities." + "reason": "Both are test/CI experiment PRs, but they touch different code paths and files: packaging/deps versus workflow/comment handling and utilities." }, { "left": "pull_request:43532", "right": "pull_request:46114", "accept": false, - "reason": "43532 is a workflow/CI diff with multiple supporting files; 46114 is a tiny ViT test edit. Not the same underlying change." + "reason": "No same underlying bug/change; 43532 is workflow/utility plumbing, while 46114 is only a duplicated ViT test edit." }, { "left": "pull_request:45932", "right": "pull_request:46100", "accept": false, - "reason": "Despite both being test-related, 45932 is a broad Grafana/CI/model sweep across many files, while 46100 is only a focused ViT test migration. They are not the same concrete fix." + "reason": "Too broad and not the same fix: 45932 changes multiple model configs/workflows for grafana/telemetry testing, while 46100 only adds failing tests in the ViT test file." }, { "left": "pull_request:46100", "right": "pull_request:46114", "accept": true, - "reason": "These are the same PR change: identical diff in the same ViT test file, with only the title/metadata differing. They are clear duplicates." + "reason": "Exact duplicate: both modify the same ViT test file with the same inserted failing tests, and the titles indicate the same test-migration intent." } ] }, "evaluator_result": { "accept": true, - "feedback": "Grounded and conservative overall. The exact duplicate pair is 46100/46114, and the other false verdicts are justified by the packet because those PRs touch different areas or broader multi-file changes rather than a single shared bugfix. The summary is also appropriately cautious about treating only the identical ViT-test pair as a clear duplicate." + "feedback": "The summary is grounded in the packet and stays conservative. The only strong duplicate is 46100/46114: they have identical diffs to the same ViT test file. The negative pair verdicts are also appropriately cautious, since the other PRs touch different areas or broader changes despite some superficial test/CI similarity." }, "evaluator_used": true, "retried": false, @@ -18150,7 +18330,7 @@ "soft_pairs": [ "pull_request:43612|pull_request:43613" ], - "prepared_review_unit_hash": "02efe351d6725f5fc920e260da08f4d10d6c2387833727d33acb60feb1e62936", + "prepared_review_unit_hash": "8fd298253a579226270e6957ba6a0447af626e0c3597b2347835a320cd5957a3", "component_budget": { "node_count": 2, "item_count": 2, @@ -18172,18 +18352,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are two separate feature PRs for different pipelines: promptable concept segmentation vs promptable visual segmentation. They overlap in docs/pipeline scaffolding but do not fix or implement the same underlying change, so they should not be deduplicated.", - "confidence": 0.96, + "summary": "These are related but distinct PRs: one adds Promptable Concept Segmentation, the other Promptable Visual Segmentation. They share documentation and pipeline plumbing, but they target different features and code paths, so they should not be treated as duplicates.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical duplicate PR: 43612 and 43613 implement different pipelines with different model/test coverage and distinct task docs.", + "canonical_pr_reason": "PR 43613 is the broader and more feature-complete segmentation PR, with wider model/test coverage, so it is the better cluster representative if one must be chosen.", "best_issue_reason": null, - "best_pr_reason": "43613 is the stronger standalone representative if one must be picked: it has the broader visual-segmentation integration surface and more extensive model/test updates, but it is still distinct from 43612.", + "best_pr_reason": "PR 43613 is the stronger representative of the pair because it covers more integrations and tests, but it is still a separate feature from 43612.", "soft_edge_verdicts": [ { "left": "pull_request:43612", "right": "pull_request:43613", "accept": false, - "reason": "Different pipeline features and code paths: concept segmentation vs visual segmentation. Shared docs/registry plumbing is not enough to treat them as the same change." + "reason": "Reject: the titles and changed files show different features\u2014promptable concept segmentation vs promptable visual segmentation\u2014with separate pipeline modules and model/tests, so they are not the same underlying change." } ] }, @@ -18213,40 +18393,40 @@ "soft_pairs": [ "pull_request:43651|pull_request:43663" ], - "prepared_review_unit_hash": "a24e0b1e6d2d021d15e0a44b7fe559c20e95b2d3ad6f435b0bdf76304ad28c8a", + "prepared_review_unit_hash": "961f4e52d998e4215ba4738c5d46ab829144de9cf576063e9975ceb5be2750dc", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4292, - "estimated_input_tokens": 1073, - "estimated_eval_tokens": 2402 + "serialized_chars": 4293, + "estimated_input_tokens": 1074, + "estimated_eval_tokens": 2404 }, "budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4292, - "estimated_input_tokens": 1073, - "estimated_eval_tokens": 2402 + "serialized_chars": 4293, + "estimated_input_tokens": 1074, + "estimated_eval_tokens": 2404 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two open PRs touch the same file but address different trainer extension points: one changes gradient-accumulation loss scaling, the other adds an overrideable signature-column helper. They are related to Trainer customization, but not the same underlying change.", - "confidence": 0.96, + "summary": "The two PRs share a file but implement different Trainer extension points: one changes gradient-accumulation loss scaling, the other changes model signature column filtering. They are related to trainer customization, but not the same underlying bug or change.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR: PR #43651 fixes loss scaling in training_step, while PR #43663 refactors column-signature filtering for custom trainers. Different code paths and different behaviors.", + "canonical_pr_reason": "No canonical PR: 43651 and 43663 touch different Trainer behaviors and are not duplicates of the same fix or feature.", "best_issue_reason": null, - "best_pr_reason": "PR #43651 is the stronger representative of the cluster because it is a concrete behavioral fix in the training loop, whereas PR #43663 is a narrower extensibility refactor.", + "best_pr_reason": "Neither PR is a strong global duplicate representative; each addresses a separate code path in src/transformers/trainer.py.", "soft_edge_verdicts": [ { "left": "pull_request:43651", "right": "pull_request:43663", "accept": false, - "reason": "Different underlying changes: #43651 controls gradient-accumulation loss normalization, while #43663 adds a helper for inspecting model forward signatures and column filtering. Same file, but not the same bug or mergeable fix." + "reason": "Different functionality: 43651 alters loss scaling during gradient accumulation, while 43663 refactors signature column filtering for custom trainers. Shared file alone is not enough to treat them as the same change." } ] }, @@ -18276,12 +18456,12 @@ "soft_pairs": [ "pull_request:43656|pull_request:43842" ], - "prepared_review_unit_hash": "d32c6008bf71ccf8a834051cdbf182cdfd135de62492368922a0b29d74ddba0f", + "prepared_review_unit_hash": "7846a5f81fa5c5f9a90fda70b9da77d0156f4279fa8d6197a653053eeef46ff5", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 3661, + "serialized_chars": 3662, "estimated_input_tokens": 916, "estimated_eval_tokens": 2088 }, @@ -18289,7 +18469,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 3661, + "serialized_chars": 3662, "estimated_input_tokens": 916, "estimated_eval_tokens": 2088 }, @@ -18298,18 +18478,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two open PRs touch the same CLI file and both mention NameError, but they fix different typing/runtime problems in the serve CLI. They are related, not duplicates.", - "confidence": 0.91, + "summary": "Two open PRs touch the same file and both mention TypeAdapter, but they address different concrete import/type-annotation problems in the CLI serve path, so they should not be merged as duplicates.", + "confidence": 0.82, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 43842 is the tighter, more explicit fix: it directly addresses the TypeAdapter NameError when pydantic is absent and has the clearest issue-targeted title/diff. PR 43656 is a separate typing/annotation cleanup for the same file.", + "canonical_pr_reason": "PR #43842 is the more canonical representative: it has the clearer, narrower fix for the pydantic-optional import path, with a small focused annotation change and an explicit issue target.", "best_issue_reason": null, - "best_pr_reason": "PR 43842 is the best representative of the cluster because its title, target, and diff all align on one concrete failure mode in the CLI. PR 43656 is adjacent but fixes a different annotation-related problem.", + "best_pr_reason": "PR #43842 is the best overall representative because its title and diff align tightly with a single concrete bug, and it is more specific than the broader TypeAdapter/typing cleanup in #43656.", "soft_edge_verdicts": [ { "left": "pull_request:43656", "right": "pull_request:43842", "accept": false, - "reason": "Both touch src/transformers/cli/serve.py and involve typing annotations, but they fix different NameErrors: 43656 changes forward-reference handling for model/processor annotations, while 43842 fixes missing pydantic TypeAdapter import/annotation handling. They are not the same underlying bug and would not naturally merge into one PR." + "reason": "Both touch `src/transformers/cli/serve.py`, but #43656 is a broader typing/annotation cleanup for a TypeAdapter NameError, while #43842 specifically fixes the `TypeAdapter` reference when pydantic is not installed. They do not appear to fix the same underlying bug." } ] }, @@ -18502,12 +18682,12 @@ "pull_request:44044|pull_request:44154", "pull_request:44024|pull_request:44076" ], - "prepared_review_unit_hash": "c08293347c587c21e1bf5c482110dc21b57f23dbc5b0a6a281401926a15f4c63", + "prepared_review_unit_hash": "5317b2da15c4b7561dd2599265611040a8e0190bfff6cfb02a35b25d5ccb70bd", "component_budget": { "node_count": 28, "item_count": 28, "soft_pair_count": 77, - "serialized_chars": 84502, + "serialized_chars": 84501, "estimated_input_tokens": 21126, "estimated_eval_tokens": 42508 }, @@ -18515,7 +18695,7 @@ "node_count": 28, "item_count": 28, "soft_pair_count": 72, - "serialized_chars": 76934, + "serialized_chars": 76933, "estimated_input_tokens": 19234, "estimated_eval_tokens": 38724 }, @@ -18524,414 +18704,348 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is mostly a set of parallel, model-specific output-tracing refactors under the same broader initiative, not one duplicate cluster. The only clear duplicate groups are the GPT-J/CodeGen PRs and the GPT-Neo PRs; the rest touch different model code paths and should stay separate.", - "confidence": 0.93, + "summary": "This cluster is mostly a sweep of model-by-model output-tracing refactors, not one single duplicate set. The only clear duplicate subcluster is around GPT-J/CodeGen output tracing, with a smaller duplicate pair in GPT-Neo; the rest share the broad standardized-output theme but not the same concrete code path.", + "confidence": 0.82, "canonical_issue_reason": null, - "canonical_pr_reason": "44722 is the cleanest representative of the GPT-J/CodeGen duplicate subset: it overlaps heavily with 44066 and 44085 on the same code paths and looks like the most consolidated revision.", + "canonical_pr_reason": "PR 44722 is the best canonical PR: it is the clearest and most complete representative of the GPT-J/CodeGen output-tracing refactor, with the same concrete code paths and explicit issue linkage.", "best_issue_reason": null, - "best_pr_reason": "44722 is the best single PR to anchor this cluster because it is the most consolidated version of the only clearly duplicated change set, while the other PRs are mostly distinct model-by-model refactors.", + "best_pr_reason": "PR 44722 is the strongest single representative for the duplicate subcluster because it updates the same GPT-J/CodeGen tracing path and appears to subsume the earlier GPT-J variants.", "soft_edge_verdicts": [ { "left": "pull_request:43996", "right": "pull_request:44085", "accept": false, - "reason": "Different model families (CVT/FNet vs GPT-J); same initiative, but not the same concrete code path." + "reason": "Different model families and different code paths; they only share the broad output-tracing refactor theme." }, { "left": "pull_request:43996", "right": "pull_request:44044", "accept": false, - "reason": "CVT/FNet output-tracing refactor vs DeBERTa output-tracing refactor; unrelated implementations." + "reason": "FNet/CVT versus DeBERTa_v2 are unrelated model implementations, so this is not the same concrete change." }, { "left": "pull_request:44066", "right": "pull_request:44085", "accept": true, - "reason": "Both target the same GPT-J output-tracing refactor in the same file; 44085 looks like an alternate/partial revision of the same change." + "reason": "Both target the GPT-J/CodeGen output-tracing path and read like overlapping versions of the same refactor." }, { "left": "pull_request:44007", "right": "pull_request:44072", "accept": false, - "reason": "ResNet/RegNet/RT-DETR ResNet vs EfficientNet; same pattern, different code path." + "reason": "RegNet and EfficientNet are separate model paths; same theme, not the same bug or change." }, { "left": "pull_request:44072", "right": "pull_request:44722", "accept": false, - "reason": "EfficientNet is unrelated to GPT-J/CodeGen; only the refactor pattern is shared." + "reason": "EfficientNet versus GPT-J/CodeGen are unrelated implementations, so these are not duplicates." }, { "left": "pull_request:44066", "right": "pull_request:44072", "accept": false, - "reason": "GPT-J/CodeGen vs EfficientNet; different model code paths." + "reason": "Different model code paths; only the general standardized-tracing effort is shared." }, { "left": "pull_request:44013", "right": "pull_request:44044", "accept": false, - "reason": "MobileNetV2 vs DeBERTa; both standardize outputs, but they are separate model changes." + "reason": "MobileNetV2 and DeBERTa_v2 are unrelated model-specific refactors." }, { "left": "pull_request:44066", "right": "pull_request:44086", "accept": false, - "reason": "GPT-J/CodeGen refactor vs MGP-STR refactor; not the same underlying bug or change." + "reason": "GPT-J/CodeGen versus MGP-STR are different model families and different concrete changes." }, { "left": "pull_request:44066", "right": "pull_request:44071", "accept": false, - "reason": "GPT-J/CodeGen and MPT are distinct model implementations." + "reason": "These touch different model implementations, so they are not the same underlying fix." }, { "left": "pull_request:44018", "right": "pull_request:44068", "accept": true, - "reason": "Both are GPT-Neo output-tracing refactors in the same model file; they look like the same underlying change." + "reason": "Both are GPT-Neo output-tracing refactors on the same model path, with the later PR looking like a duplicate/follow-up of the same change." }, { "left": "pull_request:44066", "right": "pull_request:44068", "accept": false, - "reason": "GPT-J/CodeGen vs GPT-Neo; similar refactor style, different code paths." + "reason": "GPT-J/CodeGen and GPT-Neo are different code paths, so this is only thematic similarity." }, { "left": "pull_request:44018", "right": "pull_request:44066", "accept": false, - "reason": "GPT-Neo and GPT-J/CodeGen are different model families." + "reason": "GPT-Neo versus GPT-J/CodeGen are distinct model refactors, not one duplicated change." }, { "left": "pull_request:44019", "right": "pull_request:44722", "accept": false, - "reason": "ResNet output tracing vs GPT-J/CodeGen output tracing; different models." + "reason": "ResNet and GPT-J/CodeGen are unrelated model paths." }, { "left": "pull_request:44086", "right": "pull_request:44722", "accept": false, - "reason": "MGP-STR is unrelated to GPT-J/CodeGen." + "reason": "MGP-STR and GPT-J/CodeGen are different implementations; same theme does not make them duplicates." }, { "left": "pull_request:44019", "right": "pull_request:44085", "accept": false, - "reason": "ResNet vs GPT-J; same initiative, different concrete code path." + "reason": "ResNet versus GPT-J/CodeGen are separate model-specific changes." }, { "left": "pull_request:44018", "right": "pull_request:44019", "accept": false, - "reason": "GPT-Neo and ResNet are unrelated model implementations." + "reason": "GPT-Neo and ResNet are different model code paths." }, { "left": "pull_request:44019", "right": "pull_request:44071", "accept": false, - "reason": "ResNet vs MPT; same output-tracing theme, but not the same bug/change." + "reason": "ResNet and MPT are unrelated implementations." }, { "left": "pull_request:44068", "right": "pull_request:44722", "accept": false, - "reason": "GPT-Neo vs GPT-J/CodeGen; different model code paths." + "reason": "GPT-Neo and GPT-J/CodeGen are different model refactors, not the same concrete fix." }, { "left": "pull_request:44068", "right": "pull_request:44116", "accept": false, - "reason": "GPT-Neo and Flaubert are separate model families." + "reason": "GPT-Neo and Flaubert/XLM are distinct model paths." }, { "left": "pull_request:44071", "right": "pull_request:44722", "accept": false, - "reason": "MPT vs GPT-J/CodeGen; no shared code path." + "reason": "MPT and GPT-J/CodeGen are different concrete changes." }, { "left": "pull_request:44018", "right": "pull_request:44116", "accept": false, - "reason": "GPT-Neo vs Flaubert; different implementations." + "reason": "GPT-Neo versus Flaubert/XLM is only broad thematic overlap." }, { "left": "pull_request:44018", "right": "pull_request:44086", "accept": false, - "reason": "GPT-Neo vs MGP-STR; unrelated changes." + "reason": "GPT-Neo and MGP-STR are unrelated model implementations." }, { "left": "pull_request:44018", "right": "pull_request:44085", "accept": false, - "reason": "GPT-Neo and GPT-J are distinct model code paths." + "reason": "GPT-Neo and GPT-J/CodeGen are different code paths." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44086", + "accept": false, + "reason": "Different model families and different files touched." }, { "left": "pull_request:44068", - "right": "44086", + "right": "pull_request:44101", "accept": false, - "reason": "GPT-Neo vs MGP-STR; same refactor pattern only." + "reason": "GPT-Neo and Flaubert/XLM are separate model-specific refactors." }, { "left": "pull_request:44059", "right": "pull_request:44086", "accept": false, - "reason": "GPT2 vs MGP-STR; not the same model change." + "reason": "GPT2 and MGP-STR are unrelated concrete changes." }, { "left": "pull_request:44085", "right": "pull_request:44086", "accept": false, - "reason": "GPT-J/CodeGen vs MGP-STR; different code paths." + "reason": "GPT-J/CodeGen and MGP-STR are different model code paths." }, { "left": "pull_request:44098", "right": "pull_request:44101", "accept": false, - "reason": "ViLT vs Flaubert/XLM; separate model families." + "reason": "ViLT and Flaubert/XLM are different implementations." }, { "left": "pull_request:44059", - "right": "pull_request:44722", - "accept": false, - "reason": "GPT2 refactor and GPT-J/CodeGen refactor are different concrete changes." - }, - { - "left": "pull_request:44018", "right": "pull_request:44071", "accept": false, - "reason": "GPT-Neo and MPT are unrelated implementations." - }, - { - "left": "pull_request:44074", - "right": "pull_request:44101", - "accept": false, - "reason": "TextNet vs Flaubert/XLM; different model code." - }, - { - "left": "pull_request:44056", - "right": "pull_request:44161", - "accept": false, - "reason": "MPNet and LongT5 are unrelated models." + "reason": "GPT2 and MPT are unrelated model refactors." }, { "left": "pull_request:43997", "right": "pull_request:44017", "accept": false, - "reason": "RegNet vs Segformer; different model families." + "reason": "RegNet and Segformer are different model code paths." }, { "left": "pull_request:43997", "right": "pull_request:44056", "accept": false, - "reason": "RegNet and MPNet are separate model code paths." + "reason": "RegNet and MPNet are unrelated model-specific changes." }, { "left": "pull_request:44071", - "right": "pull_request:44076", - "accept": false, - "reason": "MPT vs ImageGPT; same output-tracing initiative, different underlying change." - }, - { - "left": "pull_request:44059", - "right": "pull_request:44068", + "right": "pull_request:44129", "accept": false, - "reason": "GPT2 and GPT-Neo are distinct model implementations." + "reason": "MPT and SpeechT5 are different concrete refactors." }, { - "left": "pull_request:44018", - "right": "pull_request:44059", + "left": "pull_request:44074", + "right": "pull_request:44101", "accept": false, - "reason": "GPT-Neo vs GPT2; not the same bug or change." + "reason": "TextNet and Flaubert/XLM are separate model implementations." }, { - "left": "pull_request:44101", + "left": "pull_request:44056", "right": "pull_request:44161", "accept": false, - "reason": "Flaubert/XLM vs LongT5; different models." + "reason": "MPNet and LongT5 are different code paths." }, { "left": "pull_request:43997", "right": "pull_request:44018", "accept": false, - "reason": "RegNet and GPT-Neo are unrelated model code paths." + "reason": "RegNet and GPT-Neo are unrelated model refactors." }, { "left": "pull_request:43997", - "right": "pull_request:44071", + "right": "pull_request:44161", "accept": false, - "reason": "RegNet vs MPT; different models." + "reason": "RegNet and LongT5 are different concrete changes." }, { "left": "pull_request:44074", "right": "pull_request:44161", "accept": false, - "reason": "TextNet and LongT5 are separate model implementations." - }, - { - "left": "pull_request:43997", - "right": "pull_request:44074", - "accept": false, - "reason": "RegNet vs TextNet; different concrete changes." + "reason": "TextNet and LongT5 are different model-specific refactors." }, { "left": "pull_request:44010", "right": "pull_request:44071", "accept": false, - "reason": "SqueezeBert vs MPT; only the refactor pattern is shared." + "reason": "SqueezeBert and MPT are unrelated code paths." }, { "left": "pull_request:44010", "right": "pull_request:44073", "accept": false, - "reason": "SqueezeBert and VisualBert are different model code paths." + "reason": "SqueezeBert and VisualBert are different model implementations." }, { "left": "pull_request:44018", "right": "pull_request:44129", "accept": false, - "reason": "GPT-Neo vs SpeechT5; unrelated changes." + "reason": "GPT-Neo and SpeechT5 are unrelated model-specific changes." }, { - "left": "pull_request:44056", + "left": "pull_request:44073", "right": "pull_request:44101", "accept": false, - "reason": "MPNet vs Flaubert/XLM; not the same underlying issue." + "reason": "VisualBert and Flaubert/XLM are different code paths." }, { "left": "pull_request:44017", "right": "pull_request:44071", "accept": false, - "reason": "Segformer vs MPT; different model implementations." + "reason": "Segformer and MPT are unrelated model refactors." }, { - "left": "pull_request:44010", - "right": "pull_request:44018", + "left": "pull_request:44056", + "right": "pull_request:44161", "accept": false, - "reason": "SqueezeBert vs GPT-Neo; unrelated code paths." + "reason": "MPNet and LongT5 are different concrete changes." }, { "left": "pull_request:43996", "right": "pull_request:44161", "accept": false, - "reason": "CVT/FNet vs LongT5; different models." + "reason": "FNet/CVT and LongT5 are unrelated model-specific refactors." }, { "left": "pull_request:43996", "right": "pull_request:44017", "accept": false, - "reason": "CVT/FNet vs Segformer; not the same change." + "reason": "FNet/CVT and Segformer are different code paths." }, { "left": "pull_request:44010", "right": "pull_request:44044", "accept": false, - "reason": "SqueezeBert vs DeBERTa; same pattern, different model code." + "reason": "SqueezeBert and DeBERTa_v2 are different model implementations." }, { "left": "pull_request:44044", "right": "pull_request:44074", "accept": false, - "reason": "DeBERTa vs TextNet; separate model implementations." + "reason": "DeBERTa_v2 and TextNet are unrelated concrete changes." }, { "left": "pull_request:44066", "right": "pull_request:44722", "accept": true, - "reason": "Same GPT-J/CodeGen output-tracing refactor in the same files; strong overlap and same concrete change." + "reason": "Same GPT-J/CodeGen output-tracing refactor in the same file set; 44722 is a later, broader duplicate/follow-up." }, { "left": "pull_request:44085", "right": "pull_request:44722", "accept": true, - "reason": "Both are GPT-J output-tracing refactors on the same code path; 44722 is the broader/consolidated version." + "reason": "Both appear to implement the same GPT-J/CodeGen output-tracing change on the same code path." }, { "left": "pull_request:44085", "right": "pull_request:44129", "accept": false, - "reason": "GPT-J vs SpeechT5; unrelated models." + "reason": "GPT-J/CodeGen and SpeechT5 are different models; shared issue tracking is not enough to merge them." }, { "left": "pull_request:43996", "right": "pull_request:44007", "accept": false, - "reason": "CVT/FNet vs ResNet/RegNet/RT-DETR ResNet; different code paths." + "reason": "FNet/CVT and RegNet are separate model-specific refactors." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44129", + "accept": false, + "reason": "FNet/CVT and SpeechT5 are unrelated code paths." }, { "left": "pull_request:44044", "right": "pull_request:44129", "accept": false, - "reason": "DeBERTa and SpeechT5 are unrelated model families." + "reason": "DeBERTa_v2 and SpeechT5 are different model implementations." }, { "left": "pull_request:44044", "right": "pull_request:44154", "accept": false, - "reason": "DeBERTa vs VITS; distinct implementations." + "reason": "DeBERTa_v2 and VITS are unrelated concrete changes." }, { "left": "pull_request:44024", "right": "pull_request:44076", "accept": false, - "reason": "FocalNet vs ImageGPT; same refactor style, different models." - }, - { - "left": "pull_request:44019", - "right": "pull_request:44085", - "accept": false, - "reason": "ResNet vs GPT-J; not the same underlying bug/change." - }, - { - "left": "pull_request:44068", - "right": "pull_request:44116", - "accept": false, - "reason": "GPT-Neo vs Flaubert; different model code paths." - }, - { - "left": "pull_request:44073", - "right": "pull_request:44101", - "accept": false, - "reason": "VisualBert vs Flaubert/XLM; separate model families." - }, - { - "left": "pull_request:44017", - "right": "pull_request:44073", - "accept": false, - "reason": "Segformer vs VisualBert; different concrete implementations." - }, - { - "left": "pull_request:44059", - "right": "pull_request:44086", - "accept": false, - "reason": "GPT2 vs MGP-STR; unrelated changes." - }, - { - "left": "pull_request:44056", - "right": "pull_request:44068", - "accept": false, - "reason": "MPNet vs GPT-Neo; different model code paths." - }, - { - "left": "pull_request:44018", - "right": "pull_request:44066", - "accept": false, - "reason": "GPT-Neo vs GPT-J/CodeGen; same initiative but not the same change." - }, - { - "left": "pull_request:44044", - "right": "pull_request:44129", - "accept": false, - "reason": "DeBERTa vs SpeechT5; different model families." + "reason": "FocalNet and ImageGPT are different model-specific refactors." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is well grounded in the packet and stays conservative about the soft edges. The accepted PR pairs are supported by shared model/file overlap, and the rejected pairs are mostly cross-model refactors that do not look mergeable into one bugfix PR. I don\u2019t see evidence of overstatement." + "feedback": "The summary is grounded in the packet and stays conservative about cross-model overlap. The accepted soft-edge pairs are the obvious same-path duplicates (GPT-J/CodeGen and GPT-Neo), and the rejected pairs are appropriately limited to shared theme rather than the same concrete bugfix. Canonical PR 44722 is also a reasonable representative because it covers the shared GPT-J/CodeGen path and has explicit issue linkage." }, "evaluator_used": true, "retried": false, @@ -18966,12 +19080,12 @@ "pull_request:43997|pull_request:44010", "pull_request:44010|pull_request:44129" ], - "prepared_review_unit_hash": "1eb03368c265f582cd474094078dc6121751a6147ce71363cd3a0bad5ccb165f", + "prepared_review_unit_hash": "48dde5da5b8455dc2f515afc68917cb727c0d729cf4b91d10a2e05f6b5c4a79d", "component_budget": { "node_count": 28, "item_count": 28, "soft_pair_count": 77, - "serialized_chars": 84502, + "serialized_chars": 84501, "estimated_input_tokens": 21126, "estimated_eval_tokens": 42508 }, @@ -18988,42 +19102,42 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "All items are separate model-specific PRs for the same broad maintenance theme: migrating to standardized output capture/tracing. They are not duplicates because they touch different model implementations and code paths.", - "confidence": 0.93, + "summary": "All items are pull requests for model-specific output-tracing refactors in different architectures. They share a theme but not the same underlying code path, so this cluster should not be merged across models.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 44073 is the clearest representative of the shared refactor pattern, with explicit _can_record_outputs hooks plus capture_outputs/can_return_tuple plumbing in one model file.", + "canonical_pr_reason": "PR 44129 is the strongest representative: it has an explicit issue target, a concrete SpeechT5-specific tracing refactor, and enough scope to characterize the cluster theme.", "best_issue_reason": null, - "best_pr_reason": "PR 44129 is the strongest standalone anchor because it is tied to an explicit issue target and presents a concrete SpeechT5 output-tracing migration rather than just a generic refactor.", + "best_pr_reason": "44129 is the best single representative because it is explicitly tied to an issue and is the most clearly scoped, model-specific implementation of the standardized output-capture refactor.", "soft_edge_verdicts": [ { "left": "pull_request:44073", "right": "pull_request:44074", "accept": false, - "reason": "Same refactor theme, but different models (VisualBert vs TextNet) and different code paths; not the same underlying change." + "reason": "Both are output-tracing refactors, but they target different models (VisualBert vs TextNet) and different forward paths; same pattern, not the same change." }, { "left": "pull_request:44056", "right": "pull_request:44073", "accept": false, - "reason": "Both update output tracing, but they are separate model implementations (MPNet vs VisualBert) with model-specific hooks and edits." + "reason": "MPNet and VisualBert are separate model implementations with distinct encoder internals; these are parallel refactors, not one duplicate fix." }, { "left": "pull_request:44056", "right": "pull_request:44074", "accept": false, - "reason": "MPNet and TextNet are unrelated model files; similarity is only at the broad migration pattern level." + "reason": "Different model families and code paths; they adopt the same decorator pattern but do not fix the same bug or change." }, { "left": "pull_request:43997", "right": "pull_request:44010", "accept": false, - "reason": "RegNet and SqueezeBert are distinct models; these are parallel refactors, not one shared fix." + "reason": "RegNet and SqueezeBert are unrelated model modules; both PRs update output handling, but they are separate model-specific refactors." }, { "left": "pull_request:44010", "right": "pull_request:44129", "accept": false, - "reason": "SqueezeBert and SpeechT5 changes are model-specific, and 44129 also includes a concrete issue-linked SpeechT5 adjustment, so they are not mergeable as one PR." + "reason": "SqueezeBert and SpeechT5 touch different architectures and different output semantics; the shared theme is too broad to treat as a duplicate or merge into one PR." } ] }, @@ -19109,304 +19223,304 @@ "pull_request:44002|pull_request:46106", "pull_request:43998|pull_request:46106" ], - "prepared_review_unit_hash": "ca5aed8b62e524377ccbfaeefb037b9de7bc7a612bbd9d99f240ab42be54ddbf", + "prepared_review_unit_hash": "ee057249f554dbdfbed6be83892d327bac3ecb4fca0cae8afc1d3f9fee1ac127", "component_budget": { "node_count": 14, "item_count": 14, "soft_pair_count": 45, - "serialized_chars": 43782, - "estimated_input_tokens": 10946, - "estimated_eval_tokens": 22148 + "serialized_chars": 43779, + "estimated_input_tokens": 10945, + "estimated_eval_tokens": 22146 }, "budget": { "node_count": 14, "item_count": 14, "soft_pair_count": 45, - "serialized_chars": 43782, - "estimated_input_tokens": 10946, - "estimated_eval_tokens": 22148 + "serialized_chars": 43779, + "estimated_input_tokens": 10945, + "estimated_eval_tokens": 22146 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These PRs are mostly separate model-specific output-tracing refactors, plus one unrelated docs toctree tweak; they share a theme but not the same concrete code-path bug/change, so I would not merge them as duplicates.", - "confidence": 0.88, + "summary": "This cluster is a loose family of model-specific output-tracing refactors plus one unrelated docs toctree cleanup. The PRs share a template, but they target different model code paths, so they are not duplicates.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 44003 is the best representative of the cluster theme: it is the broadest output-tracing refactor, spans multiple related model files, and introduces the shared capture/output machinery.", + "canonical_pr_reason": "PR 44003 is the best representative of the core pattern: it is a substantial tracing refactor, touches two closely related Mamba-family implementations, and includes the new output-capture plumbing.", "best_issue_reason": null, - "best_pr_reason": "PR 44003 is the strongest overall representative because it makes the most substantial tracing-infrastructure change rather than a narrow one-model adjustment.", + "best_pr_reason": "PR 44003 is the strongest standalone candidate because it is the most concrete and coherent output-tracing change among the set, with broader scope than the single-model tweaks.", "soft_edge_verdicts": [ { "left": "pull_request:43998", "right": "pull_request:44002", "accept": false, - "reason": "Same output-tracing theme, but different model families and separate forward paths; not the same concrete change." + "reason": "Same high-level tracing refactor pattern, but different model families and different forward paths." }, { "left": "pull_request:43998", "right": "pull_request:44000", "accept": false, - "reason": "Both are tracing refactors, but timm_backbone and vision_text_dual_encoder are unrelated code paths." + "reason": "Different architectures and code paths; only the refactor style is shared." }, { "left": "pull_request:43998", "right": "pull_request:44001", "accept": false, - "reason": "Different model implementations; this is only broad architectural similarity, not a duplicate fix." + "reason": "Unrelated model implementations; not the same underlying bug or change." }, { "left": "pull_request:43998", "right": "pull_request:44003", "accept": false, - "reason": "timm_backbone vs mamba/falcon_mamba are distinct subsystems and changes; too broad to be one PR." + "reason": "Both are output-tracing changes, but they affect different model stacks and concrete code paths." }, { "left": "pull_request:43998", "right": "pull_request:44029", "accept": false, - "reason": "Different model stacks and output behavior; not the same underlying bug/change." + "reason": "Different models and runtime paths; similarity is only at the refactor template level." }, { "left": "pull_request:43999", "right": "pull_request:44001", "accept": false, - "reason": "MobileNetV1 and UnivNet are unrelated model paths; shared refactor pattern only." + "reason": "Separate model implementations; not a duplicate fix or mergeable single change." }, { "left": "pull_request:43999", "right": "pull_request:44002", "accept": false, - "reason": "These touch different model architectures and different forward signatures, so not duplicates." + "reason": "Different model code paths, even though both adopt the same tracing refactor style." }, { "left": "pull_request:43999", "right": "pull_request:44025", "accept": false, - "reason": "MobileNetV1 vs DepthAnything are separate models; same tracing cleanup theme does not make them the same change." + "reason": "Different model families and files; same general refactor theme is too broad to be duplicate." }, { "left": "pull_request:43999", "right": "pull_request:44026", "accept": false, - "reason": "Different model families and code paths; not one mergeable fix." + "reason": "Separate encoder/decoder wrapper code paths; not the same concrete change." }, { "left": "pull_request:43999", "right": "pull_request:44027", "accept": false, - "reason": "MobileNetV1 and SpeechEncoderDecoder do not share a concrete bug beyond generic output tracing." + "reason": "Different model wrappers and call paths; similar cleanup, not duplicate work." }, { "left": "pull_request:44000", "right": "pull_request:44001", "accept": false, - "reason": "Vision-text dual encoder and UnivNet are unrelated implementations; only the refactor style matches." + "reason": "Different models and outputs; the shared theme is only tracing refactor." }, { "left": "pull_request:44000", "right": "pull_request:44002", "accept": false, - "reason": "Different model code paths and outputs; not the same bug/change." + "reason": "Distinct architectures and forward implementations, so they are not the same change." }, { "left": "pull_request:44000", "right": "pull_request:44003", "accept": false, - "reason": "Vision-text dual encoder and mamba/falcon_mamba are separate subsystems; reject as a duplicate." + "reason": "These touch unrelated model families and cannot plausibly be merged as one concrete fix." }, { "left": "pull_request:44000", "right": "pull_request:44025", "accept": false, - "reason": "These are distinct model refactors in different directories, not one concrete fix." + "reason": "Same broad tracing cleanup category, but different model code paths and behavior." }, { "left": "pull_request:44001", "right": "pull_request:44002", "accept": false, - "reason": "UnivNet and UperNet are unrelated models; shared tracing cleanup is too broad." + "reason": "Different model implementations; no shared bug beyond the refactor pattern." }, { "left": "pull_request:44001", "right": "pull_request:44003", "accept": false, - "reason": "Different model stacks and code paths; not a single underlying issue." + "reason": "Separate model families and forward logic; not a duplicate." }, { "left": "pull_request:44001", "right": "pull_request:44025", "accept": false, - "reason": "UnivNet vs DepthAnything are separate model changes, not duplicates." + "reason": "Different model stacks; too broad to treat as the same underlying change." }, { "left": "pull_request:44002", "right": "pull_request:44029", "accept": false, - "reason": "UperNet and RWKV do not implement the same code path; same theme only." + "reason": "Different models and concrete code paths; only the output-tracing style overlaps." }, { "left": "pull_request:44002", "right": "pull_request:44003", "accept": false, - "reason": "Different architectures and output plumbing; not the same concrete bug." + "reason": "Not the same implementation or bug; just a similar refactor pattern." }, { "left": "pull_request:44003", "right": "pull_request:44028", "accept": false, - "reason": "mamba/falcon_mamba and SuperPoint are separate model families; no duplicate bug." + "reason": "Different model families and modules; these are not the same underlying change." }, { "left": "pull_request:44003", "right": "pull_request:44004", "accept": false, - "reason": "These are different model implementations and forward contracts; too broad to merge as one PR." + "reason": "Separate architectures and forward logic; similar cleanup does not make them duplicates." }, { "left": "pull_request:43998", "right": "pull_request:44004", "accept": false, - "reason": "timm_backbone and CodeGen are unrelated model paths; same refactor pattern only." + "reason": "Different model code paths; only the tracing refactor motif matches." }, { "left": "pull_request:44001", "right": "pull_request:44004", "accept": false, - "reason": "UnivNet and CodeGen are different models with different output handling; not duplicates." + "reason": "Distinct models and concrete forward behavior; not mergeable as one fix." }, { "left": "pull_request:44002", "right": "pull_request:44004", "accept": false, - "reason": "UperNet and CodeGen are separate code paths; the similarity is only at the abstraction level." + "reason": "Same broad theme, but different model implementations and behavior." }, { "left": "pull_request:44000", "right": "pull_request:44004", "accept": false, - "reason": "Vision-text dual encoder and CodeGen are unrelated model families; reject." + "reason": "Different model-specific forward code; not the same bug or change." }, { "left": "pull_request:44025", "right": "pull_request:44026", "accept": false, - "reason": "DepthAnything and VisionEncoderDecoder are different models and different forward logic; not one concrete fix." + "reason": "Related refactor style, but the PRs target different wrappers and code paths." }, { "left": "pull_request:44025", "right": "pull_request:44027", "accept": false, - "reason": "DepthAnything and SpeechEncoderDecoder are separate code paths; too broad to be duplicate." + "reason": "Different model wrappers and runtime paths; not a duplicate fix." }, { "left": "pull_request:44025", "right": "pull_request:44028", "accept": false, - "reason": "DepthAnything and SuperPoint are unrelated vision models; same tracing refactor theme only." + "reason": "Separate model implementations, so this is only thematic similarity." }, { "left": "pull_request:44025", "right": "pull_request:44029", "accept": false, - "reason": "DepthAnything and RWKV are different architectures and output semantics; not the same change." + "reason": "Different model families and output logic; not the same underlying issue." }, { "left": "pull_request:44025", "right": "pull_request:44030", "accept": false, - "reason": "DepthAnything and DPR are separate model code paths, so this is not a duplicate PR." + "reason": "Different model code paths despite the shared tracing-refactor pattern." }, { "left": "pull_request:44026", - "right": "pull_request:44029", + "right": "pull_request:44027", "accept": false, - "reason": "VisionEncoderDecoder and RWKV are unrelated model stacks; no shared concrete bug." + "reason": "Both are encoder-decoder wrappers, but they are different model classes and code paths." }, { "left": "pull_request:44026", - "right": "pull_request:44027", + "right": "pull_request:44028", "accept": false, - "reason": "Different encoder-decoder variants and forward behavior; not one mergeable fix." + "reason": "Different model implementations; the overlap is only in refactor style." }, { "left": "pull_request:44026", - "right": "pull_request:44028", + "right": "pull_request:44029", "accept": false, - "reason": "VisionEncoderDecoder and SuperPoint touch different subsystems and output contracts." + "reason": "Unrelated model families and forward paths; not duplicate work." }, { "left": "pull_request:44026", "right": "pull_request:44030", "accept": false, - "reason": "VisionEncoderDecoder and DPR are separate model paths; only the refactor pattern overlaps." + "reason": "Different concrete model behavior, so these are not the same change." }, { "left": "pull_request:44027", "right": "pull_request:44028", "accept": false, - "reason": "SpeechEncoderDecoder and SuperPoint are different models with different outputs; not duplicates." + "reason": "Different model families and files; not a single underlying bug or fix." }, { "left": "pull_request:44027", "right": "pull_request:44029", "accept": false, - "reason": "Speech encoder-decoder and RWKV do not share the same code-path issue." + "reason": "Separate code paths and architectures; only the refactor motif is shared." }, { "left": "pull_request:44027", "right": "pull_request:44030", "accept": false, - "reason": "Different model families and output plumbing; same theme is not enough." + "reason": "Different models and concrete behavior; not the same change." }, { "left": "pull_request:44028", "right": "pull_request:44029", "accept": false, - "reason": "SuperPoint and RWKV are unrelated implementations; reject as duplicate." + "reason": "Distinct model implementations and tracing paths; not duplicates." }, { "left": "pull_request:44028", "right": "pull_request:44030", "accept": false, - "reason": "SuperPoint and DPR touch different model code paths and can\u2019t plausibly be one PR." + "reason": "Different code paths and architectures, so they should stay separate." }, { "left": "pull_request:44029", "right": "pull_request:44030", "accept": false, - "reason": "RWKV and DPR are unrelated model architectures; not the same underlying bug/change." + "reason": "Different model families and forward implementations; only the refactor pattern overlaps." }, { "left": "pull_request:44027", "right": "pull_request:46106", "accept": false, - "reason": "Model code refactor vs documentation toctree edit; entirely unrelated." + "reason": "The docs toctree cleanup is unrelated to model output tracing." }, { "left": "pull_request:44026", "right": "pull_request:46106", "accept": false, - "reason": "Documentation-only toctree change, not a model tracing fix." + "reason": "Documentation navigation change, not the same kind of model refactor." }, { "left": "pull_request:44001", "right": "pull_request:46106", "accept": false, - "reason": "UnivNet model refactor and docs navigation change are unrelated." + "reason": "Docs-only toctree edit is unrelated to the model tracing PR." }, { "left": "pull_request:44002", "right": "pull_request:46106", "accept": false, - "reason": "UperNet code change vs docs toctree cleanup; not duplicates." + "reason": "Unrelated docs cleanup versus model forward/tracing changes." }, { "left": "pull_request:43998", "right": "pull_request:46106", "accept": false, - "reason": "Model refactor and documentation layout change have no shared underlying bug." + "reason": "Documentation restructuring is not the same underlying change as model output tracing." } ] }, @@ -19439,52 +19553,52 @@ "pull_request:44662|pull_request:45075", "pull_request:45075|pull_request:45401" ], - "prepared_review_unit_hash": "b997eaf3655335ae70659158855443277a543e52ac4da32354327adb6ccc7eeb", + "prepared_review_unit_hash": "056a5defb29e5d7941dbdff78b97b7324b3df56a452db784f88a882c57c9cec1", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 10913, - "estimated_input_tokens": 2729, - "estimated_eval_tokens": 5714 + "serialized_chars": 10912, + "estimated_input_tokens": 2728, + "estimated_eval_tokens": 5712 }, "budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 10913, - "estimated_input_tokens": 2729, - "estimated_eval_tokens": 5714 + "serialized_chars": 10912, + "estimated_input_tokens": 2728, + "estimated_eval_tokens": 5712 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are three separate model-integration pull requests for different models (PenguinVL, DeepSeek-OCR-2, and Voxtral TTS). They share common registry/docs boilerplate, but they do not appear to fix the same code-path problem or be mergeable into one PR.", - "confidence": 0.98, + "summary": "All three PRs are separate model-addition efforts for different architectures (PenguinVL, DeepSeek-OCR-2, and Voxtral TTS). The shared files are mostly framework boilerplate for registering a new model, not evidence of the same underlying bug or change.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: each PR adds a different model implementation with distinct files, configs, processors, and tests.", + "canonical_pr_reason": "45075 is the best representative of the cluster: it is a full, self-contained model integration with conversion, modeling, processing, tests, and config updates, whereas the others are different model families.", "best_issue_reason": null, - "best_pr_reason": "No single best PR for deduplication; these are independent model additions rather than duplicates of the same change.", + "best_pr_reason": "45075 is the strongest standalone PR in the set, but only as a representative of a new-model integration pattern; it is not a duplicate of the other two.", "soft_edge_verdicts": [ { "left": "pull_request:44662", "right": "pull_request:45401", "accept": false, - "reason": "Different model additions (PenguinVL vs Voxtral TTS). Shared auto-registry/docs files are boilerplate, not evidence of the same underlying bug or change." + "reason": "Different model families and codepaths: PenguinVL is a vision-language model, while Voxtral TTS is a text-to-speech model. Shared auto/docs registration files are boilerplate, not duplicate evidence." }, { "left": "pull_request:44662", "right": "pull_request:45075", "accept": false, - "reason": "Different models and different implementation stacks (PenguinVL vs DeepSeek-OCR-2). They overlap only in generic transformer registration/docs." + "reason": "These add different models (PenguinVL vs DeepSeek-OCR-2) with distinct configs, converters, processors, and tests. Similar registration scaffolding is not enough to treat them as the same change." }, { "left": "pull_request:45075", "right": "pull_request:45401", "accept": false, - "reason": "DeepSeek-OCR-2 and Voxtral TTS are unrelated integrations; the overlap is limited to shared model-registration scaffolding, so they should not be merged as duplicates." + "reason": "DeepSeek-OCR-2 and Voxtral TTS solve unrelated problems in different modalities (OCR vs TTS) and touch different implementation paths. They are not mergeable as one PR." } ] }, @@ -19514,7 +19628,7 @@ "soft_pairs": [ "pull_request:44979|pull_request:45363" ], - "prepared_review_unit_hash": "1b2184b07eb6cd3f1144dcd5dcf6d770e99c5b235c9727c073289a0898caed47", + "prepared_review_unit_hash": "7ec26b41425bacf77fed87d34dbf523793c66fdac61457df6a909bff0d2a06a8", "component_budget": { "node_count": 2, "item_count": 2, @@ -19536,18 +19650,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs concern fusion infrastructure, but they implement different layers: #44979 adds a core Module Fusion API, while #45363 adds KernelConfig-based n-to-1 kernel fusion and related integration/example changes. They are related, but not the same change and should not be merged as duplicates.", - "confidence": 0.91, + "summary": "These are related kernel-fusion PRs, but they do not look like the same change: one adds a generic Module Fusion API scaffold, while the other implements n-to-1 kernel fusion through KernelConfig and integration hooks. They could complement each other, but not be merged as duplicates.", + "confidence": 0.87, "canonical_issue_reason": null, - "canonical_pr_reason": "#44979 is the cleaner canonical PR for this cluster because it introduces the foundational Module Fusion API; #45363 is a broader follow-on feature built on top of fusion concepts rather than the same concrete change.", + "canonical_pr_reason": "PR #45363 is the better canonical PR because it makes the concrete feature change: KernelConfig-driven n-to-1 fusion wired into model loading/integration code, with an example showing end-to-end use. PR #44979 is more of an API foundation and does not appear to implement the same concrete behavior.", "best_issue_reason": null, - "best_pr_reason": "#44979 is the best representative PR for the cluster since it is the core, self-contained API addition. #45363 is more specialized and depends on a different integration path (KernelConfig-based fusion).", + "best_pr_reason": "PR #45363 is the strongest representative of the cluster because it targets the actual user-facing fusion mechanism and touches the code paths needed to make it work, rather than just defining the underlying module-fusion abstraction.", "soft_edge_verdicts": [ { "left": "pull_request:44979", "right": "pull_request:45363", "accept": false, - "reason": "Related fusion work, but not the same underlying bug/change: #44979 defines the module-fusion API, while #45363 implements kernel-config-driven n-to-1 fusion and usage plumbing. They would not plausibly be merged into one PR as duplicates." + "reason": "Related area, but different scope and likely different merge units: #44979 adds a generic Module Fusion API and tests, while #45363 implements KernelConfig-based n-to-1 kernel fusion across model-loading/integration paths. They are not the same concrete bug/change." } ] }, @@ -19577,7 +19691,7 @@ "soft_pairs": [ "pull_request:45086|pull_request:45317" ], - "prepared_review_unit_hash": "dd92f295ec2b7bf6c4d2a91832db4731fe53fd36fa93e8ec5135379025fc0c93", + "prepared_review_unit_hash": "f092f7d542120c92e78e9bc6a571a8b8052283cbf2603fa0e984a8088b691b85", "component_budget": { "node_count": 2, "item_count": 2, @@ -19599,18 +19713,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both pull requests address the same Mistral regex patch AttributeError in `tokenization_utils_tokenizers.py`; PR 45317 is the more complete fix because it includes the same code change plus test coverage.", - "confidence": 0.97, + "summary": "Both PRs address the same concrete AttributeError in `_patch_mistral_regex` by switching from `backend_tokenizer.pre_tokenizer` to `tokenizer.pre_tokenizer`; the second PR is the same fix plus tests, so they are duplicate candidates.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45317 is the best canonical PR because it fixes the same AttributeError as 45086 and adds accompanying tests, making it the more complete implementation of the shared bug fix.", + "canonical_pr_reason": "PR #45317 is the better canonical PR because it contains the same code fix as #45086 plus test coverage, making it the more complete representation of the fix.", "best_issue_reason": null, - "best_pr_reason": "PR 45317 is the best overall PR for this cluster: same concrete code-path fix, more complete than 45086, and includes tests in addition to the code change.", + "best_pr_reason": "PR #45317 is the best overall PR candidate because it fixes the same bug as #45086 and adds validation/tests, so it is the most complete and maintainable version of the change.", "soft_edge_verdicts": [ { "left": "pull_request:45086", "right": "pull_request:45317", "accept": true, - "reason": "Accept: both PRs fix the same AttributeError in `_patch_mistral_regex` by changing the pre-tokenizer access path in the same file; PR 45317 is effectively the fuller version with tests." + "reason": "Same underlying AttributeError fix in the same function and file; the second PR is the same change with added tests, so they could plausibly be merged into one PR." } ] }, @@ -19640,7 +19754,7 @@ "soft_pairs": [ "pull_request:45115|pull_request:45569" ], - "prepared_review_unit_hash": "9ec6720b783baf4c54243a81aa7ed8b5c541448201d3228047e89b82bb51f521", + "prepared_review_unit_hash": "192dec3a76c7033644aa9133c193fd89bdd1e4b2c26d31a81add50c73a0a0d53", "component_budget": { "node_count": 2, "item_count": 2, @@ -19662,18 +19776,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs are about NemotronH, but they address different changes: #45115 is a focused inheritance/refactor and conversion-mapping update, while #45569 is a broader model split introducing NemotronHDense/NemotronHSparse, docs, auto-mappings, and tests. They are related, but not the same underlying change set.", - "confidence": 0.91, + "summary": "The two PRs are related to NemotronH but address different changes: PR 45115 is a small conversion-mapping/refactor for checkpoint key renaming and model inheritance, while PR 45569 is a broader architecture split introducing separate NemotronHDense/NemotronHSparse models, docs, and tests. They are not duplicates.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #45569 is the better canonical representative because it is the broader, more complete NemotronH architecture change and includes the user-facing model split plus docs/tests, whereas #45115 is a narrower refactor/conversion-mapping step.", + "canonical_pr_reason": "PR 45569 is the broader, more complete implementation: it introduces the dedicated dense/sparse model packages, updates auto-mapping and docs, and includes tests. PR 45115 is narrower and more provisional, focused on conversion mapping and inheritance plumbing.", "best_issue_reason": null, - "best_pr_reason": "PR #45569 is the most complete and current implementation of the NemotronH work. It spans the concrete model/API restructuring and supporting docs/tests, making it the strongest single representative for this cluster.", + "best_pr_reason": "PR 45569 is the best representative artifact because it covers the main user-facing model restructuring and contains the fuller codepath changes, while PR 45115 appears to be an earlier, narrower refactor step.", "soft_edge_verdicts": [ { "left": "pull_request:45115", "right": "pull_request:45569", "accept": false, - "reason": "Do not merge: #45115 refactors NemotronH to inherit GraniteMoeHybrid and adjusts conversion mapping, but #45569 introduces separate NemotronHDense/NemotronHSparse models and broader API/docs/test changes. They touch overlapping files, but the concrete code-path changes are different and not safely one PR." + "reason": "Different underlying changes. 45115 mainly renames checkpoint conversion keys and adjusts inheritance/config plumbing; 45569 splits NemotronH into separate dense/sparse model families with docs and tests. They touch some shared files but would not plausibly be merged into one PR as the same fix." } ] }, @@ -19703,12 +19817,12 @@ "soft_pairs": [ "pull_request:45332|pull_request:45333" ], - "prepared_review_unit_hash": "daf34689c5ad842a8c09de16af5a2635ff6414fdced8c64e4ccefc6395da3a12", + "prepared_review_unit_hash": "536d16a6e4efe865c1728d9c35fe32533f8c0ded1d241b206ccd3270d8198e7f", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 5397, + "serialized_chars": 5398, "estimated_input_tokens": 1350, "estimated_eval_tokens": 2956 }, @@ -19716,7 +19830,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 5397, + "serialized_chars": 5398, "estimated_input_tokens": 1350, "estimated_eval_tokens": 2956 }, @@ -19725,18 +19839,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two related but distinct PRs: one introduces heterogeneous per-layer config support, the other adds heterogeneous modeling/runtime support. They overlap heavily in plumbing files, but they are not the same concrete change.", - "confidence": 0.89, + "summary": "Two related but distinct PRs: one adds heterogeneous per-layer configuration support, the other adds heterogeneous modeling/runtime support built on top of that config work. They share infrastructure files, but they are not the same underlying change.", + "confidence": 0.93, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45332 is the broader, later-stage integration piece: it adds heterogeneous model/runtime handling in addition to using the new per-layer config plumbing, so it is the better representative of the cluster.", + "canonical_pr_reason": "No single canonical PR: #45333 is the config-layer foundation, while #45332 adds broader modeling/cache/masking changes and is a separate follow-on change.", "best_issue_reason": null, - "best_pr_reason": "PR 45332 is the best single representative because it covers the larger end-to-end feature set and depends on the config machinery introduced in the other PR.", + "best_pr_reason": "#45332 is the broader end-to-end heterogeneous model support PR, so it is the better representative of the cluster theme, even though it depends on the config work in #45333.", "soft_edge_verdicts": [ { "left": "pull_request:45332", "right": "pull_request:45333", "accept": false, - "reason": "Related but not duplicate: 45333 focuses on per-layer configuration support, while 45332 focuses on modeling/runtime support for heterogeneous layers. They share infrastructure files, but they solve different layers of the feature and would not plausibly be merged as one PR." + "reason": "Related stack, but not the same concrete bug/change. #45333 introduces heterogeneous per-layer config support; #45332 extends modeling, cache, and masking behavior to use it. They could not be cleanly merged as one duplicate PR." } ] }, @@ -19766,7 +19880,7 @@ "soft_pairs": [ "pull_request:45415|pull_request:45425" ], - "prepared_review_unit_hash": "3abad42149ea336f5b5a84a7f563befda12e123146d4136aee08c94c8fbae083", + "prepared_review_unit_hash": "cf25d9cfb8ba1acaf86a97eefd8c60fa91eaab1112b6564e45049bc9414c346f", "component_budget": { "node_count": 2, "item_count": 2, @@ -19788,18 +19902,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs are type-checking related, but they are not the same change: one is a broad repo-wide typing rollout, the other adds/adjusts shared typing protocols and targeted typing fixes. I would not treat them as duplicates.", - "confidence": 0.9, + "summary": "Both PRs are typing-oriented, but they address different scopes: #45415 is a broad repo-wide type-checking sweep, while #45425 is a narrower protocol/typing refinement around modeling utilities and PEFT-related interfaces.", + "confidence": 0.89, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45425 is the more focused, self-contained typing-infrastructure change; PR 45415 is a broad type-checking sweep across many files and pyproject settings.", + "canonical_pr_reason": "PR #45425 is the better canonical representative because it is a focused, self-contained typing change; #45415 is a broad housekeeping sweep across many files and is less suitable as the cluster anchor.", "best_issue_reason": null, - "best_pr_reason": "PR 45425 is the best representative of the cluster because it has a narrower, coherent goal and a clearer code-path impact than the large typing sweep in 45415.", + "best_pr_reason": "#45425 is the strongest representative of this cluster because it has a coherent target and clearer code-path impact, whereas #45415 is a large cross-cutting type-check enablement patch.", "soft_edge_verdicts": [ { "left": "pull_request:45415", "right": "pull_request:45425", "accept": false, - "reason": "Both are typing-related, but 45415 is a broad type-checking rollout while 45425 adds typing protocols and targeted fixes; they do not appear to solve the same concrete problem and would not naturally merge into one PR." + "reason": "Reject: both are about typing, but they are not the same underlying change. #45415 is a broad type-checking cleanup across many modules; #45425 adds specific typing protocols and modeling-utils-related stubs. Shared files and subsystem are not enough to treat them as duplicates." } ] }, @@ -19836,7 +19950,7 @@ "pull_request:46109|pull_request:46151", "pull_request:46135|pull_request:46151" ], - "prepared_review_unit_hash": "70c734d0f2b6dec86990754f582a7e950aa5277607f47bb96012c09e93bd3b09", + "prepared_review_unit_hash": "316c6f1778779b75c9c161c484b3bad17aaf87b766489f7102a68d3a44064265", "component_budget": { "node_count": 5, "item_count": 5, @@ -19858,42 +19972,42 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is only loosely related by accelerator/device compatibility. Two PRs fix real model-parallel device placement bugs, while the others are mostly test expectation updates for different subsystems. They are not the same underlying change set.", - "confidence": 0.34, + "summary": "This cluster is mostly a loose similarity group around device/expectation updates, but the artifacts are not true duplicates. Two PRs are concrete model-parallel bug fixes in different models, while the others are test-only expectation adjustments in unrelated areas.", + "confidence": 0.73, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45487 is the strongest representative because it is a concrete functional fix for device placement in model-parallel inference across model implementations; the other PRs are narrower or test-only and do not appear to be duplicates of it.", + "canonical_pr_reason": "PR 45487 is the broadest concrete code fix: it addresses a real model-parallel device-placement bug in AltCLIP/ChineseCLIP model code and includes targeted test updates. It is more representative than the narrower or test-only PRs.", "best_issue_reason": null, - "best_pr_reason": "PR 45487 is the best standalone PR here because it addresses an actual code-path bug rather than only updating tests or expectations. The rest are unrelated enough that they should not be merged into a single duplicate cluster.", + "best_pr_reason": "PR 45487 is the best representative artifact in this cluster because it fixes an actual runtime bug in model code rather than just updating test expectations, and it covers the same general class of device-mismatch issue more fully than the other code-fix PR.", "soft_edge_verdicts": [ { "left": "pull_request:45487", "right": "pull_request:46101", "accept": false, - "reason": "Both are device-placement fixes, but they target different model code paths (AltCLIP/ChineseCLIP vs Qwen2 audio) and different masking logic. They are related in theme, not the same change." + "reason": "Both are model-parallel device-placement fixes, but they hit different models and different code paths (AltCLIP/ChineseCLIP vs Qwen2-Audio). Too different to merge as one PR." }, { "left": "pull_request:45487", "right": "pull_request:46151", "accept": false, - "reason": "45487 is a model-parallel bug fix; 46151 only adds test expectations for pipeline outputs on XPU/other backends. Different scope and underlying problem." + "reason": "45487 is a model-code bug fix; 46151 only adds test expectations for pipeline outputs. Different artifact types and different subsystems." }, { "left": "pull_request:46101", "right": "pull_request:46151", "accept": false, - "reason": "Qwen2 audio mask device placement and token-classification pipeline expectations are unrelated concrete changes, even though both mention accelerator support." + "reason": "One is a runtime fix in Qwen2-Audio model code; the other is a test expectation update for pipeline token classification. Not the same underlying change." }, { "left": "pull_request:46109", "right": "pull_request:46151", "accept": false, - "reason": "These are both test-oriented, but they cover different test suites and different parity behaviors. They do not look like the same underlying bug or change." + "reason": "Both touch tests, but they are for unrelated areas: distributed trainer parity assertions vs token-classification pipeline expectations. No shared bug/change." }, { "left": "pull_request:46135", "right": "pull_request:46151", "accept": false, - "reason": "Kosmos2.5 integration-test expectations for XPU are not the same as token-classification pipeline expectations. Same broad accelerator theme, different artifacts and behavior." + "reason": "These are both test-related, but they concern different models and different expectations (Kosmos2.5 integration vs token-classification pipeline). Not duplicates." } ] }, @@ -19925,12 +20039,12 @@ "pull_request:45513|pull_request:46084", "pull_request:46084|pull_request:46098" ], - "prepared_review_unit_hash": "7be3e71ff394717888aaf225aafdbb0c728f689a27ae81bc7cd201e8b06f12ca", + "prepared_review_unit_hash": "6d206235d9654178271ce29124fefb1b020db0c6530b1d78c70bb9b038776a0a", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 2, - "serialized_chars": 7427, + "serialized_chars": 7426, "estimated_input_tokens": 1857, "estimated_eval_tokens": 3970 }, @@ -19938,7 +20052,7 @@ "node_count": 3, "item_count": 3, "soft_pair_count": 2, - "serialized_chars": 7427, + "serialized_chars": 7426, "estimated_input_tokens": 1857, "estimated_eval_tokens": 3970 }, @@ -19947,31 +20061,34 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two Mamba2 PRs are closely related on the same seq_len>1 cache bug; the Qwen3.5 PR is a different model-specific cache fix and is not a duplicate. The broader Mamba2 PR is the best representative of the underlying issue cluster.", - "confidence": 0.87, + "summary": "The two Mamba2 PRs are about the same cached-forward seq_len>1 bug and are good duplicate candidates; the Qwen3.5 PR is a separate model-family fix and should not be merged with them.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #46098 is the most complete Mamba2 fix: it targets the shared issue (#46032), covers the seq_len>1 cache failure across CPU/GPU paths, and includes tests/docs, so it best represents the underlying change.", + "canonical_pr_reason": "pull_request:46098 is the most complete representative of the Mamba2 bug fix: it has the clearest title, explicitly targets the tracked issue, and includes the code fix plus tests/docs around the same cached seq_len>1 failure mode.", "best_issue_reason": null, - "best_pr_reason": "PR #46098 is the strongest canonical PR because it is the broadest, most explicit fix for the shared Mamba2 cache bug and likely subsumes the narrower code-only change in #46084.", + "best_pr_reason": "pull_request:46098 is the strongest standalone PR for the cluster because it spells out the concrete bug, covers both CPU/GPU paths, and is broader than the smaller companion fix in 46084.", "soft_edge_verdicts": [ { "left": "pull_request:45513", "right": "pull_request:46084", "accept": false, - "reason": "Both are cache/seq_len-related, but they fix different model families and different code paths; this is similar theme, not the same underlying bug or mergeable PR." + "reason": "Different models and different code paths: Qwen3.5 linear attention vs Mamba2 mixer cache handling. Similar cache-length symptoms are not enough to treat them as the same underlying change." }, { "left": "pull_request:46084", "right": "pull_request:46098", - "accept": false, - "reason": "They share issue #46032 and the Mamba2 model, but #46098 is broader and includes additional CPU-path/docs changes; too scope-different to treat as the same PR duplicate with confidence." + "accept": true, + "reason": "Both PRs fix the same Mamba2 use_cache seq_len>1 cached-forward bug against issue #46032, touch the same core files, and could plausibly be merged into one consolidated PR." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet. The Qwen3.5 PR is correctly treated as separate from the Mamba2 fixes, and the 46084/46098 pair is reasonably treated as mergeable into one PR for the same issue (#46032) since they share the same model, files, and failure mode. The canonical/standalone PR reasoning for 46098 is also supported by its broader scope and explicit issue linkage." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -19994,12 +20111,12 @@ "soft_pairs": [ "pull_request:45548|pull_request:45662" ], - "prepared_review_unit_hash": "d2d174bdc38a919975c45d09d4aec4f2c6fb1d79293af4e438c2eb7d799bcb85", + "prepared_review_unit_hash": "20b05605e6c0803317f34054cb4e88e71a54c56416d72cacf90fa219ca8aee97", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4507, + "serialized_chars": 4508, "estimated_input_tokens": 1127, "estimated_eval_tokens": 2510 }, @@ -20007,7 +20124,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4507, + "serialized_chars": 4508, "estimated_input_tokens": 1127, "estimated_eval_tokens": 2510 }, @@ -20016,18 +20133,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are related to EP, but they address different concrete failure modes and code paths: one is about loading via accelerate/ZeRO-3 and config serialization, the other is about FSDP2 expert parameters being overwritten during rank-0 broadcast. Not duplicates.", + "summary": "These PRs are related to expert-parallel loading, but they fix different failure modes: one preserves distributed config and adds an EP detection helper for DeepSpeed ZeRO-3/accelerate launch, while the other fixes expert parameters being overwritten during FSDP2/rank-0 broadcast and adds DTensor-aware handling across the loading path.", "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45662 is the stronger canonical choice because it targets a specific, concrete corruption bug in expert parameter loading/broadcasting and includes the relevant loading-path fixes.", + "canonical_pr_reason": "No single canonical PR for a duplicate pair here; the two changes target different code paths and different concrete bugs.", "best_issue_reason": null, - "best_pr_reason": "PR 45662 is the best representative of the cluster because it addresses the most explicit and severe runtime bug with a clear, narrowly-defined code path.", + "best_pr_reason": "PR #45662 is the stronger representative of the cluster because it addresses a more concrete loading corruption bug with a clearer end-user symptom and a broader, internally consistent fix across the loading stack.", "soft_edge_verdicts": [ { "left": "pull_request:45548", "right": "pull_request:45662", "accept": false, - "reason": "Both concern expert parallelism, but 45548 is about serialize/load behavior with accelerate/ZeRO-3 while 45662 fixes expert parameters being overwritten by rank-0 broadcast under FSDP2; different underlying bugs and not mergeable as one PR." + "reason": "Both involve EP/loading, but they do not appear to fix the same underlying bug. #45548 is about keeping distributed_config/EP state available for ZeRO-3 loading via accelerate launch, while #45662 prevents expert parameters from being silently overwritten during FSDP2 rank-0 broadcast and adds DTensor-aware post-shard handling. Same subsystem, different failure modes and patch shapes." } ] }, @@ -20057,7 +20174,7 @@ "soft_pairs": [ "pull_request:45639|pull_request:45645" ], - "prepared_review_unit_hash": "00bbb1687349f281fda25e889cc241f6303ce26bb44eec24ae4d8038a3d3eef9", + "prepared_review_unit_hash": "7bb44fc66c42b7ec2793ed5a1c58fd4a6d9022c436278beb967744bce4e3c511", "component_budget": { "node_count": 2, "item_count": 2, @@ -20079,18 +20196,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs target the same xdist/captured_info logging problem, with PR 45645 extending the narrower code fix in 45639 to also preserve CI log collection. They are close enough to be treated as the same change cluster, with 45645 as the broader superset implementation.", - "confidence": 0.89, + "summary": "Both PRs address the same xdist-related captured_info logging problem in Transformers testing utilities; #45645 appears to be the broader, more complete version of #45639 and adds the CI-side handling needed for the new per-worker files.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45645 is the better canonical PR because it includes the core xdist-safe captured_info fix plus the CI workflow update needed to consume multiple per-worker log files.", + "canonical_pr_reason": "#45645 is the more complete fix: it includes the core xdist-safe output-file change in testing_utils plus the workflow update to collect multiple captured_info files, so it subsumes #45639.", "best_issue_reason": null, - "best_pr_reason": "PR 45645 is the strongest representative of the cluster: it covers the shared bug fix in testing_utils and the accompanying workflow handling, making it more complete than 45639.", + "best_pr_reason": "#45645 is the best representative of the cluster because it covers the full end-to-end change needed to make captured_info logging xdist-safe and still visible in CI.", "soft_edge_verdicts": [ { "left": "pull_request:45639", "right": "pull_request:45645", "accept": true, - "reason": "Same underlying captured_info xdist collision fix; 45645 incorporates the same testing_utils change and adds the workflow adaptation, so they could plausibly be merged into one PR." + "reason": "Same underlying bug/change: xdist collisions for captured_info/debug logs in testing_utils. #45645 expands the same fix with workflow adjustments, so the pair is plausibly one combined PR." } ] }, @@ -20189,12 +20306,12 @@ "pull_request:46069|pull_request:46154", "pull_request:46073|pull_request:46154" ], - "prepared_review_unit_hash": "170058bec4011e36299fc9fb07d1780fdab2080da90a1f7b65a18c3eba0703c2", + "prepared_review_unit_hash": "ca5e1ce36097f5e07b0742dd72135940ed5b0981d1f6f063a4e1b82f0eec64dd", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 7264, + "serialized_chars": 7263, "estimated_input_tokens": 1816, "estimated_eval_tokens": 3888 }, @@ -20202,7 +20319,7 @@ "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 7264, + "serialized_chars": 7263, "estimated_input_tokens": 1816, "estimated_eval_tokens": 3888 }, @@ -20211,30 +20328,30 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "All three PRs are thematically similar (adding SDPA / attn_implementation dispatch), but they target different model implementations and code paths, so they are not duplicates.", - "confidence": 0.95, + "summary": "These are three separate model-specific PRs adding SDPA/attn_implementation dispatch support and doc badges. They are conceptually related, but they do not look like duplicates of the same concrete code change.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #46154 is the most representative of the cluster topic: it makes the same attn_implementation dispatch change for T5, with accompanying tests and docs, and is the clearest exemplar of this feature work.", + "canonical_pr_reason": "PR #46154 is the best representative of the cluster: it targets the common SDPA dispatch feature, has the broadest implementation/test footprint, and is the clearest exemplar of this family of model-specific attention-dispatch updates.", "best_issue_reason": null, - "best_pr_reason": "PR #46154 is the best single representative because it is a concrete, model-specific implementation of SDPA dispatch and appears more comprehensive than the smaller model-specific patches.", + "best_pr_reason": "PR #46154 is the strongest global representative because it combines the feature addition with more visible test coverage and a clearer, canonical title for the shared theme.", "soft_edge_verdicts": [ { "left": "pull_request:46069", "right": "pull_request:46073", "accept": false, - "reason": "Both are about attention-dispatch support, but one is for ShieldGemma2 image classification and the other is for CTRL text modeling; they touch different model code paths and would not plausibly be merged as one PR." + "reason": "Different model implementations (ShieldGemma2 vision model vs CTRL text model). Same broad attention-dispatch theme, but not the same concrete bug/change and not something to merge into one PR." }, { "left": "pull_request:46069", "right": "pull_request:46154", "accept": false, - "reason": "These are separate model-specific additions (ShieldGemma2 vs T5) with different implementations and tests; they share a feature theme, not the same underlying change." + "reason": "Both add attn_implementation support, but to different model code paths (ShieldGemma2 vs T5). This is a shared feature area, not a duplicate fix." }, { "left": "pull_request:46073", "right": "pull_request:46154", "accept": false, - "reason": "Although both add SDPA dispatch, they apply to different architectures (CTRL vs T5) and different modeling logic, so they are not the same concrete bug or change." + "reason": "CTRL and T5 are distinct architectures with separate attention implementations and tests; they are similar feature PRs, but not the same underlying change." } ] }, diff --git a/analysis/current/manifest.json b/analysis/current/manifest.json index 7236c20b676e2a2c25822162b582588847cdcfc9..a26ae41c334e735e2bfff4a4d9c08f474f7d796b 100644 --- a/analysis/current/manifest.json +++ b/analysis/current/manifest.json @@ -1,8 +1,8 @@ { - "analysis_id": "hybrid-model-20260523t000023z", + "analysis_id": "hybrid-model-20260523t060016z", "archived_artifacts": { - "hybrid": "snapshots/20260523T000023Z/analysis-runs/hybrid-model-20260523t000023z/analysis-report-hybrid.json", - "hybrid_reviews": "snapshots/20260523T000023Z/analysis-runs/hybrid-model-20260523t000023z/analysis-report-hybrid.llm-reviews.json" + "hybrid": "snapshots/20260523T060016Z/analysis-runs/hybrid-model-20260523t060016z/analysis-report-hybrid.json", + "hybrid_reviews": "snapshots/20260523T060016Z/analysis-runs/hybrid-model-20260523t060016z/analysis-report-hybrid.llm-reviews.json" }, "artifacts": { "hybrid": "analysis/current/analysis-report-hybrid.json", @@ -10,9 +10,9 @@ }, "channel": "canonical", "model": null, - "published_at": "2026-05-23T00:08:51Z", + "published_at": "2026-05-23T06:07:40Z", "repo": "huggingface/transformers", "schema_version": 1, - "snapshot_id": "20260523T000023Z", + "snapshot_id": "20260523T060016Z", "variant": "hybrid" } diff --git a/snapshots/20260523T060016Z/analysis-runs/hybrid-model-20260523t060016z/analysis-report-hybrid.json b/snapshots/20260523T060016Z/analysis-runs/hybrid-model-20260523t060016z/analysis-report-hybrid.json new file mode 100644 index 0000000000000000000000000000000000000000..8e57fc703e7061cf74bf1cd3ca4b6b6a29925e9d --- /dev/null +++ b/snapshots/20260523T060016Z/analysis-runs/hybrid-model-20260523t060016z/analysis-report-hybrid.json @@ -0,0 +1,2156 @@ +{ + "schema_version": "1.0", + "repo": "huggingface/transformers", + "snapshot_id": "20260523T060016Z", + "generated_at": "2026-05-23T06:07:39Z", + "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-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-46032-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #46032.", + "status": "open", + "confidence": 0.75, + "canonical_issue_number": 46032, + "canonical_pr_number": 46084, + "issue_numbers": [ + 46032 + ], + "pr_numbers": [ + 46084, + 46098 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target" + ], + "pr_comparisons": [ + { + "left_pr_number": 46084, + "right_pr_number": 46098, + "code_similarity": 0.43, + "size_similarity": 0.5, + "file_overlap": 0.667, + "area_overlap": 0.226, + "patch_similarity": 0.339, + "shared_filenames": [ + "src/transformers/models/mamba2/modeling_mamba2.py", + "tests/models/mamba2/test_modeling_mamba2.