diff --git a/analysis/current/analysis-report-hybrid.json b/analysis/current/analysis-report-hybrid.json index 8a1df5406fa12c14a650e02dab37e062795bef92..463256e9b94f2be0a873f7e44e891d1b279da243 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": "20260604T180024Z", - "generated_at": "2026-06-04T18:11:31Z", + "snapshot_id": "20260605T000034Z", + "generated_at": "2026-06-05T00:09:29Z", "evidence_quality": "full", "llm_enrichment": true, "meta_bugs": [ @@ -1148,6 +1148,364 @@ } ] }, + { + "cluster_id": "cluster-45363-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "open", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 46339, + "issue_numbers": [], + "pr_numbers": [ + 45363, + 45666, + 46339 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 45363, + "right_pr_number": 45666, + "code_similarity": 0.392, + "size_similarity": 0.298, + "file_overlap": 0.571, + "area_overlap": 0.329, + "patch_similarity": 0.306, + "shared_filenames": [ + "fused_qwen_example.py", + "src/transformers/integrations/__init__.py", + "src/transformers/integrations/hub_kernels.py", + "src/transformers/modeling_utils.py" + ], + "shared_file_areas": [ + { + "filename": "fused_qwen_example.py", + "left_ranges": [ + [ + 1, + 30 + ] + ], + "right_ranges": [ + [ + 1, + 49 + ] + ] + }, + { + "filename": "src/transformers/integrations/__init__.py", + "left_ranges": [ + [ + 67, + 81 + ], + [ + 226, + 240 + ] + ], + "right_ranges": [ + [ + 67, + 73 + ], + [ + 222, + 228 + ] + ] + }, + { + "filename": "src/transformers/integrations/hub_kernels.py", + "left_ranges": [ + [ + 11, + 17 + ], + [ + 22, + 35 + ], + [ + 505, + 716 + ] + ], + "right_ranges": [ + [ + 11, + 53 + ], + [ + 286, + 295 + ], + [ + 521, + 1077 + ] + ] + }, + { + "filename": "src/transformers/modeling_utils.py", + "left_ranges": [ + [ + 3738, + 3747 + ] + ], + "right_ranges": [ + [ + 1330, + 1338 + ], + [ + 3763, + 3789 + ], + [ + 4223, + 4234 + ], + [ + 4628, + 4641 + ] + ] + } + ] + }, + { + "left_pr_number": 45363, + "right_pr_number": 46339, + "code_similarity": 0.506, + "size_similarity": 0.925, + "file_overlap": 0.286, + "area_overlap": 0.545, + "patch_similarity": 0.297, + "shared_filenames": [ + "src/transformers/integrations/hub_kernels.py", + "src/transformers/modeling_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/integrations/hub_kernels.py", + "left_ranges": [ + [ + 11, + 17 + ], + [ + 22, + 35 + ], + [ + 505, + 716 + ] + ], + "right_ranges": [ + [ + 14, + 52 + ], + [ + 285, + 294 + ], + [ + 521, + 685 + ] + ] + }, + { + "filename": "src/transformers/modeling_utils.py", + "left_ranges": [ + [ + 3738, + 3747 + ] + ], + "right_ranges": [ + [ + 1346, + 1354 + ], + [ + 3785, + 3811 + ], + [ + 4245, + 4256 + ], + [ + 4650, + 4663 + ] + ] + } + ] + }, + { + "left_pr_number": 45666, + "right_pr_number": 46339, + "code_similarity": 0.399, + "size_similarity": 0.322, + "file_overlap": 0.667, + "area_overlap": 0.247, + "patch_similarity": 0.318, + "shared_filenames": [ + "src/transformers/core_model_loading.py", + "src/transformers/integrations/hub_kernels.py", + "src/transformers/modeling_utils.py", + "tests/kernels/test_kernels.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/core_model_loading.py", + "left_ranges": [ + [ + 19, + 25 + ], + [ + 80, + 86 + ] + ], + "right_ranges": [ + [ + 19, + 25 + ], + [ + 80, + 86 + ] + ] + }, + { + "filename": "src/transformers/integrations/hub_kernels.py", + "left_ranges": [ + [ + 11, + 53 + ], + [ + 286, + 295 + ], + [ + 521, + 1077 + ] + ], + "right_ranges": [ + [ + 14, + 52 + ], + [ + 285, + 294 + ], + [ + 521, + 685 + ] + ] + }, + { + "filename": "src/transformers/modeling_utils.py", + "left_ranges": [ + [ + 1330, + 1338 + ], + [ + 3763, + 3789 + ], + [ + 4223, + 4234 + ], + [ + 4628, + 4641 + ] + ], + "right_ranges": [ + [ + 1346, + 1354 + ], + [ + 3785, + 3811 + ], + [ + 4245, + 4256 + ], + [ + 4650, + 4663 + ] + ] + }, + { + "filename": "tests/kernels/test_kernels.py", + "left_ranges": [ + [ + 19, + 54 + ], + [ + 228, + 281 + ], + [ + 294, + 317 + ], + [ + 605, + 995 + ] + ], + "right_ranges": [ + [ + 21, + 28 + ], + [ + 33, + 39 + ], + [ + 90, + 103 + ], + [ + 217, + 323 + ], + [ + 336, + 359 + ] + ] + } + ] + } + ] + }, { "cluster_id": "cluster-43240-3", "summary": "Cluster of 1 issues and 2 PRs centered on issue #43240.", @@ -1615,77 +1973,31 @@ ] }, { - "cluster_id": "cluster-43656-2", + "cluster_id": "cluster-44002-2", "summary": "Cluster of 2 related pull requests linked by soft_similarity.", "status": "open", "confidence": 0.5, "canonical_issue_number": null, - "canonical_pr_number": 43656, + "canonical_pr_number": 44002, "issue_numbers": [], "pr_numbers": [ - 43656, - 43842 + 44002, + 44029 ], "evidence_types": [ "soft_similarity" ], "pr_comparisons": [ { - "left_pr_number": 43656, - "right_pr_number": 43842, - "code_similarity": 0.405, - "size_similarity": 0.5, - "file_overlap": 1.0, + "left_pr_number": 44002, + "right_pr_number": 44029, + "code_similarity": 0.135, + "size_similarity": 0.43, + "file_overlap": 0.0, "area_overlap": 0.0, - "patch_similarity": 0.036, - "shared_filenames": [ - "src/transformers/cli/serve.py" - ], - "shared_file_areas": [ - { - "filename": "src/transformers/cli/serve.py", - "left_ranges": [ - [ - 11, - 18 - ], - [ - 30, - 36 - ], - [ - 315, - 323 - ], - [ - 665, - 671 - ], - [ - 931, - 937 - ], - [ - 1843, - 1849 - ], - [ - 1868, - 1874 - ] - ], - "right_ranges": [ - [ - 54, - 61 - ], - [ - 587, - 593 - ] - ] - } - ] + "patch_similarity": 0.329, + "shared_filenames": [], + "shared_file_areas": [] } ] }, @@ -1994,15 +2306,6 @@ "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-43656-2", - "canonical_pr_number": 43656, - "duplicate_pr_numbers": [ - 43842 - ], - "target_issue_number": null, - "reason": "PRs in cluster-43656-2 are treated as duplicates because they share soft_similarity evidence." - }, { "cluster_id": "cluster-43698-3", "canonical_pr_number": 43779, @@ -2029,6 +2332,15 @@ "target_issue_number": 43979, "reason": "PRs in cluster-43979-11 are treated as duplicates because they converge on issue #43979 with closing_reference, shared_issue_target, soft_similarity evidence." }, + { + "cluster_id": "cluster-44002-2", + "canonical_pr_number": 44002, + "duplicate_pr_numbers": [ + 44029 + ], + "target_issue_number": null, + "reason": "PRs in cluster-44002-2 are treated as duplicates because they share soft_similarity evidence." + }, { "cluster_id": "cluster-44018-2", "canonical_pr_number": 44068, @@ -2047,6 +2359,16 @@ "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-45363-3", + "canonical_pr_number": 46339, + "duplicate_pr_numbers": [ + 45363, + 45666 + ], + "target_issue_number": null, + "reason": "PRs in cluster-45363-3 are treated as duplicates because they share soft_similarity evidence." + }, { "cluster_id": "cluster-45561-3", "canonical_pr_number": 45645, diff --git a/analysis/current/analysis-report-hybrid.llm-reviews.json b/analysis/current/analysis-report-hybrid.llm-reviews.json index c5efecb55bc693660dd99c23b4354d6e94001670..e73f5fc11bc4a04eb772b2f241defd7c1caa3be7 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": "20260604T180024Z", - "generated_at": "2026-06-04T18:11:31Z", + "snapshot_id": "20260605T000034Z", + "generated_at": "2026-06-05T00:09:29Z", "model": "gpt-5.4-mini?service_tier=flex", "reviews": [ { @@ -133,418 +133,334 @@ "issue:41762|issue:43596", "issue:39290|issue:41720" ], - "prepared_review_unit_hash": "53bd3fd6d0990d3a29343d0e99ae884846eb64591f640e9f1aa3fbb4f15259dc", + "prepared_review_unit_hash": "50bfea56a64ba632f40d890e975a7f65fd2cc664722718084808f796ad965790", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 64, - "serialized_chars": 46877, - "estimated_input_tokens": 11720, - "estimated_eval_tokens": 23696 + "serialized_chars": 46875, + "estimated_input_tokens": 11719, + "estimated_eval_tokens": 23694 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most pairs are unrelated. The only clear duplicate cluster is the DeepSpeed ZeRO-3 zero-sized tensor IndexError around Bert/Gemma model loading; the rest are distinct model-, backend-, or docs-specific issues.", - "confidence": 0.88, - "canonical_issue_reason": "issue:43638 is the clearest representative of the ZeRO-3 zero-sized tensor IndexError cluster: it states the exact exception, the DeepSpeed ZeRO-3 context, and a concrete trigger (non-pretrained Bert).", + "summary": "Mostly a mixed cluster of unrelated, model-specific runtime/docs issues. Only one clear duplicate pair stands out: the two ZeRO-3/BERT IndexError reports. The rest share broad subsystems (DeepSpeed, Qwen, docs, dtype/device issues) but not the same concrete bug.", + "confidence": 0.77, + "canonical_issue_reason": "issue:43596 is the best canonical issue for the only clear duplicate pair: it predates issue:43638 and describes the same ZeRO-3/BERT IndexError failure more cleanly and broadly.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43638 is the most general and explicit reproduction of the shared ZeRO-3 loading bug, so it is the best canonical issue for that cluster.", + "best_issue_reason": "issue:43596 is also the best overall issue to keep: earliest, concise, and the clearest representative of the shared concrete failure.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44322", "right": "issue:45464", "accept": false, - "reason": "Different Qwen3.5 bugs: missing config attribute vs streaming chat/completions failure." + "reason": "Different failures: missing config attribute vs streaming chat/completions error." }, { "left": "issue:45237", "right": "issue:45685", "accept": false, - "reason": "Different hardware/backend failures: AMD GPU issue vs MPS histogram op limitation." + "reason": "Different hardware/backend problems on different models; only broad runtime similarity." }, { "left": "issue:43716", "right": "issue:45237", "accept": false, - "reason": "Different models and failure modes; not the same bug." + "reason": "Image preprocessor dtype mismatch is not the same as AMD GPU execution failure." }, { "left": "issue:43828", "right": "issue:45237", "accept": false, - "reason": "Different symptoms: autocast dtype mismatch vs AMD GPU failure." + "reason": "Both are dtype/runtime issues, but on different models and code paths." }, { "left": "issue:45237", "right": "issue:45507", "accept": false, - "reason": "Different problems: AMD GPU runtime issue vs invalid method call in GraniteMoEHybrid." + "reason": "Different model/backend failures; not the same concrete bug." }, { "left": "issue:44292", "right": "issue:45237", "accept": false, - "reason": "Different models and error paths; not a duplicate." + "reason": "Different Qwen/GPT-OSS inference failures with no concrete shared code path." }, { "left": "issue:30277", "right": "issue:35545", "accept": false, - "reason": "Distributed collective mismatch in training is unrelated to ONNX export failure." + "reason": "DeepSpeed collective mismatch and ONNX export are unrelated." }, { "left": "issue:35545", "right": "issue:42915", "accept": false, - "reason": "Export-to-ONNX bug and FineGrainedFP8 config failure are different code paths." + "reason": "Export-to-ONNX bug vs FP8 model failure; different issues." }, { "left": "issue:35545", "right": "issue:41720", "accept": false, - "reason": "ONNX export issue vs auto device-mapping cuda assert; unrelated." + "reason": "Different model/runtime failures; no shared concrete bug." }, { "left": "issue:45901", "right": "issue:46015", "accept": false, - "reason": "Table QA crash and outdated RAG examples are unrelated." + "reason": "Table QA crash and outdated RAG docs are unrelated." }, { "left": "issue:44509", "right": "issue:44512", "accept": false, - "reason": "Both are v5 docs cleanups, but they target different removed docs content." + "reason": "Both are v5 docs cleanup items, but they reference different removed docs content." }, { "left": "issue:35545", "right": "issue:36010", "accept": false, - "reason": "ONNX export bug and GenerationMixin import error are separate issues." + "reason": "ONNX export error and GenerationMixin import error are different failures." }, { "left": "issue:44918", "right": "issue:46165", "accept": false, - "reason": "TRL SFT embedding unpacking and training resume with remote code are different failures." + "reason": "TRL SFT embedding unpacking and training-resume/trust-remote-code are different problems." }, { "left": "issue:44805", "right": "issue:45823", "accept": false, - "reason": "Generic mask-shape mismatch vs Gemma4 device-mapping mismatch; different bugs." + "reason": "Mask/index shape mismatch is not the same as Gemma4 device-map mismatch." }, { "left": "issue:43638", "right": "issue:44805", "accept": false, - "reason": "Different exception types and different model/runtime contexts." + "reason": "Both are shape/index errors, but the model/setup and failure path differ." }, { "left": "issue:43596", "right": "issue:44805", "accept": false, - "reason": "Zero-3 init IndexError is unrelated to mask/tensor shape mismatch." + "reason": "Different error signatures and code paths; not the same bug." }, { "left": "issue:44805", "right": "issue:46294", "accept": false, - "reason": "Mask-shape mismatch and Qwen3.5 video token mismatch are different problems." + "reason": "Mask shape mismatch is unrelated to video token/template mismatch." }, { "left": "issue:44805", "right": "issue:45161", "accept": false, - "reason": "Different failures: mask/indexing bug vs GPT-OSS TP issue." + "reason": "Different model features and failure mechanisms." }, { "left": "issue:44560", "right": "issue:46165", "accept": false, - "reason": "Video apply_chat_template failure is unrelated to resume/trust_remote_code training issue." + "reason": "Video apply_chat_template failure and training resume issue are unrelated." }, { "left": "issue:43596", "right": "issue:46165", "accept": false, - "reason": "DeepSpeed ZeRO-3 Bert loading bug is unrelated to training resume with remote code." + "reason": "Different training/setup issue; no concrete overlap beyond general model loading." }, { "left": "issue:43572", "right": "issue:46165", "accept": false, - "reason": "StableLmConfig pad_token_idx regression is unrelated." + "reason": "StableLmConfig pad token regression is unrelated to training-resume behavior." }, { "left": "issue:34634", "right": "issue:35545", "accept": false, - "reason": "BarkProcessor voice_preset bug is unrelated to ModernBERT ONNX export." + "reason": "BarkProcessor voice preset bug and ONNX export error are unrelated." }, { "left": "issue:30277", "right": "issue:43638", "accept": false, - "reason": "Both involve training/runtime issues, but the errors and code paths are different." + "reason": "Both mention DeepSpeed, but the failures are different and model-specific." }, { "left": "issue:43596", "right": "issue:43638", "accept": true, - "reason": "Same DeepSpeed ZeRO-3 index-out-of-bounds loading bug on Bert models; only the model wording differs." + "reason": "Same ZeRO-3/BERT IndexError during model init/loading; same concrete failure path, with 43596 being the cleaner canonical report." }, { "left": "issue:43039", "right": "issue:45161", "accept": false, - "reason": "Liger Kernel cross_entropy call and GPT-OSS TP failure are different code paths." + "reason": "Liger cross_entropy routing and GPT-OSS tensor-parallel failure are different bugs." }, { "left": "issue:43975", "right": "issue:44457", "accept": false, - "reason": "Incorrect detokenization and LoRA merge/reload mismatch are different bugs." + "reason": "Detokenization regression and LoRA merge/save/load inconsistency are different issues." }, { "left": "issue:44292", "right": "issue:44457", "accept": false, - "reason": "NVFP4 runtime failure is unrelated to LoRA save/reload inconsistency." + "reason": "Qwen-3 NVFP4 runtime failure and LoRA save/load mismatch are unrelated." }, { "left": "issue:34928", "right": "issue:44805", "accept": false, - "reason": "FSDP checkpointing size mismatch is unrelated to the mask indexing error." + "reason": "Activation-checkpoint/FSDP tensor-size mismatch is a different bug from mask/index errors." }, { "left": "issue:24643", "right": "issue:30277", "accept": false, - "reason": "DeepSpeed weight-dimension error and collective mismatch are different failure modes." + "reason": "Different DeepSpeed failures: 2-D weight error vs collective mismatch." }, { "left": "issue:45859", "right": "issue:45902", "accept": false, - "reason": "Same model family, but CPU offload incompatibility and meta-to-empty uninitialized memory are different concrete bugs." + "reason": "Same Qwen3.5-MoE rotary-embedding area, but CPU-offload incompatibility and meta\u2192to_empty uninitialized-memory are not clearly the same bug." }, { "left": "issue:43638", "right": "issue:45161", "accept": false, - "reason": "Zero-3 Bert loading bug is unrelated to GPT-OSS tensor-parallel failure." + "reason": "BERT/ZeRO-3 loading issue and GPT-OSS tensor-parallel issue are unrelated." }, { "left": "issue:34928", "right": "issue:36331", "accept": false, - "reason": "Checkpointing/FSDP tensor size mismatch is unrelated to CustomTrainer compute_loss signature change." + "reason": "Activation checkpointing FSDP mismatch and custom compute_loss signature error are unrelated." }, { "left": "issue:43596", "right": "issue:45161", "accept": false, - "reason": "DeepSpeed ZeRO-3 Bert loading bug is unrelated to GPT-OSS TP failure." + "reason": "Different models and failure mechanisms; no shared bug." }, { "left": "issue:43716", "right": "issue:45507", "accept": false, - "reason": "Dtype mismatch in Mistral-3 preprocessing is unrelated to GraniteMoEHybrid invalid method calls." + "reason": "Different model families and distinct runtime errors." }, { "left": "issue:45874", "right": "issue:46294", "accept": false, - "reason": "Different Gemma4 input-embedding issue vs Qwen3.5 video chat-template mismatch." + "reason": "Input-embeds reverse lookup cost and video token/template mismatch are unrelated." }, { "left": "issue:39290", "right": "issue:45081", "accept": false, - "reason": "Gemma3/vLLM attribute error and Mistral regex patch crash are unrelated." + "reason": "Gemma3 vLLM config regression and Mistral regex patch crash are different bugs." }, { "left": "issue:45823", "right": "issue:45874", "accept": false, - "reason": "Different Gemma4 bugs: device mismatch in forward vs expensive reverse embedding lookup." + "reason": "Device-map mismatch and expensive reverse embedding lookup are different forward-path issues." }, { "left": "issue:43425", "right": "issue:44292", "accept": false, - "reason": "Torch/version incompatibility is unrelated to Qwen-3 NVFP4 runtime failure." + "reason": "Torch/version compatibility issue and Qwen NVFP4 runtime failure are unrelated." }, { "left": "issue:34928", "right": "issue:41720", "accept": false, - "reason": "FSDP activation checkpointing mismatch and Qwen3 auto device-map assert are unrelated." - }, - { - "left": "issue:43828", - "right": "issue:45161", - "accept": false, - "reason": "Phi-tiny-MoE autocast dtype mismatch is unrelated to GPT-OSS TP failure." - }, - { - "left": "issue:43541", - "right": "issue:43716", - "accept": false, - "reason": "Mixtral grouped_mm tracing error and Mistral-3 preprocessor dtype mismatch are different bugs." + "reason": "FSDP activation-checkpoint mismatch and Qwen3 auto device-map assert are different bugs." }, { "left": "issue:43828", - "right": "issue:44292", - "accept": false, - "reason": "Different models and different dtype/runtime failures." - }, - { - "left": "issue:43716", - "right": "issue:45685", - "accept": false, - "reason": "Image-preprocessor dtype mismatch is unrelated to MPS histogram op failure." - }, - { - "left": "issue:44918", "right": "issue:45161", "accept": false, - "reason": "TRL embedding unpacking issue is unrelated to GPT-OSS tensor-parallel failure." - }, - { - "left": "issue:44292", - "right": "issue:44918", - "accept": false, - "reason": "Different models and code paths; not the same bug." - }, - { - "left": "issue:45507", - "right": "issue:45685", - "accept": false, - "reason": "GraniteMoEHybrid invalid method call and MPS histogram op error are unrelated." - }, - { - "left": "issue:43716", - "right": "issue:44292", - "accept": false, - "reason": "Mistral-3 dtype mismatch and Qwen-3 NVFP4 failure are different issues." - }, - { - "left": "issue:43039", - "right": "issue:44918", - "accept": false, - "reason": "Liger Kernel cross_entropy call is unrelated to TRL SFT embedding unpacking." - }, - { - "left": "issue:41762", - "right": "issue:43638", - "accept": true, - "reason": "Same DeepSpeed ZeRO-3 index-out-of-bounds loading error; only the affected model differs." - }, - { - "left": "issue:43827", - "right": "issue:44512", - "accept": false, - "reason": "Both are docs cleanup, but they refer to different removed features and should stay separate." - }, - { - "left": "issue:43716", - "right": "issue:44560", - "accept": false, - "reason": "Mistral-3 dtype mismatch is unrelated to Qwen3.5 video template token mismatch." - }, - { - "left": "issue:43828", - "right": "issue:45685", - "accept": false, - "reason": "Autocast dtype mismatch is unrelated to MPS integer histogram support." - }, - { - "left": "issue:44918", - "right": "issue:45685", - "accept": false, - "reason": "TRL trainer issue is unrelated to MPS backend limitation." - }, - { - "left": "issue:43039", - "right": "issue:43828", - "accept": false, - "reason": "Liger Kernel cross_entropy usage and Phi-tiny-MoE autocast mismatch are different bugs." + "reason": "Phi dtype/autocast issue and GPT-OSS tensor-parallel failure are unrelated." }, { "left": "issue:30277", "right": "issue:36331", "accept": false, - "reason": "DeepSpeed collective mismatch and compute_loss signature change are unrelated." + "reason": "DeepSpeed collective mismatch and compute_loss signature mismatch are different code paths." }, { "left": "issue:34928", "right": "issue:36010", "accept": false, - "reason": "FSDP checkpointing mismatch is unrelated to GenerationMixin import failure." + "reason": "FSDP recomputed-size mismatch and GenerationMixin import error are unrelated." }, { "left": "issue:45237", "right": "issue:46015", "accept": false, - "reason": "AMD GPU runtime issue is unrelated to outdated RAG docs." + "reason": "GPU/backend runtime failure and outdated RAG docs are unrelated." }, { "left": "issue:43039", "right": "issue:44805", "accept": false, - "reason": "Cross-entropy kernel usage and mask-shape mismatch are unrelated." + "reason": "Liger kernel cross-entropy routing and tensor mask shape mismatch are different issues." }, { "left": "issue:43866", "right": "issue:46402", "accept": false, - "reason": "Corrupted checkpoint and Sam3Model weight-loading regression are different issues." + "reason": "Corrupted checkpoint and missing text-encoder weights are not the same bug." }, { "left": "issue:45800", "right": "issue:46304", "accept": false, - "reason": "Torch/transformers compatibility complaint and USE_DISTRIBUTED=0 regression are unrelated." + "reason": "General torch/transformers incompatibility and USE_DISTRIBUTED=0 regression are different compatibility issues." }, { "left": "issue:43296", "right": "issue:46304", "accept": false, - "reason": "vLLM/PaddleOCR-VL load failure and distributed build regression are different bugs." + "reason": "PaddleOCR-VL/vLLM load failure and distributed-build regression are unrelated." }, { "left": "issue:41762", "right": "issue:43596", "accept": false, - "reason": "Closely related DeepSpeed ZeRO-3 IndexError reports, but the model/context differs enough to keep them separate without more evidence." + "reason": "Same broad ZeRO-3 IndexError theme, but different model family and report context; not clearly the same bug." }, { "left": "issue:39290", "right": "issue:41720", "accept": false, - "reason": "Gemma3/vLLM attribute issue is unrelated to Qwen3 auto device-map cuda assert." + "reason": "Gemma3 vLLM config regression and Qwen3 auto device-map assert are unrelated." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet, and the soft-edge calls are generally conservative. The one clear duplicate cluster is the DeepSpeed ZeRO-3 zero-sized tensor IndexError around Bert/Gemma loading, while the rest are separated by clearly different model, backend, or docs contexts." + "feedback": "The summary is grounded in the packet and stays conservative. The only accepted duplicate pair, issue:43596 vs issue:43638, is well supported by the shared ZeRO-3/BERT IndexError wording, and the remaining soft-edge rejections do not overclaim sameness." }, "evaluator_used": true, "retried": false, @@ -630,8 +546,8 @@ "issue:44292|issue:46015", "issue:39290|issue:43531", "issue:41093|issue:43638", - "issue:43854|issue:43866", "issue:43792|issue:45901", + "issue:43854|issue:43866", "issue:43425|issue:45070", "issue:43596|issue:45800", "issue:43975|issue:45901", @@ -640,8 +556,8 @@ "issue:44292|issue:45901", "issue:39290|issue:45823", "issue:45464|issue:45685", - "issue:43404|issue:43866", "issue:43541|issue:45685", + "issue:43404|issue:43866", "issue:43541|issue:43828", "issue:44841|issue:45084", "issue:36010|issue:39290", @@ -657,14 +573,14 @@ "issue:43366|issue:45464", "issue:43531|issue:45070" ], - "prepared_review_unit_hash": "0d10c028f016d239a66ca3499e4dfe0212ddf46e73cfa1385e13e6eb88278dbc", + "prepared_review_unit_hash": "1309680ea656fd2dc8d79781e0ee3f14e32da7737c8a10662811bad205f7887b", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, @@ -679,238 +595,265 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most pairs are clearly unrelated, with the only strong duplicate signal in the visible v5 AutoTokenizer wrong-token-ID regression cluster. The rest are mostly superficially similar error strings or broad subsystem overlaps without enough evidence of the same underlying bug.", - "confidence": 0.87, - "canonical_issue_reason": "issue:45920 is the best canonical issue because it is the broadest visible report of the v5 AutoTokenizer wrong-token-ID regression and spans multiple model families; issue:45812 is the narrower Granite-specific instance of the same symptom.", + "summary": "This is a heterogeneous cluster of mostly unrelated Transformers issues. The soft edges are largely superficial similarities (same model family, same generic error class, or same subsystem) rather than the same underlying bug/change, so none should be merged as duplicates.", + "confidence": 0.79, + "canonical_issue_reason": "If one issue must be treated as the representative, issue #45812 is the broadest v5 regression here because it affects multiple model families and has the clearest cross-cutting user impact. That said, the cluster does not form a single duplicate family.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45920 is the best global issue anchor in this packet because it describes the broadest visible tokenizer regression and names several affected model families, making it the most representative report here.", + "best_issue_reason": "#45812 is the best overall issue representative because it is the widest-scope regression in the set and most plausibly central to the cluster theme, even though it should not absorb the others as duplicates.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44589", "right": "issue:46304", "accept": false, - "reason": "Different incompatibility classes: missing Float8 storage support versus a USE_DISTRIBUTED=0 regression." + "reason": "Different failures: float8 storage lookup vs distributed-pytorch support. Same broad v5 timing, but not the same bug." }, { "left": "issue:45823", "right": "issue:46294", "accept": false, - "reason": "Different models and failure surfaces: Gemma4 device-map mismatch versus Qwen3.5 video/chat-template token-feature mismatch." + "reason": "Both involve Qwen3.5/Gemma4-era multimodal behavior, but one is a PLE device-mismatch forward bug and the other is a video chat-template/token mismatch." }, { "left": "issue:43901", "right": "issue:44512", "accept": false, - "reason": "Both are docs-related, but they concern unrelated documentation problems." + "reason": "Both are docs-related, but they concern different API surfaces and different kinds of stale documentation." }, { "left": "issue:45237", "right": "issue:45901", "accept": false, - "reason": "Unrelated failures in different areas: GPT-OSS AMD GPU support versus table-question-answering crashes." + "reason": "Both crashy, but on different tasks/models and with different root symptoms; too generic to be the same bug." }, { "left": "issue:44841", "right": "issue:45823", "accept": false, - "reason": "Different models and bugs: Voxtral processor failure versus Gemma4 device mismatch." + "reason": "Different model families and failures: Voxtral processor issue versus Gemma4 PLE device mismatch." }, { "left": "issue:43039", "right": "issue:44560", "accept": false, - "reason": "Different code paths and symptoms: Liger cross_entropy dispatch versus Qwen3-vl video StopIteration." + "reason": "Liger cross-entropy invocation and Qwen3.5 video StopIteration are unrelated beyond both being runtime errors." }, { "left": "issue:43975", "right": "issue:46015", "accept": false, - "reason": "Tokenizer detokenization bug versus outdated RAG examples; not the same issue." + "reason": "DeepSeek detokenization regression is a model-specific bug; the other is generic outdated RAG examples docs work." }, { "left": "issue:45812", "right": "issue:45920", - "accept": true, - "reason": "Same underlying v5 AutoTokenizer regression producing wrong token IDs across multiple model families." + "accept": false, + "reason": "Both are tokenizer regressions, but they affect different model sets and likely different code paths; not enough to unify." }, { "left": "issue:44863", "right": "issue:46015", "accept": false, - "reason": "Model-loading/architecture support issue versus outdated examples; unrelated root causes." + "reason": "NemotronH checkpoint loading and RAG docs are entirely unrelated." }, { "left": "issue:45020", "right": "issue:46304", "accept": false, - "reason": "Broad remote_code breakage is not the same as the distributed-build regression." + "reason": "Remote-code breakage and distributed build/runtime support are distinct compatibility issues." }, { "left": "issue:43866", "right": "issue:44863", "accept": false, - "reason": "Checkpoint corruption versus implementation loading support; different causes." + "reason": "Corrupted Ovis2 checkpoint versus NemotronH load failure are different model/loading problems." }, { "left": "issue:43872", "right": "issue:46304", "accept": false, - "reason": "bitsandbytes constructor mismatch is unrelated to the USE_DISTRIBUTED=0 regression." + "reason": "bitsandbytes initializer incompatibility is unrelated to USE_DISTRIBUTED=0 PyTorch support." }, { "left": "issue:44292", "right": "issue:46015", "accept": false, - "reason": "Model-loading type error versus a documentation issue; no shared bug." + "reason": "Qwen-3-8B-NVFP4 runtime failure and RAG docs are unrelated." }, { "left": "issue:39290", "right": "issue:43531", "accept": false, - "reason": "Both mention sliding_window, but they affect different models and likely different code paths." + "reason": "Both mention sliding_window, but one is a missing config attribute under vLLM and the other is a Qwen3-MoE sliding_window behavior issue." }, { "left": "issue:41093", "right": "issue:43638", "accept": false, - "reason": "Similar IndexError wording, but the shapes and contexts differ enough to treat as separate bugs." + "reason": "Both are mask/index errors with ZeRO-3 context, but one is a shape mismatch and the other is out-of-bounds during BertModel init; likely different causes." + }, + { + "left": "issue:43792", + "right": "issue:45901", + "accept": false, + "reason": "Whisper-large-v2 runtime failure and table-question-answering crash do not point to the same code path." }, { "left": "issue:43854", "right": "issue:43866", "accept": false, - "reason": "Different load failures: unit-test model loading problem versus checkpoint corruption." + "reason": "Model loading in unit tests versus a corrupted checkpoint are different failure modes." }, { - "left": "issue:43792", - "right": "issue:45901", + "left": "issue:43425", + "right": "issue:45070", + "accept": false, + "reason": "Torch 2.10 incompatibility and a v5.4.0 pydantic PretrainedConfig regression are distinct versioning issues." + }, + { + "left": "issue:43596", + "right": "issue:45800", "accept": false, - "reason": "Whisper model execution failure is unrelated to table-question-answering crashes." + "reason": "Both mention ZeRO-3-style init/version problems, but one is BertModel index failure and the other is torch 2.4.1 vs transformers 5.8.0 incompatibility." }, { "left": "issue:43975", "right": "issue:45901", "accept": false, - "reason": "Tokenizer detokenization bug versus table QA crash; unrelated." + "reason": "Different model/task contexts and different symptom classes; no shared bug evidence." }, { "left": "issue:43792", "right": "issue:45070", "accept": false, - "reason": "Whisper runtime failure and pydantic config regression are different bugs." + "reason": "Whisper loading/runtime issue versus pydantic PretrainedConfig regression are not the same problem." }, { "left": "issue:45362", "right": "issue:45464", "accept": false, - "reason": "Same model family, but different entrypoints and symptoms: chat crash versus streaming chat/completions failure." + "reason": "Both involve Qwen3.5, but one is a chat crash and the other is a streaming inference API failure; not enough to merge." + }, + { + "left": "issue:44292", + "right": "issue:45901", + "accept": false, + "reason": "Different model/application areas and no concrete shared failure mechanism." + }, + { + "left": "issue:39290", + "right": "issue:45823", + "accept": false, + "reason": "Shared version era only; Gemma3 sliding_window_pattern attribute error is unrelated to Gemma4 PLE device placement." }, { "left": "issue:45464", "right": "issue:45685", "accept": false, - "reason": "Different subsystems and failure modes: streaming inference versus MPS histogram support." + "reason": "Streaming chat/completions failure and MPS histogram int error are unrelated." + }, + { + "left": "issue:43541", + "right": "issue:45685", + "accept": false, + "reason": "Torch dynamo grouped_mm tracing issue versus MPS int histogram error are different backends and code paths." }, { "left": "issue:43404", "right": "issue:43866", "accept": false, - "reason": "Weight-tying