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/models/mamba2/modeling_mamba2.py", + "left_ranges": [ + [ + 256, + 262 + ], + [ + 303, + 353 + ], + [ + 461, + 467 + ], + [ + 471, + 491 + ], + [ + 496, + 502 + ], + [ + 599, + 609 + ] + ], + "right_ranges": [ + [ + 255, + 373 + ], + [ + 481, + 495 + ] + ] + }, + { + "filename": "tests/models/mamba2/test_modeling_mamba2.py", + "left_ranges": [ + [ + 206, + 236 + ], + [ + 285, + 322 + ] + ], + "right_ranges": [ + [ + 476, + 542 + ] + ] + } + ] + } + ] + }, + { + "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": 45613, + "issue_numbers": [ + 41115 + ], + "pr_numbers": [ + 41116, + 45613 + ], + "evidence_types": [ + "closing_reference" + ], + "pr_comparisons": [ + { + "left_pr_number": 41116, + "right_pr_number": 45613, + "code_similarity": 0.805, + "size_similarity": 0.885, + "file_overlap": 0.75, + "area_overlap": 0.854, + "patch_similarity": 0.693, + "shared_filenames": [ + "docs/source/en/_toctree.yml", + "docs/source/en/model_doc/minicpm3.md", + "src/transformers/models/auto/modeling_auto.py", + "src/transformers/models/minicpm3/__init__.py", + "src/transformers/models/minicpm3/configuration_minicpm3.py", + "src/transformers/models/minicpm3/modeling_minicpm3.py", + "src/transformers/models/minicpm3/modular_minicpm3.py", + "tests/models/minicpm3/__init__.py", + "tests/models/minicpm3/test_modeling_minicpm3.py" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/_toctree.yml", + "left_ranges": [ + [ + 713, + 720 + ] + ], + "right_ranges": [ + [ + 713, + 720 + ] + ] + }, + { + "filename": "docs/source/en/model_doc/minicpm3.md", + "left_ranges": [ + [ + 1, + 67 + ] + ], + "right_ranges": [ + [ + 1, + 45 + ] + ] + }, + { + "filename": "src/transformers/models/auto/modeling_auto.py", + "left_ranges": [ + [ + 285, + 291 + ], + [ + 700, + 706 + ], + [ + 1298, + 1304 + ] + ], + "right_ranges": [ + [ + 699, + 705 + ], + [ + 1301, + 1307 + ] + ] + }, + { + "filename": "src/transformers/models/minicpm3/__init__.py", + "left_ranges": [ + [ + 1, + 29 + ] + ], + "right_ranges": [ + [ + 1, + 29 + ] + ] + }, + { + "filename": "src/transformers/models/minicpm3/configuration_minicpm3.py", + "left_ranges": [ + [ + 1, + 141 + ] + ], + "right_ranges": [ + [ + 1, + 126 + ] + ] + }, + { + "filename": "src/transformers/models/minicpm3/modeling_minicpm3.py", + "left_ranges": [ + [ + 1, + 544 + ] + ], + "right_ranges": [ + [ + 1, + 522 + ] + ] + }, + { + "filename": "src/transformers/models/minicpm3/modular_minicpm3.py", + "left_ranges": [ + [ + 1, + 444 + ] + ], + "right_ranges": [ + [ + 1, + 342 + ] + ] + }, + { + "filename": "tests/models/minicpm3/test_modeling_minicpm3.py", + "left_ranges": [ + [ + 1, + 109 + ] + ], + "right_ranges": [ + [ + 1, + 136 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-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-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-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": 0.532, + "size_similarity": 0.619, + "file_overlap": 0.5, + "area_overlap": 0.543, + "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": [ + [ + 3529, + 3536 + ], + [ + 3753, + 3787 + ] + ], + "right_ranges": [ + [ + 3525, + 3558 + ], + [ + 3782, + 3788 + ] + ] + }, + { + "filename": "tests/utils/test_testing_utils.py", + "left_ranges": [ + [ + 1, + 86 + ] + ], + "right_ranges": [ + [ + 1, + 114 + ] + ] + } + ] + } + ] + }, + { + "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-46100-2", + "summary": "Cluster of 2 related pull requests linked by soft_similarity.", + "status": "open", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 46100, + "issue_numbers": [], + "pr_numbers": [ + 46100, + 46114 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 46100, + "right_pr_number": 46114, + "code_similarity": 1.0, + "size_similarity": 1.0, + "file_overlap": 1.0, + "area_overlap": 1.0, + "patch_similarity": 1.0, + "shared_filenames": [ + "tests/models/vit/test_modeling_vit.py" + ], + "shared_file_areas": [ + { + "filename": "tests/models/vit/test_modeling_vit.py", + "left_ranges": [ + [ + 216, + 226 + ] + ], + "right_ranges": [ + [ + 216, + 226 + ] + ] + } + ] + } + ] + }, + { + "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": 45613, + "duplicate_pr_numbers": [ + 41116 + ], + "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-46032-3", + "canonical_pr_number": 46084, + "duplicate_pr_numbers": [ + 46098 + ], + "target_issue_number": 46032, + "reason": "PRs in cluster-46032-3 are treated as duplicates because they converge on issue #46032 with closing_reference, shared_issue_target evidence." + }, + { + "cluster_id": "cluster-46100-2", + "canonical_pr_number": 46100, + "duplicate_pr_numbers": [ + 46114 + ], + "target_issue_number": null, + "reason": "PRs in cluster-46100-2 are treated as duplicates because they share soft_similarity evidence." + } + ], + "best_issue": { + "cluster_id": "cluster-45647-3", + "issue_number": 45647, + "reason": "Issue #45647 is the strongest global issue candidate because it is open, belongs to a cluster with 3 artifacts, and carries 12 discussion comments plus 8 inbound references.", + "score": 162.0 + }, + "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/20260523T060016Z/analysis-runs/hybrid-model-20260523t060016z/analysis-report-hybrid.llm-reviews.json b/snapshots/20260523T060016Z/analysis-runs/hybrid-model-20260523t060016z/analysis-report-hybrid.llm-reviews.json new file mode 100644 index 0000000000000000000000000000000000000000..b8b2133a3d711f8b82f0c66d9dd0c1ece8705089 --- /dev/null +++ b/snapshots/20260523T060016Z/analysis-runs/hybrid-model-20260523t060016z/analysis-report-hybrid.llm-reviews.json @@ -0,0 +1,20366 @@ +{ + "schema_version": "1.0", + "repo": "huggingface/transformers", + "snapshot_id": "20260523T060016Z", + "generated_at": "2026-05-23T06:07:39Z", + "model": "gpt-5.4-mini?service_tier=flex", + "reviews": [ + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 1, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "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:43572", + "issue:43596", + "issue:43638", + "issue:43716", + "issue:43827", + "issue:43828", + "issue:43866", + "issue:43901", + "issue:43975", + "issue:44292", + "issue:44322", + "issue:44457", + "issue:44509", + "issue:44512", + "issue:44560", + "issue:44805", + "issue:44841", + "issue:44863", + "issue:44918", + "issue:45081", + "issue:45161", + "issue:45237", + "issue:45464", + "issue:45507", + "issue:45685", + "issue:45812", + "issue:45823", + "issue:45859", + "issue:45874", + "issue:45901", + "issue:45902", + "issue:45920", + "issue:46015", + "issue:46165" + ], + "soft_pairs": [ + "issue:44322|issue:45464", + "issue:45237|issue:45685", + "issue:43716|issue:45237", + "issue:43828|issue:45237", + "issue:45237|issue:45507", + "issue:44292|issue:45237", + "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:44805|issue:45823", + "issue:44918|issue:46165", + "issue:43638|issue:44805", + "issue:43638|issue:46165", + "issue:43596|issue:44805", + "issue:44805|issue:45161", + "issue:44560|issue:46165", + "issue:43596|issue:46165", + "issue:44805|issue:44918", + "issue:43572|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:34928|issue:44805", + "issue:24643|issue:30277", + "issue:45859|issue:45902", + "issue:43638|issue:45161", + "issue:34928|issue:36331", + "issue:43596|issue:45161", + "issue:43716|issue:45507", + "issue:39290|issue:45081", + "issue:45823|issue:45874", + "issue:43425|issue:44292", + "issue:34928|issue:41720", + "issue:43828|issue:45161", + "issue:44918|issue:45161", + "issue:43716|issue:45685", + "issue:43828|issue:44292", + "issue:43541|issue:43716", + "issue:43716|issue:44292", + "issue:44292|issue:44918", + "issue:45507|issue:45685", + "issue:43039|issue:44918", + "issue:41762|issue:43638", + "issue:43827|issue:44512", + "issue:43716|issue:44560", + "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", + "issue:41762|issue:43596", + "issue:39290|issue:41720", + "issue:43901|issue:44512", + "issue:45237|issue:45901", + "issue:44841|issue:45823", + "issue:43039|issue:44560", + "issue:43975|issue:46015", + "issue:45812|issue:45920", + "issue:44863|issue:46015", + "issue:43866|issue:44863", + "issue:44292|issue:46015" + ], + "prepared_review_unit_hash": "212bc5f396c731663f9121863d11e41123380f1b65f665edef2960fb1b7f20f9", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 69, + "serialized_chars": 48401, + "estimated_input_tokens": 12101, + "estimated_eval_tokens": 24458 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly noisy and spans many unrelated bugs/docs issues. Only a couple of soft edges look like true duplicates: one DeepSpeed ZeRO-3 Bert init crash pair and one AutoTokenizer token-ID regression pair.", + "confidence": 0.61, + "canonical_issue_reason": "issue:45920 is the broadest and most central issue in the tokenizer-regression subcluster; it covers the same AutoTokenizer wrong-token-ID problem across multiple model families, while issue:45812 is a narrower Granite-only report.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45920 best represents the strongest duplicate subcluster because it describes the v5 AutoTokenizer regression in the most general terms and is the likely umbrella for the narrower Granite report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44322", + "right": "issue:45464", + "accept": false, + "reason": "Different models and failures: missing config attribute vs streaming chat/completions error." + }, + { + "left": "issue:45237", + "right": "issue:45685", + "accept": false, + "reason": "Different platforms and bugs: AMD GPU failure vs MPS histogram op limitation." + }, + { + "left": "issue:43716", + "right": "issue:45237", + "accept": false, + "reason": "Different code paths: image preprocessor dtype mismatch vs GPT-OSS AMD GPU issue." + }, + { + "left": "issue:43828", + "right": "issue:45237", + "accept": false, + "reason": "Both are dtype/device problems, but they affect different models and execution paths." + }, + { + "left": "issue:45237", + "right": "issue:45507", + "accept": false, + "reason": "Different model families and error modes; not the same bug." + }, + { + "left": "issue:44292", + "right": "issue:45237", + "accept": false, + "reason": "Separate inference/runtime failures with no clear shared concrete cause." + }, + { + "left": "issue:30277", + "right": "issue:35545", + "accept": false, + "reason": "Deepspeed collective mismatch and ONNX export are unrelated." + }, + { + "left": "issue:35545", + "right": "issue:42915", + "accept": false, + "reason": "Different model/export issue vs FP8 config failure." + }, + { + "left": "issue:35545", + "right": "issue:41720", + "accept": false, + "reason": "ONNX export error and auto device-map cuda assert are unrelated." + }, + { + "left": "issue:45901", + "right": "issue:46015", + "accept": false, + "reason": "Table QA crash vs outdated RAG docs; different artifacts and causes." + }, + { + "left": "issue:44509", + "right": "issue:44512", + "accept": false, + "reason": "Both are docs cleanup, but they reference different removed features and are not the same issue." + }, + { + "left": "issue:35545", + "right": "issue:36010", + "accept": false, + "reason": "Export-to-ONNX and GenerationMixin import failure are separate regressions." + }, + { + "left": "issue:44805", + "right": "issue:45823", + "accept": false, + "reason": "Different symptoms: mask-shape IndexError vs Gemma4 device mismatch." + }, + { + "left": "issue:44918", + "right": "issue:46165", + "accept": false, + "reason": "TRL SFT embedding unpacking and resume/trust-remote-code training are unrelated." + }, + { + "left": "issue:43638", + "right": "issue:44805", + "accept": false, + "reason": "Both mention training/ZeRO-3 style failures, but the concrete errors and model paths differ." + }, + { + "left": "issue:43638", + "right": "issue:46165", + "accept": false, + "reason": "Different training scenarios and error mechanisms." + }, + { + "left": "issue:43596", + "right": "issue:44805", + "accept": false, + "reason": "Both are IndexError reports, but one is ZeRO-3 Bert init and the other is a different shape mismatch." + }, + { + "left": "issue:44805", + "right": "issue:45161", + "accept": false, + "reason": "Generic tensor-shape failure vs GPT-OSS MoE TP issue are not the same bug." + }, + { + "left": "issue:44560", + "right": "issue:46165", + "accept": false, + "reason": "Video StopIteration and model-resume/trust-remote-code are unrelated." + }, + { + "left": "issue:43596", + "right": "issue:46165", + "accept": false, + "reason": "Different training/loading failures with no shared concrete code path." + }, + { + "left": "issue:44805", + "right": "issue:44918", + "accept": false, + "reason": "Shape mismatch error and Qwen3.5 embedding unpacking are different issues." + }, + { + "left": "issue:43572", + "right": "issue:46165", + "accept": false, + "reason": "StableLm pad_token_idx regression is unrelated to training resume problems." + }, + { + "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": "Deepspeed collective mismatch and Bert ZeRO-3 index error are different failures." + }, + { + "left": "issue:43596", + "right": "issue:43638", + "accept": true, + "reason": "Same DeepSpeed ZeRO-3/BertModel loading crash with index-0-out-of-bounds on zero-sized tensors; the second report is the same bug on a narrower Bert variant." + }, + { + "left": "issue:43039", + "right": "issue:45161", + "accept": false, + "reason": "Liger kernel cross-entropy call and GPT-OSS MoE TP issue are unrelated." + }, + { + "left": "issue:43975", + "right": "issue:44457", + "accept": false, + "reason": "Detokenization regression and LoRA merge/save mismatch are different problems." + }, + { + "left": "issue:44292", + "right": "issue:44457", + "accept": false, + "reason": "NVFP4 runtime error and LoRA save/reload mismatch are unrelated." + }, + { + "left": "issue:34928", + "right": "issue:44805", + "accept": false, + "reason": "Activation-checkpointing/FSDP recompute mismatch is not the same as the mask-shape IndexError." + }, + { + "left": "issue:24643", + "right": "issue:30277", + "accept": false, + "reason": "DeepSpeed weight-dimension error and collective mismatch are different bugs." + }, + { + "left": "issue:45859", + "right": "issue:45902", + "accept": false, + "reason": "Same class, but one is CPU-offload incompatibility and the other is uninitialized-memory on materialization; not clearly the same fix." + }, + { + "left": "issue:43638", + "right": "issue:45161", + "accept": false, + "reason": "Bert ZeRO-3 init crash and GPT-OSS TP issue are unrelated." + }, + { + "left": "issue:34928", + "right": "issue:36331", + "accept": false, + "reason": "FSDP checkpointing tensor-size mismatch vs custom trainer compute_loss signature breakage are different issues." + }, + { + "left": "issue:43596", + "right": "issue:45161", + "accept": false, + "reason": "Different models, different failure modes." + }, + { + "left": "issue:43716", + "right": "issue:45507", + "accept": false, + "reason": "Dtype mismatch in Mistral-3 image preprocessing vs GraniteMoEHybrid invalid method." + }, + { + "left": "issue:39290", + "right": "issue:45081", + "accept": false, + "reason": "Gemma3/vLLM config attribute error and Mistral tokenizer regex patch crash are unrelated." + }, + { + "left": "issue:45823", + "right": "issue:45874", + "accept": false, + "reason": "Both are Gemma4-related, but one is device placement and the other is expensive reverse-embedding lookup; not the same bug." + }, + { + "left": "issue:43425", + "right": "issue:44292", + "accept": false, + "reason": "Torch version compatibility issue vs Qwen-3 runtime error are unrelated." + }, + { + "left": "issue:34928", + "right": "issue:41720", + "accept": false, + "reason": "FSDP activation checkpointing mismatch and Qwen3 auto device-map cuda assert are unrelated." + }, + { + "left": "issue:43828", + "right": "issue:45161", + "accept": false, + "reason": "Phi-tiny-MoE autocast dtype mismatch is not the same as GPT-OSS TP failure." + }, + { + "left": "issue:44918", + "right": "issue:45161", + "accept": false, + "reason": "TRI/SFT embedding unpacking and GPT-OSS TP failure are different code paths." + }, + { + "left": "issue:43716", + "right": "issue:45685", + "accept": false, + "reason": "Different model/platform bugs." + }, + { + "left": "issue:43828", + "right": "issue:44292", + "accept": false, + "reason": "Both involve runtime errors, but on different models and mechanisms." + }, + { + "left": "issue:43541", + "right": "issue:43716", + "accept": false, + "reason": "Mixtral grouped_mm tracing error and Mistral-3 dtype mismatch are different failures." + }, + { + "left": "issue:43716", + "right": "issue:44292", + "accept": false, + "reason": "Mistral-3 preprocessing dtype mismatch vs Qwen NVFP4 runtime error are unrelated." + }, + { + "left": "issue:44292", + "right": "issue:44918", + "accept": false, + "reason": "Qwen NVFP4 runtime failure and Qwen3.5 embedding unpacking are different bugs." + }, + { + "left": "issue:45507", + "right": "issue:45685", + "accept": false, + "reason": "Invalid method call in GraniteMoEHybrid vs MPS histogram int limitation are unrelated." + }, + { + "left": "issue:43039", + "right": "issue:44918", + "accept": false, + "reason": "Liger kernel cross-entropy call and TRL embedding unpacking are unrelated." + }, + { + "left": "issue:41762", + "right": "issue:43638", + "accept": false, + "reason": "Both mention ZeRO-3 loading issues, but Gemma3 and Bert failures are different concrete bugs." + }, + { + "left": "issue:43827", + "right": "issue:44512", + "accept": false, + "reason": "Both are docs issues after v5 removals, but they concern different removed features and are separate cleanups." + }, + { + "left": "issue:43716", + "right": "issue:44560", + "accept": false, + "reason": "Different models and error surfaces: image dtype mismatch vs video StopIteration." + }, + { + "left": "issue:43828", + "right": "issue:45685", + "accept": false, + "reason": "Dtype mismatch in autocast and MPS histogram int failure are different issues." + }, + { + "left": "issue:44918", + "right": "issue:45685", + "accept": false, + "reason": "TRl SFT embedding unpacking and MPS histogram limitation are unrelated." + }, + { + "left": "issue:43039", + "right": "issue:43828", + "accept": false, + "reason": "Liger-kernel cross-entropy usage and Phi-tiny-MoE autocast mismatch are different bugs." + }, + { + "left": "issue:30277", + "right": "issue:36331", + "accept": false, + "reason": "Collective mismatch and compute_loss signature breakage are unrelated." + }, + { + "left": "issue:34928", + "right": "issue:36010", + "accept": false, + "reason": "Checkpointing/FSDP mismatch and GenerationMixin import error are unrelated." + }, + { + "left": "issue:45237", + "right": "issue:46015", + "accept": false, + "reason": "GPT-OSS AMD GPU failure and outdated RAG docs are unrelated." + }, + { + "left": "issue:43039", + "right": "issue:44560", + "accept": false, + "reason": "Liger kernel cross-entropy call and Qwen3-vl video StopIteration are unrelated." + }, + { + "left": "issue:41762", + "right": "issue:43596", + "accept": false, + "reason": "Both are ZeRO-3-related, but Gemma3 and Bert loading crashes are distinct bugs." + }, + { + "left": "issue:39290", + "right": "issue:41720", + "accept": false, + "reason": "vLLM Gemma3 config attribute error and Qwen3 device-map cuda assert are unrelated." + }, + { + "left": "issue:43901", + "right": "issue:44512", + "accept": false, + "reason": "TextClassificationPipeline docs and transformers run-command docs are separate documentation issues." + }, + { + "left": "issue:45237", + "right": "issue:45901", + "accept": false, + "reason": "Different model/runtime issue vs table-question-answering crash." + }, + { + "left": "issue:44841", + "right": "issue:45823", + "accept": false, + "reason": "Voxtral processor failure and Gemma4 device mismatch are unrelated." + }, + { + "left": "issue:43975", + "right": "issue:46015", + "accept": false, + "reason": "Detokenization regression and outdated RAG docs are unrelated." + }, + { + "left": "issue:45812", + "right": "issue:45920", + "accept": true, + "reason": "Both report the same v5 AutoTokenizer wrong-token-ID regression across model families; 45920 is the broader umbrella report." + }, + { + "left": "issue:44863", + "right": "issue:46015", + "accept": false, + "reason": "NemotronH checkpoint loading failure and RAG docs are unrelated." + }, + { + "left": "issue:43866", + "right": "issue:44863", + "accept": false, + "reason": "Corrupted checkpoint report and NemotronH loading implementation bug are different issues." + }, + { + "left": "issue:44292", + "right": "issue:46015", + "accept": false, + "reason": "Qwen NVFP4 runtime error and RAG docs cleanup are unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The analysis is grounded in the packet and stays conservative. The two accepted duplicate pairs are well supported by the issue titles and similarity, and the broader clustering summary does not overstate the evidence. The canonical issue choice (45920 over 45812) is also reasonable given the broader v5 AutoTokenizer regression wording." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 2, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:36010", + "issue:39186", + "issue:39290", + "issue:40990", + "issue:41093", + "issue:41762", + "issue:43054", + "issue:43296", + "issue:43366", + "issue:43404", + "issue:43425", + "issue:43531", + "issue:43541", + "issue:43572", + "issue:43596", + "issue:43638", + "issue:43645", + "issue:43716", + "issue:43792", + "issue:43828", + "issue:43854", + "issue:43866", + "issue:43975", + "issue:43994", + "issue:44291", + "issue:44292", + "issue:44387", + "issue:44589", + "issue:44841", + "issue:44918", + "issue:44960", + "issue:45005", + "issue:45020", + "issue:45070", + "issue:45084", + "issue:45161", + "issue:45362", + "issue:45399", + "issue:45464", + "issue:45542", + "issue:45632", + "issue:45685", + "issue:45698", + "issue:45759", + "issue:45800", + "issue:45823", + "issue:45901", + "issue:45902" + ], + "soft_pairs": [ + "issue:39290|issue:43531", + "issue:43854|issue:43866", + "issue:43792|issue:45901", + "issue:41093|issue:43638", + "issue:43425|issue:45070", + "issue:43596|issue:45800", + "issue:43975|issue:45901", + "issue:43792|issue:45070", + "issue:45362|issue:45464", + "issue:44292|issue:45901", + "issue:39290|issue:45823", + "issue:45464|issue:45685", + "issue:43404|issue:43866", + "issue:43541|issue:45685", + "issue:43541|issue:43828", + "issue:44841|issue:45084", + "issue:36010|issue:39290", + "issue:45800|issue:45823", + "issue:41093|issue:41762", + "issue:43572|issue:45070", + "issue:43296|issue:43366", + "issue:43541|issue:45161", + "issue:44387|issue:45005", + "issue:43828|issue:45464", + "issue:44291|issue:45070", + "issue:43366|issue:45464", + "issue:45020|issue:45800", + "issue:43531|issue:45070", + "issue:43541|issue:44589", + "issue:41093|issue:43596", + "issue:43054|issue:43994", + "issue:45542|issue:45800", + "issue:45759|issue:45902", + "issue:39186|issue:44918", + "issue:43716|issue:44918", + "issue:40990|issue:44960", + "issue:45632|issue:45698", + "issue:43645|issue:45399", + "issue:43404|issue:45005" + ], + "prepared_review_unit_hash": "0e5958b90eb8bb04775c57b53e4ef61bb8800dade8c226df8ee9c2a37c10fb87", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 39, + "serialized_chars": 39124, + "estimated_input_tokens": 9781, + "estimated_eval_tokens": 19818 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose collection of unrelated model-loading, config, dtype, and runtime issues. Only the SigLIP2 pair looks like a credible duplicate; the custom-module cache-path pair is adjacent but not the same bug.", + "confidence": 0.73, + "canonical_issue_reason": "issue:45698 is the best canonical issue: it is open, user-facing, and has the clearest actionable save_pretrained/from_pretrained custom-code loading regression with broader discussion than the other nearby loading reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45698 is the strongest standalone triage target because it has a concrete reproduction, current open status, and a focused failure mode that is easier to act on than the more diffuse runtime reports in the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:39290", + "right": "issue:43531", + "accept": false, + "reason": "Same broad sliding-window theme, but different models and different concrete failures (Gemma3/vLLM attribute access vs Qwen3-MoE behavior)." + }, + { + "left": "issue:43854", + "right": "issue:43866", + "accept": false, + "reason": "A model-loading test failure and a corrupted checkpoint report are not the same underlying bug." + }, + { + "left": "issue:43792", + "right": "issue:45901", + "accept": false, + "reason": "Whisper inference failure and table-question-answering crash are unrelated code paths." + }, + { + "left": "issue:41093", + "right": "issue:43638", + "accept": false, + "reason": "Both mention index errors under ZeRO-3/BERT, but the concrete symptoms and setups differ too much to merge confidently." + }, + { + "left": "issue:43425", + "right": "issue:45070", + "accept": false, + "reason": "Torch compatibility and a PretrainedConfig/pydantic regression are separate issues." + }, + { + "left": "issue:43596", + "right": "issue:45800", + "accept": false, + "reason": "A ZeRO-3 BERT index error and a Torch/Transformers incompatibility are different bugs." + }, + { + "left": "issue:43975", + "right": "issue:45901", + "accept": false, + "reason": "Incorrect detokenization and table QA crashes are different failure modes." + }, + { + "left": "issue:43792", + "right": "issue:45070", + "accept": false, + "reason": "Whisper runtime failure is unrelated to the PretrainedConfig/pydantic regression." + }, + { + "left": "issue:45362", + "right": "issue:45464", + "accept": false, + "reason": "Both involve Qwen3.5, but one is a chat-model crash and the other is a streaming API failure; not the same code path." + }, + { + "left": "issue:44292", + "right": "issue:45901", + "accept": false, + "reason": "Qwen-3-8B-NVFP4 runtime error and table-question-answering crash are unrelated." + }, + { + "left": "issue:39290", + "right": "issue:45823", + "accept": false, + "reason": "Both are Gemma-family issues, but missing sliding_window_pattern and a device mismatch are distinct bugs." + }, + { + "left": "issue:45464", + "right": "issue:45685", + "accept": false, + "reason": "Qwen3.5 streaming inference and an MPS histogram kernel error are unrelated." + }, + { + "left": "issue:43404", + "right": "issue:43866", + "accept": false, + "reason": "Tied lm_head weights in Mistral3 and a corrupted Ovis2 checkpoint are different bugs." + }, + { + "left": "issue:43541", + "right": "issue:45685", + "accept": false, + "reason": "Grouped_mm tracing failure and MPS int histogram limitation are separate runtime issues." + }, + { + "left": "issue:43541", + "right": "issue:43828", + "accept": false, + "reason": "Both are MoE-related runtime errors, but grouped_mm tracing and autocast dtype mismatch are different concrete bugs." + }, + { + "left": "issue:44841", + "right": "issue:45084", + "accept": false, + "reason": "Processor failure for Voxtral and a compile/template-node TypeError are unrelated." + }, + { + "left": "issue:36010", + "right": "issue:39290", + "accept": false, + "reason": "ImportError for GenerationMixin and a Gemma3 sliding_window_pattern issue are unrelated." + }, + { + "left": "issue:45800", + "right": "issue:45823", + "accept": false, + "reason": "Torch/Transformers incompatibility and Gemma4 device mismatch are distinct problems." + }, + { + "left": "issue:41093", + "right": "issue:41762", + "accept": false, + "reason": "Both are ZeRO-3 index failures, but the error signatures and affected models differ enough to avoid merging." + }, + { + "left": "issue:43572", + "right": "issue:45070", + "accept": false, + "reason": "Missing pad_token_idx in StableLmConfig is unrelated to the PretrainedConfig/pydantic regression." + }, + { + "left": "issue:43296", + "right": "issue:43366", + "accept": false, + "reason": "PaddleOCR-VL loading in vLLM and GGUF gpt-oss support are different requests/bugs." + }, + { + "left": "issue:43541", + "right": "issue:45161", + "accept": false, + "reason": "Both touch MoE/GPT-OSS, but one is a grouped_mm tracing bug and the other is a tensor-parallel support problem." + }, + { + "left": "issue:44387", + "right": "issue:45005", + "accept": false, + "reason": "int4 memory growth/OOM and tied-weight translation-model issues are unrelated." + }, + { + "left": "issue:43828", + "right": "issue:45464", + "accept": false, + "reason": "dtype mismatch under autocast and streaming chat/completions failure are not the same bug." + }, + { + "left": "issue:44291", + "right": "issue:45070", + "accept": false, + "reason": "Both are v5 init/config regressions, but unexpected _is_hf_initialized and pydantic field handling are different code paths." + }, + { + "left": "issue:43366", + "right": "issue:45464", + "accept": false, + "reason": "GPT-OSS GGUF support and Qwen3.5 streaming inference are unrelated." + }, + { + "left": "issue:45020", + "right": "issue:45800", + "accept": false, + "reason": "A broad remote_code regression and a Torch/Transformers version incompatibility are different issues." + }, + { + "left": "issue:43531", + "right": "issue:45070", + "accept": false, + "reason": "Sliding_window handling in Qwen3-MoE is unrelated to the PretrainedConfig/pydantic change." + }, + { + "left": "issue:43541", + "right": "issue:44589", + "accept": false, + "reason": "Grouped_mm tracing failure and missing Float8 storage are different runtime/toolchain bugs." + }, + { + "left": "issue:41093", + "right": "issue:43596", + "accept": false, + "reason": "Both are index-out-of-bounds reports under ZeRO-3, but they affect different setups and error signatures." + }, + { + "left": "issue:43054", + "right": "issue:43994", + "accept": true, + "reason": "Both report broken SigLIP2 behavior on the same model family; bad text embeddings and nonsensical AutoModel/pipeline outputs plausibly share one underlying regression." + }, + { + "left": "issue:45542", + "right": "issue:45800", + "accept": false, + "reason": "TensorBoard-without-TF backend handling is unrelated to Torch/Transformers compatibility." + }, + { + "left": "issue:45759", + "right": "issue:45902", + "accept": false, + "reason": "Unwrapping text_config for composite Qwen configs and uninitialized inv_freq after to_empty are different bugs." + }, + { + "left": "issue:39186", + "right": "issue:44918", + "accept": false, + "reason": "FSDP 2-D weight error and Qwen3.5 embedding unpacking/trl issue are unrelated." + }, + { + "left": "issue:43716", + "right": "issue:44918", + "accept": false, + "reason": "Gemma3 image-preprocessor dtype mismatch and Qwen3.5 embedding unpacking failure are different issues." + }, + { + "left": "issue:40990", + "right": "issue:44960", + "accept": false, + "reason": "GPT-OSS perplexity regression and a GLM5 issue are unrelated." + }, + { + "left": "issue:45632", + "right": "issue:45698", + "accept": false, + "reason": "These are adjacent custom-code loading bugs, but one is a cache-path collision for local models sharing a leaf directory name and the other is wrong module resolution after save_pretrained; not clearly one mergeable fix." + }, + { + "left": "issue:43645", + "right": "issue:45399", + "accept": false, + "reason": "Jupyter custom-model initialization breakage and flash-attn fallback blocking are separate regressions." + }, + { + "left": "issue:43404", + "right": "issue:45005", + "accept": false, + "reason": "Tied-weight handling in a specific Mistral3 multimodal model and broader translation-model tied-weight issues are related in theme but not the same concrete bug." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The cluster summary is mostly reasonable, but it overstates the evidence for at least one soft-edge merge. In particular, accepting issue:45632 vs issue:45698 as the same bug is not well grounded from the packet: one is a trust_remote_code cache-path collision for local models sharing a leaf directory name, while the other is from_pretrained loading the wrong custom module after save_pretrained. Those look adjacent, but not clearly one mergeable fix. Also, phrases like \"broad impact\" and \"likely share one underlying regression\" are a bit stronger than the packet supports." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 3, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:36331", + "issue:38175", + "issue:39692", + "issue:41553", + "issue:41628", + "issue:41762", + "issue:42915", + "issue:42947", + "issue:43054", + "issue:43493", + "issue:43643", + "issue:43644", + "issue:43782", + "issue:43828", + "issue:43856", + "issue:43866", + "issue:43872", + "issue:43873", + "issue:43881", + "issue:43883", + "issue:43950", + "issue:43975", + "issue:44292", + "issue:44360", + "issue:44387", + "issue:44403", + "issue:44451", + "issue:44488", + "issue:44492", + "issue:44512", + "issue:44530", + "issue:44534", + "issue:44589", + "issue:44863", + "issue:44910", + "issue:44918", + "issue:44928", + "issue:44960", + "issue:45005", + "issue:45020", + "issue:45237", + "issue:45399", + "issue:45507", + "issue:45663", + "issue:45701", + "issue:45800", + "issue:45859", + "issue:45902" + ], + "soft_pairs": [ + "issue:44530|issue:45005", + "issue:42915|issue:45399", + "issue:43873|issue:45005", + "issue:44863|issue:45005", + "issue:44360|issue:44512", + "issue:44918|issue:45507", + "issue:44492|issue:44512", + "issue:43975|issue:44292", + "issue:44292|issue:45507", + "issue:43950|issue:44534", + "issue:43644|issue:43950", + "issue:39692|issue:43054", + "issue:43828|issue:44928", + "issue:36331|issue:41762", + "issue:41628|issue:45800", + "issue:43950|issue:45902", + "issue:43872|issue:44589", + "issue:43866|issue:44403", + "issue:43054|issue:45701", + "issue:41553|issue:43883", + "issue:38175|issue:43054", + "issue:44910|issue:45859", + "issue:45663|issue:45902", + "issue:41553|issue:43881", + "issue:42947|issue:44387", + "issue:43054|issue:43493", + "issue:44403|issue:44451", + "issue:43975|issue:45237", + "issue:43643|issue:45020", + "issue:44403|issue:44488", + "issue:42947|issue:43856", + "issue:41762|issue:42915", + "issue:43782|issue:44960", + "issue:43881|issue:43883" + ], + "prepared_review_unit_hash": "1b6c635565c820ed5d373d99e857494b922033147f4e4d92fd3d4adab909d236", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 34, + "serialized_chars": 37512, + "estimated_input_tokens": 9378, + "estimated_eval_tokens": 19012 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a grab-bag of unrelated Transformers issues. The only clear duplicate-style match is the v5 non-persistent buffer regression pair; the rest are different bugs, models, or docs problems. No PRs were provided.", + "confidence": 0.93, + "canonical_issue_reason": "Issue #43950 is the best canonical because it most clearly states the non-persistent-buffer regression in v5 and is the most specific, root-cause-oriented report among the duplicate-like pair.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #43950 is the strongest overall representative of the only true duplicate cluster here: it is precise, broad enough to subsume the related report, and directly describes the underlying regression.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44530", + "right": "issue:45005", + "accept": false, + "reason": "Different bugs: Qwen3.5 cache crash vs. tied-weights behavior in translation models." + }, + { + "left": "issue:42915", + "right": "issue:45399", + "accept": false, + "reason": "FP8 config failure and flash-attn2 fallback gating are separate issues." + }, + { + "left": "issue:43873", + "right": "issue:45005", + "accept": false, + "reason": "Quantization/offload behavior is unrelated to tied-weight handling." + }, + { + "left": "issue:44863", + "right": "issue:45005", + "accept": false, + "reason": "NemotronH checkpoint loading failure is not the same as translation-model tied weights." + }, + { + "left": "issue:44360", + "right": "issue:44512", + "accept": false, + "reason": "DSA indexer/ReLU discussion and a docs typo are not the same issue." + }, + { + "left": "issue:44918", + "right": "issue:45507", + "accept": false, + "reason": "Qwen3.5 SFT embedding unpacking and GraniteMoEHybrid invalid method are different code paths." + }, + { + "left": "issue:44492", + "right": "issue:44512", + "accept": false, + "reason": "A cache-strategy typo is not the same as stale run-command docs." + }, + { + "left": "issue:43975", + "right": "issue:44292", + "accept": false, + "reason": "Detokenization regression and NVFP4 runtime error are unrelated." + }, + { + "left": "issue:44292", + "right": "issue:45507", + "accept": false, + "reason": "Qwen-3-8B-NVFP4 failure and GraniteMoEHybrid invalid method are different bugs." + }, + { + "left": "issue:43950", + "right": "issue:44534", + "accept": true, + "reason": "Both describe the same v5 regression where non-persistent buffers are mishandled/corrupted with junk values." + }, + { + "left": "issue:43644", + "right": "issue:43950", + "accept": false, + "reason": "AutoConfig trust_remote_code field loss is a different bug from non-persistent buffer corruption." + }, + { + "left": "issue:39692", + "right": "issue:43054", + "accept": false, + "reason": "SigLIP2 doc/example errors and poor text embeddings are separate problems." + }, + { + "left": "issue:43828", + "right": "issue:44928", + "accept": false, + "reason": "Autocast dtype mismatch is not the same as the Qwen3.5 SDPA/NaN issue." + }, + { + "left": "issue:36331", + "right": "issue:41762", + "accept": false, + "reason": "Trainer kwarg TypeError and DeepSpeed ZeRO-3 loading failure are unrelated." + }, + { + "left": "issue:41628", + "right": "issue:45800", + "accept": false, + "reason": "Import error for AutoImageProcessor is different from torch/transformers version incompatibility." + }, + { + "left": "issue:43950", + "right": "issue:45902", + "accept": false, + "reason": "Both involve non-persistent buffers, but this looks like a different Qwen3_5Moe materialization path rather than the same concrete bug." + }, + { + "left": "issue:43872", + "right": "issue:44589", + "accept": false, + "reason": "bitsandbytes Int8Params incompatibility and missing Float8 storage are different failures." + }, + { + "left": "issue:43866", + "right": "issue:44403", + "accept": false, + "reason": "Corrupted checkpoint vs. generic loading noise are not the same underlying issue." + }, + { + "left": "issue:43054", + "right": "issue:45701", + "accept": false, + "reason": "SigLIP2 embedding quality regression and tokenization changes are distinct." + }, + { + "left": "issue:41553", + "right": "issue:43883", + "accept": false, + "reason": "Bad AutoTokenizer error for Voxtral and Molmo tied-weights attribute error are unrelated." + }, + { + "left": "issue:38175", + "right": "issue:43054", + "accept": false, + "reason": "Unexpected zero probabilities and worse text embeddings in SigLIP2 are different symptoms." + }, + { + "left": "issue:44910", + "right": "issue:45859", + "accept": false, + "reason": "Flash-attn illegal memory access on Qwen3.5 and CPU-offload incompatibility are different code paths." + }, + { + "left": "issue:45663", + "right": "issue:45902", + "accept": false, + "reason": "Gemma4 FSDP2 shared_kv_states failure is unrelated to Qwen rotary embedding uninitialized memory." + }, + { + "left": "issue:41553", + "right": "issue:43881", + "accept": false, + "reason": "Bad Voxtral tokenizer loading error and glm-4v-9b load failure are not the same bug." + }, + { + "left": "issue:42947", + "right": "issue:44387", + "accept": false, + "reason": "LoRA gradient checkpointing ineffectiveness and int4 quantization OOM are different issues." + }, + { + "left": "issue:43054", + "right": "issue:43493", + "accept": false, + "reason": "SigLIP2 embedding degradation and HF-vs-JAX discrepancy are related only by model family, not the same bug." + }, + { + "left": "issue:44403", + "right": "issue:44451", + "accept": false, + "reason": "Generic loading noise and ScandiBERT load failure are not the same underlying issue." + }, + { + "left": "issue:43975", + "right": "issue:45237", + "accept": false, + "reason": "Detokenization regression and AMD GPU loading failure are unrelated." + }, + { + "left": "issue:43643", + "right": "issue:45020", + "accept": false, + "reason": "Missing fields with trust_remote_code and broad remote_code breakage are too broad to treat as the same concrete bug." + }, + { + "left": "issue:44403", + "right": "issue:44488", + "accept": false, + "reason": "Generic loading noise and sleng-bert load failure are different reports." + }, + { + "left": "issue:42947", + "right": "issue:43856", + "accept": false, + "reason": "LoRA checkpointing inefficiency and Qwen3 MoE memory usage are different performance issues." + }, + { + "left": "issue:41762", + "right": "issue:42915", + "accept": false, + "reason": "Gemma3 ZeRO-3 loading failure and Qwen3Moe FP8 failure are unrelated." + }, + { + "left": "issue:43782", + "right": "issue:44960", + "accept": false, + "reason": "Qwen3VL weight_only error and GLM5 are different model/load failures." + }, + { + "left": "issue:43881", + "right": "issue:43883", + "accept": false, + "reason": "These are separate model-loading failures with different error signatures and root causes." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative on the soft-edge calls. The only accepted duplicate pair, issue:43950 with issue:44534, is well supported by the titles and packet text. No PR-pair mergeability concerns apply here because no PRs were provided." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 4, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:24643", + "issue:30064", + "issue:36331", + "issue:41093", + "issue:41553", + "issue:42491", + "issue:42915", + "issue:42947", + "issue:43257", + "issue:43366", + "issue:43404", + "issue:43638", + "issue:43742", + "issue:43792", + "issue:43824", + "issue:43827", + "issue:43828", + "issue:43854", + "issue:43866", + "issue:43872", + "issue:43940", + "issue:43950", + "issue:44220", + "issue:44291", + "issue:44368", + "issue:44479", + "issue:44488", + "issue:44509", + "issue:44530", + "issue:44560", + "issue:44661", + "issue:44805", + "issue:44863", + "issue:44928", + "issue:44991", + "issue:45020", + "issue:45081", + "issue:45305", + "issue:45313", + "issue:45325", + "issue:45356", + "issue:45399", + "issue:45538", + "issue:45588", + "issue:45685", + "issue:45823", + "issue:45902", + "issue:45923" + ], + "soft_pairs": [ + "issue:45685|issue:45823", + "issue:43824|issue:43872", + "issue:44991|issue:45399", + "issue:45020|issue:45823", + "issue:42491|issue:44368", + "issue:44661|issue:45020", + "issue:43828|issue:45823", + "issue:45081|issue:45399", + "issue:45020|issue:45356", + "issue:43742|issue:45399", + "issue:43827|issue:44509", + "issue:43404|issue:43950", + "issue:43638|issue:45823", + "issue:42915|issue:45020", + "issue:43257|issue:43866", + "issue:43792|issue:44220", + "issue:43872|issue:44291", + "issue:42915|issue:43872", + "issue:41553|issue:44991", + "issue:43940|issue:45313", + "issue:36331|issue:41093", + "issue:42947|issue:45923", + "issue:43257|issue:45305", + "issue:43854|issue:44863", + "issue:44928|issue:45588", + "issue:43872|issue:45538", + "issue:41093|issue:44805", + "issue:44479|issue:44560", + "issue:41553|issue:44488", + "issue:44530|issue:44863", + "issue:24643|issue:30064", + "issue:43366|issue:43828", + "issue:45325|issue:45902" + ], + "prepared_review_unit_hash": "d21fa50cb4c4875fa8c27c43d17c5d9f56c54250eb70039a046a247ba6b7f8b6", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 33, + "serialized_chars": 38072, + "estimated_input_tokens": 9518, + "estimated_eval_tokens": 19292 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly heterogeneous transformer/model-loading regressions with no single broad duplicate. Only one soft edge looks like a true duplicate: the two docs issues about stale v5 pipeline-task references. All other candidate pairs are different bugs or different subsystems.", + "confidence": 0.84, + "canonical_issue_reason": "Issue 44509 is the broader, more explicit version of the docs regression: it covers multiple removed pipeline tasks and subsumes 43827's narrower summary.", + "canonical_pr_reason": null, + "best_issue_reason": "44509 is the best issue candidate because it is the clearest, most comprehensive statement of the docs cleanup needed after v5 pipeline removals.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45685", + "right": "issue:45823", + "accept": false, + "reason": "Different failures: MPS histogram/int typing vs Gemma4 PLE device-mapping mismatch." + }, + { + "left": "issue:43824", + "right": "issue:43872", + "accept": false, + "reason": "Different bugs: missing Qwen2.5-VL import vs bitsandbytes __new__ kwarg incompatibility." + }, + { + "left": "issue:44991", + "right": "issue:45399", + "accept": false, + "reason": "Tokenizer load failure for a specific model vs flash-attn fallback being blocked; not the same code path." + }, + { + "left": "issue:45020", + "right": "issue:45823", + "accept": false, + "reason": "Remote-code regressions and Gemma4 device mismatch are unrelated." + }, + { + "left": "issue:42491", + "right": "issue:44368", + "accept": false, + "reason": "LoRA/Qwen3 MoE compatibility issue vs a warning about tie_word_embeddings; related theme but not the same bug." + }, + { + "left": "issue:44661", + "right": "issue:45020", + "accept": false, + "reason": "Tokenizer mapping-name edge case vs broader remote_code breakage; distinct failure modes." + }, + { + "left": "issue:43828", + "right": "issue:45823", + "accept": false, + "reason": "Autocast dtype mismatch is unrelated to Gemma4 PLE device placement." + }, + { + "left": "issue:45081", + "right": "issue:45399", + "accept": false, + "reason": "Mistral regex patch crash is a tokenizer-specific bug, not the flash-attn fallback issue." + }, + { + "left": "issue:45020", + "right": "issue:45356", + "accept": false, + "reason": "General remote_code regression vs Kimi-K2.5 tokenizer codec/regression; different underlying problems." + }, + { + "left": "issue:43742", + "right": "issue:45399", + "accept": false, + "reason": "Model-specific key error vs flash-attn fallback gating; no shared bug." + }, + { + "left": "issue:43827", + "right": "issue:44509", + "accept": true, + "reason": "These are the same docs regression: both report stale references to removed v5 pipeline tasks, with 44509 being the broader wording." + }, + { + "left": "issue:43404", + "right": "issue:43950", + "accept": false, + "reason": "Weight-tying bug in Mistral3 vs non-persistent buffer corruption on from_pretrained; different mechanisms." + }, + { + "left": "issue:43638", + "right": "issue:45823", + "accept": false, + "reason": "Zero-3/Bert index error is unrelated to Gemma4 device-map mismatch." + }, + { + "left": "issue:42915", + "right": "issue:45020", + "accept": false, + "reason": "Qwen3 MoE FP8 loading failure vs remote_code breakage; not the same bug." + }, + { + "left": "issue:43257", + "right": "issue:43866", + "accept": false, + "reason": "Qwen3 MoE weight conversion/loading issue vs Ovis2 checkpoint corruption." + }, + { + "left": "issue:43792", + "right": "issue:44220", + "accept": false, + "reason": "Whisper loading/runtime problem vs _torch_extract_fbank_features issue; different code paths." + }, + { + "left": "issue:43872", + "right": "issue:44291", + "accept": false, + "reason": "Different bitsandbytes vs init_empty_weights / _is_hf_initialized incompatibilities." + }, + { + "left": "issue:42915", + "right": "issue:43872", + "accept": false, + "reason": "Qwen3 MoE loading/FP8 issue is unrelated to bitsandbytes constructor incompatibility." + }, + { + "left": "issue:41553", + "right": "issue:44991", + "accept": false, + "reason": "Voxtral tokenizer error message vs est-roberta tokenizer loading regression; distinct models and failures." + }, + { + "left": "issue:43940", + "right": "issue:45313", + "accept": false, + "reason": "Both are DeepSpeed weight-loading issues, but one is Qwen3-Next all-missing params and the other is Qwen3.5 language_model loading; too different to merge." + }, + { + "left": "issue:36331", + "right": "issue:41093", + "accept": false, + "reason": "Trainer compute_loss signature mismatch vs tensor/mask shape mismatch; unrelated bugs." + }, + { + "left": "issue:42947", + "right": "issue:45923", + "accept": false, + "reason": "Gradient checkpointing ineffective in PEFT LoRA vs missing supports_gradient_checkpointing flag; different issues." + }, + { + "left": "issue:43257", + "right": "issue:45305", + "accept": false, + "reason": "Qwen3 MoE weight conversion under accelerate+deepspeed vs gradient averaging with GAS and DeepSpeed; distinct training bugs." + }, + { + "left": "issue:43854", + "right": "issue:44863", + "accept": false, + "reason": "Unit-test loading failure for GLM-4.7-Flash vs NemotronH checkpoint loading failure; separate model implementations." + }, + { + "left": "issue:44928", + "right": "issue:45588", + "accept": false, + "reason": "RLHF gradient explosion from 3D position_ids/SDPA fallback vs flash_attention AttributeError on s_aux=None." + }, + { + "left": "issue:43872", + "right": "issue:45538", + "accept": false, + "reason": "bitsandbytes incompatibility is unrelated to CLIPTokenizer model_max_length behavior." + }, + { + "left": "issue:41093", + "right": "issue:44805", + "accept": false, + "reason": "Both are mask/tensor index errors, but the reported shapes and contexts differ enough that they do not look like the same bug." + }, + { + "left": "issue:44479", + "right": "issue:44560", + "accept": false, + "reason": "Video input regression across multiple Qwen variants vs a specific Qwen3-vl-embedding StopIteration; related area but not the same failure." + }, + { + "left": "issue:41553", + "right": "issue:44488", + "accept": false, + "reason": "Bad AutoTokenizer error for Voxtral vs a model-loading failure for cjvt/sleng-bert; different problems." + }, + { + "left": "issue:44530", + "right": "issue:44863", + "accept": false, + "reason": "PagedAttentionCache linear_attention crash vs NemotronH checkpoint loading failure; unrelated." + }, + { + "left": "issue:24643", + "right": "issue:30064", + "accept": false, + "reason": "DeepSpeed training RuntimeError vs image processor void segmentation-map failure; no shared bug." + }, + { + "left": "issue:43366", + "right": "issue:43828", + "accept": false, + "reason": "GGUF gpt-oss support request vs autocast dtype mismatch in Phi-tiny-MoE; not duplicates." + }, + { + "left": "issue:45325", + "right": "issue:45902", + "accept": false, + "reason": "Qwen2.5-VL rope-index scaling bug vs uninitialized-memory read in Qwen3_5MoeTextRotaryEmbedding; different implementations and symptoms." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded and conservative overall. The only accepted soft edge, issue:43827 vs issue:44509, is well supported by the titles: both describe stale docs references to removed v5 pipeline tasks, with 44509 clearly broader. The remaining rejections stay cautious and do not overclaim duplication from the packet evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 5, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:36010", + "issue:38175", + "issue:39692", + "issue:41720", + "issue:42491", + "issue:42617", + "issue:42915", + "issue:43317", + "issue:43366", + "issue:43531", + "issue:43550", + "issue:43596", + "issue:43632", + "issue:43638", + "issue:43644", + "issue:43645", + "issue:43646", + "issue:43708", + "issue:43720", + "issue:43824", + "issue:43950", + "issue:43957", + "issue:43994", + "issue:44075", + "issue:44155", + "issue:44451", + "issue:44488", + "issue:44534", + "issue:44560", + "issue:44589", + "issue:44661", + "issue:44756", + "issue:44805", + "issue:44877", + "issue:44910", + "issue:44918", + "issue:44960", + "issue:44977", + "issue:45030", + "issue:45276", + "issue:45313", + "issue:45357", + "issue:45584", + "issue:45663", + "issue:45685", + "issue:45823", + "issue:45834" + ], + "soft_pairs": [ + "issue:45663|issue:45823", + "issue:43366|issue:43531", + "issue:44155|issue:45584", + "issue:43950|issue:45834", + "issue:44560|issue:45685", + "issue:43824|issue:43957", + "issue:43550|issue:44910", + "issue:42915|issue:43957", + "issue:43366|issue:44589", + "issue:43645|issue:43646", + "issue:44560|issue:44918", + "issue:42617|issue:43366", + "issue:36010|issue:42915", + "issue:44661|issue:45823", + "issue:44560|issue:44805", + "issue:43632|issue:45834", + "issue:38175|issue:43994", + "issue:43638|issue:44661", + "issue:43596|issue:44661", + "issue:44661|issue:44805", + "issue:44877|issue:45030", + "issue:43317|issue:44756", + "issue:43644|issue:44534", + "issue:38175|issue:39692", + "issue:42491|issue:43720", + "issue:41720|issue:42915", + "issue:44960|issue:44977", + "issue:45313|issue:45357", + "issue:44451|issue:44488", + "issue:42915|issue:44756", + "issue:42491|issue:45276", + "issue:43708|issue:44075" + ], + "prepared_review_unit_hash": "b0b3fe14205ef052d89f7b0d611c333ed599226ea3ec36e7a5bfc9d432787219", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 32, + "serialized_chars": 36625, + "estimated_input_tokens": 9157, + "estimated_eval_tokens": 18570 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a mix of unrelated Transformers issues with a few true duplicate pairs. The clearest duplicate groups are SigLIP2 loading/output regressions, non-persistent-buffer corruption in v5, custom model initialization failures, and a BERT model-loading regression.", + "confidence": 0.89, + "canonical_issue_reason": "I\u2019d anchor the cluster on issue 43644 because it is a clear, regression-focused description of the non-persistent-buffer bug and has an exact duplicate in 44534, making it a strong representative issue.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43644 is the best overall representative: it is specific enough to describe the core regression cleanly, yet generic enough to subsume the matching duplicate 44534.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45663", + "right": "issue:45823", + "accept": false, + "reason": "Same model family, but different symptoms and code paths: FSDP2 KeyError in attention vs device_map forward mismatch." + }, + { + "left": "issue:43366", + "right": "issue:43531", + "accept": false, + "reason": "Different problems: GGUF architecture support vs Qwen3-MoE sliding_window behavior." + }, + { + "left": "issue:44155", + "right": "issue:45584", + "accept": false, + "reason": "Unrelated models and failures: AudioFlamingo3 track leakage vs Whisper empty-transcription generation." + }, + { + "left": "issue:43950", + "right": "issue:45834", + "accept": false, + "reason": "Both are loading/runtime bugs, but the buffer-corruption regression and Kosmos2.5 OCR index error are distinct." + }, + { + "left": "issue:44560", + "right": "issue:45685", + "accept": false, + "reason": "Different model-specific failures: Qwen3-vl video StopIteration vs MPS histogram support for int." + }, + { + "left": "issue:43824", + "right": "issue:43957", + "accept": false, + "reason": "ImportError for a missing class and meta-device loading breakage are different failure modes." + }, + { + "left": "issue:43550", + "right": "issue:44910", + "accept": false, + "reason": "Different code paths: torch.compile+SDPA in Bamba vs Flash Attention on Qwen3.5 with 3D position_ids." + }, + { + "left": "issue:42915", + "right": "issue:43957", + "accept": false, + "reason": "Qwen3-MoE FP8 failure and meta-device loading regression are not the same bug." + }, + { + "left": "issue:43366", + "right": "issue:44589", + "accept": false, + "reason": "Both involve low-level model loading, but GGUF gpt-oss support and Float8 storage lookup are separate issues." + }, + { + "left": "issue:43645", + "right": "issue:43646", + "accept": true, + "reason": "Same underlying custom-model initialization regression; the notebook-specific report is just a variant of the generic failure." + }, + { + "left": "issue:44560", + "right": "issue:44918", + "accept": false, + "reason": "Different Qwen3-family failures: video StopIteration vs TRL SFT unpacking of input embeddings." + }, + { + "left": "issue:42617", + "right": "issue:43366", + "accept": false, + "reason": "3d_parallel.py runtime failure and GGUF gpt-oss support are unrelated." + }, + { + "left": "issue:36010", + "right": "issue:42915", + "accept": false, + "reason": "GenerationMixin import error and Qwen3-MoE FP8 failure do not share a concrete code-path problem." + }, + { + "left": "issue:44661", + "right": "issue:45823", + "accept": false, + "reason": "Tokenizer-mapping/model-addition bug and Gemma4 device-mismatch forward bug are unrelated." + }, + { + "left": "issue:44560", + "right": "issue:44805", + "accept": false, + "reason": "Both are runtime errors, but StopIteration in video handling and mask shape mismatch are different bugs." + }, + { + "left": "issue:43632", + "right": "issue:45834", + "accept": false, + "reason": "_is_hf_initialized flag regression and Kosmos2.5 long-OCR index error are unrelated." + }, + { + "left": "issue:38175", + "right": "issue:43994", + "accept": true, + "reason": "Same SigLIP2 issue: incorrect loading/processing leads to zero or nonsensical outputs for the same model." + }, + { + "left": "issue:43638", + "right": "issue:44661", + "accept": false, + "reason": "Deepspeed ZeRO-3 Bert index error and add-new-model-like tokenizer-mapping failure are different failures." + }, + { + "left": "issue:43596", + "right": "issue:44661", + "accept": false, + "reason": "Both mention index/init problems, but the Bert ZeRO-3 error and tokenizer-mapping bug are not the same code path." + }, + { + "left": "issue:44661", + "right": "issue:44805", + "accept": false, + "reason": "Tokenizer-mapping bug and mask/tensor shape mismatch are unrelated." + }, + { + "left": "issue:44877", + "right": "issue:45030", + "accept": false, + "reason": "Strict config validation on granite_speech and glm4v config validation are separate model/config issues." + }, + { + "left": "issue:43317", + "right": "issue:44756", + "accept": false, + "reason": "Device-map dequantized-model loading and Strix Halo mmap OOM are different concerns." + }, + { + "left": "issue:43644", + "right": "issue:44534", + "accept": true, + "reason": "Exact same non-persistent-buffer corruption regression; 44534 is essentially a re-report of 43644." + }, + { + "left": "issue:38175", + "right": "issue:39692", + "accept": false, + "reason": "SigLIP2 zero-probability output and a documentation example with multiple errors are not the same underlying bug." + }, + { + "left": "issue:42491", + "right": "issue:43720", + "accept": false, + "reason": "LoRA qwen3_moe compatibility across hf versions and BitNet packed-weight loading are unrelated." + }, + { + "left": "issue:41720", + "right": "issue:42915", + "accept": false, + "reason": "Qwen3 auto device-mapping cuda assert and Qwen3-MoE FineGrainedFP8 failure are different issues." + }, + { + "left": "issue:44960", + "right": "issue:44977", + "accept": false, + "reason": "GLM5 and Qwen3.5 flash-attention generation failures are different models and different failures." + }, + { + "left": "issue:45313", + "right": "issue:45357", + "accept": false, + "reason": "ZeRO-3 weight loading for Qwen3.5 language_model and incorrect visual-encoder keys in save_pretrained are separate bugs." + }, + { + "left": "issue:44451", + "right": "issue:44488", + "accept": true, + "reason": "Same loading failure class for the same BERT-style models; 44488 is a duplicate report of the model-loading regression." + }, + { + "left": "issue:42915", + "right": "issue:44756", + "accept": false, + "reason": "Qwen3-MoE FP8 failure and Strix Halo mmap OOM are unrelated." + }, + { + "left": "issue:42491", + "right": "issue:45276", + "accept": false, + "reason": "Qwen3 MoE LoRA compatibility and Gemma4 resize_token_embeddings behavior are different problems." + }, + { + "left": "issue:43708", + "right": "issue:44075", + "accept": false, + "reason": "Trainer resume/max_steps calculation and SGD args being ignored are different Trainer/optimizer issues." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative overall. The accepted duplicate pairs are plausible from the titles/body metadata: 43644/44534 is effectively the same non-persistent-buffer regression, 43645/43646 are clearly the same custom-model initialization bug, 44451/44488 are the same BERT loading regression, and 38175/43994 are reasonably treated as a SigLIP2 duplicate pair. The false soft-edge verdicts are mostly cautious and the cluster-level summary correctly reflects that this set is largely unrelated issues with a few duplicate groups." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 6, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:36010", + "issue:39404", + "issue:39692", + "issue:41720", + "issue:42491", + "issue:42947", + "issue:43257", + "issue:43335", + "issue:43381", + "issue:43404", + "issue:43425", + "issue:43454", + "issue:43644", + "issue:43645", + "issue:43854", + "issue:43856", + "issue:43873", + "issue:43931", + "issue:43957", + "issue:43994", + "issue:44155", + "issue:44164", + "issue:44291", + "issue:44292", + "issue:44387", + "issue:44402", + "issue:44488", + "issue:44756", + "issue:44910", + "issue:44912", + "issue:44928", + "issue:44960", + "issue:44991", + "issue:45216", + "issue:45276", + "issue:45292", + "issue:45335", + "issue:45362", + "issue:45406", + "issue:45488", + "issue:45588", + "issue:45663", + "issue:45704", + "issue:45759", + "issue:45812", + "issue:45823", + "issue:45859", + "issue:46050" + ], + "soft_pairs": [ + "issue:43854|issue:43931", + "issue:43257|issue:43854", + "issue:45488|issue:45812", + "issue:44488|issue:44991", + "issue:39692|issue:43994", + "issue:43644|issue:43645", + "issue:43856|issue:43873", + "issue:42491|issue:43931", + "issue:43856|issue:44387", + "issue:45362|issue:45406", + "issue:44960|issue:46050", + "issue:42491|issue:45292", + "issue:44910|issue:45588", + "issue:44164|issue:44291", + "issue:36010|issue:39404", + "issue:45276|issue:45335", + "issue:43404|issue:43454", + "issue:43425|issue:43957", + "issue:41720|issue:44155", + "issue:39692|issue:44960", + "issue:42947|issue:45704", + "issue:44402|issue:44488", + "issue:43873|issue:44387", + "issue:45663|issue:45759", + "issue:44291|issue:44756", + "issue:45663|issue:45859", + "issue:44155|issue:45823", + "issue:43335|issue:45663", + "issue:44292|issue:44912", + "issue:44387|issue:44756", + "issue:43957|issue:44756", + "issue:43854|issue:45216", + "issue:45276|issue:45292", + "issue:42947|issue:43381", + "issue:44164|issue:44756", + "issue:43381|issue:44928" + ], + "prepared_review_unit_hash": "0ce7477a3c19eb43ffa84d6f03d00715bae1123b1ae74b5fd0a48f7a84fa59e0", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 36, + "serialized_chars": 38638, + "estimated_input_tokens": 9660, + "estimated_eval_tokens": 19576 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most items are unrelated model-specific regressions or broad theme overlaps. The only clear duplicate family is the resize_token_embeddings failure affecting output_embeddings / model-specific embedding modules across related architectures.", + "confidence": 0.73, + "canonical_issue_reason": "Issue 45292 is the cleanest, most general statement of the resize_token_embeddings regression and best serves as the duplicate anchor for the embedding-resize subgroup.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45292 is the strongest overall issue representative in this cluster because it is concise, model-agnostic, and directly captures the shared failure mode behind the related resize_token_embeddings reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43854", + "right": "issue:43931", + "accept": false, + "reason": "Different model/loading failures: GLM-4.7 unit-test loading vs Qwen3-VL weight-shape mismatch." + }, + { + "left": "issue:43257", + "right": "issue:43854", + "accept": false, + "reason": "Qwen3 MoE conversion/offload problem is unrelated to GLM-4.7 test loading." + }, + { + "left": "issue:45488", + "right": "issue:45812", + "accept": false, + "reason": "Both are tokenizer regressions, but they target different model families and likely different code paths." + }, + { + "left": "issue:44488", + "right": "issue:44991", + "accept": false, + "reason": "Both concern tokenizer loading on transformers 5.x, but the model-specific failures are distinct." + }, + { + "left": "issue:39692", + "right": "issue:43994", + "accept": false, + "reason": "SigLIP2 docs/example errors and bad AutoModel/pipeline results are related only at a very broad model-family level." + }, + { + "left": "issue:43644", + "right": "issue:43645", + "accept": false, + "reason": "Different regressions: junk non-persistent buffers vs Jupyter custom-model initialization." + }, + { + "left": "issue:43856", + "right": "issue:43873", + "accept": false, + "reason": "Training memory usage and quantized offloading behavior are not the same underlying bug." + }, + { + "left": "issue:42491", + "right": "issue:43931", + "accept": false, + "reason": "Unrelated issues: Qwen3 MoE LoRA compatibility vs Qwen3-VL load mismatch." + }, + { + "left": "issue:43856", + "right": "issue:44387", + "accept": false, + "reason": "Both involve Qwen3/quantization memory, but one is training efficiency and the other is reserved-memory OOM under int4." + }, + { + "left": "issue:45362", + "right": "issue:45406", + "accept": false, + "reason": "Different crashes in different subsystems/models: chat loading vs serve processor attribute error." + }, + { + "left": "issue:44960", + "right": "issue:46050", + "accept": false, + "reason": "GLM5 loading and sinq save/load corruption are different bugs." + }, + { + "left": "issue:42491", + "right": "issue:45292", + "accept": false, + "reason": "Qwen3 MoE LoRA compatibility is unrelated to the embedding resize regression." + }, + { + "left": "issue:44910", + "right": "issue:45588", + "accept": false, + "reason": "Flash-Attention position_ids crash and flash_attention.py s_aux=None crash are different code-path failures." + }, + { + "left": "issue:44164", + "right": "issue:44291", + "accept": false, + "reason": "save/from_pretrained extra_state handling and init_empty_weights argument handling are separate loading issues." + }, + { + "left": "issue:36010", + "right": "issue:39404", + "accept": false, + "reason": "ImportError for GenerationMixin and Whisper return_language pipeline break are unrelated." + }, + { + "left": "issue:45276", + "right": "issue:45335", + "accept": true, + "reason": "Both report the same resize_token_embeddings failure class: resizing does not propagate to model-specific embedding/output modules." + }, + { + "left": "issue:43404", + "right": "issue:43454", + "accept": false, + "reason": "Similar symptom (lm_head not tied), but different model implementations and not enough evidence of a shared code path." + }, + { + "left": "issue:43425", + "right": "issue:43957", + "accept": false, + "reason": "Torch version incompatibility and meta-device loading failures are different issues." + }, + { + "left": "issue:41720", + "right": "issue:44155", + "accept": false, + "reason": "Qwen3 auto device-map crash and AudioFlamingo3 batched inference leakage are unrelated." + }, + { + "left": "issue:39692", + "right": "issue:44960", + "accept": false, + "reason": "SigLIP2 example problems and GLM5 loading are different model regressions." + }, + { + "left": "issue:42947", + "right": "issue:45704", + "accept": false, + "reason": "Gradient checkpointing/LoRA behavior is unrelated to T5 apex RMSNorm memory leak." + }, + { + "left": "issue:44402", + "right": "issue:44488", + "accept": false, + "reason": "Different tokenizer-loading/model-vocab issues with no clear shared underlying bug." + }, + { + "left": "issue:43873", + "right": "issue:44387", + "accept": false, + "reason": "Quantization offloading misbehavior and int4 reserved-memory OOM are related only at a broad theme level." + }, + { + "left": "issue:45663", + "right": "issue:45759", + "accept": false, + "reason": "Gemma4 FSDP2 shared_kv_states bug and Qwen multimodal text_config unwrapping are different model/config issues." + }, + { + "left": "issue:44291", + "right": "issue:44756", + "accept": false, + "reason": "Meta init TypeError and Strix Halo mmap OOM are unrelated." + }, + { + "left": "issue:45663", + "right": "issue:45859", + "accept": false, + "reason": "Different models and failures: Gemma4 FSDP2 KeyError vs Qwen3_5Moe CPU offload incompatibility." + }, + { + "left": "issue:44155", + "right": "issue:45823", + "accept": false, + "reason": "AudioFlamingo3 embedding/token leak and Gemma4 device mismatch are unrelated." + }, + { + "left": "issue:43335", + "right": "issue:45663", + "accept": false, + "reason": "Sparse-layer config bug and Gemma4 attention KeyError are unrelated." + }, + { + "left": "issue:44292", + "right": "issue:44912", + "accept": false, + "reason": "Qwen NVFP4 loading failure and git-oss MXFP4 fallback are different quantization bugs." + }, + { + "left": "issue:44387", + "right": "issue:44756", + "accept": false, + "reason": "Reserved-memory OOM under int4 and Strix Halo mmap OOM are different failure modes." + }, + { + "left": "issue:43957", + "right": "issue:44756", + "accept": false, + "reason": "Meta-device loading regressions and mmap/OOM on Strix Halo are unrelated." + }, + { + "left": "issue:43854", + "right": "issue:45216", + "accept": false, + "reason": "GLM-4.7 load/test failure and Qwen3.5 save_pretrained regression are different issues." + }, + { + "left": "issue:45276", + "right": "issue:45292", + "accept": true, + "reason": "Same underlying bug: resize_token_embeddings does not update output/embedded weights correctly." + }, + { + "left": "issue:42947", + "right": "issue:43381", + "accept": false, + "reason": "Gradient checkpointing in PEFT LoRA and eval-mode checkpointing are distinct behaviors." + }, + { + "left": "issue:44164", + "right": "issue:44756", + "accept": false, + "reason": "Extra_state save/load failure and mmap/OOM on Strix Halo are unrelated." + }, + { + "left": "issue:43381", + "right": "issue:44928", + "accept": false, + "reason": "Eval-mode gradient checkpointing and RLHF NaN/SDPA fallback are different bugs." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The clustering is grounded in the packet and the soft-edge calls are conservative overall. The resize_token_embeddings group is the only clearly mergeable duplicate family, and using issue 45292 as the canonical anchor is reasonable given its model-agnostic wording. No obvious overstatement beyond the evidence shown." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 7, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:36032", + "issue:39404", + "issue:43257", + "issue:43299", + "issue:43425", + "issue:43452", + "issue:43596", + "issue:43611", + "issue:43645", + "issue:43646", + "issue:43708", + "issue:43716", + "issue:43828", + "issue:43856", + "issue:43901", + "issue:43906", + "issue:43931", + "issue:43957", + "issue:43994", + "issue:44079", + "issue:44164", + "issue:44265", + "issue:44402", + "issue:44451", + "issue:44479", + "issue:44509", + "issue:44530", + "issue:44589", + "issue:44671", + "issue:44683", + "issue:44756", + "issue:44805", + "issue:44863", + "issue:44877", + "issue:44898", + "issue:44912", + "issue:44918", + "issue:44991", + "issue:45084", + "issue:45292", + "issue:45335", + "issue:45588", + "issue:45663", + "issue:45684", + "issue:45698", + "issue:45701", + "issue:45759", + "issue:45823" + ], + "soft_pairs": [ + "issue:45684|issue:45698", + "issue:44479|issue:44877", + "issue:45588|issue:45663", + "issue:43646|issue:43708", + "issue:43645|issue:44164", + "issue:43828|issue:43957", + "issue:45292|issue:45335", + "issue:44402|issue:44451", + "issue:43425|issue:43828", + "issue:43257|issue:44863", + "issue:44683|issue:45823", + "issue:44589|issue:44756", + "issue:39404|issue:45701", + "issue:43425|issue:45823", + "issue:43425|issue:43716", + "issue:43611|issue:43645", + "issue:43906|issue:44530", + "issue:44530|issue:45084", + "issue:39404|issue:44912", + "issue:44265|issue:44898", + "issue:43901|issue:44509", + "issue:44479|issue:44805", + "issue:43994|issue:44079", + "issue:43425|issue:44918", + "issue:44589|issue:45084", + "issue:43299|issue:43931", + "issue:43596|issue:44683", + "issue:44164|issue:45759", + "issue:39404|issue:44671", + "issue:36032|issue:43452", + "issue:44671|issue:44991", + "issue:43856|issue:44756" + ], + "prepared_review_unit_hash": "4547e36f0048699a8b0d6ddcc47820d6d9ebbf8d929ee6bacf4938de29314f68", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 37355, + "estimated_input_tokens": 9339, + "estimated_eval_tokens": 18934 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose theme of Transformers v5 regressions across unrelated models/code paths. Only one soft pair looks like a true duplicate; the rest are just similarly worded issues and should stay separate.", + "confidence": 0.57, + "canonical_issue_reason": "#45292 is the broadest, most central report: it describes the underlying resize_token_embeddings propagation bug, and #45335 looks like a model-specific instance of the same failure.", + "canonical_pr_reason": null, + "best_issue_reason": "#45292 is the best representative issue because it is the most general, actionable statement of the bug and can subsume the narrower t5gemma-specific report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45684", + "right": "issue:45698", + "accept": false, + "reason": "Permissions propagation and wrong custom-module loading are different save/load problems, not the same bug." + }, + { + "left": "issue:44479", + "right": "issue:44877", + "accept": false, + "reason": "A video-input regression in Qwen VL models is unrelated to strict config blocking granite_speech loading." + }, + { + "left": "issue:45588", + "right": "issue:45663", + "accept": false, + "reason": "Flash-attention crashing on s_aux=None is a different code path from Gemma4 FSDP2 shared_kv_states rebuilding." + }, + { + "left": "issue:43646", + "right": "issue:43708", + "accept": false, + "reason": "Custom model initialization breakage and Trainer resume max_steps miscalculation are unrelated issues." + }, + { + "left": "issue:43645", + "right": "issue:44164", + "accept": false, + "reason": "Notebook custom-model initialization and extra_state save/load failures are separate lifecycle bugs." + }, + { + "left": "issue:43828", + "right": "issue:43957", + "accept": false, + "reason": "These are both loading/runtime mismatches, but one is autocast dtype behavior and the other is meta-device loading." + }, + { + "left": "issue:45292", + "right": "issue:45335", + "accept": true, + "reason": "Both report resize_token_embeddings not updating downstream embeddings/output layers; one is a general report and the other a t5gemma-specific manifestation." + }, + { + "left": "issue:44402", + "right": "issue:44451", + "accept": false, + "reason": "Tokenizer vocab-size mismatch and inability to load a specific model are not clearly the same underlying bug." + }, + { + "left": "issue:43425", + "right": "issue:43828", + "accept": false, + "reason": "Torch 2.10 compatibility and autocast dtype mismatch are different failures." + }, + { + "left": "issue:43257", + "right": "issue:44863", + "accept": false, + "reason": "Qwen3 MoE conversion under accelerate+deepspeed and NemotronH checkpoint loading are distinct model-loading problems." + }, + { + "left": "issue:44683", + "right": "issue:45823", + "accept": false, + "reason": "Compiled flex_attention on torch>=2.9 and Gemma4 PLE device-map mismatch do not share the same concrete code-path." + }, + { + "left": "issue:44589", + "right": "issue:44756", + "accept": false, + "reason": "Float8 storage lookup failure and mmap-induced OOM are unrelated issues." + }, + { + "left": "issue:39404", + "right": "issue:45701", + "accept": false, + "reason": "Whisper return_language pipeline regression is unrelated to tokenization changes in general." + }, + { + "left": "issue:43425", + "right": "issue:45823", + "accept": false, + "reason": "Torch version incompatibility is not the same as Gemma4 PLE forward device mismatch." + }, + { + "left": "issue:43425", + "right": "issue:43716", + "accept": false, + "reason": "Torch 2.10 incompatibility and Mistral-3 image preprocessor dtype mismatch are different regressions." + }, + { + "left": "issue:43611", + "right": "issue:43645", + "accept": false, + "reason": "Base_model_prefix loading and Jupyter custom-model initialization are separate Transformers 5 breakages." + }, + { + "left": "issue:43906", + "right": "issue:44530", + "accept": false, + "reason": "A reproduction of another issue and a Qwen3.5 PagedAttentionCache crash are not the same bug." + }, + { + "left": "issue:44530", + "right": "issue:45084", + "accept": false, + "reason": "PagedAttentionCache linear_attention failure and non-template-node compile errors are different compiler/runtime problems." + }, + { + "left": "issue:39404", + "right": "issue:44912", + "accept": false, + "reason": "Whisper pipeline behavior and MXFP4 quantization fallback are unrelated regressions." + }, + { + "left": "issue:44265", + "right": "issue:44898", + "accept": false, + "reason": "torch.export compatibility with torch_compilable_check is a different failure from Perceiver resolution handling." + }, + { + "left": "issue:43901", + "right": "issue:44509", + "accept": false, + "reason": "Both are docs-related, but they document different removed/deprecated pipeline behaviors and should not be merged." + }, + { + "left": "issue:44479", + "right": "issue:44805", + "accept": false, + "reason": "A Qwen video-input regression and a generic mask/index shape error are not clearly the same issue." + }, + { + "left": "issue:43994", + "right": "issue:44079", + "accept": false, + "reason": "Incorrect inference results and ModelOutput key assignment with None are different problems." + }, + { + "left": "issue:43425", + "right": "issue:44918", + "accept": false, + "reason": "Torch compatibility and TRL embedding unpacking are unrelated." + }, + { + "left": "issue:44589", + "right": "issue:45084", + "accept": false, + "reason": "Storage-object lookup failure is unrelated to compile-time non-template-node errors." + }, + { + "left": "issue:43299", + "right": "issue:43931", + "accept": false, + "reason": "Qwen3VL MoE loading regression and a specific weight-shape mismatch on Qwen3-VL-30B-A3B-Instruct are not the same concrete bug." + }, + { + "left": "issue:43596", + "right": "issue:44683", + "accept": false, + "reason": "DeepSpeed zero3/BertModel index errors and flex_attention compilation failures are separate code paths." + }, + { + "left": "issue:44164", + "right": "issue:45759", + "accept": false, + "reason": "extra_state save/load handling and unwrapping text_config in from_config are different model-serialization issues." + }, + { + "left": "issue:39404", + "right": "issue:44671", + "accept": false, + "reason": "Whisper pipeline regression and CamemBERT MLM prediction correctness are unrelated." + }, + { + "left": "issue:36032", + "right": "issue:43452", + "accept": false, + "reason": "Tokenizer add_special_tokens conflict and gguf_file from_pretrained breakage are different tokenizer/model-loading bugs." + }, + { + "left": "issue:44671", + "right": "issue:44991", + "accept": false, + "reason": "CamemBERT prediction correctness and tokenizer loading for EMBEDDIA/est-roberta are not the same underlying bug." + }, + { + "left": "issue:43856", + "right": "issue:44756", + "accept": false, + "reason": "Qwen3 MoE training memory usage and Strix Halo mmap OOM are different performance/memory issues." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative about the soft pairs. The only accepted duplicate pair, issue:45292 vs issue:45335, is well supported by the titles, and the other pairwise rejections do not overclaim equivalence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 8, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:32090", + "issue:34567", + "issue:36032", + "issue:39404", + "issue:40444", + "issue:40990", + "issue:42915", + "issue:43089", + "issue:43262", + "issue:43317", + "issue:43377", + "issue:43421", + "issue:43452", + "issue:43482", + "issue:43541", + "issue:43582", + "issue:43611", + "issue:43632", + "issue:43643", + "issue:43644", + "issue:43645", + "issue:43646", + "issue:43716", + "issue:43883", + "issue:44164", + "issue:44206", + "issue:44220", + "issue:44451", + "issue:44464", + "issue:44479", + "issue:44488", + "issue:44530", + "issue:44534", + "issue:44661", + "issue:44683", + "issue:44737", + "issue:44811", + "issue:44849", + "issue:44987", + "issue:44991", + "issue:45081", + "issue:45084", + "issue:45325", + "issue:45507", + "issue:45584", + "issue:45859", + "issue:45987" + ], + "soft_pairs": [ + "issue:32090|issue:39404", + "issue:44991|issue:45081", + "issue:44164|issue:44479", + "issue:44737|issue:45859", + "issue:43582|issue:43632", + "issue:44661|issue:44683", + "issue:44488|issue:45081", + "issue:44488|issue:44987", + "issue:43611|issue:43644", + "issue:40990|issue:43421", + "issue:43716|issue:45084", + "issue:44206|issue:44220", + "issue:36032|issue:42915", + "issue:43643|issue:44530", + "issue:34567|issue:40444", + "issue:43883|issue:44488", + "issue:44811|issue:45584", + "issue:43541|issue:45084", + "issue:42915|issue:44530", + "issue:44451|issue:44987", + "issue:44464|issue:45987", + "issue:43644|issue:43646", + "issue:43452|issue:43482", + "issue:43317|issue:43482", + "issue:36032|issue:43645", + "issue:44849|issue:45325", + "issue:44987|issue:44991", + "issue:45084|issue:45507", + "issue:43883|issue:44991", + "issue:43262|issue:43377", + "issue:43089|issue:45987", + "issue:43645|issue:44534", + "issue:43646|issue:44534" + ], + "prepared_review_unit_hash": "94b2982e355b2a834df6cb1dbd3513ef648ba574813a190d787b268117342a0a", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 33, + "serialized_chars": 37025, + "estimated_input_tokens": 9257, + "estimated_eval_tokens": 18770 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster looks mostly heterogeneous. There is a loose Transformers v5 subgroup around loading/initialization/buffer regressions, but the visible evidence does not support treating it as a confirmed duplicate set. I would avoid a strong canonical anchor and treat 44534 only as a lightweight representative of that weak subtheme, not the whole cluster.", + "confidence": 0.62, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "44534 is the clearest representative of the loose Transformers v5 buffer/init regression subtheme, but only as a provisional topic anchor. It is not strong enough evidence to collapse the cluster into one duplicate issue.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:32090", + "right": "issue:39404", + "accept": false, + "reason": "Different subsystems and symptoms: Trainer GPU broadcast TypeError vs Whisper pipeline return_language regression." + }, + { + "left": "issue:44991", + "right": "issue:45081", + "accept": false, + "reason": "Both are loading-related regressions, but they affect different paths and models; not the same concrete bug." + }, + { + "left": "issue:44164", + "right": "issue:44479", + "accept": false, + "reason": "save/from_pretrained extra_state handling is unrelated to the Qwen video-input regression." + }, + { + "left": "issue:44737", + "right": "issue:45859", + "accept": false, + "reason": "Both involve device behavior, but the model-specific failures and code paths differ." + }, + { + "left": "issue:43582", + "right": "issue:43632", + "accept": false, + "reason": "Apple Silicon warmup TypeError and _is_hf_initialized flag regression are different defects." + }, + { + "left": "issue:44661", + "right": "issue:44683", + "accept": false, + "reason": "Tokenizer mapping logic and compiled flex_attention failures are unrelated." + }, + { + "left": "issue:44488", + "right": "issue:45081", + "accept": false, + "reason": "Generic model loading failure vs Mistral tokenizer regex patch crash; same area, different bug." + }, + { + "left": "issue:44488", + "right": "issue:44987", + "accept": false, + "reason": "Both are loading complaints, but the models and failure modes differ too much to treat as duplicates." + }, + { + "left": "issue:43611", + "right": "issue:43644", + "accept": false, + "reason": "The titles suggest a related Transformers v5 theme, but the concrete symptoms differ and the packet does not support a confirmed duplicate." + }, + { + "left": "issue:40990", + "right": "issue:43421", + "accept": false, + "reason": "Perplexity regression is unrelated to runtime special-token/post-processor update behavior." + }, + { + "left": "issue:43716", + "right": "issue:45084", + "accept": false, + "reason": "Different bug classes: image preprocessor dtype mismatch vs template-node compilation error." + }, + { + "left": "issue:44206", + "right": "issue:44220", + "accept": false, + "reason": "Both are audio-feature related, but not clearly the same underlying defect or code path." + }, + { + "left": "issue:36032", + "right": "issue:42915", + "accept": false, + "reason": "T5 tokenizer conflict and Qwen3Moe FP8 failure are unrelated." + }, + { + "left": "issue:43643", + "right": "issue:44530", + "accept": false, + "reason": "AutoConfig trust_remote_code field loss is unrelated to the Qwen3.5 PagedAttention crash." + }, + { + "left": "issue:34567", + "right": "issue:40444", + "accept": false, + "reason": "TrainerState token counting and multi-image IterableDataset finetuning failures are unrelated." + }, + { + "left": "issue:43883", + "right": "issue:44488", + "accept": false, + "reason": "Missing tied-weights metadata and BERT loading failure are different issues." + }, + { + "left": "issue:44811", + "right": "issue:45584", + "accept": false, + "reason": "Both are Whisper issues, but the specific symptoms and code paths differ." + }, + { + "left": "issue:43541", + "right": "issue:45084", + "accept": false, + "reason": "Torch dynamo grouped_mm tracing failure is unrelated to non-template-node compilation errors." + }, + { + "left": "issue:42915", + "right": "issue:44530", + "accept": false, + "reason": "Both are Qwen-family regressions, but the FP8 config failure and linear_attention cache crash are different bugs." + }, + { + "left": "issue:44451", + "right": "issue:44987", + "accept": false, + "reason": "Both are model-loading failures, but there is no evidence of the same root cause." + }, + { + "left": "issue:44464", + "right": "issue:45987", + "accept": false, + "reason": "Chunked generation inconsistency and seq_length return-type contract issue are unrelated." + }, + { + "left": "issue:43644", + "right": "issue:43646", + "accept": false, + "reason": "Custom model initialization breakage is only superficially related to the non-persistent-buffer junk regression." + }, + { + "left": "issue:43452", + "right": "issue:43482", + "accept": false, + "reason": "Both touch GGUF loading, but the visible descriptions point to different concrete failures." + }, + { + "left": "issue:43317", + "right": "issue:43482", + "accept": false, + "reason": "GPU+CPU offload loading and Qwen2.5-GGUF loading are different failure modes." + }, + { + "left": "issue:36032", + "right": "issue:43645", + "accept": false, + "reason": "Tokenizer method conflict and notebook custom model initialization are unrelated." + }, + { + "left": "issue:44849", + "right": "issue:45325", + "accept": false, + "reason": "Different Qwen model regressions with different symptoms and code paths." + }, + { + "left": "issue:44987", + "right": "issue:44991", + "accept": false, + "reason": "Both are transformers>=5 loading failures, but for different models and no shared concrete bug is visible." + }, + { + "left": "issue:45084", + "right": "issue:45507", + "accept": false, + "reason": "Non-template-node compilation error and GraniteMoE invalid-method call are unrelated." + }, + { + "left": "issue:43883", + "right": "issue:44991", + "accept": false, + "reason": "Missing tied-weights metadata and tokenizer loading regression are unrelated." + }, + { + "left": "issue:43262", + "right": "issue:43377", + "accept": false, + "reason": "Audio chat-template sample-rate defaulting and MIMI padding-mask mismatch are different bugs." + }, + { + "left": "issue:43089", + "right": "issue:45987", + "accept": false, + "reason": "Generation performance overhead and seq_length contract regression are not the same issue." + }, + { + "left": "issue:43645", + "right": "issue:44534", + "accept": false, + "reason": "Custom model initialization in notebooks is distinct from the non-persistent-buffer junk regression." + }, + { + "left": "issue:43646", + "right": "issue:44534", + "accept": false, + "reason": "Custom model initialization breakage is only loosely related to the buffer-junk regression; different concrete defects." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The summary is a bit too strong for the packet. The only pair the analyst marked as a true duplicate, issue:43611 vs issue:43644, is not clearly supported by the visible titles, and the packet\u2019s own soft-pair signal marks it as deterministic_accept=false. Because of that, calling this a clear duplicate theme and using 44534 as the canonical anchor overstates the evidence. The conservative read is that the cluster is mostly heterogeneous, with at most a loose Transformers v5 similarity around buffer/init regressions, not a confirmed duplicate set." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 9, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:33666", + "issue:36032", + "issue:40444", + "issue:40990", + "issue:41553", + "issue:41720", + "issue:42175", + "issue:42915", + "issue:43296", + "issue:43322", + "issue:43377", + "issue:43404", + "issue:43425", + "issue:43531", + "issue:43540", + "issue:43582", + "issue:43611", + "issue:43632", + "issue:43742", + "issue:43792", + "issue:43819", + "issue:43994", + "issue:44062", + "issue:44079", + "issue:44220", + "issue:44265", + "issue:44291", + "issue:44403", + "issue:44451", + "issue:44466", + "issue:44479", + "issue:44534", + "issue:44589", + "issue:44610", + "issue:44683", + "issue:44756", + "issue:44811", + "issue:44877", + "issue:44987", + "issue:45020", + "issue:45081", + "issue:45084", + "issue:45127", + "issue:45216", + "issue:45325", + "issue:45357", + "issue:45542", + "issue:45663" + ], + "soft_pairs": [ + "issue:43404|issue:44466", + "issue:43322|issue:43540", + "issue:44403|issue:45081", + "issue:43994|issue:44610", + "issue:42915|issue:45084", + "issue:43377|issue:43819", + "issue:44220|issue:44683", + "issue:44265|issue:44610", + "issue:44451|issue:45081", + "issue:44987|issue:45081", + "issue:42915|issue:44291", + "issue:43632|issue:44756", + "issue:45216|issue:45357", + "issue:36032|issue:43742", + "issue:44062|issue:44987", + "issue:42915|issue:44589", + "issue:44479|issue:45325", + "issue:43404|issue:45127", + "issue:43296|issue:43531", + "issue:40444|issue:45325", + "issue:43531|issue:44877", + "issue:36032|issue:41720", + "issue:43611|issue:44534", + "issue:43425|issue:44403", + "issue:43632|issue:44534", + "issue:40990|issue:44811", + "issue:43792|issue:44589", + "issue:44079|issue:45663", + "issue:44683|issue:45020", + "issue:41553|issue:43582", + "issue:42175|issue:45542", + "issue:33666|issue:40444" + ], + "prepared_review_unit_hash": "a836a7a71932389f811ad07c23e8015dac0b2582c6a6143c6b918461a768d631", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 37185, + "estimated_input_tokens": 9297, + "estimated_eval_tokens": 18850 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is thematically related but not a true duplicate set: it mixes many independent Transformers v5/model-loading regressions, tokenizer errors, and multimodal preprocessing bugs. Only a couple of soft pairs look like the same underlying bug; most are merely similar by subsystem or model family.", + "confidence": 0.82, + "canonical_issue_reason": "issue:45020 is the broadest umbrella item in the set: it describes recent Transformers versions breaking remote_code model loading, which best matches the cluster\u2019s repeated pattern of version-regression breakages across models.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45020 is the strongest representative to surface because it is broad, regression-focused, and most likely to anchor the shared version-breakage theme across the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43404", + "right": "issue:44466", + "accept": false, + "reason": "Both concern tied weights in related models, but one is a missing tie in Mistral3ForConditionalGeneration and the other is device-dependent save_pretrained serialization; different bugs." + }, + { + "left": "issue:43322", + "right": "issue:43540", + "accept": false, + "reason": "Different model families and failure modes: Llava Next segfault vs Qwen3OmniMoe video-input ValueError." + }, + { + "left": "issue:44403", + "right": "issue:45081", + "accept": false, + "reason": "Generic loading noise vs a specific Mistral tokenizer crash; not the same underlying issue." + }, + { + "left": "issue:43994", + "right": "issue:44610", + "accept": false, + "reason": "Unrelated vision bugs: SigLIP2 bad outputs vs OmDet-Turbo processor/model input-size mismatch." + }, + { + "left": "issue:42915", + "right": "issue:45084", + "accept": false, + "reason": "Qwen3Moe FP8 config failure and a template compilation TypeError are different code paths." + }, + { + "left": "issue:43377", + "right": "issue:43819", + "accept": false, + "reason": "Both are audio/codec related, but one is batch padding-mask behavior in MIMI and the other is DAC STE mismatch; not duplicates." + }, + { + "left": "issue:44220", + "right": "issue:44683", + "accept": false, + "reason": "Feature extraction helper failure and compiled flex_attention regression are unrelated." + }, + { + "left": "issue:44265", + "right": "issue:44610", + "accept": false, + "reason": "Different subsystems: torch.export with torch_compilable_check vs OmDet-Turbo preprocessing size bug." + }, + { + "left": "issue:44451", + "right": "issue:45081", + "accept": false, + "reason": "ScandiBERT loading failure and a Mistral regex/tokenizer crash are unrelated model-specific bugs." + }, + { + "left": "issue:44987", + "right": "issue:45081", + "accept": false, + "reason": "Physical-intelligence/fast loading regression is not the same as the Mistral tokenizer backend crash." + }, + { + "left": "issue:42915", + "right": "issue:44291", + "accept": false, + "reason": "Both mention newer Transformers behavior, but FP8 config failure and init_empty_weights/_is_hf_initialized are distinct bugs." + }, + { + "left": "issue:43632", + "right": "issue:44756", + "accept": false, + "reason": "_is_hf_initialized breakage and Strix Halo mmap OOM are unrelated." + }, + { + "left": "issue:45216", + "right": "issue:45357", + "accept": true, + "reason": "Same Qwen3.5 save_pretrained regression family: both report incorrect serialized checkpoint contents/keys after v5.4+." + }, + { + "left": "issue:36032", + "right": "issue:43742", + "accept": false, + "reason": "T5 tokenizer add_special_tokens conflict and MobileLLM key error are unrelated loading issues." + }, + { + "left": "issue:44062", + "right": "issue:44987", + "accept": false, + "reason": "AddedToken special-argument conflict is a tokenizer construction bug, not the same as loading physical-intelligence/fast." + }, + { + "left": "issue:42915", + "right": "issue:44589", + "accept": false, + "reason": "FP8 config failure and missing Float8 storage type are different failure points and likely different fixes." + }, + { + "left": "issue:44479", + "right": "issue:45325", + "accept": true, + "reason": "Same Qwen2.5-VL temporal-position/rope-index regression in v5.3.0; 45325 is a narrower symptom of the broader video/input regression." + }, + { + "left": "issue:43404", + "right": "issue:45127", + "accept": false, + "reason": "Both involve tied embeddings, but one is a model implementation bug and the other is LoRA merge collapse with extended vocab; not the same issue." + }, + { + "left": "issue:43296", + "right": "issue:43531", + "accept": false, + "reason": "PaddleOCR-VL loading and Qwen3-MoE sliding_window behavior are unrelated." + }, + { + "left": "issue:40444", + "right": "issue:45325", + "accept": false, + "reason": "IterableDataset multi-image training failure is not the same as the Qwen2.5-VL rope-index temporal-position bug." + }, + { + "left": "issue:43531", + "right": "issue:44877", + "accept": false, + "reason": "Qwen3-MoE sliding_window bug and granite_speech strict-config loading are separate problems." + }, + { + "left": "issue:36032", + "right": "issue:41720", + "accept": false, + "reason": "Tokenizer construction conflict vs Qwen3 auto device mapping cudaErrorAssert are unrelated." + }, + { + "left": "issue:43611", + "right": "issue:44534", + "accept": false, + "reason": "Both are v5 regressions, but base_model_prefix loading and non-persistent buffer junk are different code paths." + }, + { + "left": "issue:43425", + "right": "issue:44403", + "accept": false, + "reason": "Torch 2.10 compatibility and generic loading noise are not the same issue." + }, + { + "left": "issue:43632", + "right": "issue:44534", + "accept": false, + "reason": "_is_hf_initialized handling and non-persistent buffer initialization are different regression classes." + }, + { + "left": "issue:40990", + "right": "issue:44811", + "accept": false, + "reason": "High perplexity on gpt-oss-20b and Whisper batch_decode skip_special_tokens are unrelated." + }, + { + "left": "issue:43792", + "right": "issue:44589", + "accept": false, + "reason": "Whisper runtime failure and missing Float8 storage are unrelated errors." + }, + { + "left": "issue:44079", + "right": "issue:45663", + "accept": false, + "reason": "Different KeyError mechanisms in different models; no shared underlying bug is evident." + }, + { + "left": "issue:44683", + "right": "issue:45020", + "accept": false, + "reason": "Compiled flex_attention on torch>=2.9 and remote_code loading regressions are separate concerns." + }, + { + "left": "issue:41553", + "right": "issue:43582", + "accept": false, + "reason": "AutoTokenizer Voxtral error-message issue and Apple Silicon TypeError are unrelated." + }, + { + "left": "issue:42175", + "right": "issue:45542", + "accept": false, + "reason": "Both touch TensorFlow backend packaging, but one is missing tf in the torch extra and the other is an undefined backend error when only tensorboard is installed; not clearly the same bug." + }, + { + "left": "issue:33666", + "right": "issue:40444", + "accept": false, + "reason": "Both are Qwen VL training issues, but multi-GPU training and multi-image iterable finetuning failures are different problems." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: it correctly characterizes the cluster as mostly unrelated Transformers v5/model-loading regressions and multimodal edge cases, with only a small number of plausible duplicate pairs. The canonical issue choice is defensible as a broad umbrella regression, and the soft-edge verdicts are generally conservative rather than overclaiming duplicates." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 10, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:35707", + "issue:40444", + "issue:41720", + "issue:42175", + "issue:42915", + "issue:43296", + "issue:43317", + "issue:43377", + "issue:43381", + "issue:43421", + "issue:43454", + "issue:43482", + "issue:43525", + "issue:43531", + "issue:43611", + "issue:43632", + "issue:43644", + "issue:43646", + "issue:43749", + "issue:43792", + "issue:43818", + "issue:43856", + "issue:43957", + "issue:43994", + "issue:44079", + "issue:44265", + "issue:44291", + "issue:44403", + "issue:44462", + "issue:44466", + "issue:44534", + "issue:44568", + "issue:44589", + "issue:44829", + "issue:44898", + "issue:44991", + "issue:45072", + "issue:45081", + "issue:45325", + "issue:45357", + "issue:45381", + "issue:45479", + "issue:45491", + "issue:45538", + "issue:45800", + "issue:45812", + "issue:45834", + "issue:45923" + ], + "soft_pairs": [ + "issue:44403|issue:44991", + "issue:43421|issue:44568", + "issue:35707|issue:40444", + "issue:43818|issue:45357", + "issue:43317|issue:43856", + "issue:44462|issue:45812", + "issue:43381|issue:45923", + "issue:43482|issue:45538", + "issue:44291|issue:44589", + "issue:44829|issue:45479", + "issue:43482|issue:43525", + "issue:44079|issue:44534", + "issue:43377|issue:45491", + "issue:43531|issue:44291", + "issue:41720|issue:45081", + "issue:40444|issue:45381", + "issue:43749|issue:44079", + "issue:44265|issue:45072", + "issue:43644|issue:44079", + "issue:42175|issue:43296", + "issue:43632|issue:45072", + "issue:45325|issue:45381", + "issue:43611|issue:43646", + "issue:43792|issue:43957", + "issue:43994|issue:44898", + "issue:45325|issue:45834", + "issue:43454|issue:44466", + "issue:42915|issue:43482", + "issue:43296|issue:45800" + ], + "prepared_review_unit_hash": "4f123c347033a7bd5895d969630551ed9e845ee306f620d21555b6a0f237250c", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36641, + "estimated_input_tokens": 9161, + "estimated_eval_tokens": 18578 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly heterogeneous; the only convincing duplicate-like edge is the pair of Transformers v5 model-initialization/load regressions around custom models/base_model_prefix.", + "confidence": 0.46, + "canonical_issue_reason": "issue:43646 is the broader, more representative statement of the v5 model-initialization regression and can absorb the narrower base_model_prefix-loading report.", + "canonical_pr_reason": null, + "best_issue_reason": "Best anchor for the only close duplicate subcluster; broad enough to represent the v5 init/load breakage.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44403", + "right": "issue:44991", + "accept": false, + "reason": "Both involve tokenization/loading, but the symptoms and target objects differ substantially." + }, + { + "left": "issue:43421", + "right": "issue:44568", + "accept": false, + "reason": "Related tokenizer behavior, but one is runtime post-processor updates and the other is missing BOS/EOS insertion." + }, + { + "left": "issue:35707", + "right": "issue:40444", + "accept": false, + "reason": "Different workflows and models: generation with embeds/past kv vs multimodal finetuning with multiple images." + }, + { + "left": "issue:43818", + "right": "issue:45357", + "accept": false, + "reason": "Both are multimodal model issues, but the concrete bugs are different (video tower weights/temporal attention vs save_pretrained visual keys)." + }, + { + "left": "issue:43317", + "right": "issue:43856", + "accept": false, + "reason": "Device offload/loading bug vs MoE training memory usage; same broad area only." + }, + { + "left": "issue:44462", + "right": "issue:45812", + "accept": false, + "reason": "Both are AutoTokenizer-related, but one ignores tokenizer.json while the other returns wrong Granite token IDs." + }, + { + "left": "issue:43381", + "right": "issue:45923", + "accept": false, + "reason": "One is eval-mode checkpointing behavior; the other is a missing capability flag on a trust_remote_code model." + }, + { + "left": "issue:43482", + "right": "issue:45538", + "accept": false, + "reason": "Different tokenizer/model_max_length problems with no concrete shared failure mode." + }, + { + "left": "issue:44291", + "right": "issue:44589", + "accept": false, + "reason": "Both are init/load regressions, but one is an unexpected argument and the other is a storage type lookup failure." + }, + { + "left": "issue:44829", + "right": "issue:45479", + "accept": false, + "reason": "Training collapse from flash attention vs degenerate zero loss in single-label classification; distinct bugs." + }, + { + "left": "issue:43482", + "right": "issue:43525", + "accept": false, + "reason": "Separate loading failures: GGUF compatibility vs a missing pad_token_id on Llama4Config." + }, + { + "left": "issue:44079", + "right": "issue:44534", + "accept": false, + "reason": "ModelOutput key assignment and non-persistent buffer initialization are unrelated code paths." + }, + { + "left": "issue:43377", + "right": "issue:45491", + "accept": false, + "reason": "Padding-mask batching discrepancy vs NaNs from mixed-length sliding-window attention; different failures." + }, + { + "left": "issue:43531", + "right": "issue:44291", + "accept": false, + "reason": "Sliding-window MoE issue vs init_empty_weights argument regression; not the same bug." + }, + { + "left": "issue:41720", + "right": "issue:45081", + "accept": false, + "reason": "CUDA assert during device mapping is unrelated to the Mistral tokenizer regex crash." + }, + { + "left": "issue:40444", + "right": "issue:45381", + "accept": false, + "reason": "Multi-image finetuning failure vs Qwen2.5-VL video position-id regression." + }, + { + "left": "issue:43749", + "right": "issue:44079", + "accept": false, + "reason": "FSDP CPU RAM-efficient loading and ModelOutput key handling are unrelated." + }, + { + "left": "issue:44265", + "right": "issue:45072", + "accept": false, + "reason": "torch.export compatibility issue vs bfloat16 dtype mismatch; different concrete problems." + }, + { + "left": "issue:43644", + "right": "issue:44079", + "accept": false, + "reason": "Non-persistent buffer junk and ModelOutput key assignment are separate regressions." + }, + { + "left": "issue:42175", + "right": "issue:43296", + "accept": false, + "reason": "Packaging/backend dependency issue vs a specific model load failure in vLLM." + }, + { + "left": "issue:43632", + "right": "issue:45072", + "accept": false, + "reason": "_is_hf_initialized regression and dtype mismatch are not the same underlying bug." + }, + { + "left": "issue:45325", + "right": "issue:45381", + "accept": false, + "reason": "Same model family and positional-id area, but still-image temporal scaling and video vision_position_ids look like distinct symptoms." + }, + { + "left": "issue:43611", + "right": "issue:43646", + "accept": true, + "reason": "Both describe the same Transformers v5 model-init/load regression; the broader issue covers the base_model_prefix-specific report." + }, + { + "left": "issue:43792", + "right": "issue:43957", + "accept": false, + "reason": "Whisper runtime failure vs meta-device loading regression; unrelated." + }, + { + "left": "issue:43994", + "right": "issue:44898", + "accept": false, + "reason": "Different vision-model failures with different repros and failure modes." + }, + { + "left": "issue:45325", + "right": "issue:45834", + "accept": false, + "reason": "Qwen2.5-VL positional-id bug vs Kosmos2.5 OCR index error; no shared code path." + }, + { + "left": "issue:43454", + "right": "issue:44466", + "accept": false, + "reason": "Tied-weight generation bug vs device-dependent serialization inconsistency; related theme, but not the same concrete bug." + }, + { + "left": "issue:42915", + "right": "issue:43482", + "accept": false, + "reason": "Qwen3Moe FP8 failure and GGUF loading failure are unrelated." + }, + { + "left": "issue:43296", + "right": "issue:45800", + "accept": false, + "reason": "Different incompatibility reports with different stacks and failure surfaces." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The assessment is grounded in the packet and stays conservative. The only accepted soft edge, issue:43611 \u2194 issue:43646, is well supported by the titles and the high similarity score as the same Transformers v5 model-init/load regression. The rejected edges are also phrased cautiously and do not overclaim equivalence. The summary matches the evidence that the cluster is mostly heterogeneous with one clear duplicate-like subcluster." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 11, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:34634", + "issue:35707", + "issue:39692", + "issue:40444", + "issue:40990", + "issue:41720", + "issue:42947", + "issue:43257", + "issue:43262", + "issue:43296", + "issue:43317", + "issue:43322", + "issue:43377", + "issue:43454", + "issue:43526", + "issue:43550", + "issue:43572", + "issue:43606", + "issue:43701", + "issue:43720", + "issue:43746", + "issue:43749", + "issue:43819", + "issue:43873", + "issue:43881", + "issue:43883", + "issue:44186", + "issue:44265", + "issue:44291", + "issue:44466", + "issue:44479", + "issue:44488", + "issue:44534", + "issue:44568", + "issue:44610", + "issue:44877", + "issue:44898", + "issue:44991", + "issue:45072", + "issue:45325", + "issue:45357", + "issue:45584", + "issue:45701", + "issue:45759", + "issue:45823", + "issue:45920", + "issue:45923", + "issue:46050" + ], + "soft_pairs": [ + "issue:44479|issue:45823", + "issue:43701|issue:45325", + "issue:44291|issue:44877", + "issue:43606|issue:44265", + "issue:43881|issue:44488", + "issue:43873|issue:45701", + "issue:43262|issue:43746", + "issue:40990|issue:45584", + "issue:34634|issue:35707", + "issue:44568|issue:45920", + "issue:35707|issue:41720", + "issue:44186|issue:44265", + "issue:43606|issue:43819", + "issue:41720|issue:45823", + "issue:43454|issue:43550", + "issue:43377|issue:44610", + "issue:39692|issue:40444", + "issue:43317|issue:43873", + "issue:43454|issue:45072", + "issue:43377|issue:44898", + "issue:43257|issue:45923", + "issue:43572|issue:45759", + "issue:40990|issue:42947", + "issue:43322|issue:46050", + "issue:43526|issue:43883", + "issue:43454|issue:43720", + "issue:41720|issue:43322", + "issue:43296|issue:44877", + "issue:43749|issue:44534", + "issue:43881|issue:44991", + "issue:44466|issue:45357" + ], + "prepared_review_unit_hash": "a949bc1f42f415472518c95ee6f1009aad0a03c99068433d3eb4327bb73d439f", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 37220, + "estimated_input_tokens": 9305, + "estimated_eval_tokens": 18866 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly a loose topical cluster around loading/offloading, quantization, tokenization, and multimodal model regressions, but the soft pairs do not look like true duplicates of the same concrete bug. I\u2019d treat this as a broad, mixed issue cluster rather than a duplicate set.", + "confidence": 0.67, + "canonical_issue_reason": "issue:43873 is the broadest, most central representative of the cluster because it is an open, discussion-heavy report about offloading + quantization, which is the most recurring theme among the nearby issues.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43873 is the best single representative issue because it captures the widest shared theme in this cluster (offloading/quantization) and is more general than the narrower model-specific failures.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44479", + "right": "issue:45823", + "accept": false, + "reason": "Both are multimodal regressions, but they concern different models and different failure modes (video-input regression vs device mismatch in Gemma4). Not the same bug." + }, + { + "left": "issue:43701", + "right": "issue:45325", + "accept": false, + "reason": "Resume-from-checkpoint key mismatch and Qwen2.5-VL rope-index scaling are unrelated code paths." + }, + { + "left": "issue:44291", + "right": "issue:44877", + "accept": false, + "reason": "`init_empty_weights` argument handling and strict config loading for granite_speech are different loading/config issues." + }, + { + "left": "issue:43606", + "right": "issue:44265", + "accept": false, + "reason": "CPU-offload device mismatch in Bark and `torch.export` failure with `torch_compilable_check` are unrelated." + }, + { + "left": "issue:43881", + "right": "issue:44488", + "accept": false, + "reason": "Both are loading failures, but for different models and likely different root causes; too broad to merge." + }, + { + "left": "issue:43873", + "right": "issue:45701", + "accept": false, + "reason": "Quantization/offloading behavior and tokenizer-version changes are different problems." + }, + { + "left": "issue:43262", + "right": "issue:43746", + "accept": false, + "reason": "Audio chat-template sample-rate handling and GraniteSpeech PEFT checkpoint loading are unrelated." + }, + { + "left": "issue:40990", + "right": "issue:45584", + "accept": false, + "reason": "High perplexity on a benchmark and Whisper empty-transcription generation failure are different bugs." + }, + { + "left": "issue:34634", + "right": "issue:35707", + "accept": false, + "reason": "Bark voice preset support and progressive generation with `inputs_embeds`/`past_key_values` are unrelated." + }, + { + "left": "issue:44568", + "right": "issue:45920", + "accept": false, + "reason": "Both involve tokenization, but one is a single tokenizer special-token bug while the other is a broader regression across many models; not clearly the same concrete defect." + }, + { + "left": "issue:35707", + "right": "issue:41720", + "accept": false, + "reason": "Generation with `inputs_embeds`/cache and Qwen3 auto device-mapping CUDA asserts are unrelated." + }, + { + "left": "issue:44186", + "right": "issue:44265", + "accept": false, + "reason": "Tokenizer padding/truncation crash and `torch.export` failures from `torch_compilable_check` are different subsystems." + }, + { + "left": "issue:43606", + "right": "issue:43819", + "accept": false, + "reason": "Bark CPU-offload device mismatch and DAC missing-STE behavior are unrelated model bugs." + }, + { + "left": "issue:41720", + "right": "issue:45823", + "accept": false, + "reason": "Both mention device mapping, but the models and observed failures differ enough that this is not the same bug." + }, + { + "left": "issue:43454", + "right": "issue:43550", + "accept": false, + "reason": "Weight tying in AyaVision and `torch.compile`/SDPA failure in Bamba are unrelated." + }, + { + "left": "issue:43377", + "right": "issue:44610", + "accept": false, + "reason": "MIMI padding-mask inconsistency and OmDet-Turbo input-size mismatch are unrelated preprocessing/model issues." + }, + { + "left": "issue:39692", + "right": "issue:40444", + "accept": false, + "reason": "Documentation/example errors for SigLIP2 and Qwen2.5-VL dataset multi-image finetuning failure are different issues." + }, + { + "left": "issue:43317", + "right": "issue:43873", + "accept": false, + "reason": "Both relate to offloading and quantization, but one is about loading a dequantized model with `device_map=auto` and the other is a broader offloading regression; not enough evidence they are the same concrete bug." + }, + { + "left": "issue:43454", + "right": "issue:45072", + "accept": false, + "reason": "Garbage text from untied `lm_head` weights and dtype mismatches in bfloat16 inference are unrelated." + }, + { + "left": "issue:43377", + "right": "issue:44898", + "accept": false, + "reason": "Padding-mask output drift in MIMI and Perceiver non-default-resolution failure are different model-specific bugs." + }, + { + "left": "issue:43257", + "right": "issue:45923", + "accept": false, + "reason": "Qwen3 MoE weight conversion/loading and missing `supports_gradient_checkpointing` flag are unrelated." + }, + { + "left": "issue:43572", + "right": "issue:45759", + "accept": false, + "reason": "A missing config field in StableLM and `from_config` unwrapping for composite Qwen multimodal configs are different problems." + }, + { + "left": "issue:40990", + "right": "issue:42947", + "accept": false, + "reason": "Perplexity regression and gradient checkpointing with LoRA are unrelated." + }, + { + "left": "issue:43322", + "right": "issue:46050", + "accept": false, + "reason": "Llava-next loading segfault and sinq save/load quantization breakage are different failure modes." + }, + { + "left": "issue:43526", + "right": "issue:43883", + "accept": false, + "reason": "BeitImageProcessorFast label reduction and Molmo tied-weights attribute error are unrelated." + }, + { + "left": "issue:43454", + "right": "issue:43720", + "accept": false, + "reason": "Weight tying serialization and packed-weight unpacking during accelerate loading are different bugs." + }, + { + "left": "issue:41720", + "right": "issue:43322", + "accept": false, + "reason": "Qwen3 auto device-map CUDA assert and Llava Next segfault are different model/load failures." + }, + { + "left": "issue:43296", + "right": "issue:44877", + "accept": false, + "reason": "PaddleOCR-VL load failure in vLLM and strict granite_speech config loading are unrelated." + }, + { + "left": "issue:43749", + "right": "issue:44534", + "accept": false, + "reason": "FSDP CPU RAM-efficient loading and non-persistent buffer junk values are different v5 regressions." + }, + { + "left": "issue:43881", + "right": "issue:44991", + "accept": false, + "reason": "A model loading failure and a tokenizer-loading regression are both loading-related but not the same bug." + }, + { + "left": "issue:44466", + "right": "issue:45357", + "accept": false, + "reason": "Serialization of tied weights and incorrect visual-encoder keys in save_pretrained are different save/export 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": 18, + "review_unit_index": 12, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:36010", + "issue:36331", + "issue:40990", + "issue:41720", + "issue:42617", + "issue:42915", + "issue:43295", + "issue:43316", + "issue:43317", + "issue:43404", + "issue:43450", + "issue:43526", + "issue:43531", + "issue:43550", + "issue:43572", + "issue:43577", + "issue:43606", + "issue:43708", + "issue:43720", + "issue:43873", + "issue:43881", + "issue:44008", + "issue:44186", + "issue:44387", + "issue:44423", + "issue:44479", + "issue:44610", + "issue:44671", + "issue:44743", + "issue:44756", + "issue:44877", + "issue:44898", + "issue:44913", + "issue:44918", + "issue:44977", + "issue:44991", + "issue:45072", + "issue:45081", + "issue:45161", + "issue:45357", + "issue:45406", + "issue:45632", + "issue:45684", + "issue:45701", + "issue:45704", + "issue:45812", + "issue:45823" + ], + "soft_pairs": [ + "issue:41720|issue:44977", + "issue:42617|issue:42915", + "issue:44008|issue:45823", + "issue:40990|issue:45701", + "issue:44387|issue:45704", + "issue:43404|issue:45357", + "issue:44186|issue:45072", + "issue:36331|issue:42617", + "issue:43577|issue:43708", + "issue:44877|issue:44991", + "issue:43873|issue:44756", + "issue:36331|issue:42915", + "issue:43450|issue:45701", + "issue:43550|issue:43720", + "issue:44423|issue:45406", + "issue:45632|issue:45684", + "issue:41720|issue:42617", + "issue:44479|issue:44918", + "issue:44610|issue:44898", + "issue:43550|issue:43606", + "issue:44186|issue:44610", + "issue:42915|issue:43317", + "issue:43295|issue:45701", + "issue:41720|issue:45161", + "issue:44671|issue:44743", + "issue:43526|issue:43881", + "issue:36010|issue:42617", + "issue:45081|issue:45812", + "issue:43531|issue:43572", + "issue:43316|issue:44913" + ], + "prepared_review_unit_hash": "dde36e2275f8963edbe32d8ef0117a805df94b2b7c7e0c6d82a85323b590caf8", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 36282, + "estimated_input_tokens": 9071, + "estimated_eval_tokens": 18398 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is heterogeneous and looks like a loose set of Transformers regressions around tokenization, model loading, offloading/quantization, and processor/model-specific bugs. None of the soft pairs appear to be the same underlying bug or change, so I would not merge any of them as duplicates.", + "confidence": 0.84, + "canonical_issue_reason": "issue:43873 is the best representative because it is the broadest open issue in the offloading/quantization family and has the strongest connectivity to several other memory/offload-related reports, even though the cluster is not a true duplicate set.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43873 is the best single issue to anchor this cluster: it is open, broad, and closest to the recurring offloading/quantization theme that shows up in several nearby reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:41720", + "right": "issue:44977", + "accept": false, + "reason": "Different failure modes: auto device-mapping cuda assert vs flash-attention generation behavior on Qwen3.5. Same family, not the same bug." + }, + { + "left": "issue:42617", + "right": "issue:42915", + "accept": false, + "reason": "Unrelated problems: running 3d_parallel.py versus FineGrainedFP8Config model failure. No shared concrete code path." + }, + { + "left": "issue:44008", + "right": "issue:45823", + "accept": false, + "reason": "Different Gemma models and different bugs: a variable-name collision in Gemma3n forward vs a Gemma4 PLE device mismatch." + }, + { + "left": "issue:40990", + "right": "issue:45701", + "accept": false, + "reason": "One is an evaluation/perplexity anomaly, the other is a tokenization regression across versions. Not the same underlying issue." + }, + { + "left": "issue:44387", + "right": "issue:45704", + "accept": false, + "reason": "Different mechanisms and subsystems: int4 quantization memory growth vs an apex FusedRMSNorm memory leak in T5." + }, + { + "left": "issue:43404", + "right": "issue:45357", + "accept": false, + "reason": "Different models and code paths: Mistral3 lm_head tying bug vs Qwen3.5 visual-encoder key saving regression." + }, + { + "left": "issue:44186", + "right": "issue:45072", + "accept": false, + "reason": "Tokenizer crash in LayoutLMv2 is unrelated to dtype mismatch issues in SwitchTransformers/TimmWrapperModel." + }, + { + "left": "issue:36331", + "right": "issue:42617", + "accept": false, + "reason": "Trainer compute_loss signature regression and 3d_parallel.py runtime failure are unrelated." + }, + { + "left": "issue:43577", + "right": "issue:43708", + "accept": false, + "reason": "BLIP2 dtype propagation and Trainer resume/max_steps calculation are separate bugs in different layers." + }, + { + "left": "issue:44877", + "right": "issue:44991", + "accept": false, + "reason": "Both are loading regressions, but one is strict config validation and the other is tokenizer loading for a different model; not the same fix." + }, + { + "left": "issue:43873", + "right": "issue:44756", + "accept": false, + "reason": "Both mention offloading/memory, but one is quantization offloading behavior and the other is mmap-related OOM on Strix Halo; different root causes." + }, + { + "left": "issue:36331", + "right": "issue:42915", + "accept": false, + "reason": "Trainer loss API regression and Qwen3Moe FP8 loading failure do not share the same code path." + }, + { + "left": "issue:43450", + "right": "issue:45701", + "accept": false, + "reason": "Video processor batch-shape bug is unrelated to a tokenizer version regression." + }, + { + "left": "issue:43550", + "right": "issue:43720", + "accept": false, + "reason": "Different models and failures: torch.compile/SDPA on Bamba vs packed-weight unpacking during accelerate loading for BitNet." + }, + { + "left": "issue:44423", + "right": "issue:45406", + "accept": false, + "reason": "Continuous-batching multimodal crash and Gemma4Processor missing _tokenizer are different processor/serve failures." + }, + { + "left": "issue:45632", + "right": "issue:45684", + "accept": false, + "reason": "Cache-path collision for local models is unrelated to permissions propagation in save_pretrained." + }, + { + "left": "issue:41720", + "right": "issue:42617", + "accept": false, + "reason": "Qwen3 auto device-mapping cuda assert and 3d_parallel.py execution failure are not the same bug." + }, + { + "left": "issue:44479", + "right": "issue:44918", + "accept": false, + "reason": "Video-input regression and trl SFT embedding unpacking are distinct bugs despite both involving multimodal/training workflows." + }, + { + "left": "issue:44610", + "right": "issue:44898", + "accept": false, + "reason": "Both are image-shape related, but one is an OmDet-Turbo processor resolution mismatch and the other is a Perceiver interpolation issue." + }, + { + "left": "issue:43550", + "right": "issue:43606", + "accept": false, + "reason": "Different models and different failures: Bamba torch.compile SDPA issue vs Bark CPU-offload device mismatch." + }, + { + "left": "issue:44186", + "right": "issue:44610", + "accept": false, + "reason": "Tokenizer crash on NER inputs is not the same as an image-processor resolution mismatch." + }, + { + "left": "issue:42915", + "right": "issue:43317", + "accept": false, + "reason": "Both touch quantized/model loading, but one is Qwen3Moe FP8 config failure and the other is dequantized model offload loading. Not a single bug." + }, + { + "left": "issue:43295", + "right": "issue:45701", + "accept": false, + "reason": "Custom processor/tokenizer access regression is different from the broader tokenization-version behavior change." + }, + { + "left": "issue:44671", + "right": "issue:44743", + "accept": false, + "reason": "CamemBERT masked-LM prediction regression and Qwen3.5 recurrent-state reset are unrelated." + }, + { + "left": "issue:43526", + "right": "issue:43881", + "accept": false, + "reason": "BEiT label reduction bug and glm-4v-9b load failure are separate issues." + }, + { + "left": "issue:36010", + "right": "issue:42617", + "accept": false, + "reason": "GenerationMixin import failure is unrelated to the 3d_parallel.py runtime problem." + }, + { + "left": "issue:45081", + "right": "issue:45812", + "accept": false, + "reason": "Both are tokenizer-related, but one is a regex-patch crash during loading and the other is silent token-ID corruption in Granite. Different code paths and fixes." + }, + { + "left": "issue:43531", + "right": "issue:43572", + "accept": false, + "reason": "Sliding-window behavior in Qwen3-MoE and missing pad_token_idx in StableLmConfig are different issues." + }, + { + "left": "issue:43316", + "right": "issue:44913", + "accept": false, + "reason": "Gemma3TextConfig API mismatch and GPTNeoX rotary_pct reload regression are unrelated config bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 13, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:29159", + "issue:30064", + "issue:33357", + "issue:34567", + "issue:34634", + "issue:35707", + "issue:36032", + "issue:41720", + "issue:42490", + "issue:42915", + "issue:43322", + "issue:43329", + "issue:43388", + "issue:43421", + "issue:43531", + "issue:43540", + "issue:43550", + "issue:43606", + "issue:43653", + "issue:43720", + "issue:43746", + "issue:43782", + "issue:43873", + "issue:43927", + "issue:44077", + "issue:44186", + "issue:44361", + "issue:44451", + "issue:44492", + "issue:44509", + "issue:44514", + "issue:44545", + "issue:44589", + "issue:44625", + "issue:44898", + "issue:44991", + "issue:45059", + "issue:45072", + "issue:45381", + "issue:45406", + "issue:45464", + "issue:45479", + "issue:45701", + "issue:45704", + "issue:45759", + "issue:45812", + "issue:45859" + ], + "soft_pairs": [ + "issue:43540|issue:43782", + "issue:42915|issue:43322", + "issue:43421|issue:45812", + "issue:44186|issue:44361", + "issue:43388|issue:45059", + "issue:29159|issue:36032", + "issue:45406|issue:45464", + "issue:30064|issue:34634", + "issue:33357|issue:35707", + "issue:41720|issue:43746", + "issue:43606|issue:43720", + "issue:43550|issue:43653", + "issue:43421|issue:45701", + "issue:43873|issue:45704", + "issue:44492|issue:44509", + "issue:43606|issue:43653", + "issue:34634|issue:41720", + "issue:42490|issue:43329", + "issue:43531|issue:45759", + "issue:34567|issue:35707", + "issue:44514|issue:44545", + "issue:43421|issue:43927", + "issue:44625|issue:45479", + "issue:44898|issue:45072", + "issue:44451|issue:44991", + "issue:43531|issue:44589", + "issue:44077|issue:44625", + "issue:43782|issue:45381", + "issue:45759|issue:45859" + ], + "prepared_review_unit_hash": "fd833ca7908358b643b58713e063661e220a3ad8d86db753b2eb043c1fedbd01", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 35961, + "estimated_input_tokens": 8991, + "estimated_eval_tokens": 18238 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly heterogeneous. The only clear duplicate is the Qwen2_5_VLProcessor batched-input crash pair (44514/44545). The EMBEDDIA/est-roberta and v5 tokenizer-load reports are only softly similar from this packet and should not be treated as a duplicate match.", + "confidence": 0.82, + "canonical_issue_reason": "44545 is the best canonical anchor because it is an exact match to 44514 on the same `Qwen2_5_VLProcessor.apply_chat_template` crash with `padding=False`, with slightly stronger reference signals.", + "canonical_pr_reason": null, + "best_issue_reason": "44545 is the strongest overall issue candidate because it is the clearest confirmed duplicate target in the set and has the strongest duplicate evidence.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43540", + "right": "issue:43782", + "accept": false, + "reason": "Related Qwen multimodal loading failures, but the concrete code paths and symptoms differ." + }, + { + "left": "issue:42915", + "right": "issue:43322", + "accept": false, + "reason": "Different models and failure modes: FP8 config loading vs Llava Next segfault." + }, + { + "left": "issue:43421", + "right": "issue:45812", + "accept": false, + "reason": "Both are tokenizer-related, but one is runtime post-processor behavior and the other is a token-id regression for Granite models." + }, + { + "left": "issue:44186", + "right": "issue:44361", + "accept": false, + "reason": "Both involve tokenizers, but they affect different tokenizers and different crash conditions." + }, + { + "left": "issue:43388", + "right": "issue:45059", + "accept": false, + "reason": "Unrelated bugs: metrics label gathering vs SAM3 text/bounding-box behavior." + }, + { + "left": "issue:29159", + "right": "issue:36032", + "accept": false, + "reason": "Both mention tokenizers, but the first is slow/fast inconsistency and the second is a load-time name collision." + }, + { + "left": "issue:45406", + "right": "issue:45464", + "accept": false, + "reason": "Different products and failure modes: serve processor attribute error vs streaming inference failure." + }, + { + "left": "issue:30064", + "right": "issue:34634", + "accept": false, + "reason": "Different subsystems and bugs: image segmentation-map processing vs Bark voice preset handling." + }, + { + "left": "issue:33357", + "right": "issue:35707", + "accept": false, + "reason": "MacOS CLIP bus error and generation with `inputs_embeds`/`past_key_values` are unrelated." + }, + { + "left": "issue:41720", + "right": "issue:43746", + "accept": false, + "reason": "Different model-specific failures: auto device mapping assert vs local PEFT checkpoint loading." + }, + { + "left": "issue:43606", + "right": "issue:43720", + "accept": false, + "reason": "Both touch loading/offload, but the concrete bugs are device mismatch vs packed-weight unpacking." + }, + { + "left": "issue:43550", + "right": "issue:43653", + "accept": false, + "reason": "Different areas entirely: torch.compile SDPA failure vs tokenizer special-token registration." + }, + { + "left": "issue:43421", + "right": "issue:45701", + "accept": false, + "reason": "A generic tokenization-change report is not the same as the runtime post-processor update bug." + }, + { + "left": "issue:43873", + "right": "issue:45704", + "accept": false, + "reason": "Quantization/offloading behavior and a T5 RMSNorm memory leak are distinct issues." + }, + { + "left": "issue:44492", + "right": "issue:44509", + "accept": false, + "reason": "A cache-strategy typo and docs mentioning removed pipelines are unrelated documentation issues." + }, + { + "left": "issue:43606", + "right": "issue:43653", + "accept": false, + "reason": "Bark CPU-offload device mismatch is unrelated to BigBirdTokenizer mask-token registration." + }, + { + "left": "issue:34634", + "right": "issue:41720", + "accept": false, + "reason": "Different models and bugs: BarkProcessor voice preset vs Qwen3 auto device mapping assert." + }, + { + "left": "issue:42490", + "right": "issue:43329", + "accept": false, + "reason": "Local-path import warning is unrelated to multimodal video token counting." + }, + { + "left": "issue:43531", + "right": "issue:45759", + "accept": false, + "reason": "Both mention Qwen configs, but one is sliding-window behavior and the other is config unwrapping in `from_config`." + }, + { + "left": "issue:34567", + "right": "issue:35707", + "accept": false, + "reason": "Trainer token counting and progressive generation are unrelated." + }, + { + "left": "issue:44514", + "right": "issue:44545", + "accept": true, + "reason": "Same underlying bug and effectively the same report: identical Qwen2_5_VLProcessor batched-input crash with `padding=False`." + }, + { + "left": "issue:43421", + "right": "issue:43927", + "accept": false, + "reason": "Runtime special-token/post-processor handling is not the same as DiaConfig losing custom token IDs on save/load." + }, + { + "left": "issue:44625", + "right": "issue:45479", + "accept": false, + "reason": "Both involve classification configs, but one is Qwen3.5 `num_labels` propagation and the other is a broader loss-behavior bug." + }, + { + "left": "issue:44898", + "right": "issue:45072", + "accept": false, + "reason": "Different model families and failure modes: Perceiver image resolution handling vs bfloat16 dtype mismatch." + }, + { + "left": "issue:44451", + "right": "issue:44991", + "accept": false, + "reason": "Both are tokenizer-load failures, but the packet only shows different models and different concrete failures; that is not enough to conclude the same regression." + }, + { + "left": "issue:43531", + "right": "issue:44589", + "accept": false, + "reason": "Qwen3-MoE sliding-window behavior and a Float8 storage lookup error are unrelated." + }, + { + "left": "issue:44077", + "right": "issue:44625", + "accept": false, + "reason": "PatchTSMixer post-init validation is unrelated to Qwen3.5 label propagation." + }, + { + "left": "issue:43782", + "right": "issue:45381", + "accept": false, + "reason": "Both are Qwen vision-related, but one is a `from_pretrained` weight-only load error and the other is a video-position-id correctness bug." + }, + { + "left": "issue:45759", + "right": "issue:45859", + "accept": false, + "reason": "Config unwrapping for multimodal models and CPU-offload rotary-embedding compatibility are different bugs." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The 44514/44545 duplicate call is well-grounded, but the 44451/44991 pair is overstated. From the packet we only have two different tokenizer-load failures on different models; that is not enough to conclude the same underlying regression. The summary should be more conservative there, e.g. treat it as a soft similarity rather than a clear duplicate." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 14, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:30064", + "issue:36010", + "issue:39692", + "issue:40444", + "issue:41720", + "issue:42886", + "issue:43381", + "issue:43421", + "issue:43531", + "issue:43540", + "issue:43596", + "issue:43618", + "issue:43645", + "issue:43653", + "issue:43697", + "issue:43720", + "issue:43746", + "issue:43819", + "issue:43844", + "issue:43873", + "issue:43994", + "issue:44008", + "issue:44062", + "issue:44186", + "issue:44220", + "issue:44368", + "issue:44451", + "issue:44610", + "issue:44661", + "issue:44671", + "issue:44843", + "issue:44918", + "issue:44936", + "issue:44987", + "issue:45042", + "issue:45081", + "issue:45183", + "issue:45356", + "issue:45357", + "issue:45479", + "issue:45588", + "issue:45663", + "issue:45704", + "issue:45759", + "issue:45859", + "issue:45907", + "issue:45923", + "issue:46050" + ], + "soft_pairs": [ + "issue:44987|issue:45042", + "issue:44368|issue:45357", + "issue:44062|issue:44451", + "issue:45588|issue:45759", + "issue:45704|issue:45923", + "issue:39692|issue:44610", + "issue:43746|issue:46050", + "issue:43421|issue:43653", + "issue:44918|issue:44936", + "issue:45663|issue:45923", + "issue:42886|issue:44843", + "issue:43381|issue:43844", + "issue:43421|issue:44186", + "issue:43531|issue:43645", + "issue:45704|issue:45859", + "issue:41720|issue:45704", + "issue:43994|issue:44671", + "issue:43596|issue:44220", + "issue:43746|issue:43873", + "issue:39692|issue:43873", + "issue:45081|issue:45356", + "issue:40444|issue:43746", + "issue:45356|issue:45907", + "issue:30064|issue:36010", + "issue:43720|issue:43819", + "issue:44220|issue:44661", + "issue:43653|issue:43720", + "issue:43540|issue:44008", + "issue:43618|issue:43697", + "issue:45183|issue:45479" + ], + "prepared_review_unit_hash": "3716efbfc479daffc4a880ce53dd45b88411f72025544cc3b53434f1628b463e", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 37217, + "estimated_input_tokens": 9305, + "estimated_eval_tokens": 18866 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is not a duplicate set: the items span many unrelated bugs/features across tokenization, multimodal models, training, offloading, and documentation. No single issue or PR is a good canonical representative.", + "confidence": 0.98, + "canonical_issue_reason": "No canonical issue: the artifacts describe distinct failures in different subsystems and models, not one underlying bug.", + "canonical_pr_reason": null, + "best_issue_reason": "No issue in this set is broadly representative of the others; choosing one would be arbitrary and misleading.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44987", + "right": "issue:45042", + "accept": false, + "reason": "One is a physical-intelligence model loading failure; the other is a PIL backend image-processor torchvision dependency regression. Different code paths and symptoms." + }, + { + "left": "issue:44368", + "right": "issue:45357", + "accept": false, + "reason": "Qwen3.5 warning about tie_word_embeddings is unrelated to Qwen3.5 save_pretrained writing incorrect visual encoder keys." + }, + { + "left": "issue:44062", + "right": "issue:44451", + "accept": false, + "reason": "Different tokenizer bugs: AddedToken special-flag argument collision vs inability to load a specific model. Not the same underlying defect." + }, + { + "left": "issue:45588", + "right": "issue:45759", + "accept": false, + "reason": "Flash-attention sink-less model crash is unrelated to AutoModelForCausalLM.from_config not unwrapping text_config for multimodal Qwen configs." + }, + { + "left": "issue:45704", + "right": "issue:45923", + "accept": false, + "reason": "T5 apex.FusedRMSNorm memory leak and missing supports_gradient_checkpointing on a trust_remote_code model are unrelated." + }, + { + "left": "issue:39692", + "right": "issue:44610", + "accept": false, + "reason": "SigLIP2 example/model-processor mismatch and quantization errors are unrelated to OmDet-Turbo processor output size mismatch." + }, + { + "left": "issue:43746", + "right": "issue:46050", + "accept": false, + "reason": "PEFT adapter local-checkpoint loading failure is unrelated to sinq quantized model save/load breakage." + }, + { + "left": "issue:43421", + "right": "issue:43653", + "accept": false, + "reason": "Runtime post-processor update for special tokens and BigBirdTokenizer mask-token registration are separate tokenizer behaviors, not one bug." + }, + { + "left": "issue:44918", + "right": "issue:44936", + "accept": false, + "reason": "TRL SFT input-embedding unpacking and trainer.evaluate() after trainer.train() are different trainer workflows with different failure modes." + }, + { + "left": "issue:45663", + "right": "issue:45923", + "accept": false, + "reason": "Gemma-4 FSDP2 shared_kv_states KeyError is unrelated to missing gradient-checkpointing support on a different model." + }, + { + "left": "issue:42886", + "right": "issue:44843", + "accept": false, + "reason": "Offline tokenizer cache loading failure is distinct from unconditional model_info() calls in _patch_mistral_regex." + }, + { + "left": "issue:43381", + "right": "issue:43844", + "accept": false, + "reason": "Gradient checkpointing failing in eval mode is not the same as abnormal gradient increase with HfDeepSpeedConfig + ZeRO-3." + }, + { + "left": "issue:43421", + "right": "issue:44186", + "accept": false, + "reason": "TokenizerBackend special-token post-processor update and LayoutLMv2Tokenizer NER/batched padding crashes are different tokenizer bugs." + }, + { + "left": "issue:43531", + "right": "issue:43645", + "accept": false, + "reason": "Qwen3-MoE sliding_window issue and Jupyter custom-model initialization breakage do not share a concrete code path." + }, + { + "left": "issue:45704", + "right": "issue:45859", + "accept": false, + "reason": "T5 Apex RMSNorm leak and Qwen3.5MoETextRotaryEmbedding CPU-offload incompatibility are unrelated implementation issues." + }, + { + "left": "issue:41720", + "right": "issue:45704", + "accept": false, + "reason": "Qwen3 auto device mapping cudaErrorAssert and T5 apex RMSNorm memory leak are unrelated failures." + }, + { + "left": "issue:43994", + "right": "issue:44671", + "accept": false, + "reason": "SigLIP2 nonsensical outputs with AutoModel/pipeline and CamemBERT masked LM regression are separate model-specific bugs." + }, + { + "left": "issue:43596", + "right": "issue:44220", + "accept": false, + "reason": "DeepSpeed ZeRO-3 init with BertModel index error is unrelated to _torch_extract_fbank_features()." + }, + { + "left": "issue:43746", + "right": "issue:43873", + "accept": false, + "reason": "PEFT adapter loading from local checkpoints and quantization/offloading behavior are different problems, not one mergeable fix." + }, + { + "left": "issue:39692", + "right": "issue:43873", + "accept": false, + "reason": "SigLIP2 docs/model example errors and quantization offloading behavior are too broad and unrelated to be the same bug." + }, + { + "left": "issue:45081", + "right": "issue:45356", + "accept": false, + "reason": "Mistral regex AttributeError when fixing tokenizer regex and Kimi-K2.5 codec/regression warning are different tokenizer paths." + }, + { + "left": "issue:40444", + "right": "issue:43746", + "accept": false, + "reason": "IterableDataset multi-image Qwen2.5-VL finetuning failure is unrelated to GraniteSpeech PEFT local checkpoint loading." + }, + { + "left": "issue:45356", + "right": "issue:45907", + "accept": false, + "reason": "Kimi tokenizer codec/regression issue and list_repo_files failing to forward revision are unrelated areas." + }, + { + "left": "issue:30064", + "right": "issue:36010", + "accept": false, + "reason": "Image processor void segmentation map failure and GenerationMixin import error are completely unrelated." + }, + { + "left": "issue:43720", + "right": "issue:43819", + "accept": false, + "reason": "BitNet packed-weight unpacking during accelerate loading and DAC.from_latents forward mismatch are different code paths and defects." + }, + { + "left": "issue:44220", + "right": "issue:44661", + "accept": false, + "reason": "Audio fbank feature extraction issue and add-new-model-like TOKENIZER_MAPPING_NAMES failure are unrelated." + }, + { + "left": "issue:43653", + "right": "issue:43720", + "accept": false, + "reason": "BigBirdTokenizer special-token registration and BitNet packed-weight loading are unrelated; shared theme is only model loading/tokenization broadly." + }, + { + "left": "issue:43540", + "right": "issue:44008", + "accept": false, + "reason": "Qwen3OmniMoe video-input ValueError and Gemma 3n audio_mel_mask variable collision are different multimodal model bugs." + }, + { + "left": "issue:43618", + "right": "issue:43697", + "accept": false, + "reason": "CLIPOutput attentions assignment regression and RTDetrV2 output differences in v5 are not the same underlying issue." + }, + { + "left": "issue:45183", + "right": "issue:45479", + "accept": false, + "reason": "XOR validation error messaging for input_ids/inputs_embeds and single-label num_labels=1 zero-loss behavior are unrelated classification/model-input 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": 18, + "review_unit_index": 15, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:30064", + "issue:36331", + "issue:39692", + "issue:41720", + "issue:43295", + "issue:43322", + "issue:43335", + "issue:43388", + "issue:43421", + "issue:43531", + "issue:43540", + "issue:43550", + "issue:43577", + "issue:43643", + "issue:43644", + "issue:43653", + "issue:43716", + "issue:43749", + "issue:43756", + "issue:43782", + "issue:43818", + "issue:43819", + "issue:43950", + "issue:44008", + "issue:44062", + "issue:44186", + "issue:44442", + "issue:44448", + "issue:44488", + "issue:44561", + "issue:44671", + "issue:44811", + "issue:44821", + "issue:44843", + "issue:44877", + "issue:44898", + "issue:44977", + "issue:44991", + "issue:45020", + "issue:45072", + "issue:45081", + "issue:45084", + "issue:45381", + "issue:45584", + "issue:45685", + "issue:45834", + "issue:45987", + "issue:46032" + ], + "soft_pairs": [ + "issue:43550|issue:45072", + "issue:43550|issue:43819", + "issue:43388|issue:46032", + "issue:43716|issue:44008", + "issue:43295|issue:45834", + "issue:44062|issue:44488", + "issue:43322|issue:45685", + "issue:43335|issue:43756", + "issue:44186|issue:44898", + "issue:43950|issue:45987", + "issue:43653|issue:43819", + "issue:43531|issue:43643", + "issue:44843|issue:45081", + "issue:43643|issue:44877", + "issue:43540|issue:45381", + "issue:30064|issue:36331", + "issue:44448|issue:44671", + "issue:43577|issue:43818", + "issue:43421|issue:44811", + "issue:43782|issue:44977", + "issue:41720|issue:44991", + "issue:44821|issue:45084", + "issue:39692|issue:45834", + "issue:45584|issue:45834", + "issue:44186|issue:44442", + "issue:43644|issue:43749", + "issue:44561|issue:45020" + ], + "prepared_review_unit_hash": "64b9ac42c5f7521af769ecd4e8d58ef9131d34af2218bfc71b80548826a966bb", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 36044, + "estimated_input_tokens": 9011, + "estimated_eval_tokens": 18278 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Loose cluster around Transformers loading/tokenizer/config regressions, but the items mostly describe different model-specific failures or different code paths. None of the soft pairs look like true duplicates or the same fixable change.", + "confidence": 0.74, + "canonical_issue_reason": "issue:43643 is the best anchor: it\u2019s a concrete, reproducible loading regression in a core API path (`AutoConfig.from_pretrained` + `trust_remote_code`) and sits near several other loading/config-related reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43643 is the strongest global representative because it is specific enough to be actionable yet broad enough to cover a family of nearby loading regressions without being model-specific.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43550", + "right": "issue:45072", + "accept": false, + "reason": "Both mention Bamba/SDPA-style failures, but one is a `torch.compile` + SDPA bug and the other is a bf16 dtype mismatch; different failure modes and fixes." + }, + { + "left": "issue:43550", + "right": "issue:43819", + "accept": false, + "reason": "`torch.compile`/SDPA on Bamba is unrelated to DAC `from_latents` vs forward-pass mismatch; different model families and code paths." + }, + { + "left": "issue:43388", + "right": "issue:46032", + "accept": false, + "reason": "Both are cache/batch-behavior bugs, but one is metrics label truncation and the other is Mamba2 `use_cache` corruption; not the same underlying bug." + }, + { + "left": "issue:43716", + "right": "issue:44008", + "accept": false, + "reason": "Both are dtype/model-input mismatches, but one is Mistral image preprocessor vs model dtype and the other is a Gemma3n variable-name collision; unrelated fixes." + }, + { + "left": "issue:43295", + "right": "issue:45834", + "accept": false, + "reason": "Both involve loading/processing regressions, but one is a processor/tokenizer API break and the other is a Kosmos2.5 OCR index error; no shared code-path." + }, + { + "left": "issue:44062", + "right": "issue:44488", + "accept": false, + "reason": "Both are tokenizer-load failures, but one is an `AddedToken` construction error and the other is a model loading failure for `cjvt/sleng-bert`; different root causes." + }, + { + "left": "issue:43322", + "right": "issue:45685", + "accept": false, + "reason": "Segmentation fault loading Llava Next is unrelated to the MPS histogram dtype error in moe; different subsystems and bug classes." + }, + { + "left": "issue:43335", + "right": "issue:43756", + "accept": false, + "reason": "SwitchTransformers sparse-layer construction and Smollm3 RoPE-layer omission are both model-architecture bugs, but they are not the same defect or fix." + }, + { + "left": "issue:44186", + "right": "issue:44898", + "accept": false, + "reason": "LayoutLMv2 tokenizer NER/padding crash and Perceiver interpolation failure are distinct tokenizer vs vision-model issues." + }, + { + "left": "issue:43950", + "right": "issue:45987", + "accept": false, + "reason": "Both touch cache/buffer contracts, but one is a `from_pretrained` non-persistent-buffer corruption regression and the other is `StaticCache.get_seq_length()` returning the wrong type; not duplicates." + }, + { + "left": "issue:43653", + "right": "issue:43819", + "accept": false, + "reason": "BigBirdTokenizer special-token registration and DAC `from_latents` mismatch are unrelated tokenizer vs audio-model bugs." + }, + { + "left": "issue:43531", + "right": "issue:43643", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior and `AutoConfig.from_pretrained` missing fields are different model/config issues with no common fix." + }, + { + "left": "issue:44843", + "right": "issue:45081", + "accept": false, + "reason": "Offline-mode `model_info()` call in `_patch_mistral_regex` and the `backend_tokenizer` AttributeError are both Mistral-related but distinct failures at different points." + }, + { + "left": "issue:43643", + "right": "issue:44877", + "accept": false, + "reason": "Both concern config/loading, but one is missing fields from `trust_remote_code` configs and the other is strict config blocking `granite_speech`; different symptoms and likely patches." + }, + { + "left": "issue:43540", + "right": "issue:45381", + "accept": false, + "reason": "Qwen3OmniMoe video-input validation and qwen2.5-vl video vision-position ids are separate multimodal bugs." + }, + { + "left": "issue:30064", + "right": "issue:36331", + "accept": false, + "reason": "Image segmentation-map processing and `CustomTrainer.compute_loss()` keyword mismatch are unrelated." + }, + { + "left": "issue:44448", + "right": "issue:44671", + "accept": false, + "reason": "Pegasus v4/v5 output difference and CamemBERT masked-LM predictions are different model behavior regressions, not the same bug." + }, + { + "left": "issue:43577", + "right": "issue:43818", + "accept": false, + "reason": "Blip2 dtype handling and Video-LLaVA temporal-attention/weight-sharing issues are unrelated model-loading defects." + }, + { + "left": "issue:43421", + "right": "issue:44811", + "accept": false, + "reason": "TokenizerBackend post-processor update on special-token changes and Whisper `batch_decode` skipping special tokens are different tokenizer behaviors." + }, + { + "left": "issue:43782", + "right": "issue:44977", + "accept": false, + "reason": "Qwen3VL `weight_only=True` load failure and Qwen3.5 flash-attention generation issue are separate loading vs inference-path bugs." + }, + { + "left": "issue:41720", + "right": "issue:44991", + "accept": false, + "reason": "Qwen3 auto device mapping CUDA assert and tokenizer loading for EMBEDDIA/est-roberta are unrelated failures." + }, + { + "left": "issue:44821", + "right": "issue:45084", + "accept": false, + "reason": "AutoImageProcessor URL loading and template-node compilation errors are unrelated API failures." + }, + { + "left": "issue:39692", + "right": "issue:45834", + "accept": false, + "reason": "SigLIP2 doc-example errors and Kosmos2.5 OCR index error are not the same bug or change." + }, + { + "left": "issue:45584", + "right": "issue:45834", + "accept": false, + "reason": "Whisper empty-transcription generation and Kosmos2.5 long-OCR index error are different multimodal inference bugs." + }, + { + "left": "issue:44186", + "right": "issue:44442", + "accept": false, + "reason": "LayoutLMv2 tokenizer NER crash and FastSpeech2ConformerTokenizer auto-loading failure are both tokenizer-related but not the same concrete problem." + }, + { + "left": "issue:43644", + "right": "issue:43749", + "accept": false, + "reason": "Non-persistent buffer corruption and FSDP CPU RAM-efficient loading are both loading-related, but they affect different mechanisms and fixes." + }, + { + "left": "issue:44561", + "right": "issue:45020", + "accept": false, + "reason": "Both are broad remote-code breakages, but one is `is_torch_fx_available` removal and the other is a general recent-version complaint; too vague 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": 1, + "component_count": 18, + "review_unit_index": 16, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:28218", + "issue:29159", + "issue:34634", + "issue:36010", + "issue:36331", + "issue:42175", + "issue:43262", + "issue:43329", + "issue:43526", + "issue:43577", + "issue:43582", + "issue:43606", + "issue:43650", + "issue:43653", + "issue:43697", + "issue:43720", + "issue:43742", + "issue:43749", + "issue:43782", + "issue:43819", + "issue:43825", + "issue:43827", + "issue:43906", + "issue:43994", + "issue:44060", + "issue:44291", + "issue:44368", + "issue:44479", + "issue:44485", + "issue:44509", + "issue:44554", + "issue:44610", + "issue:44671", + "issue:44805", + "issue:44857", + "issue:44912", + "issue:44936", + "issue:45020", + "issue:45072", + "issue:45081", + "issue:45084", + "issue:45325", + "issue:45406", + "issue:45538", + "issue:45600", + "issue:45701", + "issue:45834" + ], + "soft_pairs": [ + "issue:43606|issue:45072", + "issue:44479|issue:45834", + "issue:43577|issue:43582", + "issue:43825|issue:43994", + "issue:34634|issue:36010", + "issue:43825|issue:43827", + "issue:44485|issue:44509", + "issue:44805|issue:44936", + "issue:44671|issue:45701", + "issue:28218|issue:29159", + "issue:44610|issue:45072", + "issue:43720|issue:45072", + "issue:43782|issue:45325", + "issue:42175|issue:45081", + "issue:43749|issue:44291", + "issue:43582|issue:44857", + "issue:45081|issue:45406", + "issue:43650|issue:43906", + "issue:43262|issue:43819", + "issue:43526|issue:43697", + "issue:44060|issue:44368", + "issue:34634|issue:36331", + "issue:43653|issue:44912", + "issue:45084|issue:45538", + "issue:43526|issue:44554", + "issue:43742|issue:45020", + "issue:43329|issue:45600" + ], + "prepared_review_unit_hash": "9d801691627dc0d817433e5eac028c5a2783eb613d93fbb39754c2c6f0f7eb5e", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 27, + "serialized_chars": 35511, + "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 not a true duplicate set: it spans many unrelated tokenizer, model-loading, processor, and training bugs. A few pairs are superficially similar, but none look like the same concrete underlying defect or a single PR-worthy fix.", + "confidence": 0.93, + "canonical_issue_reason": "No strong canonical issue exists because the items are heterogeneous. If forced, issue 45020 is the broadest umbrella report, but it is still not a duplicate target for the rest of the cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45020 is the most globally representative only in the sense that it is broad and user-facing, but it does not capture the other reports as one bug, so it is a weak canonical choice.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43606", + "right": "issue:45072", + "accept": false, + "reason": "Both mention inference mismatches, but they affect different models and likely different code paths/root causes." + }, + { + "left": "issue:44479", + "right": "issue:45834", + "accept": false, + "reason": "One is a video regression in Qwen VL models; the other is a Kosmos2.5 OCR length/index error. Different bugs." + }, + { + "left": "issue:43577", + "right": "issue:43582", + "accept": false, + "reason": "Blip2 dtype loading and Apple Silicon allocator TypeError are unrelated." + }, + { + "left": "issue:43825", + "right": "issue:43994", + "accept": false, + "reason": "A pipeline error message about removed translation support is unrelated to SigLIP2 producing bad outputs." + }, + { + "left": "issue:34634", + "right": "issue:36010", + "accept": false, + "reason": "BarkProcessor voice preset handling and GenerationMixin import failure are different subsystems and defects." + }, + { + "left": "issue:43825", + "right": "issue:43827", + "accept": false, + "reason": "These are related follow-ups around v5 pipeline removals, but one is an error message issue and the other is docs cleanup, not the same bug." + }, + { + "left": "issue:44485", + "right": "issue:44509", + "accept": false, + "reason": "GLM-5 RoPE implementation discussion is unrelated to documentation still referencing removed pipeline tasks." + }, + { + "left": "issue:44805", + "right": "issue:44936", + "accept": false, + "reason": "Different failures: a mask/index shape mismatch versus trainer.evaluate() breaking after train()." + }, + { + "left": "issue:44671", + "right": "issue:45701", + "accept": false, + "reason": "CamemBERT masked-LM regression and a broad tokenization-change report are not the same concrete issue." + }, + { + "left": "issue:28218", + "right": "issue:29159", + "accept": false, + "reason": "Both are tokenizer-related, but one is about added-token spacing and the other about slow/fast tokenizer inconsistency; too broad to merge." + }, + { + "left": "issue:44610", + "right": "issue:45072", + "accept": false, + "reason": "Different models and different dtype/input-size problems; not the same bug." + }, + { + "left": "issue:43720", + "right": "issue:45072", + "accept": false, + "reason": "BitNet packed-weight loading and generic bfloat16 dtype mismatches are unrelated." + }, + { + "left": "issue:43782", + "right": "issue:45325", + "accept": false, + "reason": "Qwen3VL weight_only loading failure and Qwen2.5-VL rope-index scaling are separate bugs." + }, + { + "left": "issue:42175", + "right": "issue:45081", + "accept": false, + "reason": "Package/backend installation issue versus a tokenizer patch crash; no shared underlying defect." + }, + { + "left": "issue:43749", + "right": "issue:44291", + "accept": false, + "reason": "Both touch model loading, but one is FSDP CPU RAM-efficient loading and the other is init_empty_weights argument handling." + }, + { + "left": "issue:43582", + "right": "issue:44857", + "accept": false, + "reason": "Apple Silicon warmup TypeError and CUDA AMP loss crash are unrelated runtime failures." + }, + { + "left": "issue:45081", + "right": "issue:45406", + "accept": false, + "reason": "Different objects and different missing-attribute failures; not the same code path." + }, + { + "left": "issue:43650", + "right": "issue:43906", + "accept": false, + "reason": "One is a nonspecific request, the other a reproduction of another issue; not duplicates." + }, + { + "left": "issue:43262", + "right": "issue:43819", + "accept": false, + "reason": "Audio chat-template sampling-rate defaulting and DAC latent reconstruction mismatch are unrelated." + }, + { + "left": "issue:43526", + "right": "issue:43697", + "accept": false, + "reason": "BeitImageProcessorFast label reduction and RTDetrV2 output drift are separate bugs." + }, + { + "left": "issue:44060", + "right": "issue:44368", + "accept": false, + "reason": "Both involve tied-weights warnings, but the concrete problems differ and do not look like one mergeable fix." + }, + { + "left": "issue:34634", + "right": "issue:36331", + "accept": false, + "reason": "BarkProcessor voice preset and CustomTrainer.compute_loss signature mismatch are unrelated." + }, + { + "left": "issue:43653", + "right": "issue:44912", + "accept": false, + "reason": "Tokenizer special-token registration and MXFP4 quantization fallback are different issues." + }, + { + "left": "issue:45084", + "right": "issue:45538", + "accept": false, + "reason": "Template compilation error and CLIPTokenizer max-length bug are unrelated." + }, + { + "left": "issue:43526", + "right": "issue:44554", + "accept": false, + "reason": "A label-reduction bug is unrelated to an MPS attention correctness issue." + }, + { + "left": "issue:43742", + "right": "issue:45020", + "accept": false, + "reason": "A specific MobileLLM key error is not the same underlying problem as the broad remote_code regression." + }, + { + "left": "issue:43329", + "right": "issue:45600", + "accept": false, + "reason": "Multimodal token counting bug and removed auto-mapping references 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": 18, + "review_unit_index": 17, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:34567", + "issue:36010", + "issue:36331", + "issue:38175", + "issue:39692", + "issue:40990", + "issue:42490", + "issue:43257", + "issue:43262", + "issue:43329", + "issue:43450", + "issue:43452", + "issue:43454", + "issue:43531", + "issue:43550", + "issue:43577", + "issue:43606", + "issue:43638", + "issue:43643", + "issue:43645", + "issue:43696", + "issue:43720", + "issue:43746", + "issue:43749", + "issue:43819", + "issue:43854", + "issue:43927", + "issue:43931", + "issue:44387", + "issue:44479", + "issue:44568", + "issue:44589", + "issue:44610", + "issue:44617", + "issue:44756", + "issue:44811", + "issue:44849", + "issue:44912", + "issue:44936", + "issue:44977", + "issue:44991", + "issue:45042", + "issue:45356", + "issue:45459", + "issue:45701", + "issue:45759", + "issue:45803", + "issue:45859" + ], + "soft_pairs": [ + "issue:44589|issue:45042", + "issue:43746|issue:44811", + "issue:45459|issue:45803", + "issue:43638|issue:44936", + "issue:43452|issue:44991", + "issue:44387|issue:44617", + "issue:36010|issue:43645", + "issue:43550|issue:44912", + "issue:43819|issue:43927", + "issue:43854|issue:44977", + "issue:43257|issue:43329", + "issue:39692|issue:44912", + "issue:43720|issue:44610", + "issue:38175|issue:44610", + "issue:43454|issue:43577", + "issue:36010|issue:36331", + "issue:43531|issue:45042", + "issue:43257|issue:43696", + "issue:43262|issue:44811", + "issue:43329|issue:44479", + "issue:42490|issue:43257", + "issue:43749|issue:44756", + "issue:43643|issue:45759", + "issue:43329|issue:43450", + "issue:34567|issue:40990", + "issue:43606|issue:45859", + "issue:43257|issue:43931", + "issue:44568|issue:45356", + "issue:44849|issue:45701" + ], + "prepared_review_unit_hash": "99f221b6c374693eaccfdc53c5cd1769c27b79d6a95988430adc737dd47c335a", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36617, + "estimated_input_tokens": 9155, + "estimated_eval_tokens": 18566 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated Transformer bugs with one plausible duplicate pair around the Qwen video-input regression. I would keep only issue:43329 \u2194 issue:44479 as a duplicate/soft-duplicate candidate; the rest differ by model family, failure mode, or code path.", + "confidence": 0.77, + "canonical_issue_reason": "issue:44479 is the best canonical issue in this packet because its title describes a broader video-input regression across several Qwen VL variants, which makes it a reasonable anchor for the accepted soft pair.