bug in Mistral3 is unrelated to a corrupted checkpoint report." + "reason": "Tied lm_head weights in Mistral3 is a model implementation bug; Ovis2 corrupted checkpoint is a load/data issue." }, { "left": "issue:43541", - "right": "issue:45685", + "right": "issue:43828", "accept": false, - "reason": "Different backend failures: torch dynamo/grouped_mm tracing versus MPS histogram op support." + "reason": "Both are MoE-related, but one is a grouped_mm tracing failure and the other is an autocast dtype mismatch; different underlying bugs." }, { - "left": "issue:43541", - "right": "issue:43828", + "left": "issue:44841", + "right": "issue:45084", "accept": false, - "reason": "Both involve MoE or mixed precision, but the concrete failures and models differ." + "reason": "Processor failure for Voxtral and template-node compilation errors are unrelated." }, { - "left": "issue:44805", - "right": "issue:44918", + "left": "issue:36010", + "right": "issue:39290", "accept": false, - "reason": "Both are runtime errors, but one is a mask-shape IndexError and the other is an embedding-unpacking failure." + "reason": "ImportError for GenerationMixin and Gemma3 sliding_window_pattern error are separate API regressions." }, { - "left": "issue:44387", - "right": "issue:45005", + "left": "issue:45800", + "right": "issue:45823", "accept": false, - "reason": "Quantization OOM regression versus tied-weights issues; different code paths." + "reason": "Torch/transformers version incompatibility and Gemma4 device mismatch are distinct bugs." }, { - "left": "issue:43541", - "right": "issue:45161", + "left": "issue:41093", + "right": "issue:41762", "accept": false, - "reason": "Different models and problems: Mixtral tracing error versus GPT-OSS tensor-parallel-only failure." + "reason": "Both are index errors with ZeRO-3, but one is a mask/tensor length mismatch and the other is zero-sized tensor access while loading Gemma3ForConditionalGeneration." }, { - "left": "issue:43828", - "right": "issue:45464", + "left": "issue:43296", + "right": "issue:43366", "accept": false, - "reason": "Dtype mismatch under autocast is unrelated to the Qwen3.5 streaming API failure." + "reason": "PaddleOCR-VL loading in vLLM and GGUF gpt-oss support are different model-compatibility requests." }, { - "left": "issue:44291", + "left": "issue:43572", "right": "issue:45070", "accept": false, - "reason": "Unexpected _is_hf_initialized argument and a PretrainedConfig pydantic-field regression are not the same bug." + "reason": "Missing pad_token_idx in StableLmConfig is a model config regression; pydantic PretrainedConfig breakage is a framework-level API issue." }, { - "left": "issue:43366", - "right": "issue:45464", + "left": "issue:44805", + "right": "issue:44918", "accept": false, - "reason": "GGUF gpt-oss support request is unrelated to the Qwen3.5 streaming API bug." + "reason": "Both are tensor/embedding shape problems, but one is a mask indexing mismatch and the other is unpacking Qwen3.5 input embeddings with TRL SFT trainer." }, { - "left": "issue:43531", - "right": "issue:45070", + "left": "issue:44387", + "right": "issue:45005", "accept": false, - "reason": "Sliding-window model issue versus pydantic PretrainedConfig regression; different root causes." + "reason": "Int4 OOM regression and tied-weights translation-model issues are different change sets and failure modes." }, { - "left": "issue:36010", - "right": "issue:39290", + "left": "issue:43541", + "right": "issue:45161", "accept": false, - "reason": "GenerationMixin import failure and Gemma3 sliding_window_pattern are unrelated." + "reason": "Grouped_mm tracing crash and GPT-OSS only-TP-not-working are separate MoE/tracing versus parallelism issues." }, { - "left": "issue:45800", - "right": "issue:45823", + "left": "issue:43828", + "right": "issue:45464", "accept": false, - "reason": "Torch/transformers compatibility issue versus Gemma4 forward device mismatch." + "reason": "Autocast dtype mismatch for Phi-tiny-MoE and Qwen3.5 streaming API failure are unrelated." }, { - "left": "issue:41093", - "right": "issue:41762", + "left": "issue:44291", + "right": "issue:45070", "accept": false, - "reason": "Both are Zero-3-related IndexErrors, but the reported shapes and model contexts do not indicate the same bug." + "reason": "init_empty_weights/_is_hf_initialized TypeError and pydantic PretrainedConfig regression are related only by broad v5 breakage, not by code path." }, { - "left": "issue:43296", - "right": "issue:43366", + "left": "issue:43366", + "right": "issue:45464", "accept": false, - "reason": "PaddleOCR-VL load failure and GGUF gpt-oss support request are different model-specific issues." + "reason": "GGUF gpt-oss support and Qwen3.5 streaming failure are different features and bugs." }, { - "left": "issue:43572", + "left": "issue:43531", "right": "issue:45070", "accept": false, - "reason": "Missing pad_token_idx in StableLmConfig and a pydantic PretrainedConfig regression are different config bugs." + "reason": "Qwen3-MoE sliding_window behavior and pydantic model-field regression are not the same issue." } ] }, - "evaluator_result": { - "accept": false, - "feedback": "The pairwise judgments are mostly conservative and grounded in the titles, but the top-level issue reasoning overstates the packet: it refers to an \"accepted duplicate family,\" which is not established in the provided data. I\u2019d want the global-anchor/canonical-issue claims phrased more cautiously and only in terms visible in the packet (broadest v5 tokenizer regression, multiple model families, etc.)." - }, - "evaluator_used": true, - "retried": true, - "accepted_nontrivial_soft_edge": true, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -1010,14 +953,14 @@ "issue:43866|issue:44403", "issue:38175|issue:43054" ], - "prepared_review_unit_hash": "db9132f1972d069a9d611394def93a6cf6aaaae2dcf439f7658a73da63139c23", + "prepared_review_unit_hash": "d63fea1f3dee2113b0e6c092a44ac5c63fd28c3373d7891f3f00f12e64d6b20f", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 47, @@ -1032,219 +975,219 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a heterogeneous set of unrelated closed issues. The only clear duplicate thread is the Transformers 5.x non-persistent buffer corruption regression around from_pretrained / buffer materialization.", - "confidence": 0.91, - "canonical_issue_reason": "issue:43950 is the clearest root report of the non-persistent buffer corruption regression, with the most direct repro and wording; issue:43644 is the earlier broader report and issue:44534 is a rephrased follow-up.", + "summary": "Only the non-persistent-buffer corruption chain looks like a clear duplicate set. The SigLIP2, Qwen video, custom-module caching, and other same-subsystem pairs are too weakly evidenced from titles alone to merge conservatively.", + "confidence": 0.84, + "canonical_issue_reason": "issue:43950 is the clearest duplicate hub: it states the non-persistent-buffer corruption regression directly and sits between the exact/near-exact duplicate phrasing in issue:43644 and issue:44534.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43950 is the best representative duplicate target because it describes the concrete failure mode precisely and anchors the confirmed duplicate chain in this cluster.", + "best_issue_reason": "issue:43950 is the best global issue candidate because it most precisely captures a concrete regression with the strongest duplicate evidence and a clear underlying fix target.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43716", "right": "issue:46294", "accept": false, - "reason": "Different models and different failure modes (dtype mismatch vs video chat template token/feature mismatch)." + "reason": "Different model families and failure modes; only a broad multimodal theme overlaps." }, { "left": "issue:45020", "right": "issue:45800", "accept": false, - "reason": "One is a remote_code regression, the other is a torch/transformers version compatibility problem; not the same bug." + "reason": "One is a remote_code regression; the other is a torch/transformers version incompatibility." }, { "left": "issue:43541", "right": "issue:44589", "accept": false, - "reason": "Unrelated runtime errors in different code paths (grouped_mm tracing vs missing Float8 storage)." + "reason": "Different runtime failures in unrelated code paths." }, { "left": "issue:44479", "right": "issue:46294", "accept": false, - "reason": "Both mention Qwen video, but one is a broad v5.3.0 regression and the other is a specific apply_chat_template mismatch; not clearly the same bug." + "reason": "Both involve Qwen video inputs, but the packet does not show the same concrete bug or a duplicate chain." }, { "left": "issue:41093", "right": "issue:43596", "accept": false, - "reason": "Both are IndexErrors under DeepSpeed, but the concrete shapes and trigger differ." + "reason": "Both are index errors, but the shapes, triggers, and model contexts differ." }, { "left": "issue:43054", "right": "issue:43994", - "accept": false, - "reason": "Same model family, but one reports degraded embeddings and the other nonsensical pipeline output; too vague to be the same concrete bug." + "accept": true, + "reason": "Both report the same SigLIP2 bad-output regression: degraded/nonsensical text-embedding behavior for google/siglip2-base-patch16-224." }, { "left": "issue:45759", "right": "issue:45902", "accept": false, - "reason": "Different issues: composite config unwrapping vs uninitialized rotary embedding memory after materialization." + "reason": "Different composite-config loading bug vs buffer-materialization bug." }, { "left": "issue:45542", "right": "issue:45800", "accept": false, - "reason": "TensorBoard/TensorFlow backend handling is unrelated to the torch 2.4.1/transformers 5.8.0 incompatibility report." + "reason": "TensorBoard/TensorFlow backend issue is unrelated to version compatibility." }, { "left": "issue:39186", "right": "issue:44918", "accept": false, - "reason": "FSDP 2-D weight error and Qwen3.5 SFT embedding unpacking are different failures." + "reason": "Different model setups and different concrete failures." }, { "left": "issue:43716", "right": "issue:44918", "accept": false, - "reason": "Different models and subsystems; image preprocessor dtype mismatch is unrelated to TRL SFT embedding unpacking." + "reason": "Different model families and unrelated dtype vs embedding-unpack failures." }, { "left": "issue:40990", "right": "issue:44960", "accept": false, - "reason": "Completely different models and symptoms (perplexity vs GLM5-related report)." + "reason": "Completely different models and symptoms; only the general training/eval context overlaps." }, { "left": "issue:45632", "right": "issue:45698", "accept": false, - "reason": "Both involve custom-code loading, but one is a leaf-directory cache collision and the other is wrong module loading after save_pretrained; not clearly the same bug." + "reason": "Cache-path collision and wrong custom module after save/load are related, but not the same bug from the packet alone." }, { "left": "issue:43645", "right": "issue:45399", "accept": false, - "reason": "Custom model init in Jupyter and flash-attn2 fallback gating are unrelated." + "reason": "Notebook model initialization regression is unrelated to flash-attn fallback gating." }, { "left": "issue:43404", "right": "issue:45005", "accept": false, - "reason": "Both mention tied weights, but they concern different model families and different manifestations." + "reason": "Both mention tied weights, but the concrete model classes and bugs differ." }, { "left": "issue:42915", "right": "issue:45399", "accept": false, - "reason": "FineGrainedFP8 failure and flash-attn2 fallback checks are unrelated." + "reason": "FP8 config failure is unrelated to flash-attn fallback checks." }, { "left": "issue:44530", "right": "issue:45005", "accept": false, - "reason": "PagedAttentionCache group-type crash is unrelated to translation-model tied-weight issues." + "reason": "PagedAttention cache crash is not the same as a tied-weights loading regression." }, { "left": "issue:43873", "right": "issue:45005", "accept": false, - "reason": "Quantization/offloading behavior does not match the tied-weights regression." + "reason": "Quantization/offloading behavior is a different problem from tied-weight loading." }, { "left": "issue:44863", "right": "issue:45005", "accept": false, - "reason": "NemotronH checkpoint-loading failure is unrelated to translation-model weight tying." + "reason": "NemotronH checkpoint loading is a separate model-specific bug." }, { "left": "issue:44360", "right": "issue:44512", "accept": false, - "reason": "A DSA indexer bug and a docs typo are unrelated." + "reason": "Indexer discussion is unrelated to a docs wording issue." }, { "left": "issue:44918", "right": "issue:45507", "accept": false, - "reason": "Different model families and different error mechanisms." + "reason": "Different models and failures; no shared code-path bug is evident." }, { "left": "issue:44492", "right": "issue:44512", "accept": false, - "reason": "A cache-strategy typo and a removed CLI docs mention are unrelated." + "reason": "Separate documentation/wording issues." }, { "left": "issue:43975", "right": "issue:44292", "accept": false, - "reason": "Detokenization regression and Qwen NVFP4 runtime error are different bugs." + "reason": "Different models and symptoms; no concrete shared bug is shown." }, { "left": "issue:44292", "right": "issue:45507", "accept": false, - "reason": "Different model-specific runtime failures." + "reason": "Different model-loading/runtime issues." }, { "left": "issue:43950", "right": "issue:44534", "accept": true, - "reason": "Both report the same non-persistent-buffer corruption regression in Transformers 5.x; 44534 is a reworded follow-up to 43950." + "reason": "Same non-persistent-buffer corruption regression, described with different wording." }, { "left": "issue:39692", "right": "issue:43054", "accept": false, - "reason": "SigLIP2 docs example problems and SigLIP2 embedding quality regression are related only at a broad model level, not the same concrete bug." + "reason": "Docs-example errors and SigLIP2 embedding-quality regressions are related only thematically." }, { "left": "issue:43644", "right": "issue:43950", "accept": true, - "reason": "Same regression: non-persistent buffers get corrupted/junk-filled during loading/materialization in Transformers 5.x." + "reason": "Exact same non-persistent-buffer junk/corruption regression." }, { "left": "issue:43828", "right": "issue:44928", "accept": false, - "reason": "Autocast dtype mismatch in Phi-tiny-MoE and Qwen3.5 RLHF NaN explosion are different issues." + "reason": "Both involve numeric instability, but the trigger and failure mechanism differ." }, { "left": "issue:36331", "right": "issue:41762", "accept": false, - "reason": "Trainer API keyword mismatch and DeepSpeed ZeRO-3 load error are unrelated." + "reason": "Trainer API signature mismatch and ZeRO-3 loading error are unrelated." }, { "left": "issue:43872", "right": "issue:44589", "accept": false, - "reason": "Different quantization-related incompatibilities with different symptoms." + "reason": "Different quantization/storage incompatibilities." }, { "left": "issue:43950", "right": "issue:45902", "accept": false, - "reason": "Both mention buffer/materialization issues, but 45902 is a specific uninitialized rotary embedding case and not the same regression." + "reason": "Both touch non-persistent buffers, but the triggering paths differ enough to keep separate." }, { "left": "issue:43054", "right": "issue:45701", "accept": false, - "reason": "Model embedding quality and tokenization changes across versions are different problems." + "reason": "SigLIP2 output quality and tokenization-version changes are distinct issues." }, { "left": "issue:43866", "right": "issue:44403", "accept": false, - "reason": "Checkpoint corruption and noisy loading are unrelated." + "reason": "Checkpoint corruption is unrelated to loading noise/warnings." }, { "left": "issue:38175", "right": "issue:43054", "accept": false, - "reason": "Both are SigLIP2 complaints, but zero probabilities and poor embeddings are not specific enough to treat as the same bug." + "reason": "Same model family, but the packet does not demonstrate the same underlying bug." } ] }, "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet and the soft-edge verdicts are conservative overall. The accepted duplicate thread around 43644/43950/44534 is well supported by the issue titles, and the other pair rejections avoid over-claiming sameness. Minor caution: a couple of phrases like \"most direct repro\" go a bit beyond what the packet explicitly shows, but not enough to invalidate the cluster assessment." + "accept": false, + "feedback": "The summary is mostly grounded, but a few soft-edge accepts overstate the evidence. In particular, `issue:45632` vs `issue:45698` is not clearly the same bug from the packet alone (cache-path collision vs wrong custom module after save/load), and `issue:44479` vs `issue:46294` is also only title-level similarity rather than a demonstrated duplicate chain. `issue:38175` vs `issue:43054` is plausible, but still weakly supported. Because the instructions call for conservative soft-edge verdicts, I can\u2019t accept this as-is." }, "evaluator_used": true, - "retried": false, + "retried": true, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -1312,8 +1255,8 @@ ], "soft_pairs": [ "issue:41553|issue:43883", - "issue:44910|issue:45859", "issue:45663|issue:45902", + "issue:44910|issue:45859", "issue:42947|issue:44387", "issue:43054|issue:43493", "issue:41553|issue:43881", @@ -1343,226 +1286,226 @@ "issue:43792|issue:44220", "issue:43257|issue:45305" ], - "prepared_review_unit_hash": "173a96ffbeddc66583a1e2fd4f903d1559bf57ceb1133086007f453a9906736e", + "prepared_review_unit_hash": "d84ed3afc9f7cb676581f522bf8a3bc3d04106e9783cc33999f079ab2edc2605", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 31, - "serialized_chars": 36798, - "estimated_input_tokens": 9200, - "estimated_eval_tokens": 18656 + "serialized_chars": 36795, + "estimated_input_tokens": 9199, + "estimated_eval_tokens": 18654 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a grab bag of distinct v5 regression reports around model loading, tokenizers, MoE, quantization, and docs. Only two soft pairs look like true duplicates: the stale v5 docs/pipeline references and the `_is_hf_initialized` init_empty_weights/Int8Params TypeError.", - "confidence": 0.57, - "canonical_issue_reason": "issue:45020 is the broadest umbrella report for recent Transformers v5 loading/remote-code regressions, making it the best representative of the cluster\u2019s shared theme.", + "summary": "Most items are separate model-specific loading or training regressions, not duplicates. Only two pairs look like clear duplicates: the v5 pipeline-docs cleanup and the `_is_hf_initialized` constructor incompatibility.", + "confidence": 0.76, + "canonical_issue_reason": "Issue 45020 is the broadest representative of the cluster\u2019s main theme: recent Transformers v5 loading regressions affecting remote-code/model loading across multiple models.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45020 is the most central and general issue here; the rest are narrower, model-specific failures or unrelated regressions.", + "best_issue_reason": "Issue 45020 is the strongest single anchor for triage because it is cross-cutting, clearly describes a user-facing regression, and is broad enough to represent the surrounding loading-breakage reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:41553", "right": "issue:43883", "accept": false, - "reason": "Different models and failure modes: Voxtral has a bad AutoTokenizer error message, Molmo is a missing `all_tied_weights_keys` attribute." + "reason": "Different models and different failure modes: tokenizer error messaging vs missing tied-weights attribute on Molmo." }, { - "left": "issue:44910", - "right": "issue:45859", + "left": "issue:45663", + "right": "issue:45902", "accept": false, - "reason": "Both involve Qwen3.5/MoE, but one is a FlashAttention illegal memory access from `position_ids`, the other is a CPU offload compatibility bug in rotary embedding." + "reason": "Different models and code paths: Gemma4 FSDP2 shared_kv_states bug vs Qwen3_5MoE rotary embedding materialization bug." }, { - "left": "issue:45663", - "right": "issue:45902", + "left": "issue:44910", + "right": "issue:45859", "accept": false, - "reason": "Distinct bugs: Gemma4 FSDP2 `shared_kv_states` reconstruction vs Qwen3_5Moe `inv_freq` uninitialized memory after `to_empty(cuda)`." + "reason": "Both are Qwen3-family issues, but one is Flash Attention position_ids handling and the other is CPU-offload compatibility in rotary embedding." }, { "left": "issue:42947", "right": "issue:44387", "accept": false, - "reason": "Different regressions: ineffective gradient checkpointing with LoRA versus extra reserved memory/OOM under int4 quantization." + "reason": "Different regressions: PEFT LoRA gradient checkpointing vs int4 quantization memory growth." }, { "left": "issue:43054", "right": "issue:43493", "accept": false, - "reason": "Related model family, but not the same bug: one reports worse SigLIP2 text embeddings, the other a HF-vs-JAX implementation discrepancy." + "reason": "Related to SigLIP2, but one is a quality complaint and the other is an HF-vs-JAX implementation discrepancy; not clearly the same bug." }, { "left": "issue:41553", "right": "issue:43881", "accept": false, - "reason": "Separate loading problems for different models; no clear shared code-path or symptom beyond generic load failure." + "reason": "Different model-specific loading problems with no shared code-path evidence." }, { "left": "issue:44403", "right": "issue:44451", "accept": false, - "reason": "One is cosmetic loading noise, the other is an actual model load failure for ScandiBERT." + "reason": "One is a noisy-warning complaint, the other is a concrete model load failure." }, { "left": "issue:43975", "right": "issue:45520", "accept": false, - "reason": "Unrelated subsystems: tokenizer detokenization bug versus `flash_attn` import-time `KeyError`." + "reason": "Different subsystems: tokenizer detokenization regression vs import_utils flash-attn KeyError." }, { "left": "issue:44403", "right": "issue:44488", "accept": false, - "reason": "Generic warning/noise issue is not the same as the specific `cjvt/sleng-bert` loading failure." + "reason": "Same general loading area, but one is noise and the other is a model-specific load failure." }, { "left": "issue:43643", "right": "issue:45020", "accept": false, - "reason": "Both involve `trust_remote_code`, but one is missing fields in `AutoConfig`, while the other is a broader remote-code breakage report." + "reason": "Both involve remote-code loading, but 43643 is a specific missing-fields bug in AutoConfig while 45020 is a broader compatibility report across models." }, { "left": "issue:42947", "right": "issue:43856", "accept": false, - "reason": "Different problems: LoRA gradient checkpointing versus Qwen3 MoE training memory usage." + "reason": "Different problems: LoRA/gradient-checkpointing behavior vs Qwen3 MoE training memory efficiency." }, { "left": "issue:41762", "right": "issue:42915", "accept": false, - "reason": "Different models and errors: Gemma3 ZeRO-3 index error versus Qwen3 MOE FineGrainedFP8 failure." + "reason": "Both are model-loading issues, but Gemma3 ZeRO-3 index error and Qwen3 MoE FP8 failure are unrelated code paths." }, { "left": "issue:45685", "right": "issue:45823", "accept": false, - "reason": "MPS histogram int limitation in MoE is unrelated to Gemma4 PLE device mismatch in forward." + "reason": "Different device/runtime issues: MPS histogram op limitation vs Gemma4 PLE device mismatch." }, { "left": "issue:43782", "right": "issue:44960", "accept": false, - "reason": "Different model families and failure types; one is a Qwen3VL `weight_only` load error, the other is an unspecified GLM5 issue." + "reason": "Different models and symptoms; no evidence they share the same underlying bug." }, { "left": "issue:43824", "right": "issue:43872", "accept": false, - "reason": "Import error for `Qwen2_5_VLForConditionalGeneration` is unrelated to the bitsandbytes `_is_hf_initialized` compatibility problem." + "reason": "Different incompatibilities: missing Qwen2_5_VL import vs bitsandbytes `_is_hf_initialized` constructor error." }, { "left": "issue:44991", "right": "issue:45399", "accept": false, - "reason": "Tokenizer load failure for a specific model versus FlashAttention2 fallback being blocked by unrelated checks." + "reason": "Tokenizer load failure and flash-attn fallback logic are unrelated." }, { "left": "issue:43881", "right": "issue:43883", "accept": false, - "reason": "Both are loading failures, but for different models and root causes; no evidence they are the same bug." + "reason": "Different model failures: GLM-4V loading vs Molmo tied-weights attribute error." }, { "left": "issue:45020", "right": "issue:45823", "accept": false, - "reason": "Remote-code loading regressions are not the same as Gemma4 PLE device placement bugs." + "reason": "Remote-code loading regressions and Gemma4 PLE device placement are unrelated." }, { "left": "issue:42491", "right": "issue:44368", "accept": false, - "reason": "Same broad Qwen3.5 LoRA area, but one is a hf4.x\u2192hf5.x model incompatibility and the other is a tie-word-embeddings warning." + "reason": "Both mention Qwen MoE/LoRA, but one is a version-compatibility failure and the other is a tied-embeddings warning." }, { "left": "issue:44661", "right": "issue:45020", "accept": false, - "reason": "`add-new-model-like` failing in `TOKENIZER_MAPPING_NAMES` is a different registry bug than remote-code loading breakage." + "reason": "A tokenizer-mapping edge case is not the same bug as remote-code breakage." }, { "left": "issue:43828", "right": "issue:45823", "accept": false, - "reason": "Autocast dtype mismatch in Phi-tiny-MoE and Gemma4 PLE device mismatch are different code paths." + "reason": "Different runtime failures: autocast dtype mismatch vs device_map placement mismatch." }, { "left": "issue:45081", "right": "issue:45399", "accept": false, - "reason": "Mistral regex patch AttributeError is unrelated to FlashAttention2 fallback gating." + "reason": "Tokenizer regex patch crash and flash-attn fallback checks are separate issues." }, { "left": "issue:45715", "right": "issue:46315", "accept": false, - "reason": "Both are tokenizer performance regressions, but they affect different APIs and likely different hot paths (`convert_ids_to_tokens` vs `convert_tokens_to_ids`)." + "reason": "Both are tokenizer performance regressions, but they affect opposite conversion directions and different hot paths." }, { "left": "issue:45020", "right": "issue:45356", "accept": false, - "reason": "`trust_remote_code` loading breakage is not the same as the Kimi-K2.5 tokenizer codec regression." + "reason": "General remote-code breakage is not the same as the Kimi-K2.5 tokenizer codec regression." }, { "left": "issue:43742", "right": "issue:45399", "accept": false, - "reason": "Different areas: MobileLLM loading key error versus FlashAttention2 fallback selection." + "reason": "KeyError loading MobileLLM and flash-attn fallback gating are unrelated." }, { "left": "issue:43404", "right": "issue:43950", "accept": false, - "reason": "`lm_head` tying bug and non-persistent buffer corruption are separate state-management regressions." + "reason": "lm_head tying in Mistral3 is unrelated to silent corruption of non-persistent buffers." }, { "left": "issue:43827", "right": "issue:44509", "accept": true, - "reason": "These describe the same docs regression: stale v5 documentation still referencing removed pipeline tasks like summarization/translation/text2text-generation." + "reason": "Same docs cleanup: both note pipeline tasks removed in v5 and still referenced in documentation." }, { "left": "issue:43872", "right": "issue:44291", "accept": true, - "reason": "Same underlying compatibility bug: `init_empty_weights` / bitsandbytes passes `_is_hf_initialized` into `Int8Params.__new__`, causing the same TypeError." + "reason": "Same underlying `_is_hf_initialized` incompatibility during loading/init_empty_weights, described from two angles." }, { "left": "issue:42915", "right": "issue:43872", "accept": false, - "reason": "Different issues: Qwen3 MOE FineGrainedFP8 failure versus bitsandbytes `_is_hf_initialized` TypeError." + "reason": "Different Qwen/DeepSpeed and bitsandbytes incompatibility bugs." }, { "left": "issue:43792", "right": "issue:44220", "accept": false, - "reason": "Insufficient evidence that the Whisper load failure and `_torch_extract_fbank_features()` issue are the same bug; could be related but not clearly identical." + "reason": "The Whisper load failure is too generic to confirm it is the same bug as the fbank-feature helper issue." }, { "left": "issue:43257", "right": "issue:45305", "accept": false, - "reason": "Different DeepSpeed-related problems: Qwen3 MOE weight conversion on load versus gradient averaging under GAS with `model_accepts_loss_kwargs=True`." + "reason": "Both involve DeepSpeed/Qwen3-ish training, but one is weight conversion and the other is gradient averaging with GAS." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative on the soft edges. The two accepted pairs are well supported as same-bug/doc-duplicate cases, and the rejected pairs avoid overclaiming shared root cause." + "feedback": "The summary is well-grounded in the packet and stays conservative on the soft-edge calls. The two accepted pairs are plausibly the same bugfix/doc cleanup, and the rejected pairs are mostly justified by model- or code-path-specific differences. No clear overstatement of evidence." }, "evaluator_used": true, "retried": false, @@ -1635,10 +1578,10 @@ "soft_pairs": [ "issue:36331|issue:41093", "issue:42947|issue:45923", - "issue:43940|issue:45313", "issue:41553|issue:44991", - "issue:43854|issue:44863", + "issue:43940|issue:45313", "issue:44928|issue:45588", + "issue:43854|issue:44863", "issue:41093|issue:44805", "issue:44479|issue:44560", "issue:44863|issue:46402", @@ -1666,235 +1609,235 @@ "issue:38175|issue:43994", "issue:43638|issue:44661" ], - "prepared_review_unit_hash": "acaa79f4964227d1522ad4e4eb03c600d87eaaf1e19d2c290994db6c1ca86393", + "prepared_review_unit_hash": "5d256cfc11c36bff7172663c9ddf9a61ab39c79142c828b71679e26358c3b64e", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 32, - "serialized_chars": 37482, - "estimated_input_tokens": 9371, - "estimated_eval_tokens": 18998 + "serialized_chars": 37486, + "estimated_input_tokens": 9372, + "estimated_eval_tokens": 19000 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab bag of unrelated Transformers regressions across loading, tokenization, attention, and training. Two likely duplicate mini-clusters stand out: the Transformers 5.0.0 custom model initialization breakage, and the SigLIP2 zero-probability/nonsensical-output regression.", - "confidence": 0.87, - "canonical_issue_reason": "issue:43646 is the clearest broad duplicate target for the Transformers 5.0.0 custom model initialization regression; issue:43645 looks like a narrower notebook-specific repro of the same breakage.", + "summary": "Mostly a grab bag of unrelated Transformers bug reports, with a few clear duplicate-like pairs around the same model or error path. I\u2019d keep the more general custom-model init issue as the canonical issue.", + "confidence": 0.81, + "canonical_issue_reason": "issue:43646 is the broad, general statement of the custom-model initialization regression; issue:43645 looks like a notebook-specific variant of the same problem.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43994 is the strongest standalone issue to keep for the SigLIP2 cluster: it has a concrete model-level repro and matches issue:38175\u2019s zero-probability / nonsensical-output symptoms.", + "best_issue_reason": "issue:43646 is the cleanest representative of the duplicate pair because it describes the underlying regression without the extra notebook-specific constraint.