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44479 is the strongest single issue to represent the cluster because it has the broadest affected-model scope among the paired video-regression reports and reads like the more general symptom-level entry.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44589", + "right": "issue:45042", + "accept": false, + "reason": "Different bug classes: float8 storage lookup vs PIL/torchvision dependency handling." + }, + { + "left": "issue:43746", + "right": "issue:44811", + "accept": false, + "reason": "Both involve loading/processing, but they target different components and failure modes." + }, + { + "left": "issue:45459", + "right": "issue:45803", + "accept": false, + "reason": "Both mention exception handling, but in different code paths and with different symptoms." + }, + { + "left": "issue:43638", + "right": "issue:44936", + "accept": false, + "reason": "Separate training/evaluation failures without a shared concrete root cause." + }, + { + "left": "issue:43452", + "right": "issue:44991", + "accept": false, + "reason": "Both are loading regressions, but for different assets and likely different code paths." + }, + { + "left": "issue:44387", + "right": "issue:44617", + "accept": false, + "reason": "OOM reports for different models and memory paths." + }, + { + "left": "issue:36010", + "right": "issue:43645", + "accept": false, + "reason": "ImportError for GenerationMixin is unrelated to notebook custom-model initialization." + }, + { + "left": "issue:43550", + "right": "issue:44912", + "accept": false, + "reason": "Different model/runtime failures: torch.compile+SDPA vs MXFP4 loading behavior." + }, + { + "left": "issue:43819", + "right": "issue:43927", + "accept": false, + "reason": "Distinct bugs: DAC latent mismatch vs DiaConfig token-ID persistence." + }, + { + "left": "issue:43854", + "right": "issue:44977", + "accept": false, + "reason": "Both are generation/loading issues at a broad level, but the models and symptoms differ." + }, + { + "left": "issue:43257", + "right": "issue:43329", + "accept": false, + "reason": "Qwen3 MoE loading/conversion is unrelated to multimodal video token counting." + }, + { + "left": "issue:39692", + "right": "issue:44912", + "accept": false, + "reason": "Documentation/example failures are not the same as quantized model loading behavior." + }, + { + "left": "issue:43720", + "right": "issue:44610", + "accept": false, + "reason": "Packed-weight loading and image-size mismatch are unrelated problems." + }, + { + "left": "issue:38175", + "right": "issue:44610", + "accept": false, + "reason": "Different model output/shape problems with no clear shared code path." + }, + { + "left": "issue:43454", + "right": "issue:43577", + "accept": false, + "reason": "Weight tying failure and dtype propagation are separate loading defects." + }, + { + "left": "issue:36010", + "right": "issue:36331", + "accept": false, + "reason": "Importing GenerationMixin and custom compute_loss signature handling are unrelated API issues." + }, + { + "left": "issue:43531", + "right": "issue:45042", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior is unrelated to PIL backend requirements." + }, + { + "left": "issue:43257", + "right": "issue:43696", + "accept": false, + "reason": "A loading/conversion bug and a plain CUDA OOM are not the same issue." + }, + { + "left": "issue:43262", + "right": "issue:44811", + "accept": false, + "reason": "Audio chat-template sampling-rate defaults and Whisper batch_decode token skipping are different behaviors." + }, + { + "left": "issue:43329", + "right": "issue:44479", + "accept": true, + "reason": "These look like the same underlying Qwen video-input regression: 43329 points to the broken video-token counting branch, and 44479 reports the broader downstream failure across the same model family." + }, + { + "left": "issue:42490", + "right": "issue:43257", + "accept": false, + "reason": "Local-path import warnings and Qwen3 MoE loading/conversion are unrelated." + }, + { + "left": "issue:43749", + "right": "issue:44756", + "accept": false, + "reason": "Different loading/memory issues: FSDP CPU RAM efficient loading vs mmap on Strix Halo OOM." + }, + { + "left": "issue:43643", + "right": "issue:45759", + "accept": false, + "reason": "Missing config fields with trust_remote_code is different from unwrapping text_config in composite multimodal configs." + }, + { + "left": "issue:43329", + "right": "issue:43450", + "accept": false, + "reason": "Video token counting and batched video processor shape are different code paths." + }, + { + "left": "issue:34567", + "right": "issue:40990", + "accept": false, + "reason": "Trainer token accounting and GPT-oss perplexity are unrelated." + }, + { + "left": "issue:43606", + "right": "issue:45859", + "accept": false, + "reason": "CPU offload device mismatch is unrelated to Qwen3_5Moe rotary embedding compatibility." + }, + { + "left": "issue:43257", + "right": "issue:43931", + "accept": false, + "reason": "Different Qwen loading failures with different symptoms and likely different root causes." + }, + { + "left": "issue:44568", + "right": "issue:45356", + "accept": false, + "reason": "Tokenizer special-token insertion and codec handling regressions are separate bugs." + }, + { + "left": "issue:44849", + "right": "issue:45701", + "accept": false, + "reason": "Hidden-state behavior in Qwen3.5 is not the same as broader tokenization changes across versions." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The cluster-level conclusion is mostly conservative, and the one accepted soft pair (issue:43329 vs issue:44479) is plausible. However, the canonical/best-issue rationale overstates what is supported by the packet: the metadata only shows titles and discussion counts, so claims like \"clearly reproducible,\" \"user-facing,\" and \"umbrella symptom\" are not directly grounded. Please soften those statements to match the evidence available from the packet alone." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 18, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:24643", + "issue:28282", + "issue:29127", + "issue:29159", + "issue:33357", + "issue:34567", + "issue:34689", + "issue:41720", + "issue:42175", + "issue:42915", + "issue:43257", + "issue:43317", + "issue:43404", + "issue:43452", + "issue:43540", + "issue:43577", + "issue:43582", + "issue:43697", + "issue:43742", + "issue:43749", + "issue:43756", + "issue:43782", + "issue:43824", + "issue:43906", + "issue:43927", + "issue:44062", + "issue:44291", + "issue:44479", + "issue:44488", + "issue:44560", + "issue:44568", + "issue:44737", + "issue:44821", + "issue:44912", + "issue:44938", + "issue:44987", + "issue:44991", + "issue:45042", + "issue:45072", + "issue:45137", + "issue:45357", + "issue:45381", + "issue:45405", + "issue:45488", + "issue:45538", + "issue:45588", + "issue:45907", + "issue:45987" + ], + "soft_pairs": [ + "issue:44568|issue:45488", + "issue:29159|issue:34567", + "issue:42915|issue:43742", + "issue:44291|issue:45042", + "issue:42915|issue:44821", + "issue:43452|issue:43582", + "issue:44987|issue:45538", + "issue:41720|issue:43540", + "issue:43697|issue:45357", + "issue:42915|issue:45538", + "issue:43927|issue:44912", + "issue:44479|issue:45381", + "issue:33357|issue:34567", + "issue:42175|issue:43824", + "issue:43540|issue:44479", + "issue:43317|issue:45538", + "issue:43540|issue:44560", + "issue:43257|issue:43756", + "issue:43577|issue:45072", + "issue:44987|issue:45907", + "issue:44062|issue:44991", + "issue:45137|issue:45405", + "issue:24643|issue:29159", + "issue:28282|issue:34689", + "issue:43742|issue:44291", + "issue:42175|issue:44488", + "issue:42915|issue:43749", + "issue:44737|issue:45588", + "issue:43404|issue:45987", + "issue:42175|issue:44938", + "issue:43906|issue:45381", + "issue:28282|issue:29127", + "issue:43582|issue:43782" + ], + "prepared_review_unit_hash": "64efa54ddb92119ebb1d85079f07cb2a53af388dfb2e103738475ba9717b6f3b", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 33, + "serialized_chars": 37685, + "estimated_input_tokens": 9422, + "estimated_eval_tokens": 19100 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab-bag of unrelated Transformers bugs. The only clear duplicate-like pair is the Qwen video-input regression pair; the rest are only superficially similar by subsystem or model family and should stay separate.", + "confidence": 0.89, + "canonical_issue_reason": "issue:44479 is the best representative of the only coherent subgroup: a broad v5.3.0 Qwen video-input regression affecting multiple models, with a narrower companion issue (45381) pointing at the same area.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44479 is the strongest cluster anchor because it describes a concrete regression, spans multiple affected Qwen variants, and has the clearest overlap with the one acceptable soft pair.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44568", + "right": "issue:45488", + "accept": false, + "reason": "Both are tokenizer regressions, but one is about add_special_tokens/BOS-EOS behavior and the other about LlamaTokenizer overriding tokenizer.json pre-tokenization; different bugs." + }, + { + "left": "issue:29159", + "right": "issue:34567", + "accept": false, + "reason": "Tokenizer slow/fast inconsistency and TrainerState token counting are unrelated code paths and symptoms." + }, + { + "left": "issue:42915", + "right": "issue:43742", + "accept": false, + "reason": "Qwen3Moe FP8 loading failure and MobileLLM key error are different model-loading problems." + }, + { + "left": "issue:44291", + "right": "issue:45042", + "accept": false, + "reason": "Unexpected _is_hf_initialized during init_empty_weights is unrelated to the PIL backend's torchvision dependency regression." + }, + { + "left": "issue:42915", + "right": "issue:44821", + "accept": false, + "reason": "Quantization/model-loading failure and AutoImageProcessor URL loading are not the same bug." + }, + { + "left": "issue:43452", + "right": "issue:43582", + "accept": false, + "reason": "gguf_file tokenizer/model loading and AppleSilicon caching_allocator_warmup TypeError are unrelated." + }, + { + "left": "issue:44987", + "right": "issue:45538", + "accept": false, + "reason": "A model-loading regression for a specific repo and CLIPTokenizer's absurd model_max_length are different issues." + }, + { + "left": "issue:41720", + "right": "issue:43540", + "accept": false, + "reason": "Both mention Qwen and failures, but one is device-mapping/cuda assert and the other is video-input processing; not the same bug." + }, + { + "left": "issue:43697", + "right": "issue:45357", + "accept": false, + "reason": "RTDetrV2 output divergence and Qwen3.5 save_pretrained key corruption are unrelated." + }, + { + "left": "issue:42915", + "right": "issue:45538", + "accept": false, + "reason": "Qwen3Moe FP8 loading and CLIP tokenizer max_length are unrelated." + }, + { + "left": "issue:43927", + "right": "issue:44912", + "accept": false, + "reason": "DiaConfig token ID persistence and MXFP4 quantization fallback are different failure modes." + }, + { + "left": "issue:44479", + "right": "issue:45381", + "accept": true, + "reason": "45381 is a narrower manifestation of the same Qwen video-input regression covered by 44479, both centered on broken video position handling in v5.3.0." + }, + { + "left": "issue:33357", + "right": "issue:34567", + "accept": false, + "reason": "MacOS CLIP bus error and TrainerState token counter stalling are unrelated." + }, + { + "left": "issue:42175", + "right": "issue:43824", + "accept": false, + "reason": "Missing TensorFlow in the torch extra and missing Qwen2_5_VLForConditionalGeneration export are separate packaging/import issues." + }, + { + "left": "issue:43540", + "right": "issue:44479", + "accept": false, + "reason": "Both are video-related, but they affect different Qwen model families and likely different code paths." + }, + { + "left": "issue:43317", + "right": "issue:45538", + "accept": false, + "reason": "Device-map/offload dequantization failure is unrelated to CLIPTokenizer's model_max_length bug." + }, + { + "left": "issue:43257", + "right": "issue:43756", + "accept": false, + "reason": "Qwen3 MoE weight conversion under accelerate+deepspeed and Smollm3 RoPE layer loss are unrelated." + }, + { + "left": "issue:43577", + "right": "issue:45072", + "accept": false, + "reason": "Blip2 dtype persistence and SwitchTransformers/TimmWrapper bfloat16 mismatches are distinct dtype issues with different models." + }, + { + "left": "issue:44987", + "right": "issue:45907", + "accept": false, + "reason": "Model loading failure and list_repo_files not forwarding revision are unrelated." + }, + { + "left": "issue:44062", + "right": "issue:44991", + "accept": false, + "reason": "AddedToken special-kwarg TypeError and tokenizer loading failure for est-roberta are not the same defect." + }, + { + "left": "issue:45137", + "right": "issue:45405", + "accept": false, + "reason": "DeepSpeed ZeRO3 deque underflow is unrelated to the PEFT minimum-version pin issue." + }, + { + "left": "issue:24643", + "right": "issue:29159", + "accept": false, + "reason": "DeepSpeed weight dimensionality error and tokenizer slow/fast inconsistency are unrelated." + }, + { + "left": "issue:28282", + "right": "issue:34689", + "accept": false, + "reason": "Missing PyTorch import and Llama 3.2 Vision model-loading regression are different issues." + }, + { + "left": "issue:43742", + "right": "issue:44291", + "accept": false, + "reason": "MobileLLM key error and unexpected _is_hf_initialized during init_empty_weights are different loading bugs." + }, + { + "left": "issue:42175", + "right": "issue:44488", + "accept": false, + "reason": "Torch extra backend packaging and inability to load cjvt/sleng-bert are unrelated." + }, + { + "left": "issue:42915", + "right": "issue:43749", + "accept": false, + "reason": "Qwen3Moe FP8 config failure and FSDP CPU RAM efficient loading broken are different issues." + }, + { + "left": "issue:44737", + "right": "issue:45588", + "accept": false, + "reason": "XLNet CPU-side arange and flash_attention s_aux=None crash are unrelated code paths." + }, + { + "left": "issue:43404", + "right": "issue:45987", + "accept": false, + "reason": "lm_head weight tying in Mistral3 and StaticCache returning a tensor instead of int are unrelated." + }, + { + "left": "issue:43782", + "right": "issue:43582", + "accept": false, + "reason": "Qwen3VL weight_only loading error and AppleSilicon caching_allocator_warmup TypeError are unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The clustering is grounded in the packet: the Qwen v5.3.0 video-input regression pair (44479/45381) is the only clearly related subgroup, and the other soft-edge rejections are conservative and model/code-path specific. No overstatement stands out from the provided titles and descriptions." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 19, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:24643", + "issue:29127", + "issue:34567", + "issue:36331", + "issue:38175", + "issue:41762", + "issue:43452", + "issue:43493", + "issue:43526", + "issue:43531", + "issue:43540", + "issue:43577", + "issue:43643", + "issue:43688", + "issue:43697", + "issue:43720", + "issue:43742", + "issue:43749", + "issue:43756", + "issue:43994", + "issue:44062", + "issue:44291", + "issue:44488", + "issue:44514", + "issue:44589", + "issue:44625", + "issue:44792", + "issue:44821", + "issue:44843", + "issue:44877", + "issue:44898", + "issue:44913", + "issue:44936", + "issue:44964", + "issue:45042", + "issue:45081", + "issue:45216", + "issue:45290", + "issue:45325", + "issue:45357", + "issue:45375", + "issue:45381", + "issue:45405", + "issue:45479", + "issue:45600", + "issue:45823", + "issue:45920", + "issue:46121" + ], + "soft_pairs": [ + "issue:38175|issue:44898", + "issue:43540|issue:45823", + "issue:43493|issue:43994", + "issue:43526|issue:45479", + "issue:44291|issue:44821", + "issue:43756|issue:45325", + "issue:34567|issue:36331", + "issue:44913|issue:46121", + "issue:44062|issue:45081", + "issue:44792|issue:44964", + "issue:45357|issue:45920", + "issue:45042|issue:45081", + "issue:43452|issue:44488", + "issue:43742|issue:44589", + "issue:43531|issue:44821", + "issue:43643|issue:45375", + "issue:45405|issue:45600", + "issue:43720|issue:44898", + "issue:43697|issue:45216", + "issue:44625|issue:44877", + "issue:24643|issue:29127", + "issue:44514|issue:45381", + "issue:41762|issue:44936", + "issue:43688|issue:43749", + "issue:43452|issue:43577", + "issue:43452|issue:44843", + "issue:44514|issue:45290" + ], + "prepared_review_unit_hash": "bdf7406f496e1b7144f52f893e469ccc011a692d222abdaf7b07ad534eb2bdfa", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 36219, + "estimated_input_tokens": 9055, + "estimated_eval_tokens": 18366 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly unrelated issue reports; only the two SigLIP2 reports look like the same underlying model-quality bug. No PRs are present in this cluster.", + "confidence": 0.66, + "canonical_issue_reason": "Issue 45920 is the broadest, most representative open report: it describes a cross-model tokenizer regression with multiple affected checkpoints and clear reproduction potential.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45920 is the best issue candidate because it is open, spans several models, and is phrased as a higher-level regression rather than a narrow single-model symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:38175", + "right": "issue:44898", + "accept": false, + "reason": "Both are vision-model bugs, but they concern different models and different failure modes; not the same underlying defect." + }, + { + "left": "issue:43540", + "right": "issue:45823", + "accept": false, + "reason": "Different models and different runtime problems: video-input handling versus device-mismatch in forward pass." + }, + { + "left": "issue:43493", + "right": "issue:43994", + "accept": true, + "reason": "Both report SigLIP2 producing incorrect outputs; one frames it as HF-vs-JAX discrepancy and the other as nonsensical AutoModel/pipeline results, which plausibly stem from the same model implementation bug." + }, + { + "left": "issue:43526", + "right": "issue:45479", + "accept": false, + "reason": "Unrelated bugs in different components: image processor label reduction vs sequence-classification zero-loss behavior." + }, + { + "left": "issue:44291", + "right": "issue:44821", + "accept": false, + "reason": "Different loading paths and failures; init_empty_weights/_is_hf_initialized is not the same issue as loading AutoImageProcessor from URL." + }, + { + "left": "issue:43756", + "right": "issue:45325", + "accept": false, + "reason": "Both involve RoPE, but they are distinct model-specific bugs with different code paths and symptoms." + }, + { + "left": "issue:34567", + "right": "issue:36331", + "accept": false, + "reason": "TrainerState token counting and custom compute_loss signature regression are separate trainer issues." + }, + { + "left": "issue:44913", + "right": "issue:46121", + "accept": false, + "reason": "Both touch rope-config handling, but one is a reload defaulting bug and the other is a type error during validation; different failures." + }, + { + "left": "issue:44062", + "right": "issue:45081", + "accept": false, + "reason": "Different tokenizer-related crashes in different helpers; no shared concrete bug." + }, + { + "left": "issue:44792", + "right": "issue:44964", + "accept": false, + "reason": "Both are multimodal-model reports, but they concern different models and failures." + }, + { + "left": "issue:45357", + "right": "issue:45920", + "accept": false, + "reason": "Qwen3.5 save_pretrained visual-key corruption is unrelated to the tokenizer ID regression in 45920." + }, + { + "left": "issue:45042", + "right": "issue:45081", + "accept": false, + "reason": "Image-processor torchvision dependency regression is unrelated to the Mistral regex/backend_tokenizer crash." + }, + { + "left": "issue:43452", + "right": "issue:44488", + "accept": false, + "reason": "Both are load failures, but one is about gguf_file/AutoModel loading and the other about a specific model not loading; not enough evidence of the same bug." + }, + { + "left": "issue:43742", + "right": "issue:44589", + "accept": false, + "reason": "KeyError while loading a model and missing Float8 storage are different deserialization issues." + }, + { + "left": "issue:43531", + "right": "issue:44821", + "accept": false, + "reason": "Qwen3-MoE sliding-window behavior and AutoImageProcessor URL loading are unrelated." + }, + { + "left": "issue:43643", + "right": "issue:45375", + "accept": false, + "reason": "Strict config/missing fields with trust_remote_code is not the same as a missing deepstack_visual_indexes field in Qwen3_5MoeVisionConfig." + }, + { + "left": "issue:45405", + "right": "issue:45600", + "accept": false, + "reason": "Release/versioning mismatch for MIN_PEFT_VERSION is unrelated to removed auto_mappings references." + }, + { + "left": "issue:43720", + "right": "issue:44898", + "accept": false, + "reason": "BitNet packed-weight loading and Perceiver positional-encoding failure are different issues." + }, + { + "left": "issue:43697", + "right": "issue:45216", + "accept": false, + "reason": "RTDetrV2 output drift and Qwen3.5 save_pretrained regression are different code paths and symptoms." + }, + { + "left": "issue:44625", + "right": "issue:44877", + "accept": false, + "reason": "Qwen3.5 num_labels propagation and granite_speech strict-config loading are separate config bugs." + }, + { + "left": "issue:24643", + "right": "issue:29127", + "accept": false, + "reason": "DeepSpeed training weight-shape error and LayoutLMv3 error-message clarity are unrelated." + }, + { + "left": "issue:44514", + "right": "issue:45381", + "accept": false, + "reason": "Both concern Qwen2.5-VL, but one is chat-template batching/padding and the other is video vision_position_ids; different bugs." + }, + { + "left": "issue:41762", + "right": "issue:44936", + "accept": false, + "reason": "Gemma3 DeepSpeed ZeRO-3 loading failure and trainer.evaluate() regression are unrelated." + }, + { + "left": "issue:43688", + "right": "issue:43749", + "accept": false, + "reason": "Auxiliary-loss normalization and FSDP CPU RAM efficient loading are different issues." + }, + { + "left": "issue:43452", + "right": "issue:43577", + "accept": false, + "reason": "gguf_file loading breakage and Blip2 dtype propagation are unrelated loading problems." + }, + { + "left": "issue:43452", + "right": "issue:44843", + "accept": false, + "reason": "gguf_file/AutoModel loading failure and offline model_info call in _patch_mistral_regex are different bugs." + }, + { + "left": "issue:44514", + "right": "issue:45290", + "accept": false, + "reason": "Same helper function, but the failing conditions differ materially: batched padding=False vs assistant tool-call messages with no content." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded overall: the selection of issue 45920 is supported by the packet (open, multi-model, v5 regression), and the lone accepted soft pair between the two SigLIP2 reports is a conservative match. The other pairwise rejections are appropriately cautious, and there are no PRs in this cluster." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 20, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:28282", + "issue:29159", + "issue:30064", + "issue:32090", + "issue:33357", + "issue:33666", + "issue:35141", + "issue:38175", + "issue:41720", + "issue:43295", + "issue:43531", + "issue:43550", + "issue:43577", + "issue:43643", + "issue:43646", + "issue:43653", + "issue:43688", + "issue:43697", + "issue:43742", + "issue:43749", + "issue:43819", + "issue:43906", + "issue:43927", + "issue:44361", + "issue:44442", + "issue:44464", + "issue:44466", + "issue:44484", + "issue:44493", + "issue:44561", + "issue:44568", + "issue:44704", + "issue:44743", + "issue:44811", + "issue:44843", + "issue:44936", + "issue:44977", + "issue:44991", + "issue:45005", + "issue:45356", + "issue:45440", + "issue:45491", + "issue:45701", + "issue:45704", + "issue:45823", + "issue:46002", + "issue:46050" + ], + "soft_pairs": [ + "issue:44493|issue:46050", + "issue:44466|issue:45005", + "issue:43819|issue:44811", + "issue:30064|issue:33666", + "issue:44936|issue:45823", + "issue:44484|issue:44704", + "issue:43295|issue:45356", + "issue:44464|issue:44743", + "issue:43653|issue:44361", + "issue:44361|issue:44442", + "issue:28282|issue:32090", + "issue:28282|issue:33357", + "issue:41720|issue:43906", + "issue:38175|issue:43295", + "issue:28282|issue:35141", + "issue:45491|issue:46002", + "issue:43577|issue:43697", + "issue:43653|issue:44568", + "issue:43688|issue:45704", + "issue:43688|issue:45440", + "issue:44843|issue:44991", + "issue:43646|issue:43749", + "issue:43688|issue:45701", + "issue:29159|issue:33357", + "issue:43531|issue:43742", + "issue:43653|issue:43927", + "issue:43643|issue:44561", + "issue:44464|issue:44977", + "issue:43550|issue:43927" + ], + "prepared_review_unit_hash": "ac81d24d938d679b1d12a3af3a5b61f79311008003b27770de18da239fe5481f", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 35812, + "estimated_input_tokens": 8953, + "estimated_eval_tokens": 18162 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most pairs are clearly unrelated. The tied-weights pair is the only close match, but with just titles and no shared targets or code evidence, it is safer to treat it as related rather than a confirmed duplicate.", + "confidence": 0.76, + "canonical_issue_reason": "issue:45005 is the best umbrella issue by title because it is broader and more general, but the evidence is only enough to justify cautious relatedness, not a definitive subsumption of issue:44466.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45005 is the strongest single issue to keep because it names the v5 tied-weights problem at a broader level and is the likeliest representative of the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44493", + "right": "issue:46050", + "accept": false, + "reason": "Different failure modes: position-id key handling vs quantized save/load breakage." + }, + { + "left": "issue:44466", + "right": "issue:45005", + "accept": false, + "reason": "The titles are closely related, but the packet does not provide enough evidence to confirm the same underlying bug or a clear duplicate relationship." + }, + { + "left": "issue:43819", + "right": "issue:44811", + "accept": false, + "reason": "Different code paths: DAC latent decoding vs Whisper batch decoding behavior." + }, + { + "left": "issue:30064", + "right": "issue:33666", + "accept": false, + "reason": "Segmentation-map preprocessing and multi-GPU training are unrelated problems." + }, + { + "left": "issue:44936", + "right": "issue:45823", + "accept": false, + "reason": "Trainer lifecycle failure vs device-mapping mismatch in Gemma4." + }, + { + "left": "issue:44484", + "right": "issue:44704", + "accept": false, + "reason": "A parameter/documentation question vs a cached_file kwargs-forwarding bug." + }, + { + "left": "issue:43295", + "right": "issue:45356", + "accept": false, + "reason": "Both are tokenizer-related regressions, but they affect different models and mechanisms." + }, + { + "left": "issue:44464", + "right": "issue:44743", + "accept": false, + "reason": "Both involve generation/state behavior, but the concrete bugs differ." + }, + { + "left": "issue:43653", + "right": "issue:44361", + "accept": false, + "reason": "Different tokenizer failures: special-token registration vs task-time AttributeError." + }, + { + "left": "issue:44361", + "right": "issue:44442", + "accept": false, + "reason": "MLukeTokenizer task failure and FastSpeech2ConformerTokenizer loading are unrelated." + }, + { + "left": "issue:28282", + "right": "issue:32090", + "accept": false, + "reason": "Import-time PyTorch missing error is unrelated to a Trainer broadcast TypeError." + }, + { + "left": "issue:28282", + "right": "issue:33357", + "accept": false, + "reason": "Dependency import error and MacOS bus error are different failure modes." + }, + { + "left": "issue:41720", + "right": "issue:43906", + "accept": false, + "reason": "The reproduction issue does not clearly match the specific CUDA assert / device-mapping bug." + }, + { + "left": "issue:38175", + "right": "issue:43295", + "accept": false, + "reason": "Zero probabilities in SigLIP2 are unrelated to the processor/tokenizer regression." + }, + { + "left": "issue:28282", + "right": "issue:35141", + "accept": false, + "reason": "PyTorch import failure and tied-embedding reinitialization are unrelated." + }, + { + "left": "issue:45491", + "right": "issue:46002", + "accept": false, + "reason": "Different NaN sources: sliding-window attention vs classification-head init." + }, + { + "left": "issue:43577", + "right": "issue:43697", + "accept": false, + "reason": "Blip2 dtype-casting and RTDetr output divergence are different bugs." + }, + { + "left": "issue:43653", + "right": "issue:44568", + "accept": false, + "reason": "Both mention special tokens, but they concern different token-handling bugs." + }, + { + "left": "issue:43688", + "right": "issue:45704", + "accept": false, + "reason": "Aux-loss normalization and apex RMSNorm memory leak are unrelated." + }, + { + "left": "issue:43688", + "right": "issue:45440", + "accept": false, + "reason": "MoE auxiliary-loss scaling and DeepSeekV3 implementation divergence are not the same issue." + }, + { + "left": "issue:44843", + "right": "issue:44991", + "accept": false, + "reason": "Offline hub lookup bug and tokenizer-load regression for a specific model are different code paths." + }, + { + "left": "issue:43646", + "right": "issue:43749", + "accept": false, + "reason": "Custom model initialization breakage and FSDP CPU RAM-efficient loading are unrelated." + }, + { + "left": "issue:29159", + "right": "issue:33357", + "accept": false, + "reason": "Tokenizer inconsistency and a MacOS bus error do not share the same root cause." + }, + { + "left": "issue:43531", + "right": "issue:43742", + "accept": false, + "reason": "Qwen3 sliding-window behavior and MobileLLM loading key errors are different problems." + }, + { + "left": "issue:43653", + "right": "issue:43927", + "accept": false, + "reason": "Special-token registration and saved custom token ID persistence are related broadly, but not the same concrete bug." + }, + { + "left": "issue:43550", + "right": "issue:43927", + "accept": false, + "reason": "Torch.compile/SDPA failure is unrelated to DiaConfig token-ID persistence." + }, + { + "left": "issue:43643", + "right": "issue:44561", + "accept": false, + "reason": "Missing fields with trust_remote_code and removal of is_torch_fx_available are different regressions." + }, + { + "left": "issue:44464", + "right": "issue:44977", + "accept": false, + "reason": "Chunked generation with compiled forward is not the same as flash-attention generation on Qwen3.5." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The tied-weights match (issue:44466 vs issue:45005) is plausible from the titles, but the packet only supports cautious relatedness, not the stronger claim that 45005 clearly subsumes 44466 or that they are definitively the same underlying bug. The rest of the soft-edge false verdicts are conservative enough, but this accepted pair overstates the evidence, so I can\u2019t accept the summary as written." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 21, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:28218", + "issue:32090", + "issue:33357", + "issue:33666", + "issue:34567", + "issue:36010", + "issue:36331", + "issue:39692", + "issue:41720", + "issue:43122", + "issue:43262", + "issue:43278", + "issue:43295", + "issue:43452", + "issue:43493", + "issue:43504", + "issue:43525", + "issue:43550", + "issue:43572", + "issue:43577", + "issue:43606", + "issue:43688", + "issue:43720", + "issue:43742", + "issue:43756", + "issue:43825", + "issue:43906", + "issue:43927", + "issue:44361", + "issue:44451", + "issue:44464", + "issue:44466", + "issue:44479", + "issue:44554", + "issue:44568", + "issue:44610", + "issue:44625", + "issue:44821", + "issue:44849", + "issue:44913", + "issue:44964", + "issue:45357", + "issue:45405", + "issue:45588", + "issue:45812", + "issue:45823", + "issue:46002", + "issue:46050" + ], + "soft_pairs": [ + "issue:44479|issue:44625", + "issue:43550|issue:44361", + "issue:43577|issue:45588", + "issue:41720|issue:44464", + "issue:43577|issue:44554", + "issue:32090|issue:34567", + "issue:43122|issue:43906", + "issue:43504|issue:43720", + "issue:43504|issue:43550", + "issue:43504|issue:44610", + "issue:32090|issue:33666", + "issue:44466|issue:46050", + "issue:33666|issue:36010", + "issue:43742|issue:44821", + "issue:45823|issue:46050", + "issue:43452|issue:44451", + "issue:43606|issue:43927", + "issue:43720|issue:43927", + "issue:43278|issue:43688", + "issue:43262|issue:43295", + "issue:44849|issue:45357", + "issue:43525|issue:43572", + "issue:33666|issue:36331", + "issue:43493|issue:43825", + "issue:43720|issue:44361", + "issue:44625|issue:46002", + "issue:43756|issue:44913", + "issue:44568|issue:45812", + "issue:28218|issue:33357", + "issue:44964|issue:45405", + "issue:39692|issue:43493" + ], + "prepared_review_unit_hash": "2bb8e27306b5c7264f2984a01db02689089e9b690e8bc298dac00dd2fa3f33c4", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 37103, + "estimated_input_tokens": 9276, + "estimated_eval_tokens": 18808 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are a grab-bag of unrelated Transformers bugs and regressions. None of the soft pairs look like the same underlying issue or change, so all candidate duplicate links should be rejected.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44479", + "right": "issue:44625", + "accept": false, + "reason": "Different Qwen3/VL regression vs config propagation bug; no shared code-path or fix." + }, + { + "left": "issue:43550", + "right": "issue:44361", + "accept": false, + "reason": "Bamba torch.compile/SDPA failure and MLukeTokenizer task AttributeError are unrelated." + }, + { + "left": "issue:43577", + "right": "issue:45588", + "accept": false, + "reason": "Blip2 dtype loading bug vs flash_attention s_aux=None crash; different subsystems." + }, + { + "left": "issue:41720", + "right": "issue:44464", + "accept": false, + "reason": "Qwen3 device-mapping CUDA assert and chunked-generation compiled-forward inconsistency are distinct bugs." + }, + { + "left": "issue:43577", + "right": "issue:44554", + "accept": false, + "reason": "dtype loading issue is unrelated to an MPS attention correctness bug." + }, + { + "left": "issue:32090", + "right": "issue:34567", + "accept": false, + "reason": "Trainer broadcast TypeError and num_input_tokens_seen bookkeeping are separate trainer issues." + }, + { + "left": "issue:43122", + "right": "issue:43906", + "accept": false, + "reason": "Tokenization-version regression and an isolated reproduction are not enough to confirm the same bug." + }, + { + "left": "issue:43504", + "right": "issue:43720", + "accept": false, + "reason": "BEiT legacy-field loading failure is unrelated to BitNet packed-weight loading." + }, + { + "left": "issue:43504", + "right": "issue:43550", + "accept": false, + "reason": "Pretrained preset load failure and Bamba SDPA compile failure affect different paths." + }, + { + "left": "issue:43504", + "right": "issue:44610", + "accept": false, + "reason": "Legacy-field load bug vs OmDet-Turbo input-size mismatch are unrelated." + }, + { + "left": "issue:32090", + "right": "issue:33666", + "accept": false, + "reason": "Trainer broadcast error and Qwen2-VL multi-GPU training are different problems." + }, + { + "left": "issue:44466", + "right": "issue:46050", + "accept": false, + "reason": "Serialization of tied lm_head weights is not the same as sinq quantized save/load breakage." + }, + { + "left": "issue:33666", + "right": "issue:36010", + "accept": false, + "reason": "Multi-GPU training issue and GenerationMixin import error do not share a concrete cause." + }, + { + "left": "issue:43742", + "right": "issue:44821", + "accept": false, + "reason": "MobileLLM key error and AutoImageProcessor URL loading failure are unrelated loading bugs." + }, + { + "left": "issue:45823", + "right": "issue:46050", + "accept": false, + "reason": "Gemma4 device-map mismatch and quantized model save/load corruption are different failure modes." + }, + { + "left": "issue:43452", + "right": "issue:44451", + "accept": false, + "reason": "gguf_file loader breakage is not the same as ScandiBERT loading failure." + }, + { + "left": "issue:43606", + "right": "issue:43927", + "accept": false, + "reason": "CPU offload device mismatch and DiaConfig token-ID loss on save/load are distinct." + }, + { + "left": "issue:43720", + "right": "issue:43927", + "accept": false, + "reason": "Packed-weight unpacking during accelerate loading is unrelated to DiaConfig serialization of custom token IDs." + }, + { + "left": "issue:43278", + "right": "issue:43688", + "accept": false, + "reason": "Embedding dtype drift and auxiliary-loss normalization are different training behaviors." + }, + { + "left": "issue:43262", + "right": "issue:43295", + "accept": false, + "reason": "Audio chat-template sampling-rate default and processor.tokenizer/image handling regression are different processor bugs." + }, + { + "left": "issue:44849", + "right": "issue:45357", + "accept": false, + "reason": "Qwen3.5 hidden-states bug and visual-encoder save_pretrained regression are not the same issue." + }, + { + "left": "issue:43525", + "right": "issue:43572", + "accept": false, + "reason": "Missing pad_token_id in Llama4Config and missing pad_token_idx in StableLmConfig are similar in theme but different concrete bugs." + }, + { + "left": "issue:33666", + "right": "issue:36331", + "accept": false, + "reason": "Qwen2-VL multi-GPU training and compute_loss signature mismatch are unrelated trainer/model issues." + }, + { + "left": "issue:43493", + "right": "issue:43825", + "accept": false, + "reason": "SigLIP2 implementation discrepancy and pipeline translation-message wording are unrelated." + }, + { + "left": "issue:43720", + "right": "issue:44361", + "accept": false, + "reason": "BitNet loading/unpacking and MLukeTokenizer task AttributeError do not share a code path." + }, + { + "left": "issue:44625", + "right": "issue:46002", + "accept": false, + "reason": "num_labels propagation and classification-head NaNs are different bugs." + }, + { + "left": "issue:43756", + "right": "issue:44913", + "accept": false, + "reason": "Smollm3 RoPE-layer count issue and GPTNeoX rotary_pct reload bug are not the same defect." + }, + { + "left": "issue:44568", + "right": "issue:45812", + "accept": false, + "reason": "mdeberta special-token regression and Granite wrong token IDs are separate tokenizer bugs." + }, + { + "left": "issue:28218", + "right": "issue:33357", + "accept": false, + "reason": "Tokenizer added-space bug and MacOS CLIP bus error are unrelated." + }, + { + "left": "issue:44964", + "right": "issue:45405", + "accept": false, + "reason": "Phi-4 multimodal load failure and unreleased MIN_PEFT_VERSION bump are unrelated." + }, + { + "left": "issue:39692", + "right": "issue:43493", + "accept": false, + "reason": "SigLIP2 docs example errors and SigLIP2 implementation discrepancy are different issues: documentation vs model behavior." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 22, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:29942", + "issue:30333", + "issue:33357", + "issue:33666", + "issue:39692", + "issue:43257", + "issue:43278", + "issue:43295", + "issue:43299", + "issue:43334", + "issue:43493", + "issue:43504", + "issue:43540", + "issue:43572", + "issue:43575", + "issue:43606", + "issue:43646", + "issue:43653", + "issue:43701", + "issue:43756", + "issue:43824", + "issue:43844", + "issue:43927", + "issue:43994", + "issue:44112", + "issue:44291", + "issue:44361", + "issue:44423", + "issue:44464", + "issue:44466", + "issue:44488", + "issue:44493", + "issue:44589", + "issue:44734", + "issue:44792", + "issue:44843", + "issue:44977", + "issue:45083", + "issue:45200", + "issue:45229", + "issue:45405", + "issue:45520", + "issue:45584", + "issue:45759", + "issue:45803", + "issue:45823", + "issue:45987" + ], + "soft_pairs": [ + "issue:43540|issue:44977", + "issue:43493|issue:43844", + "issue:45200|issue:45229", + "issue:43278|issue:43844", + "issue:43334|issue:43572", + "issue:45803|issue:45987", + "issue:44488|issue:44843", + "issue:43257|issue:43299", + "issue:43756|issue:45083", + "issue:43334|issue:45759", + "issue:39692|issue:43295", + "issue:45229|issue:45823", + "issue:43646|issue:44466", + "issue:43824|issue:44589", + "issue:33357|issue:33666", + "issue:43504|issue:43606", + "issue:44589|issue:45520", + "issue:43572|issue:44291", + "issue:43927|issue:44361", + "issue:44464|issue:45584", + "issue:43606|issue:44112", + "issue:43504|issue:43653", + "issue:43575|issue:45405", + "issue:43278|issue:43701", + "issue:44792|issue:45405", + "issue:43994|issue:44493", + "issue:44423|issue:44734", + "issue:29942|issue:30333" + ], + "prepared_review_unit_hash": "2879e682be0b243af5041e7f813092ad7e27a6cdd0b4b03ed26ea5167b74024a", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 35667, + "estimated_input_tokens": 8917, + "estimated_eval_tokens": 18090 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No true duplicates here: the soft pairs are mostly superficial title similarity across unrelated regressions, model-loading issues, tokenizer/config bugs, and CI/test failures. Treat the items as separate issues rather than one dedup cluster.", + "confidence": 0.89, + "canonical_issue_reason": "No single canonical issue stands out; the cluster is too heterogeneous to safely collapse into one duplicate representative.", + "canonical_pr_reason": null, + "best_issue_reason": "If forced to pick one representative, 'Transformers 5.0.0 breaks custom model initialization' is the broadest version-5 loading regression, but it still does not cover the rest of the items as duplicates.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43540", + "right": "issue:44977", + "accept": false, + "reason": "Different models and failure modes: video-input processing vs flash-attention generation." + }, + { + "left": "issue:43493", + "right": "issue:43844", + "accept": false, + "reason": "Both are training-related, but SigLIP2 implementation discrepancy is not the same bug as ZeRO-3 gradient growth." + }, + { + "left": "issue:45200", + "right": "issue:45229", + "accept": false, + "reason": "Same model family, but missing text-only token defaults is unrelated to multi-GPU inference OOM." + }, + { + "left": "issue:43278", + "right": "issue:43844", + "accept": false, + "reason": "Embedding dtype drift during evaluate is a different code-path from DeepSpeed gradient instability." + }, + { + "left": "issue:43334", + "right": "issue:43572", + "accept": false, + "reason": "Both are config-field regressions, but they affect different models and different missing attributes." + }, + { + "left": "issue:45803", + "right": "issue:45987", + "accept": false, + "reason": "Unrelated components: Fuyu tensor conversion vs StaticCache API return-type contract." + }, + { + "left": "issue:44488", + "right": "issue:44843", + "accept": false, + "reason": "Both are loading bugs, but offline hub access and the mistral-regex patch are different root causes." + }, + { + "left": "issue:43257", + "right": "issue:43299", + "accept": false, + "reason": "Related Qwen3 MoE area, but accelerate+deepspeed weight conversion is not the same as Qwen3VL loading failure." + }, + { + "left": "issue:43756", + "right": "issue:45083", + "accept": false, + "reason": "Different subsystems entirely: RoPE-layer handling vs feature-extraction length helper behavior." + }, + { + "left": "issue:43334", + "right": "issue:45759", + "accept": false, + "reason": "Both touch Qwen multimodal config loading, but missing pad_token_id and not unwrapping text_config are distinct bugs." + }, + { + "left": "issue:39692", + "right": "issue:43295", + "accept": false, + "reason": "Both mention processor/tokenizer mismatch, but one is a docs/quantization example problem and the other is a runtime regression." + }, + { + "left": "issue:45229", + "right": "issue:45823", + "accept": false, + "reason": "Same Gemma4 family, but CUDA OOM and device mismatch are different failure modes." + }, + { + "left": "issue:43646", + "right": "issue:44466", + "accept": false, + "reason": "Both are v5 regressions, but custom model initialization is unrelated to lm_head serialization/tied-weight behavior." + }, + { + "left": "issue:43824", + "right": "issue:44589", + "accept": false, + "reason": "Different loading/storage issues with different modules and symptoms." + }, + { + "left": "issue:33357", + "right": "issue:33666", + "accept": false, + "reason": "Both are multimodal issues, but a Mac bus error in CLIP is not the same as Qwen2-VL multi-GPU training." + }, + { + "left": "issue:43504", + "right": "issue:43606", + "accept": false, + "reason": "Legacy-field load failure and CPU-offload device mismatch are separate bugs." + }, + { + "left": "issue:44589", + "right": "issue:45520", + "accept": false, + "reason": "Both are low-level import/storage errors, but they are different code-paths and causes." + }, + { + "left": "issue:43572", + "right": "issue:44291", + "accept": false, + "reason": "Both are 5.0-era init regressions, but missing pad_token_idx and unexpected _is_hf_initialized argument are different issues." + }, + { + "left": "issue:43927", + "right": "issue:44361", + "accept": false, + "reason": "Token-ID persistence across save/load is unrelated to MLLukeTokenizer task-time AttributeError." + }, + { + "left": "issue:44464", + "right": "issue:45584", + "accept": false, + "reason": "Compiled-forward chunked generation and empty-transcription token alignment are different bugs." + }, + { + "left": "issue:43606", + "right": "issue:44112", + "accept": false, + "reason": "A real CPU-offload mismatch is not the same as a stale GraniteSpeech CI test." + }, + { + "left": "issue:43504", + "right": "issue:43653", + "accept": false, + "reason": "Pretrained-model legacy field handling and BigBird special-token registration are unrelated." + }, + { + "left": "issue:43575", + "right": "issue:45405", + "accept": false, + "reason": "Tensor-parallel OOM and unreleased dependency version bump are not duplicates." + }, + { + "left": "issue:43278", + "right": "issue:43701", + "accept": false, + "reason": "Training/eval dtype drift is unrelated to resume_from_checkpoint key mismatch." + }, + { + "left": "issue:44792", + "right": "issue:45405", + "accept": false, + "reason": "janus image-generation test failure has nothing to do with the PEFT version bump." + }, + { + "left": "issue:43994", + "right": "issue:44493", + "accept": false, + "reason": "SigLIP2 AutoModel/pipeline correctness and position-id warnings are different bugs." + }, + { + "left": "issue:44423", + "right": "issue:44734", + "accept": false, + "reason": "Both are serve crashes, but continuous-batching string handling and KV-cache indexing are distinct code-path failures." + }, + { + "left": "issue:29942", + "right": "issue:30333", + "accept": false, + "reason": "Completely unrelated: Flash Attention 2 tests vs MLflow reporting after training failure." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 23, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:15354", + "issue:28218", + "issue:29159", + "issue:32090", + "issue:34567", + "issue:37428", + "issue:38175", + "issue:42490", + "issue:43232", + "issue:43295", + "issue:43450", + "issue:43493", + "issue:43540", + "issue:43550", + "issue:43572", + "issue:43582", + "issue:43696", + "issue:43701", + "issue:43824", + "issue:43844", + "issue:43873", + "issue:43957", + "issue:44315", + "issue:44466", + "issue:44521", + "issue:44554", + "issue:44610", + "issue:44849", + "issue:44898", + "issue:44964", + "issue:44977", + "issue:44987", + "issue:45072", + "issue:45084", + "issue:45127", + "issue:45198", + "issue:45200", + "issue:45290", + "issue:45325", + "issue:45356", + "issue:45406", + "issue:45412", + "issue:45446", + "issue:45520", + "issue:45588", + "issue:45823", + "issue:46050", + "issue:46121" + ], + "soft_pairs": [ + "issue:43873|issue:46050", + "issue:45406|issue:46121", + "issue:45588|issue:46050", + "issue:42490|issue:43696", + "issue:44521|issue:45290", + "issue:43844|issue:44964", + "issue:28218|issue:34567", + "issue:29159|issue:32090", + "issue:45084|issue:45520", + "issue:44554|issue:45446", + "issue:43582|issue:44977", + "issue:43540|issue:45200", + "issue:15354|issue:38175", + "issue:45200|issue:45823", + "issue:43232|issue:44849", + "issue:45072|issue:45198", + "issue:37428|issue:43957", + "issue:43493|issue:43701", + "issue:44315|issue:44849", + "issue:43295|issue:44610", + "issue:43572|issue:44987", + "issue:43295|issue:43450", + "issue:44554|issue:45356", + "issue:43572|issue:43957", + "issue:37428|issue:43824", + "issue:45200|issue:45325", + "issue:43295|issue:43873", + "issue:44898|issue:45412", + "issue:44466|issue:45127", + "issue:43550|issue:43844" + ], + "prepared_review_unit_hash": "ad0c694f0fd3c8cee93ccce42744ad069c9acbfe1ca4d66f14ea4a43a03e11c2", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36893, + "estimated_input_tokens": 9224, + "estimated_eval_tokens": 18704 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a heterogeneous false-positive cluster: the items span unrelated Transformers bugs across tokenization, generation, loading, multimodal processors, quantization, and training. I do not see any soft pair that looks like the same underlying bug/change.", + "confidence": 0.96, + "canonical_issue_reason": "No single canonical issue is safe to choose here; the cluster is too mixed and the soft links do not describe true duplicates.", + "canonical_pr_reason": null, + "best_issue_reason": "No best issue stands out as a representative duplicate target because the items cover several unrelated failure modes.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43873", + "right": "issue:46050", + "accept": false, + "reason": "Both involve quantization/offloading, but one is an offloading behavior issue and the other is broken save/load after sinq quantization." + }, + { + "left": "issue:45406", + "right": "issue:46121", + "accept": false, + "reason": "Gemma4 processor `_tokenizer` access and rope-validation list handling are unrelated bugs in different code paths." + }, + { + "left": "issue:45588", + "right": "issue:46050", + "accept": false, + "reason": "A flash-attention `s_aux=None` crash is unrelated to quantized model persistence after save/load." + }, + { + "left": "issue:42490", + "right": "issue:43696", + "accept": false, + "reason": "A trailing-slash import warning and a CUDA OOM are clearly unrelated." + }, + { + "left": "issue:44521", + "right": "issue:45290", + "accept": false, + "reason": "Both touch chat templating, but one is multimodal assistant-mask generation and the other is tool-call/no-content tokenization handling." + }, + { + "left": "issue:43844", + "right": "issue:44964", + "accept": false, + "reason": "ZeRO-3 gradient growth and a Phi-4 multimodal loading failure are unrelated." + }, + { + "left": "issue:28218", + "right": "issue:34567", + "accept": false, + "reason": "Tokenizer spacing after added tokens does not match the TrainerState token-count update bug." + }, + { + "left": "issue:29159", + "right": "issue:32090", + "accept": false, + "reason": "Slow/fast tokenizer inconsistency is unrelated to the NoneType broadcast error in Trainer." + }, + { + "left": "issue:45084", + "right": "issue:45520", + "accept": false, + "reason": "Template-node compilation failure and a Python 3.13 flash_attn import KeyError are unrelated." + }, + { + "left": "issue:44554", + "right": "issue:45446", + "accept": false, + "reason": "An MPS attention correctness issue is unrelated to the flex_attention PyTorch version check." + }, + { + "left": "issue:43582", + "right": "issue:44977", + "accept": false, + "reason": "Apple Silicon allocator warmup TypeError and Qwen3.5 flash-attention generation bugs are unrelated." + }, + { + "left": "issue:43540", + "right": "issue:45200", + "accept": false, + "reason": "Video-input handling in Qwen3OmniMoe is unrelated to Gemma 4 text-only `mm_token_type_ids` defaults." + }, + { + "left": "issue:15354", + "right": "issue:38175", + "accept": false, + "reason": "TorchScript GeneratorExp export failure and SigLIP2 zero probabilities are unrelated." + }, + { + "left": "issue:45200", + "right": "issue:45823", + "accept": false, + "reason": "Gemma4 text-only defaults and Gemma4 device-map forward mismatch are different bugs in different code paths." + }, + { + "left": "issue:43232", + "right": "issue:44849", + "accept": false, + "reason": "Generation kwargs after `sync_gpus` and Qwen3.5 hidden-state output are unrelated generation issues." + }, + { + "left": "issue:45072", + "right": "issue:45198", + "accept": false, + "reason": "dtype mismatches in bfloat16 inference are unrelated to Wav2Vec2 save/tokenization failures." + }, + { + "left": "issue:37428", + "right": "issue:43957", + "accept": false, + "reason": "A missing flash-attention import and meta-device model loading failures are different problems." + }, + { + "left": "issue:43493", + "right": "issue:43701", + "accept": false, + "reason": "SigLIP2 implementation discrepancy is unrelated to checkpoint key mismatches on resume." + }, + { + "left": "issue:44315", + "right": "issue:44849", + "accept": false, + "reason": "Liger Kernel not applied in `model_init` and Qwen3.5 hidden-state output are unrelated." + }, + { + "left": "issue:43295", + "right": "issue:44610", + "accept": false, + "reason": "Processor/tokenizer regression and OmDet-Turbo image-size mismatch are different processor bugs." + }, + { + "left": "issue:43572", + "right": "issue:44987", + "accept": false, + "reason": "Missing `pad_token_idx` in StableLmConfig and a transformers>=5.1 loading failure are unrelated." + }, + { + "left": "issue:43295", + "right": "issue:43450", + "accept": false, + "reason": "Processor/tokenizer regression is unrelated to video processor batch-shape errors." + }, + { + "left": "issue:44554", + "right": "issue:45356", + "accept": false, + "reason": "An MPS attention correctness bug is unrelated to a Kimi-K2.5 tokenizer regression." + }, + { + "left": "issue:43572", + "right": "issue:43957", + "accept": false, + "reason": "A config default bug and a model-loading meta-device regression are not the same issue." + }, + { + "left": "issue:37428", + "right": "issue:43824", + "accept": false, + "reason": "Both are import errors, but they involve different missing symbols and different root causes." + }, + { + "left": "issue:45200", + "right": "issue:45325", + "accept": false, + "reason": "Gemma4 `mm_token_type_ids` defaults and Qwen2.5-VL rope-index scaling are unrelated multimodal bugs." + }, + { + "left": "issue:43295", + "right": "issue:43873", + "accept": false, + "reason": "Processor/tokenizer regression is unrelated to quantization offloading behavior." + }, + { + "left": "issue:44898", + "right": "issue:45412", + "accept": false, + "reason": "Perceiver resolution handling and RT-DETR memory release are unrelated." + }, + { + "left": "issue:44466", + "right": "issue:45127", + "accept": false, + "reason": "Tied-weight serialization differences and LoRA merge collapse are different bugs despite both involving weights." + }, + { + "left": "issue:43550", + "right": "issue:43844", + "accept": false, + "reason": "torch.compile SDPA failure and ZeRO-3 gradient growth are unrelated training/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": 18, + "review_unit_index": 24, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:28218", + "issue:35141", + "issue:37428", + "issue:38175", + "issue:42490", + "issue:42915", + "issue:43262", + "issue:43278", + "issue:43334", + "issue:43335", + "issue:43388", + "issue:43425", + "issue:43450", + "issue:43493", + "issue:43525", + "issue:43526", + "issue:43550", + "issue:43575", + "issue:43606", + "issue:43653", + "issue:43819", + "issue:43824", + "issue:43825", + "issue:43827", + "issue:43856", + "issue:43901", + "issue:43975", + "issue:44112", + "issue:44206", + "issue:44442", + "issue:44448", + "issue:44514", + "issue:44521", + "issue:44610", + "issue:44743", + "issue:44792", + "issue:44857", + "issue:45083", + "issue:45127", + "issue:45198", + "issue:45245", + "issue:45290", + "issue:45381", + "issue:45412", + "issue:45520", + "issue:45563", + "issue:45584", + "issue:45593" + ], + "soft_pairs": [ + "issue:44792|issue:45563", + "issue:43334|issue:43525", + "issue:43824|issue:43975", + "issue:42915|issue:43824", + "issue:43388|issue:43526", + "issue:43450|issue:43575", + "issue:42490|issue:43493", + "issue:43278|issue:43856", + "issue:43975|issue:45520", + "issue:43653|issue:44112", + "issue:28218|issue:35141", + "issue:38175|issue:43493", + "issue:44442|issue:45198", + "issue:44514|issue:44521", + "issue:38175|issue:45412", + "issue:44206|issue:45584", + "issue:43827|issue:43901", + "issue:43262|issue:43450", + "issue:43819|issue:44112", + "issue:44610|issue:44857", + "issue:43819|issue:45083", + "issue:37428|issue:43425", + "issue:43335|issue:43550", + "issue:43606|issue:45083", + "issue:44857|issue:45593", + "issue:45127|issue:45245", + "issue:45290|issue:45381", + "issue:44448|issue:44743", + "issue:43278|issue:43825" + ], + "prepared_review_unit_hash": "f7b3011ff890841b68050d1c2ca734e59954fecac20c9293dcc17e0263afb650", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36535, + "estimated_input_tokens": 9134, + "estimated_eval_tokens": 18524 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is mostly a grab bag of unrelated Transformer bugs and regressions. None of the soft pairs are strong duplicates under a strict same-root-cause standard. No PRs are present; issue 44521 is the best single issue to anchor on because it has the broadest, most active bug report among the items.", + "confidence": 0.89, + "canonical_issue_reason": "Issue 44521 has the richest discussion/inbound references and a broad, clearly stated bug in a widely used API, making it the best anchor despite the cluster being heterogeneous.", + "canonical_pr_reason": null, + "best_issue_reason": "44521 is the most representative and actionable issue in the set; it is active, well-described, and general enough to serve as the closest canonical issue.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44792", + "right": "issue:45563", + "accept": false, + "reason": "Both involve generation-related behavior, but one is a Janus image-generation test and the other is a stale paging warning; different code paths and symptoms." + }, + { + "left": "issue:43334", + "right": "issue:43525", + "accept": false, + "reason": "Same missing-attribute symptom, but for different config classes/models; no evidence they share the same underlying bug." + }, + { + "left": "issue:43824", + "right": "issue:43975", + "accept": false, + "reason": "An import error for a Qwen VL class and a detokenization regression in DeepSeek are unrelated." + }, + { + "left": "issue:42915", + "right": "issue:43824", + "accept": false, + "reason": "FineGrainedFP8Config failure for Qwen3Moe is unrelated to a missing import for a Qwen VL model class." + }, + { + "left": "issue:43388", + "right": "issue:43526", + "accept": false, + "reason": "Both concern labels, but one is evaluation batch gathering and the other is BEiT image preprocessing; different bug surfaces." + }, + { + "left": "issue:43450", + "right": "issue:43575", + "accept": false, + "reason": "Batched video processor shape handling and tensor-parallel OOM are different failure modes in different areas." + }, + { + "left": "issue:42490", + "right": "issue:43493", + "accept": false, + "reason": "Local-path import warning and SigLIP2 parity discrepancy are unrelated." + }, + { + "left": "issue:43278", + "right": "issue:43856", + "accept": false, + "reason": "Dtype drift in evaluate and MoE memory usage are different issues." + }, + { + "left": "issue:43975", + "right": "issue:45520", + "accept": false, + "reason": "Detokenization regression and a Python 3.13 flash_attn import KeyError are unrelated." + }, + { + "left": "issue:43653", + "right": "issue:44112", + "accept": false, + "reason": "Tokenizer special-token registration and a stale CI device-override test do not describe the same bug." + }, + { + "left": "issue:28218", + "right": "issue:35141", + "accept": false, + "reason": "Tokenizer spacing after added tokens and embedding reinitialization after resize are distinct tokenizer/model-state issues." + }, + { + "left": "issue:38175", + "right": "issue:43493", + "accept": false, + "reason": "Both are about SigLIP2, but one is zero probabilities and the other is a general HF-vs-JAX discrepancy; not enough to treat as the same defect." + }, + { + "left": "issue:44442", + "right": "issue:45198", + "accept": false, + "reason": "Different audio-tokenizer/model workflows with different failures; no shared code path is evident." + }, + { + "left": "issue:44514", + "right": "issue:44521", + "accept": false, + "reason": "Both touch apply_chat_template, but one crashes on batched padding=False input while the other produces all-zero assistant masks for multimodal inputs; not the same bug." + }, + { + "left": "issue:38175", + "right": "issue:45412", + "accept": false, + "reason": "SigLIP2 probability issues and RT-DETR memory-release behavior are unrelated." + }, + { + "left": "issue:44206", + "right": "issue:45584", + "accept": false, + "reason": "A feature extractor unsupported-arg crash and Whisper empty-transcription generation failure are separate problems." + }, + { + "left": "issue:43827", + "right": "issue:43901", + "accept": false, + "reason": "Both are doc mismatches, but they cover different pipeline APIs and different stale behavior; not one underlying change." + }, + { + "left": "issue:43262", + "right": "issue:43450", + "accept": false, + "reason": "Audio chat-template sampling-rate defaulting and video processor batched-shape handling are unrelated." + }, + { + "left": "issue:43819", + "right": "issue:44112", + "accept": false, + "reason": "DAC latent mismatch and a GraniteSpeech CI test issue are different code paths." + }, + { + "left": "issue:44610", + "right": "issue:44857", + "accept": false, + "reason": "Processor image-size mismatch and AMP crash in a different loss are distinct." + }, + { + "left": "issue:43819", + "right": "issue:45083", + "accept": false, + "reason": "DAC latent conversion and qwen3_omni_moe feature-extraction length behavior are unrelated." + }, + { + "left": "issue:37428", + "right": "issue:43425", + "accept": false, + "reason": "An import error for one helper is not the same as broad Torch 2.10 incompatibility." + }, + { + "left": "issue:43335", + "right": "issue:43550", + "accept": false, + "reason": "A sparse-layer config bug and a torch.compile/SDPA failure in Bamba are different issues." + }, + { + "left": "issue:43606", + "right": "issue:45083", + "accept": false, + "reason": "CPU-offload device mismatch and qwen3_omni_moe feature-length behavior are unrelated." + }, + { + "left": "issue:44857", + "right": "issue:45593", + "accept": false, + "reason": "A float16 AMP crash and missing auxiliary losses are both detection-related but clearly different defects." + }, + { + "left": "issue:45127", + "right": "issue:45245", + "accept": false, + "reason": "LoRA/extended-vocab collapse and a category-count runtime limit are unrelated." + }, + { + "left": "issue:45290", + "right": "issue:45381", + "accept": false, + "reason": "Assistant tool-call chat-template crash and VL video position-id bug are different multimodal issues." + }, + { + "left": "issue:44448", + "right": "issue:44743", + "accept": false, + "reason": "Pegasus v4/v5 output drift and Qwen3_5 recurrent-state reset are unrelated." + }, + { + "left": "issue:43824", + "right": "issue:45083", + "accept": false, + "reason": "A missing Qwen2.5-VL import and qwen3_omni_moe feature-length behavior are unrelated." + }, + { + "left": "issue:43278", + "right": "issue:43825", + "accept": false, + "reason": "Embedding dtype drift and a pipeline error-message 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": 18, + "review_unit_index": 25, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:28218", + "issue:29127", + "issue:29942", + "issue:32090", + "issue:34689", + "issue:35141", + "issue:37428", + "issue:38175", + "issue:39692", + "issue:40444", + "issue:42490", + "issue:43335", + "issue:43504", + "issue:43526", + "issue:43550", + "issue:43606", + "issue:43611", + "issue:43653", + "issue:43701", + "issue:43720", + "issue:43723", + "issue:43819", + "issue:43854", + "issue:43931", + "issue:44112", + "issue:44206", + "issue:44292", + "issue:44351", + "issue:44360", + "issue:44448", + "issue:44462", + "issue:44493", + "issue:44554", + "issue:44561", + "issue:44779", + "issue:44855", + "issue:44857", + "issue:44977", + "issue:45072", + "issue:45083", + "issue:45200", + "issue:45229", + "issue:45310", + "issue:45478", + "issue:45520", + "issue:45588", + "issue:45823", + "issue:45859" + ], + "soft_pairs": [ + "issue:29942|issue:32090", + "issue:45200|issue:45588", + "issue:37428|issue:44351", + "issue:43611|issue:43931", + "issue:44206|issue:45823", + "issue:43335|issue:43653", + "issue:43550|issue:44112", + "issue:43819|issue:44493", + "issue:44855|issue:45083", + "issue:43335|issue:43504", + "issue:45588|issue:45859", + "issue:29127|issue:32090", + "issue:43335|issue:45083", + "issue:39692|issue:43526", + "issue:45310|issue:45478", + "issue:39692|issue:44493", + "issue:44360|issue:44561", + "issue:43720|issue:44112", + "issue:44292|issue:45520", + "issue:43335|issue:44112", + "issue:43335|issue:45072", + "issue:44448|issue:44779", + "issue:42490|issue:43854", + "issue:44977|issue:45229", + "issue:38175|issue:44493", + "issue:43606|issue:44857", + "issue:34689|issue:35141", + "issue:44857|issue:45072", + "issue:44448|issue:44977", + "issue:40444|issue:43526", + "issue:28218|issue:32090", + "issue:43701|issue:45083", + "issue:44561|issue:44779", + "issue:43723|issue:44554", + "issue:44462|issue:44561", + "issue:43550|issue:45083" + ], + "prepared_review_unit_hash": "ff18e0df2cba74b745b002ce2a94aea40dabb7d261ee2789bdc40e923860a87d", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 36, + "serialized_chars": 38447, + "estimated_input_tokens": 9612, + "estimated_eval_tokens": 19480 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is mostly unrelated issues; the only clear duplicate pair is the Qwen3.5 MoE from_pretrained regression (45310/45478). No PRs are present.", + "confidence": 0.86, + "canonical_issue_reason": "issue:45310 is the earliest and slightly more detailed report for the only true duplicate pair in the set, so it is the best canonical issue to anchor that regression.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45310 is the clearest global anchor here: it matches its duplicate exactly, names the concrete regression, and is the earliest report among the pair.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:29942", + "right": "issue:32090", + "accept": false, + "reason": "Different bugs: Flash Attention 2 test failures vs Trainer/_gpu_broadcast_one NoneType broadcast error." + }, + { + "left": "issue:45200", + "right": "issue:45588", + "accept": false, + "reason": "Different code paths and failures: Gemma4 token-type IDs vs flash_attention.py AttributeError on s_aux=None." + }, + { + "left": "issue:37428", + "right": "issue:44351", + "accept": false, + "reason": "Both are import errors, but for different missing symbols and likely different underlying breakages." + }, + { + "left": "issue:43611", + "right": "issue:43931", + "accept": false, + "reason": "Model-loading regression with base_model_prefix is not the same as a Qwen3-VL weight shape mismatch." + }, + { + "left": "issue:44206", + "right": "issue:45823", + "accept": false, + "reason": "Unsupported center arg in a feature extractor is unrelated to Gemma4 PLE device mismatch during forward." + }, + { + "left": "issue:43335", + "right": "issue:43653", + "accept": false, + "reason": "SwitchTransformers sparse-layer config bug and BigBirdTokenizer special-token bug are unrelated." + }, + { + "left": "issue:43550", + "right": "issue:44112", + "accept": false, + "reason": "torch.compile/SDPA failure in Bamba is unrelated to a GraniteSpeech CI device-override test issue." + }, + { + "left": "issue:43819", + "right": "issue:44493", + "accept": false, + "reason": "DAC.from_latents/STE mismatch is a different bug than unexpected position-id keys in many models." + }, + { + "left": "issue:44855", + "right": "issue:45083", + "accept": false, + "reason": "Python 3.13 IndentationError in DebertaV2Model is unrelated to qwen3_omni_moe length helper behavior." + }, + { + "left": "issue:43335", + "right": "issue:43504", + "accept": false, + "reason": "Different BEIT/SwitchTransformers model-loading/configuration problems." + }, + { + "left": "issue:45588", + "right": "issue:45859", + "accept": false, + "reason": "Both involve Qwen-related attention code, but one is a flash_attention s_aux=None crash and the other is CPU-offload rotary embedding incompatibility." + }, + { + "left": "issue:29127", + "right": "issue:32090", + "accept": false, + "reason": "LayoutLMv3 error messaging and Trainer GPU broadcast TypeError are unrelated." + }, + { + "left": "issue:43335", + "right": "issue:45083", + "accept": false, + "reason": "SwitchTransformers sparse-layer creation and qwen3_omni_moe feature-length helper are different bugs." + }, + { + "left": "issue:39692", + "right": "issue:43526", + "accept": false, + "reason": "SigLIP2 docs/example issues are unrelated to BeitImageProcessorFast reduce_labels behavior." + }, + { + "left": "issue:45310", + "right": "issue:45478", + "accept": true, + "reason": "Same concrete regression, same title, same version range, and 45478 is a later duplicate report of 45310." + }, + { + "left": "issue:39692", + "right": "issue:44493", + "accept": false, + "reason": "Documentation/example errors in SigLIP2 are unrelated to position-id key warnings." + }, + { + "left": "issue:44360", + "right": "issue:44561", + "accept": false, + "reason": "DSA indexer ReLU discussion is unrelated to removal of is_torch_fx_available breaking trust_remote_code models." + }, + { + "left": "issue:43720", + "right": "issue:44112", + "accept": false, + "reason": "Packed-weights loading failure in BitNet is unrelated to a stale CI device-override test." + }, + { + "left": "issue:44292", + "right": "issue:45520", + "accept": false, + "reason": "Qwen-3-8B-NVFP4 runtime error is different from the Python 3.13 flash_attn import KeyError." + }, + { + "left": "issue:43335", + "right": "issue:44112", + "accept": false, + "reason": "Different subsystems and failure modes: SwitchTransformers config bug vs GraniteSpeech CI device mismatch." + }, + { + "left": "issue:43335", + "right": "issue:45072", + "accept": false, + "reason": "Sparse-layer config and bfloat16 dtype mismatch are unrelated." + }, + { + "left": "issue:44448", + "right": "issue:44779", + "accept": false, + "reason": "Pegasus output drift across versions is not the same as Deepseek tokenizer regression." + }, + { + "left": "issue:42490", + "right": "issue:43854", + "accept": false, + "reason": "Local-path import warning and GLM-4.7-Flash loading failure are distinct model-loading issues." + }, + { + "left": "issue:44977", + "right": "issue:45229", + "accept": false, + "reason": "Qwen3.5 flash-attention generation bug and Gemma4 multi-GPU CUDA OOM are different problems." + }, + { + "left": "issue:38175", + "right": "issue:44493", + "accept": false, + "reason": "SigLIP2 zero probabilities and unexpected position-id keys are unrelated." + }, + { + "left": "issue:43606", + "right": "issue:44857", + "accept": false, + "reason": "CPU-offload device mismatch in bark-small is unrelated to LwDetrImageLoss AMP crash." + }, + { + "left": "issue:34689", + "right": "issue:35141", + "accept": false, + "reason": "Llama 3.2 Vision model-loading regression and embedding reinitialization after resize are different bugs." + }, + { + "left": "issue:44857", + "right": "issue:45072", + "accept": false, + "reason": "float16 AMP crash in LwDetrImageLoss is unrelated to bfloat16 dtype mismatches." + }, + { + "left": "issue:44448", + "right": "issue:44977", + "accept": false, + "reason": "Pegasus v4/v5 output differences and Qwen3.5 flash-attention generation failures are unrelated." + }, + { + "left": "issue:40444", + "right": "issue:43526", + "accept": false, + "reason": "IterableDataset multi-image finetuning failure is unrelated to BeitImageProcessorFast label reduction." + }, + { + "left": "issue:28218", + "right": "issue:32090", + "accept": false, + "reason": "Tokenizer added-token spacing bug is unrelated to the Trainer broadcast NoneType error." + }, + { + "left": "issue:43701", + "right": "issue:45083", + "accept": false, + "reason": "Checkpoint key mismatch is unrelated to qwen3_omni_moe helper output lengths." + }, + { + "left": "issue:44561", + "right": "issue:44779", + "accept": false, + "reason": "Both mention version 5 regressions, but one is an API removal breakage and the other is tokenizer output drift." + }, + { + "left": "issue:43723", + "right": "issue:44554", + "accept": false, + "reason": "AutoTokenizer loading issue in v5 is unrelated to an MPS attention correctness problem." + }, + { + "left": "issue:44462", + "right": "issue:44561", + "accept": false, + "reason": "Ignoring tokenizer.json and removal of is_torch_fx_available are different regression classes." + }, + { + "left": "issue:43550", + "right": "issue:45083", + "accept": false, + "reason": "Bamba SDPA compile failure is unrelated to qwen3_omni_moe feature-length helper behavior." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet: all items are issues, and 45310/45478 is the only exact-title duplicate pair with the earlier report correctly identified as canonical. The soft-edge verdicts are conservative and the distinctions are generally supported by the titles/body metadata." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 26, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:10000", + "issue:15354", + "issue:25251", + "issue:29942", + "issue:32090", + "issue:33357", + "issue:35141", + "issue:37428", + "issue:41720", + "issue:42490", + "issue:43065", + "issue:43278", + "issue:43329", + "issue:43381", + "issue:43412", + "issue:43575", + "issue:43653", + "issue:43873", + "issue:43927", + "issue:43976", + "issue:43994", + "issue:44186", + "issue:44206", + "issue:44361", + "issue:44448", + "issue:44466", + "issue:44479", + "issue:44568", + "issue:44743", + "issue:44857", + "issue:44871", + "issue:44977", + "issue:44987", + "issue:45198", + "issue:45229", + "issue:45245", + "issue:45372", + "issue:45406", + "issue:45412", + "issue:45440", + "issue:45459", + "issue:45491", + "issue:45563", + "issue:45588", + "issue:45698", + "issue:45712", + "issue:45803", + "issue:45823" + ], + "soft_pairs": [ + "issue:32090|issue:33357", + "issue:29942|issue:35141", + "issue:37428|issue:44987", + "issue:43873|issue:45229", + "issue:43927|issue:45698", + "issue:41720|issue:45229", + "issue:43278|issue:43329", + "issue:44977|issue:45563", + "issue:44568|issue:45491", + "issue:10000|issue:15354", + "issue:45372|issue:45588", + "issue:42490|issue:43575", + "issue:43412|issue:45588", + "issue:45198|issue:45803", + "issue:44857|issue:45588", + "issue:43976|issue:45440", + "issue:45372|issue:45406", + "issue:45459|issue:45712", + "issue:45588|issue:45823", + "issue:43278|issue:43381", + "issue:44448|issue:44466", + "issue:43653|issue:44743", + "issue:45245|issue:45563", + "issue:25251|issue:32090", + "issue:45372|issue:45459", + "issue:44206|issue:44479", + "issue:43065|issue:43994", + "issue:44186|issue:45198", + "issue:43065|issue:45412", + "issue:44871|issue:44977", + "issue:44361|issue:45198" + ], + "prepared_review_unit_hash": "acc87e52515044ec5c2a808ca31bef2fbe61d739d9ce258a2d6b9b267a555248", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 37309, + "estimated_input_tokens": 9328, + "estimated_eval_tokens": 18912 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is largely heterogeneous: these are distinct bug reports across different models, components, and failure modes. I rejected all soft pairs as true duplicates; none look like the same underlying bug or change.", + "confidence": 0.95, + "canonical_issue_reason": "If one issue must represent the cluster, issue #43873 is the most active and broadly framed (open, multiple references, discussion-heavy). That said, it is not a true semantic canonical for the whole set because the cluster is not actually cohesive.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #43873 is the best global representative only by breadth and activity, not because it matches the other issues semantically.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:32090", + "right": "issue:33357", + "accept": false, + "reason": "Different failures: Trainer/_gpu_broadcast_one TypeError vs MacOS bus error on a CLIP model." + }, + { + "left": "issue:29942", + "right": "issue:35141", + "accept": false, + "reason": "Flash Attention 2 test failures are unrelated to token embedding reinitialization in post_init." + }, + { + "left": "issue:37428", + "right": "issue:44987", + "accept": false, + "reason": "Import error for a flash-attention utility is unrelated to a model-loading failure for physical-intelligence/fast." + }, + { + "left": "issue:43873", + "right": "issue:45229", + "accept": false, + "reason": "Quantization/offloading behavior and Gemma4 multi-GPU OOM are different code paths and symptoms." + }, + { + "left": "issue:43927", + "right": "issue:45698", + "accept": false, + "reason": "Saving/loading custom token IDs is a different bug from loading the wrong custom module after save_pretrained." + }, + { + "left": "issue:41720", + "right": "issue:45229", + "accept": false, + "reason": "Auto device mapping cudaErrorAssert on Qwen3 is distinct from Gemma4 CUDA OOM." + }, + { + "left": "issue:43278", + "right": "issue:43329", + "accept": false, + "reason": "Embedding dtype drift in eval is unrelated to a multimodal token counting bug in the video branch." + }, + { + "left": "issue:44977", + "right": "issue:45563", + "accept": false, + "reason": "Qwen3.5 flash-attention generation regression is not the same as a stale num_return_sequences warning." + }, + { + "left": "issue:44568", + "right": "issue:45491", + "accept": false, + "reason": "Tokenizer BOS/EOS handling is unrelated to Gemma3 mixed-length NaN embeddings from attention masking." + }, + { + "left": "issue:10000", + "right": "issue:15354", + "accept": false, + "reason": "German DistilBertModel issue and torch.jit.script GeneratorExp export failure are different model/runtime problems." + }, + { + "left": "issue:45372", + "right": "issue:45588", + "accept": false, + "reason": "Missing ReasoningEffort import for Gemma4 processor loading is unrelated to an AttributeError in flash_attention.py on s_aux=None." + }, + { + "left": "issue:42490", + "right": "issue:43575", + "accept": false, + "reason": "Trailing-slash local path import warning and Qwen2-57B-A14B-Instruct TP OOM are unrelated." + }, + { + "left": "issue:43412", + "right": "issue:45588", + "accept": false, + "reason": "Gemma3n executorch export failures and flash_attention.py s_aux=None crashes affect different pipelines." + }, + { + "left": "issue:45198", + "right": "issue:45803", + "accept": false, + "reason": "Wav2Vec2 save/tokenization failures are not the same as FuyuBatchFeature swallowing errors with a bare except." + }, + { + "left": "issue:44857", + "right": "issue:45588", + "accept": false, + "reason": "LwDetrImageLoss AMP/CUDA crash is a different failure mode than a flash_attention attribute error." + }, + { + "left": "issue:43976", + "right": "issue:45440", + "accept": false, + "reason": "Python version compatibility in Transformers 5.1.0 is unrelated to DeepseekV3MoE divergence." + }, + { + "left": "issue:45372", + "right": "issue:45406", + "accept": false, + "reason": "Both involve Gemma4-related loading, but one is an import dependency issue and the other is a missing _tokenizer attribute in transformers serve." + }, + { + "left": "issue:45459", + "right": "issue:45712", + "accept": false, + "reason": "Tokenizer error masking without protobuf and leftover dummy classes failing repo checks are unrelated." + }, + { + "left": "issue:45588", + "right": "issue:45823", + "accept": false, + "reason": "flash_attention.py s_aux=None crash is different from Gemma4 PLE device mismatch under device_map=auto." + }, + { + "left": "issue:43278", + "right": "issue:43381", + "accept": false, + "reason": "Eval-time dtype change for embeddings is not the same as gradient checkpointing being disallowed in eval mode." + }, + { + "left": "issue:44448", + "right": "issue:44466", + "accept": false, + "reason": "Different output across Transformers versions for Pegasus is unrelated to device-dependent lm_head.weight serialization." + }, + { + "left": "issue:43653", + "right": "issue:44743", + "accept": false, + "reason": "BigBirdTokenizer special-token registration and Qwen3.5 recurrent-state reset are separate tokenizer/model bugs." + }, + { + "left": "issue:45245", + "right": "issue:45563", + "accept": false, + "reason": "Category-count limit error has no relation to a stale generate() warning." + }, + { + "left": "issue:25251", + "right": "issue:32090", + "accept": false, + "reason": "Pipeline top_k output nesting and Trainer broadcast TypeError are unrelated." + }, + { + "left": "issue:45372", + "right": "issue:45459", + "accept": false, + "reason": "Gemma4 processor import dependency failure is different from protobuf exception masking in tokenizers." + }, + { + "left": "issue:44206", + "right": "issue:44479", + "accept": false, + "reason": "Unsupported center arg in LasrFeatureExtractor is unrelated to the Qwen2.5/Qwen3.5 video input regression." + }, + { + "left": "issue:43065", + "right": "issue:43994", + "accept": false, + "reason": "Dummy Conv2d in Sam3PixelDecoder and SigLIP2 nonsensical outputs are different model-specific issues." + }, + { + "left": "issue:44186", + "right": "issue:45198", + "accept": false, + "reason": "LayoutLMv2 tokenizer crashes on NER/padding are unrelated to Wav2Vec2 save_pretrained/tokenization failures." + }, + { + "left": "issue:43065", + "right": "issue:45412", + "accept": false, + "reason": "A dummy Conv2d placeholder bug is unrelated to RT-DETR memory not being released." + }, + { + "left": "issue:44871", + "right": "issue:44977", + "accept": false, + "reason": "Gemma3 eos_token_id configuration mismatch is not the same as Qwen3.5 flash-attention generation failure." + }, + { + "left": "issue:44361", + "right": "issue:45198", + "accept": false, + "reason": "MLukeTokenizer AttributeError and Wav2Vec2 tokenization/save failures are different tokenizer 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": 18, + "review_unit_index": 27, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:29127", + "issue:29942", + "issue:30064", + "issue:32090", + "issue:33290", + "issue:33357", + "issue:34567", + "issue:35141", + "issue:36010", + "issue:39692", + "issue:42490", + "issue:42915", + "issue:42994", + "issue:43066", + "issue:43257", + "issue:43278", + "issue:43381", + "issue:43421", + "issue:43526", + "issue:43611", + "issue:43653", + "issue:43701", + "issue:43756", + "issue:43761", + "issue:43824", + "issue:43873", + "issue:43994", + "issue:44186", + "issue:44265", + "issue:44279", + "issue:44442", + "issue:44451", + "issue:44462", + "issue:44610", + "issue:44625", + "issue:45103", + "issue:45198", + "issue:45372", + "issue:45412", + "issue:45440", + "issue:45491", + "issue:45520", + "issue:45584", + "issue:45593", + "issue:45834", + "issue:45859", + "issue:46121" + ], + "soft_pairs": [ + "issue:44451|issue:45520", + "issue:43611|issue:46121", + "issue:42915|issue:42994", + "issue:43066|issue:44462", + "issue:32090|issue:33290", + "issue:42490|issue:43873", + "issue:29942|issue:30064", + "issue:33357|issue:39692", + "issue:29127|issue:34567", + "issue:43994|issue:45412", + "issue:44265|issue:45198", + "issue:43257|issue:43278", + "issue:43756|issue:45440", + "issue:43421|issue:45491", + "issue:43653|issue:44442", + "issue:43381|issue:43701", + "issue:44610|issue:45412", + "issue:43824|issue:44279", + "issue:45440|issue:45593", + "issue:29127|issue:33357", + "issue:29942|issue:36010", + "issue:43526|issue:45834", + "issue:43761|issue:45859", + "issue:43526|issue:44186", + "issue:45103|issue:45372", + "issue:44625|issue:45584", + "issue:29127|issue:35141" + ], + "prepared_review_unit_hash": "a8285994e2aeb8a5e2df7d57323787947952aae91c50865ab165f01aaaf65da6", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 27, + "serialized_chars": 35182, + "estimated_input_tokens": 8796, + "estimated_eval_tokens": 17848 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a heterogeneous mix of unrelated Transformers issues; a few pairs are superficially similar by subsystem, but none look like the same underlying bug/change. No soft-edge duplicate should be merged.", + "confidence": 0.87, + "canonical_issue_reason": "issue:43873 is the broadest and most discussion-heavy unresolved bug here, so it is the best anchor for the cluster even though the rest of the items do not form a true duplicate set.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43873 is the strongest overall representative by scope, activity, and inbound references; it is the best issue to anchor triage for this mixed cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44451", + "right": "issue:45520", + "accept": false, + "reason": "Different failures: model loading for a specific checkpoint vs a Python 3.13 flash_attn import-path KeyError." + }, + { + "left": "issue:43611", + "right": "issue:46121", + "accept": false, + "reason": "Unrelated code paths: base_model_prefix loading regression vs rope-param validation on a JSON-loaded list." + }, + { + "left": "issue:42915", + "right": "issue:42994", + "accept": false, + "reason": "Both touch quantization, but one is a Qwen3Moe FP8 runtime failure and the other is quantized model saving; not the same bug." + }, + { + "left": "issue:43066", + "right": "issue:44462", + "accept": false, + "reason": "Both involve tokenizers/loading, but one is a v5 decoder-type warning and the other is AutoTokenizer ignoring tokenizer.json." + }, + { + "left": "issue:32090", + "right": "issue:33290", + "accept": false, + "reason": "Trainer broadcast TypeError and DeepSpeed Adafactor OOM are unrelated failure modes." + }, + { + "left": "issue:42490", + "right": "issue:43873", + "accept": false, + "reason": "Local-path trailing-slash import warning is unrelated to quantization/offloading behavior." + }, + { + "left": "issue:29942", + "right": "issue:30064", + "accept": false, + "reason": "Flash Attention 2 test failures and void segmentation-map processing are different subsystems and bugs." + }, + { + "left": "issue:33357", + "right": "issue:39692", + "accept": false, + "reason": "MacOS bus error on a CLIP model is not the same as SigLIP2 documentation/example errors." + }, + { + "left": "issue:29127", + "right": "issue:34567", + "accept": false, + "reason": "LayoutLMv3 error-message clarity and TrainerState token counting are unrelated issues." + }, + { + "left": "issue:43994", + "right": "issue:45412", + "accept": false, + "reason": "SigLIP2 inference correctness and RT-DETR memory release are different code paths." + }, + { + "left": "issue:44265", + "right": "issue:45198", + "accept": false, + "reason": "torch.export support for torch_compilable_check is unrelated to Wav2Vec2 save_pretrained/tokenization failures." + }, + { + "left": "issue:43257", + "right": "issue:43278", + "accept": false, + "reason": "Qwen3 MoE weight conversion under accelerate+deepspeed is separate from embedding dtype drift between train/eval." + }, + { + "left": "issue:43756", + "right": "issue:45440", + "accept": false, + "reason": "Smollm3 RoPE-layer drop and DeepseekV3MoE divergence are distinct model-implementation issues." + }, + { + "left": "issue:43421", + "right": "issue:45491", + "accept": false, + "reason": "Runtime special-token post-processing and Gemma3 NaN embeddings on GPU are not the same failure." + }, + { + "left": "issue:43653", + "right": "issue:44442", + "accept": false, + "reason": "BigBirdTokenizer special-token registration and AutoTokenizer failing to load FastSpeech2ConformerTokenizer are different tokenizer bugs." + }, + { + "left": "issue:43381", + "right": "issue:43701", + "accept": false, + "reason": "Gradient checkpointing in eval mode and resume_from_checkpoint key mismatch are unrelated training-path issues." + }, + { + "left": "issue:44610", + "right": "issue:45412", + "accept": false, + "reason": "Processor image-size mismatch is unrelated to RT-DETR garbage-collection memory release." + }, + { + "left": "issue:43824", + "right": "issue:44279", + "accept": false, + "reason": "Importing a missing Qwen2.5-VL class and a generic dependency issue are not the same underlying bug." + }, + { + "left": "issue:45440", + "right": "issue:45593", + "accept": false, + "reason": "DeepseekV3MoE implementation divergence and D-FINE auxiliary-loss behavior are different model bugs." + }, + { + "left": "issue:29127", + "right": "issue:33357", + "accept": false, + "reason": "LayoutLMv3 box-information messaging and CLIP bus errors on MacOS are unrelated." + }, + { + "left": "issue:29942", + "right": "issue:36010", + "accept": false, + "reason": "Flash Attention 2 test failures are not the same as the GenerationMixin import breakage." + }, + { + "left": "issue:43526", + "right": "issue:45834", + "accept": false, + "reason": "BeitImageProcessorFast label reduction and Kosmos2.5 long-OCR index errors are unrelated processing bugs." + }, + { + "left": "issue:43761", + "right": "issue:45859", + "accept": false, + "reason": "CLIPVisionModel missing hidden_states and Qwen3.5 MoE CPU-offload incompatibility are different regressions." + }, + { + "left": "issue:43526", + "right": "issue:44186", + "accept": false, + "reason": "BeitImageProcessorFast reduce_labels bug and LayoutLMv2Tokenizer crashes are unrelated." + }, + { + "left": "issue:45103", + "right": "issue:45372", + "accept": false, + "reason": "auto_docstring annotation handling and Mistral ReasoningEffort import failure affect different components and code paths." + }, + { + "left": "issue:44625", + "right": "issue:45584", + "accept": false, + "reason": "Qwen3.5 num_labels propagation and Whisper empty-transcription generation are unrelated bugs." + }, + { + "left": "issue:29127", + "right": "issue:35141", + "accept": false, + "reason": "LayoutLMv3 error-message clarity and embedding reinitialization after resize 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": 18, + "review_unit_index": 28, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:33290", + "issue:34567", + "issue:34689", + "issue:35141", + "issue:42490", + "issue:42915", + "issue:43232", + "issue:43262", + "issue:43316", + "issue:43334", + "issue:43381", + "issue:43388", + "issue:43441", + "issue:43525", + "issue:43526", + "issue:43531", + "issue:43575", + "issue:43611", + "issue:43646", + "issue:43701", + "issue:43723", + "issue:43761", + "issue:43873", + "issue:43931", + "issue:44062", + "issue:44077", + "issue:44079", + "issue:44188", + "issue:44206", + "issue:44387", + "issue:44442", + "issue:44464", + "issue:44479", + "issue:44743", + "issue:44811", + "issue:44857", + "issue:44898", + "issue:44977", + "issue:45003", + "issue:45072", + "issue:45245", + "issue:45290", + "issue:45440", + "issue:45479", + "issue:45491", + "issue:45588", + "issue:45803", + "issue:45907" + ], + "soft_pairs": [ + "issue:43334|issue:43531", + "issue:44442|issue:45072", + "issue:42915|issue:43931", + "issue:44977|issue:45245", + "issue:43526|issue:44857", + "issue:42490|issue:43381", + "issue:44464|issue:45491", + "issue:43575|issue:43701", + "issue:43262|issue:44206", + "issue:43575|issue:44387", + "issue:44743|issue:44811", + "issue:43646|issue:44077", + "issue:43441|issue:45588", + "issue:44062|issue:45907", + "issue:43761|issue:44079", + "issue:45003|issue:45440", + "issue:43723|issue:45479", + "issue:43232|issue:43388", + "issue:44479|issue:45290", + "issue:44442|issue:44898", + "issue:33290|issue:34567", + "issue:33290|issue:34689", + "issue:43316|issue:43525", + "issue:42490|issue:44743", + "issue:33290|issue:35141", + "issue:44442|issue:45803", + "issue:43381|issue:43873", + "issue:43761|issue:44188", + "issue:34689|issue:43611" + ], + "prepared_review_unit_hash": "a9955991463e0d523388434a13b24b0b94aa9afb12bf109835dccac9e7217a91", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36460, + "estimated_input_tokens": 9115, + "estimated_eval_tokens": 18486 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: most items are unrelated regressions or bugs across loading, generation, audio, quantization, configs, and trainer utilities. None of the proposed soft pairs look like true duplicates or mergeable PR-equivalents. The broadest central issue is the quantization/offloading memory report.", + "confidence": 0.41, + "canonical_issue_reason": "issue:43873 is the broadest and most central report, with the highest discussion/inbound-reference activity, and it captures a recurring quantization/offloading failure more generally than the other issues.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43873 is the best cluster anchor because it is broad, actively discussed, and most likely to subsume related quantization/offloading follow-ups.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43334", + "right": "issue:43531", + "accept": false, + "reason": "Different model families and failures: Qwen3-VL missing pad_token_id vs Qwen3-MoE sliding_window behavior." + }, + { + "left": "issue:44442", + "right": "issue:45072", + "accept": false, + "reason": "Tokenizer loading failure vs dtype mismatch in inference; different code paths and symptoms." + }, + { + "left": "issue:42915", + "right": "issue:43931", + "accept": false, + "reason": "Qwen3Moe FP8 config failure and Qwen3-VL weight-shape mismatch are distinct model-loading bugs." + }, + { + "left": "issue:44977", + "right": "issue:45245", + "accept": false, + "reason": "Flash-attention generation regression is unrelated to the category-count RuntimeError." + }, + { + "left": "issue:43526", + "right": "issue:44857", + "accept": false, + "reason": "BEiT label reduction bug and LwDetr AMP crash are different image-processing/loss issues." + }, + { + "left": "issue:42490", + "right": "issue:43381", + "accept": false, + "reason": "Trailing-slash import warning is unrelated to the eval-mode gradient-checkpointing restriction." + }, + { + "left": "issue:44464", + "right": "issue:45491", + "accept": false, + "reason": "Both involve generation, but one is compiled forward chunked generation and the other is Gemma3 mixed-length sliding-window NaNs." + }, + { + "left": "issue:43575", + "right": "issue:43701", + "accept": false, + "reason": "Tensor-parallel OOM during load is unrelated to checkpoint key mismatches." + }, + { + "left": "issue:43262", + "right": "issue:44206", + "accept": false, + "reason": "Audio chat-template sampling-rate default and unsupported `center` arg are different audio-processing bugs." + }, + { + "left": "issue:43575", + "right": "issue:44387", + "accept": false, + "reason": "Both mention OOM, but one is TP model loading and the other is int4 reserved-memory growth; different causes." + }, + { + "left": "issue:44743", + "right": "issue:44811", + "accept": false, + "reason": "Qwen3.5 recurrent-state reset and Whisper batch_decode token handling are unrelated." + }, + { + "left": "issue:43646", + "right": "issue:44077", + "accept": false, + "reason": "Generic v5 custom-init breakage and a PatchTSMixer-specific `post_init` policy bug are not the same issue." + }, + { + "left": "issue:43441", + "right": "issue:45588", + "accept": false, + "reason": "Different flash-attention failures with different models and root causes." + }, + { + "left": "issue:44062", + "right": "issue:45907", + "accept": false, + "reason": "Tokenizer AddedToken construction error is unrelated to hub file-list revision forwarding." + }, + { + "left": "issue:43761", + "right": "issue:44079", + "accept": false, + "reason": "CLIPVisionModel hidden_states regression and ModelOutput key assignment are distinct output-handling bugs." + }, + { + "left": "issue:45003", + "right": "issue:45440", + "accept": false, + "reason": "sys.modules access safety and DeepSeekV3 implementation divergence are unrelated problems." + }, + { + "left": "issue:43723", + "right": "issue:45479", + "accept": false, + "reason": "AutoTokenizer loading in v5 is unrelated to the single-label-classification zero-loss bug." + }, + { + "left": "issue:43232", + "right": "issue:43388", + "accept": false, + "reason": "Generation kwargs syncing and gather_for_metrics label dropping are separate distributed/runtime issues." + }, + { + "left": "issue:44479", + "right": "issue:45290", + "accept": false, + "reason": "Video-input regression and chat-template tool-call crash affect different multimodal paths." + }, + { + "left": "issue:44442", + "right": "issue:44898", + "accept": false, + "reason": "Tokenizer loading and Perceiver image-size handling are different subsystems." + }, + { + "left": "issue:33290", + "right": "issue:34567", + "accept": false, + "reason": "Optimizer OOM in deepspeed is unrelated to TrainerState token-count tracking." + }, + { + "left": "issue:33290", + "right": "issue:34689", + "accept": false, + "reason": "AdaFactor/deepspeed OOM and Llama 3.2 vision model-loading regression are unrelated." + }, + { + "left": "issue:43316", + "right": "issue:43525", + "accept": false, + "reason": "Both are config discrepancies, but they concern different model families and different missing fields." + }, + { + "left": "issue:42490", + "right": "issue:44743", + "accept": false, + "reason": "Local-path import warning and Qwen3.5 recurrent-state reset do not share a code path." + }, + { + "left": "issue:33290", + "right": "issue:35141", + "accept": false, + "reason": "Optimizer OOM is unrelated to embedding reinitialization during post_init." + }, + { + "left": "issue:44442", + "right": "issue:45803", + "accept": false, + "reason": "Tokenizer loading failure is unrelated to Fuyu's broad exception handling bug." + }, + { + "left": "issue:43381", + "right": "issue:43873", + "accept": false, + "reason": "Gradient checkpointing in eval mode and quantization offloading failures are different behaviors." + }, + { + "left": "issue:43761", + "right": "issue:44188", + "accept": false, + "reason": "Hidden-states regression in CLIPVisionModel is unrelated to torch.compile attention-kernel divergence." + }, + { + "left": "issue:34689", + "right": "issue:43611", + "accept": false, + "reason": "Both are model-loading regressions, but they affect different versions, model families, and root causes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 29, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:16998", + "issue:22355", + "issue:25251", + "issue:28282", + "issue:33290", + "issue:34689", + "issue:36010", + "issue:36331", + "issue:41628", + "issue:42907", + "issue:43065", + "issue:43295", + "issue:43316", + "issue:43404", + "issue:43441", + "issue:43493", + "issue:43540", + "issue:43550", + "issue:43606", + "issue:43643", + "issue:43701", + "issue:43716", + "issue:43720", + "issue:43749", + "issue:43825", + "issue:43827", + "issue:43873", + "issue:43931", + "issue:44077", + "issue:44162", + "issue:44190", + "issue:44442", + "issue:44479", + "issue:44496", + "issue:44514", + "issue:44534", + "issue:44617", + "issue:44792", + "issue:44855", + "issue:44871", + "issue:44877", + "issue:44913", + "issue:45137", + "issue:45216", + "issue:45372", + "issue:45584", + "issue:45698", + "issue:45759" + ], + "soft_pairs": [ + "issue:43720|issue:44442", + "issue:44077|issue:44479", + "issue:43493|issue:43827", + "issue:43931|issue:45216", + "issue:42907|issue:45698", + "issue:34689|issue:36010", + "issue:43316|issue:45759", + "issue:41628|issue:45372", + "issue:33290|issue:36331", + "issue:43749|issue:43931", + "issue:43873|issue:44871", + "issue:43643|issue:44913", + "issue:44190|issue:44496", + "issue:44855|issue:45137", + "issue:25251|issue:28282", + "issue:43540|issue:44514", + "issue:43316|issue:44877", + "issue:43295|issue:43825", + "issue:44162|issue:44792", + "issue:43065|issue:44617", + "issue:43316|issue:43716", + "issue:43404|issue:43441", + "issue:43550|issue:43701", + "issue:43606|issue:43701", + "issue:16998|issue:22355", + "issue:44077|issue:44534", + "issue:44871|issue:45584" + ], + "prepared_review_unit_hash": "918271073fec3cb53d44c6a9094f8ace771c596c0132d651603145d51a13661d", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 35668, + "estimated_input_tokens": 8917, + "estimated_eval_tokens": 18090 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is very loose: most items are separate Transformers bug reports spanning config, loading, processors, training, and multimodal edge cases. None of the soft pairs look like true duplicates of the same concrete bug or fix.", + "confidence": 0.34, + "canonical_issue_reason": "issue:43873 is the best single issue to anchor the cluster because it is open, has the most inbound references, and is the broadest/most discussion-heavy of the loading/offload-related reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43873 is the strongest representative issue: it is still open, relatively central to the cluster\u2019s recurring runtime/loading failures, and more likely than the others to absorb related follow-up.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43720", + "right": "issue:44442", + "accept": false, + "reason": "Different failures in different areas: packed-weight loading for BitNet vs tokenizer loading for a FastSpeech2Conformer tokenizer." + }, + { + "left": "issue:44077", + "right": "issue:44479", + "accept": false, + "reason": "Unrelated code paths: PatchTSMixer post_init policy vs Qwen video-input regression." + }, + { + "left": "issue:43493", + "right": "issue:43827", + "accept": false, + "reason": "Different problems: SigLIP2 implementation discrepancy vs docs still mentioning pipeline() after v5 removals." + }, + { + "left": "issue:43931", + "right": "issue:45216", + "accept": false, + "reason": "Both involve Qwen models, but one is a weight-shape load failure and the other is a save_pretrained checkpoint regression." + }, + { + "left": "issue:42907", + "right": "issue:45698", + "accept": false, + "reason": "Related to save/load behavior in a broad sense, but the concrete bugs differ: dequantized model saving vs wrong custom module restoration." + }, + { + "left": "issue:34689", + "right": "issue:36010", + "accept": false, + "reason": "Different import/load failures affecting different APIs and model families." + }, + { + "left": "issue:43316", + "right": "issue:45759", + "accept": false, + "reason": "Both concern config/model API mismatches, but they are different composite-model behaviors and not the same bug." + }, + { + "left": "issue:41628", + "right": "issue:45372", + "accept": false, + "reason": "Both are ImportErrors, but for unrelated symbols and packages with different fixes." + }, + { + "left": "issue:33290", + "right": "issue:36331", + "accept": false, + "reason": "Completely different issues: optimizer memory usage vs trainer signature mismatch." + }, + { + "left": "issue:43749", + "right": "issue:43931", + "accept": false, + "reason": "Different loading subsystems and symptoms: FSDP CPU RAM efficient loading vs Qwen3-VL shape mismatch." + }, + { + "left": "issue:43873", + "right": "issue:44871", + "accept": false, + "reason": "Offloading/quantization behavior is unrelated to Gemma-3 eos_token_id configuration." + }, + { + "left": "issue:43643", + "right": "issue:44913", + "accept": false, + "reason": "Both involve config persistence, but one is missing fields from trust_remote_code and the other is a specific attribute not round-tripping." + }, + { + "left": "issue:44190", + "right": "issue:44496", + "accept": false, + "reason": "Dataset loading for an example script is unrelated to an unrecognized model/config.json error." + }, + { + "left": "issue:44855", + "right": "issue:45137", + "accept": false, + "reason": "No shared underlying bug: Python 3.13 decorator parsing vs DeepSpeed ZeRO3 deque underflow." + }, + { + "left": "issue:25251", + "right": "issue:28282", + "accept": false, + "reason": "Pipeline output nesting and missing PyTorch are unrelated issues." + }, + { + "left": "issue:43540", + "right": "issue:44514", + "accept": false, + "reason": "Both are Qwen multimodal processor bugs, but the concrete failures differ: video processing ValueError vs batched chat-template crash." + }, + { + "left": "issue:43316", + "right": "issue:44877", + "accept": false, + "reason": "Different config issues: API discrepancy in Gemma3TextConfig vs strict config blocking granite_speech loading." + }, + { + "left": "issue:43295", + "right": "issue:43825", + "accept": false, + "reason": "Different regressions: processor/tokenizer handling vs a misleading pipeline() error message about translation tasks." + }, + { + "left": "issue:44162", + "right": "issue:44792", + "accept": false, + "reason": "Different models and code paths: ESM2 breakage vs janus image-generation test failure." + }, + { + "left": "issue:43065", + "right": "issue:44617", + "accept": false, + "reason": "Both mention Sam3, but one is a dummy Conv2d implementation detail and the other is a CUDA OOM symptom." + }, + { + "left": "issue:43316", + "right": "issue:43716", + "accept": false, + "reason": "Different multimodal config issues with no evidence they stem from the same bug." + }, + { + "left": "issue:43404", + "right": "issue:43441", + "accept": false, + "reason": "Same model family, but different failures: untied lm_head weights vs FlashAttention compatibility." + }, + { + "left": "issue:43550", + "right": "issue:43701", + "accept": false, + "reason": "Completely unrelated: torch.compile/SDPA failure vs resume_from_checkpoint key mismatch." + }, + { + "left": "issue:43606", + "right": "issue:43701", + "accept": false, + "reason": "CPU offload device mismatch is unrelated to checkpoint key mismatch." + }, + { + "left": "issue:16998", + "right": "issue:22355", + "accept": false, + "reason": "Different import/config questions with no shared underlying bug." + }, + { + "left": "issue:44077", + "right": "issue:44534", + "accept": false, + "reason": "PatchTSMixer post_init policy is unrelated to non-persistent buffer initialization in v5." + }, + { + "left": "issue:44871", + "right": "issue:45584", + "accept": false, + "reason": "Gemma eos_token_id inconsistency and Whisper empty-transcription failure 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": 18, + "review_unit_index": 30, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:32090", + "issue:35141", + "issue:41950", + "issue:42617", + "issue:43295", + "issue:43388", + "issue:43452", + "issue:43475", + "issue:43504", + "issue:43525", + "issue:43531", + "issue:43575", + "issue:43646", + "issue:43673", + "issue:43688", + "issue:43742", + "issue:43761", + "issue:43827", + "issue:43906", + "issue:43931", + "issue:43994", + "issue:44060", + "issue:44190", + "issue:44242", + "issue:44464", + "issue:44479", + "issue:44496", + "issue:44521", + "issue:44534", + "issue:44617", + "issue:44734", + "issue:44743", + "issue:44792", + "issue:44871", + "issue:44913", + "issue:44964", + "issue:44977", + "issue:45127", + "issue:45200", + "issue:45290", + "issue:45381", + "issue:45406", + "issue:45563", + "issue:45584", + "issue:45600", + "issue:45698", + "issue:46050" + ], + "soft_pairs": [ + "issue:42617|issue:43575", + "issue:43673|issue:44743", + "issue:43575|issue:43742", + "issue:41950|issue:43475", + "issue:43388|issue:43688", + "issue:43295|issue:44871", + "issue:43761|issue:44479", + "issue:43452|issue:44496", + "issue:45127|issue:45200", + "issue:44871|issue:46050", + "issue:43646|issue:44913", + "issue:43504|issue:45698", + "issue:32090|issue:35141", + "issue:44734|issue:45406", + "issue:44617|issue:45584", + "issue:45563|issue:45584", + "issue:43761|issue:44242", + "issue:43525|issue:43531", + "issue:44792|issue:44977", + "issue:43931|issue:44060", + "issue:44479|issue:44734", + "issue:43906|issue:45290", + "issue:44464|issue:45563", + "issue:44190|issue:44964", + "issue:43827|issue:43994", + "issue:44534|issue:44913", + "issue:44792|issue:45600", + "issue:44521|issue:45381" + ], + "prepared_review_unit_hash": "d6bf92b2a553e26dfa7135d05a19239505367461fa6d32d7bd3a1210245c26c3", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 35673, + "estimated_input_tokens": 8919, + "estimated_eval_tokens": 18094 + }, + "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 bugs and regressions; the soft pairs share broad subsystem similarity but not the same concrete underlying defect, so none should be merged as duplicates. There are no PRs in the cluster.", + "confidence": 0.61, + "canonical_issue_reason": "issue:43761 is the cleanest single anchor: a concrete v5 regression in multimodal/vision forward behavior with a clear symptom and repro. It\u2019s the best representative issue if this cluster needs one, though the overall set is too heterogeneous to treat as one duplicate group.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43761 is the most representative issue because it is specific, user-facing, and fits the common v5 regression/multimodal theme better than the more isolated loading/config or script failures. It is still only a cluster anchor, not evidence of duplication with the others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:42617", + "right": "issue:43575", + "accept": false, + "reason": "Different failures: 3d_parallel.py execution vs tensor-parallel OOM while loading a specific Qwen2 model." + }, + { + "left": "issue:43673", + "right": "issue:44743", + "accept": false, + "reason": "Both involve cache/state, but one is a GenerationMixin chunked_prefill cache regression and the other is recurrent-state reset in qwen3_5; different code paths and symptoms." + }, + { + "left": "issue:43575", + "right": "issue:43742", + "accept": false, + "reason": "OOM during tensor-parallel model load is not the same bug as a key error loading MobileLLM-125M config." + }, + { + "left": "issue:41950", + "right": "issue:43475", + "accept": false, + "reason": "Pipeline processor lookup for video classification is unrelated to Sam3 vision encoder output missing an attribute." + }, + { + "left": "issue:43388", + "right": "issue:43688", + "accept": false, + "reason": "Gathering metrics dropping labels and auxiliary-loss normalization are distinct training bugs." + }, + { + "left": "issue:43295", + "right": "issue:44871", + "accept": false, + "reason": "Custom processor/tokenizer API regression is unrelated to an eos_token_id configuration mismatch." + }, + { + "left": "issue:43761", + "right": "issue:44479", + "accept": false, + "reason": "Both are multimodal regressions, but one is CLIPVision hidden_states handling and the other is Qwen video-input position-id behavior; not the same defect." + }, + { + "left": "issue:43452", + "right": "issue:44496", + "accept": false, + "reason": "Broken gguf loading for tokenizer/model classes is different from an unrecognized-model config.json missing model_type." + }, + { + "left": "issue:45127", + "right": "issue:45200", + "accept": false, + "reason": "LoRA merge with extended vocab and Gemma 4 mm_token_type_ids defaulting are separate model-behavior issues." + }, + { + "left": "issue:44871", + "right": "issue:46050", + "accept": false, + "reason": "EOS-token config inconsistency is unrelated to save/load breakage for sinq-quantized models." + }, + { + "left": "issue:43646", + "right": "issue:44913", + "accept": false, + "reason": "Broad custom-model initialization breakage is not the same as GPTNeoX rotary_pct failing to round-trip on reload." + }, + { + "left": "issue:43504", + "right": "issue:45698", + "accept": false, + "reason": "Legacy-field pretrained loading failure and wrong custom module resolution after save_pretrained are different load-path bugs." + }, + { + "left": "issue:32090", + "right": "issue:35141", + "accept": false, + "reason": "Trainer GPU broadcast TypeError is unrelated to output-embedding reinitialization during post_init after resizing embeddings." + }, + { + "left": "issue:44734", + "right": "issue:45406", + "accept": false, + "reason": "KV-cache continuation tensor indexing in serve is a different failure from Gemma4Processor missing _tokenizer." + }, + { + "left": "issue:44617", + "right": "issue:45584", + "accept": false, + "reason": "SAM3 video OOM and Whisper empty-transcription generation failure are unrelated model-specific bugs." + }, + { + "left": "issue:45563", + "right": "issue:45584", + "accept": false, + "reason": "A stale warning for num_return_sequences is not the same as Whisper failing on empty transcription." + }, + { + "left": "issue:43761", + "right": "issue:44242", + "accept": false, + "reason": "CLIPVision hidden_states regression and missing load-balancing loss when output_router_logits=False are different issues." + }, + { + "left": "issue:43525", + "right": "issue:43531", + "accept": false, + "reason": "Missing pad_token_id in Llama4Config and Qwen3-MoE sliding_window behavior are distinct model/config bugs." + }, + { + "left": "issue:44792", + "right": "issue:44977", + "accept": false, + "reason": "Janus image-generation test failure and Qwen3.5 flash-attention generation issue are not the same underlying bug." + }, + { + "left": "issue:43931", + "right": "issue:44060", + "accept": false, + "reason": "Weight-shape mismatch while loading Qwen3-VL is unrelated to the tied-weights warning that binds embed_tokens.weight to linear_attn.dt_bias." + }, + { + "left": "issue:44479", + "right": "issue:44734", + "accept": false, + "reason": "Qwen video-input regression and serve KV-cache continuation crash are separate subsystems and code paths." + }, + { + "left": "issue:43906", + "right": "issue:45290", + "accept": false, + "reason": "An isolated reproduction of another issue is not the same as an apply_chat_template crash on tool-call assistant messages." + }, + { + "left": "issue:44464", + "right": "issue:45563", + "accept": false, + "reason": "Compiled-forward chunked-generation inconsistency and a stale warning for paged generate are different generation bugs." + }, + { + "left": "issue:44190", + "right": "issue:44964", + "accept": false, + "reason": "Local dataset loading in an image-classification script is unrelated to loading a multimodal model checkpoint." + }, + { + "left": "issue:43827", + "right": "issue:43994", + "accept": false, + "reason": "Docs still referencing pipeline() and SigLIP2 producing nonsensical outputs are not the same defect." + }, + { + "left": "issue:44534", + "right": "issue:44913", + "accept": false, + "reason": "Non-persistent buffer junk in v5 is not the same as GPTNeoXConfig rotary_pct resetting on reload." + }, + { + "left": "issue:44792", + "right": "issue:45600", + "accept": false, + "reason": "A Janus generate-images test failure is unrelated to auto_mappings referencing removed Sam3LiteText configs." + }, + { + "left": "issue:44521", + "right": "issue:45381", + "accept": false, + "reason": "All-zero assistant masks in apply_chat_template and wrong vision_position_ids for Qwen2.5-VL video input are different multimodal 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": 18, + "review_unit_index": 31, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:35141", + "issue:36246", + "issue:38617", + "issue:39692", + "issue:41628", + "issue:41720", + "issue:41950", + "issue:42175", + "issue:42490", + "issue:43065", + "issue:43295", + "issue:43335", + "issue:43479", + "issue:43540", + "issue:43550", + "issue:43644", + "issue:43650", + "issue:43976", + "issue:44016", + "issue:44077", + "issue:44079", + "issue:44112", + "issue:44190", + "issue:44220", + "issue:44279", + "issue:44462", + "issue:44496", + "issue:44610", + "issue:44734", + "issue:44737", + "issue:44855", + "issue:44913", + "issue:44933", + "issue:44938", + "issue:45127", + "issue:45245", + "issue:45276", + "issue:45357", + "issue:45362", + "issue:45375", + "issue:45542", + "issue:45584", + "issue:45588", + "issue:45600", + "issue:45803", + "issue:45907", + "issue:46082", + "issue:46121" + ], + "soft_pairs": [ + "issue:41720|issue:45245", + "issue:44855|issue:45588", + "issue:43065|issue:44610", + "issue:44933|issue:45362", + "issue:35141|issue:42490", + "issue:43335|issue:44855", + "issue:44077|issue:44079", + "issue:43550|issue:44190", + "issue:43976|issue:44855", + "issue:44737|issue:45803", + "issue:43644|issue:44913", + "issue:45127|issue:45357", + "issue:43540|issue:44734", + "issue:44462|issue:45907", + "issue:44112|issue:45600", + "issue:44938|issue:45542", + "issue:43335|issue:44190", + "issue:46082|issue:46121", + "issue:39692|issue:44190", + "issue:44496|issue:45803", + "issue:45276|issue:45584", + "issue:36246|issue:41628", + "issue:43295|issue:45276", + "issue:43479|issue:45375", + "issue:43650|issue:44220", + "issue:45245|issue:45584", + "issue:36246|issue:38617", + "issue:41950|issue:44016", + "issue:42175|issue:44279" + ], + "prepared_review_unit_hash": "ef442f064f25dd664cac2ac4adb1a28e7acdc321a290854c47d8712ab74353dc", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36542, + "estimated_input_tokens": 9136, + "estimated_eval_tokens": 18528 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous; the soft-similarity edges are mostly superficial and do not describe the same underlying bug or change. No PRs are present. The only plausible mini-cluster is around token-embedding resize behavior, centered on issue 35141.", + "confidence": 0.84, + "canonical_issue_reason": "Issue 35141 is the clearest root-cause report for the only related subtopic here: resize_token_embeddings / post_init reinitialization when embeddings are untied.", + "canonical_pr_reason": null, + "best_issue_reason": "35141 is the best anchor because it states the underlying code-path problem more generally than the model-specific symptom in 45276.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:41720", + "right": "issue:45245", + "accept": false, + "reason": "Different failures: cudaErrorAssert during Qwen3 auto device mapping vs a category-count limit runtime error." + }, + { + "left": "issue:44855", + "right": "issue:45588", + "accept": false, + "reason": "Unrelated issues: Python 3.13 import parsing failure vs flash_attention crashing on s_aux=None." + }, + { + "left": "issue:43065", + "right": "issue:44610", + "accept": false, + "reason": "Different models and different bugs: Sam3PixelDecoder dummy conv vs OmDet-Turbo image-size mismatch." + }, + { + "left": "issue:44933", + "right": "issue:45362", + "accept": false, + "reason": "Import alias problem vs Qwen3.5 chat crash; no shared code path." + }, + { + "left": "issue:35141", + "right": "issue:42490", + "accept": false, + "reason": "Resize-token-embedding/post_init bug is unrelated to import warning on trailing slash paths." + }, + { + "left": "issue:43335", + "right": "issue:44855", + "accept": false, + "reason": "SwitchTransformers sparse-layer construction bug is unrelated to DebertaV2 import parsing on Python 3.13." + }, + { + "left": "issue:44077", + "right": "issue:44079", + "accept": false, + "reason": "Optional post_init enforcement and ModelOutput None-key assignment are distinct behaviors in different code paths." + }, + { + "left": "issue:43550", + "right": "issue:44190", + "accept": false, + "reason": "torch.compile + SDPA issue is unrelated to local dataset loading in a training script." + }, + { + "left": "issue:43976", + "right": "issue:44855", + "accept": false, + "reason": "Python-version support regression is unrelated to the DebertaV2 JIT parsing bug." + }, + { + "left": "issue:44737", + "right": "issue:45803", + "accept": false, + "reason": "CPU device placement bug in XLNet is unrelated to FuyuBatchFeature swallowing exceptions." + }, + { + "left": "issue:43644", + "right": "issue:44913", + "accept": false, + "reason": "Both involve config/state persistence, but one is non-persistent buffers and the other is GPTNeoX rotary_pct reload; different bugs." + }, + { + "left": "issue:45127", + "right": "issue:45357", + "accept": false, + "reason": "LoRA merge with extended vocab and incorrect visual encoder keys are different Qwen-related issues." + }, + { + "left": "issue:43540", + "right": "issue:44734", + "accept": false, + "reason": "Video-input validation error in Qwen3OmniMoe is unrelated to serve KV-cache tensor indexing." + }, + { + "left": "issue:44462", + "right": "issue:45907", + "accept": false, + "reason": "AutoTokenizer repository-file selection and list_repo_files revision forwarding are separate hub API behaviors." + }, + { + "left": "issue:44112", + "right": "issue:45600", + "accept": false, + "reason": "CI test flake in GraniteSpeech is unrelated to stale Sam3LiteText auto_mappings references." + }, + { + "left": "issue:44938", + "right": "issue:45542", + "accept": false, + "reason": "Python 3.14 import/load failure and missing TensorFlow backend are different environment/setup problems." + }, + { + "left": "issue:43335", + "right": "issue:44190", + "accept": false, + "reason": "SwitchTransformers sparse-layer bug is unrelated to local dataset loading in run_image_classification_no_trainer.py." + }, + { + "left": "issue:46082", + "right": "issue:46121", + "accept": false, + "reason": "LlamaConfig head_dim validation and rope-parameter JSON list handling are distinct config-validation bugs." + }, + { + "left": "issue:39692", + "right": "issue:44190", + "accept": false, + "reason": "SigLIP2 doc example errors are unrelated to loading a local dataset in the image-classification script." + }, + { + "left": "issue:44496", + "right": "issue:45803", + "accept": false, + "reason": "Model config recognition failure is unrelated to exception swallowing in FuyuBatchFeature." + }, + { + "left": "issue:45276", + "right": "issue:45584", + "accept": false, + "reason": "Both involve model/input handling, but Gemma4 resize-token-embedding behavior is unrelated to Whisper empty-transcription failure." + }, + { + "left": "issue:36246", + "right": "issue:41628", + "accept": false, + "reason": "Two different missing imports for different symbols; same symptom class, not the same bug." + }, + { + "left": "issue:43295", + "right": "issue:45276", + "accept": false, + "reason": "Processor/tokenizer regression in v4.57.5 is unrelated to Gemma4 resize_token_embeddings not updating embeddings." + }, + { + "left": "issue:43479", + "right": "issue:45375", + "accept": false, + "reason": "Default config initialization and strict-dropping of a vision field are different multimodal config bugs." + }, + { + "left": "issue:43650", + "right": "issue:44220", + "accept": false, + "reason": "One report is unusable/title-only; the other is a specific _torch_extract_fbank_features issue." + }, + { + "left": "issue:45245", + "right": "issue:45584", + "accept": false, + "reason": "Category-count overflow and Whisper empty-transcription are unrelated." + }, + { + "left": "issue:36246", + "right": "issue:38617", + "accept": false, + "reason": "Both are import errors, but they concern different missing names in different modules." + }, + { + "left": "issue:41950", + "right": "issue:44016", + "accept": false, + "reason": "Video-classification pipeline image-processor lookup and a notebook syntax error are unrelated." + }, + { + "left": "issue:42175", + "right": "issue:44279", + "accept": false, + "reason": "TensorFlow-not-in-backend packaging issue is unrelated to a generic dependency problem." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 32, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:38175", + "issue:42490", + "issue:42907", + "issue:43065", + "issue:43295", + "issue:43299", + "issue:43335", + "issue:43381", + "issue:43404", + "issue:43441", + "issue:43479", + "issue:43525", + "issue:43526", + "issue:43531", + "issue:43643", + "issue:43673", + "issue:43824", + "issue:43854", + "issue:43901", + "issue:43976", + "issue:44016", + "issue:44062", + "issue:44075", + "issue:44112", + "issue:44186", + "issue:44220", + "issue:44246", + "issue:44279", + "issue:44336", + "issue:44368", + "issue:44496", + "issue:44561", + "issue:44683", + "issue:44734", + "issue:44933", + "issue:44964", + "issue:44987", + "issue:45003", + "issue:45042", + "issue:45216", + "issue:45335", + "issue:45357", + "issue:45446", + "issue:45584", + "issue:45593", + "issue:45600", + "issue:45632", + "issue:45759" + ], + "soft_pairs": [ + "issue:44186|issue:44734", + "issue:44683|issue:45446", + "issue:44246|issue:45003", + "issue:44016|issue:44220", + "issue:43404|issue:45216", + "issue:38175|issue:43065", + "issue:44734|issue:45632", + "issue:43643|issue:44496", + "issue:44561|issue:45632", + "issue:44964|issue:45600", + "issue:44112|issue:44964", + "issue:43854|issue:44964", + "issue:43976|issue:44987", + "issue:42907|issue:43441", + "issue:42490|issue:45632", + "issue:43531|issue:45600", + "issue:43295|issue:45335", + "issue:43526|issue:43901", + "issue:43673|issue:44336", + "issue:44496|issue:45042", + "issue:44016|issue:44062", + "issue:43335|issue:45593", + "issue:43824|issue:44933", + "issue:43299|issue:45216", + "issue:44075|issue:44368", + "issue:43976|issue:44279", + "issue:43381|issue:45593", + "issue:43901|issue:44496", + "issue:45335|issue:45584", + "issue:44496|issue:45759", + "issue:43299|issue:45357", + "issue:43479|issue:43525" + ], + "prepared_review_unit_hash": "758c0c9199ab992b3762f684657e5724eba506a842df1ca831230c7a8c923ad1", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 37249, + "estimated_input_tokens": 9313, + "estimated_eval_tokens": 18882 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These issues are a heterogeneous mix of unrelated model-loading, config, import, and docs/test regressions. None of the soft pairs look like true duplicates or the same concrete fix, so no soft edge is accepted.", + "confidence": 0.93, + "canonical_issue_reason": "Issue 45632 is the cleanest single library-level bug in the set: a precise `trust_remote_code` cache-path collision for local models with the same leaf directory name. The rest are mostly unrelated one-off model or tooling regressions, so there is no strong cluster-wide duplicate target.", + "canonical_pr_reason": null, + "best_issue_reason": "45632 is the best representative issue to keep because it has a narrow, actionable root cause and a fixable central code path affecting multiple local-model setups.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44186", + "right": "issue:44734", + "accept": false, + "reason": "Different failures: tokenizer/NER padding crash vs serving KV-cache continuation tensor indexing." + }, + { + "left": "issue:44683", + "right": "issue:45446", + "accept": false, + "reason": "Both touch flex_attention, but one is a torch-version check bug and the other is a compiled flex_attention failure on torch>=2.9." + }, + { + "left": "issue:44246", + "right": "issue:45003", + "accept": false, + "reason": "Import slowness from package loading is not the same as unsafe `sys.modules` access in `modeling_utils`." + }, + { + "left": "issue:44016", + "right": "issue:44220", + "accept": false, + "reason": "Notebook syntax error and `_torch_extract_fbank_features()` are unrelated bugs." + }, + { + "left": "issue:43404", + "right": "issue:45216", + "accept": false, + "reason": "Different code paths: weight tying in Mistral3 vs incorrect `save_pretrained` output for Qwen3.5." + }, + { + "left": "issue:38175", + "right": "issue:43065", + "accept": false, + "reason": "SigLIP zero-probabilities and a dummy `Conv2d` in Sam3PixelDecoder are distinct model bugs." + }, + { + "left": "issue:44734", + "right": "issue:45632", + "accept": false, + "reason": "Serving KV-cache indexing and local trust_remote_code cache-path collisions are unrelated." + }, + { + "left": "issue:43643", + "right": "issue:44496", + "accept": false, + "reason": "Missing fields in `AutoConfig.from_pretrained(..., trust_remote_code=True)` is different from an unrecognized-model/config.json failure." + }, + { + "left": "issue:44561", + "right": "issue:45632", + "accept": false, + "reason": "Removal of `is_torch_fx_available` breaking remote-code models is a different regression than cache-path collision." + }, + { + "left": "issue:44964", + "right": "issue:45600", + "accept": false, + "reason": "Phi-4 multimodal load failure and removed Sam3LiteText config references are different model-specific regressions." + }, + { + "left": "issue:44112", + "right": "issue:44964", + "accept": false, + "reason": "A stale GraniteSpeech CI test is not the same bug as loading Phi-4 multimodal." + }, + { + "left": "issue:43854", + "right": "issue:44964", + "accept": false, + "reason": "GLM-4.7-Flash loading issues are unrelated to the Phi-4 multimodal loading failure." + }, + { + "left": "issue:43976", + "right": "issue:44987", + "accept": false, + "reason": "Python-version compatibility and a specific model-loading regression are not the same underlying bug." + }, + { + "left": "issue:42907", + "right": "issue:43441", + "accept": false, + "reason": "Saving dequantized Ministral/Devstral models and FlashAttention load failures are distinct problems." + }, + { + "left": "issue:42490", + "right": "issue:45632", + "accept": false, + "reason": "A warning for local paths ending with a slash is a different path-handling issue than cache-key collisions across leaf directories." + }, + { + "left": "issue:43531", + "right": "issue:45600", + "accept": false, + "reason": "Qwen3-MoE sliding-window behavior and Sam3LiteText auto-mapping cleanup are unrelated." + }, + { + "left": "issue:43295", + "right": "issue:45335", + "accept": false, + "reason": "Processor/tokenizer regression and `resize_token_embeddings` not updating decoder embeddings are different bugs." + }, + { + "left": "issue:43526", + "right": "issue:43901", + "accept": false, + "reason": "BEiT `reduce_labels` output corruption is unrelated to a docs mismatch about `return_all_scores`." + }, + { + "left": "issue:43673", + "right": "issue:44336", + "accept": false, + "reason": "Chunked-prefill cache regression and ANSI escape output in loading_report are unrelated." + }, + { + "left": "issue:44496", + "right": "issue:45042", + "accept": false, + "reason": "Unrecognized model/config handling and PIL backend requiring torchvision are different code paths." + }, + { + "left": "issue:44016", + "right": "issue:44062", + "accept": false, + "reason": "Notebook syntax error is unrelated to the `AddedToken(..., special=...)` argument conflict." + }, + { + "left": "issue:43335", + "right": "issue:45593", + "accept": false, + "reason": "Sparse-layer creation in SwitchTransformersConfig and missing auxiliary losses in D-FINE are unrelated." + }, + { + "left": "issue:43824", + "right": "issue:44933", + "accept": false, + "reason": "Missing `Qwen2_5_VLForConditionalGeneration` export and a nonexistent `image_utils` import are different import bugs." + }, + { + "left": "issue:43299", + "right": "issue:45216", + "accept": false, + "reason": "Qwen3VL loading and Qwen3.5 checkpoint saving are related only by family name, not the same fix." + }, + { + "left": "issue:44075", + "right": "issue:44368", + "accept": false, + "reason": "Unused SGD args and a `tie_word_embeddings` warning are unrelated." + }, + { + "left": "issue:43381", + "right": "issue:45593", + "accept": false, + "reason": "Gradient checkpointing in eval mode and D-FINE auxiliary-loss behavior do not share a code path." + }, + { + "left": "issue:43901", + "right": "issue:44496", + "accept": false, + "reason": "Docs mentioning `return_all_scores` is unrelated to model/config loading failure." + }, + { + "left": "issue:45335", + "right": "issue:45584", + "accept": false, + "reason": "Embedding resize propagation and Whisper empty-transcription generation are unrelated." + }, + { + "left": "issue:44496", + "right": "issue:45759", + "accept": false, + "reason": "Unrecognized model/config parsing and `AutoModelForCausalLM.from_config` not unwrapping `text_config` are different issues." + }, + { + "left": "issue:43479", + "right": "issue:43525", + "accept": false, + "reason": "Phi4Multimodal default config initialization and missing `pad_token_id` on Llama4Config are separate config bugs." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 33, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:33290", + "issue:38175", + "issue:39692", + "issue:41950", + "issue:42994", + "issue:43295", + "issue:43299", + "issue:43317", + "issue:43441", + "issue:43450", + "issue:43577", + "issue:43650", + "issue:43653", + "issue:43854", + "issue:43901", + "issue:43976", + "issue:44016", + "issue:44242", + "issue:44279", + "issue:44291", + "issue:44297", + "issue:44351", + "issue:44462", + "issue:44466", + "issue:44521", + "issue:44617", + "issue:44623", + "issue:44683", + "issue:44749", + "issue:44779", + "issue:45020", + "issue:45081", + "issue:45200", + "issue:45216", + "issue:45276", + "issue:45278", + "issue:45325", + "issue:45405", + "issue:45542", + "issue:45593", + "issue:45600", + "issue:45656", + "issue:45712", + "issue:45812", + "issue:45823", + "issue:45834", + "issue:45907", + "issue:46132" + ], + "soft_pairs": [ + "issue:44521|issue:45325", + "issue:44242|issue:45200", + "issue:44351|issue:45278", + "issue:41950|issue:45020", + "issue:44779|issue:45812", + "issue:41950|issue:45823", + "issue:43577|issue:45216", + "issue:43976|issue:45405", + "issue:44683|issue:45542", + "issue:38175|issue:44617", + "issue:44279|issue:45405", + "issue:39692|issue:43901", + "issue:43299|issue:43854", + "issue:43295|issue:44779", + "issue:43650|issue:46132", + "issue:44462|issue:45081", + "issue:44297|issue:44462", + "issue:44016|issue:44291", + "issue:43450|issue:44279", + "issue:45081|issue:45907", + "issue:41950|issue:43650", + "issue:43441|issue:45834", + "issue:45600|issue:45712", + "issue:44466|issue:45276", + "issue:42994|issue:43317", + "issue:33290|issue:45656", + "issue:43653|issue:45593", + "issue:44623|issue:44749" + ], + "prepared_review_unit_hash": "c5755cbebd0bbc5b68e9ddc55dc2096b9645978acc2737819b6a49c4f8de0723", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36066, + "estimated_input_tokens": 9017, + "estimated_eval_tokens": 18290 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: mostly unrelated Transformers bugs spanning tokenizers, multimodal processors, model loading/saving, quantization/offload, deepspeed, CI, and docs. I do not see any true duplicate pair that should be merged; the soft links are mostly broad topical similarity rather than the same underlying defect.", + "confidence": 0.96, + "canonical_issue_reason": "No single canonical issue cleanly represents the cluster because the items are about different code paths and regressions, not one duplicate bug.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44521 is the strongest single representative only because it is open, has the most inbound references, and is a concrete user-facing multimodal bug; it still does not subsume the rest of the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44521", + "right": "issue:45325", + "accept": false, + "reason": "Different bugs: assistant mask generation in chat templates vs Qwen2.5-VL RoPE/temporal position scaling." + }, + { + "left": "issue:44242", + "right": "issue:45200", + "accept": false, + "reason": "Different models and failure modes: MoE load-balancing loss logic vs Gemma4 mm_token_type_ids defaults." + }, + { + "left": "issue:44351", + "right": "issue:45278", + "accept": false, + "reason": "A single HybridCache import error is not the same as broad import failures across many modules." + }, + { + "left": "issue:41950", + "right": "issue:45020", + "accept": false, + "reason": "Video-classification pipeline processor lookup is unrelated to remote_code model loading regressions." + }, + { + "left": "issue:44779", + "right": "issue:45812", + "accept": false, + "reason": "Both involve tokenizers, but they affect different model families and likely different code paths." + }, + { + "left": "issue:41950", + "right": "issue:45823", + "accept": false, + "reason": "Processor lookup bug in a pipeline is not the same as Gemma4 device-mapping mismatch in forward." + }, + { + "left": "issue:43577", + "right": "issue:45216", + "accept": false, + "reason": "Loading dtype propagation and save_pretrained checkpoint serialization are distinct issues." + }, + { + "left": "issue:43976", + "right": "issue:45405", + "accept": false, + "reason": "Python version compatibility and an unreleased dependency minimum are different release blockers." + }, + { + "left": "issue:44683", + "right": "issue:45542", + "accept": false, + "reason": "Torch flex_attention compile failure and TensorFlow backend detection are unrelated." + }, + { + "left": "issue:38175", + "right": "issue:44617", + "accept": false, + "reason": "Unexpected zero probabilities and CUDA OOM are different symptoms with different likely causes." + }, + { + "left": "issue:44279", + "right": "issue:45405", + "accept": false, + "reason": "Generic dependency issue is not the same as a specific PEFT version bump problem." + }, + { + "left": "issue:39692", + "right": "issue:43901", + "accept": false, + "reason": "Doc example errors in SigLIP2 are not the same as stale return_all_scores documentation." + }, + { + "left": "issue:43299", + "right": "issue:43854", + "accept": false, + "reason": "Both are model-loading regressions, but for different models and different underlying failures." + }, + { + "left": "issue:43295", + "right": "issue:44779", + "accept": false, + "reason": "Processor.tokenizer regression and Deepseek tokenizer output regression are separate code paths." + }, + { + "left": "issue:43650", + "right": "issue:46132", + "accept": false, + "reason": "Placeholder issue title versus AttentionInterface.register behavior change; no shared bug." + }, + { + "left": "issue:44462", + "right": "issue:45081", + "accept": false, + "reason": "AutoTokenizer ignoring tokenizer.json and Mistral regex patch crashing are distinct tokenizer-loading problems." + }, + { + "left": "issue:44297", + "right": "issue:44462", + "accept": false, + "reason": "Tokenizer save_pretrained metadata mismatch is not the same as AutoTokenizer preferring the wrong source file." + }, + { + "left": "issue:44016", + "right": "issue:44291", + "accept": false, + "reason": "Notebook syntax error and init_empty_weights TypeError are unrelated failures." + }, + { + "left": "issue:43450", + "right": "issue:44279", + "accept": false, + "reason": "Batched video processor shape bug is unrelated to a dependency issue." + }, + { + "left": "issue:45081", + "right": "issue:45907", + "accept": false, + "reason": "Tokenizer regex crash and list_repo_files revision forwarding are different subsystems entirely." + }, + { + "left": "issue:41950", + "right": "issue:43650", + "accept": false, + "reason": "A real video-classification bug is not related to a meaningless placeholder issue." + }, + { + "left": "issue:43441", + "right": "issue:45834", + "accept": false, + "reason": "Ministral FlashAttention failure and Kosmos2.5 OCR index error are unrelated model-specific bugs." + }, + { + "left": "issue:45600", + "right": "issue:45712", + "accept": false, + "reason": "Removed config references in auto_mappings and leftover dummy class cleanup are separate CI issues." + }, + { + "left": "issue:44466", + "right": "issue:45276", + "accept": false, + "reason": "Inconsistent lm_head serialization and resize_token_embeddings behavior are different model-state bugs." + }, + { + "left": "issue:42994", + "right": "issue:43317", + "accept": false, + "reason": "Quantized save failure and dequantized load/offload failure are different phases and code paths." + }, + { + "left": "issue:33290", + "right": "issue:45656", + "accept": false, + "reason": "Adafactor+deepspeed OOM and optimizer.step being called twice with deepspeed are not the same defect." + }, + { + "left": "issue:43653", + "right": "issue:45593", + "accept": false, + "reason": "BigBirdTokenizer special-token registration and D-FINE auxiliary-loss omission are unrelated." + }, + { + "left": "issue:44623", + "right": "issue:44749", + "accept": false, + "reason": "Missing processor save files and a data-filtering slowdown after upgrade 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": 18, + "review_unit_index": 34, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:41628", + "issue:42175", + "issue:42491", + "issue:43065", + "issue:43262", + "issue:43299", + "issue:43441", + "issue:43450", + "issue:43525", + "issue:43643", + "issue:43650", + "issue:43673", + "issue:43723", + "issue:43761", + "issue:43784", + "issue:43824", + "issue:43881", + "issue:43901", + "issue:43976", + "issue:43994", + "issue:44016", + "issue:44162", + "issue:44188", + "issue:44242", + "issue:44246", + "issue:44261", + "issue:44263", + "issue:44297", + "issue:44336", + "issue:44361", + "issue:44373", + "issue:44521", + "issue:44589", + "issue:44617", + "issue:44623", + "issue:44671", + "issue:44683", + "issue:44779", + "issue:44898", + "issue:44987", + "issue:45290", + "issue:45362", + "issue:45397", + "issue:45542", + "issue:45561", + "issue:45563", + "issue:45593", + "issue:45823" + ], + "soft_pairs": [ + "issue:43650|issue:45290", + "issue:44623|issue:45397", + "issue:43065|issue:44898", + "issue:44261|issue:44263", + "issue:43441|issue:44683", + "issue:43723|issue:44162", + "issue:42175|issue:43976", + "issue:44361|issue:45593", + "issue:44016|issue:45823", + "issue:44671|issue:44779", + "issue:43262|issue:44521", + "issue:43673|issue:43761", + "issue:43901|issue:43994", + "issue:44987|issue:45542", + "issue:43784|issue:45542", + "issue:43824|issue:43976", + "issue:43450|issue:44617", + "issue:44297|issue:45397", + "issue:43525|issue:43881", + "issue:43901|issue:44246", + "issue:44188|issue:44336", + "issue:43525|issue:43643", + "issue:41628|issue:43976", + "issue:44589|issue:45362", + "issue:45561|issue:45563", + "issue:42491|issue:43299", + "issue:44336|issue:44373", + "issue:44242|issue:45593" + ], + "prepared_review_unit_hash": "164c246caf3163616da69b149c913d33bee3979c83af8728e1915825540f3ac6", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 35687, + "estimated_input_tokens": 8922, + "estimated_eval_tokens": 18100 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is largely a grab-bag of distinct regressions across model loading, chat templates, processors, attention kernels, and docs. I did not find any soft pair that is clearly the same underlying bug/change closely enough to merge.", + "confidence": 0.61, + "canonical_issue_reason": "Issue #44521 is the clearest representative: it is a concrete, actively discussed regression in a core multimodal API, with multiple inbound refs and broad user impact.", + "canonical_pr_reason": null, + "best_issue_reason": "#44521 is the best anchor because it is open, specific, and central to a widely used code path; it is more representative than niche model-specific or docs-only reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43650", + "right": "issue:45290", + "accept": false, + "reason": "Completely different problems: a generic placeholder issue versus a chat-template crash on tool-call assistant messages." + }, + { + "left": "issue:44623", + "right": "issue:45397", + "accept": false, + "reason": "Different code paths: missing files on processor save versus ZeRO3/from_pretrained loading failure for Gemma4." + }, + { + "left": "issue:43065", + "right": "issue:44898", + "accept": false, + "reason": "Both touch vision models, but one is a dummy Conv2d in Sam3PixelDecoder and the other is Perceiver interpolation failure; not the same bug." + }, + { + "left": "issue:44261", + "right": "issue:44263", + "accept": false, + "reason": "Unrelated model internals: MLA q_a_layernorm precision/config issue versus GlmMoeDsaIndexer torch.split return handling." + }, + { + "left": "issue:43441", + "right": "issue:44683", + "accept": false, + "reason": "Both mention attention, but one is a Ministral FlashAttention regression and the other is a torch>=2.9 compiled flex_attention failure." + }, + { + "left": "issue:43723", + "right": "issue:44162", + "accept": false, + "reason": "Tokenizer loading regression and ESM2 model breakage are different issues and likely different subsystems." + }, + { + "left": "issue:42175", + "right": "issue:43976", + "accept": false, + "reason": "Package/backend dependency installation issue versus Python version compatibility breakage; not the same defect." + }, + { + "left": "issue:44361", + "right": "issue:45593", + "accept": false, + "reason": "Different behaviors in different components: MLukeTokenizer task attribute error versus D-FINE auxiliary-loss logic." + }, + { + "left": "issue:44016", + "right": "issue:45823", + "accept": false, + "reason": "Notebook syntax error is unrelated to Gemma4 device-map forward mismatch." + }, + { + "left": "issue:44671", + "right": "issue:44779", + "accept": false, + "reason": "Different model families and failure modes: CamemBERT masked-LM predictions versus Deepseek tokenizer output regression." + }, + { + "left": "issue:43262", + "right": "issue:44521", + "accept": false, + "reason": "Both involve chat-template-related APIs, but one is audio sampling-rate defaulting and the other is multimodal assistant-mask generation; distinct bugs." + }, + { + "left": "issue:43673", + "right": "issue:43761", + "accept": false, + "reason": "Generation cache missing in chunked prefill is unrelated to CLIPVisionModel hidden_states handling." + }, + { + "left": "issue:43901", + "right": "issue:43994", + "accept": false, + "reason": "Docs mismatch for return_all_scores is unrelated to SigLIP2 model/pipeline output quality." + }, + { + "left": "issue:44987", + "right": "issue:45542", + "accept": false, + "reason": "Model loading failure for physical-intelligence/fast is unrelated to the TensorFlow backend install problem." + }, + { + "left": "issue:43784", + "right": "issue:45542", + "accept": false, + "reason": "Sentence-transformers import NameError and missing TensorFlow backend are different dependency/import regressions." + }, + { + "left": "issue:43824", + "right": "issue:43976", + "accept": false, + "reason": "ImportError for a specific Qwen VL class is unrelated to the Python version compatibility report." + }, + { + "left": "issue:43450", + "right": "issue:44617", + "accept": false, + "reason": "Video processor batched-shape bug and Sam3Video CUDA OOM are different problems, not one code-path fix." + }, + { + "left": "issue:44297", + "right": "issue:45397", + "accept": false, + "reason": "Tokenizer save_pretrained metadata mismatch is unrelated to Gemma4 zero3 loading." + }, + { + "left": "issue:43525", + "right": "issue:43881", + "accept": false, + "reason": "Missing pad_token_id on Llama4Config and glm-4v-9b loading failure are not the same underlying bug." + }, + { + "left": "issue:43901", + "right": "issue:44246", + "accept": false, + "reason": "A docs discrepancy and an occasional slow import are unrelated." + }, + { + "left": "issue:44188", + "right": "issue:44336", + "accept": false, + "reason": "Attention-kernel divergence under torch.compile is unrelated to ANSI codes in loading_report." + }, + { + "left": "issue:43525", + "right": "issue:43643", + "accept": false, + "reason": "Both involve config-ish loading behavior, but one is a missing config attribute and the other is trust_remote_code returning incomplete fields; different defects." + }, + { + "left": "issue:41628", + "right": "issue:43976", + "accept": false, + "reason": "ImportError for AutoImageProcessor is unrelated to the Python version support issue." + }, + { + "left": "issue:44589", + "right": "issue:45362", + "accept": false, + "reason": "Float8 storage lookup failure and Qwen3.5 chat crash are different runtime issues." + }, + { + "left": "issue:45561", + "right": "issue:45563", + "accept": false, + "reason": "pytest-xdist file race and stale num_return_sequences warning are unrelated test/runtime behaviors." + }, + { + "left": "issue:42491", + "right": "issue:43299", + "accept": false, + "reason": "Both mention Qwen3 MoE loading, but one is a LoRA model compatibility issue and the other is a Qwen3VL MoE loading regression; not clearly the same concrete bug." + }, + { + "left": "issue:44336", + "right": "issue:44373", + "accept": false, + "reason": "Terminal ANSI-code leakage is unrelated to the wrong docstring for position_ids." + }, + { + "left": "issue:44242", + "right": "issue:45593", + "accept": false, + "reason": "Router-logits load balancing loss logic is unrelated to D-FINE auxiliary losses when denoising is off." + } + ] + }, + "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": 18, + "review_unit_index": 35, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:36246", + "issue:41950", + "issue:42757", + "issue:42886", + "issue:43232", + "issue:43257", + "issue:43262", + "issue:43352", + "issue:43526", + "issue:43531", + "issue:43596", + "issue:43673", + "issue:43901", + "issue:44162", + "issue:44220", + "issue:44242", + "issue:44246", + "issue:44295", + "issue:44336", + "issue:44351", + "issue:44485", + "issue:44496", + "issue:44561", + "issue:44749", + "issue:44779", + "issue:44857", + "issue:44871", + "issue:44877", + "issue:44987", + "issue:44995", + "issue:45003", + "issue:45137", + "issue:45200", + "issue:45250", + "issue:45292", + "issue:45362", + "issue:45375", + "issue:45397", + "issue:45412", + "issue:45459", + "issue:45561", + "issue:45584", + "issue:45701", + "issue:45759", + "issue:45987", + "issue:46082", + "issue:46123", + "issue:46132" + ], + "soft_pairs": [ + "issue:44295|issue:45397", + "issue:44246|issue:44496", + "issue:41950|issue:44220", + "issue:43531|issue:45362", + "issue:42886|issue:46132", + "issue:43526|issue:45003", + "issue:36246|issue:42757", + "issue:36246|issue:44351", + "issue:44336|issue:44485", + "issue:45003|issue:45561", + "issue:44246|issue:44871", + "issue:43352|issue:45250", + "issue:43673|issue:44561", + "issue:36246|issue:43596", + "issue:44242|issue:44857", + "issue:44779|issue:45292", + "issue:44162|issue:45200", + "issue:44295|issue:44749", + "issue:43901|issue:45003", + "issue:45412|issue:46123", + "issue:44496|issue:45003", + "issue:45292|issue:45584", + "issue:45987|issue:46082", + "issue:44877|issue:45375", + "issue:45292|issue:45701", + "issue:43257|issue:45137", + "issue:43262|issue:44246", + "issue:45375|issue:45759", + "issue:44987|issue:45459", + "issue:43232|issue:44336", + "issue:44995|issue:45200" + ], + "prepared_review_unit_hash": "9d6ca5a83bd0ea7b7d318152f201b88a6c37e80aa720ec68ecfcd91a79f25b40", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 36859, + "estimated_input_tokens": 9215, + "estimated_eval_tokens": 18686 + }, + "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-, tokenizer-, pipeline-, config-, or Deepspeed-specific bugs. I do not see any soft edge that safely represents the same underlying issue/change, so no merges are justified.", + "confidence": 0.43, + "canonical_issue_reason": "issue:45701 is the broadest open regression report with multiple inbound references; it is the best representative placeholder in an otherwise mixed cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45701 is the most globally useful issue here because it is broad, user-facing, and has the strongest evidence of attracting related reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44295", + "right": "issue:45397", + "accept": false, + "reason": "Different failures: position_ids buffer access vs Gemma-4 Zero3 from_pretrained loading." + }, + { + "left": "issue:44246", + "right": "issue:44496", + "accept": false, + "reason": "Generic import slowdown versus unrecognized-model config error; unrelated code paths." + }, + { + "left": "issue:41950", + "right": "issue:44220", + "accept": false, + "reason": "Both involve processors, but one is video-classification pipeline lookup and the other is fbank feature extraction." + }, + { + "left": "issue:43531", + "right": "issue:45362", + "accept": false, + "reason": "Both mention Qwen, but one is a sliding_window bug and the other is a chat crash; different underlying problems." + }, + { + "left": "issue:42886", + "right": "issue:46132", + "accept": false, + "reason": "Offline tokenizer cache loading and AttentionInterface.register behavior are unrelated." + }, + { + "left": "issue:43526", + "right": "issue:45003", + "accept": false, + "reason": "BeitImageProcessorFast label reduction bug is unrelated to unsafe sys.modules access." + }, + { + "left": "issue:36246", + "right": "issue:42757", + "accept": false, + "reason": "Different missing imports from different packages: Qwen image processor vs huggingface_hub helper." + }, + { + "left": "issue:36246", + "right": "issue:44351", + "accept": false, + "reason": "Both are import errors, but they concern different symbols and different fixes." + }, + { + "left": "issue:44336", + "right": "issue:44485", + "accept": false, + "reason": "ANSI-color output bug is unrelated to GLM-5 RoPE implementation." + }, + { + "left": "issue:45003", + "right": "issue:45561", + "accept": false, + "reason": "Unsafe sys.modules access and pytest-xdist file races are unrelated." + }, + { + "left": "issue:44246", + "right": "issue:44871", + "accept": false, + "reason": "Import performance regression vs Gemma eos_token_id config inconsistency." + }, + { + "left": "issue:43352", + "right": "issue:45250", + "accept": false, + "reason": "Generic Flash Attention topic, but one is a model-specific unsupported warning and the other is a broader feature request." + }, + { + "left": "issue:43673", + "right": "issue:44561", + "accept": false, + "reason": "Chunked-prefill cache bug and trust_remote_code breakage from removed helper are different issues." + }, + { + "left": "issue:36246", + "right": "issue:43596", + "accept": false, + "reason": "Qwen import failure and DeepSpeed/BertModel index error are unrelated." + }, + { + "left": "issue:44242", + "right": "issue:44857", + "accept": false, + "reason": "Load-balancing loss omission is unrelated to float16 AMP crash in LwDetrImageLoss." + }, + { + "left": "issue:44779", + "right": "issue:45292", + "accept": false, + "reason": "Tokenizer output regression and resize_token_embeddings not updating output embeddings are different bugs." + }, + { + "left": "issue:44162", + "right": "issue:45200", + "accept": false, + "reason": "ESM2 breakage and Gemma-4 mm_token_type_ids defaulting are unrelated model issues." + }, + { + "left": "issue:44295", + "right": "issue:44749", + "accept": false, + "reason": "Position_ids buffer access and slower filtering after upgrade are not the same bug." + }, + { + "left": "issue:43901", + "right": "issue:45003", + "accept": false, + "reason": "Documentation mismatch in TextClassificationPipeline is unrelated to sys.modules access." + }, + { + "left": "issue:45412", + "right": "issue:46123", + "accept": false, + "reason": "RT-DETR memory cleanup and MaskGenerationPipeline final-batch dropping are unrelated pipeline bugs." + }, + { + "left": "issue:44496", + "right": "issue:45003", + "accept": false, + "reason": "Model config recognition failure is unrelated to sys.modules lookup safety." + }, + { + "left": "issue:45292", + "right": "issue:45584", + "accept": false, + "reason": "Embedding resize bug and Whisper empty-transcription failure are different code paths." + }, + { + "left": "issue:45987", + "right": "issue:46082", + "accept": false, + "reason": "Cache length return type and LlamaConfig head_dim validation are unrelated." + }, + { + "left": "issue:44877", + "right": "issue:45375", + "accept": false, + "reason": "Both are config-validation issues, but they affect different models and different missing fields." + }, + { + "left": "issue:45292", + "right": "issue:45701", + "accept": false, + "reason": "Embedding resize behavior and version-dependent tokenization regression are not the same underlying issue." + }, + { + "left": "issue:43257", + "right": "issue:45137", + "accept": false, + "reason": "Both touch DeepSpeed, but one is Qwen3 MoE weight conversion and the other is a deque underflow during ZeRO3 init." + }, + { + "left": "issue:43262", + "right": "issue:44246", + "accept": false, + "reason": "Audio chat-template sampling-rate defaulting and import slowdown are unrelated." + }, + { + "left": "issue:45375", + "right": "issue:45759", + "accept": false, + "reason": "Missing vision-config field and text_config unwrapping for composite Qwen configs are distinct composite-model bugs." + }, + { + "left": "issue:44987", + "right": "issue:45459", + "accept": false, + "reason": "Loading a specific model under v5.1.0 and protobuf-masked tokenizer errors are unrelated." + }, + { + "left": "issue:43232", + "right": "issue:44336", + "accept": false, + "reason": "Generation kwargs update after sync_gpus and ANSI code emission are unrelated." + }, + { + "left": "issue:44995", + "right": "issue:45200", + "accept": false, + "reason": "GlmMoeDsa second-pass cache crash and Gemma-4 token type defaults 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": 18, + "review_unit_index": 36, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:28282", + "issue:30990", + "issue:42886", + "issue:42994", + "issue:43065", + "issue:43232", + "issue:43408", + "issue:43475", + "issue:43482", + "issue:43540", + "issue:43643", + "issue:43646", + "issue:43653", + "issue:43673", + "issue:43931", + "issue:44038", + "issue:44060", + "issue:44077", + "issue:44162", + "issue:44297", + "issue:44361", + "issue:44368", + "issue:44466", + "issue:44623", + "issue:44734", + "issue:44749", + "issue:44869", + "issue:44871", + "issue:44933", + "issue:44977", + "issue:44995", + "issue:45137", + "issue:45229", + "issue:45230", + "issue:45278", + "issue:45310", + "issue:45335", + "issue:45356", + "issue:45381", + "issue:45397", + "issue:45459", + "issue:45563", + "issue:45588", + "issue:45600", + "issue:45656", + "issue:46077", + "issue:46123", + "issue:46132" + ], + "soft_pairs": [ + "issue:43475|issue:43540", + "issue:43065|issue:43408", + "issue:42886|issue:45229", + "issue:44466|issue:45335", + "issue:43653|issue:44162", + "issue:45397|issue:45656", + "issue:44060|issue:44077", + "issue:44995|issue:45588", + "issue:44368|issue:44977", + "issue:45137|issue:45600", + "issue:44368|issue:45563", + "issue:43673|issue:44734", + "issue:43931|issue:44368", + "issue:43065|issue:46123", + "issue:44933|issue:45459", + "issue:44869|issue:45356", + "issue:44297|issue:44623", + "issue:43232|issue:43673", + "issue:44361|issue:44749", + "issue:43646|issue:44162", + "issue:44871|issue:46132", + "issue:43673|issue:45278", + "issue:44623|issue:45656", + "issue:44749|issue:45656", + "issue:44749|issue:45310", + "issue:42994|issue:43482", + "issue:43475|issue:45381", + "issue:28282|issue:30990", + "issue:45230|issue:45397", + "issue:43475|issue:43643", + "issue:44038|issue:44623", + "issue:46077|issue:46123" + ], + "prepared_review_unit_hash": "e14707550622f4978da4b5b86376520a0d0547b4195608b8ae980a76fdd464ff", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 37143, + "estimated_input_tokens": 9286, + "estimated_eval_tokens": 18828 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are largely unrelated issue reports across different models, tokenizer paths, generation flows, and serialization behaviors. I do not see any soft pair that is safely mergeable as the same underlying bug/change, so all soft edges should be rejected. If a single representative issue is needed, 45397 is the clearest standalone bug report with a concrete failing path.", + "confidence": 0.93, + "canonical_issue_reason": "issue:45397 is the most focused, actionable bug report here: a concrete `gemma-4 zero3 from_pretrained` failure with a narrow code path and likely direct fix.