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:36331", "right": "issue:41093", "accept": false, - "reason": "Different bugs: a custom trainer API kwarg mismatch vs a tensor/mask shape mismatch." + "reason": "Both are training-time errors, but the compute_loss kwarg mismatch and the mask/tensor shape mismatch are different failures and different code paths." }, { "left": "issue:42947", "right": "issue:45923", "accept": false, - "reason": "Both mention gradient checkpointing, but one is ineffective LoRA checkpointing behavior and the other is a missing capability flag." - }, - { - "left": "issue:43940", - "right": "issue:45313", - "accept": false, - "reason": "Both are ZeRO-3 loading failures, but they affect different Qwen model families and report different concrete failures." + "reason": "Both mention gradient checkpointing, but one is about PEFT LoRA behavior and the other is a missing flag on a trust_remote_code model; not the same bug." }, { "left": "issue:41553", "right": "issue:44991", "accept": false, - "reason": "Tokenizer loading failures, but for different models and different failure modes." + "reason": "Both concern tokenizer loading, but they affect different models and one is an error-message issue while the other is a hard load failure." }, { - "left": "issue:43854", - "right": "issue:44863", + "left": "issue:43940", + "right": "issue:45313", "accept": false, - "reason": "Different model-loading regressions on different architectures; not the same underlying bug." + "reason": "Both are ZeRO-3 weight-loading problems, but they target different Qwen model families and appear to be separate model-specific failures." }, { "left": "issue:44928", "right": "issue:45588", "accept": false, - "reason": "One is a Qwen3.5 RLHF gradient explosion from 3D position_ids/SDPA behavior; the other is a flash_attention.py None-handling crash." + "reason": "One is a Qwen3.5 RLHF gradient explosion from 3D position_ids; the other is a flash-attention AttributeError on sink-less models. Different issues." + }, + { + "left": "issue:43854", + "right": "issue:44863", + "accept": false, + "reason": "Different models and different loading failures; too broad to treat as the same underlying bug." }, { "left": "issue:41093", "right": "issue:44805", - "accept": false, - "reason": "Both are mask/index shape errors, but the reported shapes and contexts differ enough to indicate separate bugs." + "accept": true, + "reason": "These describe the same mask-vs-indexed-tensor shape mismatch IndexError, with essentially the same error signature." }, { "left": "issue:44479", "right": "issue:44560", - "accept": false, - "reason": "Both involve video on Qwen variants, but one is a broad video-input regression and the other is a StopIteration in a different embedding path." + "accept": true, + "reason": "Both are Qwen3-VL family video-input regressions in the same release line, and the StopIteration looks like the same broken video path." }, { "left": "issue:44863", "right": "issue:46402", "accept": false, - "reason": "Separate loading problems for different model families and different code paths." + "reason": "Different architectures and different loading failures; no strong evidence of a shared root cause." }, { "left": "issue:41553", "right": "issue:44488", "accept": false, - "reason": "Both are model/tokenizer loading issues, but they involve unrelated models and symptoms." + "reason": "Both are loading-related, but one is Voxtral AutoTokenizer messaging and the other is a separate tokenizer load failure for a different model." }, { "left": "issue:24643", "right": "issue:30064", "accept": false, - "reason": "Training with DeepSpeed weight-shape error vs image processor failing on void segmentation maps; unrelated." + "reason": "Completely different subsystems: DeepSpeed training vs image processor segmentation maps." }, { "left": "issue:44530", "right": "issue:44863", "accept": false, - "reason": "PagedAttentionCache invalid group-type crash vs NemotronH checkpoint loading failure; different code paths." + "reason": "Qwen3.5 PagedAttentionCache/linear_attention crash is unrelated to NemotronH checkpoint loading." }, { "left": "issue:45325", "right": "issue:45902", "accept": false, - "reason": "Different Qwen3.5-related bugs: one is temporal position_id scaling, the other is uninitialized memory after meta-to-empty materialization." + "reason": "Both touch rotary/position handling, but one is Qwen2.5-VL video still-image scaling and the other is a meta/to_empty initialization memory bug." }, { "left": "issue:43366", "right": "issue:43828", "accept": false, - "reason": "GGUF gpt-oss support vs Phi MoE autocast dtype mismatch; unrelated." + "reason": "Different models and different failure modes: GGUF architecture support vs autocast dtype mismatch." }, { "left": "issue:45663", "right": "issue:45823", "accept": false, - "reason": "Same model family, but different bugs: per-layer shared_kv_states rebuild vs device_map device mismatch." + "reason": "Same model family, but one is an FSDP2 KeyError in attention and the other is a device-map forward mismatch; different code paths." }, { "left": "issue:43366", "right": "issue:43531", "accept": false, - "reason": "GGUF architecture support issue vs Qwen3-MoE sliding_window behavior; unrelated." + "reason": "No overlap beyond broad model-family similarity; gpt-oss GGUF support is unrelated to Qwen3-MoE sliding_window behavior." }, { "left": "issue:44155", "right": "issue:45584", "accept": false, - "reason": "AudioFlamingo3 track leakage vs Whisper empty-transcription handling; different bugs." + "reason": "AudioFlamingo3 batched inference leakage and Whisper empty-transcription generation are unrelated bugs." }, { "left": "issue:43950", "right": "issue:45834", "accept": false, - "reason": "Non-persistent buffer corruption regression vs Kosmos2.5 long-OCR index error; different code paths." + "reason": "Non-persistent buffer corruption during from_pretrained is unrelated to Kosmos2.5 OCR index errors." }, { "left": "issue:44560", "right": "issue:45685", "accept": false, - "reason": "Qwen3-vl video StopIteration vs MoE MPS histogram kernel not implemented for Int." + "reason": "Video-input regression and an MPS histogram-not-implemented MoE issue are unrelated." }, { "left": "issue:43824", "right": "issue:43957", "accept": false, - "reason": "Missing import symbol for a specific class vs meta-device model loading regression." + "reason": "Import/export of a specific class and meta-device loading breakage are different issues." }, { "left": "issue:42915", "right": "issue:43957", "accept": false, - "reason": "GenerationMixin import failure vs meta-device loading breakage; unrelated failures." + "reason": "FineGrainedFP8Config failure for Qwen3Moe is unrelated to meta-device model loading." }, { "left": "issue:43366", "right": "issue:44589", "accept": false, - "reason": "GGUF support request vs Float8 storage lookup error; not the same bug." + "reason": "GGUF support for gpt-oss and Float8 storage lookup failure are different problems." }, { "left": "issue:43550", "right": "issue:44910", "accept": false, - "reason": "Bamba torch.compile/SDPA failure vs Qwen3.5 Flash Attention illegal-memory-access bug from 3D position_ids." + "reason": "Different code paths: torch.compile/SDPA on Bamba versus flash-attention with 3D position_ids on Qwen3.5." }, { "left": "issue:43645", "right": "issue:43646", "accept": true, - "reason": "Same Transformers 5.0.0 regression: custom model initialization breaks, with 43645 just being a notebook-specific repro of the broader issue." + "reason": "Same custom model initialization regression; one report is just notebook-specific while the other is the general failure." }, { "left": "issue:44560", "right": "issue:44918", "accept": false, - "reason": "Different failures: video StopIteration in Qwen3-vl-embedding vs TRL SFT unpacking of Qwen3.5 embeddings." + "reason": "Both involve Qwen3.5/VL embeddings, but one is a video StopIteration regression and the other is TRL SFT unpacking failure." }, { "left": "issue:42617", "right": "issue:43366", "accept": false, - "reason": "3d_parallel.py execution failure vs GGUF model architecture support; unrelated." + "reason": "A script execution problem and GGUF architecture support are not the same bug." }, { "left": "issue:36010", "right": "issue:42915", "accept": false, - "reason": "Missing GenerationMixin import vs FineGrainedFP8Config failure in Qwen3Moe; different problems." + "reason": "GenerationMixin import failure and Qwen3Moe FP8 config failure are unrelated." }, { "left": "issue:44661", "right": "issue:45823", "accept": false, - "reason": "Tokenizer mapping-name bug vs Gemma4 PLE device mismatch; unrelated." + "reason": "Tokenizer-mapping add-new-model-like failure is unrelated to Gemma4 PLE device mismatch." }, { "left": "issue:44560", "right": "issue:44805", "accept": false, - "reason": "Different symptoms and contexts: video StopIteration vs mask shape mismatch." + "reason": "Both are IndexError-shaped titles, but one is a Qwen3-VL video issue and the other is a generic mask/tensor mismatch; not enough to merge." }, { "left": "issue:43632", "right": "issue:45834", "accept": false, - "reason": "_is_hf_initialized regression vs Kosmos2.5 OCR indexing bug; unrelated." + "reason": "_is_hf_initialized regression and Kosmos2.5 OCR long-input indexing are unrelated." }, { "left": "issue:38175", "right": "issue:43994", "accept": true, - "reason": "Same underlying SigLIP2-base-patch16-224 inference regression: one report shows zero probabilities, the other nonsensical outputs from AutoModel/pipeline." + "reason": "Same model and same symptom class: incorrect/zeroed outputs from siglip2-base-patch16-224, likely the same underlying regression." }, { "left": "issue:43638", "right": "issue:44661", "accept": false, - "reason": "DeepSpeed ZeRO-3/Bert index error vs tokenizer mapping-name failure; unrelated." + "reason": "DeepSpeed zero3 with a non-pretrained Bert model is unrelated to tokenizer mapping support failures." } ] }, "evaluator_result": { - "accept": false, - "feedback": "The summary is mostly conservative, but one key claim is overstated: issue:43994 does not have downstream references in the packet (inbound_references is 0), so that part of the best-issue rationale is unsupported. Keep the 43994/38175 pairing as a likely duplicate cluster, but ground it only in the shared SigLIP2 regression and repro symptoms. The rest of the soft-edge verdicts look appropriately cautious." + "accept": true, + "feedback": "The summary is grounded in the packet and stays mostly conservative. The accepted soft-edge pairs are plausibly supported by the titles alone, and the rejected pairs are not overclaimed. The canonical custom-model init choice also looks well supported." }, "evaluator_used": true, - "retried": true, + "retried": false, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -1919,8 +1862,7 @@ "issue:42915", "issue:43257", "issue:43317", - "issue:43404", - "issue:43454", + "issue:43425", "issue:43596", "issue:43644", "issue:43645", @@ -1930,6 +1872,7 @@ "issue:43856", "issue:43873", "issue:43931", + "issue:43957", "issue:43994", "issue:44075", "issue:44155", @@ -1980,242 +1923,242 @@ "issue:39692|issue:43994", "issue:45488|issue:45812", "issue:43856|issue:43873", - "issue:43854|issue:43931", "issue:43257|issue:43854", + "issue:43854|issue:43931", "issue:43856|issue:44387", - "issue:45362|issue:45406", "issue:44960|issue:46050", + "issue:45362|issue:45406", "issue:42491|issue:45292", "issue:42491|issue:43931", - "issue:44910|issue:45588", "issue:44164|issue:44291", + "issue:44910|issue:45588", "issue:45276|issue:45335", - "issue:43404|issue:43454", - "issue:41720|issue:44155" + "issue:41720|issue:44155", + "issue:43425|issue:43957" ], - "prepared_review_unit_hash": "d27bfa787da9b42891026160a0bd4e3a2a46cc7621621cd9b672437f3bd59da2", + "prepared_review_unit_hash": "b8a6bf62ad043f7ef0acf1ff79816fba6d788c8e8d9db5de4b2b14fe060856a0", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 31, - "serialized_chars": 36446, - "estimated_input_tokens": 9112, - "estimated_eval_tokens": 18480 + "serialized_chars": 36396, + "estimated_input_tokens": 9099, + "estimated_eval_tokens": 18454 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a collection of unrelated bug reports across different models and subsystems, with a few clear duplicate subclusters around config-validation load failures, non-persistent-buffer initialization, and resize_token_embeddings regressions. The BERT-loading pair issue:44451 vs issue:44488 is not strong enough to merge on the available evidence.", - "confidence": 0.79, - "canonical_issue_reason": "issue:45276 is the best canonical for the resize_token_embeddings subcluster: it states the core bug most generally, and it aligns with the related Gemma/T5Gemma reports about embeddings not updating after resizing.", + "summary": "This cluster is mostly a loose set of Transformers v5-era regression reports around loading, quantization, offload, and model-specific configuration, with only a few clear duplicate pairs. The strongest duplicate signal is the non-persistent buffer corruption report, while most other pairs are only thematically similar and should stay separate.", + "confidence": 0.61, + "canonical_issue_reason": "issue:43644 is the cleanest anchor: it describes a concrete Transformers 5 regression about non-persistent buffers being filled with junk, and issue:44534 is a near-exact duplicate with the same symptom.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45488 is the strongest standalone report overall: it describes a broad, silent tokenizer regression that breaks a major model family and has clear user impact.", + "best_issue_reason": "issue:43644 is the most representative and actionable issue in the set, with the clearest duplicate relationship and a concrete library-wide v5 regression rather than a model-specific one.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44877", "right": "issue:45030", "accept": false, - "reason": "Related loading failures, but the packet does not establish the same root cause; this is too speculative to merge as duplicates." + "reason": "Both are config-loading breakages, but they affect different models and appear to stem from different validation paths." }, { "left": "issue:43596", "right": "issue:44661", "accept": false, - "reason": "Different bugs: a DeepSpeed/BertModel zero-sized tensor error versus add-new-model-like failing on TOKENIZER_MAPPING_NAMES." + "reason": "One is a DeepSpeed/BertModel init failure; the other is an add-new-model-like tokenizer mapping issue." }, { "left": "issue:44661", "right": "issue:44805", "accept": false, - "reason": "No shared code path is evident; tokenizer-mapping registration failure and tensor mask shape mismatch are different failures." + "reason": "Tokenizer mapping failure and mask-shape IndexError are unrelated code paths." }, { "left": "issue:43317", "right": "issue:44756", "accept": false, - "reason": "Different problems: dequantized-model offload loading failure versus a Strix Halo mmap/OOM workaround." + "reason": "Both mention loading/offload behavior, but one is dequantized gpu+cpu offload and the other is mmap/OOM on Strix Halo." }, { "left": "issue:38175", "right": "issue:39692", "accept": false, - "reason": "Same model family, but different concrete failures: zero probabilities at inference versus docs/example model-processor and quantization errors." + "reason": "Same model family, but one is zero probabilities and the other is a docs-example mix of model/processor mismatch and quantization errors." }, { "left": "issue:43644", "right": "issue:44534", "accept": true, - "reason": "Both describe the same Transformers v5 non-persistent-buffer initialization regression where buffers get filled with junk." + "reason": "They describe the same Transformers 5 bug: non-persistent buffers getting corrupted/junk values." }, { "left": "issue:42491", "right": "issue:43720", "accept": false, - "reason": "Different failure modes: qwen3_moe LoRA incompatibility across versions versus BitNet packed-weight unpacking during accelerate loading." + "reason": "LoRA/Qwen3 MoE version incompatibility and BitNet packed-weight loading are different failures." }, { "left": "issue:41720", "right": "issue:42915", "accept": false, - "reason": "Both involve Qwen3, but one is auto device mapping cuda asserts and the other is FineGrainedFP8Config loading; not the same bug." + "reason": "Both are Qwen3-related, but one is a device-map CUDA assert and the other is a FineGrainedFP8Config failure." }, { "left": "issue:44960", "right": "issue:44977", "accept": false, - "reason": "Different models and symptoms: GLM5 versus Qwen3.5 flash-attention generation failure." + "reason": "Different models and different failure modes; too broad to be the same bug." }, { "left": "issue:42915", "right": "issue:44756", "accept": false, - "reason": "Unrelated: FineGrainedFP8/Qwen3Moe issue versus mmap/OOM on Strix Halo." + "reason": "Qwen3Moe FP8 loading and Strix Halo mmap/OOM are unrelated." }, { "left": "issue:44451", "right": "issue:44488", "accept": false, - "reason": "Both are model-loading failures for BERT-like models, but the packet does not establish a shared root cause or the same failing code path." + "reason": "Both are load failures for different models, but there is no concrete evidence they share the same underlying bug." }, { "left": "issue:45313", "right": "issue:45357", "accept": false, - "reason": "Different code paths: DeepSpeed ZeRO-3 language_model loading versus incorrect visual-encoder keys in save_pretrained." + "reason": "DeepSpeed ZeRO-3 weight loading and incorrect visual encoder save keys are different issues." }, { "left": "issue:42491", "right": "issue:45276", "accept": false, - "reason": "Different bugs: qwen3_moe LoRA/version incompatibility versus resize_token_embeddings not propagating to embeddings." + "reason": "Qwen3 MoE LoRA incompatibility is not the same as resize_token_embeddings propagation in Gemma4." }, { "left": "issue:43708", "right": "issue:44075", "accept": false, - "reason": "Trainer resume/max_steps calculation bug is unrelated to SGD args not being used." + "reason": "Trainer checkpoint-resume math and SGD args not being used are unrelated." }, { "left": "issue:44488", "right": "issue:44991", "accept": false, - "reason": "Both are loading failures, but one is a model load regression and the other is tokenizer loading for a specific repo; no clear same root cause." + "reason": "Both are tokenizer/model load complaints, but for different models and without a shared concrete failure path." }, { "left": "issue:43644", "right": "issue:43645", "accept": false, - "reason": "Different regressions: junk-filled non-persistent buffers versus Jupyter custom-model initialization failure." + "reason": "Non-persistent buffer corruption and Jupyter custom-model initialization are distinct regressions." }, { "left": "issue:39692", "right": "issue:43994", "accept": false, - "reason": "Both mention SigLIP2, but one is docs/example breakage and the other is incorrect inference outputs; not the same concrete bug." + "reason": "Both involve SigLIP2, but one is a docs-example/mismatch report and the other is a nonsensical-results inference report; not clearly the same bug." }, { "left": "issue:45488", "right": "issue:45812", "accept": false, - "reason": "Different tokenizer regressions: LlamaTokenizer pre-tokenizer override versus Granite token-id mis-generation." + "reason": "Tokenizer behavior regressions, but they affect different tokenizer stacks and model families." }, { "left": "issue:43856", "right": "issue:43873", "accept": false, - "reason": "Different areas: Qwen3 MoE training memory efficiency versus quantization offloading behavior." + "reason": "Both mention memory/offloading around Qwen3 MoE, but one is training memory inefficiency and the other is quantization offload behavior." }, { - "left": "issue:43854", - "right": "issue:43931", + "left": "issue:43257", + "right": "issue:43854", "accept": false, - "reason": "Different model-loading failures with different models and error types." + "reason": "Qwen3 MoE weight conversion under accelerate+deepspeed is unrelated to GLM-4.7-Flash loading in tests." }, { - "left": "issue:43257", - "right": "issue:43854", + "left": "issue:43854", + "right": "issue:43931", "accept": false, - "reason": "Different bugs: accelerate+deepspeed MOE weight conversion versus GLM-4.7-Flash loading in tests." + "reason": "Different model-loading failures with different symptoms and code paths." }, { "left": "issue:43856", "right": "issue:44387", "accept": false, - "reason": "Training-memory inefficiency and int4 CUDA reserved-memory OOM are related only broadly, not the same bug." + "reason": "Both are memory-related, but one is training inefficiency and the other is int4 quantization reserved-memory/OOM regression." }, { - "left": "issue:45362", - "right": "issue:45406", + "left": "issue:44960", + "right": "issue:46050", "accept": false, - "reason": "Different model/processor crashes: Qwen3.5 chat crash versus Gemma4Processor missing _tokenizer." + "reason": "GLM5 issue and quantized save/load breakage are separate problems." }, { - "left": "issue:44960", - "right": "issue:46050", + "left": "issue:45362", + "right": "issue:45406", "accept": false, - "reason": "GLM5 issue versus quantized model broken after save/load are unrelated." + "reason": "Qwen3.5 chat crash and Gemma4Processor serve crash are unrelated." }, { "left": "issue:42491", "right": "issue:45292", "accept": false, - "reason": "Different bugs: qwen3_moe version incompatibility versus resize_token_embeddings not updating output embeddings." + "reason": "LoRA/hf4.x versus resize_token_embeddings propagation are different bugs." }, { "left": "issue:42491", "right": "issue:43931", "accept": false, - "reason": "Different failures: qwen3_moe branch incompatibility versus Qwen3-VL weight-shape mismatch." + "reason": "Qwen3 MoE version incompatibility is not the same as Qwen3-VL shape mismatch." }, { - "left": "issue:44910", - "right": "issue:45588", + "left": "issue:44164", + "right": "issue:44291", "accept": false, - "reason": "Both touch flash attention, but one is packed-sequence misinterpretation and the other is an s_aux=None AttributeError." + "reason": "save/from_pretrained extra_state handling and init_empty_weights argument validation are different loading issues." }, { - "left": "issue:44164", - "right": "issue:44291", + "left": "issue:44910", + "right": "issue:45588", "accept": false, - "reason": "Different serialization/loading regressions: extra_state handling versus init_empty_weights unexpected argument." + "reason": "Both touch flash_attention, but one is a packed-sequence position_ids crash and the other is an s_aux=None AttributeError." }, { "left": "issue:45276", "right": "issue:45335", "accept": true, - "reason": "Same resize_token_embeddings regression: both report target embeddings not being updated in related model variants." + "reason": "Both point to the same resize_token_embeddings propagation bug across model-specific embedding/output modules." }, { - "left": "issue:43404", - "right": "issue:43454", + "left": "issue:41720", + "right": "issue:44155", "accept": false, - "reason": "Both are lm_head tying problems, but they affect different model classes and are not clearly the same concrete fix." + "reason": "CUDA assert during Qwen3 auto device map is unrelated to AudioFlamingo3 batched inference token/embedding leakage." }, { - "left": "issue:41720", - "right": "issue:44155", + "left": "issue:43425", + "right": "issue:43957", "accept": false, - "reason": "Unrelated: cuda assert during auto device mapping versus AudioFlamingo3 batched inference token/embedding leakage." + "reason": "Torch 2.10 incompatibility and meta-device loading regressions are different problems." } ] }, "evaluator_result": { - "accept": false, - "feedback": "One soft-edge verdict is too strong for the evidence in the packet: issue:44451 vs issue:44488 are both model-loading failures, but the packet does not establish that they share the same root cause. The acceptance claim (\u201csame load regression affecting different BERT-like models\u201d) goes beyond the titles/metadata, so this is not conservative enough. The rest of the summary is mostly grounded, but this overstates one duplicate pair." + "accept": true, + "feedback": "The analysis is grounded in the packet and stays conservative on the soft edges. The two accepted duplicate pairs are well supported by the issue titles/body themes, and the rejected pairs do not overclaim similarity. The summary is appropriately cautious overall." }, "evaluator_used": true, - "retried": true, + "retried": false, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -2239,12 +2182,13 @@ "issue:42947", "issue:43335", "issue:43381", + "issue:43404", "issue:43425", + "issue:43454", "issue:43611", "issue:43645", "issue:43646", "issue:43708", - "issue:43716", "issue:43828", "issue:43854", "issue:43873", @@ -2282,7 +2226,7 @@ "issue:46315" ], "soft_pairs": [ - "issue:43425|issue:43957", + "issue:43404|issue:43454", "issue:39692|issue:44960", "issue:44402|issue:44488", "issue:42947|issue:45704", @@ -2299,8 +2243,8 @@ "issue:42947|issue:43381", "issue:45276|issue:45292", "issue:44164|issue:44756", - "issue:43854|issue:45216", "issue:43381|issue:44928", + "issue:43854|issue:45216", "issue:44479|issue:44877", "issue:45684|issue:45698", "issue:45588|issue:45663", @@ -2315,223 +2259,222 @@ "issue:39404|issue:45701", "issue:44683|issue:45823", "issue:43611|issue:43645", - "issue:43425|issue:45823", - "issue:43425|issue:43716" + "issue:43425|issue:45823" ], - "prepared_review_unit_hash": "27560eb294d5eed38b7f9336c7602e9072714cf7dce32354d4191b49c52ccb2d", + "prepared_review_unit_hash": "2d2fd9a038855907456d2c3c9611352f94442864a49190ca8f782db10c7aba68", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 35, - "serialized_chars": 37650, - "estimated_input_tokens": 9413, - "estimated_eval_tokens": 19082 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 34, + "serialized_chars": 37967, + "estimated_input_tokens": 9492, + "estimated_eval_tokens": 19240 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a set of unrelated Transformers issues grouped by loose textual similarity. The only clear duplicate family is the resize_token_embeddings regression affecting output embeddings / decoder embeddings in Gemma4/T5Gemma variants.", - "confidence": 0.86, - "canonical_issue_reason": "Issue 45292 is the best canonical issue because it states the core regression in the most general form: resize_token_embeddings does not update output embeddings. The neighboring reports are narrower model-specific manifestations of the same behavior.", + "summary": "This cluster is mostly a grab-bag of unrelated Transformers bug reports. The only clear duplicate family here is the resize_token_embeddings regression around downstream embeddings; most other pairs are only loosely related by subsystem and should stay separate.", + "confidence": 0.64, + "canonical_issue_reason": "issue:45292 is the most general, model-agnostic report of the resize_token_embeddings/output_embeddings regression, so it works best as the cluster anchor.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45292 is the strongest global issue target because it is broad, concise, and describes the shared underlying bug rather than a single model variant.", + "best_issue_reason": "issue:45292 is the best duplicate target because its title captures the shared API-level bug without narrowing to one model family.", "best_pr_reason": null, "soft_edge_verdicts": [ { - "left": "issue:43425", - "right": "issue:43957", + "left": "issue:43404", + "right": "issue:43454", "accept": false, - "reason": "Both are version/runtime breakages, but one is Torch compatibility and the other is meta-device model loading; they do not describe the same bug." + "reason": "Both mention missing weight tying/generation quality, but they are different model-specific failures and not the same concrete bug." }, { "left": "issue:39692", "right": "issue:44960", "accept": false, - "reason": "Completely different topics: SigLIP2 documentation errors vs. a GLM5 model-loading issue." + "reason": "Unrelated topics: SigLIP2 documentation example errors vs a generic GLM5 issue." }, { "left": "issue:44402", "right": "issue:44488", "accept": false, - "reason": "Both are model-loading complaints, but the model names and failure modes differ; not enough evidence of the same underlying bug." + "reason": "Both are model-loading complaints, but the underlying failures appear different (vocab mismatch vs a specific model load regression)." }, { "left": "issue:42947", "right": "issue:45704", "accept": false, - "reason": "Gradient checkpointing with LoRA and an apex RMSNorm memory leak are unrelated bugs." + "reason": "Different bugs: LoRA gradient checkpointing behavior vs a T5 RMSNorm memory leak." }, { "left": "issue:36010", "right": "issue:39404", "accept": false, - "reason": "ImportError for GenerationMixin and Whisper pipeline return_language are different failures with no shared code-path indicated." + "reason": "ImportError for GenerationMixin is unrelated to Whisper pipeline return_language behavior." }, { "left": "issue:43873", "right": "issue:44387", "accept": false, - "reason": "Both mention quantization, but one is offloading behavior and the other is increased reserved memory/OOM under int4; too broad to treat as the same bug." + "reason": "Both involve quantization, but one is offloading behavior and the other is CUDA memory/OOM regression." }, { "left": "issue:45663", "right": "issue:45759", "accept": false, - "reason": "Gemma4 FSDP2 shared_kv_states and Qwen multimodal config unwrapping are different model-integration problems." + "reason": "Different code paths: Gemma4 FSDP attention state vs Qwen multimodal config unwrapping." }, { "left": "issue:44291", "right": "issue:44756", "accept": false, - "reason": "Unexpected _is_hf_initialized during init_empty_weights is unrelated to disabling mmap to avoid OOM." + "reason": "init_empty_weights/_is_hf_initialized is unrelated to mmap-related OOM on Strix Halo." }, { "left": "issue:45663", "right": "issue:45859", "accept": false, - "reason": "Different models and failure modes: Gemma4 FSDP2 KeyError vs Qwen3_5Moe CPU offload incompatibility." + "reason": "Different model families and failures: Gemma4 FSDP attention vs Qwen3_5Moe CPU offload compatibility." }, { "left": "issue:44155", "right": "issue:45823", "accept": false, - "reason": "AudioFlamingo3 batch leakage and Gemma4 device mismatch are unrelated issues." + "reason": "AudioFlamingo3 batch leakage and Gemma4 device mismatch are unrelated bugs." }, { "left": "issue:44292", "right": "issue:44912", "accept": false, - "reason": "Both involve quantization, but NVFP4 loading failure and MXFP4 fallback to bf16 are not the same bug." + "reason": "Both are quantization load issues, but the concrete models and failure modes differ." }, { "left": "issue:44387", "right": "issue:44756", "accept": false, - "reason": "Int4 quantization OOM and Strix Halo mmap OOM are separate memory issues." + "reason": "OOM under int4 quantization is not the same as disabling mmap to avoid OOM on Strix Halo." }, { "left": "issue:43335", "right": "issue:45663", "accept": false, - "reason": "SwitchTransformers sparse-layer creation and Gemma4 FSDP2 attention-state rebuilding are unrelated." + "reason": "Sparse-layer construction in SwitchTransformers is unrelated to Gemma4 FSDP attention state rebuilding." }, { "left": "issue:43957", "right": "issue:44756", "accept": false, - "reason": "Meta-device loading breakage is not the same as Strix Halo mmap/OOM behavior." + "reason": "meta-device model loading breakage is a different issue from Strix Halo mmap OOM." }, { "left": "issue:42947", "right": "issue:43381", "accept": false, - "reason": "Both mention gradient checkpointing, but one is ineffective with PEFT LoRA and the other is incompatibility in eval mode; different bugs." + "reason": "Both mention gradient checkpointing, but one is ineffective under LoRA and the other is an eval-mode restriction." }, { "left": "issue:45276", "right": "issue:45292", "accept": true, - "reason": "Same underlying regression: resize_token_embeddings does not propagate updated sizes to model output embeddings; 45276 is a model-specific manifestation." + "reason": "Same underlying regression: resize_token_embeddings does not propagate to downstream output embeddings; 45276 is the Gemma4-specific instance." }, { "left": "issue:44164", "right": "issue:44756", "accept": false, - "reason": "save/from_pretrained extra_state handling and Strix Halo mmap OOM are unrelated." + "reason": "save/from_pretrained extra_state handling is unrelated to mmap/OOM behavior." }, { - "left": "issue:43854", - "right": "issue:45216", + "left": "issue:43381", + "right": "issue:44928", "accept": false, - "reason": "Loading GLM-4.7-Flash in tests and Qwen3.5 save_pretrained regression are different issues." + "reason": "Gradient checkpointing in eval mode is unrelated to RLHF NaN explosions from 3D position_ids and SDPA fallback." }, { - "left": "issue:43381", - "right": "issue:44928", + "left": "issue:43854", + "right": "issue:45216", "accept": false, - "reason": "Gradient checkpointing in eval mode and RLHF NaN from 3D position_ids / SDPA fallback are unrelated." + "reason": "Different save/load regressions affecting different models and checkpoints." }, { "left": "issue:44479", "right": "issue:44877", "accept": false, - "reason": "Video-input regression in Qwen VL models and strict config blocking granite_speech are different problems." + "reason": "Video input regression in Qwen variants is unrelated to granite_speech config loading." }, { "left": "issue:45684", "right": "issue:45698", "accept": false, - "reason": "Read-only permissions propagation during save_pretrained and loading the wrong custom module after save_pretrained are distinct save/load issues." + "reason": "Both involve save_pretrained/custom modules, but one is permissions propagation and the other is loading the wrong module." }, { "left": "issue:45588", "right": "issue:45663", "accept": false, - "reason": "Flash-attention sinkless-model crash and Gemma4 FSDP2 shared_kv_states KeyError are unrelated." + "reason": "Flash-attention sink-less-model crash is unrelated to Gemma4 FSDP shared_kv_states." }, { "left": "issue:43646", "right": "issue:43708", "accept": false, - "reason": "Custom model initialization in notebooks and Trainer max_steps calculation on checkpoint resume are unrelated." + "reason": "Custom model initialization failure and checkpoint-resume max_steps miscalculation are different bugs." }, { "left": "issue:43645", "right": "issue:44164", "accept": false, - "reason": "Notebook custom-model initialization and extra_state save/load handling are different code paths." + "reason": "Notebook custom-model initialization and extra_state save/load handling are unrelated." }, { "left": "issue:45292", "right": "issue:45335", "accept": true, - "reason": "Same resize_token_embeddings regression, with 45335 showing the decoder.embed_tokens variant of the same missing propagation." + "reason": "Same API regression class: resize_token_embeddings fails to update the relevant downstream embeddings, with different model-specific manifestations." }, { "left": "issue:45812", "right": "issue:46315", "accept": false, - "reason": "Wrong token IDs from AutoTokenizer and slower convert_tokens_to_ids are different tokenizer regressions." + "reason": "Token-ID correctness regression and tokenization performance regression are different bugs." }, { "left": "issue:43828", "right": "issue:43957", "accept": false, - "reason": "Autocast dtype mismatch for Phi-tiny-MoE and meta-device loading breakage are unrelated." + "reason": "Autocast dtype mismatch in Phi-tiny-MoE is unrelated to meta-device model loading breakage." }, { "left": "issue:43425", "right": "issue:43828", "accept": false, - "reason": "Torch version incompatibility and autocast dtype mismatch are not the same bug." + "reason": "Torch 2.10 compatibility and autocast dtype mismatch are different failure classes." }, { "left": "issue:44402", "right": "issue:44451", "accept": false, - "reason": "Both are model-loading failures, but they concern different models and no common failure mode is shown." + "reason": "Tokenizer vocab-size mismatch and a specific model loading failure are not the same bug." }, { "left": "issue:44589", "right": "issue:44756", "accept": false, - "reason": "Float8 storage lookup error and mmap OOM on Strix Halo are unrelated." + "reason": "Missing Float8 storage type and mmap-related OOM are unrelated." }, { "left": "issue:39404", "right": "issue:45701", "accept": false, - "reason": "Whisper pipeline return_language regression and tokenizer-version behavior change are different issues." + "reason": "Whisper pipeline return_language regression is unrelated to tokenizer-version changes." }, { "left": "issue:44683", @@ -2543,25 +2486,19 @@ "left": "issue:43611", "right": "issue:43645", "accept": false, - "reason": "Base_model_prefix loading failure and notebook custom-model initialization are separate custom-model breakages." + "reason": "Both concern Transformers 5 custom-model behavior, but they describe different regressions and code paths." }, { "left": "issue:43425", "right": "issue:45823", "accept": false, - "reason": "Torch 2.10 incompatibility and Gemma4 PLE device mismatch do not share the same underlying bug." - }, - { - "left": "issue:43425", - "right": "issue:43716", - "accept": false, - "reason": "Torch compatibility and Mistral-3 image-preprocessor dtype mismatch are different failures." + "reason": "Torch version incompatibility is unrelated to Gemma4 PLE device placement." } ] }, "evaluator_result": { "accept": true, - "feedback": "The cluster summary is well grounded in the packet, and the soft-edge decisions are conservative. The only clearly supported duplicate family is the resize_token_embeddings/output_embeddings regression across 45276, 45292, and 45335; the other rejected pairs do not have enough evidence of a shared underlying bug." + "feedback": "The summary is well grounded in the packet: the cluster is mostly heterogeneous, and the only clearly supported duplicate family is the resize_token_embeddings/output_embeddings regression centered on issue:45292 with the Gemma4 and t5gemma variants. The rejected soft-edge pairs are phrased conservatively and do not overclaim sameness; the accepted pairs within the resize_token_embeddings family look like legitimate single-bug variants." }, "evaluator_used": true, "retried": false, @@ -2632,14 +2569,15 @@ "issue:46402" ], "soft_pairs": [ + "issue:43425|issue:43716", "issue:44530|issue:45084", "issue:43906|issue:44530", "issue:39404|issue:44912", "issue:44265|issue:44898", "issue:44479|issue:44805", "issue:43906|issue:46326", - "issue:46050|issue:46326", "issue:44589|issue:45084", + "issue:46050|issue:46326", "issue:43901|issue:44509", "issue:43425|issue:44918", "issue:43299|issue:43931", @@ -2653,238 +2591,244 @@ "issue:44164|issue:44479", "issue:39404|issue:44671", "issue:44737|issue:45859", - "issue:44488|issue:45081", "issue:43582|issue:43632", + "issue:44488|issue:45081", "issue:44661|issue:44683", "issue:44488|issue:44987", - "issue:43611|issue:43644", "issue:32090|issue:39404", - "issue:46285|issue:46402", + "issue:43611|issue:43644", "issue:43716|issue:45084", + "issue:46285|issue:46402", "issue:40990|issue:43421", "issue:36032|issue:42915", "issue:43643|issue:44530" ], - "prepared_review_unit_hash": "23fca553bd9a56c734302a0a661de0566b842ed484b0b614b4ad4a5635c04fd1", + "prepared_review_unit_hash": "358b24bdc4efbbfae0ea3c8f533959461009987a0cf97b7b3b5eb7edf929b93e", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 32, - "serialized_chars": 37406, - "estimated_input_tokens": 9352, - "estimated_eval_tokens": 18960 + "soft_pair_count": 33, + "serialized_chars": 37710, + "estimated_input_tokens": 9428, + "estimated_eval_tokens": 19112 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is a grab-bag of mostly unrelated Transformers issues, with several broad v5 regressions and many model-specific loading or runtime bugs. None of the soft pairs look like the same underlying defect closely enough to merge.", - "confidence": 0.93, - "canonical_issue_reason": "#43611 is the best canonical representative because it describes a broad, user-facing Transformers v5 loading regression that can subsume a family of similar breakages; the other issues are narrower, model-specific, or unrelated.", + "summary": "These items span many unrelated Transformers regressions; none of the soft pairs look like the same underlying bug/change, so all soft edges are rejected. The most general issue in the set is the v5 model-loading regression around `base_model_prefix`, but there is no strong single duplicate cluster overall.", + "confidence": 0.81, + "canonical_issue_reason": "Issue 43611 is the broadest, most user-facing loading regression in the set and is the closest thing to a canonical issue, but the overall cluster is heterogeneous.", "canonical_pr_reason": null, - "best_issue_reason": "#43611 is the strongest issue to anchor the cluster: it is broad, clearly regression-oriented, and about a core loading path rather than a niche edge case.", + "best_issue_reason": "43611 best represents the recurring v5 model-loading breakages because it is broad, concrete, and likely to subsume adjacent loading regressions better than the more model-specific reports.