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45397 is the best standalone issue candidate because it is specific, reproducible, and not just a broad symptom or regression description.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43475", + "right": "issue:43540", + "accept": false, + "reason": "Both mention video/SAM/Qwen families, but they describe different model paths and different failures; not the same bug." + }, + { + "left": "issue:43065", + "right": "issue:43408", + "accept": false, + "reason": "Same broad SAM3 area, but one is a dummy Conv2d concern and the other is a model-type mismatch warning; different underlying issues." + }, + { + "left": "issue:42886", + "right": "issue:45229", + "accept": false, + "reason": "Tokenizer offline-cache loading and multi-GPU CUDA OOM are unrelated failure modes." + }, + { + "left": "issue:44466", + "right": "issue:45335", + "accept": false, + "reason": "Both touch tied/embedding behavior, but one is about serialization inconsistency and the other about resize_token_embeddings not updating a decoder embedding; different code paths." + }, + { + "left": "issue:43653", + "right": "issue:44162", + "accept": false, + "reason": "BigBirdTokenizer special-token decode bug and ESM2 breakage are unrelated." + }, + { + "left": "issue:45397", + "right": "issue:45656", + "accept": false, + "reason": "Gemma-4 from_pretrained on ZeRO3 and optimizer stepping twice under deepspeed are different bugs in different subsystems." + }, + { + "left": "issue:44060", + "right": "issue:44077", + "accept": false, + "reason": "One is a tied-weights warning in Qwen3-Next, the other is a PatchTSMixer post_init policy issue; not the same problem." + }, + { + "left": "issue:44995", + "right": "issue:45588", + "accept": false, + "reason": "Stale indexer cache on second forward pass and flash_attention None-handling crash are distinct runtime failures." + }, + { + "left": "issue:44368", + "right": "issue:44977", + "accept": false, + "reason": "A tie_word_embeddings warning during fine-tuning is not the same as a flash-attention generation failure." + }, + { + "left": "issue:45137", + "right": "issue:45600", + "accept": false, + "reason": "DeepSpeed deque underflow and auto_mappings CI breakage are unrelated." + }, + { + "left": "issue:44368", + "right": "issue:45563", + "accept": false, + "reason": "A tied-embedding warning and a stale num_return_sequences warning affect different generation logic." + }, + { + "left": "issue:43673", + "right": "issue:44734", + "accept": false, + "reason": "Both involve cache/generation, but one is missing cache in chunked prefill and the other is a KV-cache continuation indexing crash; not clearly the same bug." + }, + { + "left": "issue:43931", + "right": "issue:44368", + "accept": false, + "reason": "Qwen3-VL weight-shape mismatch and a tied-weights warning are different issues." + }, + { + "left": "issue:43065", + "right": "issue:46123", + "accept": false, + "reason": "SAM3 pixel decoder initialization and MaskGenerationPipeline partial-batch handling are unrelated." + }, + { + "left": "issue:44933", + "right": "issue:45459", + "accept": false, + "reason": "Nonexistent image_utils import and protobuf decode-error handling are different tokenizer/import problems." + }, + { + "left": "issue:44869", + "right": "issue:45356", + "accept": false, + "reason": "Whisper timestamp decode crash and Kimi-K2.5 tokenizer regression are unrelated tokenization bugs." + }, + { + "left": "issue:44297", + "right": "issue:44623", + "accept": false, + "reason": "Tokenizer save_pretrained metadata mismatch and processor.save_pretrained missing files are both serialization-related but not the same concrete failure." + }, + { + "left": "issue:43232", + "right": "issue:43673", + "accept": false, + "reason": "Generation kwargs update after sync_gpus and missing cache in chunked prefill are separate generation-path issues." + }, + { + "left": "issue:44361", + "right": "issue:44749", + "accept": false, + "reason": "MLukeTokenizer attribute errors and a performance regression after upgrade are unrelated." + }, + { + "left": "issue:43646", + "right": "issue:44162", + "accept": false, + "reason": "Custom model initialization breakage and ESM2 breakage are different root problems." + }, + { + "left": "issue:44871", + "right": "issue:46132", + "accept": false, + "reason": "Gemma3 eos_token_id inconsistency and AttentionInterface.register behavior change are unrelated." + }, + { + "left": "issue:43673", + "right": "issue:45278", + "accept": false, + "reason": "A specific chunked_prefill cache bug and broad import errors after upgrade are too different to merge." + }, + { + "left": "issue:44623", + "right": "issue:45656", + "accept": false, + "reason": "Processor save_pretrained file emission and optimizer double-step under deepspeed are unrelated." + }, + { + "left": "issue:44749", + "right": "issue:45656", + "accept": false, + "reason": "Performance slowdown in filtering data and double optimizer stepping are not the same bug." + }, + { + "left": "issue:44749", + "right": "issue:45310", + "accept": false, + "reason": "Data filtering slowdown and Qwen3.5 Moe from_pretrained failure are unrelated." + }, + { + "left": "issue:42994", + "right": "issue:43482", + "accept": false, + "reason": "Quantized model saving failure and GGUF loading failure are opposite directions and distinct code paths." + }, + { + "left": "issue:43475", + "right": "issue:45381", + "accept": false, + "reason": "Both concern video/VL models, but the reported attributes and failures differ; not a single concrete bug." + }, + { + "left": "issue:28282", + "right": "issue:30990", + "accept": false, + "reason": "AutoModel missing PyTorch and Sentence Transformers hanging on load are unrelated environment/runtime issues." + }, + { + "left": "issue:45230", + "right": "issue:45397", + "accept": false, + "reason": "A placeholder generic bug report cannot be confidently merged with a specific Gemma-4 loading bug." + }, + { + "left": "issue:44038", + "right": "issue:44623", + "accept": false, + "reason": "Qwen3-VL-Moe from_pretrained error and processor.save_pretrained missing files are different failures." + }, + { + "left": "issue:46077", + "right": "issue:46123", + "accept": false, + "reason": "Bidirectional-attention mask construction and final partial-batch handling in MaskGenerationPipeline 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": 18, + "review_unit_index": 37, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:30990", + "issue:35141", + "issue:36010", + "issue:41628", + "issue:42175", + "issue:42222", + "issue:42371", + "issue:42673", + "issue:42831", + "issue:42898", + "issue:42994", + "issue:43232", + "issue:43278", + "issue:43377", + "issue:43475", + "issue:43479", + "issue:43504", + "issue:43525", + "issue:43526", + "issue:43673", + "issue:43704", + "issue:43824", + "issue:43873", + "issue:43976", + "issue:44038", + "issue:44188", + "issue:44222", + "issue:44242", + "issue:44351", + "issue:44496", + "issue:44568", + "issue:44661", + "issue:44792", + "issue:44869", + "issue:44933", + "issue:44945", + "issue:44964", + "issue:45081", + "issue:45092", + "issue:45103", + "issue:45137", + "issue:45200", + "issue:45310", + "issue:45375", + "issue:45478", + "issue:45561", + "issue:45712", + "issue:45753" + ], + "soft_pairs": [ + "issue:41628|issue:44933", + "issue:42371|issue:45092", + "issue:43475|issue:44661", + "issue:42994|issue:43525", + "issue:45375|issue:45478", + "issue:45081|issue:45103", + "issue:45310|issue:45375", + "issue:43873|issue:44945", + "issue:41628|issue:44351", + "issue:43232|issue:44188", + "issue:42673|issue:43704", + "issue:42831|issue:44222", + "issue:44945|issue:45753", + "issue:36010|issue:41628", + "issue:44222|issue:45310", + "issue:42994|issue:43824", + "issue:44964|issue:45137", + "issue:42175|issue:45712", + "issue:43673|issue:44188", + "issue:43232|issue:44792", + "issue:44038|issue:45310", + "issue:44496|issue:45561", + "issue:43526|issue:45561", + "issue:43504|issue:45561", + "issue:43377|issue:44242", + "issue:42222|issue:45200", + "issue:42175|issue:44933", + "issue:42898|issue:44568", + "issue:43976|issue:44933", + "issue:30990|issue:35141", + "issue:43479|issue:44869", + "issue:43278|issue:45137" + ], + "prepared_review_unit_hash": "c1494400716b455290108e593ba195e3d31b190270143f93c739b5153dd17d3b", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 37234, + "estimated_input_tokens": 9309, + "estimated_eval_tokens": 18874 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The packet is mostly heterogeneous, but it does contain two separate exact-duplicate issue families: the Qwen3.5 MoE `from_pretrained` error reports (45310\u219445478) and the Qwen3ForCausalLM VRAM leak reports (42673\u219443704). The Qwen3_5MoeVisionConfig config-field report (45375) is a different bug and should not be merged into the Qwen3.5 MoE family.", + "confidence": 0.86, + "canonical_issue_reason": "Issue 45310 is the best canonical issue for the Qwen3.5 MoE `from_pretrained` duplicate family: it clearly states the failure mode and is directly duplicated by 45478. The packet also contains a separate exact-duplicate family at 42673/43704, but 45310 is the stronger single canonical representative.", + "canonical_pr_reason": null, + "best_issue_reason": "45310 is the strongest single issue representative because it has an exact-title duplicate and a concrete, well-scoped bug description. It is a better canonical pick than the other Qwen3.5-related reports, while 42673 is best treated as the canonical for a separate duplicate family.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:41628", + "right": "issue:44933", + "accept": false, + "reason": "Both are import-related, but they concern different missing symbols and different modules; not the same bug." + }, + { + "left": "issue:42371", + "right": "issue:45092", + "accept": false, + "reason": "TF32 API guidance and InternVL2 meta-init incompatibility are unrelated failure modes." + }, + { + "left": "issue:43475", + "right": "issue:44661", + "accept": false, + "reason": "One is a missing attribute on SAM3 vision output; the other is a model-addition helper failure. Different code paths." + }, + { + "left": "issue:42994", + "right": "issue:43525", + "accept": false, + "reason": "Quantized model saving and missing `pad_token_id` in `Llama4Config` are different issues." + }, + { + "left": "issue:45375", + "right": "issue:45478", + "accept": false, + "reason": "45375 is a config-field report (`deepstack_visual_indexes`), not the same `from_pretrained` failure as 45478." + }, + { + "left": "issue:45081", + "right": "issue:45103", + "accept": false, + "reason": "Both crash with `AttributeError`, but in different utilities and on different objects; not the same underlying bug." + }, + { + "left": "issue:45310", + "right": "issue:45375", + "accept": false, + "reason": "45310 is a Qwen3.5 MoE `from_pretrained` error; 45375 reports a different config-field omission." + }, + { + "left": "issue:43873", + "right": "issue:44945", + "accept": false, + "reason": "Quantization offloading and pipeline-parallel output corruption are separate problems." + }, + { + "left": "issue:41628", + "right": "issue:44351", + "accept": false, + "reason": "Both are import errors, but for different names and different release breakages." + }, + { + "left": "issue:43232", + "right": "issue:44188", + "accept": false, + "reason": "Generation kwargs propagation and attention-kernel divergence under `torch.compile` are unrelated." + }, + { + "left": "issue:42673", + "right": "issue:43704", + "accept": true, + "reason": "Exact same VRAM leak report for Qwen3ForCausalLM in multiple dataloader threads." + }, + { + "left": "issue:42831", + "right": "issue:44222", + "accept": false, + "reason": "FineGrainedFP8 accuracy regression and FP8 `save_pretrained` moe are different symptoms and likely different fixes." + }, + { + "left": "issue:44945", + "right": "issue:45753", + "accept": false, + "reason": "Pipeline parallelism output errors and Qwen3_5 infinite-loop-on-image are distinct bugs." + }, + { + "left": "issue:36010", + "right": "issue:41628", + "accept": false, + "reason": "GenerationMixin import removal and AutoImageProcessor import failure are separate import breakages." + }, + { + "left": "issue:44222", + "right": "issue:45310", + "accept": false, + "reason": "FP8 `save_pretrained` moe and Qwen3.5 MoE `from_pretrained` are different issues." + }, + { + "left": "issue:42994", + "right": "issue:43824", + "accept": false, + "reason": "Quantized saving failure and Qwen2.5-VL import failure do not describe the same bug." + }, + { + "left": "issue:44964", + "right": "issue:45137", + "accept": false, + "reason": "Phi-4 multimodal loading and DeepSpeed ZeRO3 deque underflow are unrelated." + }, + { + "left": "issue:42175", + "right": "issue:45712", + "accept": false, + "reason": "Packaging dependency omission and dummy class cleanup are unrelated maintenance issues." + }, + { + "left": "issue:43673", + "right": "issue:44188", + "accept": false, + "reason": "Generation cache missing in chunked prefill and attention-kernel branching divergence are different defects." + }, + { + "left": "issue:43232", + "right": "issue:44792", + "accept": false, + "reason": "Generation kwargs handling and janus image-generation test failure are not the same bug." + }, + { + "left": "issue:44038", + "right": "issue:45310", + "accept": false, + "reason": "Both mention Qwen MoE, but they target different model variants and different version breakages; too broad to merge." + }, + { + "left": "issue:44496", + "right": "issue:45561", + "accept": false, + "reason": "Model config recognition and xdist race on `captured_info.txt` are unrelated." + }, + { + "left": "issue:43526", + "right": "issue:45561", + "accept": false, + "reason": "Fast image label reduction bug and xdist file race are different areas and failures." + }, + { + "left": "issue:43504", + "right": "issue:45561", + "accept": false, + "reason": "Beit pretrained-load legacy-field issue is unrelated to the xdist race." + }, + { + "left": "issue:43377", + "right": "issue:44242", + "accept": false, + "reason": "MIMI padding-mask mismatch and router-logit load-balancing loss are different model behaviors." + }, + { + "left": "issue:42222", + "right": "issue:45200", + "accept": false, + "reason": "VitPose breakage and Gemma 4 token-type defaulting are unrelated." + }, + { + "left": "issue:42175", + "right": "issue:44933", + "accept": false, + "reason": "Backend packaging and missing image_utils export are different problems." + }, + { + "left": "issue:42898", + "right": "issue:44568", + "accept": false, + "reason": "Tokenizer cleanup behavior change and BOS/EOS special-token regression are related to tokenization, but not the same concrete failure." + }, + { + "left": "issue:43976", + "right": "issue:44933", + "accept": false, + "reason": "Python version compatibility and missing image_utils import are unrelated." + }, + { + "left": "issue:30990", + "right": "issue:35141", + "accept": false, + "reason": "Sentence-Transformers loading hang and embedding reinitialization are distinct issues." + }, + { + "left": "issue:43479", + "right": "issue:44869", + "accept": false, + "reason": "Multimodal config initialization and Whisper timestamp decode crash are unrelated." + }, + { + "left": "issue:43278", + "right": "issue:45137", + "accept": false, + "reason": "Embedding dtype drift during evaluation and DeepSpeed deque underflow are different bugs." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The summary is a bit too strong for the packet. The Qwen3.5 pair is supported for issue:45310 \u2194 issue:45478 (identical titles), but issue:45375 is a different config-field report and is not grounded as the same bug. Also, the Qwen3ForCausalLM VRAM leak pair issue:42673 \u2194 issue:43704 is an exact-title duplicate in the packet, so the writeup should treat it separately and avoid implying there is only one substantial duplicate family. Several accepted soft-edge verdicts therefore look non-conservative." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 38, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:22355", + "issue:28282", + "issue:30333", + "issue:33290", + "issue:34689", + "issue:35141", + "issue:36331", + "issue:41628", + "issue:42757", + "issue:42831", + "issue:42898", + "issue:43116", + "issue:43479", + "issue:43644", + "issue:43723", + "issue:43784", + "issue:43824", + "issue:43874", + "issue:44060", + "issue:44079", + "issue:44188", + "issue:44190", + "issue:44206", + "issue:44222", + "issue:44295", + "issue:44297", + "issue:44373", + "issue:44448", + "issue:44484", + "issue:44568", + "issue:44623", + "issue:44734", + "issue:44779", + "issue:44933", + "issue:44977", + "issue:45071", + "issue:45092", + "issue:45103", + "issue:45125", + "issue:45245", + "issue:45310", + "issue:45356", + "issue:45405", + "issue:45656", + "issue:45712", + "issue:45812", + "issue:46082", + "issue:46132" + ], + "soft_pairs": [ + "issue:43479|issue:44079", + "issue:44623|issue:45310", + "issue:42898|issue:45812", + "issue:44188|issue:44373", + "issue:30333|issue:33290", + "issue:43784|issue:43824", + "issue:44734|issue:45103", + "issue:43723|issue:44568", + "issue:42898|issue:44779", + "issue:42898|issue:45356", + "issue:43874|issue:45092", + "issue:44060|issue:45125", + "issue:45405|issue:45712", + "issue:42898|issue:44448", + "issue:44295|issue:45656", + "issue:44448|issue:46132", + "issue:44933|issue:45712", + "issue:44222|issue:44297", + "issue:43116|issue:44190", + "issue:44297|issue:45656", + "issue:43479|issue:43644", + "issue:44190|issue:46082", + "issue:42831|issue:44623", + "issue:30333|issue:36331", + "issue:44222|issue:44623", + "issue:45245|issue:46132", + "issue:30333|issue:34689", + "issue:41628|issue:42757", + "issue:22355|issue:28282", + "issue:44206|issue:45103", + "issue:45071|issue:46082", + "issue:28282|issue:30333", + "issue:30333|issue:35141", + "issue:44297|issue:45310", + "issue:43723|issue:44977", + "issue:44222|issue:44484" + ], + "prepared_review_unit_hash": "e62a25818d3d5fd749843b59b4b847541142639c998716fced81c7ac2fe86a85", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 36, + "serialized_chars": 38511, + "estimated_input_tokens": 9628, + "estimated_eval_tokens": 19512 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous: import errors, tokenizer regressions, training-loop bugs, config issues, and save/load problems. I don\u2019t see any soft-edge pair that clearly describes the same underlying bug/change, so all candidates should be kept separate.", + "confidence": 0.94, + "canonical_issue_reason": "issue:42898 is the broadest, highest-signal regression report in the set (v5 tokenizer behavior), so it is the closest thing to a cluster representative, though the cluster is not actually a duplicate set.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:42898 is the most central and reusable issue title among these mixed reports, but it should not be treated as a duplicate umbrella for the others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43479", + "right": "issue:44079", + "accept": false, + "reason": "Both involve None handling, but they affect different classes and code paths (config defaults vs ModelOutput key assignment)." + }, + { + "left": "issue:44623", + "right": "issue:45310", + "accept": false, + "reason": "Different features and failures: processor save_pretrained file output vs Qwen3.5 MoE from_pretrained loading." + }, + { + "left": "issue:42898", + "right": "issue:45812", + "accept": false, + "reason": "Both are tokenizer regressions, but they concern different models and distinct behavior changes." + }, + { + "left": "issue:44188", + "right": "issue:44373", + "accept": false, + "reason": "Attention-kernel divergence and a docstring issue are unrelated." + }, + { + "left": "issue:30333", + "right": "issue:33290", + "accept": false, + "reason": "MLFlow reporting on failure and a DeepSpeed OOM in Adafactor are different training bugs." + }, + { + "left": "issue:43784", + "right": "issue:43824", + "accept": false, + "reason": "Both are import errors, but for different symbols, packages, and root causes." + }, + { + "left": "issue:44734", + "right": "issue:45103", + "accept": false, + "reason": "Serving KV-cache indexing and auto_docstring annotation handling are unrelated code paths." + }, + { + "left": "issue:43723", + "right": "issue:44568", + "accept": false, + "reason": "Both hit tokenizer behavior in v5, but one is a loading issue and the other is special-token insertion semantics." + }, + { + "left": "issue:42898", + "right": "issue:44779", + "accept": false, + "reason": "Different tokenizer regressions for different model families; not the same concrete bug." + }, + { + "left": "issue:42898", + "right": "issue:45356", + "accept": false, + "reason": "Both are tokenizer regressions, but the reported codec/warning issue is distinct from clean_up_tokenization_spaces behavior." + }, + { + "left": "issue:43874", + "right": "issue:45092", + "accept": false, + "reason": "Different multimodal loading failures affecting different architectures and mechanisms." + }, + { + "left": "issue:44060", + "right": "issue:45125", + "accept": false, + "reason": "Qwen model-specific tied-weight warning and tensor-parallel plan omission are separate issues." + }, + { + "left": "issue:45405", + "right": "issue:45712", + "accept": false, + "reason": "A dependency version bump and leftover dummy classes are unrelated repository hygiene problems." + }, + { + "left": "issue:42898", + "right": "issue:44448", + "accept": false, + "reason": "Both are v4\u2192v5 behavioral regressions, but they affect different models and output paths." + }, + { + "left": "issue:44295", + "right": "issue:45656", + "accept": false, + "reason": "Buffer access/readback and optimizer stepping with DeepSpeed are unrelated bugs." + }, + { + "left": "issue:44933", + "right": "issue:45712", + "accept": false, + "reason": "Public import exposure from image_utils and dummy-class leakage into dir(transformers) are different export problems." + }, + { + "left": "issue:44222", + "right": "issue:44297", + "accept": false, + "reason": "Both involve save_pretrained, but one is MoE FP8 saving and the other is tokenizer config metadata." + }, + { + "left": "issue:43116", + "right": "issue:44190", + "accept": false, + "reason": "Different example scripts and different failures: multi-label inference output vs local dataset loading." + }, + { + "left": "issue:44297", + "right": "issue:45656", + "accept": false, + "reason": "Tokenizer save metadata and DeepSpeed optimizer stepping are unrelated." + }, + { + "left": "issue:43479", + "right": "issue:43644", + "accept": false, + "reason": "Both mention initialization/defaults, but one is multimodal config defaults and the other is non-persistent buffer filling." + }, + { + "left": "issue:44190", + "right": "issue:46082", + "accept": false, + "reason": "Dataset loading in an example script and LlamaConfig validation are unrelated." + }, + { + "left": "issue:42831", + "right": "issue:44623", + "accept": false, + "reason": "FP8 accuracy and processor save_pretrained file emission are different concerns." + }, + { + "left": "issue:30333", + "right": "issue:36331", + "accept": false, + "reason": "MLFlow reporting on failed jobs and Trainer compute_loss signature mismatch are separate training-loop regressions." + }, + { + "left": "issue:44222", + "right": "issue:44623", + "accept": false, + "reason": "Both touch save_pretrained, but the artifacts and failure modes are different." + }, + { + "left": "issue:45245", + "right": "issue:46132", + "accept": false, + "reason": "Category-count limits and AttentionInterface registration behavior are unrelated." + }, + { + "left": "issue:30333", + "right": "issue:34689", + "accept": false, + "reason": "Training-job failure reporting and Llama Vision model loading are unrelated." + }, + { + "left": "issue:41628", + "right": "issue:42757", + "accept": false, + "reason": "Import errors in different dependency namespaces with different missing symbols." + }, + { + "left": "issue:22355", + "right": "issue:28282", + "accept": false, + "reason": "Both are import/setup failures, but one is a missing transformers submodule and the other is missing PyTorch." + }, + { + "left": "issue:44206", + "right": "issue:45103", + "accept": false, + "reason": "Feature extractor argument compatibility and docstring processing are unrelated." + }, + { + "left": "issue:45071", + "right": "issue:46082", + "accept": false, + "reason": "Config type-checking in v5.4 and explicit head_dim validation are different config issues." + }, + { + "left": "issue:28282", + "right": "issue:30333", + "accept": false, + "reason": "Missing PyTorch import and MLFlow reporting on failed training are unrelated." + }, + { + "left": "issue:30333", + "right": "issue:35141", + "accept": false, + "reason": "Training failure reporting and embedding reinitialization after resize are different bugs." + }, + { + "left": "issue:44297", + "right": "issue:45310", + "accept": false, + "reason": "Tokenizer save metadata and Qwen3.5 MoE loading are unrelated." + }, + { + "left": "issue:43723", + "right": "issue:44977", + "accept": false, + "reason": "Both are tokenizer-related regressions, but one is AutoTokenizer loading and the other is DeepSeek tokenization correctness." + }, + { + "left": "issue:44222", + "right": "issue:44484", + "accept": false, + "reason": "FP8 save_pretrained and max_shard_size documentation/behavior are not the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 18, + "review_unit_index": 39, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:22355", + "issue:28218", + "issue:30990", + "issue:38175", + "issue:39401", + "issue:41628", + "issue:42757", + "issue:42831", + "issue:42886", + "issue:42907", + "issue:42994", + "issue:43066", + "issue:43232", + "issue:43479", + "issue:43723", + "issue:43756", + "issue:43761", + "issue:43824", + "issue:44038", + "issue:44060", + "issue:44261", + "issue:44263", + "issue:44295", + "issue:44360", + "issue:44373", + "issue:44442", + "issue:44483", + "issue:44485", + "issue:44514", + "issue:44568", + "issue:44623", + "issue:44829", + "issue:44871", + "issue:44913", + "issue:44991", + "issue:45092", + "issue:45103", + "issue:45278", + "issue:45310", + "issue:45357", + "issue:45406", + "issue:45593", + "issue:45753", + "issue:45834", + "issue:46002", + "issue:46050", + "issue:46082" + ], + "soft_pairs": [ + "issue:43232|issue:45834", + "issue:44263|issue:45092", + "issue:43761|issue:46082", + "issue:44483|issue:45406", + "issue:44373|issue:44485", + "issue:44295|issue:44623", + "issue:44263|issue:44360", + "issue:44514|issue:45103", + "issue:43723|issue:44871", + "issue:30990|issue:38175", + "issue:45092|issue:45357", + "issue:42907|issue:46050", + "issue:43723|issue:44991", + "issue:44829|issue:46002", + "issue:43723|issue:44442", + "issue:44261|issue:45092", + "issue:41628|issue:43824", + "issue:42831|issue:45310", + "issue:43479|issue:44913", + "issue:44263|issue:44373", + "issue:42831|issue:44038", + "issue:39401|issue:44568", + "issue:42757|issue:45278", + "issue:42886|issue:42994", + "issue:43232|issue:45593", + "issue:42831|issue:45753", + "issue:43756|issue:44060", + "issue:42886|issue:43066", + "issue:22355|issue:28218" + ], + "prepared_review_unit_hash": "41d4c6bf7abd1fd83f5eb2239cf8e08f390bb3602c2ec9018366c9d34090af64", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 35480, + "estimated_input_tokens": 8870, + "estimated_eval_tokens": 17996 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly unrelated bug reports; the soft pairs are only superficially similar by subsystem/topic and do not look like the same concrete underlying bug or change. The most central issue is the v5 tokenizer-loading regression cluster.", + "confidence": 0.79, + "canonical_issue_reason": "issue:43723 is the broadest and most representative tokenizer-loading regression report, and it has the strongest thematic overlap with multiple tokenizer-related soft candidates.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43723 is the best single issue to anchor this cluster because it captures the recurring v5 tokenizer-loading failure pattern better than the narrower, model-specific reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43232", + "right": "issue:45834", + "accept": false, + "reason": "Different failures: generation kwargs handling after sync_gpus vs Kosmos2.5 OCR index error." + }, + { + "left": "issue:44263", + "right": "issue:45092", + "accept": false, + "reason": "Unrelated bugs: torch.split return values in a DSA indexer vs old InternVL2 checkpoints failing under v5 meta init." + }, + { + "left": "issue:43761", + "right": "issue:46082", + "accept": false, + "reason": "Different model behaviors: CLIPVisionModel hidden_states regression vs LlamaConfig head_dim validation." + }, + { + "left": "issue:44483", + "right": "issue:45406", + "accept": false, + "reason": "Both are v5-era failures, but one is /v1/chat/completions request handling and the other is a Gemma4Processor attribute error." + }, + { + "left": "issue:44373", + "right": "issue:44485", + "accept": false, + "reason": "Docstring typo for position_ids is not the same bug as GLM-5 RoPE implementation." + }, + { + "left": "issue:44295", + "right": "issue:44623", + "accept": false, + "reason": "Registering position_ids as a buffer causing read errors is distinct from processor.save_pretrained missing files." + }, + { + "left": "issue:44263", + "right": "issue:44360", + "accept": false, + "reason": "Same general DSA indexer area, but different concrete issues: split outputs vs missing ReLU." + }, + { + "left": "issue:44514", + "right": "issue:45103", + "accept": false, + "reason": "Both involve processor/template code, but one is batched apply_chat_template padding behavior and the other is auto_docstring crashing on future annotations." + }, + { + "left": "issue:43723", + "right": "issue:44871", + "accept": false, + "reason": "Tokenizer loading regression vs eos_token_id config inconsistency; related area but different underlying bug." + }, + { + "left": "issue:30990", + "right": "issue:38175", + "accept": false, + "reason": "Sentence Transformers loading hang is unrelated to SigLIP2 zero-probability outputs." + }, + { + "left": "issue:45092", + "right": "issue:45357", + "accept": false, + "reason": "Meta-initialization incompatibility for old InternVL2 checkpoints is different from incorrect visual encoder keys on save_pretrained." + }, + { + "left": "issue:42907", + "right": "issue:46050", + "accept": false, + "reason": "Both mention quantization/save issues, but failing to save dequantized Ministrals/Devstrals is not the same as sinq quantized save_pretrained/from_pretrained breakage." + }, + { + "left": "issue:43723", + "right": "issue:44991", + "accept": false, + "reason": "Both are tokenizer-load regressions, but the affected models and failure modes are not specific enough to treat as the same bug." + }, + { + "left": "issue:44829", + "right": "issue:46002", + "accept": false, + "reason": "Degenerate training with flash_attention_3 is unrelated to NaNs after checkout + init." + }, + { + "left": "issue:43723", + "right": "issue:44442", + "accept": false, + "reason": "AutoTokenizer failing on FastSpeech2ConformerTokenizer may be related by theme, but it is not clearly the same bug as issue loading tokenizer in v5." + }, + { + "left": "issue:44261", + "right": "issue:45092", + "accept": false, + "reason": "MLA q_a_layernorm precision error and old InternVL2 meta-init incompatibility are different classes of bugs." + }, + { + "left": "issue:41628", + "right": "issue:43824", + "accept": false, + "reason": "Both are import errors, but for different symbols/models and likely different causes." + }, + { + "left": "issue:42831", + "right": "issue:45310", + "accept": false, + "reason": "FineGrainedFP8 accuracy issue is not the same as Qwen3.5 Moe from_pretrained error." + }, + { + "left": "issue:43479", + "right": "issue:44913", + "accept": false, + "reason": "Phi4MultimodalConfig None-handling and GPTNeoXConfig rotary_pct reload behavior are unrelated configuration bugs." + }, + { + "left": "issue:44263", + "right": "issue:44373", + "accept": false, + "reason": "Indexer output handling vs a wrong docstring are not the same issue." + }, + { + "left": "issue:42831", + "right": "issue:44038", + "accept": false, + "reason": "FineGrainedFP8 accuracy and Qwen3-VL-Moe bug are different model-specific problems." + }, + { + "left": "issue:39401", + "right": "issue:44568", + "accept": false, + "reason": "Qwen3 wrong offset_mapping and mdeberta-v3 add_special_tokens BOS/EOS regression are distinct tokenizer bugs." + }, + { + "left": "issue:42757", + "right": "issue:45278", + "accept": false, + "reason": "Both mention imports after version changes, but one is a specific missing huggingface_hub symbol and the other is a broad import-error report." + }, + { + "left": "issue:42886", + "right": "issue:42994", + "accept": false, + "reason": "Offline cache loading failure is a different problem from quantized model saving failure." + }, + { + "left": "issue:43232", + "right": "issue:45593", + "accept": false, + "reason": "Generation kwargs handling and D-FINE auxiliary-loss behavior are unrelated." + }, + { + "left": "issue:42831", + "right": "issue:45753", + "accept": false, + "reason": "FineGrainedFP8 accuracy issue and Qwen3_5 infinite loop on a specific image are unrelated." + }, + { + "left": "issue:43756", + "right": "issue:44060", + "accept": false, + "reason": "Smollm3 RoPE layer mismatch is not the same as Qwen3-Next tied-weights warning." + }, + { + "left": "issue:42886", + "right": "issue:43066", + "accept": false, + "reason": "Tokenizer cache/offline loading and wrong tokenizer decoder type are different bugs." + }, + { + "left": "issue:22355", + "right": "issue:28218", + "accept": false, + "reason": "Missing transformers.onnx import and extra-space after added token are unrelated tokenizer/module 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": 18, + "review_unit_index": 40, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:28218", + "issue:30990", + "issue:33357", + "issue:36296", + "issue:39401", + "issue:41628", + "issue:42371", + "issue:42907", + "issue:43116", + "issue:43122", + "issue:43232", + "issue:43408", + "issue:43452", + "issue:43475", + "issue:43576", + "issue:43646", + "issue:43784", + "issue:43824", + "issue:43937", + "issue:43976", + "issue:44038", + "issue:44117", + "issue:44222", + "issue:44261", + "issue:44263", + "issue:44295", + "issue:44297", + "issue:44351", + "issue:44360", + "issue:44373", + "issue:44485", + "issue:44488", + "issue:44514", + "issue:44521", + "issue:44625", + "issue:44671", + "issue:44704", + "issue:44743", + "issue:44779", + "issue:44829", + "issue:44945", + "issue:44964", + "issue:45200", + "issue:45278", + "issue:45290", + "issue:45600", + "issue:46002", + "issue:46132" + ], + "soft_pairs": [ + "issue:44671|issue:46132", + "issue:43576|issue:43784", + "issue:43122|issue:44521", + "issue:43646|issue:46002", + "issue:44117|issue:44704", + "issue:30990|issue:33357", + "issue:42371|issue:44263", + "issue:44295|issue:44297", + "issue:43232|issue:44743", + "issue:41628|issue:45278", + "issue:44704|issue:45290", + "issue:42907|issue:44945", + "issue:43122|issue:44514", + "issue:43784|issue:45600", + "issue:39401|issue:44488", + "issue:44261|issue:44373", + "issue:43976|issue:44945", + "issue:43408|issue:43475", + "issue:42907|issue:43937", + "issue:44360|issue:44485", + "issue:39401|issue:44521", + "issue:44625|issue:44829", + "issue:43116|issue:45200", + "issue:42907|issue:44964", + "issue:44038|issue:44297", + "issue:28218|issue:30990", + "issue:44038|issue:44295", + "issue:43122|issue:44779", + "issue:42907|issue:43452", + "issue:43824|issue:44351", + "issue:36296|issue:43116", + "issue:44038|issue:44222" + ], + "prepared_review_unit_hash": "239cec6d47de06d6b3bb5aaeb8d04e548c3a9cdec0a7288ea02505445f35b4f2", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 36790, + "estimated_input_tokens": 9198, + "estimated_eval_tokens": 18652 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: most items are unrelated issue reports spanning tokenization, model loading, generation, import errors, and model-specific regressions. I would not merge any of the soft pairs, and there are no PRs to consider.", + "confidence": 0.86, + "canonical_issue_reason": "issue:43122 is the broadest tokenizer-regression report here and the closest thing to a representative topic anchor, though the cluster is too mixed to treat as a true duplicate set.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43122 is the most reusable issue to keep as a representative for the tokenizer-regression subset; it has the widest apparent topical overlap among the listed issues.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44671", + "right": "issue:46132", + "accept": false, + "reason": "Different bugs: masked-LM prediction quality vs AttentionInterface.register changing callback behavior." + }, + { + "left": "issue:43576", + "right": "issue:43784", + "accept": false, + "reason": "Both are import/loading failures, but one is the v5 env command and the other is a sentence-transformers import NameError." + }, + { + "left": "issue:43122", + "right": "issue:44521", + "accept": false, + "reason": "Both involve tokenization/chat handling, but one is a versioned tokenizer regression and the other is multimodal assistant mask generation." + }, + { + "left": "issue:43646", + "right": "issue:46002", + "accept": false, + "reason": "Custom model initialization failure is not the same concrete bug as NaNs in classification heads after checkout/init." + }, + { + "left": "issue:44117", + "right": "issue:44704", + "accept": false, + "reason": "TOKENIZER_MAPPING_NAMES handling and AutoProcessor kwargs forwarding are unrelated code paths." + }, + { + "left": "issue:30990", + "right": "issue:33357", + "accept": false, + "reason": "Sentence-transformers loading stall and a MacOS bus error on a CLIP model are different failures." + }, + { + "left": "issue:42371", + "right": "issue:44263", + "accept": false, + "reason": "TF32 settings guidance and a torch.split return-value bug in GlmMoeDsaIndexer are unrelated." + }, + { + "left": "issue:44295", + "right": "issue:44297", + "accept": false, + "reason": "Reading position_ids after buffer registration is unrelated to tokenizer_class mismatch on save_pretrained." + }, + { + "left": "issue:43232", + "right": "issue:44743", + "accept": false, + "reason": "Both touch generation/cache logic, but they describe different failures in different code paths." + }, + { + "left": "issue:41628", + "right": "issue:45278", + "accept": false, + "reason": "A specific missing AutoImageProcessor import is not the same as broad import errors after upgrade." + }, + { + "left": "issue:44704", + "right": "issue:45290", + "accept": false, + "reason": "cached_file kwarg forwarding and apply_chat_template crashes are distinct processor issues." + }, + { + "left": "issue:42907", + "right": "issue:44945", + "accept": false, + "reason": "Saving dequantized models and incorrect output under pipeline parallelism are unrelated." + }, + { + "left": "issue:43122", + "right": "issue:44514", + "accept": false, + "reason": "Both mention tokenizer/template behavior, but one is a tokenizer regression across versions and the other is a batched apply_chat_template crash." + }, + { + "left": "issue:43784", + "right": "issue:45600", + "accept": false, + "reason": "Importing sentence-transformers with latest transformers and broken auto_mappings CI are different import-related regressions." + }, + { + "left": "issue:39401", + "right": "issue:44488", + "accept": false, + "reason": "Wrong offset_mapping in Qwen3 tokenizer is unrelated to failing to load a specific model." + }, + { + "left": "issue:44261", + "right": "issue:44373", + "accept": false, + "reason": "A precision/config issue in MLA q_a_layernorm is not the same as a wrong docstring for position_ids." + }, + { + "left": "issue:43976", + "right": "issue:44945", + "accept": false, + "reason": "Python version compatibility breakage and pipeline-parallel output corruption are unrelated." + }, + { + "left": "issue:43408", + "right": "issue:43475", + "accept": false, + "reason": "Both are SAM3-related, but one is a model-type warning and the other is a missing output attribute." + }, + { + "left": "issue:42907", + "right": "issue:43937", + "accept": false, + "reason": "Dequantized save failures and invalid GenerationConfig are different bugs." + }, + { + "left": "issue:44360", + "right": "issue:44485", + "accept": false, + "reason": "A missing ReLU in the DSA indexer and GLM-5 RoPE implementation are not the same issue." + }, + { + "left": "issue:39401", + "right": "issue:44521", + "accept": false, + "reason": "Wrong offset_mapping and all-zero assistant masks are different tokenizer/chat-template failures." + }, + { + "left": "issue:44625", + "right": "issue:44829", + "accept": false, + "reason": "num_labels propagation and flash_attention_3 degenerate training are unrelated." + }, + { + "left": "issue:43116", + "right": "issue:45200", + "accept": false, + "reason": "Empty multi-label classification results and Gemma 4 mm_token_type_ids defaults are different problems." + }, + { + "left": "issue:42907", + "right": "issue:44964", + "accept": false, + "reason": "Dequantized save_pretrained failure is unrelated to loading Phi-4 multimodal." + }, + { + "left": "issue:44038", + "right": "issue:44297", + "accept": false, + "reason": "Qwen3-VL-Moe bug and tokenizer_class mismatch on save are unrelated." + }, + { + "left": "issue:28218", + "right": "issue:30990", + "accept": false, + "reason": "Tokenizer added-space behavior and Sentence Transformers loading hang are unrelated." + }, + { + "left": "issue:44038", + "right": "issue:44295", + "accept": false, + "reason": "Qwen3-VL-Moe behavior and position_ids buffer access are unrelated." + }, + { + "left": "issue:43122", + "right": "issue:44779", + "accept": false, + "reason": "Both are tokenizer regressions in v5, but they involve different models and concrete failure modes." + }, + { + "left": "issue:43824", + "right": "issue:44351", + "accept": false, + "reason": "Missing Qwen2_5_VLForConditionalGeneration and missing HybridCache are separate import errors." + }, + { + "left": "issue:36296", + "right": "issue:43116", + "accept": false, + "reason": "Tensor parallel training instability and empty classification outputs are unrelated." + }, + { + "left": "issue:44038", + "right": "issue:44222", + "accept": false, + "reason": "Qwen3-VL-Moe regressions and FP8 save_pretrained moe are different bugs, despite both mentioning moe." + } + ] + }, + "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": 18, + "review_unit_index": 41, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:22355", + "issue:29127", + "issue:29159", + "issue:33453", + "issue:38617", + "issue:39401", + "issue:42222", + "issue:42371", + "issue:42673", + "issue:42757", + "issue:43064", + "issue:43066", + "issue:43116", + "issue:43122", + "issue:43278", + "issue:43441", + "issue:43493", + "issue:43611", + "issue:43673", + "issue:43784", + "issue:43825", + "issue:43827", + "issue:43874", + "issue:43937", + "issue:43976", + "issue:43992", + "issue:44038", + "issue:44075", + "issue:44351", + "issue:44380", + "issue:44485", + "issue:44704", + "issue:44779", + "issue:44936", + "issue:44945", + "issue:44977", + "issue:45072", + "issue:45083", + "issue:45216", + "issue:45278", + "issue:45357", + "issue:45446", + "issue:45478", + "issue:45563", + "issue:45753", + "issue:46002", + "issue:46077" + ], + "soft_pairs": [ + "issue:45216|issue:45478", + "issue:42757|issue:44351", + "issue:43278|issue:44945", + "issue:43784|issue:45446", + "issue:22355|issue:29159", + "issue:43122|issue:44977", + "issue:43493|issue:45753", + "issue:38617|issue:45278", + "issue:42673|issue:45072", + "issue:43066|issue:44779", + "issue:22355|issue:29127", + "issue:43992|issue:44704", + "issue:43937|issue:44485", + "issue:43611|issue:46002", + "issue:43976|issue:45446", + "issue:42222|issue:43441", + "issue:43116|issue:43493", + "issue:42371|issue:43874", + "issue:44075|issue:44936", + "issue:43064|issue:43673", + "issue:44038|issue:45478", + "issue:45357|issue:45478", + "issue:43116|issue:43825", + "issue:45083|issue:45446", + "issue:43937|issue:45563", + "issue:44380|issue:46077", + "issue:43827|issue:45446", + "issue:33453|issue:39401" + ], + "prepared_review_unit_hash": "66236ff23197e74d8ed3e9c9f33582c64f95fe410e637dd786767c6e98c5fbc5", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 35098, + "estimated_input_tokens": 8775, + "estimated_eval_tokens": 17806 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is not a coherent duplicate cluster: the items span many unrelated bugs across tokenization, imports, training, generation, multimodal models, and docs. Under strict triage, every soft pair should be rejected.", + "confidence": 0.86, + "canonical_issue_reason": "No single canonical issue stands out because the reports are heterogeneous and do not share one underlying defect.", + "canonical_pr_reason": null, + "best_issue_reason": "If one representative issue must be chosen, issue 43122 is the cleanest broad regression-style tokenizer report; however, it is only a weak global representative because the set is not a true duplicate cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45216", + "right": "issue:45478", + "accept": false, + "reason": "Both concern Qwen3.5 save/load failures, but one is a bad checkpoint saved by save_pretrained and the other is a from_pretrained error; different concrete code paths." + }, + { + "left": "issue:42757", + "right": "issue:44351", + "accept": false, + "reason": "Both are import errors, but they involve different missing symbols from different packages/modules and are not the same bug." + }, + { + "left": "issue:43278", + "right": "issue:44945", + "accept": false, + "reason": "One is a dtype mismatch between train/evaluate; the other is incorrect output under pipeline parallelism. Different failure modes and subsystems." + }, + { + "left": "issue:43784", + "right": "issue:45446", + "accept": false, + "reason": "Unrelated import/version issues: sentence-transformers import failure versus a PyTorch version check in flex_attention." + }, + { + "left": "issue:22355", + "right": "issue:29159", + "accept": false, + "reason": "Missing transformers.onnx module import is unrelated to slow/fast tokenizer behavioral inconsistency." + }, + { + "left": "issue:43122", + "right": "issue:44977", + "accept": false, + "reason": "Tokenizer behavior regression across versions is unrelated to flash-attention generation failure on Qwen3.5." + }, + { + "left": "issue:43493", + "right": "issue:45753", + "accept": false, + "reason": "Different models and different symptoms: SigLIP2 implementation discrepancy versus a Qwen3.5 infinite loop on a specific image." + }, + { + "left": "issue:38617", + "right": "issue:45278", + "accept": false, + "reason": "One is a specific missing import symbol; the other is broad import breakage after an upgrade. Not the same underlying defect." + }, + { + "left": "issue:42673", + "right": "issue:45072", + "accept": false, + "reason": "VRAM leak in dataloader threads is unrelated to bfloat16 dtype mismatches in inference." + }, + { + "left": "issue:43066", + "right": "issue:44779", + "accept": false, + "reason": "Wrong tokenizer decoder type and Deepseek tokenization regression are different tokenizer bugs with no shared concrete code-path." + }, + { + "left": "issue:22355", + "right": "issue:29127", + "accept": false, + "reason": "Module import failure and a LayoutLMv3 error-message clarity issue are unrelated." + }, + { + "left": "issue:43992", + "right": "issue:44704", + "accept": false, + "reason": "Missing embed_tokens weights during from_pretrained is unrelated to AutoProcessor not forwarding kwargs to cached_file." + }, + { + "left": "issue:43937", + "right": "issue:44485", + "accept": false, + "reason": "GenerationConfig validation failure and GLM-5 RoPE discussion are different model-generation issues." + }, + { + "left": "issue:43611", + "right": "issue:46002", + "accept": false, + "reason": "Base_model_prefix loading regression and NaNs after checkout/init are unrelated symptoms and code paths." + }, + { + "left": "issue:43976", + "right": "issue:45446", + "accept": false, + "reason": "Python version support mismatch is not the same bug as the flex_attention import/version check." + }, + { + "left": "issue:42222", + "right": "issue:43441", + "accept": false, + "reason": "VitPose model breakage and Ministral flash-attention failure are different model-specific issues." + }, + { + "left": "issue:43116", + "right": "issue:43493", + "accept": false, + "reason": "Empty multi-label classification results in an example script is unrelated to a SigLIP2 implementation discrepancy." + }, + { + "left": "issue:42371", + "right": "issue:43874", + "accept": false, + "reason": "TF32 behavior control messaging/API and a missing image-processor method are unrelated." + }, + { + "left": "issue:44075", + "right": "issue:44936", + "accept": false, + "reason": "Ignored SGD args and trainer.evaluate() failing after train() are different trainer/optimizer problems." + }, + { + "left": "issue:43064", + "right": "issue:43673", + "accept": false, + "reason": "Wrong optimizer state placement under FSDP2/PEFT is unrelated to missing GenerationMixin cache in chunked prefill." + }, + { + "left": "issue:44038", + "right": "issue:45478", + "accept": false, + "reason": "These are different Qwen3.5/Qwen3-VL-Moe issues with different failing operations and symptoms." + }, + { + "left": "issue:45357", + "right": "issue:45478", + "accept": false, + "reason": "Incorrect visual encoder keys saved by save_pretrained is not the same as a from_pretrained loading error." + }, + { + "left": "issue:43116", + "right": "issue:43825", + "accept": false, + "reason": "Example script multi-label classification failure and pipeline translation-task messaging are unrelated." + }, + { + "left": "issue:45083", + "right": "issue:45446", + "accept": false, + "reason": "Helper-function length behavior in qwen3_omni_moe is unrelated to a PyTorch version check bug." + }, + { + "left": "issue:43937", + "right": "issue:45563", + "accept": false, + "reason": "Invalid GenerationConfig and a stale warning about num_return_sequences are not the same issue." + }, + { + "left": "issue:44380", + "right": "issue:46077", + "accept": false, + "reason": "GPT2 attention scaling ignored in backends versus Gemma4 causal-mask construction are different code paths and models." + }, + { + "left": "issue:43827", + "right": "issue:45446", + "accept": false, + "reason": "Docs still referencing pipeline() and a version check for AuxRequest are unrelated." + }, + { + "left": "issue:33453", + "right": "issue:39401", + "accept": false, + "reason": "Tokenizer loading regression and Qwen3 offset_mapping errors are distinct tokenizer bugs, not one underlying defect." + } + ] + }, + "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": 18, + "review_unit_index": 42, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:38617", + "issue:39401", + "issue:41628", + "issue:42175", + "issue:42222", + "issue:42371", + "issue:42490", + "issue:42617", + "issue:42673", + "issue:42831", + "issue:43012", + "issue:43066", + "issue:43122", + "issue:43408", + "issue:43421", + "issue:43450", + "issue:43577", + "issue:43582", + "issue:43742", + "issue:43906", + "issue:43937", + "issue:43986", + "issue:44117", + "issue:44220", + "issue:44261", + "issue:44291", + "issue:44410", + "issue:44466", + "issue:44483", + "issue:44484", + "issue:44496", + "issue:44779", + "issue:44829", + "issue:44933", + "issue:44977", + "issue:45071", + "issue:45081", + "issue:45216", + "issue:45278", + "issue:45310", + "issue:45341", + "issue:45405", + "issue:45446", + "issue:45464", + "issue:45542", + "issue:45561", + "issue:45588", + "issue:46002" + ], + "soft_pairs": [ + "issue:44117|issue:45588", + "issue:39401|issue:44779", + "issue:42673|issue:43906", + "issue:43937|issue:44977", + "issue:44466|issue:46002", + "issue:44291|issue:45071", + "issue:43450|issue:43986", + "issue:45405|issue:45446", + "issue:45341|issue:45561", + "issue:38617|issue:41628", + "issue:44117|issue:45081", + "issue:44410|issue:45310", + "issue:44829|issue:45216", + "issue:39401|issue:43421", + "issue:42617|issue:45341", + "issue:42673|issue:43582", + "issue:44484|issue:45310", + "issue:42371|issue:44261", + "issue:42490|issue:42673", + "issue:44933|issue:45278", + "issue:43408|issue:43742", + "issue:38617|issue:44220", + "issue:42222|issue:45542", + "issue:43937|issue:44496", + "issue:44483|issue:45464", + "issue:42175|issue:42222", + "issue:43066|issue:43122", + "issue:42371|issue:42831", + "issue:43012|issue:43577" + ], + "prepared_review_unit_hash": "d51b4fa44b61f8e535b8745e83c365c516754e4f997592b0864346879b9864d8", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36289, + "estimated_input_tokens": 9073, + "estimated_eval_tokens": 18402 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a set of unrelated Transformers regressions that only share broad surface areas like tokenizers, imports, or model loading. None of the soft pairs look like safe duplicates, so I would not merge any of them. The most representative issue is the broad tokenizer-regression report in 43122.", + "confidence": 0.34, + "canonical_issue_reason": "issue:43122 is the broadest, most central regression report here: same tokenizer behaving differently across 4.57.3 to 5.0, which best captures the tokenizer-related subset of the cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43122 is the best standalone anchor for this cluster because it is the most general user-facing regression and has the clearest version-bump framing.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44117", + "right": "issue:45588", + "accept": false, + "reason": "Import-time tokenizer mapping fallback failure vs flash-attention crash on sink-less models; different code paths and symptoms." + }, + { + "left": "issue:39401", + "right": "issue:44779", + "accept": false, + "reason": "Both are tokenizer bugs, but one is offset mapping and the other is model-specific output regression; not the same underlying defect." + }, + { + "left": "issue:42673", + "right": "issue:43906", + "accept": false, + "reason": "VRAM leak in multi-threaded dataloader use vs an isolated reproduction of another issue; no clear shared bug signature." + }, + { + "left": "issue:43937", + "right": "issue:44977", + "accept": false, + "reason": "GenerationConfig validation failure for GLM-5 vs flash-attention generation failure for Qwen3.5; different failure modes." + }, + { + "left": "issue:44466", + "right": "issue:46002", + "accept": false, + "reason": "Serialization/tied-weights inconsistency vs NaNs after checkout+init; these are distinct model-state problems." + }, + { + "left": "issue:44291", + "right": "issue:45071", + "accept": false, + "reason": "Unexpected _is_hf_initialized argument during init_empty_weights loading vs PretrainedConfig type-checking breakage; related version area but not the same bug." + }, + { + "left": "issue:43450", + "right": "issue:43986", + "accept": false, + "reason": "Incorrect batched video-processor shape vs missing-torchvision crash in AutoProcessor; different layers of the video stack." + }, + { + "left": "issue:45405", + "right": "issue:45446", + "accept": false, + "reason": "Dependency version bump/release issue vs a PyTorch version guard bug in flex_attention; unrelated." + }, + { + "left": "issue:45341", + "right": "issue:45561", + "accept": false, + "reason": "General testing_utils bug vs xdist race on captured_info.txt; the latter is a specific concurrency issue, not the same defect." + }, + { + "left": "issue:38617", + "right": "issue:41628", + "accept": false, + "reason": "Different missing imports from different modules; both are import errors, but not the same missing symbol or code path." + }, + { + "left": "issue:44117", + "right": "issue:45081", + "accept": false, + "reason": "TOKENIZER_MAPPING_NAMES returning None vs Mistral regex patch crashing on backend_tokenizer absence; distinct tokenizer internals." + }, + { + "left": "issue:44410", + "right": "issue:45310", + "accept": false, + "reason": "Qwen3Next missing projections in layer 0 vs Qwen3.5 MoE from_pretrained failure; different model families and bugs." + }, + { + "left": "issue:44829", + "right": "issue:45216", + "accept": false, + "reason": "Flash-attention-3 training degeneration vs save_pretrained checkpoint regression; different subsystems and symptoms." + }, + { + "left": "issue:39401", + "right": "issue:43421", + "accept": false, + "reason": "Offset mapping bug vs post-processor not updating after runtime special-token changes; both tokenizer-related but not the same issue." + }, + { + "left": "issue:42617", + "right": "issue:45341", + "accept": false, + "reason": "3d_parallel.py runtime failure vs a small testing_utils bug; no overlap in underlying code path." + }, + { + "left": "issue:42673", + "right": "issue:43582", + "accept": false, + "reason": "Multi-thread VRAM leak vs Apple Silicon TypeError in caching_allocator_warmup; unrelated runtime failures." + }, + { + "left": "issue:44484", + "right": "issue:45310", + "accept": false, + "reason": "chat/completions API request handling failure vs Qwen3.5 MoE loading error; different APIs and failure points." + }, + { + "left": "issue:42371", + "right": "issue:44261", + "accept": false, + "reason": "TF32 control API warning vs missing rms_norm_eps causing precision drift; not the same bug." + }, + { + "left": "issue:42490", + "right": "issue:42673", + "accept": false, + "reason": "Local-path trailing-slash import warning vs VRAM leak in multi-threaded loading; unrelated issues." + }, + { + "left": "issue:44933", + "right": "issue:45278", + "accept": false, + "reason": "Specific missing import from image_utils vs broad import breakage after upgrade; the latter is a wider symptom, not the same concrete defect." + }, + { + "left": "issue:43408", + "right": "issue:43742", + "accept": false, + "reason": "Model-type mismatch warning for sam3 variants vs key error loading MobileLLM; different loaders and errors." + }, + { + "left": "issue:38617", + "right": "issue:44220", + "accept": false, + "reason": "Configuration import error vs audio feature extraction issue; completely different subsystems." + }, + { + "left": "issue:42222", + "right": "issue:45542", + "accept": false, + "reason": "ViTPose model breakage vs TensorFlow backend install ambiguity; only broad backend context overlaps." + }, + { + "left": "issue:43937", + "right": "issue:44496", + "accept": false, + "reason": "GLM-5 generation config invalid vs unrecognized model type in config.json; both load-time validation issues but with different roots." + }, + { + "left": "issue:44483", + "right": "issue:45464", + "accept": false, + "reason": "General chat/completions request rejection vs streaming inference failure on Qwen3.5-0.8B; not enough evidence for one bug." + }, + { + "left": "issue:42175", + "right": "issue:42222", + "accept": false, + "reason": "Missing TensorFlow in torch-only install vs vitpose model breakage; different packaging/model problems." + }, + { + "left": "issue:43066", + "right": "issue:43122", + "accept": false, + "reason": "Wrong tokenizer decoder type vs changed tokenization output across versions; closely related theme, but not the same concrete defect." + }, + { + "left": "issue:42371", + "right": "issue:42831", + "accept": false, + "reason": "TF32 API settings warning vs FineGrainedFP8 accuracy regression; unrelated numeric-precision issues." + }, + { + "left": "issue:43012", + "right": "issue:43577", + "accept": false, + "reason": "Compilation warning for bf16 models vs BLIP2 dtype fields staying float32 on load; different dtype-handling 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": 18, + "review_unit_index": 43, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:38617", + "issue:41628", + "issue:42222", + "issue:42757", + "issue:43012", + "issue:43064", + "issue:43097", + "issue:43335", + "issue:43408", + "issue:43441", + "issue:43452", + "issue:43502", + "issue:43531", + "issue:43673", + "issue:43873", + "issue:43874", + "issue:43931", + "issue:43986", + "issue:43992", + "issue:44038", + "issue:44062", + "issue:44075", + "issue:44117", + "issue:44261", + "issue:44263", + "issue:44315", + "issue:44351", + "issue:44368", + "issue:44410", + "issue:44484", + "issue:44485", + "issue:44493", + "issue:44704", + "issue:44829", + "issue:44987", + "issue:45003", + "issue:45071", + "issue:45103", + "issue:45341", + "issue:45372", + "issue:45375", + "issue:45459", + "issue:45478", + "issue:45542", + "issue:45588", + "issue:45656", + "issue:46050", + "issue:46123" + ], + "soft_pairs": [ + "issue:44075|issue:45656", + "issue:43673|issue:45341", + "issue:44410|issue:45478", + "issue:44484|issue:45478", + "issue:43097|issue:44493", + "issue:41628|issue:45071", + "issue:43874|issue:44261", + "issue:44263|issue:44485", + "issue:45071|issue:45542", + "issue:42222|issue:45459", + "issue:43012|issue:43408", + "issue:43441|issue:44829", + "issue:43931|issue:44410", + "issue:43673|issue:46123", + "issue:43931|issue:44038", + "issue:43097|issue:43452", + "issue:43335|issue:45071", + "issue:45003|issue:45341", + "issue:43986|issue:45341", + "issue:44117|issue:44987", + "issue:43986|issue:44484", + "issue:43408|issue:44368", + "issue:43874|issue:44263", + "issue:43531|issue:45341", + "issue:43986|issue:45372", + "issue:38617|issue:42757", + "issue:44261|issue:44485", + "issue:43064|issue:44315", + "issue:43097|issue:46050", + "issue:43097|issue:45588", + "issue:44062|issue:44117", + "issue:38617|issue:44351", + "issue:44410|issue:45375", + "issue:44038|issue:44410", + "issue:43502|issue:43992", + "issue:43097|issue:43873", + "issue:44704|issue:45103", + "issue:43986|issue:44117" + ], + "prepared_review_unit_hash": "4d5047fe6b8deb1fe7a596022e02081bb061a56f13cf2845168f96353098a8d0", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 38, + "serialized_chars": 39066, + "estimated_input_tokens": 9767, + "estimated_eval_tokens": 19790 + }, + "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; several pairs share only superficial themes like imports, loading, or v5 regressions, but none look like the same underlying defect closely enough to merge.", + "confidence": 0.68, + "canonical_issue_reason": "issue:43097 is the strongest cluster anchor: it\u2019s a broadly impactful v5 regression report, has inbound references, and sits near several weak similarity links, even though the neighboring issues are not true duplicates.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45372 is the best standalone issue to keep: it is open, concrete, and currently actionable, with a clear processor-loading breakage and active discussion.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44075", + "right": "issue:45656", + "accept": false, + "reason": "Different optimizer bugs: unused SGD args vs optimizer stepping twice under deepspeed." + }, + { + "left": "issue:43673", + "right": "issue:45341", + "accept": false, + "reason": "Unrelated failures: generation cache/chunked prefill vs a testing utility bug." + }, + { + "left": "issue:44410", + "right": "issue:45478", + "accept": false, + "reason": "Different Qwen model-path issues: missing projections in qwen3next vs a from_pretrained load error for Qwen3.5 MoE." + }, + { + "left": "issue:44484", + "right": "issue:45478", + "accept": false, + "reason": "No shared underlying bug; one is a save_pretrained shard-size question, the other is a model loading failure." + }, + { + "left": "issue:43097", + "right": "issue:44493", + "accept": false, + "reason": "Both are v5-related regressions, but one is a removed config flag and the other is unexpected position-id keys." + }, + { + "left": "issue:41628", + "right": "issue:45071", + "accept": false, + "reason": "Both involve import/type issues, but they affect different APIs and code paths." + }, + { + "left": "issue:43874", + "right": "issue:44261", + "accept": false, + "reason": "Different model components: image processor missing a method vs MLA q_a_layernorm precision/config handling." + }, + { + "left": "issue:44263", + "right": "issue:44485", + "accept": false, + "reason": "Both are GLM-family issues, but one is a torch.split return handling bug and the other is a RoPE implementation concern." + }, + { + "left": "issue:45071", + "right": "issue:45542", + "accept": false, + "reason": "Unrelated: PreTrainedConfig type-checking regression vs TensorFlow/tensorboard backend detection." + }, + { + "left": "issue:42222", + "right": "issue:45459", + "accept": false, + "reason": "Different failures: vitpose model breakage vs tokenizer protobuf error masking." + }, + { + "left": "issue:43012", + "right": "issue:43408", + "accept": false, + "reason": "Both are warnings in different model paths; they do not describe the same bug." + }, + { + "left": "issue:43441", + "right": "issue:44829", + "accept": false, + "reason": "Both mention FlashAttention, but one is a load-time failure and the other is degenerate training behavior." + }, + { + "left": "issue:43931", + "right": "issue:44410", + "accept": false, + "reason": "Distinct model-loading issues for different model families and different failure modes." + }, + { + "left": "issue:43673", + "right": "issue:46123", + "accept": false, + "reason": "Different pipeline behaviors: generation cache during chunked prefill vs final partial batch handling in a mask pipeline." + }, + { + "left": "issue:43931", + "right": "issue:44038", + "accept": false, + "reason": "Both are Qwen model issues, but the concrete failures differ and are not the same code-path bug." + }, + { + "left": "issue:43097", + "right": "issue:43452", + "accept": false, + "reason": "Different regressions: removed tie_embeddings flag vs gguf_file loading in tokenizer/model loaders." + }, + { + "left": "issue:43335", + "right": "issue:45071", + "accept": false, + "reason": "Unrelated: SwitchTransformers sparse-layer config bug vs PretrainedConfig type-checking regression." + }, + { + "left": "issue:45003", + "right": "issue:45341", + "accept": false, + "reason": "Different subsystems: unsafe sys.modules access vs a testing utility bug." + }, + { + "left": "issue:43986", + "right": "issue:45341", + "accept": false, + "reason": "Completely different areas: AutoProcessor video-model crash vs testing utilities." + }, + { + "left": "issue:44117", + "right": "issue:44987", + "accept": false, + "reason": "Both are from_pretrained-related, but they target different assumptions and different failures." + }, + { + "left": "issue:43986", + "right": "issue:44484", + "accept": false, + "reason": "No common underlying bug: video processor dependency handling vs save_pretrained shard sizing." + }, + { + "left": "issue:43408", + "right": "issue:44368", + "accept": false, + "reason": "Both are warnings, but they concern different model/config mismatches." + }, + { + "left": "issue:43874", + "right": "issue:44263", + "accept": false, + "reason": "Different bugs in different GLM components; not the same defect." + }, + { + "left": "issue:43531", + "right": "issue:45341", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior is unrelated to testing_utils." + }, + { + "left": "issue:43986", + "right": "issue:45372", + "accept": false, + "reason": "Both involve loading processors, but the missing torchvision path and missing ReasoningEffort import are different causes." + }, + { + "left": "issue:38617", + "right": "issue:42757", + "accept": false, + "reason": "Both are ImportError reports, but they concern different symbols from different dependencies and different breakages." + }, + { + "left": "issue:44261", + "right": "issue:44485", + "accept": false, + "reason": "Different GLM-related correctness issues; no shared code-path evidence." + }, + { + "left": "issue:43064", + "right": "issue:44315", + "accept": false, + "reason": "Training/optimizer-state bug vs Liger kernel application under model_init; not the same problem." + }, + { + "left": "issue:43097", + "right": "issue:46050", + "accept": false, + "reason": "Different serialization regressions: config flag removal vs quantized model save/load breakage." + }, + { + "left": "issue:43097", + "right": "issue:45588", + "accept": false, + "reason": "Unrelated: tie_embeddings config removal vs flash_attention AttributeError on sink-less models." + }, + { + "left": "issue:44062", + "right": "issue:44117", + "accept": false, + "reason": "Tokenizer construction bug vs tokenizer mapping lookup assumption; related area but different defects." + }, + { + "left": "issue:38617", + "right": "issue:44351", + "accept": false, + "reason": "Both are import errors, but they are for different missing names and different modules." + }, + { + "left": "issue:44410", + "right": "issue:45375", + "accept": false, + "reason": "Different vision/config issues: missing attention projections vs a dropped config field under strict handling." + }, + { + "left": "issue:44038", + "right": "issue:44410", + "accept": false, + "reason": "Both are Qwen variants, but the concrete failures are different model-internal bugs." + }, + { + "left": "issue:43502", + "right": "issue:43992", + "accept": false, + "reason": "Different from_pretrained failures: local_files_only network leakage vs missing embed_tokens weight." + }, + { + "left": "issue:43097", + "right": "issue:43873", + "accept": false, + "reason": "Different problem classes: config removal regression vs quantization/offloading behavior." + }, + { + "left": "issue:44704", + "right": "issue:45103", + "accept": false, + "reason": "Unrelated kwargs-passing bug vs auto_docstring annotation crash." + }, + { + "left": "issue:43986", + "right": "issue:44117", + "accept": false, + "reason": "Different loading failures in different subsystems; not a duplicate." + } + ] + }, + "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": 18, + "review_unit_index": 44, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:36683", + "issue:38617", + "issue:42490", + "issue:42757", + "issue:42907", + "issue:43012", + "issue:43064", + "issue:43066", + "issue:43208", + "issue:43299", + "issue:43352", + "issue:43475", + "issue:43502", + "issue:43526", + "issue:43550", + "issue:43582", + "issue:43618", + "issue:43756", + "issue:43761", + "issue:43792", + "issue:43824", + "issue:43825", + "issue:43867", + "issue:43901", + "issue:43906", + "issue:43976", + "issue:43986", + "issue:43992", + "issue:44038", + "issue:44222", + "issue:44230", + "issue:44368", + "issue:44380", + "issue:44483", + "issue:44484", + "issue:44485", + "issue:44514", + "issue:44655", + "issue:44683", + "issue:45276", + "issue:45290", + "issue:45381", + "issue:45478", + "issue:45479", + "issue:45753", + "issue:46077", + "issue:46123", + "issue:46132" + ], + "soft_pairs": [ + "issue:43824|issue:43986", + "issue:43066|issue:43906", + "issue:43208|issue:43550", + "issue:43825|issue:44655", + "issue:43992|issue:44484", + "issue:43012|issue:43550", + "issue:43012|issue:44368", + "issue:43352|issue:44380", + "issue:42907|issue:45753", + "issue:42907|issue:44655", + "issue:43792|issue:46123", + "issue:43352|issue:43976", + "issue:44380|issue:45479", + "issue:44380|issue:46132", + "issue:43064|issue:44380", + "issue:44483|issue:44514", + "issue:44222|issue:45478", + "issue:36683|issue:43582", + "issue:43208|issue:43526", + "issue:43012|issue:44683", + "issue:42757|issue:43824", + "issue:42490|issue:43012", + "issue:43618|issue:43761", + "issue:43618|issue:44380", + "issue:44038|issue:44230", + "issue:38617|issue:43502", + "issue:44483|issue:45290", + "issue:44483|issue:45381", + "issue:43756|issue:44485", + "issue:43299|issue:43867", + "issue:36683|issue:43475", + "issue:43299|issue:45753", + "issue:43901|issue:44655", + "issue:45276|issue:46077" + ], + "prepared_review_unit_hash": "a15949a66d907da2f680c917de67ae8fd47a64a2da2cd41b422246d4dcedd72f", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 34, + "serialized_chars": 37720, + "estimated_input_tokens": 9430, + "estimated_eval_tokens": 19116 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No true duplicates stand out in this noisy cluster. The soft-similarity pairs mostly share a subsystem or model family, but they describe different failures and should stay separate. No PRs are present; I\u2019d use issue #43012 as the canonical representative only because it is a clear, broadly phrased regression report.", + "confidence": 0.33, + "canonical_issue_reason": "#43012 is the cleanest, most generally framed bug report in the set and is a better representative than the many model-specific or UX-only reports.", + "canonical_pr_reason": null, + "best_issue_reason": "#43012 is a concrete, actionable runtime regression with a clear symptom and wider applicability than the other items.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43824", + "right": "issue:43986", + "accept": false, + "reason": "Both are import/load failures, but one is a missing model class export and the other is a torchvision dependency crash; different root causes." + }, + { + "left": "issue:43066", + "right": "issue:43906", + "accept": false, + "reason": "These concern different v5 issues: tokenizer decoder typing vs an isolated reproduction of another bug." + }, + { + "left": "issue:43208", + "right": "issue:43550", + "accept": false, + "reason": "Different model families and failures: xLSTM training bugs vs Bamba torch.compile/SDPA behavior." + }, + { + "left": "issue:43825", + "right": "issue:44655", + "accept": false, + "reason": "Pipeline task-message behavior and pipeline saving are separate code paths and symptoms." + }, + { + "left": "issue:43992", + "right": "issue:44484", + "accept": false, + "reason": "Missing checkpoint weight loading is unrelated to the default shard-size question in save_pretrained()." + }, + { + "left": "issue:43012", + "right": "issue:43550", + "accept": false, + "reason": "One is a PyTorch precision warning during compile; the other is a model-specific compile failure under SDPA." + }, + { + "left": "issue:43012", + "right": "issue:44368", + "accept": false, + "reason": "Different symptoms and code paths: compile-time warning vs repeated tie_word_embeddings warning in LoRA fine-tuning." + }, + { + "left": "issue:43352", + "right": "issue:44380", + "accept": false, + "reason": "Both mention attention backends, but one is unsupported FlashAttention 2 for a model and the other is ignored GPT-2 attention scaling." + }, + { + "left": "issue:42907", + "right": "issue:45753", + "accept": false, + "reason": "Saving dequantized models and an image-triggered infinite loop are unrelated problems." + }, + { + "left": "issue:42907", + "right": "issue:44655", + "accept": false, + "reason": "Model serialization after dequantization and saving Pipeline objects involve different save paths and failures." + }, + { + "left": "issue:43792", + "right": "issue:46123", + "accept": false, + "reason": "Whisper model loading failure and MaskGenerationPipeline partial-batch dropping are unrelated." + }, + { + "left": "issue:43352", + "right": "issue:43976", + "accept": false, + "reason": "A model capability error is unrelated to the Python-version compatibility regression." + }, + { + "left": "issue:44380", + "right": "issue:45479", + "accept": false, + "reason": "Attention-scaling being ignored is not the same bug as degenerate zero loss in sequence classification." + }, + { + "left": "issue:44380", + "right": "issue:46132", + "accept": false, + "reason": "Attention backend scaling and AttentionInterface.register behavior are distinct issues." + }, + { + "left": "issue:43064", + "right": "issue:44380", + "accept": false, + "reason": "Distributed optimizer-state corruption is unrelated to GPT-2 attention scaling under SDPA/FlashAttention." + }, + { + "left": "issue:44483", + "right": "issue:44514", + "accept": false, + "reason": "Chat-completions request handling and batched chat-template crashes are different interfaces and failures." + }, + { + "left": "issue:44222", + "right": "issue:45478", + "accept": false, + "reason": "Both involve Qwen3.5 MoE, but one is FP8 save_pretrained and the other is a from_pretrained regression; not the same concrete bug." + }, + { + "left": "issue:36683", + "right": "issue:43582", + "accept": false, + "reason": "Gemma3Config vocab_size and Apple Silicon allocator warmup errors are unrelated." + }, + { + "left": "issue:43208", + "right": "issue:43526", + "accept": false, + "reason": "xLSTM training bugs and BeitImageProcessorFast label reduction are different subsystems and defects." + }, + { + "left": "issue:43012", + "right": "issue:44683", + "accept": false, + "reason": "Both mention compilation, but one is a warning on bfloat16 models and the other is a torch>=2.9 flex_attention failure." + }, + { + "left": "issue:42757", + "right": "issue:43824", + "accept": false, + "reason": "Different missing imports from different packages with different fixes and call sites." + }, + { + "left": "issue:42490", + "right": "issue:43012", + "accept": false, + "reason": "Local-path import warning and bfloat16 compile warning are unrelated." + }, + { + "left": "issue:43618", + "right": "issue:43761", + "accept": false, + "reason": "Both are CLIP regressions, but one drops attentions and the other drops hidden_states; not the same behavior." + }, + { + "left": "issue:43618", + "right": "issue:44380", + "accept": false, + "reason": "CLIP output field regression is unrelated to GPT-2 attention scaling semantics." + }, + { + "left": "issue:44038", + "right": "issue:44230", + "accept": false, + "reason": "Different Qwen3-VL-Moe problems: general loading bug vs FP8 inference support." + }, + { + "left": "issue:38617", + "right": "issue:43502", + "accept": false, + "reason": "A missing import symbol and unwanted network requests under local_files_only are different bugs." + }, + { + "left": "issue:44483", + "right": "issue:45290", + "accept": false, + "reason": "Chat-completions API rejection and apply_chat_template tool-call crashes are different entry points." + }, + { + "left": "issue:44483", + "right": "issue:45381", + "accept": false, + "reason": "API request acceptance and Qwen2.5-VL video vision_position_ids are unrelated failures." + }, + { + "left": "issue:43756", + "right": "issue:44485", + "accept": false, + "reason": "Both mention RoPE, but the model/configuration defects are different and not mergeable into one fix." + }, + { + "left": "issue:43299", + "right": "issue:43867", + "accept": false, + "reason": "Qwen3VL MoE loading and sorted-state_dict load errors are separate loading-path bugs." + }, + { + "left": "issue:43299", + "right": "issue:45753", + "accept": false, + "reason": "Model-loading regression and image-specific infinite loop are unrelated." + }, + { + "left": "issue:43901", + "right": "issue:44655", + "accept": false, + "reason": "Documentation mismatch for return_all_scores and pipeline serialization failure are different issues." + }, + { + "left": "issue:45276", + "right": "issue:46077", + "accept": false, + "reason": "Embedding resizing and bidirectional attention mask construction are different Gemma4 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": 18, + "review_unit_index": 45, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:36010", + "issue:36296", + "issue:36683", + "issue:41669", + "issue:42490", + "issue:42757", + "issue:42890", + "issue:42898", + "issue:42994", + "issue:43010", + "issue:43064", + "issue:43066", + "issue:43352", + "issue:43525", + "issue:43575", + "issue:43611", + "issue:43618", + "issue:43867", + "issue:43931", + "issue:43937", + "issue:43976", + "issue:43992", + "issue:44075", + "issue:44077", + "issue:44242", + "issue:44246", + "issue:44263", + "issue:44355", + "issue:44380", + "issue:44492", + "issue:44625", + "issue:44655", + "issue:44855", + "issue:44938", + "issue:45071", + "issue:45216", + "issue:45278", + "issue:45310", + "issue:45357", + "issue:45375", + "issue:45381", + "issue:45440", + "issue:45561", + "issue:45593", + "issue:45850", + "issue:46002", + "issue:46077", + "issue:46082" + ], + "soft_pairs": [ + "issue:42994|issue:44655", + "issue:36010|issue:42757", + "issue:42898|issue:43066", + "issue:44077|issue:44380", + "issue:43352|issue:43937", + "issue:43867|issue:45216", + "issue:36683|issue:44077", + "issue:43611|issue:44655", + "issue:44625|issue:46077", + "issue:43976|issue:44938", + "issue:46077|issue:46082", + "issue:43352|issue:45593", + "issue:36683|issue:43525", + "issue:43618|issue:45216", + "issue:44263|issue:45440", + "issue:36683|issue:44242", + "issue:43352|issue:43931", + "issue:43064|issue:44492", + "issue:43010|issue:45278", + "issue:43992|issue:45310", + "issue:43867|issue:45375", + "issue:44355|issue:44855", + "issue:41669|issue:44246", + "issue:42890|issue:45561", + "issue:36683|issue:43064", + "issue:44938|issue:45071", + "issue:43867|issue:45357", + "issue:43010|issue:45850", + "issue:44938|issue:45381", + "issue:43010|issue:43976", + "issue:43618|issue:44077", + "issue:43992|issue:46002", + "issue:42490|issue:44075", + "issue:36296|issue:43575" + ], + "prepared_review_unit_hash": "659923f8cd3f90101a57fd04b16b378368ea41468d70a68be844d52d467e8353", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 34, + "serialized_chars": 37351, + "estimated_input_tokens": 9338, + "estimated_eval_tokens": 18932 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is dominated by unrelated Transformers v5 regression reports: import failures, model-loading/config issues, save_pretrained regressions, and a few training/runtime bugs. The soft pairs are mostly thematic similarities rather than true duplicates, so I would not merge any of them.", + "confidence": 0.7, + "canonical_issue_reason": "issue:45278 is the broadest concrete umbrella in this set: a v5 upgrade causing many import failures. It best represents the cluster\u2019s recurring version-regression theme without being as vague as the other broad reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45278 is the most suitable issue to anchor this cluster because it is broad, concrete, and closest to the shared upgrade-breakage pattern across several items.