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:43425", + "right": "issue:43716", + "accept": false, + "reason": "Torch version incompatibility vs. Qwen3-Mistral image preprocessor/model dtype mismatch; different failure modes and code paths." + }, { "left": "issue:44530", "right": "issue:45084", "accept": false, - "reason": "Different failures: a Qwen3.5 PagedAttentionCache crash versus a tokenizer/template compilation TypeError." + "reason": "PagedAttentionCache runtime crash on Qwen3.5 vs tokenizer template compilation error; unrelated subsystems." }, { "left": "issue:43906", "right": "issue:44530", "accept": false, - "reason": "Unrelated code paths and symptoms; one is a reproduction of an older issue, the other is a cache/group-type crash on Qwen3.5." + "reason": "Isolated reproduction of another issue vs Qwen3.5 cache crash; no shared bug surface." }, { "left": "issue:39404", "right": "issue:44912", "accept": false, - "reason": "Whisper pipeline language-return regression and GPT-OSS MXFP4 quantization fallback are unrelated bugs." + "reason": "Whisper pipeline `return_language` regression vs MXFP4 quantization fallback; different model/features." }, { "left": "issue:44265", "right": "issue:44898", "accept": false, - "reason": "torch.export with torch_compilable_check is a different export-path bug than Perceiver image-size handling." + "reason": "`torch.export` failure around `torch_compilable_check` vs Perceiver interpolation/image-size issue; distinct problems." }, { "left": "issue:44479", "right": "issue:44805", "accept": false, - "reason": "Different model families and failures: video-input regression versus a mask/index shape mismatch." + "reason": "Video-input regression in VL models vs mask-shape index error; not the same underlying defect." }, { "left": "issue:43906", "right": "issue:46326", "accept": false, - "reason": "One is a reproduction of an older issue; the other is a mistral_common tokenizer-specific continue_final_message bug." + "reason": "Reproduction of issue 38071 vs mistral_common tokenizer `continue_final_message` bug; different tokenizer behavior." }, { - "left": "issue:46050", - "right": "issue:46326", + "left": "issue:44589", + "right": "issue:45084", "accept": false, - "reason": "Quantized save/load corruption is not the same as the mistral_common final-message handling issue." + "reason": "Float8 storage lookup error vs template compilation error; unrelated runtime failures." }, { - "left": "issue:44589", - "right": "issue:45084", + "left": "issue:46050", + "right": "issue:46326", "accept": false, - "reason": "Float8 storage lookup failure and template-node compilation errors are different defects." + "reason": "Quantization save/load corruption vs tokenizer final-message handling; different code paths." }, { "left": "issue:43901", "right": "issue:44509", "accept": false, - "reason": "Both are docs-oriented, but they concern different pipeline deprecations and different documentation fixes." + "reason": "Docs mismatch for `return_all_scores` vs docs mentioning removed pipeline tasks; documentation-only but not the same issue." }, { "left": "issue:43425", "right": "issue:44918", "accept": false, - "reason": "Torch version incompatibility and Qwen3.5/TRL embedding unpacking are separate issues." + "reason": "Torch incompatibility vs TRL SFT unpacking embeddings; different components and failure signatures." }, { "left": "issue:43299", "right": "issue:43931", "accept": false, - "reason": "Both involve Qwen3-VL, but one is a v5 loading regression for MoE models and the other is a specific weight-shape mismatch." + "reason": "Qwen3VL MoE loading regression vs Qwen3-VL weight shape mismatch; related model family but different concrete bug." }, { "left": "issue:43994", "right": "issue:44079", "accept": false, - "reason": "SigLIP2 bad inference results and ModelOutput key assignment with None are unrelated." + "reason": "SigLIP2 nonsensical outputs vs `ModelOutput` key assignment bug; one is inference quality, the other data-structure behavior." }, { "left": "issue:36032", "right": "issue:43452", "accept": false, - "reason": "A T5 tokenizer method conflict is not the same bug as gguf_file breaking AutoTokenizer/AutoModel loading." + "reason": "Tokenizer special-token method conflict vs gguf file loading breakage; different tokenizer/load paths." }, { "left": "issue:44164", "right": "issue:45759", "accept": false, - "reason": "save/from_pretrained extra_state handling and composite Qwen config unwrapping are different loading/saving paths." + "reason": "`extra_state` save/load handling vs composite Qwen config `text_config` unwrapping; both loading-related but not the same defect." }, { "left": "issue:44671", "right": "issue:44991", "accept": false, - "reason": "CamemBERT MLM prediction regression and EMBEDDIA tokenizer loading failure are different model-specific bugs." + "reason": "CamemBERT masked-LM prediction regression vs tokenizer load failure for a specific model; unrelated model-specific issues." }, { "left": "issue:43596", "right": "issue:44683", "accept": false, - "reason": "DeepSpeed ZeRO-3/BertModel initialization and compiled flex_attention on newer torch versions are unrelated." + "reason": "DeepSpeed ZeRO3/BertModel indexing error vs compiled flex_attention on newer torch; different stack and root cause." }, { "left": "issue:44991", "right": "issue:45081", "accept": false, - "reason": "Different tokenizer problems: EMBEDDIA load failure versus Mistral regex patch crashing on backend_tokenizer." + "reason": "Tokenizer load failure for EMBEDDIA/est-roberta vs Mistral regex patch crash; same broad area, different bugs." }, { "left": "issue:43856", "right": "issue:44756", "accept": false, - "reason": "MoE training memory inefficiency and Strix Halo mmap OOM are different memory issues with different causes." + "reason": "Qwen3 MoE training memory inefficiency vs mmap OOM on Strix Halo; not the same underlying issue." }, { "left": "issue:44164", "right": "issue:44479", "accept": false, - "reason": "Extra_state save/load handling has nothing to do with the Qwen video-input regression." + "reason": "Save/from_pretrained extra_state handling vs VL video-input regression; no shared code path." }, { "left": "issue:39404", "right": "issue:44671", "accept": false, - "reason": "Whisper pipeline behavior and CamemBERT masked-LM predictions are separate model regressions." + "reason": "Whisper pipeline language return regression vs CamemBERT masked LM prediction issue; unrelated models and symptoms." }, { "left": "issue:44737", "right": "issue:45859", "accept": false, - "reason": "Both involve device/offload behavior, but the XLNet CPU arange bug and Qwen3_5MoE rotary embedding incompatibility are distinct." + "reason": "XLNet CPU arange/device bug vs Qwen3_5Moe CPU offload incompatibility; both device-related but different code paths." }, { - "left": "issue:44488", - "right": "issue:45081", + "left": "issue:43582", + "right": "issue:43632", "accept": false, - "reason": "Loading cjvt/sleng-bert and Mistral regex patching are different tokenizer-loading problems." + "reason": "Apple Silicon allocator warmup TypeError vs `_is_hf_initialized` flag regression; different internal failures." }, { - "left": "issue:43582", - "right": "issue:43632", + "left": "issue:44488", + "right": "issue:45081", "accept": false, - "reason": "Apple Silicon warmup TypeError and the _is_hf_initialized v5 regression are unrelated." + "reason": "Loading `cjvt/sleng-bert` failure vs Mistral regex patch crash; distinct tokenizer-loading problems." }, { "left": "issue:44661", "right": "issue:44683", "accept": false, - "reason": "Tokenizer mapping-name failure and compiled flex_attention are on different subsystems entirely." + "reason": "`add-new-model-like` generator logic vs compiled flex_attention; unrelated feature/tooling areas." }, { "left": "issue:44488", "right": "issue:44987", "accept": false, - "reason": "They are both loading failures, but for different models and different failure modes." - }, - { - "left": "issue:43611", - "right": "issue:43644", - "accept": false, - "reason": "Both are v5 regressions, but one is about base_model_prefix loading and the other about non-persistent buffers being filled with junk." + "reason": "Model load failure for `cjvt/sleng-bert` vs `physical-intelligence/fast` load regression; different affected models and likely causes." }, { "left": "issue:32090", "right": "issue:39404", "accept": false, - "reason": "Trainer broadcast NoneType and Whisper return_language pipeline behavior are unrelated." + "reason": "Trainer `_gpu_broadcast_one` NoneType TypeError vs Whisper `return_language` pipeline regression; unrelated behavior." }, { - "left": "issue:46285", - "right": "issue:46402", + "left": "issue:43611", + "right": "issue:43644", "accept": false, - "reason": "CLIP text-model structure inconsistency and Sam3 text-encoder weights not loading are different model-specific issues." + "reason": "`base_model_prefix` loading regression vs non-persistent buffer junk initialization; both v5-related but different bugs." }, { "left": "issue:43716", "right": "issue:45084", "accept": false, - "reason": "Mistral-3 dtype mismatch during preprocessing/model use is unrelated to the template-node compilation error." + "reason": "Image preprocessor/model dtype mismatch vs tokenizer template compilation error; no shared underlying change." + }, + { + "left": "issue:46285", + "right": "issue:46402", + "accept": false, + "reason": "CLIP text-model structure inconsistency vs Sam3 text encoder weights not loading; separate model architecture/loading issues." }, { "left": "issue:40990", "right": "issue:43421", "accept": false, - "reason": "Perplexity quality on gpt-oss and tokenizer post-processor updates after runtime special-token changes are different bugs." + "reason": "High perplexity on gpt-oss-20b vs runtime special-token/post-processor update behavior; unrelated problems." }, { "left": "issue:36032", "right": "issue:42915", "accept": false, - "reason": "T5 tokenizer method-name conflict and Qwen3Moe FineGrainedFP8Config failure are unrelated." + "reason": "Tokenizer method name conflict vs Qwen3Moe FineGrainedFP8Config failure; different tokenizer/model-loading paths." }, { "left": "issue:43643", "right": "issue:44530", "accept": false, - "reason": "AutoConfig trust_remote_code field loss and Qwen3.5 PagedAttentionCache crashes are different code paths and symptoms." + "reason": "Missing fields from `trust_remote_code=True` config objects vs Qwen3.5 cache crash; different APIs and failure modes." } ] }, @@ -2979,8 +2923,8 @@ "issue:43262|issue:43377", "issue:43645|issue:44534", "issue:43646|issue:44534", - "issue:43404|issue:44466", "issue:44403|issue:45081", + "issue:43404|issue:44466", "issue:42915|issue:45084", "issue:43883|issue:44991", "issue:43089|issue:45987", @@ -2994,244 +2938,244 @@ "issue:42915|issue:44291", "issue:44220|issue:44683" ], - "prepared_review_unit_hash": "e57898660f82826c15f72c66d687e11995b0434f6c55b5fd3b38cbc45bbe74f4", + "prepared_review_unit_hash": "dd254cdcf3bf9ad46bb27d4dda0415656d71964f39f0bbe813097e96d1f6c69d", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 35, - "serialized_chars": 38323, - "estimated_input_tokens": 9581, - "estimated_eval_tokens": 19418 + "serialized_chars": 38320, + "estimated_input_tokens": 9580, + "estimated_eval_tokens": 19416 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is loosely centered on Transformers v5/v5.x regressions around model/tokenizer loading, init, and generation, but the listed issues are largely distinct bugs affecting different code paths and models. I found no soft-edge pair that clearly looks like the same underlying change/bug.", - "confidence": 0.34, - "canonical_issue_reason": "issue:43646 is the broadest representative: it describes a generic Transformers 5.0 custom-model initialization regression, which best matches the cluster\u2019s overall v5 load/init theme.", + "summary": "This cluster is a loose set of mostly unrelated Transformers regressions and model-loading issues. The soft pairs generally share only broad subsystem similarity, not the same concrete bug or fixable code path, so they should be rejected as duplicates.", + "confidence": 0.68, + "canonical_issue_reason": "issue:43646 is the broadest representative in the set: it describes a general Transformers 5.0.0 custom model initialization regression, which is closer to the recurring v5 initialization/load failures than the more specific or unrelated issues nearby.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43646 is the most general and actionable issue in the set, with a clear regression framing and broader scope than the model-specific reports.", + "best_issue_reason": "issue:43646 is the best issue candidate because it is the most general and reusable description of the recurring v5 initialization breakage, while the other candidates are narrower symptoms (Jupyter-only, non-persistent buffers, tokenizer/model-specific loading failures).", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:34567", "right": "issue:40444", "accept": false, - "reason": "Trainer token accounting vs multimodal iterable-dataset failure; unrelated bugs." + "reason": "Unrelated bugs: token-count tracking in TrainerState vs multimodal IterableDataset loading." }, { "left": "issue:40444", "right": "issue:46326", "accept": false, - "reason": "Different areas and failure modes: multimodal dataset handling vs mistral_common tokenizer messaging." + "reason": "Both mention model-processing failures, but one is Qwen2.5-VL multi-image dataset handling and the other is mistral_common continue_final_message behavior." }, { "left": "issue:44451", "right": "issue:44987", "accept": false, - "reason": "Both are model-load failures, but for different models and likely different tokenizer/model code paths." + "reason": "Different model-loading failures for different checkpoints; no evidence of the same underlying bug." }, { "left": "issue:43541", "right": "issue:45084", "accept": false, - "reason": "Torch dynamo grouped_mm tracing error is unrelated to template-node compilation failure." + "reason": "One is a grouped_mm tracing failure in Mixtral; the other is a template compilation error. Different code paths." }, { "left": "issue:44811", "right": "issue:45584", "accept": false, - "reason": "Both involve Whisper, but one is batch_decode special-token handling and the other is empty-transcription generation after align_special_tokens." + "reason": "Both involve Whisper generation/token handling, but one is batch_decode skip_special_tokens and the other is empty transcription after align_special_tokens." }, { "left": "issue:42175", "right": "issue:46178", "accept": false, - "reason": "Backend packaging omission is not the same as an unsupported backend-mapping error for tensorflow_text." + "reason": "Backend packaging vs backend registry validation are related at a high level, but not the same defect." }, { "left": "issue:42915", "right": "issue:44530", "accept": false, - "reason": "Different models and subsystems: FP8 config failure vs paged attention cache linear_attention crash." + "reason": "Different model families and failures: FineGrainedFP8Config on Qwen3Moe vs PagedAttentionCache linear_attention crash on Qwen3.5." }, { "left": "issue:43644", "right": "issue:43646", "accept": false, - "reason": "Related to v5 initialization regressions, but one is junk non-persistent buffers and the other is custom model init failure." + "reason": "Both are Transformers 5 regressions, but one is non-persistent buffer initialization and the other is custom model initialization." }, { "left": "issue:44206", "right": "issue:44220", "accept": false, - "reason": "Both are audio-feature related, but they point to different failures and code paths." + "reason": "Similar audio preprocessing area, but the failures differ: unsupported center arg vs _torch_extract_fbank_features behavior." }, { "left": "issue:43452", "right": "issue:43482", "accept": false, - "reason": "Both mention GGUF loading, but one is a generic gguf_file/from_pretrained breakage and the other is a specific Qwen2.5-GGUF v5 failure." + "reason": "Both are gguf-related loading problems, but one targets AutoTokenizer/AutoModelForCausalLM with gguf_file and the other is Qwen2.5-GGUF under v5." }, { "left": "issue:43883", "right": "issue:44488", "accept": false, - "reason": "Molmo tied-weights attribute error vs a separate model-loading failure; no shared underlying bug." + "reason": "Same broad loader regression theme, but one is missing all_tied_weights_keys and the other is a specific model load failure." }, { "left": "issue:36032", "right": "issue:43645", "accept": false, - "reason": "Tokenizer method-name conflict and notebook custom-model init regression are unrelated." + "reason": "Tokenizer class/method conflict vs notebook custom model initialization; unrelated." }, { "left": "issue:43317", "right": "issue:43482", "accept": false, - "reason": "GPU/CPU offload dequantization failure is not the same as a Qwen2.5 GGUF loading regression." + "reason": "Device-map offload/dequantized model loading vs GGUF model loading; different failure surfaces." }, { "left": "issue:44464", "right": "issue:45987", "accept": false, - "reason": "Chunked-generation inconsistency and StaticCache seq-length type mismatch are different bugs." + "reason": "Chunked generation inconsistencies vs StaticCache.get_seq_length return type; not the same bug." }, { "left": "issue:45084", "right": "issue:45507", "accept": false, - "reason": "Template compilation TypeError vs GraniteMoEHybrid invalid-method call; different concrete failures." + "reason": "Template compilation error vs GraniteMoEHybrid invalid method call; different underlying issues." }, { "left": "issue:44987", "right": "issue:44991", "accept": false, - "reason": "Different models and symptoms: one model fails to load, the other tokenizer loading fails for a different repo." + "reason": "Both are v5 loading regressions, but they affect different models and likely different loader/tokenizer paths." }, { "left": "issue:44849", "right": "issue:45325", "accept": false, - "reason": "Qwen3.5 hidden-states bug vs Qwen2.5-VL RoPE position-id scaling bug; unrelated." + "reason": "Different Qwen3.5 bugs: hidden-state output vs rope index scaling for Qwen2.5-VL." }, { "left": "issue:43746", "right": "issue:46326", "accept": false, - "reason": "PEFT adapter loading vs continue_final_message behavior in mistral_common tokenizer; not the same bug." + "reason": "PEFT/local-checkpoint loading for GraniteSpeech is unrelated to mistral_common continue_final_message handling." }, { "left": "issue:43262", "right": "issue:43377", "accept": false, - "reason": "Audio chat-template sampling-rate default and MIMI padding-mask mismatch are different audio bugs." + "reason": "Audio sampling-rate defaulting vs MIMI encoder padding-mask correctness; different components and symptoms." }, { "left": "issue:43645", "right": "issue:44534", "accept": false, - "reason": "Both are v5 regressions, but custom model init in notebooks is distinct from non-persistent buffer junk initialization." + "reason": "Notebook/custom model initialization bug vs non-persistent buffer junk; related only by being v5 regressions." }, { "left": "issue:43646", "right": "issue:44534", "accept": false, - "reason": "Same reasoning: different init-related regressions with different observed failures." + "reason": "General custom initialization regression vs buffer initialization corruption; not the same code path." }, { - "left": "issue:43404", - "right": "issue:44466", + "left": "issue:44403", + "right": "issue:45081", "accept": false, - "reason": "Tied lm_head weights missing vs serialization differing by device are related broadly, but not the same concrete bug." + "reason": "Unnecessary loading noise vs Mistral regex patch backend_tokenizer AttributeError; unrelated." }, { - "left": "issue:44403", - "right": "issue:45081", + "left": "issue:43404", + "right": "issue:44466", "accept": false, - "reason": "Logging/noise issue vs Mistral tokenizer regex patch crash; different scopes and failures." + "reason": "lm_head tying bug in Mistral3 vs device-dependent lm_head serialization inconsistency; related around weights, but not the same defect." }, { "left": "issue:42915", - "right": "issue:44291", + "right": "issue:45084", "accept": false, - "reason": "FP8-config model failure and init_empty_weights argument error are separate initialization/configuration bugs." + "reason": "Qwen3Moe FP8 config failure and template compilation failure are unrelated." }, { "left": "issue:43883", "right": "issue:44991", "accept": false, - "reason": "Molmo tied-weights attribute error is unrelated to tokenizer loading for EMBEDDIA/est-roberta." + "reason": "Different loading failures for different models and causes; no shared concrete bug." }, { "left": "issue:43089", "right": "issue:45987", "accept": false, - "reason": "Generation performance overhead is unrelated to StaticCache seq-length contract issues." + "reason": "Generation performance overhead vs cache length return type contract; different problem classes." }, { "left": "issue:43994", "right": "issue:44610", "accept": false, - "reason": "Nonsensical outputs from one model and processor/model input-size mismatch in another are distinct problems." + "reason": "SigLIP2 incorrect outputs via AutoModel/pipeline vs OmDet-Turbo processor/model resolution mismatch; not the same issue." }, { "left": "issue:43322", "right": "issue:43540", "accept": false, - "reason": "Llava Next segfault and Qwen3OmniMoe video-processing ValueError are different model-loading/processing failures." + "reason": "Segfault on Llava Next loading vs Qwen3OmniMoe video-input ValueError; both multimodal, but distinct failures." }, { "left": "issue:43377", "right": "issue:43819", "accept": false, - "reason": "Both are audio reconstruction/comparison bugs, but they involve different models and different missing mechanisms (padding mask vs STE)." + "reason": "MIMI batched-vs-single padding-mask mismatch vs DAC.from_latents missing STE mismatch; unrelated audio-model correctness bugs." }, { "left": "issue:44451", "right": "issue:45081", "accept": false, - "reason": "Generic model load failure and Mistral regex-patch tokenizer crash do not share the same root cause." + "reason": "Different model-specific loading failure vs Mistral tokenizer regex patch crash." }, { "left": "issue:44987", "right": "issue:45081", "accept": false, - "reason": "Different failure surfaces: loading a specific model repo vs tokenizer regex patch crash." + "reason": "Model loading regression for physical-intelligence/fast vs Mistral tokenizer regex backend issue; different subsystems." }, { "left": "issue:44265", "right": "issue:44610", "accept": false, - "reason": "torch.export failure with torch_compilable_check is unrelated to OmDet processor image-size mismatch." + "reason": "torch.export/exportability with torch_compilable_check vs processor size mismatch; not the same defect." }, { "left": "issue:44991", "right": "issue:46315", "accept": false, - "reason": "Tokenizer load failure and slow convert_tokens_to_ids performance regression are different bugs." + "reason": "Tokenizer loading failure for a specific model vs slow convert_tokens_to_ids regression; different functions and symptoms." }, { "left": "issue:42915", "right": "issue:44291", "accept": false, - "reason": "Different error classes and code paths despite both being v5-era regressions." + "reason": "FineGrainedFP8Config on Qwen3Moe vs init_empty_weights/_is_hf_initialized argument mismatch; unrelated initialization paths." }, { "left": "issue:44220", "right": "issue:44683", "accept": false, - "reason": "Audio fbank extraction issue and compiled flex_attention torch>=2.9 failure are unrelated." + "reason": "Audio feature extraction issue vs compiled flex_attention on torch>=2.9; unrelated." } ] }, @@ -3305,8 +3249,8 @@ ], "soft_pairs": [ "issue:45216|issue:45357", - "issue:44062|issue:44987", "issue:36032|issue:43742", + "issue:44062|issue:44987", "issue:46209|issue:46326", "issue:43632|issue:44756", "issue:42915|issue:44589", @@ -3315,8 +3259,8 @@ "issue:43404|issue:45127", "issue:43531|issue:44877", "issue:40444|issue:45325", - "issue:43611|issue:44534", "issue:43296|issue:43531", + "issue:43611|issue:44534", "issue:43425|issue:44403", "issue:40990|issue:44811", "issue:43792|issue:44589", @@ -3334,218 +3278,203 @@ "issue:44829|issue:45479", "issue:43482|issue:43525" ], - "prepared_review_unit_hash": "4d2d30ce9149dbf23e51fb608a3aa96dbff1191552b5d42fff0a9d2456de563f", + "prepared_review_unit_hash": "624248e23865daf1e07182c9a36a9e271ad65da8ec7db4a6240d8068ca30b4ca", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 29, - "serialized_chars": 35806, - "estimated_input_tokens": 8952, - "estimated_eval_tokens": 18160 + "serialized_chars": 35802, + "estimated_input_tokens": 8951, + "estimated_eval_tokens": 18158 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly unrelated issues were grouped together. I only accept two true duplicate pairs: the Qwen3.5 save_pretrained regression and the AutoTokenizer/tokenizer.json Granite regression. The rest are adjacent by subsystem or model family, but not the same underlying bug.", - "confidence": 0.92, - "canonical_issue_reason": "Issue 45812 is the broadest, highest-signal representative of a real duplicate set: it describes the tokenizer regression at the root level, while 44462 reads like a narrower symptom. It also has more inbound references.", + "summary": "This cluster is heterogeneous: most items are distinct regressions or feature requests across tokenizers, Qwen/VL, FP8, checkpointing, and loading paths. There is no strong duplicate set, so the best single issue is only an umbrella representative, not a true canonical duplicate.", + "confidence": 0.31, + "canonical_issue_reason": "issue:45020 is the broadest umbrella report in the set, covering recent-version breakages in model loading/remote code, but it still does not subsume the many unrelated specific bugs here.", "canonical_pr_reason": null, - "best_issue_reason": "45812 is the best single issue to keep for this cluster because it captures the clearest cross-cutting bug and is the most likely root report among the few plausible duplicates.", + "best_issue_reason": "issue:45020 is the best representative because it captures the widest cross-cutting regression theme among these otherwise unrelated issues.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45216", "right": "issue:45357", - "accept": true, - "reason": "Same Qwen3.5 save_pretrained regression: both report incorrect checkpoint serialization, with 45357 reading as a later/continued symptom." - }, - { - "left": "issue:44062", - "right": "issue:44987", "accept": false, - "reason": "Both are tokenizer-related, but the concrete failures differ: AddedToken argument collision vs loading a specific model tokenizer in transformers 5.1." + "reason": "Both are Qwen3.5 save_pretrained regressions, but they describe different failures: overall checkpoint corruption vs incorrect visual-encoder keys. Not the same concrete bug." }, { "left": "issue:36032", "right": "issue:43742", "accept": false, - "reason": "Different models and different errors; one is a T5 tokenizer method conflict, the other is a key error loading MobileLLM-125M." + "reason": "Tokenizer constructor conflict in T5 vs a KeyError while loading MobileLLM-125M. Different subsystems and symptoms." + }, + { + "left": "issue:44062", + "right": "issue:44987", + "accept": false, + "reason": "AddedToken special-argument duplication is a tokenizer API bug; loading physical-intelligence/fast is a model loading regression. No shared concrete code path." }, { "left": "issue:46209", "right": "issue:46326", "accept": false, - "reason": "Completely different code paths: FP8 inference on B200 vs mistral_common continue_final_message behavior." + "reason": "FP8 inference failure on B200 and continue_final_message tokenizer behavior are unrelated." }, { "left": "issue:43632", "right": "issue:44756", "accept": false, - "reason": "One is an _is_hf_initialized regression, the other is an mmap/OOM workaround on Strix Halo; not the same bug." + "reason": "_is_hf_initialized init regression and Strix Halo mmap OOM are unrelated bugs." }, { "left": "issue:42915", "right": "issue:44589", "accept": false, - "reason": "Both involve FP8, but one is a Qwen3Moe FineGrainedFP8Config failure and the other is a missing Float8 storage type; distinct failures." + "reason": "Both mention FP8, but one is a Qwen3MoE FineGrainedFP8Config failure and the other is a missing Float8 storage type error. Different failure modes." }, { "left": "issue:44479", "right": "issue:45325", "accept": false, - "reason": "Same model family, but different regressions: video input handling versus still-image temporal position_id scaling." + "reason": "Both involve Qwen VL media handling, but one is a broad video-input regression while the other is a specific still-image temporal position_id scaling bug." }, { "left": "issue:36032", "right": "issue:41720", "accept": false, - "reason": "No shared underlying bug; tokenizer method conflict vs auto device mapping cuda assert." + "reason": "T5 tokenizer load error and Qwen3 auto device mapping cuda assert are unrelated." }, { "left": "issue:43404", "right": "issue:45127", "accept": false, - "reason": "Both mention weights/tie_word_embeddings, but one is a missing lm_head tie in a model class and the other is LoRA merge collapse; not the same fix." + "reason": "Mistral3 lm_head tying bug and LoRA merge collapse with extended vocab are different issues, even if both touch tied weights." }, { "left": "issue:43531", "right": "issue:44877", "accept": false, - "reason": "Different issues in different components: Qwen3-MoE sliding_window behavior vs strict config loading for granite_speech." + "reason": "Qwen3-MoE sliding_window behavior and granite_speech strict-config loading are unrelated." }, { "left": "issue:40444", "right": "issue:45325", "accept": false, - "reason": "Same Qwen2.5-VL area, but one is iterable-dataset multi-image finetuning failure and the other is a rope/position-id scaling regression." + "reason": "Both are Qwen2.5-VL-related, but one is training with multiple images per prompt and the other is a still-image position-id scaling regression." }, { - "left": "issue:43611", - "right": "issue:44534", + "left": "issue:43296", + "right": "issue:43531", "accept": false, - "reason": "Unrelated regressions: base_model_prefix loading broke in v5, while non-persistent buffers are being filled with junk." + "reason": "PaddleOCR-VL loading in vLLM and Qwen3-MoE sliding_window are different model/load-path bugs." }, { - "left": "issue:43296", - "right": "issue:43531", + "left": "issue:43611", + "right": "issue:44534", "accept": false, - "reason": "Different model families and different failure modes; PaddleOCR-VL loading vs Qwen3-MoE sliding_window." + "reason": "base_model_prefix loading breakage and non-persistent buffer junk are unrelated implementation issues." }, { "left": "issue:43425", "right": "issue:44403", "accept": false, - "reason": "Torch version incompatibility is not the same as noisy load logs; these are separate problems." + "reason": "Torch 2.10 incompatibility is a compatibility issue; unnecessary load noise is a separate UX/logging bug." }, { "left": "issue:40990", "right": "issue:44811", "accept": false, - "reason": "No shared root cause: perplexity issue on gpt-oss-20b vs Whisper batch_decode skip_special_tokens handling." + "reason": "High perplexity on gpt-oss-20b and Whisper batch_decode skip_special_tokens behavior are unrelated." }, { "left": "issue:43792", "right": "issue:44589", "accept": false, - "reason": "Whisper model runtime failure and Float8 storage lookup failure are unrelated." + "reason": "Whisper-large-v2 runtime failure and missing Float8 storage type are unrelated." }, { "left": "issue:43632", "right": "issue:44534", "accept": false, - "reason": "Different v5 regressions affecting different internal mechanisms; not one underlying bug." + "reason": "Different regressions: init flag handling vs buffer initialization." }, { "left": "issue:44079", "right": "issue:45663", "accept": false, - "reason": "ModelOutput key assignment bug and Gemma-4 FSDP2 shared_kv_states KeyError are distinct issues." + "reason": "ModelOutput key assignment bug and Gemma-4 FSDP2 shared_kv_states KeyError are unrelated." }, { "left": "issue:44403", "right": "issue:44991", "accept": false, - "reason": "One concerns load-time noise, the other tokenizer loading for est-roberta; not the same defect." + "reason": "Loading noise and tokenizer loading regression for EMBEDDIA/est-roberta are different issues." }, { "left": "issue:33666", "right": "issue:40444", "accept": false, - "reason": "Both are Qwen-VL training issues, but one is multi-GPU training and the other is iterable-dataset multi-image finetuning; different bugs." + "reason": "Both are Qwen-VL training reports, but multi-GPU training and iterable-dataset multi-image failure are not the same bug." }, { "left": "issue:43421", "right": "issue:44568", "accept": false, - "reason": "Both are special-token related, but one is runtime post-processor updates and the other is add_special_tokens not inserting BOS/EOS for a specific tokenizer." + "reason": "Runtime special-token post-processor update bug and mdeberta add_special_tokens BOS/EOS regression are related in theme but not the same concrete failure." }, { "left": "issue:44683", "right": "issue:45020", "accept": false, - "reason": "Different layers entirely: compiled flex_attention on torch >=2.9 vs remote_code loading regressions." + "reason": "Torch flex_attention compilation failure and remote_code loading regressions are unrelated." }, { "left": "issue:43381", "right": "issue:45923", "accept": false, - "reason": "Both mention gradient checkpointing, but one is eval-mode misuse and the other is a missing capability flag on a remote-code model." + "reason": "Eval-mode gradient checkpointing restriction and missing supports_gradient_checkpointing on a trust_remote_code model are different code paths." }, { "left": "issue:44291", "right": "issue:44589", "accept": false, - "reason": "Different failures: init_empty_weights gets an unexpected argument, while this one cannot find a Float8 storage object." + "reason": "Unexpected _is_hf_initialized argument during init_empty_weights and missing Float8 storage type are separate loading/runtime issues." }, { "left": "issue:35707", "right": "issue:40444", "accept": false, - "reason": "Different generation/training scenarios and different failure modes; not the same underlying bug." + "reason": "Progressive generation with inputs_embeds/past_key_values and Qwen2.5-VL multi-image finetuning failure are unrelated." }, { "left": "issue:44462", "right": "issue:45812", - "accept": true, - "reason": "Same tokenizer regression family: 44462 says AutoTokenizer ignores repository tokenizer.json, and 45812 reports the resulting wrong token IDs for Granite models." - }, - { - "left": "issue:43482", - "right": "issue:45538", - "accept": false, - "reason": "Unrelated issues: GGUF loading with transformers v5 vs CLIPTokenizer model_max_length semantics." - }, - { - "left": "issue:44829", - "right": "issue:45479", "accept": false, - "reason": "Both involve degenerate training, but the root causes differ: flash_attention_3 behavior vs a loss/label setup issue." + "reason": "Both concern AutoTokenizer behavior, but one ignores tokenizer.json from the repo and the other produces wrong Granite token IDs. Too little evidence of the same underlying bug." }, { "left": "issue:43482", - "right": "issue:43525", + "right": "issue:45538", "accept": false, - "reason": "No meaningful overlap beyond being load-time failures; Qwen2.5-GGUF loading and Llama4Config pad_token_id are separate." + "reason": "Qwen2.5-GGUF loading with transformers v5 and CLIPTokenizer max_length semantics are unrelated." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet and stays mostly conservative. The two accepted pairs are plausibly the same underlying regressions based on their titles, while the rejected pairs are generally framed as adjacent-but-distinct bugs rather than definite duplicates. No clear overstatement stands out from the available evidence." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -3612,8 +3541,8 @@ ], "soft_pairs": [ "issue:43317|issue:43856", - "issue:43818|issue:46402", "issue:42175|issue:45542", + "issue:43818|issue:46402", "issue:41553|issue:43582", "issue:44079|issue:44534", "issue:43818|issue:45357", @@ -3626,8 +3555,8 @@ "issue:43749|issue:44079", "issue:45325|issue:45381", "issue:43611|issue:43646", - "issue:43632|issue:45072", "issue:42175|issue:43296", + "issue:43632|issue:45072", "issue:43792|issue:43957", "issue:43994|issue:44898", "issue:45325|issue:45834", @@ -3641,218 +3570,215 @@ "issue:43296|issue:45800", "issue:43262|issue:43746" ], - "prepared_review_unit_hash": "2e44c74017f62985395dfa0eba00dc985c726cfadf891b538a2b1aabeac13413", + "prepared_review_unit_hash": "578ee650938b455b86ab8e7a7ee2321bd80b2f8aa788c28ca3afb72bbfded906", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 29, - "serialized_chars": 35884, - "estimated_input_tokens": 8971, - "estimated_eval_tokens": 18198 + "serialized_chars": 35885, + "estimated_input_tokens": 8972, + "estimated_eval_tokens": 18200 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most pairs are related only at a subsystem level and should stay separate. The only clear duplicate-family matches are the Transformers v5 model-initialization regressions (43611/43646) and, more tentatively, the Qwen2.5-VL position-id regressions (45325/45381). No PRs are present.", - "confidence": 0.69, - "canonical_issue_reason": "issue:43646 is the broadest representative of the Transformers v5 initialization regression family; it captures the underlying custom-model init breakage better than the narrower base_model_prefix symptom.", + "summary": "These items are mostly separate bug reports, not true duplicates. The only clear common theme is a broad set of Transformers v5 loading/init regressions, but the proposed soft pairs each point to different code paths or different model families, so none should be merged.", + "confidence": 0.86, + "canonical_issue_reason": "Issue 43611 is the broadest, most representative loading regression in the set; it captures a core Transformers v5 breakage affecting model loading, whereas the others are narrower variants or unrelated bugs.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43646 is the most general and reusable issue report among the accepted duplicate candidates, making it the best single issue to anchor triage.", + "best_issue_reason": "43611 is the best single issue to anchor a broader v5 initialization/loading regression cluster because it is general, concrete, and likely relevant to multiple downstream failures.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43317", "right": "issue:43856", "accept": false, - "reason": "Different bugs: one is load/offload failure with dequantized models, the other is Qwen3 MoE training memory usage." + "reason": "Both mention device/offload behavior, but one is a dequantized model load failure and the other is Qwen3 MoE training memory usage; different bug classes." }, { - "left": "issue:43818", - "right": "issue:46402", + "left": "issue:42175", + "right": "issue:45542", "accept": false, - "reason": "Unrelated model families and failure modes: Video-LLaVA weight sharing vs Sam3 text-encoder loading." + "reason": "Both involve TensorFlow/backend packaging, but one is a missing dependency in the torch extra and the other is a tensorboard-without-TF runtime backend error." }, { - "left": "issue:42175", - "right": "issue:45542", + "left": "issue:43818", + "right": "issue:46402", "accept": false, - "reason": "Both involve TensorFlow availability, but one is packaging extras and the other is an undefined backend error; not the same bug." + "reason": "Different models and failures: Video-LLaVA video_tower wiring vs Sam3Model text encoder weights not loading." }, { "left": "issue:41553", "right": "issue:43582", "accept": false, - "reason": "Completely different areas: tokenizer error message vs Apple Silicon warmup TypeError." + "reason": "AutoTokenizer/Voxtral error messaging and Apple Silicon caching_allocator_warmup TypeError are unrelated." }, { "left": "issue:44079", "right": "issue:44534", "accept": false, - "reason": "One is ModelOutput key assignment, the other is non-persistent buffer serialization; different code paths." + "reason": "ModelOutput key assignment bug and non-persistent buffer junk are distinct internal behaviors." }, { "left": "issue:43818", "right": "issue:45357", "accept": false, - "reason": "Different product areas: Video-LLaVA architecture bug vs Qwen3.5 save_pretrained visual-key regression." + "reason": "Video-LLaVA tower/weight-sharing bug is unrelated to Qwen3.5 save_pretrained key serialization." }, { "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": "Qwen3-MoE sliding_window behavior and init_empty_weights/_is_hf_initialized TypeError are different regressions." }, { "left": "issue:41720", "right": "issue:45081", "accept": false, - "reason": "Qwen3 device-mapping crash and Mistral tokenizer patch crash are unrelated." + "reason": "Qwen3 auto device mapping cuda assert and Mistral regex patching crash are unrelated code paths." }, { "left": "issue:43377", "right": "issue:45491", "accept": false, - "reason": "Both are batching/padding-related, but MIMI encoder padding-mask mismatch and Gemma3 NaNs are different bugs." + "reason": "MIMI padding-mask batching bug and Gemma3 NaN embeddings from sliding-window all-padding windows are different issues." }, { "left": "issue:40444", "right": "issue:45381", "accept": false, - "reason": "Both touch Qwen2.5-VL, but multiple-images-per-prompt finetuning failure is not the same as video position-id regression." + "reason": "IterableDataset multi-image finetuning failure is not the same as Qwen2.5-VL video vision_position_ids regression." }, { "left": "issue:43644", "right": "issue:44079", "accept": false, - "reason": "Tied-weight serialization inconsistency and ModelOutput key handling are different issues." + "reason": "Both touch model internals, but one is non-persistent buffer serialization and the other is ModelOutput key handling." }, { "left": "issue:44265", "right": "issue:45072", "accept": false, - "reason": "torch.export/compiler check failure and bfloat16 dtype mismatch are unrelated." + "reason": "torch.export with torch_compilable_check is a different failure mode from bf16 dtype mismatches in inference." }, { "left": "issue:43749", "right": "issue:44079", "accept": false, - "reason": "FSDP CPU RAM efficient loading is a loading-path bug, not a ModelOutput-key issue." + "reason": "FSDP CPU RAM efficient loading and ModelOutput key assignment are unrelated." }, { "left": "issue:45325", "right": "issue:45381", - "accept": true, - "reason": "Same Qwen2.5-VL visual position-id bug family; both report incorrect rope/vision position ids in the video/image path." + "accept": false, + "reason": "Both are Qwen2.5-VL position-id issues, but one is still-image RoPE scaling and the other is video input vision_position_ids; not the same bug." }, { "left": "issue:43611", "right": "issue:43646", - "accept": true, - "reason": "Same Transformers v5 model-init/load regression; base_model_prefix failures and custom initialization failures are the same underlying breakage." - }, - { - "left": "issue:43632", - "right": "issue:45072", "accept": false, - "reason": "_is_hf_initialized handling in v5 and dtype mismatch in inference are different regressions." + "reason": "Both are Transformers v5 regressions, but base_model_prefix loading failure and custom model initialization breakage are distinct concrete problems." }, { "left": "issue:42175", "right": "issue:43296", "accept": false, - "reason": "Package/backend installation issue vs vLLM model-load failure; not duplicates." + "reason": "TensorFlow packaging/backend issue and PaddleOCR-VL loading failure in vLLM are unrelated." + }, + { + "left": "issue:43632", + "right": "issue:45072", + "accept": false, + "reason": "_is_hf_initialized flag breakage and bfloat16 dtype mismatch are different failure modes." }, { "left": "issue:43792", "right": "issue:43957", "accept": false, - "reason": "Whisper runtime failure and meta-device loading breakage are distinct problems." + "reason": "Whisper can't run and torch.device('meta') loading failures affect different model-loading paths." }, { "left": "issue:43994", "right": "issue:44898", "accept": false, - "reason": "Different vision-model symptoms: SigLIP2 nonsense outputs vs Perceiver non-default resolution failure." + "reason": "Siglip2 nonsensical outputs and Perceiver non-default resolution failure are different vision-model bugs." }, { "left": "issue:45325", "right": "issue:45834", "accept": false, - "reason": "Both mention OCR/video-like inputs, but Qwen2.5-VL position ids and Kosmos2.5 long OCR index errors are not the same bug." + "reason": "Both are long-context/video-style positional issues, but one is Qwen2.5-VL RoPE scaling and the other is Kosmos2.5 OCR index overflow." }, { "left": "issue:43454", "right": "issue:44466", "accept": false, - "reason": "Both involve lm_head/tied weights, but one is a broken tie in model config and the other is device-dependent serialization inconsistency." + "reason": "AyaVision lm_head tying and inconsistent lm_head serialization are related only loosely; they are not the same concrete bug." }, { "left": "issue:42915", "right": "issue:43482", "accept": false, - "reason": "Different model-loading failures: Qwen3Moe FP8 config vs Qwen2.5 GGUF with transformers v5." + "reason": "Qwen3Moe FP8 failure and Qwen2.5-GGUF loading under v5 are unrelated." }, { "left": "issue:43701", "right": "issue:45325", "accept": false, - "reason": "Checkpoint resume key mismatch is unrelated to Qwen2.5-VL position-id computation." + "reason": "resume_from_checkpoint key mismatch is unrelated to Qwen2.5-VL position-id scaling." }, { "left": "issue:44479", "right": "issue:45823", "accept": false, - "reason": "Both are device/offload-related, but Qwen video regression and Gemma4 PLE forward mismatch are separate bugs." + "reason": "Qwen video-input regression and Gemma4 PLE device mismatch are different model-specific failures." }, { "left": "issue:44291", "right": "issue:44877", "accept": false, - "reason": "Unexpected _is_hf_initialized argument and strict granite_speech config loading are different issues." + "reason": "init_empty_weights TypeError and strict config blocking granite_speech config are different issues." }, { "left": "issue:43873", "right": "issue:45701", "accept": false, - "reason": "Quantization offload behavior and tokenization changes across versions are unrelated." + "reason": "Quantization/offloading behavior and tokenization changes across versions are not duplicates." }, { "left": "issue:43606", "right": "issue:44265", "accept": false, - "reason": "CPU offload device mismatch and torch.export/compiler-check failure are different code paths." + "reason": "CPU offload device mismatch and torch.export/torch_compilable_check failure are unrelated." }, { "left": "issue:43296", "right": "issue:45800", "accept": false, - "reason": "PaddleOCR-VL/vLLM load failure and torch/transformers version incompatibility are not the same bug." + "reason": "PaddleOCR-VL load failure in vLLM and torch 2.4.1/transformers 5.8 incompatibility are different problems." }, { "left": "issue:43262", "right": "issue:43746", "accept": false, - "reason": "Audio chat-template sampling-rate default and GraniteSpeech PEFT checkpoint loading are unrelated." + "reason": "Audio chat-template sample-rate defaulting and GraniteSpeech PEFT checkpoint loading are unrelated." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet, and the soft-edge calls are conservative overall. The two accepted duplicate-family pairs are reasonably supported by the issue titles, and the many rejections avoid overstating similarity. No PRs are present, so the null PR fields are appropriate." - }, - "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 }, @@ -3934,8 +3860,8 @@ "issue:43257|issue:45923", "issue:43572|issue:45759", "issue:43317|issue:43873", - "issue:43377|issue:44898", "issue:40990|issue:42947", + "issue:43377|issue:44898", "issue:43322|issue:46050", "issue:43454|issue:43720", "issue:41720|issue:43322", @@ -3949,20 +3875,20 @@ "issue:40990|issue:45701", "issue:43881|issue:44991" ], - "prepared_review_unit_hash": "6e157776a4ad77e721deb1c5b1a02fa5219f3331e063296532503f0eb32a4833", + "prepared_review_unit_hash": "a95972fe420ce92760dbf0d0f1010eb0b6a5d4ba370778af94ce55603fe55465", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 30, - "serialized_chars": 36233, + "serialized_chars": 36234, "estimated_input_tokens": 9059, "estimated_eval_tokens": 18374 }, @@ -3971,192 +3897,192 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly a grab-bag of model-specific regression reports, but the closest common thread is loading/offloading/quantization behavior. The best umbrella issue is the generic offloading + quantization report.", - "confidence": 0.68, - "canonical_issue_reason": "issue:43873 is the broadest and most reusable issue in the set: it captures the shared offloading/quantization failure mode without being tied to one model, and it has enough discussion to serve as an umbrella.", + "summary": "No strong duplicate cluster emerges here. The soft links are mostly broad thematic overlaps (device_map/offloading, tokenizer regressions, serialization, generation) rather than the same underlying bug. No PRs are present.", + "confidence": 0.94, + "canonical_issue_reason": "issue:41720 is the clearest concrete runtime regression in the set and the most central anchor for the device-mapping/offloading subgroup, even though the neighboring reports are not true duplicates.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43873 is the strongest global representative because it is open, broadly phrased, and closest to the recurring loading/offload regressions that show up in nearby issues.", + "best_issue_reason": "issue:41720 is the best single issue to anchor triage because it is specific, reproducible, and has the strongest thematic overlap with several other candidates, but it should not absorb the others as duplicates.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:40990", "right": "issue:45584", "accept": false, - "reason": "Different symptoms and domains: perplexity on GPT-OSS vs Whisper empty-transcription generation." + "reason": "Different bugs: high perplexity on WikiText-2 vs Whisper empty-transcription generation after token alignment." }, { "left": "issue:34634", "right": "issue:35707", "accept": false, - "reason": "Unrelated models and bugs: Bark voice preset vs progressive generation with inputs_embeds/past_key_values." + "reason": "Processor voice_preset bug and progressive generation with inputs_embeds/past_key_values are unrelated code paths." }, { "left": "issue:35707", "right": "issue:41720", "accept": false, - "reason": "Different code paths: generation with cached embeddings vs Qwen3 auto device mapping crash." + "reason": "Generation cache behavior and Qwen3 auto device-map CUDA assert are different failures." }, { "left": "issue:44568", "right": "issue:45920", "accept": false, - "reason": "Both tokenizer-related, but one is special-token insertion for a specific tokenizer and the other is wrong token IDs across many models; not the same concrete bug." + "reason": "One is a missing BOS/EOS insertion bug for a specific tokenizer; the other is a broad wrong-token-IDs regression across multiple models." }, { "left": "issue:44186", "right": "issue:44265", "accept": false, - "reason": "Different subsystems: LayoutLMv2 tokenizer padding/NER crash vs torch.export failure on torch_compilable_check." + "reason": "Tokenizer NER padding/truncation crash is not the same as torch.export.export failing on torch_compilable_check." }, { "left": "issue:41720", "right": "issue:45823", "accept": false, - "reason": "Both mention device_map=auto, but they are different models and different failure modes (assert vs PLE forward mismatch)." + "reason": "Both involve device_map=auto, but they affect different models and fail differently (CUDA assert vs device mismatch in forward)." }, { "left": "issue:43606", "right": "issue:43819", "accept": false, - "reason": "Both are device-related, but one is CPU offload mismatch for Bark and the other is DAC latent/STE mismatch." + "reason": "Bark CPU offload mismatch and DAC.from_latents missing STE are separate model/runtime bugs." }, { "left": "issue:43454", "right": "issue:43550", "accept": false, - "reason": "Unrelated bugs: weight tying in AyaVision vs torch.compile/SDPA on Bamba." + "reason": "Weight tying/garbage text generation is unrelated to torch.compile + SDPA failures." }, { "left": "issue:39692", "right": "issue:40444", "accept": false, - "reason": "Different problems: SigLIP2 docs/model-processor mismatch vs Qwen2.5-VL finetuning with multiple images." + "reason": "Documentation example errors and multi-image IterableDataset finetuning failure are different problems." }, { "left": "issue:43881", "right": "issue:44488", "accept": false, - "reason": "Both are load failures, but for different models and likely different causes; not enough to be the same bug." + "reason": "Different models with different load failures; no shared concrete code-path." }, { "left": "issue:46305", "right": "issue:46315", "accept": false, - "reason": "Completely different areas: streaming window masking vs tokenizer conversion performance." + "reason": "Sliding-window off-by-one in masking is unrelated to tokenizer conversion performance regression." }, { "left": "issue:43377", "right": "issue:44610", "accept": false, - "reason": "Different model components: MIMI encoder padding-mask correctness vs OmDet-Turbo processor image size mismatch." + "reason": "Missing padding-mask support in MIMI encoder and OmDet processor/model size mismatch are not the same bug." }, { "left": "issue:43454", "right": "issue:45072", "accept": false, - "reason": "Different issues: tied weights/garbage generation vs bfloat16 dtype mismatch." + "reason": "lm_head tying/garbage generation is unrelated to bfloat16 dtype mismatches in inference." }, { "left": "issue:43257", "right": "issue:45923", "accept": false, - "reason": "Different bugs: Qwen3 MoE weight conversion with accelerate/deepspeed vs missing gradient-checkpointing flag." + "reason": "MOE weight conversion with accelerate+deepspeed and a missing supports_gradient_checkpointing flag are unrelated." }, { "left": "issue:43572", "right": "issue:45759", "accept": false, - "reason": "Different configuration problems: missing pad_token_idx vs from_config text_config unwrapping for composite multimodal configs." + "reason": "Missing pad_token_idx in StableLmConfig is unrelated to unwrap-text_config behavior for composite Qwen configs." }, { "left": "issue:43317", "right": "issue:43873", "accept": false, - "reason": "Related offloading/quantization theme, but the reports are still distinct: dequantized model loading on gpu+cpu offload vs generic offloading not working as expected with quantization." + "reason": "Closely themed around offloading/quantization, but not the same concrete failure; 43873 is a broad umbrella report." }, { - "left": "issue:43377", - "right": "issue:44898", + "left": "issue:40990", + "right": "issue:42947", "accept": false, - "reason": "Different model families and failures: MIMI batching/padding vs Perceiver image resolution/interpolation." + "reason": "Perplexity regression on GPT OSS is unrelated to gradient checkpointing not taking effect with PEFT LoRA." }, { - "left": "issue:40990", - "right": "issue:42947", + "left": "issue:43377", + "right": "issue:44898", "accept": false, - "reason": "Unrelated: perplexity regression vs gradient checkpointing ineffectiveness with PEFT LoRA." + "reason": "Batched-padding discrepancy in MIMI and Perceiver non-default-resolution failure are different issues." }, { "left": "issue:43322", "right": "issue:46050", "accept": false, - "reason": "Different bug classes: Llava Next load-time segfault vs sinq quantization save/load corruption." + "reason": "Segmentation fault when loading Llava Next and sinq save/load corruption are distinct load-path bugs." }, { "left": "issue:43454", "right": "issue:43720", "accept": false, - "reason": "Different code paths: lm_head weight tying vs packed-weight unpacking during accelerate loading." + "reason": "Weight tying problem and BitNet packed-weight unpacking during accelerate loading are unrelated." }, { "left": "issue:41720", "right": "issue:43322", "accept": false, - "reason": "Both are loading/runtime failures, but for different models and root causes." + "reason": "Both are model-loading/runtime failures, but the failure modes and affected models are different." }, { "left": "issue:45325", "right": "issue:46315", "accept": false, - "reason": "Different areas entirely: Qwen2.5-VL rope indexing vs tokenizer conversion speed." + "reason": "Qwen2.5-VL rope-index scaling bug is unrelated to tokenizer conversion speed regression." }, { "left": "issue:43749", "right": "issue:44534", "accept": false, - "reason": "Both involve loading/state handling, but one is FSDP CPU-efficient loading and the other is non-persistent buffer initialization; not the same bug." + "reason": "FSDP CPU RAM efficient loading and junk-filled non-persistent buffers are separate regressions." }, { "left": "issue:44991", "right": "issue:46291", "accept": false, - "reason": "Unrelated: tokenizer loading regression vs kernels dependency requiring version/revision." + "reason": "Tokenizer loading failure and kernels v0.15 import/version-revision failure do not share the same bug." }, { "left": "issue:42617", "right": "issue:42915", "accept": false, - "reason": "Different model/runtime problems: 3d_parallel.py execution vs Qwen3Moe FineGrainedFP8Config failure." + "reason": "3d_parallel.py execution failure and FineGrainedFP8Config failure are different issues." }, { "left": "issue:41720", "right": "issue:44977", "accept": false, - "reason": "Both touch Qwen3-family inference, but one is device-map crash and the other is flash-attention generation quality/behavior; not the same concrete bug." + "reason": "Same model family, but different symptoms and likely different code paths (CUDA assert vs flash-attention generation issue)." }, { "left": "issue:44466", "right": "issue:45357", "accept": false, - "reason": "Both are save_pretrained serialization issues, but they affect different tensors and different models." + "reason": "Both concern saving/serialization, but one is tied lm_head.weight behavior and the other is incorrect visual encoder keys." }, { "left": "issue:43526", "right": "issue:43883", "accept": false, - "reason": "Unrelated: BeitImageProcessorFast label reduction vs Molmo tied-weights attribute missing." + "reason": "BeitImageProcessorFast label reduction bug and Molmo missing all_tied_weights_keys are unrelated." }, { "left": "issue:40990", "right": "issue:45701", "accept": false, - "reason": "Different symptom space: perplexity issue vs tokenizer behavior changes across versions." + "reason": "Perplexity regression is not the same as version-dependent tokenization changes." }, { "left": "issue:43881", "right": "issue:44991", "accept": false, - "reason": "Different failures: model load crash vs tokenizer load regression for a different repository." + "reason": "Different models and different failure surfaces: model loading vs tokenizer loading." } ] }, @@ -4233,17 +4159,17 @@ "issue:36331|issue:42617", "issue:43065|issue:46402", "issue:44186|issue:45072", - "issue:44008|issue:45823", "issue:43577|issue:43708", - "issue:36331|issue:42915", + "issue:44008|issue:45823", "issue:44877|issue:44991", + "issue:36331|issue:42915", "issue:43450|issue:45701", "issue:43873|issue:44756", "issue:43550|issue:43720", "issue:43404|issue:45357", "issue:42175|issue:46291", - "issue:44423|issue:45406", "issue:41720|issue:42617", + "issue:44423|issue:45406", "issue:43550|issue:43606", "issue:44610|issue:44898", "issue:44186|issue:44610", @@ -4251,204 +4177,216 @@ "issue:43295|issue:45701", "issue:45632|issue:45684", "issue:43425|issue:46291", - "issue:41720|issue:45161", "issue:36010|issue:42617", + "issue:41720|issue:45161", "issue:43525|issue:46291", - "issue:44671|issue:44743", "issue:45081|issue:45812", + "issue:44671|issue:44743", "issue:43531|issue:43572", "issue:43316|issue:44913" ], - "prepared_review_unit_hash": "e57a32d3b6ab5f0bd3309d26800b2d2c94fcb84652066ca0a15c9ba4594e3b82", + "prepared_review_unit_hash": "a6b9e66582ea78f13181a010ad9bd07284df534df5e2bd50641768b19fcc7645", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 29, - "serialized_chars": 35944, - "estimated_input_tokens": 8986, - "estimated_eval_tokens": 18228 + "serialized_chars": 35945, + "estimated_input_tokens": 8987, + "estimated_eval_tokens": 18230 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: the soft pairs are mostly same-subsystem but not the same bug/change, so I reject them all. The broadest and most representative issue here is the open quantization/offload report, but there is no real duplicate canonical across the set.", - "confidence": 0.71, - "canonical_issue_reason": "Issue 43873 is the broadest, most active, and most referenced report in this set, so it is the best anchor if one must be chosen; however, the cluster does not contain a true duplicate group.", + "summary": "These items span many unrelated regressions across different subsystems. None of the soft pairs look like the same underlying bug or change, so I would reject all duplicate merges. If an anchor is needed, issue 43873 is the broadest discussion-heavy issue, but it is not a true duplicate match for the others.", + "confidence": 0.93, + "canonical_issue_reason": "Issue 43873 is the broadest and most actively discussed item here, with multiple inbound references and a recurring offloading/quantization theme, so it is the best available cluster anchor.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 43873 has the strongest combination of scope (offloading + quantization), discussion, and external references, making it the most suitable representative issue among otherwise unrelated reports.", + "best_issue_reason": "Issue 43873 is the most suitable representative issue because it has the widest scope and the strongest evidence of ongoing community impact, even though this set does not contain a clear duplicate group.