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:42994", + "right": "issue:44655", + "accept": false, + "reason": "Both involve saving, but one is quantized-model serialization and the other is Pipeline.save_pretrained. Different code paths and failure modes." + }, + { + "left": "issue:36010", + "right": "issue:42757", + "accept": false, + "reason": "Both are import errors, but they are missing different symbols from different packages. Same symptom, not the same bug." + }, + { + "left": "issue:42898", + "right": "issue:43066", + "accept": false, + "reason": "One is a v5 tokenization cleanup behavior change, the other is a tokenizer decoder type regression. Related area, different underlying issue." + }, + { + "left": "issue:44077", + "right": "issue:44380", + "accept": false, + "reason": "Unrelated failures: optional post_init handling vs GPT2 attention scaling being ignored. No shared concrete code path." + }, + { + "left": "issue:43352", + "right": "issue:43937", + "accept": false, + "reason": "Flash Attention support error for a model vs invalid GenerationConfig validation. Different subsystems." + }, + { + "left": "issue:43867", + "right": "issue:45216", + "accept": false, + "reason": "Both are loading/saving regressions, but one is a sorted state_dict load failure and the other is an incorrect Qwen3.5 checkpoint save. Not the same defect." + }, + { + "left": "issue:36683", + "right": "issue:44077", + "accept": false, + "reason": "A missing config attribute in Gemma3 is unrelated to patchtsmixer post_init restrictions. Only a superficial model/config similarity." + }, + { + "left": "issue:43611", + "right": "issue:44655", + "accept": false, + "reason": "base_model_prefix loading breakage and Pipeline.save_pretrained failure are separate serialization/loading paths." + }, + { + "left": "issue:44625", + "right": "issue:46077", + "accept": false, + "reason": "Qwen3.5 num_labels propagation and Gemma4 bidirectional attention masks are unrelated config bugs." + }, + { + "left": "issue:43976", + "right": "issue:44938", + "accept": false, + "reason": "Both mention Python version compatibility, but one is a 5.1.0/Python3.9+ mismatch and the other is a Python 3.14 import/load failure. Not the same bug." + }, + { + "left": "issue:46077", + "right": "issue:46082", + "accept": false, + "reason": "Different config validation problems: attention-mask construction vs rejecting explicit head_dim." + }, + { + "left": "issue:43352", + "right": "issue:45593", + "accept": false, + "reason": "Unsupported Flash Attention on Nemotron and missing auxiliary losses in D-FINE are unrelated model-specific issues." + }, + { + "left": "issue:36683", + "right": "issue:43525", + "accept": false, + "reason": "Both are missing config attributes, but for different models and different fields. Too broad to be a duplicate." + }, + { + "left": "issue:43618", + "right": "issue:45216", + "accept": false, + "reason": "CLIPOutput attentions assignment and Qwen3.5 checkpoint saving are distinct regressions." + }, + { + "left": "issue:44263", + "right": "issue:45440", + "accept": false, + "reason": "A torch.split indexing issue in GlmMoeDsaIndexer is unrelated to DeepSeekV3 implementation divergence." + }, + { + "left": "issue:36683", + "right": "issue:44242", + "accept": false, + "reason": "Gemma3Config missing vocab_size and load-balancing loss omission are different bugs in different model families." + }, + { + "left": "issue:43352", + "right": "issue:43931", + "accept": false, + "reason": "Flash Attention unsupported for Nemotron is unrelated to Qwen3-VL weight-shape mismatch." + }, + { + "left": "issue:43064", + "right": "issue:44492", + "accept": false, + "reason": "Trainer/FSDP2 optimizer state corruption and a cache-strategy typo are unrelated issues." + }, + { + "left": "issue:43010", + "right": "issue:45278", + "accept": false, + "reason": "No-grad decoration for cache updates and broad import failures are different problem areas." + }, + { + "left": "issue:43992", + "right": "issue:45310", + "accept": false, + "reason": "Missing embed_tokens.weight in UMT5Encoder and Qwen3.5 MoE from_pretrained errors are separate loading bugs." + }, + { + "left": "issue:43867", + "right": "issue:45375", + "accept": false, + "reason": "Sorted state_dict load failure and missing deepstack_visual_indexes in config are not the same defect." + }, + { + "left": "issue:44355", + "right": "issue:44855", + "accept": false, + "reason": "Compiled Python file loading errors and the Python 3.13 @torch.jit.script indentation parsing issue are distinct interpreter/codegen problems." + }, + { + "left": "issue:41669", + "right": "issue:44246", + "accept": false, + "reason": "Import-* performance cleanup and intermittent import slowdowns are related only at a very high level; not a duplicate." + }, + { + "left": "issue:42890", + "right": "issue:45561", + "accept": false, + "reason": "Missing set_seed in tests and xdist race on captured_info.txt are test-infrastructure issues, but different concrete failures." + }, + { + "left": "issue:36683", + "right": "issue:43064", + "accept": false, + "reason": "Gemma3Config missing vocab_size is not the same as FSDP2/PEFT optimizer state misalignment." + }, + { + "left": "issue:44938", + "right": "issue:45071", + "accept": false, + "reason": "Both are version-compatibility regressions, but one is Python 3.14 loading and the other is type-checking breakage in v5.4.0." + }, + { + "left": "issue:43867", + "right": "issue:45357", + "accept": false, + "reason": "Generic load-model state_dict sorting failure and incorrect visual encoder save keys are different serialization bugs." + }, + { + "left": "issue:43010", + "right": "issue:45850", + "accept": false, + "reason": "The no_grad cache update proposal is a specific fix request, while 45850 is a vague broad complaint about many model breakages." + }, + { + "left": "issue:44938", + "right": "issue:45381", + "accept": false, + "reason": "Python 3.14 load failure and Qwen2.5-VL video vision_position_ids are unrelated issues." + }, + { + "left": "issue:43010", + "right": "issue:43976", + "accept": false, + "reason": "Cache update no_grad decoration and Python version support for 5.1.0 are unrelated." + }, + { + "left": "issue:43618", + "right": "issue:44077", + "accept": false, + "reason": "CLIPOutput attentions regression and patchtsmixer post_init restrictions are different model-specific changes." + }, + { + "left": "issue:43992", + "right": "issue:46002", + "accept": false, + "reason": "UMT5Encoder missing weights and NaNs in classification heads after init/checkout are not the same bug." + }, + { + "left": "issue:42490", + "right": "issue:44075", + "accept": false, + "reason": "Local-path trailing-slash import warning and ignored SGD args are unrelated." + }, + { + "left": "issue:36296", + "right": "issue:43575", + "accept": false, + "reason": "Tensor parallel training bug and Qwen2-57B-A14B-A14B Instruct OOM are different runtime problems, not a single duplicate." + } + ] + }, + "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": 18, + "review_unit_index": 46, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:28282", + "issue:30333", + "issue:30990", + "issue:31795", + "issue:33453", + "issue:34689", + "issue:36296", + "issue:42548", + "issue:42757", + "issue:42831", + "issue:42890", + "issue:43010", + "issue:43065", + "issue:43232", + "issue:43299", + "issue:43502", + "issue:43576", + "issue:43618", + "issue:43723", + "issue:43827", + "issue:43867", + "issue:43874", + "issue:44062", + "issue:44112", + "issue:44222", + "issue:44230", + "issue:44315", + "issue:44351", + "issue:44355", + "issue:44360", + "issue:44387", + "issue:44484", + "issue:44485", + "issue:44556", + "issue:44655", + "issue:44749", + "issue:44857", + "issue:44961", + "issue:44964", + "issue:44977", + "issue:45071", + "issue:45125", + "issue:45341", + "issue:45375", + "issue:45440", + "issue:45478", + "issue:45600", + "issue:46132" + ], + "soft_pairs": [ + "issue:28282|issue:33453", + "issue:43867|issue:44484", + "issue:43618|issue:44857", + "issue:44360|issue:45440", + "issue:43618|issue:44977", + "issue:42890|issue:44964", + "issue:42890|issue:44112", + "issue:42890|issue:45600", + "issue:43502|issue:44062", + "issue:43576|issue:46132", + "issue:43723|issue:44556", + "issue:44387|issue:44556", + "issue:42548|issue:44351", + "issue:42890|issue:45341", + "issue:43010|issue:43065", + "issue:44222|issue:45125", + "issue:43502|issue:44485", + "issue:44355|issue:44749", + "issue:45125|issue:45478", + "issue:30333|issue:33453", + "issue:43010|issue:43232", + "issue:45125|issue:45375", + "issue:30990|issue:31795", + "issue:43502|issue:44222", + "issue:43299|issue:44556", + "issue:42831|issue:43874", + "issue:44315|issue:44360", + "issue:43502|issue:44961", + "issue:43827|issue:44961", + "issue:30333|issue:36296", + "issue:44355|issue:45071", + "issue:44355|issue:44655", + "issue:33453|issue:34689", + "issue:42757|issue:43576", + "issue:43576|issue:45071", + "issue:44222|issue:44230", + "issue:33453|issue:36296" + ], + "prepared_review_unit_hash": "736a8e2a852d9fe0084970f99dbbebf6357114db6da7d9835fadbb6ec5408569", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 37, + "serialized_chars": 38152, + "estimated_input_tokens": 9538, + "estimated_eval_tokens": 19332 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is very heterogeneous. The soft pairs mostly share only broad wording like \u201cv5\u201d, \u201cloading\u201d, or \u201cregression\u201d, but they point to different code paths, models, or tooling issues. I would not merge any of the soft edges.", + "confidence": 0.76, + "canonical_issue_reason": "issue:43723 is the broadest, most user-facing loading regression in the set and best fits the loose v5 breakage theme, even though the overall cluster is too mixed for a strong duplicate canonical.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43723 is the clearest general anchor: it is concrete, actionable, and broadly about v5 tokenizer loading breakage rather than a model-specific or test-only failure.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:28282", + "right": "issue:33453", + "accept": false, + "reason": "Missing PyTorch import error vs tokenizer-loading regression; same general area, different bug." + }, + { + "left": "issue:43867", + "right": "issue:44484", + "accept": false, + "reason": "State-dict load ordering issue vs a save_pretrained shard-size question; unrelated root causes." + }, + { + "left": "issue:43618", + "right": "issue:44857", + "accept": false, + "reason": "CLIP attentions regression vs LwDetr AMP/CUDA crash; different models and failures." + }, + { + "left": "issue:44360", + "right": "issue:45440", + "accept": false, + "reason": "DSA indexer/ReLU discussion vs DeepseekV3MoE implementation divergence; not the same bug." + }, + { + "left": "issue:43618", + "right": "issue:44977", + "accept": false, + "reason": "CLIP attentions assignment bug vs Qwen3.5 flash-attention generation bug; different code paths." + }, + { + "left": "issue:42890", + "right": "issue:44964", + "accept": false, + "reason": "Flaky integration test from missing seed vs Phi-4 multimodal load failure; unrelated." + }, + { + "left": "issue:42890", + "right": "issue:44112", + "accept": false, + "reason": "Test flakiness vs CI stale device override failure; both test-related, but not the same issue." + }, + { + "left": "issue:42890", + "right": "issue:45600", + "accept": false, + "reason": "Missing set_seed flake vs auto_mappings CI break; different causes." + }, + { + "left": "issue:43502", + "right": "issue:44062", + "accept": false, + "reason": "local_files_only network leakage vs AddedToken kwarg collision; unrelated failures." + }, + { + "left": "issue:43576", + "right": "issue:46132", + "accept": false, + "reason": "v5 env command breakage vs AttentionInterface.register behavior change; different features." + }, + { + "left": "issue:43723", + "right": "issue:44556", + "accept": false, + "reason": "Both are loading regressions, but tokenizer.from_pretrained failure and checkpoint reload failure are distinct paths." + }, + { + "left": "issue:44387", + "right": "issue:44556", + "accept": false, + "reason": "Int4 memory regression vs checkpoint reload regression; different symptoms and fixes." + }, + { + "left": "issue:42548", + "right": "issue:44351", + "accept": false, + "reason": "Import errors for different symbols from different packages; not duplicates." + }, + { + "left": "issue:42890", + "right": "issue:45341", + "accept": false, + "reason": "SamHQ test flake vs a testing_utils bug; different scopes." + }, + { + "left": "issue:43010", + "right": "issue:43065", + "accept": false, + "reason": "@no_grad cache update fix vs dummy Conv2d in Sam3PixelDecoder; unrelated changes." + }, + { + "left": "issue:44222", + "right": "issue:45125", + "accept": false, + "reason": "FP8 save_pretrained bug vs missing tensor-parallel plan; same broad area, different concrete problems." + }, + { + "left": "issue:43502", + "right": "issue:44485", + "accept": false, + "reason": "Network call leakage vs GLM-5 RoPE discussion; no shared bug." + }, + { + "left": "issue:44355", + "right": "issue:44749", + "accept": false, + "reason": "Compiled Python file errors vs slower filtering after upgrade; different symptoms." + }, + { + "left": "issue:45125", + "right": "issue:45478", + "accept": false, + "reason": "Missing _tp_plan vs Qwen3.5 Moe from_pretrained error; related family, but not the same fixable bug." + }, + { + "left": "issue:30333", + "right": "issue:33453", + "accept": false, + "reason": "MLflow reporting failure vs tokenizer loading regression; unrelated." + }, + { + "left": "issue:43010", + "right": "issue:43232", + "accept": false, + "reason": "Cache update no_grad fix vs generation kwarg-sync issue; different code paths." + }, + { + "left": "issue:45125", + "right": "issue:45375", + "accept": false, + "reason": "Missing tensor-parallel plan vs missing config field; different model metadata bugs." + }, + { + "left": "issue:30990", + "right": "issue:31795", + "accept": false, + "reason": "Sentence-Transformers loading stall vs documentation confusion; not duplicates." + }, + { + "left": "issue:43502", + "right": "issue:44222", + "accept": false, + "reason": "local_files_only API leakage vs FP8 save_pretrained bug; different features." + }, + { + "left": "issue:43299", + "right": "issue:44556", + "accept": false, + "reason": "Both are model-loading regressions, but they affect different model families and likely need different fixes." + }, + { + "left": "issue:42831", + "right": "issue:43874", + "accept": false, + "reason": "FP8 accuracy issue vs missing image-patch-count method; different underlying defects." + }, + { + "left": "issue:44315", + "right": "issue:44360", + "accept": false, + "reason": "Liger Kernel not applied via model_init vs DSA indexer ReLU discussion; unrelated." + }, + { + "left": "issue:43502", + "right": "issue:44961", + "accept": false, + "reason": "Real bug vs placeholder title; no duplicate signal." + }, + { + "left": "issue:43827", + "right": "issue:44961", + "accept": false, + "reason": "Docs still mention pipeline() vs placeholder title; not a duplicate pair." + }, + { + "left": "issue:30333", + "right": "issue:36296", + "accept": false, + "reason": "MLflow reporting bug vs tensor-parallel training bug; different subsystems." + }, + { + "left": "issue:44355", + "right": "issue:45071", + "accept": false, + "reason": "Compiled-file runtime errors vs PretrainedConfig type-checking regression; different problems." + }, + { + "left": "issue:44355", + "right": "issue:44655", + "accept": false, + "reason": "Compiled-file runtime errors vs pipeline save_pretrained problem; unrelated." + }, + { + "left": "issue:33453", + "right": "issue:34689", + "accept": false, + "reason": "Tokenizer-loading regression vs Llama 3.2 model-loading regression; similar wording, different concrete failures." + }, + { + "left": "issue:42757", + "right": "issue:43576", + "accept": false, + "reason": "Missing export from huggingface_hub vs broken transformers env command; distinct bugs." + }, + { + "left": "issue:43576", + "right": "issue:45071", + "accept": false, + "reason": "CLI env regression vs config type-checking regression; not the same issue." + }, + { + "left": "issue:44222", + "right": "issue:44230", + "accept": false, + "reason": "FP8 save_pretrained bug vs FP8 inference support request/bug; related area but not one fix." + }, + { + "left": "issue:33453", + "right": "issue:36296", + "accept": false, + "reason": "Tokenizer-loading regression vs tensor-parallel training bug; different root causes 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": 18, + "review_unit_index": 47, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:31795", + "issue:36296", + "issue:38617", + "issue:41628", + "issue:41669", + "issue:42371", + "issue:42548", + "issue:42890", + "issue:43352", + "issue:43408", + "issue:43525", + "issue:43526", + "issue:43572", + "issue:43576", + "issue:43673", + "issue:43824", + "issue:43901", + "issue:43976", + "issue:44038", + "issue:44230", + "issue:44246", + "issue:44261", + "issue:44273", + "issue:44295", + "issue:44297", + "issue:44315", + "issue:44355", + "issue:44360", + "issue:44393", + "issue:44485", + "issue:44492", + "issue:44556", + "issue:44623", + "issue:44704", + "issue:44908", + "issue:44961", + "issue:45003", + "issue:45071", + "issue:45092", + "issue:45125", + "issue:45230", + "issue:45278", + "issue:45362", + "issue:45468", + "issue:45542", + "issue:45753", + "issue:45850", + "issue:46095" + ], + "soft_pairs": [ + "issue:43576|issue:43824", + "issue:41628|issue:42548", + "issue:31795|issue:38617", + "issue:44961|issue:45468", + "issue:38617|issue:43576", + "issue:44908|issue:44961", + "issue:43673|issue:44492", + "issue:44485|issue:44961", + "issue:36296|issue:43901", + "issue:41628|issue:44355", + "issue:36296|issue:43526", + "issue:45362|issue:45850", + "issue:44623|issue:45230", + "issue:44261|issue:44360", + "issue:44273|issue:45542", + "issue:44273|issue:45003", + "issue:44038|issue:45125", + "issue:44393|issue:45071", + "issue:44393|issue:45753", + "issue:44230|issue:46095", + "issue:43576|issue:44556", + "issue:41669|issue:43673", + "issue:43525|issue:45850", + "issue:44556|issue:45092", + "issue:43572|issue:45850", + "issue:44492|issue:45850", + "issue:42371|issue:44492", + "issue:45278|issue:45850", + "issue:42890|issue:44908", + "issue:44246|issue:44273", + "issue:44704|issue:44908", + "issue:43408|issue:44315", + "issue:43976|issue:44393", + "issue:43352|issue:44393", + "issue:43576|issue:44908", + "issue:44297|issue:45230", + "issue:44295|issue:45230", + "issue:44908|issue:45071" + ], + "prepared_review_unit_hash": "cbc8b9166ba34a4ffa013b9da34de4989b09fe92509b3e0785279957ea74b557", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 38, + "serialized_chars": 38435, + "estimated_input_tokens": 9609, + "estimated_eval_tokens": 19474 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of unrelated Transformers v5 regressions: import failures, loading/lazy-loading issues, model-specific config bugs, and a few docs/test reports. None of the soft pairs looks like the same concrete bug/change, so I would not merge any of them as duplicates. The broadest umbrella issue is the general v5 breakage report.", + "confidence": 0.48, + "canonical_issue_reason": "Issue 45850 is the broadest umbrella in the set: it explicitly frames a general 'v5 breaks too many models' regression, which is the best catch-all for the cluster\u2019s version-5 compatibility complaints.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45850 is the most suitable global representative because it is the broadest open regression report and can anchor follow-up links to more specific v5 breakages.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43576", + "right": "issue:43824", + "accept": false, + "reason": "Both are v5-related, but one is a broken `transformers env` command and the other is an ImportError for a specific model class; different bugs." + }, + { + "left": "issue:41628", + "right": "issue:42548", + "accept": false, + "reason": "These are different missing imports (`AutoImageProcessor` vs `PreTrainedModel`) and do not point to the same failure path." + }, + { + "left": "issue:31795", + "right": "issue:38617", + "accept": false, + "reason": "Documentation confusion in `model.forward` is unrelated to an ImportError for `layer_type_validation`." + }, + { + "left": "issue:44961", + "right": "issue:45468", + "accept": false, + "reason": "A meaningless title (`racoon`) does not match the Gemma-4 audio positional encoding bug." + }, + { + "left": "issue:38617", + "right": "issue:43576", + "accept": false, + "reason": "One is a specific import failure, the other is a broken CLI command; only a broad v5 theme is shared." + }, + { + "left": "issue:44908", + "right": "issue:44961", + "accept": false, + "reason": "The scheduler kwargs bug has nothing to do with the unrelated `racoon` issue." + }, + { + "left": "issue:43673", + "right": "issue:44492", + "accept": false, + "reason": "Chunked-prefill cache missing is a runtime bug; the cache-strategies item is just a typo/wording issue." + }, + { + "left": "issue:44485", + "right": "issue:44961", + "accept": false, + "reason": "GLM-5 RoPE implementation and `racoon` are clearly unrelated." + }, + { + "left": "issue:36296", + "right": "issue:43901", + "accept": false, + "reason": "Tensor-parallel training is unrelated to a documentation mismatch about `return_all_scores`." + }, + { + "left": "issue:41628", + "right": "issue:44355", + "accept": false, + "reason": "A missing `AutoImageProcessor` import and errors in compiled Python files are different problems." + }, + { + "left": "issue:36296", + "right": "issue:43526", + "accept": false, + "reason": "Tensor-parallel training and `BeitImageProcessorFast.reduce_labels` are different code paths." + }, + { + "left": "issue:45362", + "right": "issue:45850", + "accept": false, + "reason": "Both are broad regression complaints, but one is a specific Qwen3.5 chat crash and the other is a general v5 breakage meta-report." + }, + { + "left": "issue:44623", + "right": "issue:45230", + "accept": false, + "reason": "`processor.save_pretrained(...)` missing files is a concrete bug; `Bug report` is too vague to be the same issue." + }, + { + "left": "issue:44261", + "right": "issue:44360", + "accept": false, + "reason": "MLA layernorm precision/config issue and a DSA indexer ReLU omission are unrelated." + }, + { + "left": "issue:44273", + "right": "issue:45542", + "accept": false, + "reason": "Lazy loading malfunction and a TensorBoard-without-TensorFlow backend error are different failures." + }, + { + "left": "issue:44273", + "right": "issue:45003", + "accept": false, + "reason": "Lazy loading and unsafe `sys.modules` access may be adjacent internals, but they are not the same concrete bug." + }, + { + "left": "issue:44038", + "right": "issue:45125", + "accept": false, + "reason": "A Qwen3-VL bug report and missing `_tp_plan` for Qwen3_5Moe tensor parallelism are different model-specific issues." + }, + { + "left": "issue:44393", + "right": "issue:45071", + "accept": false, + "reason": "A Qwen3-VL 2D bbox error is unrelated to `PretrainedConfig` type-checking breakage." + }, + { + "left": "issue:44393", + "right": "issue:45753", + "accept": false, + "reason": "Both mention Qwen3-VL/Qwen3_5, but one is bounding-box output corruption and the other is an infinite loop on a specific image; not the same bug." + }, + { + "left": "issue:44230", + "right": "issue:46095", + "accept": false, + "reason": "FP8 inference support for Qwen models and a DeepSeekV4 weight-conversion question are unrelated." + }, + { + "left": "issue:43576", + "right": "issue:44556", + "accept": false, + "reason": "Broken `transformers env` and inability to reload old checkpoints after upgrading are separate regression classes." + }, + { + "left": "issue:41669", + "right": "issue:43673", + "accept": false, + "reason": "Import-time `import *` slowdown and missing GenerationMixin cache in chunked prefill are unrelated." + }, + { + "left": "issue:43525", + "right": "issue:45850", + "accept": false, + "reason": "Missing `pad_token_id` in Llama4Config is a specific model config bug, not the same as the broad v5 breakage complaint." + }, + { + "left": "issue:44556", + "right": "issue:45092", + "accept": false, + "reason": "Both are checkpoint-compatibility issues after v5, but they affect different model families and likely different load paths." + }, + { + "left": "issue:43572", + "right": "issue:45850", + "accept": false, + "reason": "StableLmConfig missing `pad_token_idx` is a specific config regression, not the same as the generic v5 complaint." + }, + { + "left": "issue:44492", + "right": "issue:45850", + "accept": false, + "reason": "A typo in cache-strategy docs/code is not the same as broad model breakage." + }, + { + "left": "issue:42371", + "right": "issue:44492", + "accept": false, + "reason": "TF32 settings behavior and cache-strategies typo are unrelated." + }, + { + "left": "issue:45278", + "right": "issue:45850", + "accept": false, + "reason": "The import-errors-on-upgrade report is related at a high level, but it is still a different concrete problem than the generic 'many models break' meta-issue." + }, + { + "left": "issue:42890", + "right": "issue:44908", + "accept": false, + "reason": "A missing `set_seed()` in tests is unrelated to scheduler kwargs handling." + }, + { + "left": "issue:44246", + "right": "issue:44273", + "accept": false, + "reason": "Slow import timing and broken lazy loading are both import-related, but they are not the same underlying bug." + }, + { + "left": "issue:44704", + "right": "issue:44908", + "accept": false, + "reason": "Both mention kwargs, but one is `AutoProcessor.from_pretrained` not forwarding them and the other is an LR scheduler ignoring them." + }, + { + "left": "issue:43408", + "right": "issue:44315", + "accept": false, + "reason": "A model-type mismatch warning and Liger Kernel not applying under `model_init` are different issues." + }, + { + "left": "issue:43976", + "right": "issue:44393", + "accept": false, + "reason": "Python-version compatibility breakage is unrelated to a Qwen3-VL bounding-box hallucination/error." + }, + { + "left": "issue:43352", + "right": "issue:44393", + "accept": false, + "reason": "Flash Attention 2 support for Nemotron and Qwen3-VL bbox output problems are different model features and bugs." + }, + { + "left": "issue:43576", + "right": "issue:44908", + "accept": false, + "reason": "Broken `transformers env` and scheduler kwargs ignored do not share the same code path." + }, + { + "left": "issue:44297", + "right": "issue:45230", + "accept": false, + "reason": "Tokenizer config mismatch and a generic bug report are not the same issue." + }, + { + "left": "issue:44295", + "right": "issue:45230", + "accept": false, + "reason": "Position_ids buffer read errors are unrelated to the vague bug report." + }, + { + "left": "issue:44908", + "right": "issue:45071", + "accept": false, + "reason": "Scheduler kwargs handling and `PretrainedConfig` type-checking breakage 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": 18, + "review_unit_index": 48, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:31515", + "issue:31795", + "issue:33453", + "issue:34689", + "issue:37428", + "issue:38617", + "issue:41669", + "issue:42371", + "issue:42913", + "issue:43066", + "issue:43298", + "issue:43502", + "issue:43519", + "issue:43525", + "issue:43572", + "issue:43576", + "issue:43606", + "issue:43618", + "issue:43704", + "issue:43827", + "issue:43874", + "issue:44077", + "issue:44230", + "issue:44261", + "issue:44263", + "issue:44273", + "issue:44315", + "issue:44351", + "issue:44360", + "issue:44393", + "issue:44485", + "issue:44492", + "issue:44623", + "issue:44655", + "issue:44829", + "issue:44861", + "issue:44908", + "issue:44945", + "issue:44998", + "issue:45092", + "issue:45230", + "issue:45290", + "issue:45310", + "issue:45341", + "issue:45468", + "issue:45478", + "issue:45850", + "issue:46095" + ], + "soft_pairs": [ + "issue:41669|issue:44351", + "issue:41669|issue:45341", + "issue:44492|issue:44945", + "issue:43874|issue:45850", + "issue:44393|issue:45478", + "issue:44077|issue:44861", + "issue:43827|issue:44908", + "issue:41669|issue:44908", + "issue:31795|issue:33453", + "issue:44315|issue:44829", + "issue:43576|issue:44273", + "issue:44360|issue:45468", + "issue:44655|issue:44861", + "issue:45230|issue:45310", + "issue:44273|issue:44908", + "issue:43519|issue:44485", + "issue:43519|issue:44908", + "issue:31795|issue:34689", + "issue:41669|issue:43704", + "issue:31795|issue:37428", + "issue:44315|issue:45092", + "issue:44485|issue:45468", + "issue:43704|issue:44485", + "issue:43298|issue:46095", + "issue:43502|issue:43519", + "issue:42371|issue:43704", + "issue:43519|issue:44623", + "issue:43519|issue:43704", + "issue:43606|issue:45468", + "issue:44263|issue:45468", + "issue:44261|issue:45468", + "issue:42913|issue:43066", + "issue:44908|issue:44998", + "issue:44492|issue:44998", + "issue:44861|issue:45290", + "issue:43618|issue:44998", + "issue:44998|issue:45850", + "issue:44998|issue:45341", + "issue:43525|issue:44861", + "issue:43572|issue:44861", + "issue:31515|issue:38617", + "issue:44230|issue:44623" + ], + "prepared_review_unit_hash": "6811cc7eba357048bca7c967009d8d1285b6da2c66f46ffdb0cf161aa4987ff2", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 42, + "serialized_chars": 39658, + "estimated_input_tokens": 9915, + "estimated_eval_tokens": 20086 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly noisy false-positive similarity across unrelated Transformers issues. The clearest representative is the Qwen3.5 MoE from_pretrained regression; the later same-title report looks like the closest duplicate anchor. No PRs are present.", + "confidence": 0.69, + "canonical_issue_reason": "issue:45310 is the earliest concrete report of a specific from_pretrained regression for Qwen3.5 Moe and is the best anchor for the duplicated same-title follow-up.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45310 is the most useful issue to keep as the representative bug report: specific, version-scoped, and directly tied to a reproducible model-loading failure.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:41669", + "right": "issue:44351", + "accept": false, + "reason": "Both are import-related, but one is about import-star performance/scale and the other is a missing public import; different failure modes." + }, + { + "left": "issue:41669", + "right": "issue:45341", + "accept": false, + "reason": "Unrelated: model import mechanics vs a small testing_utils bug." + }, + { + "left": "issue:44492", + "right": "issue:44945", + "accept": false, + "reason": "Cache-strategy typo and pipeline-parallelism output bug are different code paths." + }, + { + "left": "issue:43874", + "right": "issue:45850", + "accept": false, + "reason": "A missing image-processor method is too specific to merge with a broad '5.0 breaks too many models' complaint." + }, + { + "left": "issue:44393", + "right": "issue:45478", + "accept": false, + "reason": "Different model families and different bugs: 2D bbox output error vs Qwen3.5 MoE from_pretrained failure." + }, + { + "left": "issue:44077", + "right": "issue:44861", + "accept": false, + "reason": "One is a validation policy issue, the other is an AttributeError in tied-weight key handling." + }, + { + "left": "issue:43827", + "right": "issue:44908", + "accept": false, + "reason": "Docs still referencing pipeline() and scheduler kwargs handling are unrelated." + }, + { + "left": "issue:41669", + "right": "issue:44908", + "accept": false, + "reason": "Import-star cleanup and scheduler kwargs are unrelated." + }, + { + "left": "issue:31795", + "right": "issue:33453", + "accept": false, + "reason": "Documentation confusion in forward signatures does not match a tokenizer loading regression." + }, + { + "left": "issue:44315", + "right": "issue:44829", + "accept": false, + "reason": "Both involve training/model creation, but they are different concrete bugs and not one mergeable fix." + }, + { + "left": "issue:43576", + "right": "issue:44273", + "accept": false, + "reason": "Broken env command and lazy loading are separate issues." + }, + { + "left": "issue:44360", + "right": "issue:45468", + "accept": false, + "reason": "DSA indexer activation behavior and Gemma audio positional encoding are unrelated." + }, + { + "left": "issue:44655", + "right": "issue:44861", + "accept": false, + "reason": "Saving Pipeline objects and tied-weight key crashes do not share the same underlying defect." + }, + { + "left": "issue:45230", + "right": "issue:45310", + "accept": false, + "reason": "'Bug report' is too generic to establish the same concrete from_pretrained regression." + }, + { + "left": "issue:44273", + "right": "issue:44908", + "accept": false, + "reason": "Lazy loading and inverse_sqrt scheduler kwargs are different subsystems." + }, + { + "left": "issue:43519", + "right": "issue:44485", + "accept": false, + "reason": "Timestamp calculation in Qwen3VL and GLM-5 RoPE implementation are different model-specific bugs." + }, + { + "left": "issue:43519", + "right": "issue:44908", + "accept": false, + "reason": "A processor timestamp bug is unrelated to scheduler kwarg propagation." + }, + { + "left": "issue:31795", + "right": "issue:34689", + "accept": false, + "reason": "Forward-doc confusion and Llama 3.2 Vision loading regression are not the same bug." + }, + { + "left": "issue:41669", + "right": "issue:43704", + "accept": false, + "reason": "Import-star cleanup and VRAM leak under dataloader threads are unrelated." + }, + { + "left": "issue:31795", + "right": "issue:37428", + "accept": false, + "reason": "Forward docs and a missing flash-attention import are unrelated." + }, + { + "left": "issue:44315", + "right": "issue:45092", + "accept": false, + "reason": "Both mention model_init/meta-init context, but the concrete failures are different and not one code-path fix." + }, + { + "left": "issue:44485", + "right": "issue:45468", + "accept": false, + "reason": "Different model implementations and different symptoms." + }, + { + "left": "issue:43704", + "right": "issue:44485", + "accept": false, + "reason": "VRAM leak and RoPE implementation are unrelated." + }, + { + "left": "issue:43298", + "right": "issue:46095", + "accept": false, + "reason": "Continuous batching for audio models and a DeepSeek weight-conversion question are unrelated." + }, + { + "left": "issue:43502", + "right": "issue:43519", + "accept": false, + "reason": "Local-files-only network leakage and timestamp calculation are unrelated." + }, + { + "left": "issue:42371", + "right": "issue:43704", + "accept": false, + "reason": "TF32 settings guidance and VRAM leak are unrelated." + }, + { + "left": "issue:43519", + "right": "issue:44623", + "accept": false, + "reason": "Processor timestamp math and missing processor save files are separate bugs." + }, + { + "left": "issue:43519", + "right": "issue:43704", + "accept": false, + "reason": "Timestamp calculation and VRAM leakage do not share a code path." + }, + { + "left": "issue:43606", + "right": "issue:45468", + "accept": false, + "reason": "CPU offload device mismatch and Gemma audio positional encoding are unrelated." + }, + { + "left": "issue:44263", + "right": "issue:45468", + "accept": false, + "reason": "torch.split return handling in an indexer and Gemma audio positional encoding are different defects." + }, + { + "left": "issue:44261", + "right": "issue:45468", + "accept": false, + "reason": "RMS norm epsilon precision issue in MLA and Gemma audio positional encoding are unrelated." + }, + { + "left": "issue:42913", + "right": "issue:43066", + "accept": false, + "reason": "Both are tokenizer v5 issues, but one is broad behavior drift while the other is a specific decoder-type problem; not enough to treat as the same concrete bug." + }, + { + "left": "issue:44908", + "right": "issue:44998", + "accept": false, + "reason": "Scheduler kwargs and an unrelated placeholder issue are not duplicates." + }, + { + "left": "issue:44492", + "right": "issue:44998", + "accept": false, + "reason": "Cache-strategy typo and an unrelated placeholder issue are not duplicates." + }, + { + "left": "issue:44861", + "right": "issue:45290", + "accept": false, + "reason": "Tied-weight key handling and chat-template tool-call crashes are different code paths." + }, + { + "left": "issue:43618", + "right": "issue:44998", + "accept": false, + "reason": "CLIPOutput attentions assignment and an unrelated placeholder issue are not duplicates." + }, + { + "left": "issue:44998", + "right": "issue:45850", + "accept": false, + "reason": "A placeholder issue cannot be merged with a broad v5 complaints issue." + }, + { + "left": "issue:44998", + "right": "issue:45341", + "accept": false, + "reason": "An unrelated placeholder issue and a testing_utils bug are not duplicates." + }, + { + "left": "issue:43525", + "right": "issue:44861", + "accept": false, + "reason": "Missing pad_token_id on Llama4Config and tied-weight key AttributeError are different issues." + }, + { + "left": "issue:43572", + "right": "issue:44861", + "accept": false, + "reason": "Missing pad_token_idx on StableLmConfig and tied-weight key handling are unrelated." + }, + { + "left": "issue:31515", + "right": "issue:38617", + "accept": false, + "reason": "Slow checkpoint loading and missing layer_type_validation import are unrelated." + }, + { + "left": "issue:44230", + "right": "issue:44623", + "accept": false, + "reason": "fp8 inference support request and processor save_pretrained missing files 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": 18, + "review_unit_index": 49, + "review_unit_count": 49, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-414", + "nodes": [ + "issue:30333", + "issue:31515", + "issue:31795", + "issue:33453", + "issue:36296", + "issue:43298", + "issue:44230", + "issue:45310" + ], + "soft_pairs": [ + "issue:30333|issue:31515", + "issue:31515|issue:36296", + "issue:31515|issue:31795", + "issue:44230|issue:45310", + "issue:31515|issue:33453", + "issue:43298|issue:44230" + ], + "prepared_review_unit_hash": "b87fac37ae6cd5746a3974ae02fa2b9277272c07e1847aeee05a4ca6f6f70f5e", + "component_budget": { + "node_count": 414, + "item_count": 414, + "soft_pair_count": 1552, + "serialized_chars": 800272, + "estimated_input_tokens": 200068, + "estimated_eval_tokens": 400392 + }, + "budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 6, + "serialized_chars": 6307, + "estimated_input_tokens": 1577, + "estimated_eval_tokens": 3410 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These issues are mostly unrelated and do not form a duplicate cluster. The only apparent thematic overlap is between the two Qwen/MoE loading issues, but they still describe different failure modes and scopes.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "issue:45310 is the strongest representative of the cluster\u2019s bug-report style: it is concrete, recent, and describes a specific regression in from_pretrained for a named model family. That said, the set overall is too heterogeneous for a true canonical issue.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:30333", + "right": "issue:31515", + "accept": false, + "reason": "Different subsystems and symptoms: MLflow reporting after failed training vs slow checkpoint loading in from_pretrained." + }, + { + "left": "issue:31515", + "right": "issue:36296", + "accept": false, + "reason": "One is about slow checkpoint load time; the other is a tensor-parallel training bug. No same code-path evidence." + }, + { + "left": "issue:31515", + "right": "issue:31795", + "accept": false, + "reason": "Checkpoint load performance is unrelated to documentation confusion about input_ids and past_key_values." + }, + { + "left": "issue:44230", + "right": "issue:45310", + "accept": false, + "reason": "Both mention Qwen3.5/MoE and loading/inference, but one is fp8 inference support and the other is a transformers>=5.4.0 from_pretrained error. Different concrete fixes." + }, + { + "left": "issue:31515", + "right": "issue:33453", + "accept": false, + "reason": "Slow checkpoint loading and tokenizer loading regression are both loading-related but not the same bug or code path." + }, + { + "left": "issue:43298", + "right": "issue:44230", + "accept": false, + "reason": "Continuous batching support for audio models and fp8 inference support are separate feature/compatibility issues." + } + ] + }, + "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": 18, + "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": "f8550c830647b2cd6150b5eb785f204a3c0e55af4d46d489eb30168238ebaef2", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 2875, + "estimated_input_tokens": 719, + "estimated_eval_tokens": 1694 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 2875, + "estimated_input_tokens": 719, + "estimated_eval_tokens": 1694 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The three issues are not duplicates; they cover unrelated topics: tokenizer truncation behavior, a ReduceLROnPlateau-like training utility, and Wav2Vec2 word time offsets. No soft edge should be accepted.", + "confidence": 0.97, + "canonical_issue_reason": "Issue 10519 is the clearest standalone request: it specifies one concrete behavior change for truncation direction and is well-scoped.", + "canonical_pr_reason": null, + "best_issue_reason": "10519 is the best representative issue only because it is the most concrete and actionable; the other two are separate feature requests in different areas.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10519", + "right": "issue:11307", + "accept": false, + "reason": "Different subsystems and user goals: tokenizer truncation direction vs Wav2Vec2 word boundary timestamps." + }, + { + "left": "issue:10519", + "right": "issue:11005", + "accept": false, + "reason": "Completely unrelated requests: truncation behavior vs learning-rate scheduler-like functionality." + }, + { + "left": "issue:11005", + "right": "issue:11307", + "accept": false, + "reason": "No shared bug or change path; one is about optimization scheduling, the other about speech model offsets." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 3, + "component_count": 18, + "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": "be741c67b264bcfbb39dad868fecd1110b0e115a7ff4098c8594216ca60801d3", + "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": "These three issues are only loosely related by transformers/model loading or tokenization themes, but the titles point to different problems: tokenizer mismatch, private-model auth/error messaging, and a generic inability to load a specific model. I would not merge any pair as duplicates.", + "confidence": 0.84, + "canonical_issue_reason": "Issue 16618 is the broadest and most discussion-heavy representative of the cluster, but it still appears to describe a different failure mode than the others.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 16618 is the best single representative if one must be chosen, because it is the most general and likely to attract the most follow-up context.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:13244", + "right": "issue:16618", + "accept": false, + "reason": "Tokenizer output mismatch for TAPAS is a different bug from a generic model-loading failure; no evidence they share the same code path." + }, + { + "left": "issue:15980", + "right": "issue:16618", + "accept": false, + "reason": "One is specifically about a bad auth-related error when downloading a private model, while the other is a generic failure loading bert-base-uncased; titles do not indicate the same underlying issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 4, + "component_count": 18, + "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": "5864f33bdb333612b76f9a9d25ea04893163a2675ae634a49d79bb0829303bae", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1589, + "estimated_input_tokens": 398, + "estimated_eval_tokens": 1052 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1589, + "estimated_input_tokens": 398, + "estimated_eval_tokens": 1052 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two separate community-contribution issues for different feature areas: Flash Attention 2 support vs GGUF support. They are not duplicates and should remain distinct.", + "confidence": 0.97, + "canonical_issue_reason": "Issue 26350 is the older and more discussed thread, so it is the better canonical issue if one must be chosen for the cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 26350 is the strongest representative because it has higher discussion activity and is the earlier, more established request.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:26350", + "right": "issue:33260", + "accept": false, + "reason": "Different underlying requests and code paths: one is about Flash Attention 2 architecture support, the other about GGUF support. They are not 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": 5, + "component_count": 18, + "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-27088-5", + "nodes": [ + "issue:27088", + "issue:28936", + "issue:43834", + "issue:44365", + "issue:44370" + ], + "soft_pairs": [ + "issue:27088|issue:44370", + "issue:27088|issue:44365", + "issue:27088|issue:43834", + "issue:27088|issue:28936", + "issue:28936|issue:43834", + "issue:28936|issue:44365", + "issue:28936|issue:44370", + "issue:43834|issue:44370", + "issue:43834|issue:44365", + "issue:44365|issue:44370" + ], + "prepared_review_unit_hash": "3ede41fe204c778e76c6b85a718d198f529edc9792a3afdea77c837270da12ed", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 10, + "serialized_chars": 6443, + "estimated_input_tokens": 1611, + "estimated_eval_tokens": 3478 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 10, + "serialized_chars": 6443, + "estimated_input_tokens": 1611, + "estimated_eval_tokens": 3478 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster mixes two distinct language-specific translation requests (Turkish, Spanish) with three near-identical generic template issues. Only the three generic template issues look like true duplicates of the same underlying item.", + "confidence": 0.74, + "canonical_issue_reason": "Issue 43834 is the earliest of the three near-identical generic template issues ('[i18n-] Translating docs to '), so it is the best duplicate representative within that subgroup.", + "canonical_pr_reason": null, + "best_issue_reason": "43834 is the cleanest representative of the duplicate template subgroup and is more suitable as the cluster anchor than the language-specific requests, which are distinct translation targets.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:27088", + "right": "issue:44370", + "accept": false, + "reason": "Turkish-specific request vs generic placeholder; not the same concrete issue." + }, + { + "left": "issue:27088", + "right": "issue:44365", + "accept": false, + "reason": "Turkish-specific request vs generic placeholder; not the same concrete issue." + }, + { + "left": "issue:27088", + "right": "issue:43834", + "accept": false, + "reason": "Turkish-specific request vs generic placeholder; not the same concrete issue." + }, + { + "left": "issue:27088", + "right": "issue:28936", + "accept": false, + "reason": "Different target languages (Turkish vs Spanish), so these are distinct translation requests." + }, + { + "left": "issue:28936", + "right": "issue:43834", + "accept": false, + "reason": "Spanish-specific request vs generic placeholder; not the same concrete issue." + }, + { + "left": "issue:28936", + "right": "issue:44365", + "accept": false, + "reason": "Spanish-specific request vs generic placeholder; not the same concrete issue." + }, + { + "left": "issue:28936", + "right": "issue:44370", + "accept": false, + "reason": "Spanish-specific request vs generic placeholder; not the same concrete issue." + }, + { + "left": "issue:43834", + "right": "issue:44370", + "accept": true, + "reason": "Near-identical generic template issue; appears to be the same underlying request duplicated." + }, + { + "left": "issue:43834", + "right": "issue:44365", + "accept": true, + "reason": "Near-identical generic template issue; appears to be the same underlying request duplicated." + }, + { + "left": "issue:44365", + "right": "issue:44370", + "accept": true, + "reason": "Near-identical generic template issue; appears to be the same underlying request duplicated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet: the two language-specific issues are distinct requests, and the three generic-template issues are near-identical and plausibly duplicates of one another. The soft-edge verdicts are conservative, and the representative choice of 43834 among the generic subgroup is reasonable." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 6, + "component_count": 18, + "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-36817-16", + "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" + ], + "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: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:44121|issue:45322", + "issue:43837|issue:45027", + "issue:37015|issue:45027", + "issue:43837|issue:45177", + "issue:45322|issue:46133" + ], + "prepared_review_unit_hash": "490eb4cb209d028eeadc68d474a8dfad4eaedb59a352a5ee6ccb1b9fce042d92", + "component_budget": { + "node_count": 16, + "item_count": 16, + "soft_pair_count": 36, + "serialized_chars": 21658, + "estimated_input_tokens": 5415, + "estimated_eval_tokens": 11086 + }, + "budget": { + "node_count": 16, + "item_count": 16, + "soft_pair_count": 36, + "serialized_chars": 21658, + "estimated_input_tokens": 5415, + "estimated_eval_tokens": 11086 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is a loose set of model-request issues across different architectures and modalities. They share a broad \"add support for model X\" theme, but none of the pairs look like true duplicates of the same underlying bug/change.", + "confidence": 0.67, + "canonical_issue_reason": "issue:42738 is the broadest umbrella topic here ('BERT-like models with RoPE') and best represents the shared model-architecture-support theme, though it is still not a duplicate of the other specific model requests.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:42738 is the most general and reusable representative issue in the set; the rest are mostly specific requests for unrelated models or modalities.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:36817", + "right": "issue:41211", + "accept": false, + "reason": "Both are model-add requests, but they concern different models and different architectures; not the same code-path change." + }, + { + "left": "issue:36817", + "right": "issue:43671", + "accept": false, + "reason": "EuroBert config support and Qwen3-TTS support are unrelated model additions." + }, + { + "left": "issue:36817", + "right": "issue:36879", + "accept": false, + "reason": "Different model families entirely: EuroBert vs RF-DETR." + }, + { + "left": "issue:36879", + "right": "issue:41211", + "accept": false, + "reason": "RF-DETR and DEIMv2 are separate model integrations with different implementation needs." + }, + { + "left": "issue:36879", + "right": "issue:43909", + "accept": false, + "reason": "RF-DETR is an object-detection model request; LFM2.5 Audio is an audio model request." + }, + { + "left": "issue:36879", + "right": "issue:44121", + "accept": false, + "reason": "Different model families and likely different adapters/config paths." + }, + { + "left": "issue:36879", + "right": "issue:43837", + "accept": false, + "reason": "RF-DETR and Qwen3-ASR are unrelated capabilities and would not merge into one PR." + }, + { + "left": "issue:36879", + "right": "issue:43671", + "accept": false, + "reason": "Different Qwen3 TTS/ASR request vs RF-DETR object-detection support." + }, + { + "left": "issue:37015", + "right": "issue:43837", + "accept": false, + "reason": "NeoBERT and Qwen3-ASR are unrelated model additions." + }, + { + "left": "issue:37015", + "right": "issue:44121", + "accept": false, + "reason": "NeoBERT and circuit-sparsity/circuitgpt are different model families with distinct implementation work." + }, + { + "left": "issue:37015", + "right": "issue:43909", + "accept": false, + "reason": "NeoBERT and LFM2.5 Audio are not the same underlying change." + }, + { + "left": "issue:37015", + "right": "issue:42503", + "accept": false, + "reason": "NeoBERT and ModernVBERT are different models despite superficial BERT naming overlap." + }, + { + "left": "issue:41115", + "right": "issue:43837", + "accept": false, + "reason": "MiniCPM3 architecture support and Qwen3-ASR support are different model requests." + }, + { + "left": "issue:41115", + "right": "issue:44121", + "accept": false, + "reason": "MiniCPM3 and circuit-sparsity/circuitgpt are unrelated additions." + }, + { + "left": "issue:41115", + "right": "issue:42503", + "accept": false, + "reason": "MiniCPM3 and ModernVBERT target different model families." + }, + { + "left": "issue:41115", + "right": "issue:45027", + "accept": false, + "reason": "MiniCPM3 and Voxtral TTS support are different code paths and modalities." + }, + { + "left": "issue:41211", + "right": "issue:43671", + "accept": false, + "reason": "DEIMv2 and Qwen3-TTS are distinct model integrations." + }, + { + "left": "issue:42503", + "right": "issue:43837", + "accept": false, + "reason": "ModernVBERT and Qwen3-ASR are different model additions." + }, + { + "left": "issue:42503", + "right": "issue:43671", + "accept": false, + "reason": "ModernVBERT and Qwen3-TTS are unrelated model requests." + }, + { + "left": "issue:42503", + "right": "issue:42738", + "accept": false, + "reason": "ModernVBERT is a specific model request; the RoPE issue is a broad architectural topic, not the same change." + }, + { + "left": "issue:42738", + "right": "issue:45027", + "accept": false, + "reason": "Broad BERT-like/RoPE architecture support is unrelated to Voxtral TTS support." + }, + { + "left": "issue:43671", + "right": "issue:43837", + "accept": false, + "reason": "Qwen3-TTS and Qwen3-ASR are separate capabilities and likely separate model classes." + }, + { + "left": "issue:43671", + "right": "issue:45027", + "accept": false, + "reason": "Qwen3-TTS and Voxtral-4B-TTS are different TTS model requests, not the same underlying fix." + }, + { + "left": "issue:43671", + "right": "issue:45177", + "accept": false, + "reason": "Qwen3-TTS support and DeepSeek-OCR2 are unrelated model integrations." + }, + { + "left": "issue:43671", + "right": "issue:44121", + "accept": false, + "reason": "TTS support and weight-sparse transformer support are different changes." + }, + { + "left": "issue:43671", + "right": "issue:43909", + "accept": false, + "reason": "Qwen3-TTS and LFM2.5 Audio are both audio-related but not the same model or implementation path." + }, + { + "left": "issue:43837", + "right": "issue:43909", + "accept": false, + "reason": "Qwen3-ASR and LFM2.5 Audio are different audio model requests with different support needs." + }, + { + "left": "issue:43837", + "right": "issue:45322", + "accept": false, + "reason": "Qwen3-ASR and EUPE are unrelated model families and modalities." + }, + { + "left": "issue:43837", + "right": "issue:44121", + "accept": false, + "reason": "ASR support and weight-sparse transformer support are not the same underlying code change." + }, + { + "left": "issue:43909", + "right": "issue:44205", + "accept": false, + "reason": "LFM2.5 Audio and SAM3-LiteText target different modalities and implementations." + }, + { + "left": "issue:44121", + "right": "issue:45322", + "accept": false, + "reason": "Circuit-sparsity/circuitgpt and EUPE are different research models, not one duplicate request." + }, + { + "left": "issue:43837", + "right": "issue:45027", + "accept": false, + "reason": "Qwen3-ASR and Voxtral TTS are different audio models and would not be merged as one PR." + }, + { + "left": "issue:37015", + "right": "issue:45027", + "accept": false, + "reason": "NeoBERT and Voxtral-4B-TTS are unrelated model support requests." + }, + { + "left": "issue:43837", + "right": "issue:45177", + "accept": false, + "reason": "Qwen3-ASR and DeepSeek-OCR2 are distinct model additions." + }, + { + "left": "issue:45322", + "right": "issue:46133", + "accept": false, + "reason": "EUPE and TIPSv2 are both vision-language oriented, but they are different models and different architectural changes." + } + ] + }, + "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": 18, + "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-37934-10", + "nodes": [ + "issue:37934", + "issue:42936", + "issue:43240", + "issue:43668", + "issue:43717", + "issue:43813", + "issue:44031", + "issue:46018", + "issue:46143", + "issue:46144" + ], + "soft_pairs": [ + "issue:37934|issue:43668", + "issue:42936|issue:43240", + "issue:42936|issue:46018", + "issue:42936|issue:46144", + "issue:42936|issue:44031", + "issue:42936|issue:43813", + "issue:43240|issue:46144", + "issue:43240|issue:43668", + "issue:43668|issue:44031", + "issue:43668|issue:43813", + "issue:42936|issue:43668", + "issue:43717|issue:46144", + "issue:46143|issue:46144" + ], + "prepared_review_unit_hash": "9a7209a1b83271e9e20d0910d09b1cd2ebfcf19d74d16194f919779c203d524f", + "component_budget": { + "node_count": 10, + "item_count": 10, + "soft_pair_count": 13, + "serialized_chars": 10295, + "estimated_input_tokens": 2574, + "estimated_eval_tokens": 5404 + }, + "budget": { + "node_count": 10, + "item_count": 10, + "soft_pair_count": 13, + "serialized_chars": 10295, + "estimated_input_tokens": 2574, + "estimated_eval_tokens": 5404 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This is a loose similarity set with mostly unrelated issues. The only near-adjacent pair is the RoFormer kwargs/attention-interface reports, but they still look like different defects, so I would not merge any pair as duplicates.", + "confidence": 0.67, + "canonical_issue_reason": "issue:46143 is the closest thing to a cluster center: it describes a concrete, actionable RoFormer forwarding bug. The other items are either unrelated model/config issues or adjacent but distinct RoFormer concerns.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:46143 is the best representative issue overall because it is the clearest and broadest statement of the only recurring theme in the set (missing kwargs propagation).", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:37934", + "right": "issue:43668", + "accept": false, + "reason": "Different models and different bug classes: a possible norm-definition question vs an incorrect type hint." + }, + { + "left": "issue:42936", + "right": "issue:43240", + "accept": false, + "reason": "Both mention missing usage of a setting/kwargs, but one is about Mask2Former ignore_value and the other about loss-function kwargs forwarding; not the same bug." + }, + { + "left": "issue:42936", + "right": "issue:46018", + "accept": false, + "reason": "Mask2Former ignore_value and DeepSeek-V4 shared expert gating are unrelated model behaviors." + }, + { + "left": "issue:42936", + "right": "issue:46144", + "accept": false, + "reason": "Different subsystems and different symptoms; ignore_value handling is unrelated to RoFormer attention interface usage." + }, + { + "left": "issue:42936", + "right": "issue:44031", + "accept": false, + "reason": "Mask2Former ignore_value and tokenizer regex warnings are unrelated." + }, + { + "left": "issue:42936", + "right": "issue:43813", + "accept": false, + "reason": "Unrelated: ignore_value handling vs a typo in quantization-operation naming." + }, + { + "left": "issue:43240", + "right": "issue:46144", + "accept": false, + "reason": "Both involve kwargs, but one is about loss calculation forwarding and the other is about RoFormer attention architecture; not the same concrete bug." + }, + { + "left": "issue:43240", + "right": "issue:43668", + "accept": false, + "reason": "Loss kwargs forwarding is unrelated to a ModernBERT config type-hint issue." + }, + { + "left": "issue:43668", + "right": "issue:44031", + "accept": false, + "reason": "Incorrect type hint in ModernBERTConfig is unrelated to tokenizer regex warnings." + }, + { + "left": "issue:43668", + "right": "issue:43813", + "accept": false, + "reason": "Different modules and bug types: type-hint error vs typo in a conversion attribute name." + }, + { + "left": "issue:42936", + "right": "issue:43668", + "accept": false, + "reason": "No shared underlying defect; these are distinct correctness/documentation issues." + }, + { + "left": "issue:43717", + "right": "issue:46144", + "accept": false, + "reason": "Mamba init_weights usage and RoFormer attention-interface usage are unrelated architecture issues." + }, + { + "left": "issue:46143", + "right": "issue:46144", + "accept": false, + "reason": "Both are RoFormer-related, but one is about missing kwargs propagation through methods and the other about not using the attention interface; overlapping area, different underlying defect." + } + ] + }, + "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": 18, + "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": "6c8cfeb31812421cbf4ba4606606b49b7c59067aabbc1c1c0fe2924892d780cb", + "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 do not look like duplicates: one is about block decoding, the other is a request for DeepSeek v3.2 support. They appear unrelated, so the soft edge should be rejected.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43037 is the better standalone representative because it is a specific feature request; issue 41084 is terse and appears to target a different topic entirely.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:41084", + "right": "issue:43037", + "accept": false, + "reason": "Different underlying topics: block decoding vs model/version support. No shared issue target or code-path indication." + } + ] + }, + "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": 18, + "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": "fd979d2cf4c4cfbbe62b4f3504812e3ac3067435ea07c930ac5c51e8cb4fc855", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1583, + "estimated_input_tokens": 396, + "estimated_eval_tokens": 1048 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1583, + "estimated_input_tokens": 396, + "estimated_eval_tokens": 1048 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues are only loosely related through custom trainers, but they describe different problems in different Trainer code paths: column pruning customizability vs FLOPS estimation.", + "confidence": 0.95, + "canonical_issue_reason": "issue #43598 is the broader, more central Trainer-behavior concern; it reads like the more general customization problem, while #43600 is a narrower FLOPS-estimation follow-up.", + "canonical_pr_reason": null, + "best_issue_reason": "#43598 is the better representative issue because it captures the broader API/customization theme and is more likely to subsume adjacent Trainer customizability concerns.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43598", + "right": "issue:43600", + "accept": false, + "reason": "Reject: they share the idea of 'custom trainers' but not the same underlying bug. #43598 is about `remove_unused_column` behavior; #43600 is about FLOPS estimation logic for custom trainers. 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": 10, + "component_count": 18, + "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": "ea3b9f4900ebe61695c95359754f127d983f84d7ffbd26a000daa25f88de8276", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1612, + "estimated_input_tokens": 403, + "estimated_eval_tokens": 1062 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1612, + "estimated_input_tokens": 403, + "estimated_eval_tokens": 1062 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are distinct feature requests: one asks for Qwen3-Omni auto-class registration, while the other asks for AutoModelForSequenceClassification support for Qwen3.5. They target different model families and different auto classes, so they are not duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "issue:43684 is the broader and cleaner representative of the cluster, but it is still a separate request from the Qwen3.5 sequence-classification support issue.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43684", + "right": "issue:44405", + "accept": false, + "reason": "Different underlying changes: Qwen3-Omni registration for AutoModel/AutoModelForConditionalGeneration versus Qwen3.5 sequence-classification support. They do not describe the same bug or fix." + } + ] + }, + "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": 18, + "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": "2ad89ded5360b1c2d8d0eb735b7e66209ea9bf17d2dcf8c6d4e0fca5e65e2697", + "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": "Issues 43845, 43846, and 43859 share the exact same title text and have similar low-activity metadata, so they form the strongest duplicate cluster visible in the packet. Issue 43835 has a different title and weaker similarity signals, so it does not match the cluster as well.", + "confidence": 0.92, + "canonical_issue_reason": "43845 is the earliest of the three identical-title issues and is the cleanest representative based on title match, similar body length/activity, and timestamps.", + "canonical_pr_reason": null, + "best_issue_reason": "43845 is the best global issue candidate because it anchors the only clearly repeated title group and predates the other matching issues.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43835", + "right": "issue:43845", + "accept": false, + "reason": "Different titles and no stronger shared signals beyond soft similarity." + }, + { + "left": "issue:43835", + "right": "issue:43846", + "accept": false, + "reason": "Different titles and no shared targets, filenames, or other concrete match signals." + }, + { + "left": "issue:43835", + "right": "issue:43859", + "accept": false, + "reason": "Different titles; the visible metadata does not show a same-issue match." + }, + { + "left": "issue:43845", + "right": "issue:43846", + "accept": true, + "reason": "Exact same title text with close body length and activity/timestamp patterns; this is a plausible duplicate pair." + }, + { + "left": "issue:43845", + "right": "issue:43859", + "accept": true, + "reason": "Exact same title text and similar lightweight metadata; they look like the same report cluster." + }, + { + "left": "issue:43846", + "right": "issue:43859", + "accept": true, + "reason": "Exact same title text with similar metadata signals, supporting a duplicate relationship." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The cluster call is a bit too strong for the packet. The data only shows identical titles for 43845/43846/43859 plus soft-similarity scores; it does not support calling them \"near-identical placeholders\" or asserting any shared underlying content beyond title similarity. Likewise, 43835 can be treated as a weaker/non-matching candidate because its title differs, but \"likely unrelated content\" is not grounded in the packet. Please keep the wording limited to the visible signals (title similarity, body length, activity, timestamps) and avoid content claims." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 12, + "component_count": 18, + "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": "701141eb8dc963e2a11d463685997aafff1f540690157089b490c12463667893", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 9, + "serialized_chars": 5788, + "estimated_input_tokens": 1447, + "estimated_eval_tokens": 3150 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 9, + "serialized_chars": 5788, + "estimated_input_tokens": 1447, + "estimated_eval_tokens": 3150 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All items look like unrelated low-signal or spammy issues; there is no evidence they describe the same underlying bug. If one must be kept as the most plausible real issue, #44418 is the least obviously spam-like.", + "confidence": 0.9, + "canonical_issue_reason": "#44418 has the only non-spam, issue-shaped title (\u201cDocumentation Enhancement Suggestion\u201d), so it is the best candidate to anchor the cluster even though the body is still too sparse to confirm legitimacy.", + "canonical_pr_reason": null, + "best_issue_reason": "#44418 is the most suitable global representative because it is the only item that reads like a plausible product issue rather than obvious spam or filler text.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44371", + "right": "issue:44572", + "accept": false, + "reason": "Both are titled only \"\" with no body, targets, or links; identical spam labels do not indicate the same underlying issue." + }, + { + "left": "issue:44371", + "right": "issue:44573", + "accept": false, + "reason": "Both are titled only \"\" and contain no actionable evidence of a shared bug or change." + }, + { + "left": "issue:44371", + "right": "issue:44418", + "accept": false, + "reason": "One is generic spam, the other is a vague documentation request; they do not describe the same concrete issue." + }, + { + "left": "issue:44371", + "right": "issue:44937", + "accept": false, + "reason": "\"\" and \"Check out \\\"Google LLC\\\"\" are both spam-like, but there is no sign they refer to the same underlying problem." + }, + { + "left": "issue:44418", + "right": "issue:44572", + "accept": false, + "reason": "A generic documentation enhancement and a spam placeholder are not the same issue." + }, + { + "left": "issue:44418", + "right": "issue:44573", + "accept": false, + "reason": "The titles and available metadata do not point to the same bug/change; one is a vague doc suggestion, the other is spam." + }, + { + "left": "issue:44572", + "right": "issue:44573", + "accept": false, + "reason": "Both are spam placeholders, but there is no concrete shared bug, change, or target to merge them as duplicates." + }, + { + "left": "issue:44572", + "right": "issue:44937", + "accept": false, + "reason": "Spam placeholder vs. spammy brand mention; similar low quality is not enough to treat them as the same issue." + }, + { + "left": "issue:44573", + "right": "issue:44937", + "accept": false, + "reason": "No shared code-path, target, or substantive description is present; these are just separate spam-like issues." + } + ] + }, + "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": 18, + "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": "7dc1f8a861ea4de3f0b55ef1ca251eee7fc0c8a8260a01f567af86056fa8e4dc", + "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 pair looks like a likely duplicate/near-duplicate issue set: the titles are identical and the body lengths are very close, but the evidence is sparse, so confidence should stay moderate. Issue 45145 is the better canonical representative because it was created first and has the only inbound reference in the pair.", + "confidence": 0.78, + "canonical_issue_reason": "Issue 45145 is the best canonical representative because it was created first and has the only inbound reference among these two near-duplicate issues.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45145 is the strongest overall issue candidate for the same reason: earliest creation time and slightly more external reference activity.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45145", + "right": "issue:45151", + "accept": true, + "reason": "Same exact title and very similar body size make these likely duplicate/near-duplicate reports, though the available evidence is limited and does not prove identical content." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The comparison is plausible, but the verdict is a bit too confident for the evidence shown. The packet only provides identical titles, similar body lengths, and a soft_similarity signal; it does not give direct body-content evidence or explicit duplicate links. Please soften the language from \"same underlying issue\" to \"likely duplicate/near-duplicate,\" and reduce the confidence accordingly. The canonical choice is grounded in earlier creation time and one inbound reference, but \"more established in discussion/links\" is slightly overstated given the sparse activity." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 14, + "component_count": 18, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45295-2", + "nodes": [ + "issue:45295", + "issue:45373" + ], + "soft_pairs": [], + "prepared_review_unit_hash": "f5fd95fada2f530ab0999e343c17c16266cde0c7182bb22509972ca44827cd49", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 0, + "serialized_chars": 1286, + "estimated_input_tokens": 322, + "estimated_eval_tokens": 900 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 0, + "serialized_chars": 1286, + "estimated_input_tokens": 322, + "estimated_eval_tokens": 900 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues describe the same missing Gemma 4 sequence-classification support; the later, more specific open issue is the better cluster representative.", + "confidence": 0.94, + "canonical_issue_reason": "issue #45373 is the clearest canonical tracker: it is open, explicitly names the missing class `Gemma4ForSequenceClassification`, and states the module-level gap relative to Gemma 2/3.", + "canonical_pr_reason": null, + "best_issue_reason": "issue #45373 best captures the concrete request and likely remaining work, while #45295 is a shorter earlier duplicate with less detail and is already closed.", + "best_pr_reason": null, + "soft_edge_verdicts": [] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 15, + "component_count": 18, + "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": "b9e7be76507fc1b84bb5e45f31bca4829ea9965b0a67e1bdaeb27dc5e9f8e23d", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1668, + "estimated_input_tokens": 417, + "estimated_eval_tokens": 1090 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1668, + "estimated_input_tokens": 417, + "estimated_eval_tokens": 1090 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are two separate 4-bit inference bugs: one is a Gemma4 audio-module uint8/torch.finfo TypeError, the other is a BitsAndBytesConfig lm_head exclusion regression causing an AssertionError. They are related only at a broad quantization-inference level, not the same underlying defect.", + "confidence": 0.97, + "canonical_issue_reason": "issue #45674 is the better representative of the cluster because it is framed as a concrete configuration/regression bug in BitsAndBytesConfig with a clear failure mode and broader applicability than the Gemma4-specific report.", + "canonical_pr_reason": null, + "best_issue_reason": "issue #45674 is the strongest standalone issue to keep because it describes a specific, actionable regression with an explicit API trigger and a clear assertion failure path.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45672", + "right": "issue:45674", + "accept": false, + "reason": "Different underlying problems: Gemma4 audio modules hit a torch.finfo TypeError on uint8 weights, while the BitsAndBytesConfig issue is about llm_int8_skip_modules clearing the default lm_head exclusion and triggering an AssertionError. Same broad area, but not the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 16, + "component_count": 18, + "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": "630b139043b751d835d2c24ebb7e29ca8e6af3ecd21166c2503654b065d68c82", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1477, + "estimated_input_tokens": 370, + "estimated_eval_tokens": 996 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1477, + "estimated_input_tokens": 370, + "estimated_eval_tokens": 996 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two closed issues are not duplicates: their titles are different personal-name/date entries with no shared problem statement, code path, or references. The soft similarity appears superficial only.", + "confidence": 0.98, + "canonical_issue_reason": "No canonical issue here; the items look unrelated rather than like two reports of the same underlying bug/request.", + "canonical_pr_reason": null, + "best_issue_reason": "Neither issue is a good global representative for a duplicate cluster because there is no real cluster to collapse.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45762", + "right": "issue:45763", + "accept": false, + "reason": "Different names and dates, no body, links, or shared context 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": 17, + "component_count": 18, + "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": "90279dcfb8e03d81bff1fb84dcac3dbf0c35aafc2c63ab347f1377d072d3d6b6", + "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 look like unrelated low-signal entries (personal names plus dates) with no shared technical context, 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": "issue:45764", + "right": "issue:45765", + "accept": false, + "reason": "Titles do not describe the same bug or request; they are distinct name/date strings with no body, links, or code context indicating a shared 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": 18, + "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": "3a29482d2f409117b5b66dca55d74aaef52b17948db4a666e59a4dfb5bfb7ced", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 8, + "serialized_chars": 5726, + "estimated_input_tokens": 1432, + "estimated_eval_tokens": 3120 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 8, + "serialized_chars": 5726, + "estimated_input_tokens": 1432, + "estimated_eval_tokens": 3120 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "There are two exact duplicate issue pairs in the cluster: 45870/45871 and 45872/45873. Issue 45869 is related only at a broad financial-data level, but it is a distinct request. The two DataForge issues describe different underlying asks, so they should not be merged together.", + "confidence": 0.84, + "canonical_issue_reason": "45872 is the strongest representative of the cluster: it is the earlier copy of an exact duplicate pair (45872/45873) and has the most descriptive body/title among the clearly duplicated issues.", + "canonical_pr_reason": null, + "best_issue_reason": "45872 is the best global representative because it is one of the exact-duplicate heads and appears more fully specified than the other duplicate pair.