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44387", "right": "issue:45704", "accept": false, - "reason": "Both involve memory symptoms, but one is int4 quantization OOM and the other is an apex RMSNorm leak; different code paths and fixes." + "reason": "Different problems: int4 quantization OOM vs T5/apex RMSNorm memory leak." }, { "left": "issue:36331", "right": "issue:42617", "accept": false, - "reason": "Trainer compute_loss signature breakage is unrelated to not being able to run 3d_parallel.py." + "reason": "Trainer custom compute_loss keyword mismatch is unrelated to failing to run a 3d_parallel script." }, { "left": "issue:43065", "right": "issue:46402", "accept": false, - "reason": "Both are Sam3-related, but one is a dummy Conv2d in the pixel decoder and the other is missing text-encoder weights." + "reason": "Different SAM3 bugs: dummy Conv2d vs text encoder weights not loading." }, { "left": "issue:44186", "right": "issue:45072", "accept": false, - "reason": "Tokenizer padding/NER crash and bfloat16 dtype mismatch are unrelated." - }, - { - "left": "issue:44008", - "right": "issue:45823", - "accept": false, - "reason": "Different models and different failures: variable-name collision in Gemma3n vs device mismatch in Gemma4 PLE." + "reason": "Tokenizer crash on NER/padding is unrelated to dtype mismatch in bfloat16 inference." }, { "left": "issue:43577", "right": "issue:43708", "accept": false, - "reason": "BLIP2 dtype loading bug and Trainer resume/max_steps regression are separate problems." + "reason": "BLIP2 dtype propagation bug is unrelated to Trainer resume/max_steps recalculation." }, { - "left": "issue:36331", - "right": "issue:42915", + "left": "issue:44008", + "right": "issue:45823", "accept": false, - "reason": "Unexpected compute_loss kwarg and Qwen3 MoE FP8 failure do not share the same underlying bug." + "reason": "Different Gemma models and failures: variable-name collision in forward vs PLE device mismatch." }, { "left": "issue:44877", "right": "issue:44991", "accept": false, - "reason": "Both are loading issues, but one is strict config handling and the other is tokenizer loading for a specific model." + "reason": "Strict config loading for granite_speech is unrelated to tokenizer loading for EMBEDDIA/est-roberta." + }, + { + "left": "issue:36331", + "right": "issue:42915", + "accept": false, + "reason": "CustomTrainer loss signature mismatch is unrelated to Qwen3MoE FP8 failure." }, { "left": "issue:43450", "right": "issue:45701", "accept": false, - "reason": "Video processor batching shape bug is unrelated to tokenization changing across versions." + "reason": "Video processor batched-shape bug is unrelated to a broad tokenization version-change report." }, { "left": "issue:43873", "right": "issue:44756", "accept": false, - "reason": "Both involve memory/offload, but one is quantization offloading and the other is mmap/OOM on Strix Halo." + "reason": "Both involve loading/offloading, but one is quantization offload behavior and the other is mmap OOM on Strix Halo; not the same bug." }, { "left": "issue:43550", "right": "issue:43720", "accept": false, - "reason": "torch.compile/SDPA failure and accelerate packed-weight loading failure are different code paths." + "reason": "Both are BitNet/SDPA vs packed-weight-unpacking issues; different failure paths and fixes." }, { "left": "issue:43404", "right": "issue:45357", "accept": false, - "reason": "Mistral lm_head tying and Qwen3.5 visual encoder key saving are unrelated serialization/model issues." + "reason": "Mistral3 lm_head tying bug is unrelated to Qwen3.5 visual encoder key saving." }, { "left": "issue:42175", "right": "issue:46291", "accept": false, - "reason": "Packaging the torch extra and kernels version/revision requirements are unrelated dependency issues." + "reason": "TensorFlow backend packaging issue is unrelated to kernels version/revision import failure in evaluate." + }, + { + "left": "issue:41720", + "right": "issue:42617", + "accept": false, + "reason": "Qwen3 auto device-map cuda assert is unrelated to a script execution failure." }, { "left": "issue:44423", "right": "issue:45406", "accept": false, - "reason": "Both affect serving, but one is continuous-batching with multimodal inputs and the other is a Gemma4Processor attribute error." + "reason": "Continuous-batching multimodal crash is unrelated to Gemma4Processor missing _tokenizer in serve." + }, + { + "left": "issue:43550", + "right": "issue:43606", + "accept": false, + "reason": "Different subsystems and symptoms: torch.compile+SDPA in Bamba vs CPU offload device mismatch in bark-small." }, { "left": "issue:44610", "right": "issue:44898", "accept": false, - "reason": "Both are image-resolution mismatches, but they affect different models and processor/model contracts." + "reason": "OmDet processor output size bug is different from Perceiver non-default resolution failure." }, { "left": "issue:44186", "right": "issue:44610", "accept": false, - "reason": "Tokenizer crash on NER/batched padding is unrelated to OmDet-Turbo image size mismatch." + "reason": "Tokenizer/NER padding crash and image processor resolution mismatch are unrelated." }, { "left": "issue:42915", "right": "issue:43317", "accept": false, - "reason": "Qwen3 MoE FP8 failure and dequantized model offload failure are different bugs." + "reason": "Qwen3Moe FP8 failure is unrelated to dequantized model loading with auto device_map offload." }, { "left": "issue:43295", "right": "issue:45701", "accept": false, - "reason": "Regression around processor.tokenizer/images is not the same as a general tokenization change across versions." + "reason": "Processor/tokenizer regression is unrelated to a general tokenization-version report." }, { "left": "issue:45632", "right": "issue:45684", "accept": false, - "reason": "Cache-path collision and read-only permission propagation are separate save/load issues." + "reason": "Both touch save/load paths, but cache-path collisions and read-only permission propagation are different bugs." }, { "left": "issue:43425", "right": "issue:46291", "accept": false, - "reason": "Torch 2.10 incompatibility and kernels import/version requirements are unrelated." + "reason": "Torch 2.10 incompatibility is unrelated to kernels version/revision requirements." }, { - "left": "issue:41720", - "right": "issue:45161", + "left": "issue:36010", + "right": "issue:42617", "accept": false, - "reason": "cudaErrorAssert on auto device mapping is not the same as TP failing for GPT-OSS MoE." + "reason": "GenerationMixin import error is unrelated to failing to run 3d_parallel.py." }, { - "left": "issue:36010", - "right": "issue:42617", + "left": "issue:41720", + "right": "issue:45161", "accept": false, - "reason": "GenerationMixin import error and 3d_parallel.py runtime failure are unrelated." + "reason": "Both mention model dispatching, but one is a cuda assert on Qwen3 and the other is only-TP not working on GPT-OSS MoE." }, { "left": "issue:43525", "right": "issue:46291", "accept": false, - "reason": "Missing pad_token_id in Llama4Config is unrelated to kernels import/version handling." + "reason": "Missing pad_token_id in Llama4Config is unrelated to kernels import/version requirements." }, { - "left": "issue:44671", - "right": "issue:44743", + "left": "issue:45081", + "right": "issue:45812", "accept": false, - "reason": "CamemBERT masked-LM regression and Qwen3.5 recurrent-state reset are different model-specific bugs." + "reason": "Mistral regex patch crash is unrelated to Granite token ID regression." }, { - "left": "issue:45081", - "right": "issue:45812", + "left": "issue:44671", + "right": "issue:44743", "accept": false, - "reason": "Mistral regex patch crash and Granite token-ID regression involve different tokenizer paths and failures." + "reason": "CamemBERT masked-LM predictions and Qwen3.5 recurrent-state handling are different code paths and failures." }, { "left": "issue:43531", "right": "issue:43572", "accept": false, - "reason": "Qwen3-MoE sliding_window behavior and StableLm pad_token_idx regression are unrelated config issues." + "reason": "Qwen3-MoE sliding-window behavior is unrelated to StableLmConfig missing pad_token_idx." }, { "left": "issue:43316", "right": "issue:44913", "accept": false, - "reason": "Gemma3TextConfig API discrepancy and GPTNeoX rotary_pct reload regression are different config bugs." + "reason": "Gemma3TextConfig API mismatch is unrelated to GPTNeoX rotary_pct reload regression." } ] }, @@ -4540,8 +4478,8 @@ "issue:44530|issue:46291", "issue:43421|issue:45701", "issue:45357|issue:46402", - "issue:44492|issue:44509", "issue:43606|issue:43653", + "issue:44492|issue:44509", "issue:43531|issue:45759", "issue:34634|issue:41720", "issue:42490|issue:43329", @@ -4551,20 +4489,20 @@ "issue:44451|issue:44991", "issue:43531|issue:44589" ], - "prepared_review_unit_hash": "7011f91bff7b82282dc749d02e763fe07497cdd0c91f0d241a4d63124e68e88f", + "prepared_review_unit_hash": "f6cc990cdb85435f074956b0b2f28365e4cd453d282d782424a3909e51fe0248", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 36231, + "serialized_chars": 36229, "estimated_input_tokens": 9058, "estimated_eval_tokens": 18372 }, @@ -4573,186 +4511,186 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab bag of unrelated Transformers issues. The only clear duplicate is the Qwen2.5-VL `apply_chat_template` crash on batched input with `padding=False` (44514/44545).", - "confidence": 0.64, - "canonical_issue_reason": "Issue 44514 is the earliest and most explicit report of the only confirmed duplicate bug in the set, with a concrete reproducer and clear failure condition.", + "summary": "Only one soft edge looks like a true duplicate: the two Qwen2_5_VLProcessor.apply_chat_template crash reports. The rest are distinct bugs across different components and should stay separate.", + "confidence": 0.96, + "canonical_issue_reason": "issue:44514 is the cleanest canonical anchor for the batched apply_chat_template crash: it was filed slightly earlier and describes the failure mode directly. issue:44545 is essentially the same report with a near-identical title.", "canonical_pr_reason": null, - "best_issue_reason": "44514 is the strongest representative because it is a precise, reproducible bug report and anchors the only accepted duplicate pair.", + "best_issue_reason": "issue:44514 best represents the duplicated Qwen2_5_VLProcessor.padding=False crash because it is specific, actionable, and the duplicate pair maps to the same underlying code path.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43540", "right": "issue:43782", "accept": false, - "reason": "Both involve Qwen multimodal paths, but one is a video-input ValueError and the other is a `from_pretrained`/`weight_only` load error in a different class and code path." + "reason": "Different models and different failure modes: Qwen3OmniMoe video-input ValueError vs Qwen3VL from_pretrained weight_only load error." }, { "left": "issue:42915", "right": "issue:43322", "accept": false, - "reason": "Different models and failure modes: FineGrainedFP8 loading for Qwen3-MoE vs a Llava Next segmentation fault." + "reason": "Different multimodal model families and distinct load-time errors; no shared code-path evidence." }, { "left": "issue:44186", "right": "issue:44361", "accept": false, - "reason": "Tokenizer bugs, but for different tokenizers, tasks, and error mechanisms; not the same underlying issue." + "reason": "Both are tokenizer-related but the bugs differ: NER/batched padding crash in LayoutLMv2Tokenizer vs AttributeError in MLukeTokenizer." }, { "left": "issue:43421", "right": "issue:45812", "accept": false, - "reason": "Both are tokenizer-related, but one is runtime post-processor synchronization and the other is a Granite token-id regression." + "reason": "Both involve tokenizers post-processing, but one is runtime special-token updates while the other is a Granite AutoTokenizer regression producing wrong IDs." }, { "left": "issue:36331", "right": "issue:41720", "accept": false, - "reason": "Unrelated: trainer loss signature mismatch vs Qwen3 auto device-mapping CUDA assert." + "reason": "Unrelated areas: trainer compute_loss signature mismatch vs Qwen3 auto device mapping CUDA assert." }, { "left": "issue:43526", "right": "issue:43881", "accept": false, - "reason": "ImageProcessor label reduction bug and glm-4v model loading failure are separate problems." + "reason": "Different bugs in different models: BeitImageProcessorFast label reduction vs glm-4v-9b loading failure." }, { "left": "issue:29159", "right": "issue:36032", "accept": false, - "reason": "Both mention tokenizers, but one is slow/fast behavioral inconsistency and the other is a T5Tokenizer load-time attribute conflict." + "reason": "Tokenizer inconsistency vs T5Tokenizer class/method conflict; these are not the same bug." }, { "left": "issue:33357", "right": "issue:35707", "accept": false, - "reason": "MacOS CLIP bus error and progressive generation with `inputs_embeds`/`past_key_values` are different code paths." + "reason": "A MacOS bus error on a CLIP community model is unrelated to generation with inputs_embeds and past_key_values." }, { "left": "issue:43388", "right": "issue:45059", "accept": false, - "reason": "Metrics label-truncation bug vs SAM3 text/bounding-box behavior; no shared concrete defect." + "reason": "Different subsystems and symptoms: gather_for_metrics label truncation vs SAM3 text/bbox behavior." }, { "left": "issue:30064", "right": "issue:34634", "accept": false, - "reason": "Void segmentation map processing and BarkProcessor `voice_preset` are unrelated image/audio processor issues." + "reason": "Both affect processors, but one is void segmentation-map handling and the other is BarkProcessor voice_preset support." }, { "left": "issue:43550", "right": "issue:43653", "accept": false, - "reason": "`torch.compile`/SDPA failure in Bamba is unrelated to BigBirdTokenizer special-token registration." + "reason": "Different code paths: torch.compile+SDPA failure in Bamba vs tokenizer mask-token registration/empty decode issue in BigBirdTokenizer." }, { "left": "issue:43606", "right": "issue:43720", "accept": false, - "reason": "Both touch loading/offload behavior, but they are different models and different bugs (CPU offload mismatch vs unpacking packed weights)." + "reason": "Both are CI/loading failures, but one is CPU offload device mismatch and the other is packed-weight unpacking in accelerate loading." }, { "left": "issue:45406", "right": "issue:45464", "accept": false, - "reason": "Serving crash from missing `_tokenizer` and streaming inference failure on Qwen3.5 are distinct runtime issues." + "reason": "Different crash surfaces: Gemma4Processor missing _tokenizer in serve vs chat/completions streaming failure on Qwen3.5-0.8B." }, { "left": "issue:43873", "right": "issue:45704", "accept": false, - "reason": "Quantization/offloading behavior and a T5 RMSNorm memory leak are not the same bug." + "reason": "Quantization/offloading behavior is broad, but these are unrelated specific bugs: offloading with quantization vs T5 using apex.FusedRMSNorm memory leak." }, { "left": "issue:41720", "right": "issue:43746", "accept": false, - "reason": "Auto device mapping CUDA assert vs PEFT local checkpoint loading; different failure surfaces." + "reason": "Different model families and causes: Qwen3 auto device mapping assert vs GraniteSpeech PEFT local checkpoint loading." }, { "left": "issue:44530", "right": "issue:46291", "accept": false, - "reason": "PagedAttention cache group-type crash and `evaluate` import/version requirement are unrelated." + "reason": "PagedAttentionCache linear_attention group error is unrelated to kernels import/version requirements in evaluate." }, { "left": "issue:43421", "right": "issue:45701", "accept": false, - "reason": "Runtime special-token updates and version-dependent tokenization are too broad and not the same concrete defect." + "reason": "Both involve tokenization behavior broadly, but one is runtime post-processor updates and the other is a version-dependent tokenization difference report." }, { "left": "issue:45357", "right": "issue:46402", "accept": false, - "reason": "Both involve Qwen/Sam3 multimodal weight handling, but one is incorrect save_pretrained keys and the other is missing text-encoder weights on load." + "reason": "Both concern Qwen3.5 checkpoints, but one is save_pretrained visual encoder keys and the other is loading text encoder weights; different operations and failures." }, { - "left": "issue:44492", - "right": "issue:44509", + "left": "issue:43606", + "right": "issue:43653", "accept": false, - "reason": "A typo in cache-strategy docs and obsolete pipeline-task documentation are unrelated documentation issues." + "reason": "No shared underlying bug: CPU offload device mismatch vs BigBirdTokenizer special-token registration." }, { - "left": "issue:43606", - "right": "issue:43653", + "left": "issue:44492", + "right": "issue:44509", "accept": false, - "reason": "Device-mismatch on CPU offload is not the same underlying bug as a tokenizer mask-token registration problem." + "reason": "A typo in cache strategies is unrelated to documentation still mentioning removed pipeline tasks." }, { "left": "issue:43531", "right": "issue:45759", "accept": false, - "reason": "Qwen3-MoE sliding-window behavior and composite config `text_config` unwrapping are separate model/config issues." + "reason": "Different Qwen MoE configuration issues: sliding_window support vs from_config not unwrapping text_config." }, { "left": "issue:34634", "right": "issue:41720", "accept": false, - "reason": "BarkProcessor `voice_preset` and Qwen3 auto device mapping are unrelated." + "reason": "Different symptoms and model families: BarkProcessor voice_preset vs Qwen3 auto device mapping assert." }, { "left": "issue:42490", "right": "issue:43329", "accept": false, - "reason": "Trailing-slash import warning and undefined video-branch variables in multimodal token counting are different bugs." + "reason": "Local-path trailing-slash import warning is unrelated to a video-branch undefined-variable bug in multimodal token counting." }, { "left": "issue:34567", "right": "issue:35707", "accept": false, - "reason": "TrainerState token counters and progressive generation with embeddings/past KV are unrelated." + "reason": "TrainerState token counting is unrelated to progressive generation with inputs_embeds/past_key_values." }, { "left": "issue:44514", "right": "issue:44545", "accept": true, - "reason": "Same processor, same method, same batched `padding=False` crash; this is a direct duplicate." + "reason": "These are the same underlying crash report: Qwen2_5_VLProcessor.apply_chat_template fails on batched input when padding=False." }, { "left": "issue:44625", "right": "issue:45479", "accept": false, - "reason": "Qwen3.5 config propagation and a general zero-loss classification issue are not the same defect." + "reason": "Different classification bugs: num_labels propagation in Qwen3.5 vs degenerate zero loss across sequence-classification models." }, { "left": "issue:44451", "right": "issue:44991", "accept": false, - "reason": "Different models and tokenizer load failures; not a duplicate." + "reason": "Both are tokenizer loading regressions, but they affect different checkpoints and likely different root causes." }, { "left": "issue:43531", "right": "issue:44589", "accept": false, - "reason": "Qwen3-MoE sliding-window behavior and missing `Float8_e4m3fnStorage` are unrelated." + "reason": "Different model/config problems: Qwen3-MoE sliding_window vs missing Float8 storage type during loading." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is conservative and grounded in the packet. The only accepted duplicate pair, issue:44514 / issue:44545, is strongly supported by the nearly identical titles and shared failure mode, and the rejected soft edges are explained cautiously without overclaiming. No clear overstatement stands out." + "feedback": "The summary is grounded in the packet and stays conservative. The only accepted soft edge, issue:44514 \u2194 issue:44545, is a clear duplicate by title and timing. The remaining pairs are reasonably rejected as distinct bugs based on their titles and scopes, without overstating shared root cause evidence." }, "evaluator_used": true, "retried": false, @@ -4854,223 +4792,226 @@ "issue:45356|issue:45907", "issue:30064|issue:36010" ], - "prepared_review_unit_hash": "a0e70ac03b9be15595bee2d41bc7c934deab096870a253bcc7bafafe56b011f6", + "prepared_review_unit_hash": "bf9844fa89c752951063787a1e740edb163382279cc2e30139c9bf4e35e187cf", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 37086, - "estimated_input_tokens": 9272, - "estimated_eval_tokens": 18800 + "serialized_chars": 37078, + "estimated_input_tokens": 9270, + "estimated_eval_tokens": 18796 }, "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 and feature requests. The soft pairs are mostly thematic lookalikes (tokenizers, Qwen, quantization, training), but they do not appear to be the same underlying defect or change.", - "confidence": 0.89, - "canonical_issue_reason": "No clear canonical issue: the items span unrelated subsystems and failure modes, so none is a strong duplicate representative for the whole cluster.", + "summary": "The cluster is mostly heterogeneous, with one strong duplicate pair around offline/Mistral-regex tokenizer loading (42886 \u2194 44843). A second Mistral-regex pair (45081 \u2194 45356) is related but too different in failure mode to treat as the same bug.", + "confidence": 0.84, + "canonical_issue_reason": "44843 is the best canonical issue for the only clearly duplicate subgroup: it names the concrete HF_HUB_OFFLINE regression in _patch_mistral_regex and matches 42886 closely.", "canonical_pr_reason": null, - "best_issue_reason": "No single best issue emerges because the cluster is heterogeneous and the soft-paired items are not true duplicates.", + "best_issue_reason": "44843 is the strongest standalone issue here because it captures the specific offline tokenizer regression and serves as the clearest representative of the duplicate subgroup.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44898", "right": "issue:45072", "accept": false, - "reason": "Both are vision-related, but one is Perceiver resolution handling and the other is a bfloat16 dtype mismatch in different models/code paths." + "reason": "Different models and failure modes: image-size/interpolation behavior vs dtype mismatch in bfloat16." }, { "left": "issue:44077", "right": "issue:44625", "accept": false, - "reason": "Unrelated problems: patchtsmixer post_init policy vs Qwen3.5 num_labels propagation." + "reason": "Unrelated config issues: optional post_init validation vs num_labels propagation." }, { "left": "issue:45759", "right": "issue:45859", "accept": false, - "reason": "Both involve Qwen configs, but one is composite config unwrapping and the other is CPU-offload incompatibility in a rotary embedding path." + "reason": "Different code paths: composite config unwrapping vs CPU-offload incompatibility in rotary embedding." }, { "left": "issue:43421", "right": "issue:43927", "accept": false, - "reason": "Token ID / special-token handling is broadly related, but one is runtime post-processor updates and the other is save/load config persistence." + "reason": "Both tokenizer-related, but one is runtime post-processor updates and the other is save/load losing custom token IDs." }, { "left": "issue:45588", "right": "issue:45759", "accept": false, - "reason": "Different code paths: flash-attention sink handling vs Qwen multimodal config unwrapping." + "reason": "Flash-attention sinkless-model crash is unrelated to Qwen config unwrapping." }, { "left": "issue:44062", "right": "issue:44451", "accept": false, - "reason": "AddedToken argument duplication is a tokenizer construction bug; loading ScandiBERT is a separate model-loading failure." + "reason": "AddedToken keyword conflict is not the same as failing to load a specific model." }, { "left": "issue:44368", "right": "issue:45357", "accept": false, - "reason": "One is a warning about tie_word_embeddings, the other is incorrect visual-encoder key serialization." + "reason": "Different regressions: LoRA warning about tie_word_embeddings vs incorrect visual encoder keys on save_pretrained." }, { "left": "issue:44987", "right": "issue:45042", "accept": false, - "reason": "Loading a specific model versus PIL image processors wrongly requiring torchvision; different failures." + "reason": "Loading a specific model failing is not the same bug as PIL image processors requiring torchvision." }, { "left": "issue:44918", "right": "issue:44936", "accept": false, - "reason": "Trainer/SFT embedding unpacking and evaluate-after-train are distinct trainer lifecycle bugs." + "reason": "Different trainer failures: embedding unpacking in SFT vs evaluate() after train()." }, { "left": "issue:43782", "right": "issue:45381", "accept": false, - "reason": "Both are Qwen vision issues, but one is weight_only loading and the other is incorrect video position IDs." + "reason": "Different multimodal bugs: weight_only loading error vs wrong video vision_position_ids." }, { "left": "issue:39692", "right": "issue:44610", "accept": false, - "reason": "Documentation example errors for SigLIP2 are unrelated to OmDet-Turbo processor resolution mismatch." + "reason": "Both mention processor/model mismatch, but they involve different models and different failures." }, { "left": "issue:43746", "right": "issue:46050", "accept": false, - "reason": "PEFT local checkpoint loading and quantized save/load corruption are separate persistence paths." + "reason": "Local checkpoint loading with PEFT adapters is not the same as quantized-model save/load corruption." }, { "left": "issue:42886", "right": "issue:44843", - "accept": false, - "reason": "Both mention HF_HUB_OFFLINE, but 44843 is specifically _patch_mistral_regex/model_info, not the general cache-load failure in 42886." + "accept": true, + "reason": "Same offline tokenizer regression: HF_HUB_OFFLINE breaks tokenizer loading via the unconditional model_info call in the Mistral regex path." }, { "left": "issue:45704", "right": "issue:45923", "accept": false, - "reason": "A T5 RMSNorm memory leak is unrelated to a missing supports_gradient_checkpointing flag." + "reason": "Apex RMSNorm memory leak and missing supports_gradient_checkpointing are unrelated bugs." }, { "left": "issue:43381", "right": "issue:43844", "accept": false, - "reason": "Gradient checkpointing in eval mode and gradient blow-up under ZeRO-3 are different problems." + "reason": "Eval-mode checkpointing restriction is a different issue from abnormal gradient growth under ZeRO-3." }, { "left": "issue:43421", "right": "issue:43653", "accept": false, - "reason": "Both touch special tokens, but one is runtime post-processing updates and the other is BigBirdTokenizer mask-token registration." + "reason": "Special-token post-processor updating and BigBird mask-token registration are related, but not the same underlying bug." }, { "left": "issue:45663", "right": "issue:45923", "accept": false, - "reason": "Gemma4 shared_kv_states rebuild bug is unrelated to a trust_remote_code flag omission." + "reason": "FSDP2 shared_kv_states KeyError is unrelated to the gradient-checkpointing capability flag." }, { "left": "issue:45704", "right": "issue:45859", "accept": false, - "reason": "Different failures: apex FusedRMSNorm leak vs Qwen3_5MoeTextRotaryEmbedding CPU-offload incompatibility." + "reason": "Apex RMSNorm memory leak and CPU-offload incompatibility in Qwen rotary embedding are unrelated." }, { "left": "issue:41720", "right": "issue:45704", "accept": false, - "reason": "Qwen3 auto device mapping CUDA assert and T5 RMSNorm memory leak are unrelated." + "reason": "Auto device mapping cudaErrorAssert and an RMSNorm memory leak are different bugs." }, { "left": "issue:43531", "right": "issue:43645", "accept": false, - "reason": "Sliding-window behavior in Qwen3-MoE is separate from Jupyter custom-model initialization in Transformers 5.0." + "reason": "Sliding-window behavior in Qwen3-MoE is unrelated to Jupyter custom-model initialization." }, { "left": "issue:43994", "right": "issue:44671", "accept": false, - "reason": "Both are bad outputs, but they affect different models and likely different preprocessing/model-selection bugs." + "reason": "Both are incorrect-output reports, but on different models and mechanisms." }, { "left": "issue:43421", "right": "issue:44186", "accept": false, - "reason": "Runtime special-token updates and LayoutLMv2 NER/padding crashes are not the same bug." + "reason": "Token post-processor updates and LayoutLMv2 NER/padding crashes are distinct tokenizer issues." }, { "left": "issue:43746", "right": "issue:43873", "accept": false, - "reason": "PEFT checkpoint loading and quantization/offloading behavior are related only at a high level, not the same code path." + "reason": "PEFT local checkpoint loading is unrelated to quantization/offloading behavior." }, { "left": "issue:45081", "right": "issue:45356", "accept": false, - "reason": "Both involve fix_mistral_regex, but one crashes on backend_tokenizer access while the other is a codec-handling regression in a different tokenizer." + "reason": "Both touch the Mistral-regex/tokenizer path, but the actual failures differ: backend_tokenizer AttributeError crash vs Kimi-K2.5 codec/warning regression." }, { "left": "issue:39692", "right": "issue:43873", "accept": false, - "reason": "SigLIP2 docs/example mistakes are unrelated to quantization offloading behavior." + "reason": "SigLIP2 doc/example problems are unrelated to quantization/offloading not working as expected." }, { "left": "issue:43540", "right": "issue:46294", "accept": false, - "reason": "Both are video-input issues in Qwen models, but they fail at different stages with different root causes." + "reason": "Both are video-input issues, but they affect different Qwen families and different failure points." }, { "left": "issue:43596", "right": "issue:44220", "accept": false, - "reason": "DeepSpeed zero3/BertModel indexing and audio fbank extraction are unrelated." + "reason": "Deepspeed zero3/BertModel indexing and _torch_extract_fbank_features() are unrelated." }, { "left": "issue:42915", "right": "issue:43746", "accept": false, - "reason": "FineGrainedFP8Config failure in Qwen3Moe is unrelated to GraniteSpeech local checkpoint loading." + "reason": "Qwen3Moe FP8 configuration failure is unrelated to GraniteSpeech PEFT checkpoint loading." }, { "left": "issue:45356", "right": "issue:45907", "accept": false, - "reason": "Tokenizer regression in Kimi-K2.5 is unrelated to list_repo_files missing revision forwarding." + "reason": "Kimi-K2.5 tokenizer regression is unrelated to list_repo_files not forwarding revision." }, { "left": "issue:30064", "right": "issue:36010", "accept": false, - "reason": "Void segmentation map processing and a GenerationMixin import error are completely different issues." + "reason": "Void segmentation map processing and a GenerationMixin import error are completely different." } ] }, - "evaluator_result": null, - "evaluator_used": false, - "retried": false, - "accepted_nontrivial_soft_edge": false, - "error_kind": null, - "error_message": null + "evaluator_result": { + "accept": false, + "feedback": "The clustering is mostly grounded, and 42886\u219444843 is a defensible duplicate link. But the accept on 45081\u219445356 is too strong for the packet: the titles show related Mistral/fix_mistral_regex code paths, yet one is a backend_tokenizer AttributeError crash and the other is a Kimi-K2.5 codec/warning regression. That is similar enough to group softly, but not enough evidence to call it a true duplicate with confidence. The overall summary should stay more conservative." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null }, { "label": "issue", @@ -5149,228 +5090,216 @@ "issue:43716|issue:44008", "issue:43653|issue:43819", "issue:43531|issue:43643", - "issue:43335|issue:43756", "issue:43388|issue:46032", + "issue:43335|issue:43756", "issue:43950|issue:45987", - "issue:44843|issue:45081", "issue:43643|issue:44877", + "issue:44843|issue:45081", "issue:30064|issue:36331", "issue:44448|issue:44671", "issue:41720|issue:44991", "issue:43540|issue:45381", "issue:43421|issue:44811", - "issue:43577|issue:43818", "issue:44821|issue:45084", + "issue:43577|issue:43818", "issue:39692|issue:45834" ], - "prepared_review_unit_hash": "0ae1db94a7e5ad43226dc81a019702be45f0502d1d5159c79c951129dfd3952f", + "prepared_review_unit_hash": "6cf40e4c047972f2bde3ad60fcac2c15679e27eb684011772e0d36491a59fe25", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 28, - "serialized_chars": 35954, - "estimated_input_tokens": 8989, - "estimated_eval_tokens": 18234 + "serialized_chars": 35960, + "estimated_input_tokens": 8990, + "estimated_eval_tokens": 18236 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is highly heterogeneous: most items are unrelated bugs spanning tokenizers, multimodal processors, model loading, caching, and model-specific forward paths. No soft edge looks like the same underlying duplicate, though a few are loosely related by theme.", - "confidence": 0.91, - "canonical_issue_reason": "issue:44843 is the closest representative of the only obvious subcluster: a concrete tokenizer-loading regression with a direct code-path failure and a nearby related Mistral regex issue. Even so, the overall cluster is too mixed to treat it as a strong canonical duplicate set.", + "summary": "This cluster is a grab-bag of unrelated bug reports spanning tokenizers, multimodal models, cache/dtype issues, and model-specific regressions. None of the soft pairs look like the same underlying bug or change, so I would not merge any of them into a duplicate cluster.", + "confidence": 0.98, + "canonical_issue_reason": "No single canonical issue fits this cluster: the items target different models, different code paths, and different failure modes.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44843 is the best single issue to anchor the cluster because it is a specific, reproducible regression in core loading logic and is the only item that clearly sits near another issue in the set. It is still not a safe umbrella for the rest of the cluster.", + "best_issue_reason": "No global best issue; the reports are heterogeneous and none is a clear umbrella for the rest.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43720", "right": "issue:43819", "accept": false, - "reason": "Different failures: accelerate loading of packed BitNet weights vs DAC latent reconstruction without STE." + "reason": "BitNet packed-weight loading during accelerate vs DAC latent/STE mismatch; different models and different code paths." }, { "left": "issue:43653", "right": "issue:43720", "accept": false, - "reason": "Tokenizer special-token registration bug and BitNet packed-weight loading bug are unrelated." + "reason": "BigBirdTokenizer special-token decode bug is unrelated to BitNet packed-weight unpacking in accelerate loading." }, { "left": "issue:44220", "right": "issue:44661", "accept": false, - "reason": "Audio feature extraction issue vs add-new-model-like failing inside TOKENIZER_MAPPING_NAMES; different code paths." + "reason": "Audio fbank feature extraction vs add-new-model-like tokenizer mapping; unrelated failures." }, { "left": "issue:43550", "right": "issue:45072", "accept": false, - "reason": "Different model/runtime problems: torch.compile+SDPA on Bamba vs bfloat16 dtype mismatches in SwitchTransformers/TimmWrapperModel." + "reason": "One is Bamba torch.compile with SDPA, the other is SwitchTransformers/TimmWrapperModel bfloat16 dtype mismatch; not the same bug." }, { "left": "issue:43540", "right": "issue:44008", "accept": false, - "reason": "Video-input processing failure in Qwen3OmniMoe vs Gemma3n variable-name collision; not the same bug." + "reason": "Qwen3OmniMoe video-input ValueError and Gemma3n audio_mel_mask name collision are different model-specific issues." }, { "left": "issue:43295", "right": "issue:45834", "accept": false, - "reason": "Processor.tokenizer regression and long OCR index error are separate failures in different models." + "reason": "processor.tokenizer regression in v4.57.5 is unrelated to Kosmos2.5 OCR index handling." }, { "left": "issue:43618", "right": "issue:43697", "accept": false, - "reason": "CLIP attentions serialization issue and RTDetrV2 output divergence are different model behaviors." + "reason": "Missing CLIPOutput attentions and RTDetrV2 output drift in v5 are different regressions." }, { "left": "issue:44062", "right": "issue:44488", "accept": false, - "reason": "AddedToken special-argument duplication vs loading a specific tokenizer model are unrelated." + "reason": "AddedToken special-arg duplication is unrelated to loading cjvt/sleng-bert." }, { "left": "issue:43550", "right": "issue:43819", "accept": false, - "reason": "Both are model bugs, but one is compile/SDPA and the other is DAC latent/forward mismatch; not mergeable as one fix." + "reason": "Bamba SDPA/torch.compile failure and DAC from_latents mismatch are unrelated." }, { "left": "issue:45183", "right": "issue:45479", "accept": false, - "reason": "Validation error messaging for input_ids/inputs_embeds is unrelated to single-label classification zero-loss behavior." + "reason": "Input validation XOR messaging is unrelated to single-label-classification zero-loss behavior." }, { "left": "issue:44186", "right": "issue:44898", "accept": false, - "reason": "LayoutLMv2 tokenizer padding/NER crash and Perceiver image-classification resolution issue are distinct." + "reason": "LayoutLMv2Tokenizer NER/padding crash and Perceiver resize/interpolate issue are different subsystems." }, { "left": "issue:43322", "right": "issue:45685", "accept": false, - "reason": "Llava Next loading crash and MPS histogram int support error are completely different subsystems." + "reason": "Llava Next loading segfault and MPS histogram int error do not share a concrete code-path." }, { "left": "issue:43716", "right": "issue:44008", "accept": false, - "reason": "Both touch multimodal inputs, but one is image-preprocessor dtype mismatch and the other is audio mask attribute collision." + "reason": "Mistral-3 image-preprocessor dtype mismatch and Gemma3n variable collision are separate bugs." }, { "left": "issue:43653", "right": "issue:43819", "accept": false, - "reason": "Special-token registration in BigBirdTokenizer is unrelated to DAC forward/from_latents mismatch." + "reason": "BigBirdTokenizer mask-token decode bug and DAC latent/forward mismatch are unrelated." }, { "left": "issue:43531", "right": "issue:43643", "accept": false, - "reason": "Sliding-window behavior in Qwen3-MoE and missing fields with trust_remote_code are unrelated." + "reason": "Qwen3-MoE sliding_window bug and trust_remote_code missing-fields regression are different issues." }, { - "left": "issue:43335", - "right": "issue:43756", + "left": "issue:43388", + "right": "issue:46032", "accept": false, - "reason": "SwitchTransformers sparse-layer creation and Smollm3 RoPE-layer dropping are different configuration bugs." + "reason": "gather_for_metrics tuple-label trimming and Mamba2Mixer use_cache corruption are not the same underlying defect." }, { - "left": "issue:43388", - "right": "issue:46032", + "left": "issue:43335", + "right": "issue:43756", "accept": false, - "reason": "Metrics label trimming bug and Mamba2 cache wrong-results bug are separate code paths." + "reason": "Sparse-layer creation with zero layers and Smollm3 RoPE-layer dropping are distinct model-configuration bugs." }, { "left": "issue:43950", "right": "issue:45987", "accept": false, - "reason": "Non-persistent buffer corruption on load is unrelated to StaticCache get_seq_length shape/type contract." + "reason": "Non-persistent buffer corruption on from_pretrained is unrelated to StaticCache.get_seq_length returning a tensor." }, { - "left": "issue:44843", - "right": "issue:45081", + "left": "issue:43643", + "right": "issue:44877", "accept": false, - "reason": "Same helper name, but different failures: offline model_info call vs backend_tokenizer AttributeError. Not the same underlying bug." + "reason": "Missing fields with trust_remote_code and strict config blocking granite_speech are separate config-loading problems." }, { - "left": "issue:43643", - "right": "issue:44877", + "left": "issue:44843", + "right": "issue:45081", "accept": false, - "reason": "Missing fields from trust_remote_code config and strict granite_speech config rejection are separate loading issues." + "reason": "Both touch Mistral regex patching, but one is offline model_info calls and the other is backend_tokenizer attribute access; different failure points and fixes." }, { "left": "issue:30064", "right": "issue:36331", "accept": false, - "reason": "Void segmentation maps in image processor and CustomTrainer compute_loss signature mismatch are unrelated." + "reason": "Void segmentation maps in Image Processor vs CustomTrainer.compute_loss kwarg change are unrelated." }, { "left": "issue:44448", "right": "issue:44671", "accept": false, - "reason": "Pegasus v4/v5 output differences and CamemBERT masked-LM prediction regressions are model-specific and distinct." + "reason": "Both mention v4/v5 output differences, but they concern different models and different prediction paths." }, { "left": "issue:41720", "right": "issue:44991", "accept": false, - "reason": "A800 cudaErrorAssert with auto device mapping and tokenizer loading failure for est-roberta are different problems." + "reason": "Qwen3 auto-device-map CUDA assert and est-roberta tokenizer loading failure are unrelated." }, { "left": "issue:43540", "right": "issue:45381", "accept": false, - "reason": "Both involve video inputs, but they affect different models and different failures (processing error vs vision_position_ids mismatch)." + "reason": "Qwen3OmniMoe video-input ValueError and Qwen2.5-VL vision_position_ids bug are model-specific and not the same fix." }, { "left": "issue:43421", "right": "issue:44811", "accept": false, - "reason": "Runtime special-token/post-processor update bug and Whisper batch_decode skip_special_tokens bug are not the same code path." - }, - { - "left": "issue:43577", - "right": "issue:43818", - "accept": false, - "reason": "Blip2 dtype loading mismatch and Video-LLaVA missing temporal attention/shared weights are unrelated." + "reason": "Runtime post-processor updates for special tokens and Whisper batch_decode skip_special_tokens behavior are different tokenizer features." }, { "left": "issue:44821", "right": "issue:45084", "accept": false, - "reason": "AutoImageProcessor URL loading and template-node compilation failure are unrelated." - }, - { - "left": "issue:39692", - "right": "issue:45834", - "accept": false, - "reason": "SigLIP2 doc-example errors and Kosmos2.5 long-OCR index error do not share the same underlying bug." + "reason": "AutoImageProcessor URL loading and template-node compilation errors are unrelated." }, { - "left": "issue:43653", - "right": "issue:43819", + "left": "issue:43577", + "right": "issue:43818", "accept": false, - "reason": "Tokenizer special-token registration and DAC latent/forward mismatch are unrelated." + "reason": "BLIP2 dtype propagation and Video-LLaVA temporal-attention/weight-sharing issues are different model families and bugs." }, { - "left": "issue:43531", - "right": "issue:43643", + "left": "issue:39692", + "right": "issue:45834", "accept": false, - "reason": "Duplicate of the earlier pair: sliding-window MoE behavior is not the same as trust_remote_code config loading." + "reason": "SigLIP2 doc example errors and Kosmos2.5 OCR index error are unrelated." } ] }, @@ -5450,20 +5379,20 @@ "issue:43606|issue:45072", "issue:44479|issue:45834", "issue:43825|issue:43994", - "issue:44805|issue:44936", "issue:34634|issue:36010", + "issue:44805|issue:44936", "issue:43577|issue:43582", "issue:44485|issue:44509", - "issue:44561|issue:45020", "issue:45987|issue:46192", "issue:28218|issue:29159", + "issue:44561|issue:45020", "issue:43720|issue:45072", "issue:44610|issue:45072", "issue:44671|issue:45701", "issue:43782|issue:45325", "issue:43825|issue:43827", - "issue:43906|issue:46209", "issue:43582|issue:44857", + "issue:43906|issue:46209", "issue:43749|issue:44291", "issue:45081|issue:45406", "issue:42175|issue:45081", @@ -5472,212 +5401,209 @@ "issue:34634|issue:36331", "issue:45003|issue:46285" ], - "prepared_review_unit_hash": "724dd2db3cc1034403828bc4e5e4ad17b0fdb1eb0fa923e567ab491ccb5865e9", + "prepared_review_unit_hash": "f7b90a0e6c326257821ca03dfbb61cd3df486e89da0fe5d01c7f4fa68fc29141", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 28, - "serialized_chars": 35678, - "estimated_input_tokens": 8920, - "estimated_eval_tokens": 18096 + "serialized_chars": 35685, + "estimated_input_tokens": 8922, + "estimated_eval_tokens": 18100 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is mostly heterogeneous, with only one plausible duplicate pair in the tokenizer behavior area. The strongest representative issue is the broad tokenization-regression report, while the rest are distinct bugs across unrelated subsystems.", - "confidence": 0.28, - "canonical_issue_reason": "issue:45701 is the best umbrella representative: it is broad, recent, open, and captures the recurring theme of version-driven tokenization changes better than the more specific reports.", + "summary": "The set is highly heterogeneous: most items are unrelated point bugs across tokenizers, model loading, training, vision/audio processors, and docs. All soft-similarity pairs look like false positives, so no merges are warranted.", + "confidence": 0.79, + "canonical_issue_reason": "issue:45020 is the broadest open-ended regression report in the set ('Recent transformers versions break models using `remote_code`'), so it is the closest thing to an umbrella issue; however, the cluster overall is mixed and it does not actually subsume the other reports.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45701 is the most cluster-like issue overall; it has the broadest scope and the highest likelihood of subsuming adjacent tokenizer-regression reports.", + "best_issue_reason": "issue:45020 is the best single representative because it is the most generic version-regression report and could attract related breakages; the other issues are narrower and largely unrelated.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43782", "right": "issue:44977", "accept": false, - "reason": "Same model family, but different failure modes: loading with weight_only vs generation under flash-attention." + "reason": "Different failures: Qwen3VL `weight_only=True` loading error vs Qwen3.5 flash-attention generation bug." }, { "left": "issue:45584", "right": "issue:45834", "accept": false, - "reason": "Whisper empty transcription failure and Kosmos2.5 long-OCR index error are unrelated code paths." + "reason": "Whisper empty-transcription bug after `align_special_tokens` is unrelated to Kosmos2.5 long-OCR index errors." }, { "left": "issue:44186", "right": "issue:44442", "accept": false, - "reason": "Tokenizer crash on NER padding/truncation is not the same as AutoTokenizer failing to load a tokenizer class." + "reason": "LayoutLMv2 tokenizer NER/padding crash and AutoTokenizer failing to load a different tokenizer are separate code paths." }, { "left": "issue:43644", "right": "issue:43749", "accept": false, - "reason": "Non-persistent buffer corruption and FSDP RAM-efficient loading are different bugs in different loading paths." + "reason": "Non-persistent buffer junk initialization is not the same as FSDP CPU RAM efficient loading being broken." }, { "left": "issue:43606", "right": "issue:45072", "accept": false, - "reason": "CPU-offload device mismatch and bfloat16 dtype mismatch are distinct failure modes." + "reason": "CPU-offload device mismatch for Bark is a different bug from dtype mismatches in SwitchTransformers/TimmWrapperModel." }, { "left": "issue:44479", "right": "issue:45834", "accept": false, - "reason": "Video-input regression for Qwen VL models is unrelated to Kosmos2.5 long OCR indexing." + "reason": "Qwen VL video-input regression and Kosmos2.5 OCR index error are unrelated model-specific failures." }, { "left": "issue:43825", "right": "issue:43994", "accept": false, - "reason": "One is an error-message wording issue; the other is a model-output correctness bug." + "reason": "Wrong pipeline error messaging about removed translation tasks is not the same as SigLIP2 producing bad outputs." }, { - "left": "issue:44805", - "right": "issue:44936", + "left": "issue:34634", + "right": "issue:36010", "accept": false, - "reason": "Mask-shape IndexError during indexing is not the same as evaluate() failing after train()." + "reason": "BarkProcessor `voice_preset` issue and `GenerationMixin` import failure are unrelated." }, { - "left": "issue:34634", - "right": "issue:36010", + "left": "issue:44805", + "right": "issue:44936", "accept": false, - "reason": "BarkProcessor voice_preset and GenerationMixin import failure are unrelated issues." + "reason": "Mask shape/indexing error is a different failure than `trainer.evaluate()` breaking after `train()`." }, { "left": "issue:43577", "right": "issue:43582", "accept": false, - "reason": "dtype propagation for BLIP2 and Apple Silicon warmup TypeError are separate bugs." + "reason": "BLIP2 dtype staying float32 and Apple Silicon warmup `TypeError` are unrelated problems." }, { "left": "issue:44485", "right": "issue:44509", "accept": false, - "reason": "GLM-5 RoPE discussion and docs mentioning removed pipeline tasks are not the same underlying defect." - }, - { - "left": "issue:44561", - "right": "issue:45020", - "accept": false, - "reason": "A specific helper removal breaking trust_remote_code is not the same as the broader remote_code regression report." + "reason": "GLM-5 RoPE discussion and docs still mentioning removed pipeline tasks are separate issues." }, { "left": "issue:45987", "right": "issue:46192", "accept": false, - "reason": "StaticCache seq-length contract breakage and return_tensors being ignored are unrelated API bugs." + "reason": "`StaticCache.get_seq_length()` returning a tensor and `return_tensors` being ignored are different API bugs." }, { "left": "issue:28218", "right": "issue:29159", - "accept": true, - "reason": "Both point to tokenizer behavior divergence around added tokens and slow/fast tokenizer inconsistency, plausibly the same underlying whitespace bug." + "accept": false, + "reason": "Extra space after added token and slow/fast tokenizer inconsistency are related only at a very high level, not the same concrete bug." + }, + { + "left": "issue:44561", + "right": "issue:45020", + "accept": false, + "reason": "Both mention remote-code breakage, but one is specifically `is_torch_fx_available` removal and the other is a broader version-regression report." }, { "left": "issue:43720", "right": "issue:45072", "accept": false, - "reason": "Packed-weight unpacking during accelerate loading and bfloat16 dtype mismatches are different problems." + "reason": "Packed-weight unpacking during accelerate loading is not the same as bf16 dtype mismatches." }, { "left": "issue:44610", "right": "issue:45072", "accept": false, - "reason": "Processor output size mismatch and dtype mismatch are unrelated failure modes." + "reason": "OmDet-Turbo processor producing 640px inputs is unrelated to SwitchTransformers/TimmWrapperModel dtype mismatch." }, { "left": "issue:44671", "right": "issue:45701", "accept": false, - "reason": "CamemBERT masked-LM prediction regression is not the same concrete bug as the generic tokenization-change report." + "reason": "CamemBERT masked-LM prediction regression and general tokenization changes are not the same underlying bug." }, { "left": "issue:43782", "right": "issue:45325", "accept": false, - "reason": "Qwen VL loading/weight_only error and RoPE temporal scaling are separate bugs." + "reason": "Qwen3VL `weight_only` load failure and Qwen2.5-VL RoPE scaling bug affect different models and code paths." }, { "left": "issue:43825", "right": "issue:43827", "accept": false, - "reason": "They share the v5 pipeline-removal theme, but one is runtime messaging and the other is documentation; not the same concrete issue." + "reason": "An incorrect runtime error message and stale docs references are separate artifacts, not one duplicate bug." }, { - "left": "issue:43906", - "right": "issue:46209", + "left": "issue:43582", + "right": "issue:44857", "accept": false, - "reason": "An isolated repro of an older issue and an FP8 inference failure on B200 are unrelated." + "reason": "Apple Silicon warmup `TypeError` and LwDetrImageLoss AMP/CUDA crash are unrelated." }, { - "left": "issue:43582", - "right": "issue:44857", + "left": "issue:43906", + "right": "issue:46209", "accept": false, - "reason": "Apple Silicon allocator warmup and CUDA AMP loss crash are different hardware-specific problems." + "reason": "An isolated reproduction of an earlier issue and an FP8 B200 inference failure do not describe the same bug." }, { "left": "issue:43749", "right": "issue:44291", "accept": false, - "reason": "Both involve loading, but FSDP CPU RAM-efficient loading and init_empty_weights argument handling are distinct code paths." + "reason": "FSDP CPU RAM loading failure and `init_empty_weights` unexpected `_is_hf_initialized` argument are distinct initialization/loading regressions." }, { "left": "issue:45081", "right": "issue:45406", "accept": false, - "reason": "Mistral regex patch crash and Gemma4Processor missing _tokenizer are different tokenizer/processor bugs." + "reason": "Mistral tokenizer regex patch crash and Gemma4Processor missing `_tokenizer` are different failures." }, { "left": "issue:42175", "right": "issue:45081", "accept": false, - "reason": "Backend installation extras and tokenizer regex patching are unrelated." + "reason": "Package install backend omission and Mistral tokenizer regex crash are unrelated." }, { "left": "issue:45372", "right": "issue:46326", "accept": false, - "reason": "Missing ReasoningEffort import and continue_final_message behavior are different mistral_common-related defects." + "reason": "`ReasoningEffort` import failure in Gemma 4 processor loading and `continue_final_message` behavior for mistral_common tokenizers are different bugs." }, { "left": "issue:43262", "right": "issue:43819", "accept": false, - "reason": "Sampling-rate defaulting and DAC latent/forward mismatch do not share the same bug." + "reason": "Audio chat-template sampling-rate defaulting and DAC `from_latents` mismatch are unrelated." }, { "left": "issue:34634", "right": "issue:36331", "accept": false, - "reason": "BarkProcessor voice_preset and compute_loss signature mismatch are unrelated." + "reason": "BarkProcessor voice preset and `CustomTrainer.compute_loss()` signature mismatch are unrelated." }, { "left": "issue:45003", "right": "issue:46285", "accept": false, - "reason": "Unsafe sys.modules access and CLIPTextModel structural inconsistency are different regressions." + "reason": "Unsafe `sys.modules` access in `modeling_utils` and CLIPTextModel structure inconsistency are different code-path regressions." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative: it identifies one plausible tokenizer-related duplicate pair and treats the rest as distinct, which matches the titles. The representative issue choice is also supported by the broad scope of issue:45701. No soft-edge verdict appears to overstate the evidence." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -5744,10 +5670,10 @@ "issue:46424" ], "soft_pairs": [ - "issue:43526|issue:43697", "issue:43653|issue:44912", - "issue:44060|issue:44368", + "issue:43526|issue:43697", "issue:43742|issue:45020", + "issue:44060|issue:44368", "issue:45084|issue:45538", "issue:43526|issue:44554", "issue:43650|issue:43906", @@ -5759,10 +5685,10 @@ "issue:43550|issue:44912", "issue:44387|issue:44617", "issue:43819|issue:43927", - "issue:46421|issue:46424", "issue:43257|issue:43329", - "issue:43854|issue:44977", + "issue:46421|issue:46424", "issue:36010|issue:43645", + "issue:43854|issue:44977", "issue:43720|issue:44610", "issue:38175|issue:44610", "issue:39692|issue:44912", @@ -5774,218 +5700,215 @@ "issue:43262|issue:44811", "issue:43257|issue:43696" ], - "prepared_review_unit_hash": "782bb380ad668c4c6d035e480ae83294f8ee870085a3970ece9311e96d2c4f24", + "prepared_review_unit_hash": "102ba651a4c4e03f09d06d01e4cbef642b08cfa43b40d1bef95b3f2780e90c84", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36612, - "estimated_input_tokens": 9153, - "estimated_eval_tokens": 18562 + "serialized_chars": 36606, + "estimated_input_tokens": 9152, + "estimated_eval_tokens": 18560 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous single-bug reports, not one broad duplicate set. The only strong duplicate-like pair is the Qwen tied-weights warning reports (44060/44368); the rest are too far apart in subsystem and failure mode to merge.", - "confidence": 0.72, - "canonical_issue_reason": "No single canonical issue cleanly covers the whole cluster; if forced, 44060 is the best anchor because it states the most concrete Qwen tied-weights warning bug and appears to match 44368.", + "summary": "This set does not form one duplicate cluster; it contains many unrelated issues across tokenizers, loading, quantization, cache, and multimodal processors. All soft-similarity pairs look like false positives and should be kept separate.", + "confidence": 0.97, + "canonical_issue_reason": "No single issue is a good canonical representative because the items are thematically unrelated rather than duplicates of one underlying bug.", "canonical_pr_reason": null, - "best_issue_reason": "44060 is the best representative issue because it is the clearest, most actionable description of the only likely duplicate set in the cluster.", + "best_issue_reason": "None of the issues is suitable as a global canonical issue for this set; the strongest matches still describe different code paths or different products/models.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:43653", + "right": "issue:44912", + "accept": false, + "reason": "Different bugs: BigBirdTokenizer special-token registration vs GPT-OSS/gguf quantization loading." + }, { "left": "issue:43526", "right": "issue:43697", "accept": false, - "reason": "Different models and failure modes: BEiT label reduction vs RT-DETR output drift." + "reason": "Different subsystems and symptoms: BEiT label reduction vs RTDetrV2 output drift." }, { - "left": "issue:43653", - "right": "issue:44912", + "left": "issue:43742", + "right": "issue:45020", "accept": false, - "reason": "Tokenizer special-token decode bug vs MXFP4 quantization fallback; unrelated code paths." + "reason": "One is a MobileLLM key error; the other is a broad remote_code regression affecting many models." }, { "left": "issue:44060", "right": "issue:44368", - "accept": true, - "reason": "Both describe the same Qwen tied-weights warning regression; one is the core warning bug and the other is the same warning surfacing during Qwen3.5 LoRA fine-tuning." - }, - { - "left": "issue:43742", - "right": "issue:45020", "accept": false, - "reason": "MobileLLM key error is too specific to be the same bug as a broad remote_code-loading regression." + "reason": "Both mention Qwen warnings, but one is a tied-weights bug and the other is a fine-tuning config warning; not the same defect." }, { "left": "issue:45084", "right": "issue:45538", "accept": false, - "reason": "Compilation error on template nodes is unrelated to CLIPTokenizer model_max_length." + "reason": "Unrelated: template compilation error vs CLIPTokenizer max length constant." }, { "left": "issue:43526", "right": "issue:44554", "accept": false, - "reason": "No shared underlying bug; image processor label reduction and MPS attention correctness are distinct." + "reason": "Unrelated vision preprocessing bug vs MPS attention correctness issue." }, { "left": "issue:43650", "right": "issue:43906", "accept": false, - "reason": "A vague request for data is not the same as an isolated reproduction of another issue." + "reason": "No meaningful relation; placeholder issue text vs a reproduction of a specific upstream bug." }, { "left": "issue:43329", "right": "issue:45600", "accept": false, - "reason": "Both touch multimodal/model-mapping areas, but they are different concrete bugs with different symptoms." + "reason": "Different areas: multimodal token counting vs auto_mappings/config cleanup." }, { "left": "issue:44589", "right": "issue:45042", "accept": false, - "reason": "Float8 storage lookup failure is unrelated to PIL backend torchvision requirements." + "reason": "Storage type import error is unrelated to PIL/image-processor torchvision dependency handling." }, { "left": "issue:43452", "right": "issue:44991", "accept": false, - "reason": "Both involve loading, but one is gguf/from_pretrained breakage and the other is tokenizer loading for a specific model." + "reason": "Both involve loading, but one is gguf_file support and the other is a tokenizer regression for a specific model; not the same code path." }, { "left": "issue:43746", "right": "issue:44811", "accept": false, - "reason": "Local PEFT checkpoint loading and Whisper batch_decode skip_special_tokens are unrelated." + "reason": "Different models and APIs: GraniteSpeech PEFT checkpoint loading vs Whisper batch_decode token handling." }, { "left": "issue:43638", "right": "issue:44936", "accept": false, - "reason": "Zero3/non-pretrained BERT indexing and trainer.evaluate after train are different trainer lifecycle bugs." + "reason": "Different training/evaluation failures with different root causes; not a duplicate." }, { "left": "issue:43550", "right": "issue:44912", "accept": false, - "reason": "torch.compile+SDPA on Bamba is not the same problem as MXFP4 quantization loading." + "reason": "Different model/runtime issues: torch.compile+SDPA vs gguf/MXFP4 quantized loading." }, { "left": "issue:44387", "right": "issue:44617", "accept": false, - "reason": "Both mention OOM, but one is int4 reserved-memory regression and the other is a model-specific memory exhaustion report." + "reason": "Both can OOM, but one is a Transformers 5.x int4 memory regression and the other is a model-specific runtime OOM." }, { "left": "issue:43819", "right": "issue:43927", "accept": false, - "reason": "DAC from_latents/STE mismatch and DiaConfig token ID persistence are unrelated." - }, - { - "left": "issue:46421", - "right": "issue:46424", - "accept": false, - "reason": "Both are cache-related, but one is recompilation from lazy init and the other is max_cache_length semantics; not the same bug." + "reason": "Different failures: DAC latent reconstruction mismatch vs DiaConfig save/load token ID loss." }, { "left": "issue:43257", "right": "issue:43329", "accept": false, - "reason": "Qwen3 MoE weight conversion and sliding_window behavior are different Qwen3 issues." + "reason": "Completely different areas: MoE weight conversion/loading vs multimodal token counting video path." }, { - "left": "issue:43854", - "right": "issue:44977", + "left": "issue:46421", + "right": "issue:46424", "accept": false, - "reason": "GLM-4.7-Flash load failure and Qwen3.5 flash-attention generation problems are unrelated." + "reason": "Related cache area, but one is lazy init causing recompilation and the other is max_cache_length handling for quantized cache; not the same bug." }, { "left": "issue:36010", "right": "issue:43645", "accept": false, - "reason": "Importing GenerationMixin and Jupyter custom-model initialization are different regressions." + "reason": "Import error for GenerationMixin is unrelated to Jupyter custom-model initialization in Transformers 5.0.0." + }, + { + "left": "issue:43854", + "right": "issue:44977", + "accept": false, + "reason": "Different model-specific generation/load problems; no shared concrete code-path defect." }, { "left": "issue:43720", "right": "issue:44610", "accept": false, - "reason": "Packed-weight loading in BitNet and OmDet-Turbo processor size mismatch are unrelated." + "reason": "Unrelated: packed-weight unloading in accelerate vs image processor input-size mismatch." }, { "left": "issue:38175", "right": "issue:44610", "accept": false, - "reason": "SigLIP2 zero probabilities and OmDet-Turbo input-size mismatch are unrelated." + "reason": "Different model families and failure modes; zero probabilities vs wrong processor size." }, { "left": "issue:39692", "right": "issue:44912", "accept": false, - "reason": "Documentation example errors and MXFP4 quantization loading are not the same underlying issue." + "reason": "Documentation/example errors are unrelated to the gguf/MXFP4 loading regression." }, { "left": "issue:36010", "right": "issue:36331", "accept": false, - "reason": "ImportError for GenerationMixin vs unexpected compute_loss keyword are unrelated training/loading bugs." + "reason": "Different APIs and failures: import path breakage vs Trainer custom-loss signature change." }, { "left": "issue:43531", "right": "issue:45042", "accept": false, - "reason": "Qwen3-MoE sliding_window and PIL backend torchvision requirements are different code paths." + "reason": "Qwen3-MoE sliding window behavior is unrelated to PIL image-processor torchvision requirements." }, { "left": "issue:45459", "right": "issue:45803", "accept": false, - "reason": "Both involve exception handling, but they affect different components and different failure modes." + "reason": "Both are exception-handling issues, but they affect different code paths and different error types; not the same bug." }, { "left": "issue:43454", "right": "issue:43577", "accept": false, - "reason": "AyaVision weight tying and Blip2 dtype handling are separate model-loading defects." + "reason": "Different model-loading/data-type problems: lm_head tying vs dtype propagation to qformer." }, { "left": "issue:42490", "right": "issue:43257", "accept": false, - "reason": "Trailing-slash path warning and Qwen3 MoE weight conversion are unrelated." + "reason": "Trailing-slash path warning is unrelated to Qwen3 MOE weight conversion with accelerate+deepspeed." }, { "left": "issue:43262", "right": "issue:44811", "accept": false, - "reason": "Audio apply_chat_template sample-rate default and Whisper batch_decode behavior are not the same bug." + "reason": "Different processor APIs and symptoms: audio chat-template sample-rate default vs Whisper decode token filtering." }, { "left": "issue:43257", "right": "issue:43696", "accept": false, - "reason": "Qwen3 MoE conversion and GPT-oss OOM are unrelated." + "reason": "MoE weight conversion/loading is unrelated to GPT-oss CUDA OOM." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The clustering is grounded in the packet and remains conservative. The single accepted soft edge (44060/44368) is plausible from the titles alone, and the rest are appropriately rejected as cross-subsystem or cross-failure-mode pairs. The summary does not overclaim a broader duplicate set, and the canonical/best-issue choices are presented cautiously rather than as certainty." - }, - "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 }, @@ -6058,9 +5981,9 @@ "issue:43606|issue:45859", "issue:34567|issue:40990", "issue:43749|issue:44756", - "issue:43257|issue:43931", "issue:44568|issue:45356", "issue:29159|issue:34567", + "issue:43257|issue:43931", "issue:42915|issue:43742", "issue:44291|issue:45042", "issue:42915|issue:44821", @@ -6073,8 +5996,8 @@ "issue:44568|issue:45488", "issue:41720|issue:43540", "issue:33357|issue:34567", - "issue:44479|issue:45381", "issue:42175|issue:43824", + "issue:44479|issue:45381", "issue:43317|issue:45538", "issue:43540|issue:44479", "issue:44062|issue:44991", @@ -6085,20 +6008,20 @@ "issue:43742|issue:44291", "issue:44987|issue:45907" ], - "prepared_review_unit_hash": "0a5399bd818fa84c0ad5cca7a31f1a8efaa98e203c491883f1ab81843f89335c", + "prepared_review_unit_hash": "fa24daff671f1bba2faeecf50b14c9b9edb5f34004add87710825838738a0d4c", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 32, - "serialized_chars": 37325, + "serialized_chars": 37328, "estimated_input_tokens": 9332, "estimated_eval_tokens": 18920 }, @@ -6107,214 +6030,211 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Most items are unrelated standalone issues; there is one likely duplicate cluster around the Qwen v5.3.0 video-input regression. No PRs are present.", - "confidence": 0.68, - "canonical_issue_reason": "issue:44479 is the broadest and most representative report of the Qwen video-input regression, naming the release and multiple affected model families.", + "summary": "This cluster is heterogeneous: it mixes unrelated tokenizer, video/VLM, loading, offload, and training-state bugs. I found no soft pair that clearly represents the same underlying fix, so all pairs should be kept separate.", + "confidence": 0.24, + "canonical_issue_reason": "Issue 45701 is the broadest, most umbrella-like report in the set and has inbound references, but there is no truly strong canonical duplicate cluster here because the topics are too mixed.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44479 is the best standalone issue to anchor the cluster because it captures the shared regression surface more generally than the model-specific reports.", + "best_issue_reason": "Issue 45701 is the closest thing to a representative issue for the cluster because it is the most general regression report and could absorb adjacent tokenization/version-related complaints, though it does not cover the whole set.