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45869", + "right": "issue:45870", + "accept": false, + "reason": "Related finance/DataForge theme, but the dataset bundle request and the prediction example are different underlying issues." + }, + { + "left": "issue:45869", + "right": "issue:45871", + "accept": false, + "reason": "Broad topical overlap only; this is a dataset bundle request, not the same change/request as the prediction example." + }, + { + "left": "issue:45870", + "right": "issue:45871", + "accept": true, + "reason": "Same title and apparently the same issue text; this looks like an exact duplicate pair." + }, + { + "left": "issue:45870", + "right": "issue:45872", + "accept": false, + "reason": "Both mention financial time series in DataForge, but one is an example/prediction request and the other is a scraping-service request." + }, + { + "left": "issue:45870", + "right": "issue:45873", + "accept": false, + "reason": "Same broad domain, but the concrete request differs from the prediction-example issue, so they are not the same underlying issue." + }, + { + "left": "issue:45871", + "right": "issue:45872", + "accept": false, + "reason": "The titles point to different concrete asks: prediction example versus scraping service." + }, + { + "left": "issue:45871", + "right": "issue:45873", + "accept": false, + "reason": "Related only at a high level; these are not the same underlying issue." + }, + { + "left": "issue:45872", + "right": "issue:45873", + "accept": true, + "reason": "Same title and matching issue shape; this appears to be an exact duplicate pair." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The cluster summary is grounded in the packet: 45870/45871 and 45872/45873 are exact duplicate pairs by identical title/body length and perfect Jaccard, while 45869 is only loosely related and distinct. The representative choice is conservative and supported by the longer body on 45872. No PR mergeability issue applies here because this is an issue-only cluster." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 27, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-39895-14", + "nodes": [ + "pull_request:39895", + "pull_request:41855", + "pull_request:42668", + "pull_request:43251", + "pull_request:43254", + "pull_request:43424", + "pull_request:43823", + "pull_request:43995", + "pull_request:44827", + "pull_request:45170", + "pull_request:45213", + "pull_request:45655", + "pull_request:46048", + "pull_request:46067" + ], + "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:46067", + "pull_request:43251|pull_request:43823", + "pull_request:43251|pull_request:44827", + "pull_request:43424|pull_request:46067", + "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:43823|pull_request:46067", + "pull_request:43995|pull_request:44827", + "pull_request:43995|pull_request:45213", + "pull_request:44827|pull_request:46067", + "pull_request:44827|pull_request:45170", + "pull_request:44827|pull_request:46048", + "pull_request:44827|pull_request:45655", + "pull_request:44827|pull_request:45213", + "pull_request:43823|pull_request:45213", + "pull_request:43251|pull_request:43254" + ], + "prepared_review_unit_hash": "b4a06a54dfcb41d8099445a4a68063d95fd60a1ec466262d7219639c15ad8285", + "component_budget": { + "node_count": 14, + "item_count": 14, + "soft_pair_count": 22, + "serialized_chars": 38370, + "estimated_input_tokens": 9593, + "estimated_eval_tokens": 19442 + }, + "budget": { + "node_count": 14, + "item_count": 14, + "soft_pair_count": 22, + "serialized_chars": 38370, + "estimated_input_tokens": 9593, + "estimated_eval_tokens": 19442 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Mostly heterogeneous PRs across unrelated models, docs, and refactors. The only clear duplicate/overlap is the pair of loss_utils.py changes for fixed_cross_entropy kwargs; everything else is unrelated and should stay separate.", + "confidence": 0.86, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 43251 is the clearest representative of the fixed_cross_entropy kwargs fix: it has the explicit issue target, broader review activity, and the same code-path change as 43254.", + "best_issue_reason": null, + "best_pr_reason": "PR 43251 is the strongest standalone PR in the cluster because it directly addresses a concrete bug report with explicit tracking, while the rest are mostly unrelated feature additions or refactors.", + "soft_edge_verdicts": [ + { + "left": "pull_request:39895", + "right": "pull_request:43424", + "accept": false, + "reason": "Different features in unrelated areas: VideoPrism model addition vs Executorch dynamic-shape test." + }, + { + "left": "pull_request:41855", + "right": "pull_request:43823", + "accept": false, + "reason": "Mistral tokenizer method additions and MobileLLM model addition are unrelated changes." + }, + { + "left": "pull_request:41855", + "right": "pull_request:43424", + "accept": false, + "reason": "Tokenizer API patch vs Executorch export test; no shared bug or code path." + }, + { + "left": "pull_request:42668", + "right": "pull_request:43823", + "accept": false, + "reason": "Processor robustness changes for existing models vs adding MobileLLM; not the same underlying fix." + }, + { + "left": "pull_request:43251", + "right": "pull_request:46067", + "accept": false, + "reason": "fixed_cross_entropy kwargs bug fix is unrelated to base_model_prefix conversion-loading changes." + }, + { + "left": "pull_request:43251", + "right": "pull_request:43823", + "accept": false, + "reason": "Loss-function fix and new model implementation are unrelated." + }, + { + "left": "pull_request:43251", + "right": "pull_request:44827", + "accept": false, + "reason": "Loss-function kwargs support and Mistral4 test fixes do not share the same underlying bug." + }, + { + "left": "pull_request:43251", + "right": "pull_request:46048", + "accept": false, + "reason": "Cross-entropy kwargs fix and TDT loss kernel integration are different changes." + }, + { + "left": "pull_request:43424", + "right": "pull_request:46067", + "accept": false, + "reason": "Executorch dynamic-shape testing is unrelated to conversion mapping/base_model_prefix handling." + }, + { + "left": "pull_request:43424", + "right": "pull_request:44827", + "accept": false, + "reason": "Executorch export tests and Mistral4 test fixes are unrelated." + }, + { + "left": "pull_request:43424", + "right": "pull_request:43823", + "accept": false, + "reason": "Exportability test additions and MobileLLM model addition are unrelated." + }, + { + "left": "pull_request:43823", + "right": "pull_request:45655", + "accept": false, + "reason": "New MobileLLM model code and config-resolution ordering fix address different problems." + }, + { + "left": "pull_request:43823", + "right": "pull_request:43995", + "accept": false, + "reason": "Adding MobileLLM and refactoring Falcon output collection are unrelated model changes." + }, + { + "left": "pull_request:43823", + "right": "pull_request:46067", + "accept": false, + "reason": "New model addition and conversion scope/base_model_prefix fix are unrelated." + }, + { + "left": "pull_request:43995", + "right": "pull_request:44827", + "accept": false, + "reason": "Falcon output-interface refactor and Mistral4 test fixes are different code paths." + }, + { + "left": "pull_request:43995", + "right": "pull_request:45213", + "accept": false, + "reason": "Falcon refactor and model-creation skill/docs PR are unrelated." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46067", + "accept": false, + "reason": "Mistral4 test fixes and conversion scoping/base_model_prefix handling are unrelated." + }, + { + "left": "pull_request:44827", + "right": "pull_request:45170", + "accept": false, + "reason": "Mistral4 test fixes and layrnorm-to-layernorm rename are different fixes." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46048", + "accept": false, + "reason": "Mistral4 tests and TDT loss kernel integration do not target the same issue." + }, + { + "left": "pull_request:44827", + "right": "pull_request:45655", + "accept": false, + "reason": "Mistral4 test changes and cls.config resolution ordering are unrelated." + }, + { + "left": "pull_request:44827", + "right": "pull_request:45213", + "accept": false, + "reason": "Mistral4 test fixes and model-creation skill/docs are unrelated." + }, + { + "left": "pull_request:43823", + "right": "pull_request:45213", + "accept": false, + "reason": "MobileLLM model addition and model-creation skill/docs are not the same change." + }, + { + "left": "pull_request:43251", + "right": "pull_request:43254", + "accept": true, + "reason": "Both PRs change the same loss_utils.py code path to add cross_entropy kwargs support, and they look like overlapping implementations of the same fix." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative: the cluster is mostly unrelated model additions, docs, and refactors, and the only clear overlap is the 43251/43254 fixed_cross_entropy pair, which share the same file, function, and issue target. The mergeability judgment for that pair is reasonable. No evident overstatement." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 2, + "component_count": 27, + "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": "6d9c1f19b2a27c051b36b6d8be08195a358cd53f156bd124d653f952561b0026", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 7225, + "estimated_input_tokens": 1807, + "estimated_eval_tokens": 3870 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 7225, + "estimated_input_tokens": 1807, + "estimated_eval_tokens": 3870 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two PRs for the same model family, but they target different deliverables: initial VibeVoice support vs. a later VibeVoice Realtime extension. Shared files mostly reflect the same subsystem and not the same concrete change.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #40546 is the more complete, non-draft baseline implementation of VibeVoice and is better suited as the representative PR for this cluster.", + "best_issue_reason": null, + "best_pr_reason": "PR #40546 is the strongest standalone representative: it introduces the core model support, has substantial review activity, and is broader/more central than the draft realtime follow-up.", + "soft_edge_verdicts": [ + { + "left": "pull_request:40546", + "right": "pull_request:42781", + "accept": false, + "reason": "Reject: #40546 adds core VibeVoice support, while #42781 adds VibeVoice Realtime and related acoustic-tokenizer/generation work. They share model infrastructure but are not the same concrete fix/change and would not plausibly be merged as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 3, + "component_count": 27, + "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": "5a44b4713bf3951383f74da078c4fd7113480be3b331c9c972d1ac9ea0ad9225", + "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 on unrelated model families and implement different features, so they should not be clustered as duplicates.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: one adds TimesFM covariate forecasting, the other adds DINOv3 image classification support; they do not fix the same underlying problem.", + "best_issue_reason": null, + "best_pr_reason": "PR #41224 is the cleaner representative of its own change because it is a focused, self-contained model support addition with clear model/code/test updates.", + "soft_edge_verdicts": [ + { + "left": "pull_request:40755", + "right": "pull_request:41224", + "accept": false, + "reason": "Different model domains and different code paths: TimesFM forecasting with covariates versus DINOv3 image classification support. They are not the same underlying bug or change and would not be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 4, + "component_count": 27, + "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": 27, + "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-41797-3", + "nodes": [ + "pull_request:41797", + "pull_request:45186", + "pull_request:45630" + ], + "soft_pairs": [ + "pull_request:41797|pull_request:45186", + "pull_request:45186|pull_request:45630" + ], + "prepared_review_unit_hash": "b6c7cec47605b04117cdc75d5581e0ed27ff81dcaf43c9c8d35076d2e69a1df9", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 10436, + "estimated_input_tokens": 2609, + "estimated_eval_tokens": 5474 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 10436, + "estimated_input_tokens": 2609, + "estimated_eval_tokens": 5474 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are three separate model-integration PRs (DeepSeek OCR, Isaac, and Kimi2-6). They share Transformers boilerplate and auto-mapping updates, but they target different model families and code paths, so they are not duplicates.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "Kimi2-6 is the broadest and most actively discussed PR in the set, touching the most integration surfaces and likely serving best as the representative item for this cluster.", + "best_issue_reason": null, + "best_pr_reason": "Kimi2-6 is the strongest representative PR overall because it has the highest discussion activity and the widest integration footprint; the other two are narrower, separate model additions.", + "soft_edge_verdicts": [ + { + "left": "pull_request:41797", + "right": "pull_request:45186", + "accept": false, + "reason": "DeepSeek OCR and Isaac are different model additions with different filenames, configs, and model code; the overlap is only shared Transformers scaffolding and docs, not the same underlying change." + }, + { + "left": "pull_request:45186", + "right": "pull_request:45630", + "accept": false, + "reason": "Isaac and Kimi2-6 integrate different models and touch different concrete implementations; shared auto-mapping and docs updates are routine for new-model PRs, not evidence of a duplicate 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": 6, + "component_count": 27, + "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": "0bbf1410d99b72a178a35dbc74ca234e253f82ebc8b8d4b6ce6e6dc1223ce5f4", + "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 separate GitHub Actions maintenance PRs: one is a broad Node 24 compatibility upgrade for several workflow actions, the other is a docker/build-push-action version bump. They overlap in workflows but do not fix the same underlying problem.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 42975 is the better canonical representative: it has a clearer, broader goal (Node 24 compatibility) and touches the major GitHub Actions version updates across workflows.", + "best_issue_reason": null, + "best_pr_reason": "PR 42975 is the best overall representative of the cluster because its title and diff describe a single, concrete maintenance objective, while PR 42976 is a narrower, separate action-version bump.", + "soft_edge_verdicts": [ + { + "left": "pull_request:42975", + "right": "pull_request:42976", + "accept": false, + "reason": "Do not merge: 42975 upgrades several GitHub Actions to Node 24-compatible versions; 42976 only bumps docker/build-push-action from v5 to v6. They are related maintenance changes but not the same underlying fix and would not naturally be one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 7, + "component_count": 27, + "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-42978-5", + "nodes": [ + "pull_request:42978", + "pull_request:43448", + "pull_request:43451", + "pull_request:45490", + "pull_request:46162" + ], + "soft_pairs": [ + "pull_request:42978|pull_request:43448", + "pull_request:42978|pull_request:43451", + "pull_request:42978|pull_request:45490", + "pull_request:43448|pull_request:43451", + "pull_request:43448|pull_request:45490", + "pull_request:43451|pull_request:45490", + "pull_request:45490|pull_request:46162" + ], + "prepared_review_unit_hash": "ad136c739266ae3b104edcfbf2daa78607ed70c2efb480ce79c5cfca94be0d06", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 7, + "serialized_chars": 16484, + "estimated_input_tokens": 4121, + "estimated_eval_tokens": 8498 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 7, + "serialized_chars": 16484, + "estimated_input_tokens": 4121, + "estimated_eval_tokens": 8498 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are separate model-addition or test-only PRs with only superficial overlap in docs and auto-mapping files. The shared filenames do not indicate the same underlying bug or change, so this does not form a duplicate cluster.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "No PR here is a true duplicate of the others; if one must be picked as the strongest standalone representative, PR 43451 is the most substantial and complete model-addition PR, but it is still distinct from the rest.", + "soft_edge_verdicts": [ + { + "left": "pull_request:42978", + "right": "pull_request:43448", + "accept": false, + "reason": "ViT NEPA vs Molmo are different model additions; overlap is limited to shared docs/auto-registration files, not the same change." + }, + { + "left": "pull_request:42978", + "right": "pull_request:43451", + "accept": false, + "reason": "ViT NEPA and Molmo2 are unrelated model integrations with different codepaths and artifacts." + }, + { + "left": "pull_request:42978", + "right": "pull_request:45490", + "accept": false, + "reason": "ViT NEPA and CTSM are distinct model additions; the common files are generic registry/docs updates only." + }, + { + "left": "pull_request:43448", + "right": "pull_request:43451", + "accept": false, + "reason": "Molmo and Molmo2 are related by name but are separate model families with different implementations and files." + }, + { + "left": "pull_request:43448", + "right": "pull_request:45490", + "accept": false, + "reason": "Molmo and CTSM are unrelated model additions; shared auto/docs files are insufficient for duplicate triage." + }, + { + "left": "pull_request:43451", + "right": "pull_request:45490", + "accept": false, + "reason": "Molmo2 and CTSM target different models and codepaths; this is not the same concrete change." + }, + { + "left": "pull_request:45490", + "right": "pull_request:46162", + "accept": false, + "reason": "CTSM model addition vs a small XPU test annotation tweak are entirely different changes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 8, + "component_count": 27, + "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-43291-5", + "nodes": [ + "pull_request:43291", + "pull_request:43665", + "pull_request:44815", + "pull_request:46102", + "pull_request:46161" + ], + "soft_pairs": [ + "pull_request:43291|pull_request:43665", + "pull_request:43291|pull_request:44815", + "pull_request:43665|pull_request:46102", + "pull_request:43665|pull_request:44815", + "pull_request:43665|pull_request:46161", + "pull_request:44815|pull_request:46161" + ], + "prepared_review_unit_hash": "1db9eaad1d51f9ce6fc5a5661285912de85f924c3e37f3c5ddfc7ee6a07c639f", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 6, + "serialized_chars": 12045, + "estimated_input_tokens": 3012, + "estimated_eval_tokens": 6280 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 6, + "serialized_chars": 12045, + "estimated_input_tokens": 3012, + "estimated_eval_tokens": 6280 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs do not form a duplicate set: they span unrelated areas (Whisper tokenization, CLIP/ViT tests, FP8 dequant/loading, FSDP tied weights, and processor subfolder fallback). Even the one soft-similar pair is not the same concrete fix.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46161 is the clearest standalone bug fix: it targets one concrete processor/tokenizer fallback path and explicitly links to an issue.", + "best_issue_reason": null, + "best_pr_reason": "PR 46161 is the best representative candidate because it is narrowly scoped, traceable to an explicit issue, and describes a single code-path fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43291", + "right": "pull_request:43665", + "accept": false, + "reason": "Whisper tokenizer decode logic vs unrelated CLIP/ViT test edits; different subsystems and different change types." + }, + { + "left": "pull_request:43291", + "right": "pull_request:44815", + "accept": false, + "reason": "Whisper decode/batch handling is unrelated to FP8 dequant/model-loading changes in a different subsystem." + }, + { + "left": "pull_request:43665", + "right": "pull_request:46102", + "accept": false, + "reason": "Test-only CLIP/ViT edits are not the same bug as FSDP tied-weights cleanup/fix." + }, + { + "left": "pull_request:43665", + "right": "pull_request:44815", + "accept": false, + "reason": "One is test scaffolding in vision models; the other is FP8/loading infrastructure. No shared underlying defect." + }, + { + "left": "pull_request:43665", + "right": "pull_request:46161", + "accept": false, + "reason": "Placeholder test changes are unrelated to tokenizer subfolder fallback behavior." + }, + { + "left": "pull_request:44815", + "right": "pull_request:46161", + "accept": false, + "reason": "Both touch tokenizer-adjacent code, but one is FP8/dequant plus loading internals while the other is processor/tokenizer subfolder fallback; 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": 9, + "component_count": 27, + "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-43488-5", + "nodes": [ + "pull_request:43488", + "pull_request:43532", + "pull_request:45932", + "pull_request:46100", + "pull_request:46114" + ], + "soft_pairs": [ + "pull_request:43488|pull_request:46114", + "pull_request:43488|pull_request:43532", + "pull_request:43532|pull_request:46114", + "pull_request:45932|pull_request:46100", + "pull_request:46100|pull_request:46114" + ], + "prepared_review_unit_hash": "472b5bce35f57f33724ff8ec20035891324f8aefe14203eaca33e079d751ee46", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 5, + "serialized_chars": 11343, + "estimated_input_tokens": 2836, + "estimated_eval_tokens": 5928 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 5, + "serialized_chars": 11343, + "estimated_input_tokens": 2836, + "estimated_eval_tokens": 5928 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is mostly a set of unrelated \"do not merge\" test PRs; the only clear duplicate pair is the two identical ViT test-migration PRs.", + "confidence": 0.88, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46100 is the cleanest canonical representative of the exact duplicate pair: it has the simpler title and the same minimal ViT test-file diff as 46114.", + "best_issue_reason": null, + "best_pr_reason": "PR 46100 is the best overall representative because it is the clearest self-contained duplicate target; 46114 is just a forked copy of the same change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43488", + "right": "pull_request:46114", + "accept": false, + "reason": "Different changes: 43488 is a setup/dependency/version-bump repo-bot check, while 46114 is a ViT test migration clone. Shared test-file presence is not enough." + }, + { + "left": "pull_request:43488", + "right": "pull_request:43532", + "accept": false, + "reason": "Both are test/CI experiment PRs, but they touch different code paths and files: packaging/deps versus workflow/comment handling and utilities." + }, + { + "left": "pull_request:43532", + "right": "pull_request:46114", + "accept": false, + "reason": "No same underlying bug/change; 43532 is workflow/utility plumbing, while 46114 is only a duplicated ViT test edit." + }, + { + "left": "pull_request:45932", + "right": "pull_request:46100", + "accept": false, + "reason": "Too broad and not the same fix: 45932 changes multiple model configs/workflows for grafana/telemetry testing, while 46100 only adds failing tests in the ViT test file." + }, + { + "left": "pull_request:46100", + "right": "pull_request:46114", + "accept": true, + "reason": "Exact duplicate: both modify the same ViT test file with the same inserted failing tests, and the titles indicate the same test-migration intent." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The only strong duplicate is 46100/46114: they have identical diffs to the same ViT test file. The negative pair verdicts are also appropriately cautious, since the other PRs touch different areas or broader changes despite some superficial test/CI similarity." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 10, + "component_count": 27, + "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": "8fd298253a579226270e6957ba6a0447af626e0c3597b2347835a320cd5957a3", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6082, + "estimated_input_tokens": 1521, + "estimated_eval_tokens": 3298 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6082, + "estimated_input_tokens": 1521, + "estimated_eval_tokens": 3298 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related but distinct PRs: one adds Promptable Concept Segmentation, the other Promptable Visual Segmentation. They share documentation and pipeline plumbing, but they target different features and code paths, so they should not be treated as duplicates.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 43613 is the broader and more feature-complete segmentation PR, with wider model/test coverage, so it is the better cluster representative if one must be chosen.", + "best_issue_reason": null, + "best_pr_reason": "PR 43613 is the stronger representative of the pair because it covers more integrations and tests, but it is still a separate feature from 43612.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43612", + "right": "pull_request:43613", + "accept": false, + "reason": "Reject: the titles and changed files show different features\u2014promptable concept segmentation vs promptable visual segmentation\u2014with separate pipeline modules and model/tests, so they are not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 11, + "component_count": 27, + "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": "961f4e52d998e4215ba4738c5d46ab829144de9cf576063e9975ceb5be2750dc", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4293, + "estimated_input_tokens": 1074, + "estimated_eval_tokens": 2404 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4293, + "estimated_input_tokens": 1074, + "estimated_eval_tokens": 2404 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two PRs share a file but implement different Trainer extension points: one changes gradient-accumulation loss scaling, the other changes model signature column filtering. They are related to trainer customization, but not the same underlying bug or change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: 43651 and 43663 touch different Trainer behaviors and are not duplicates of the same fix or feature.", + "best_issue_reason": null, + "best_pr_reason": "Neither PR is a strong global duplicate representative; each addresses a separate code path in src/transformers/trainer.py.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43651", + "right": "pull_request:43663", + "accept": false, + "reason": "Different functionality: 43651 alters loss scaling during gradient accumulation, while 43663 refactors signature column filtering for custom trainers. Shared file alone is not enough to treat them as the same change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 12, + "component_count": 27, + "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": "7846a5f81fa5c5f9a90fda70b9da77d0156f4279fa8d6197a653053eeef46ff5", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3662, + "estimated_input_tokens": 916, + "estimated_eval_tokens": 2088 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3662, + "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 file and both mention TypeAdapter, but they address different concrete import/type-annotation problems in the CLI serve path, so they should not be merged as duplicates.", + "confidence": 0.82, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #43842 is the more canonical representative: it has the clearer, narrower fix for the pydantic-optional import path, with a small focused annotation change and an explicit issue target.", + "best_issue_reason": null, + "best_pr_reason": "PR #43842 is the best overall representative because its title and diff align tightly with a single concrete bug, and it is more specific than the broader TypeAdapter/typing cleanup in #43656.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43656", + "right": "pull_request:43842", + "accept": false, + "reason": "Both touch `src/transformers/cli/serve.py`, but #43656 is a broader typing/annotation cleanup for a TypeAdapter NameError, while #43842 specifically fixes the `TypeAdapter` reference when pydantic is not installed. They do not appear to fix the same underlying bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 13, + "component_count": 27, + "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": 14, + "component_count": 27, + "review_unit_index": 1, + "review_unit_count": 2, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-43996-28", + "nodes": [ + "pull_request:43996", + "pull_request:43997", + "pull_request:44007", + "pull_request:44010", + "pull_request:44013", + "pull_request:44017", + "pull_request:44018", + "pull_request:44019", + "pull_request:44024", + "pull_request:44044", + "pull_request:44056", + "pull_request:44059", + "pull_request:44066", + "pull_request:44068", + "pull_request:44071", + "pull_request:44072", + "pull_request:44073", + "pull_request:44074", + "pull_request:44076", + "pull_request:44085", + "pull_request:44086", + "pull_request:44098", + "pull_request:44101", + "pull_request:44116", + "pull_request:44129", + "pull_request:44154", + "pull_request:44161", + "pull_request:44722" + ], + "soft_pairs": [ + "pull_request:43996|pull_request:44085", + "pull_request:43996|pull_request:44044", + "pull_request:44066|pull_request:44085", + "pull_request:44007|pull_request:44072", + "pull_request:44072|pull_request:44722", + "pull_request:44066|pull_request:44072", + "pull_request:44013|pull_request:44044", + "pull_request:44066|pull_request:44086", + "pull_request: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:44071|pull_request:44722", + "pull_request:44018|pull_request:44116", + "pull_request:44018|pull_request:44086", + "pull_request:44018|pull_request:44085", + "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:44073|pull_request:44101", + "pull_request:44017|pull_request:44071", + "pull_request:44056|pull_request:44101", + "pull_request:44059|pull_request:44071", + "pull_request:44073|pull_request:44161", + "pull_request:44059|pull_request:44068", + "pull_request:44018|pull_request:44059", + "pull_request:44101|pull_request:44161", + "pull_request:43997|pull_request:44017", + "pull_request:43997|pull_request:44056", + "pull_request:44071|pull_request:44076", + "pull_request:44074|pull_request:44101", + "pull_request:44056|pull_request:44161", + "pull_request:43997|pull_request:44018", + "pull_request:43997|pull_request:44161", + "pull_request:44074|pull_request:44161", + "pull_request:43997|pull_request:44071", + "pull_request:44010|pull_request:44071", + "pull_request:44010|pull_request:44073", + "pull_request:44018|pull_request:44129", + "pull_request:43997|pull_request:44074", + "pull_request:43997|pull_request:44129", + "pull_request:44071|pull_request:44129", + "pull_request:44010|pull_request:44018", + "pull_request:43996|pull_request:44161", + "pull_request:43996|pull_request:44017", + "pull_request:44010|pull_request:44044", + "pull_request:44044|pull_request:44074", + "pull_request:44066|pull_request:44722", + "pull_request:44085|pull_request:44722", + "pull_request:44085|pull_request:44129", + "pull_request:43996|pull_request:44007", + "pull_request:43996|pull_request:44129", + "pull_request:44044|pull_request:44129", + "pull_request:44044|pull_request:44154", + "pull_request:44024|pull_request:44076" + ], + "prepared_review_unit_hash": "5317b2da15c4b7561dd2599265611040a8e0190bfff6cfb02a35b25d5ccb70bd", + "component_budget": { + "node_count": 28, + "item_count": 28, + "soft_pair_count": 77, + "serialized_chars": 84501, + "estimated_input_tokens": 21126, + "estimated_eval_tokens": 42508 + }, + "budget": { + "node_count": 28, + "item_count": 28, + "soft_pair_count": 72, + "serialized_chars": 76933, + "estimated_input_tokens": 19234, + "estimated_eval_tokens": 38724 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a sweep of model-by-model output-tracing refactors, not one single duplicate set. The only clear duplicate subcluster is around GPT-J/CodeGen output tracing, with a smaller duplicate pair in GPT-Neo; the rest share the broad standardized-output theme but not the same concrete code path.", + "confidence": 0.82, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 44722 is the best canonical PR: it is the clearest and most complete representative of the GPT-J/CodeGen output-tracing refactor, with the same concrete code paths and explicit issue linkage.", + "best_issue_reason": null, + "best_pr_reason": "PR 44722 is the strongest single representative for the duplicate subcluster because it updates the same GPT-J/CodeGen tracing path and appears to subsume the earlier GPT-J variants.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43996", + "right": "pull_request:44085", + "accept": false, + "reason": "Different model families and different code paths; they only share the broad output-tracing refactor theme." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44044", + "accept": false, + "reason": "FNet/CVT versus DeBERTa_v2 are unrelated model implementations, so this is not the same concrete change." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44085", + "accept": true, + "reason": "Both target the GPT-J/CodeGen output-tracing path and read like overlapping versions of the same refactor." + }, + { + "left": "pull_request:44007", + "right": "pull_request:44072", + "accept": false, + "reason": "RegNet and EfficientNet are separate model paths; same theme, not the same bug or change." + }, + { + "left": "pull_request:44072", + "right": "pull_request:44722", + "accept": false, + "reason": "EfficientNet versus GPT-J/CodeGen are unrelated implementations, so these are not duplicates." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44072", + "accept": false, + "reason": "Different model code paths; only the general standardized-tracing effort is shared." + }, + { + "left": "pull_request:44013", + "right": "pull_request:44044", + "accept": false, + "reason": "MobileNetV2 and DeBERTa_v2 are unrelated model-specific refactors." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-J/CodeGen versus MGP-STR are different model families and different concrete changes." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44071", + "accept": false, + "reason": "These touch different model implementations, so they are not the same underlying fix." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44068", + "accept": true, + "reason": "Both are GPT-Neo output-tracing refactors on the same model path, with the later PR looking like a duplicate/follow-up of the same change." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44068", + "accept": false, + "reason": "GPT-J/CodeGen and GPT-Neo are different code paths, so this is only thematic similarity." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44066", + "accept": false, + "reason": "GPT-Neo versus GPT-J/CodeGen are distinct model refactors, not one duplicated change." + }, + { + "left": "pull_request:44019", + "right": "pull_request:44722", + "accept": false, + "reason": "ResNet and GPT-J/CodeGen are unrelated model paths." + }, + { + "left": "pull_request:44086", + "right": "pull_request:44722", + "accept": false, + "reason": "MGP-STR and GPT-J/CodeGen are different implementations; same theme does not make them duplicates." + }, + { + "left": "pull_request:44019", + "right": "pull_request:44085", + "accept": false, + "reason": "ResNet versus GPT-J/CodeGen are separate model-specific changes." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44019", + "accept": false, + "reason": "GPT-Neo and ResNet are different model code paths." + }, + { + "left": "pull_request:44019", + "right": "pull_request:44071", + "accept": false, + "reason": "ResNet and MPT are unrelated implementations." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44722", + "accept": false, + "reason": "GPT-Neo and GPT-J/CodeGen are different model refactors, not the same concrete fix." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44116", + "accept": false, + "reason": "GPT-Neo and Flaubert/XLM are distinct model paths." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44722", + "accept": false, + "reason": "MPT and GPT-J/CodeGen are different concrete changes." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44116", + "accept": false, + "reason": "GPT-Neo versus Flaubert/XLM is only broad thematic overlap." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-Neo and MGP-STR are unrelated model implementations." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44085", + "accept": false, + "reason": "GPT-Neo and GPT-J/CodeGen are different code paths." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44086", + "accept": false, + "reason": "Different model families and different files touched." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44101", + "accept": false, + "reason": "GPT-Neo and Flaubert/XLM are separate model-specific refactors." + }, + { + "left": "pull_request:44059", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT2 and MGP-STR are unrelated concrete changes." + }, + { + "left": "pull_request:44085", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-J/CodeGen and MGP-STR are different model code paths." + }, + { + "left": "pull_request:44098", + "right": "pull_request:44101", + "accept": false, + "reason": "ViLT and Flaubert/XLM are different implementations." + }, + { + "left": "pull_request:44059", + "right": "pull_request:44071", + "accept": false, + "reason": "GPT2 and MPT are unrelated model refactors." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44017", + "accept": false, + "reason": "RegNet and Segformer are different model code paths." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44056", + "accept": false, + "reason": "RegNet and MPNet are unrelated model-specific changes." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44129", + "accept": false, + "reason": "MPT and SpeechT5 are different concrete refactors." + }, + { + "left": "pull_request:44074", + "right": "pull_request:44101", + "accept": false, + "reason": "TextNet and Flaubert/XLM are separate model implementations." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44161", + "accept": false, + "reason": "MPNet and LongT5 are different code paths." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44018", + "accept": false, + "reason": "RegNet and GPT-Neo are unrelated model refactors." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44161", + "accept": false, + "reason": "RegNet and LongT5 are different concrete changes." + }, + { + "left": "pull_request:44074", + "right": "pull_request:44161", + "accept": false, + "reason": "TextNet and LongT5 are different model-specific refactors." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44071", + "accept": false, + "reason": "SqueezeBert and MPT are unrelated code paths." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44073", + "accept": false, + "reason": "SqueezeBert and VisualBert are different model implementations." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44129", + "accept": false, + "reason": "GPT-Neo and SpeechT5 are unrelated model-specific changes." + }, + { + "left": "pull_request:44073", + "right": "pull_request:44101", + "accept": false, + "reason": "VisualBert and Flaubert/XLM are different code paths." + }, + { + "left": "pull_request:44017", + "right": "pull_request:44071", + "accept": false, + "reason": "Segformer and MPT are unrelated model refactors." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44161", + "accept": false, + "reason": "MPNet and LongT5 are different concrete changes." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44161", + "accept": false, + "reason": "FNet/CVT and LongT5 are unrelated model-specific refactors." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44017", + "accept": false, + "reason": "FNet/CVT and Segformer are different code paths." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44044", + "accept": false, + "reason": "SqueezeBert and DeBERTa_v2 are different model implementations." + }, + { + "left": "pull_request:44044", + "right": "pull_request:44074", + "accept": false, + "reason": "DeBERTa_v2 and TextNet are unrelated concrete changes." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44722", + "accept": true, + "reason": "Same GPT-J/CodeGen output-tracing refactor in the same file set; 44722 is a later, broader duplicate/follow-up." + }, + { + "left": "pull_request:44085", + "right": "pull_request:44722", + "accept": true, + "reason": "Both appear to implement the same GPT-J/CodeGen output-tracing change on the same code path." + }, + { + "left": "pull_request:44085", + "right": "pull_request:44129", + "accept": false, + "reason": "GPT-J/CodeGen and SpeechT5 are different models; shared issue tracking is not enough to merge them." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44007", + "accept": false, + "reason": "FNet/CVT and RegNet are separate model-specific refactors." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44129", + "accept": false, + "reason": "FNet/CVT and SpeechT5 are unrelated code paths." + }, + { + "left": "pull_request:44044", + "right": "pull_request:44129", + "accept": false, + "reason": "DeBERTa_v2 and SpeechT5 are different model implementations." + }, + { + "left": "pull_request:44044", + "right": "pull_request:44154", + "accept": false, + "reason": "DeBERTa_v2 and VITS are unrelated concrete changes." + }, + { + "left": "pull_request:44024", + "right": "pull_request:44076", + "accept": false, + "reason": "FocalNet and ImageGPT are different model-specific refactors." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative about cross-model overlap. The accepted soft-edge pairs are the obvious same-path duplicates (GPT-J/CodeGen and GPT-Neo), and the rejected pairs are appropriately limited to shared theme rather than the same concrete bugfix. Canonical PR 44722 is also a reasonable representative because it covers the shared GPT-J/CodeGen path and has explicit issue linkage." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 14, + "component_count": 27, + "review_unit_index": 2, + "review_unit_count": 2, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-43996-28", + "nodes": [ + "pull_request:43997", + "pull_request:44010", + "pull_request:44056", + "pull_request:44073", + "pull_request:44074", + "pull_request:44129" + ], + "soft_pairs": [ + "pull_request:44073|pull_request:44074", + "pull_request:44056|pull_request:44073", + "pull_request:44056|pull_request:44074", + "pull_request:43997|pull_request:44010", + "pull_request:44010|pull_request:44129" + ], + "prepared_review_unit_hash": "48dde5da5b8455dc2f515afc68917cb727c0d729cf4b91d10a2e05f6b5c4a79d", + "component_budget": { + "node_count": 28, + "item_count": 28, + "soft_pair_count": 77, + "serialized_chars": 84501, + "estimated_input_tokens": 21126, + "estimated_eval_tokens": 42508 + }, + "budget": { + "node_count": 6, + "item_count": 6, + "soft_pair_count": 5, + "serialized_chars": 13068, + "estimated_input_tokens": 3267, + "estimated_eval_tokens": 6790 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All items are pull requests for model-specific output-tracing refactors in different architectures. They share a theme but not the same underlying code path, so this cluster should not be merged across models.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 44129 is the strongest representative: it has an explicit issue target, a concrete SpeechT5-specific tracing refactor, and enough scope to characterize the cluster theme.", + "best_issue_reason": null, + "best_pr_reason": "44129 is the best single representative because it is explicitly tied to an issue and is the most clearly scoped, model-specific implementation of the standardized output-capture refactor.", + "soft_edge_verdicts": [ + { + "left": "pull_request:44073", + "right": "pull_request:44074", + "accept": false, + "reason": "Both are output-tracing refactors, but they target different models (VisualBert vs TextNet) and different forward paths; same pattern, not the same change." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44073", + "accept": false, + "reason": "MPNet and VisualBert are separate model implementations with distinct encoder internals; these are parallel refactors, not one duplicate fix." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44074", + "accept": false, + "reason": "Different model families and code paths; they adopt the same decorator pattern but do not fix the same bug or change." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44010", + "accept": false, + "reason": "RegNet and SqueezeBert are unrelated model modules; both PRs update output handling, but they are separate model-specific refactors." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44129", + "accept": false, + "reason": "SqueezeBert and SpeechT5 touch different architectures and different output semantics; the shared theme is too broad to treat as a duplicate or 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": 15, + "component_count": 27, + "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-43998-14", + "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", + "pull_request:46106" + ], + "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", + "pull_request:44027|pull_request:46106", + "pull_request:44026|pull_request:46106", + "pull_request:44001|pull_request:46106", + "pull_request:44002|pull_request:46106", + "pull_request:43998|pull_request:46106" + ], + "prepared_review_unit_hash": "ee057249f554dbdfbed6be83892d327bac3ecb4fca0cae8afc1d3f9fee1ac127", + "component_budget": { + "node_count": 14, + "item_count": 14, + "soft_pair_count": 45, + "serialized_chars": 43779, + "estimated_input_tokens": 10945, + "estimated_eval_tokens": 22146 + }, + "budget": { + "node_count": 14, + "item_count": 14, + "soft_pair_count": 45, + "serialized_chars": 43779, + "estimated_input_tokens": 10945, + "estimated_eval_tokens": 22146 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is a loose family of model-specific output-tracing refactors plus one unrelated docs toctree cleanup. The PRs share a template, but they target different model code paths, so they are not duplicates.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 44003 is the best representative of the core pattern: it is a substantial tracing refactor, touches two closely related Mamba-family implementations, and includes the new output-capture plumbing.", + "best_issue_reason": null, + "best_pr_reason": "PR 44003 is the strongest standalone candidate because it is the most concrete and coherent output-tracing change among the set, with broader scope than the single-model tweaks.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43998", + "right": "pull_request:44002", + "accept": false, + "reason": "Same high-level tracing refactor pattern, but different model families and different forward paths." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44000", + "accept": false, + "reason": "Different architectures and code paths; only the refactor style is shared." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44001", + "accept": false, + "reason": "Unrelated model implementations; not the same underlying bug or change." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44003", + "accept": false, + "reason": "Both are output-tracing changes, but they affect different model stacks and concrete code paths." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44029", + "accept": false, + "reason": "Different models and runtime paths; similarity is only at the refactor template level." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44001", + "accept": false, + "reason": "Separate model implementations; not a duplicate fix or mergeable single change." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44002", + "accept": false, + "reason": "Different model code paths, even though both adopt the same tracing refactor style." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44025", + "accept": false, + "reason": "Different model families and files; same general refactor theme is too broad to be duplicate." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44026", + "accept": false, + "reason": "Separate encoder/decoder wrapper code paths; not the same concrete change." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44027", + "accept": false, + "reason": "Different model wrappers and call paths; similar cleanup, not duplicate work." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44001", + "accept": false, + "reason": "Different models and outputs; the shared theme is only tracing refactor." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44002", + "accept": false, + "reason": "Distinct architectures and forward implementations, so they are not the same change." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44003", + "accept": false, + "reason": "These touch unrelated model families and cannot plausibly be merged as one concrete fix." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44025", + "accept": false, + "reason": "Same broad tracing cleanup category, but different model code paths and behavior." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44002", + "accept": false, + "reason": "Different model implementations; no shared bug beyond the refactor pattern." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44003", + "accept": false, + "reason": "Separate model families and forward logic; not a duplicate." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44025", + "accept": false, + "reason": "Different model stacks; too broad to treat as the same underlying change." + }, + { + "left": "pull_request:44002", + "right": "pull_request:44029", + "accept": false, + "reason": "Different models and concrete code paths; only the output-tracing style overlaps." + }, + { + "left": "pull_request:44002", + "right": "pull_request:44003", + "accept": false, + "reason": "Not the same implementation or bug; just a similar refactor pattern." + }, + { + "left": "pull_request:44003", + "right": "pull_request:44028", + "accept": false, + "reason": "Different model families and modules; these are not the same underlying change." + }, + { + "left": "pull_request:44003", + "right": "pull_request:44004", + "accept": false, + "reason": "Separate architectures and forward logic; similar cleanup does not make them duplicates." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44004", + "accept": false, + "reason": "Different model code paths; only the tracing refactor motif matches." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44004", + "accept": false, + "reason": "Distinct models and concrete forward behavior; not mergeable as one fix." + }, + { + "left": "pull_request:44002", + "right": "pull_request:44004", + "accept": false, + "reason": "Same broad theme, but different model implementations and behavior." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44004", + "accept": false, + "reason": "Different model-specific forward code; not the same bug or change." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44026", + "accept": false, + "reason": "Related refactor style, but the PRs target different wrappers and code paths." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44027", + "accept": false, + "reason": "Different model wrappers and runtime paths; not a duplicate fix." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44028", + "accept": false, + "reason": "Separate model implementations, so this is only thematic similarity." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44029", + "accept": false, + "reason": "Different model families and output logic; not the same underlying issue." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44030", + "accept": false, + "reason": "Different model code paths despite the shared tracing-refactor pattern." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44027", + "accept": false, + "reason": "Both are encoder-decoder wrappers, but they are different model classes and code paths." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44028", + "accept": false, + "reason": "Different model implementations; the overlap is only in refactor style." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44029", + "accept": false, + "reason": "Unrelated model families and forward paths; not duplicate work." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44030", + "accept": false, + "reason": "Different concrete model behavior, so these are not the same change." + }, + { + "left": "pull_request:44027", + "right": "pull_request:44028", + "accept": false, + "reason": "Different model families and files; not a single underlying bug or fix." + }, + { + "left": "pull_request:44027", + "right": "pull_request:44029", + "accept": false, + "reason": "Separate code paths and architectures; only the refactor motif is shared." + }, + { + "left": "pull_request:44027", + "right": "pull_request:44030", + "accept": false, + "reason": "Different models and concrete behavior; not the same change." + }, + { + "left": "pull_request:44028", + "right": "pull_request:44029", + "accept": false, + "reason": "Distinct model implementations and tracing paths; not duplicates." + }, + { + "left": "pull_request:44028", + "right": "pull_request:44030", + "accept": false, + "reason": "Different code paths and architectures, so they should stay separate." + }, + { + "left": "pull_request:44029", + "right": "pull_request:44030", + "accept": false, + "reason": "Different model families and forward implementations; only the refactor pattern overlaps." + }, + { + "left": "pull_request:44027", + "right": "pull_request:46106", + "accept": false, + "reason": "The docs toctree cleanup is unrelated to model output tracing." + }, + { + "left": "pull_request:44026", + "right": "pull_request:46106", + "accept": false, + "reason": "Documentation navigation change, not the same kind of model refactor." + }, + { + "left": "pull_request:44001", + "right": "pull_request:46106", + "accept": false, + "reason": "Docs-only toctree edit is unrelated to the model tracing PR." + }, + { + "left": "pull_request:44002", + "right": "pull_request:46106", + "accept": false, + "reason": "Unrelated docs cleanup versus model forward/tracing changes." + }, + { + "left": "pull_request:43998", + "right": "pull_request:46106", + "accept": false, + "reason": "Documentation restructuring is not the same underlying change as model output tracing." + } + ] + }, + "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": 27, + "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-44662-3", + "nodes": [ + "pull_request:44662", + "pull_request:45075", + "pull_request:45401" + ], + "soft_pairs": [ + "pull_request:44662|pull_request:45401", + "pull_request:44662|pull_request:45075", + "pull_request:45075|pull_request:45401" + ], + "prepared_review_unit_hash": "056a5defb29e5d7941dbdff78b97b7324b3df56a452db784f88a882c57c9cec1", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 10912, + "estimated_input_tokens": 2728, + "estimated_eval_tokens": 5712 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 10912, + "estimated_input_tokens": 2728, + "estimated_eval_tokens": 5712 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All three PRs are separate model-addition efforts for different architectures (PenguinVL, DeepSeek-OCR-2, and Voxtral TTS). The shared files are mostly framework boilerplate for registering a new model, not evidence of the same underlying bug or change.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "45075 is the best representative of the cluster: it is a full, self-contained model integration with conversion, modeling, processing, tests, and config updates, whereas the others are different model families.", + "best_issue_reason": null, + "best_pr_reason": "45075 is the strongest standalone PR in the set, but only as a representative of a new-model integration pattern; it is not a duplicate of the other two.", + "soft_edge_verdicts": [ + { + "left": "pull_request:44662", + "right": "pull_request:45401", + "accept": false, + "reason": "Different model families and codepaths: PenguinVL is a vision-language model, while Voxtral TTS is a text-to-speech model. Shared auto/docs registration files are boilerplate, not duplicate evidence." + }, + { + "left": "pull_request:44662", + "right": "pull_request:45075", + "accept": false, + "reason": "These add different models (PenguinVL vs DeepSeek-OCR-2) with distinct configs, converters, processors, and tests. Similar registration scaffolding is not enough to treat them as the same change." + }, + { + "left": "pull_request:45075", + "right": "pull_request:45401", + "accept": false, + "reason": "DeepSeek-OCR-2 and Voxtral TTS solve unrelated problems in different modalities (OCR vs TTS) and touch different implementation paths. 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": 17, + "component_count": 27, + "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": "7ec26b41425bacf77fed87d34dbf523793c66fdac61457df6a909bff0d2a06a8", + "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 are related kernel-fusion PRs, but they do not look like the same change: one adds a generic Module Fusion API scaffold, while the other implements n-to-1 kernel fusion through KernelConfig and integration hooks. They could complement each other, but not be merged as duplicates.", + "confidence": 0.87, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #45363 is the better canonical PR because it makes the concrete feature change: KernelConfig-driven n-to-1 fusion wired into model loading/integration code, with an example showing end-to-end use. PR #44979 is more of an API foundation and does not appear to implement the same concrete behavior.", + "best_issue_reason": null, + "best_pr_reason": "PR #45363 is the strongest representative of the cluster because it targets the actual user-facing fusion mechanism and touches the code paths needed to make it work, rather than just defining the underlying module-fusion abstraction.", + "soft_edge_verdicts": [ + { + "left": "pull_request:44979", + "right": "pull_request:45363", + "accept": false, + "reason": "Related area, but different scope and likely different merge units: #44979 adds a generic Module Fusion API and tests, while #45363 implements KernelConfig-based n-to-1 kernel fusion across model-loading/integration paths. They are not the same concrete bug/change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 18, + "component_count": 27, + "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": "f092f7d542120c92e78e9bc6a571a8b8052283cbf2603fa0e984a8088b691b85", + "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 address the same concrete AttributeError in `_patch_mistral_regex` by switching from `backend_tokenizer.pre_tokenizer` to `tokenizer.pre_tokenizer`; the second PR is the same fix plus tests, so they are duplicate candidates.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #45317 is the better canonical PR because it contains the same code fix as #45086 plus test coverage, making it the more complete representation of the fix.", + "best_issue_reason": null, + "best_pr_reason": "PR #45317 is the best overall PR candidate because it fixes the same bug as #45086 and adds validation/tests, so it is the most complete and maintainable version of the change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45086", + "right": "pull_request:45317", + "accept": true, + "reason": "Same underlying AttributeError fix in the same function and file; the second PR is the same change with added tests, so they could plausibly be merged into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 19, + "component_count": 27, + "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": "192dec3a76c7033644aa9133c193fd89bdd1e4b2c26d31a81add50c73a0a0d53", + "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": "The two PRs are related to NemotronH but address different changes: PR 45115 is a small conversion-mapping/refactor for checkpoint key renaming and model inheritance, while PR 45569 is a broader architecture split introducing separate NemotronHDense/NemotronHSparse models, docs, and tests. They are not duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 45569 is the broader, more complete implementation: it introduces the dedicated dense/sparse model packages, updates auto-mapping and docs, and includes tests. PR 45115 is narrower and more provisional, focused on conversion mapping and inheritance plumbing.", + "best_issue_reason": null, + "best_pr_reason": "PR 45569 is the best representative artifact because it covers the main user-facing model restructuring and contains the fuller codepath changes, while PR 45115 appears to be an earlier, narrower refactor step.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45115", + "right": "pull_request:45569", + "accept": false, + "reason": "Different underlying changes. 45115 mainly renames checkpoint conversion keys and adjusts inheritance/config plumbing; 45569 splits NemotronH into separate dense/sparse model families with docs and tests. They touch some shared files but would not plausibly be merged into one PR as the same fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 20, + "component_count": 27, + "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": "536d16a6e4efe865c1728d9c35fe32533f8c0ded1d241b206ccd3270d8198e7f", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5398, + "estimated_input_tokens": 1350, + "estimated_eval_tokens": 2956 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5398, + "estimated_input_tokens": 1350, + "estimated_eval_tokens": 2956 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two related but distinct PRs: one adds heterogeneous per-layer configuration support, the other adds heterogeneous modeling/runtime support built on top of that config work. They share infrastructure files, but they are not the same underlying change.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: #45333 is the config-layer foundation, while #45332 adds broader modeling/cache/masking changes and is a separate follow-on change.", + "best_issue_reason": null, + "best_pr_reason": "#45332 is the broader end-to-end heterogeneous model support PR, so it is the better representative of the cluster theme, even though it depends on the config work in #45333.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45332", + "right": "pull_request:45333", + "accept": false, + "reason": "Related stack, but not the same concrete bug/change. #45333 introduces heterogeneous per-layer config support; #45332 extends modeling, cache, and masking behavior to use it. They could not be cleanly merged as one duplicate PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 21, + "component_count": 27, + "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": "cf25d9cfb8ba1acaf86a97eefd8c60fa91eaab1112b6564e45049bc9414c346f", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5514, + "estimated_input_tokens": 1379, + "estimated_eval_tokens": 3014 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5514, + "estimated_input_tokens": 1379, + "estimated_eval_tokens": 3014 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs are typing-oriented, but they address different scopes: #45415 is a broad repo-wide type-checking sweep, while #45425 is a narrower protocol/typing refinement around modeling utilities and PEFT-related interfaces.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #45425 is the better canonical representative because it is a focused, self-contained typing change; #45415 is a broad housekeeping sweep across many files and is less suitable as the cluster anchor.", + "best_issue_reason": null, + "best_pr_reason": "#45425 is the strongest representative of this cluster because it has a coherent target and clearer code-path impact, whereas #45415 is a large cross-cutting type-check enablement patch.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45415", + "right": "pull_request:45425", + "accept": false, + "reason": "Reject: both are about typing, but they are not the same underlying change. #45415 is a broad type-checking cleanup across many modules; #45425 adds specific typing protocols and modeling-utils-related stubs. Shared files and subsystem are not enough to treat them as duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 22, + "component_count": 27, + "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-45487-5", + "nodes": [ + "pull_request:45487", + "pull_request:46101", + "pull_request:46109", + "pull_request:46135", + "pull_request:46151" + ], + "soft_pairs": [ + "pull_request:45487|pull_request:46101", + "pull_request:45487|pull_request:46151", + "pull_request:46101|pull_request:46151", + "pull_request:46109|pull_request:46151", + "pull_request:46135|pull_request:46151" + ], + "prepared_review_unit_hash": "316c6f1778779b75c9c161c484b3bad17aaf87b766489f7102a68d3a44064265", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 5, + "serialized_chars": 12531, + "estimated_input_tokens": 3133, + "estimated_eval_tokens": 6522 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 5, + "serialized_chars": 12531, + "estimated_input_tokens": 3133, + "estimated_eval_tokens": 6522 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is mostly a loose similarity group around device/expectation updates, but the artifacts are not true duplicates. Two PRs are concrete model-parallel bug fixes in different models, while the others are test-only expectation adjustments in unrelated areas.", + "confidence": 0.73, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 45487 is the broadest concrete code fix: it addresses a real model-parallel device-placement bug in AltCLIP/ChineseCLIP model code and includes targeted test updates. It is more representative than the narrower or test-only PRs.", + "best_issue_reason": null, + "best_pr_reason": "PR 45487 is the best representative artifact in this cluster because it fixes an actual runtime bug in model code rather than just updating test expectations, and it covers the same general class of device-mismatch issue more fully than the other code-fix PR.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45487", + "right": "pull_request:46101", + "accept": false, + "reason": "Both are model-parallel device-placement fixes, but they hit different models and different code paths (AltCLIP/ChineseCLIP vs Qwen2-Audio). Too different to merge as one PR." + }, + { + "left": "pull_request:45487", + "right": "pull_request:46151", + "accept": false, + "reason": "45487 is a model-code bug fix; 46151 only adds test expectations for pipeline outputs. Different artifact types and different subsystems." + }, + { + "left": "pull_request:46101", + "right": "pull_request:46151", + "accept": false, + "reason": "One is a runtime fix in Qwen2-Audio model code; the other is a test expectation update for pipeline token classification. Not the same underlying change." + }, + { + "left": "pull_request:46109", + "right": "pull_request:46151", + "accept": false, + "reason": "Both touch tests, but they are for unrelated areas: distributed trainer parity assertions vs token-classification pipeline expectations. No shared bug/change." + }, + { + "left": "pull_request:46135", + "right": "pull_request:46151", + "accept": false, + "reason": "These are both test-related, but they concern different models and different expectations (Kosmos2.5 integration vs token-classification pipeline). Not duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 23, + "component_count": 27, + "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-45513-3", + "nodes": [ + "pull_request:45513", + "pull_request:46084", + "pull_request:46098" + ], + "soft_pairs": [ + "pull_request:45513|pull_request:46084", + "pull_request:46084|pull_request:46098" + ], + "prepared_review_unit_hash": "6d206235d9654178271ce29124fefb1b020db0c6530b1d78c70bb9b038776a0a", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 7426, + "estimated_input_tokens": 1857, + "estimated_eval_tokens": 3970 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 7426, + "estimated_input_tokens": 1857, + "estimated_eval_tokens": 3970 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two Mamba2 PRs are about the same cached-forward seq_len>1 bug and are good duplicate candidates; the Qwen3.5 PR is a separate model-family fix and should not be merged with them.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "pull_request:46098 is the most complete representative of the Mamba2 bug fix: it has the clearest title, explicitly targets the tracked issue, and includes the code fix plus tests/docs around the same cached seq_len>1 failure mode.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:46098 is the strongest standalone PR for the cluster because it spells out the concrete bug, covers both CPU/GPU paths, and is broader than the smaller companion fix in 46084.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45513", + "right": "pull_request:46084", + "accept": false, + "reason": "Different models and different code paths: Qwen3.5 linear attention vs Mamba2 mixer cache handling. Similar cache-length symptoms are not enough to treat them as the same underlying change." + }, + { + "left": "pull_request:46084", + "right": "pull_request:46098", + "accept": true, + "reason": "Both PRs fix the same Mamba2 use_cache seq_len>1 cached-forward bug against issue #46032, touch the same core files, and could plausibly be merged into one consolidated PR." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet. The Qwen3.5 PR is correctly treated as separate from the Mamba2 fixes, and the 46084/46098 pair is reasonably treated as mergeable into one PR for the same issue (#46032) since they share the same model, files, and failure mode. The canonical/standalone PR reasoning for 46098 is also supported by its broader scope and explicit issue linkage." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 24, + "component_count": 27, + "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": "20b05605e6c0803317f34054cb4e88e71a54c56416d72cacf90fa219ca8aee97", + "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 to expert-parallel loading, but they fix different failure modes: one preserves distributed config and adds an EP detection helper for DeepSpeed ZeRO-3/accelerate launch, while the other fixes expert parameters being overwritten during FSDP2/rank-0 broadcast and adds DTensor-aware handling across the loading path.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR for a duplicate pair here; the two changes target different code paths and different concrete bugs.", + "best_issue_reason": null, + "best_pr_reason": "PR #45662 is the stronger representative of the cluster because it addresses a more concrete loading corruption bug with a clearer end-user symptom and a broader, internally consistent fix across the loading stack.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45548", + "right": "pull_request:45662", + "accept": false, + "reason": "Both involve EP/loading, but they do not appear to fix the same underlying bug. #45548 is about keeping distributed_config/EP state available for ZeRO-3 loading via accelerate launch, while #45662 prevents expert parameters from being silently overwritten during FSDP2 rank-0 broadcast and adds DTensor-aware post-shard handling. Same subsystem, different failure modes and patch shapes." + } + ] + }, + "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": 27, + "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": "7bb44fc66c42b7ec2793ed5a1c58fd4a6d9022c436278beb967744bce4e3c511", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4589, + "estimated_input_tokens": 1148, + "estimated_eval_tokens": 2552 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4589, + "estimated_input_tokens": 1148, + "estimated_eval_tokens": 2552 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs address the same xdist-related captured_info logging problem in Transformers testing utilities; #45645 appears to be the broader, more complete version of #45639 and adds the CI-side handling needed for the new per-worker files.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "#45645 is the more complete fix: it includes the core xdist-safe output-file change in testing_utils plus the workflow update to collect multiple captured_info files, so it subsumes #45639.", + "best_issue_reason": null, + "best_pr_reason": "#45645 is the best representative of the cluster because it covers the full end-to-end change needed to make captured_info logging xdist-safe and still visible in CI.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45639", + "right": "pull_request:45645", + "accept": true, + "reason": "Same underlying bug/change: xdist collisions for captured_info/debug logs in testing_utils. #45645 expands the same fix with workflow adjustments, so the pair is plausibly one combined PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 26, + "component_count": 27, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "cache", + "cache_hit": true, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45737-2", + "nodes": [ + "pull_request:45737", + "pull_request:45738" + ], + "soft_pairs": [ + "pull_request:45737|pull_request:45738" + ], + "prepared_review_unit_hash": "cc2ade73a49ec6a8883a7a24c433988645588bcd049e8aad2bce56b1b24daa1c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4131, + "estimated_input_tokens": 1033, + "estimated_eval_tokens": 2322 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4131, + "estimated_input_tokens": 1033, + "estimated_eval_tokens": 2322 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs address the same MusicgenMelody cache initialization bug in `forward()`: replacing `EncoderDecoderCache(DynamicCache, DynamicCache)` with a single `DynamicCache`. PR #45737 is the more complete duplicate because it bundles the fix with a regression test, while #45738 is the fix-only variant.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #45737 is the best canonical PR because it contains the same code fix as #45738 plus a regression test, making it the most complete representation of the bug fix.", + "best_issue_reason": null, + "best_pr_reason": "PR #45737 is the strongest overall candidate: it fixes the concrete cache bug and adds a targeted regression test, whereas #45738 only changes the implementation.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45737", + "right": "pull_request:45738", + "accept": true, + "reason": "Same underlying code-path bug in `musicgen_melody` cache setup; the diffs match on the core fix and the test-only addition in #45737 could plausibly live in the same PR as #45738." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The assessment is grounded in the packet: both PRs make the same `musicgen_melody` cache initialization change, and #45737 only adds a regression test on top of that fix. The soft-edge accept is conservative enough because the diffs are mergeable into one PR for the same bugfix." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 27, + "component_count": 27, + "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-46069-3", + "nodes": [ + "pull_request:46069", + "pull_request:46073", + "pull_request:46154" + ], + "soft_pairs": [ + "pull_request:46069|pull_request:46073", + "pull_request:46069|pull_request:46154", + "pull_request:46073|pull_request:46154" + ], + "prepared_review_unit_hash": "ca5e1ce36097f5e07b0742dd72135940ed5b0981d1f6f063a4e1b82f0eec64dd", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7263, + "estimated_input_tokens": 1816, + "estimated_eval_tokens": 3888 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 7263, + "estimated_input_tokens": 1816, + "estimated_eval_tokens": 3888 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are three separate model-specific PRs adding SDPA/attn_implementation dispatch support and doc badges. They are conceptually related, but they do not look like duplicates of the same concrete code change.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #46154 is the best representative of the cluster: it targets the common SDPA dispatch feature, has the broadest implementation/test footprint, and is the clearest exemplar of this family of model-specific attention-dispatch updates.", + "best_issue_reason": null, + "best_pr_reason": "PR #46154 is the strongest global representative because it combines the feature addition with more visible test coverage and a clearer, canonical title for the shared theme.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46069", + "right": "pull_request:46073", + "accept": false, + "reason": "Different model implementations (ShieldGemma2 vision model vs CTRL text model). Same broad attention-dispatch theme, but not the same concrete bug/change and not something to merge into one PR." + }, + { + "left": "pull_request:46069", + "right": "pull_request:46154", + "accept": false, + "reason": "Both add attn_implementation support, but to different model code paths (ShieldGemma2 vs T5). This is a shared feature area, not a duplicate fix." + }, + { + "left": "pull_request:46073", + "right": "pull_request:46154", + "accept": false, + "reason": "CTRL and T5 are distinct architectures with separate attention implementations and tests; they are similar feature PRs, but not the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + } + ] +} diff --git a/snapshots/20260523T060016Z/analysis-runs/hybrid-model-20260523t060016z/manifest.json b/snapshots/20260523T060016Z/analysis-runs/hybrid-model-20260523t060016z/manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..6e06c2e971bfa19f3134ba6735b23becf3e29a56 --- /dev/null +++ b/snapshots/20260523T060016Z/analysis-runs/hybrid-model-20260523t060016z/manifest.json @@ -0,0 +1,14 @@ +{ + "analysis_id": "hybrid-model-20260523t060016z", + "artifacts": { + "hybrid": "snapshots/20260523T060016Z/analysis-runs/hybrid-model-20260523t060016z/analysis-report-hybrid.json", + "hybrid_reviews": "snapshots/20260523T060016Z/analysis-runs/hybrid-model-20260523t060016z/analysis-report-hybrid.llm-reviews.json" + }, + "channel": "canonical", + "model": null, + "published_at": "2026-05-23T06:07:40Z", + "repo": "huggingface/transformers", + "schema_version": 1, + "snapshot_id": "20260523T060016Z", + "variant": "hybrid" +} diff --git a/snapshots/20260523T060016Z/manifest.json b/snapshots/20260523T060016Z/manifest.json index f1e5940c07e656cee499be7b16b7b50bfcc215c5..d69f4c518df7e4d1b8354fbbcfb60129c2e7326d 100644 --- a/snapshots/20260523T060016Z/manifest.json +++ b/snapshots/20260523T060016Z/manifest.json @@ -32,6 +32,24 @@ "timeline_events": 60 }, "extracted_at": "2026-05-23T06:00:16Z", + "published_analysis": { + "canonical_analysis_id": "hybrid-model-20260523t060016z", + "runs": { + "hybrid-model-20260523t060016z": { + "analysis_id": "hybrid-model-20260523t060016z", + "artifacts": { + "hybrid": "snapshots/20260523T060016Z/analysis-runs/hybrid-model-20260523t060016z/analysis-report-hybrid.json", + "hybrid_reviews": "snapshots/20260523T060016Z/analysis-runs/hybrid-model-20260523t060016z/analysis-report-hybrid.llm-reviews.json" + }, + "channel": "canonical", + "manifest_path": "snapshots/20260523T060016Z/analysis-runs/hybrid-model-20260523t060016z/manifest.json", + "model": null, + "published_at": "2026-05-23T06:07:40Z", + "variant": "hybrid" + } + }, + "schema_version": 1 + }, "repo": "huggingface/transformers", "snapshot_id": "20260523T060016Z", "watermark": {