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43643", "right": "issue:45759", "accept": false, - "reason": "Different bugs: missing fields with trust_remote_code vs composite Qwen text_config unwrapping." + "reason": "Both are config/model-construction complaints, but one is AutoConfig trust_remote_code metadata loss and the other is AutoModelForCausalLM.from_config not unwrapping text_config. Different code paths and failure modes." }, { "left": "issue:43329", "right": "issue:44479", - "accept": true, - "reason": "Same underlying video-input regression in the Qwen VL path; 43329 appears to pinpoint the root helper bug behind the broader 5.3.0 regression." + "accept": false, + "reason": "Both involve video/VLM regressions, but 43329 is an undefined-name bug in a helper path while 44479 is a broader v5.3.0 video-input regression across several Qwen variants. Not the same concrete bug." }, { "left": "issue:43329", "right": "issue:43450", "accept": false, - "reason": "One is an undefined-variable bug in token counting; the other is a batched shape issue in video processors." + "reason": "One is token counting with undefined video-branch symbols; the other is batched video processor shape handling. Related subsystem, but different underlying defects." }, { "left": "issue:43606", "right": "issue:45859", "accept": false, - "reason": "Both mention CPU offload, but they affect different models and code paths." + "reason": "Both mention CPU offload, but they hit different models and different code paths: Bark device mismatch versus Qwen3.5 MoE rotary embedding compatibility." }, { "left": "issue:34567", "right": "issue:40990", "accept": false, - "reason": "Trainer token-count tracking is unrelated to perplexity on gpt-oss-20b." + "reason": "TrainerState token accounting is unrelated to perplexity on gpt-oss-20b." }, { "left": "issue:43749", "right": "issue:44756", "accept": false, - "reason": "Both concern loading/memory, but one is FSDP RAM-efficient loading and the other is mmap OOM on Strix Halo." - }, - { - "left": "issue:43257", - "right": "issue:43931", - "accept": false, - "reason": "Both are Qwen loading problems, but one is accelerate/deepspeed weight conversion and the other is a Qwen3-VL shape mismatch." + "reason": "Both are memory/loading related, but FSDP CPU RAM-efficient loading and Strix Halo mmap OOM are distinct problems." }, { "left": "issue:44568", "right": "issue:45356", "accept": false, - "reason": "Distinct tokenizer regressions affecting different models and different behaviors." + "reason": "Both are tokenizer regressions, but one is add_special_tokens not adding BOS/EOS and the other is codec/warning behavior for Kimi-K2.5. Different bugs." }, { "left": "issue:29159", "right": "issue:34567", "accept": false, - "reason": "Tokenizer slow/fast inconsistency is unrelated to TrainerState token counting." + "reason": "Tokenizer slow/fast inconsistency is unrelated to TrainerState num_input_tokens_seen not updating." + }, + { + "left": "issue:43257", + "right": "issue:43931", + "accept": false, + "reason": "Both are Qwen3 loading issues, but one is deepspeed/accelerate MoE weight conversion and the other is a Qwen3-VL shape mismatch. Not the same fix." }, { "left": "issue:42915", "right": "issue:43742", "accept": false, - "reason": "FP8 fine-grained config failure is unrelated to MobileLLM key errors." + "reason": "Qwen3Moe FP8 failure and MobileLLM key error are unrelated loading failures." }, { "left": "issue:44291", "right": "issue:45042", "accept": false, - "reason": "Init-empty-weights argument handling is unrelated to PIL/torchvision backend requirements." + "reason": "init_empty_weights/_is_hf_initialized and PIL backend torchvision dependency are different code paths with no shared concrete bug." }, { "left": "issue:42915", "right": "issue:44821", "accept": false, - "reason": "Different areas: Qwen3Moe FP8 vs AutoImageProcessor URL loading." + "reason": "FP8 Qwen3Moe loading and AutoImageProcessor-from-URL are unrelated." }, { "left": "issue:44987", "right": "issue:45538", "accept": false, - "reason": "Model-loading failure for one repo is not the same as CLIPTokenizer max-length bug." + "reason": "Loading physical-intelligence/fast on v5.1.0 and CLIPTokenizer model_max_length are separate issues." }, { "left": "issue:43452", "right": "issue:43582", "accept": false, - "reason": "gguf loading issue is unrelated to AppleSilicon allocator warmup TypeError." + "reason": "gguf_file tokenizer/model loading breakage and AppleSilicon caching_allocator_warmup TypeError are unrelated." }, { "left": "issue:43697", "right": "issue:45357", "accept": false, - "reason": "RTDetrV2 output drift and Qwen3.5 save_pretrained key regression are different bugs." + "reason": "RTDetrV2 output changes and Qwen3.5 save_pretrained visual-key regression are different model families and different failure modes." }, { "left": "issue:43927", "right": "issue:44912", "accept": false, - "reason": "Config save/load token IDs issue is unrelated to MXFP4 quantization fallback." + "reason": "DiaConfig save/load token ID loss is unrelated to MXFP4 quantization fallback on git-oss-20b." }, { "left": "issue:42915", "right": "issue:45538", "accept": false, - "reason": "Qwen3Moe FP8 and CLIPTokenizer model_max_length are unrelated." + "reason": "Qwen3Moe FP8 loading and CLIPTokenizer max length are not the same bug." }, { "left": "issue:44849", "right": "issue:45701", "accept": false, - "reason": "Qwen3.5 hidden-states behavior and generic tokenization changes are not the same bug." + "reason": "Qwen3.5 output_hidden_states regression is a model-forward bug, while 45701 is a generic tokenization/version-change report." }, { "left": "issue:44568", "right": "issue:45488", "accept": false, - "reason": "Different tokenizer regressions: special-token insertion vs LlamaTokenizer pre-tokenizer override." + "reason": "Both are tokenizer regressions, but one concerns special-token insertion and the other concerns overriding tokenizer.json pre-tokenizer settings. Different code paths." }, { "left": "issue:41720", "right": "issue:43540", "accept": false, - "reason": "Different symptoms and models: CUDA assert on A800 vs video ValueError in Qwen3OmniMoe." + "reason": "cudaErrorAssert on auto device mapping for Qwen3 and video-processing ValueError in Qwen3OmniMoe are distinct issues." }, { "left": "issue:33357", "right": "issue:34567", "accept": false, - "reason": "MacOS CLIP bus error is unrelated to TrainerState token accounting." - }, - { - "left": "issue:44479", - "right": "issue:45381", - "accept": true, - "reason": "Same Qwen video-input regression family; 45381 is a model-specific report within the broader 44479 regression." + "reason": "MacOS bus error with community CLIP and TrainerState token tracking are unrelated." }, { "left": "issue:42175", "right": "issue:43824", "accept": false, - "reason": "TensorFlow backend packaging issue is unrelated to Qwen2.5-VL import failure." + "reason": "TensorFlow backend packaging and missing Qwen2.5-VL import are different concerns." + }, + { + "left": "issue:44479", + "right": "issue:45381", + "accept": false, + "reason": "Both are video-input regressions, but 45381 is specifically wrong vision_position_ids in qwen2.5-vl, while 44479 is a broader multi-model regression. Too little evidence for one bug." }, { "left": "issue:43317", "right": "issue:45538", "accept": false, - "reason": "Device-map dequantized-model offload failure is unrelated to CLIPTokenizer max-length behavior." + "reason": "Dequantized model loading with device_map=auto/offload and CLIPTokenizer model_max_length are unrelated." }, { "left": "issue:43540", "right": "issue:44479", "accept": false, - "reason": "Both are video-related, but Qwen3OmniMoe ValueError is not clearly the same bug as the multi-model 5.3.0 regression." + "reason": "Qwen3OmniMoe video ValueError and the broader v5.3.0 video-input regression are not clearly the same underlying defect." }, { "left": "issue:44062", "right": "issue:44991", "accept": false, - "reason": "AddedToken special-argument conflict is unrelated to tokenizer loading for EMBEDDIA/est-roberta." + "reason": "AddedToken special-argument conflict and tokenizer loading for EMBEDDIA/est-roberta are different tokenizer bugs." }, { "left": "issue:43540", "right": "issue:44560", "accept": false, - "reason": "Different video-input failures in different Qwen models; not enough evidence they share one bug." + "reason": "Both are Qwen video-related, but one is a processing ValueError and the other is a StopIteration in qwen3-vl embedding. Different concrete failures." }, { "left": "issue:28282", "right": "issue:34689", "accept": false, - "reason": "PyTorch-missing import error is unrelated to Llama 3.2 model loading breakage." + "reason": "Missing PyTorch import and Llama 3.2 Vision model loading breakage are unrelated load-time errors." }, { "left": "issue:45137", "right": "issue:45405", "accept": false, - "reason": "DeepSpeed ZeRO3 deque error is unrelated to the PEFT version bump issue." + "reason": "DeepSpeed ZeRO3 deque underflow and a PEFT version bump are not the same issue." }, { "left": "issue:24643", "right": "issue:29159", "accept": false, - "reason": "DeepSpeed training weight-shape error is unrelated to tokenizer slow/fast inconsistency." + "reason": "DeepSpeed weight dimensionality error and tokenizer inconsistency are unrelated." }, { "left": "issue:43742", "right": "issue:44291", "accept": false, - "reason": "Both are loading-related, but one is a key error for a model and the other is an init_empty_weights argument regression." + "reason": "Key error loading MobileLLM and init_empty_weights unexpected argument are separate model-loading failures." }, { "left": "issue:44987", "right": "issue:45907", "accept": false, - "reason": "Repo loading failure and list_repo_files revision forwarding are different Hugging Face Hub issues." + "reason": "Loading physical-intelligence/fast and list_repo_files not forwarding revision are unrelated regressions." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "Grounded overall. The summary stays conservative, correctly centers issue:44479 as the broad Qwen video-input regression, and the accepted soft edges are plausible from the titles alone. No PRs are present in the packet." - }, - "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 }, @@ -6390,17 +6310,17 @@ "issue:43582|issue:43782", "issue:44737|issue:45588", "issue:43493|issue:43994", - "issue:34567|issue:36331", "issue:44291|issue:44821", + "issue:34567|issue:36331", "issue:43404|issue:45987", "issue:43526|issue:45479", - "issue:43756|issue:46399", "issue:44062|issue:45081", "issue:43906|issue:45381", "issue:27957|issue:28282", - "issue:43756|issue:45325", + "issue:43756|issue:46399", "issue:44913|issue:46121", "issue:44792|issue:44964", + "issue:43756|issue:45325", "issue:45042|issue:45081", "issue:43452|issue:44488", "issue:43742|issue:44589", @@ -6409,20 +6329,20 @@ "issue:45405|issue:45600", "issue:43643|issue:45375" ], - "prepared_review_unit_hash": "c09a82e0820529e2c6e04409f8c52e95ecb0b4edef0a0dc6cb8c839c1ec42536", + "prepared_review_unit_hash": "618bee7731f22f39421c60b7c68da478310cbf9bc4bfed4afd0506c08d6b7471", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 27, - "serialized_chars": 36185, + "serialized_chars": 36188, "estimated_input_tokens": 9047, "estimated_eval_tokens": 18350 }, @@ -6431,174 +6351,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous. The only strong duplicate candidate is the SigLIP2 pair, where one issue describes the underlying HF-vs-JAX implementation mismatch and the other reports the same bad outputs through AutoModel/pipeline. No PRs are present.", - "confidence": 0.68, - "canonical_issue_reason": "issue:43493 is the best canonical issue because it points to the underlying SigLIP2 implementation discrepancy; issue:43994 appears to be the same bug surfacing as nonsensical inference output.", + "summary": "The cluster is mostly heterogeneous: the issues span unrelated models, loaders, dtype paths, tokenizer bugs, and config serialization problems. Only the SigLIP2 pair (43493/43994) looks like a plausible duplicate; the rest are distinct reports and should not be merged.", + "confidence": 0.95, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "issue:43493 is the strongest representative: it is root-cause oriented, specific to the affected model, and can subsume the user-facing symptoms in issue:43994.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43577", "right": "issue:45072", "accept": false, - "reason": "Both mention dtype mismatches, but they affect different model families and code paths (Blip2/QFormer vs SwitchTransformers/TimmWrapperModel)." + "reason": "Both mention dtype mismatches, but they affect different models and code paths (BLIP2 load vs SwitchTransformers/TimmWrapper inference)." }, { "left": "issue:28282", "right": "issue:29127", "accept": false, - "reason": "Unrelated problems: one is a missing-PyTorch import error, the other is a layoutlmv3 validation/error-message clarity issue." + "reason": "One is an ImportError about missing PyTorch; the other is a layoutlmv3 error-message clarity issue. No shared bug." }, { "left": "issue:42915", "right": "issue:43749", "accept": false, - "reason": "Different failures in different loading/training mechanisms; both involve advanced memory/loading settings but not the same defect." + "reason": "Qwen3Moe FP8 failure and FSDP CPU RAM efficient loading are different mechanisms and different failure modes." }, { "left": "issue:38175", "right": "issue:44898", "accept": false, - "reason": "Same broad model area (vision), but the symptoms and mechanics differ: zero probabilities vs non-default resolution/interpolation failure." + "reason": "Both are image-model correctness reports, but SigLIP2 zero probabilities and Perceiver non-default-resolution failure are distinct bugs." }, { "left": "issue:42175", "right": "issue:44488", "accept": false, - "reason": "One is about backend dependency packaging, the other about loading a specific model; no shared concrete bug." + "reason": "Packaging/backend dependency coverage is unrelated to a model-loading failure for cjvt/sleng-bert." }, { "left": "issue:43257", "right": "issue:43756", "accept": false, - "reason": "Both are Qwen/SmolLM-related, but one is a MoE weight conversion/loading issue and the other is a RoPE layer-count/configuration bug." + "reason": "Different Qwen3/Smollm3 bugs: weight conversion under accelerate+deepspeed vs RoPE layer dropping." }, { "left": "issue:43582", "right": "issue:43782", "accept": false, - "reason": "Different subsystems: Apple Silicon allocator warmup TypeError vs Qwen3VL weight_only loading failure." + "reason": "AppleSilicon allocator TypeError and Qwen3VL weight_only loading error are unrelated." }, { "left": "issue:44737", "right": "issue:45588", "accept": false, - "reason": "One is a CPU-device placement bug in XLNet; the other is a flash-attention sink/s_aux crash. Not the same code path." + "reason": "XLNet CPU-device placement bug and flash_attention sink-less crash are different code paths." }, { "left": "issue:43493", "right": "issue:43994", "accept": true, - "reason": "Both are about SigLIP2 producing incorrect outputs; 43493 frames the implementation discrepancy, while 43994 reports the same underlying bug through AutoModel/pipeline usage." + "reason": "Both describe the same SigLIP2 model producing incorrect outputs; 43994 looks like a user-facing manifestation of the implementation discrepancy in 43493." }, { - "left": "issue:34567", - "right": "issue:36331", + "left": "issue:44291", + "right": "issue:44821", "accept": false, - "reason": "Both concern trainer behavior, but one is a state counter not updating and the other is an unexpected argument in custom compute_loss." + "reason": "Loading with init_empty_weights and AutoImageProcessor-from-URL failure are unrelated." }, { - "left": "issue:44291", - "right": "issue:44821", + "left": "issue:34567", + "right": "issue:36331", "accept": false, - "reason": "Different loading failures: unexpected `_is_hf_initialized` argument versus inability to load AutoImageProcessor from URL." + "reason": "TrainerState token counting and CustomTrainer.compute_loss signature mismatch are different trainer bugs." }, { "left": "issue:43404", "right": "issue:45987", "accept": false, - "reason": "A tied-weight bug in Mistral3 is unrelated to StaticCache returning a tensor instead of an int." + "reason": "Mistral3 lm_head tying and StaticCache.get_seq_length return type are unrelated." }, { "left": "issue:43526", "right": "issue:45479", "accept": false, - "reason": "Both are classification/label-related, but one is an image processor label-reduction bug and the other is a degenerate loss issue for single-label classification." - }, - { - "left": "issue:43756", - "right": "issue:46399", - "accept": false, - "reason": "Different rotary-related areas: one is a model architecture/config mismatch, the other is a kernel registration crash." + "reason": "BeitImageProcessorFast label reduction bug and sequence-classification zero-loss bug are different subsystems and symptoms." }, { "left": "issue:44062", "right": "issue:45081", "accept": false, - "reason": "Tokenizer construction failure vs Mistral regex patching backend attribute crash; they are separate bugs." + "reason": "Tokenizer AddedToken argument duplication and Mistral regex patch backend_tokenizer crash are unrelated tokenizer bugs." }, { "left": "issue:43906", "right": "issue:45381", "accept": false, - "reason": "One is just a reproduction of an older issue; the other is a specific Qwen2.5-VL position-id regression. Insufficient evidence they are the same bug." + "reason": "An isolated reproduction of an earlier issue and a Qwen2.5-VL video position-id regression are not the same bug." }, { "left": "issue:27957", "right": "issue:28282", "accept": false, - "reason": "XLMRoberta Flash Attention support and a missing-PyTorch ImportError are unrelated." + "reason": "XLMRoberta Flash Attention 2 support and missing PyTorch environment detection are unrelated." }, { "left": "issue:43756", - "right": "issue:45325", + "right": "issue:46399", "accept": false, - "reason": "Both touch positional/rope logic in multimodal models, but they target different models and failure modes." + "reason": "Both involve RoPE-adjacent models, but Smollm3 RoPE-layer loss and Qwen3.5 kernel registration failure are different code paths." }, { "left": "issue:44913", "right": "issue:46121", "accept": false, - "reason": "Different config serialization issues: rotary_pct reload regression vs list-handling TypeError in rope validation." + "reason": "GPTNeoX rotary_pct reload regression and JSON-list handling in rope validation are separate config bugs." }, { "left": "issue:44792", "right": "issue:44964", "accept": false, - "reason": "A failed janus test and a Phi-4 multimodal load failure are distinct symptoms on different models." + "reason": "Janus image-generation test failure and Phi-4 multimodal load failure are different model-specific problems." + }, + { + "left": "issue:43756", + "right": "issue:45325", + "accept": false, + "reason": "Smollm3 RoPE-layer dropping and Qwen2.5-VL still-image temporal position scaling are unrelated." }, { "left": "issue:45042", "right": "issue:45081", "accept": false, - "reason": "Dependency requirement regression in image processors vs a tokenizer regex patch crash; not the same defect." + "reason": "A PIL-backend torchvision dependency regression is unrelated to the Mistral tokenizer regex crash." }, { "left": "issue:43452", "right": "issue:44488", "accept": false, - "reason": "GGUF loading breaks in tokenizers/CausalLM paths, while cjvt/sleng-bert is a separate model-loading issue." + "reason": "gguf_file loading breakage and inability to load cjvt/sleng-bert are both loading issues, but not the same concrete failure." }, { "left": "issue:43742", "right": "issue:44589", "accept": false, - "reason": "KeyError loading MobileLLM-125M is unrelated to the Float8 storage-object TypeError." + "reason": "MobileLLM-125M key error and missing Float8 storage type are unrelated load/runtime issues." }, { "left": "issue:45357", "right": "issue:45920", "accept": false, - "reason": "One is a save_pretrained visual-key regression for Qwen3.5; the other is a tokenizer ID regression across many models." + "reason": "Qwen3.5 save_pretrained visual-key regression and AutoTokenizer wrong token IDs across multiple models are different regressions." }, { "left": "issue:43531", "right": "issue:44821", "accept": false, - "reason": "Qwen3-MoE sliding-window behavior is unrelated to AutoImageProcessor URL loading." + "reason": "Qwen3-MoE sliding_window behavior and AutoImageProcessor URL loading are unrelated." }, { "left": "issue:45405", "right": "issue:45600", "accept": false, - "reason": "One is a version pin bump for PEFT; the other is a broken auto_mappings reference to removed configs." + "reason": "A PEFT version bump problem and auto_mappings referencing removed configs are separate CI/package issues." + }, + { + "left": "issue:43643", + "right": "issue:45375", + "accept": false, + "reason": "Generic trust_remote_code config field loss and Qwen3_5MoeVisionConfig missing deepstack_visual_indexes are both config issues, but not the same bug." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is well grounded in the packet: it stays conservative about the cluster being mostly heterogeneous and only treats issue:43493/issue:43994 as the plausible duplicate pair. The soft-edge rejections are cautious and consistent with the titles/body signals, and no PR-pair mergeability concerns apply because there are no PRs." + "feedback": "The analysis is grounded in the packet and the conclusions are conservative overall. The one accepted pair, 43493/43994, is a plausible duplicate about SigLIP2 incorrect outputs, while the rest are reasonably rejected as different models, code paths, or bug classes. No soft-edge verdict looks overstated based on the provided titles and evidence." }, "evaluator_used": true, "retried": false, @@ -6671,8 +6597,8 @@ "soft_pairs": [ "issue:44625|issue:44877", "issue:43720|issue:44898", - "issue:43697|issue:45216", "issue:24643|issue:29127", + "issue:43697|issue:45216", "issue:41762|issue:44936", "issue:44514|issue:45381", "issue:44493|issue:46050", @@ -6686,8 +6612,8 @@ "issue:45372|issue:46170", "issue:28282|issue:32090", "issue:44936|issue:45823", - "issue:43295|issue:45356", "issue:28282|issue:33357", + "issue:43295|issue:45356", "issue:44361|issue:44442", "issue:43653|issue:44361", "issue:44484|issue:44704", @@ -6695,212 +6621,218 @@ "issue:45542|issue:46178", "issue:24643|issue:27957", "issue:43653|issue:44568", - "issue:43577|issue:43697", "issue:44514|issue:45290", + "issue:43577|issue:43697", "issue:44464|issue:44743" ], - "prepared_review_unit_hash": "255ddaee4f7c666691da35767c5354b42d633564bcf41a37e8adc4b6d66bb15a", + "prepared_review_unit_hash": "ddb3e76b0adf2c727e6872060f515c70de4e711429a863445b2412e82387fa25", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36684, - "estimated_input_tokens": 9171, - "estimated_eval_tokens": 18598 + "serialized_chars": 36685, + "estimated_input_tokens": 9172, + "estimated_eval_tokens": 18600 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a mixed cluster of unrelated Transformers issues across model loading, tokenizers, processors, training, and quantization. Only one soft pair looks like the same underlying regression: the v5 tied-weights/save_pretrained serialization bug.", - "confidence": 0.84, - "canonical_issue_reason": "Issue 45005 is the better canonical because it states the broader tied-weights regression in v5, while 44466 reads like a specific manifestation of that same save/serialization problem.", + "summary": "Most items are unrelated bug reports across tokenizer, processor, training, quantization, and serialization paths. Only one soft duplicate looks plausible: the tied-weights save/serialization regression pair (44466 \u2194 45005).", + "confidence": 0.87, + "canonical_issue_reason": "Issue 45005 is the best umbrella report: it describes the broader v5 tied-weights serialization regression, which fits the more specific lm_head.weight device-dependent serialization symptom in 44466.", "canonical_pr_reason": null, - "best_issue_reason": "45005 is the most representative issue in the only likely duplicate pair: it captures the shared root cause (tied weights breaking in v5) rather than one device-specific symptom.", + "best_issue_reason": "Issue 45005 is the strongest cluster representative because it is the broadest and most actionable issue about the underlying tied-weights/save_pretrained bug.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44625", "right": "issue:44877", "accept": false, - "reason": "Different bugs: Qwen3.5 label propagation vs granite_speech strict-config loading." + "reason": "Both are config/loading regressions, but they affect different models and different failure modes; not the same underlying bug." }, { "left": "issue:43720", "right": "issue:44898", "accept": false, - "reason": "Unrelated code paths: BitNet packed-weight loading vs Perceiver image classification interpolation." + "reason": "Unrelated code paths: BitNet accelerate weight unpacking vs Perceiver positional interpolation." }, { - "left": "issue:43697", - "right": "issue:45216", + "left": "issue:24643", + "right": "issue:29127", "accept": false, - "reason": "Different regressions in different models: RTDetrV2 output changes vs Qwen3.5 checkpoint save_pretrained corruption." + "reason": "DeepSpeed training weight-shape error and a LayoutLMv3 error-message clarity issue are not the same bug." }, { - "left": "issue:24643", - "right": "issue:29127", + "left": "issue:43697", + "right": "issue:45216", "accept": false, - "reason": "DeepSpeed weight-shape crash and LayoutLMv3 error-message clarity are not the same bug." + "reason": "Both are v5 regressions, but one is RTDetrV2 output drift and the other is Qwen3.5 checkpoint saving; different code paths." }, { "left": "issue:41762", "right": "issue:44936", "accept": false, - "reason": "Both involve training/loading, but one is Gemma3 ZeRO-3 loading and the other is trainer state after train()\u2014distinct failures." + "reason": "Gemma3 ZeRO-3 loading failure and trainer.evaluate-after-train behavior are different problems." }, { "left": "issue:44514", "right": "issue:45381", "accept": false, - "reason": "Same family (Qwen2.5-VL) but different problems: chat-template batching vs video vision_position_ids." + "reason": "Same model family, but one is chat-template batching and the other is video vision_position_ids; not the same concrete bug." }, { "left": "issue:44493", "right": "issue:46050", "accept": false, - "reason": "Position-id warnings and sinq save/from_pretrained breakage are different serialization/config issues." + "reason": "Unexpected position-id keys vs sinq quantized model save/load failure are unrelated." }, { "left": "issue:44466", "right": "issue:45005", "accept": true, - "reason": "Both describe the same tied-weights/v5 serialization regression; 44466 is a concrete device-dependent symptom of the broader 45005 issue." + "reason": "Both report the v5 tied-weights serialization/save_pretrained regression; 44466 is a more specific manifestation of the broader bug in 45005." }, { "left": "issue:43452", "right": "issue:43577", "accept": false, - "reason": "Tokenizer/model-from-pretrained GGUF breakage is unrelated to BLIP2 dtype remaining float32." + "reason": "Tokenizer/AutoModel loading with gguf_file is unrelated to Blip2 dtype propagation." }, { "left": "issue:43906", "right": "issue:46202", "accept": false, - "reason": "A reproduction of one issue and a Gemma pipeline roundtrip tokenization bug are not the same underlying defect." + "reason": "A reproduction of an older issue and a Gemma pipeline roundtrip bug are not the same underlying defect." }, { "left": "issue:43452", "right": "issue:44843", "accept": false, - "reason": "Different tokenizer loading failures: GGUF-related breakage vs offline model_info() call in mistral regex patching." + "reason": "Both involve AutoTokenizer loading, but the gguf_file breakage and offline model_info call are distinct code paths." }, { "left": "issue:30064", "right": "issue:33666", "accept": false, - "reason": "Image-segmentation map handling and Qwen2-VL multi-GPU training are unrelated." + "reason": "Void segmentation map processing is unrelated to Qwen2-VL multi-GPU training." }, { "left": "issue:43688", "right": "issue:43749", "accept": false, - "reason": "Aux-loss normalization and FSDP CPU RAM-efficient loading affect different subsystems and code paths." + "reason": "Auxiliary-loss normalization and FSDP CPU RAM efficient loading are different issues." }, { "left": "issue:43819", "right": "issue:44811", "accept": false, - "reason": "DAC latent reconstruction mismatch is unrelated to Whisper batch_decode skip_special_tokens handling." + "reason": "DAC latent decoding mismatch and Whisper batch_decode skip_special_tokens handling are unrelated." }, { "left": "issue:45372", "right": "issue:46170", "accept": false, - "reason": "Gemma 4 processor import failure and Windows ROCm torch.distributed issues are different environment/import problems." + "reason": "Gemma 4 processor import failure and Windows/ROCm torch.distributed import issues are different backend problems." }, { "left": "issue:28282", "right": "issue:32090", "accept": false, - "reason": "Missing PyTorch import error and Trainer _gpu_broadcast_one NoneType failure are unrelated." + "reason": "Missing PyTorch import and Trainer NoneType broadcast are unrelated errors." }, { "left": "issue:44936", "right": "issue:45823", "accept": false, - "reason": "Trainer evaluate-after-train failure is not the same as Gemma4 device_map=\"auto\" forward device mismatch." + "reason": "Trainer lifecycle failure and Gemma4 device_map forward mismatch are different code paths." + }, + { + "left": "issue:28282", + "right": "issue:33357", + "accept": false, + "reason": "General PyTorch-import error and a MacOS bus error with CLIP are not the same bug." }, { "left": "issue:43295", "right": "issue:45356", "accept": false, - "reason": "Both are tokenizer regressions, but one is processor/tokenizer API breakage and the other is Kimi-K2.5 codec/fix_mistral_regex behavior." + "reason": "Both are regressions, but one is custom processor/tokenizer behavior and the other is Kimi-K2.5 codec/regex handling." }, { "left": "issue:44361", "right": "issue:44442", "accept": false, - "reason": "MLukeTokenizer task AttributeError and FastSpeech2ConformerTokenizer loading failure are different tokenizer implementations." + "reason": "Different tokenizer failures on different models; no shared underlying defect is evident." }, { "left": "issue:43653", "right": "issue:44361", "accept": false, - "reason": "BigBird special-token registration/empty decode is unrelated to MLukeTokenizer task errors." + "reason": "BigBird special-token registration and MLukeTokenizer task AttributeError are unrelated." }, { "left": "issue:44484", "right": "issue:44704", "accept": false, - "reason": "save_pretrained shard-size behavior and AutoProcessor cached_file kwargs are different issues." + "reason": "save_pretrained shard-size question and AutoProcessor kwargs propagation are not the same issue." }, { "left": "issue:38175", "right": "issue:43295", "accept": false, - "reason": "SigLIP zero probabilities and processor.tokenizer regression are unrelated." + "reason": "Siglip2 zero probabilities and processor.tokenizer regression are unrelated." }, { "left": "issue:45542", "right": "issue:46178", "accept": false, - "reason": "TensorBoard-without-TensorFlow backend error and tensorflow_text backend mapping error are distinct backend-registration problems." + "reason": "Both touch backend detection, but they are different missing-backend/registration failures." }, { "left": "issue:24643", "right": "issue:27957", "accept": false, - "reason": "DeepSpeed training weight-shape error and XLMRoberta Flash Attention 2 are different failures." + "reason": "DeepSpeed training weight-shape error and XLMRoberta Flash Attention 2 support are different issues." }, { "left": "issue:43653", "right": "issue:44568", "accept": false, - "reason": "BigBird mask-token special-token bug is unrelated to mDeBERTa BOS/EOS add_special_tokens regression." + "reason": "Both involve token special handling, but the BigBird mask-token bug and mDeBERTa BOS/EOS regression are distinct." }, { - "left": "issue:43577", - "right": "issue:43697", + "left": "issue:44514", + "right": "issue:45290", "accept": false, - "reason": "BLIP2 dtype propagation and RTDetrV2 output drift are unrelated model bugs." + "reason": "Both are chat/template related, but one is batched input padding and the other is tool-call assistant messages; different bugs." }, { - "left": "issue:44514", - "right": "issue:45290", + "left": "issue:43577", + "right": "issue:43697", "accept": false, - "reason": "Both hit Qwen2.5-VL chat templating, but one is batched padding=False and the other is assistant tool-call content handling." + "reason": "Blip2 dtype propagation and RTDetrV2 output differences are unrelated." }, { "left": "issue:44464", "right": "issue:44743", "accept": false, - "reason": "Chunked generation with compiled forward and Qwen3.5 recurrent-state reset are different generation/runtime behaviors." + "reason": "Both involve generation/cache behavior, but one is chunked generation with compiled forward and the other is recurrent-state reset in modular_qwen3_5; not enough to call the same bug." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative. The only accepted soft pair, issue:44466 and issue:45005, is a plausible same-bug match based on their titles around tied weights / save_pretrained regression in v5. The other verdicts do not overstate overlap and are reasonably rejected as distinct issues." + "feedback": "The cluster summary is well grounded in the packet, and the only accepted soft edge (44466 \u2194 45005) is conservative and plausibly the same tied-weights/save_pretrained regression. The rejected pairs are also appropriately cautious given the titles alone." }, "evaluator_used": true, "retried": false, @@ -6983,19 +6915,19 @@ "issue:43688|issue:45701", "issue:44464|issue:44977", "issue:44479|issue:44625", - "issue:43643|issue:44561", "issue:43550|issue:44361", + "issue:43643|issue:44561", "issue:41720|issue:44464", "issue:44977|issue:46209", "issue:32090|issue:34567", "issue:43577|issue:45588", "issue:43577|issue:44554", - "issue:43504|issue:43720", "issue:44466|issue:46050", "issue:32090|issue:33666", + "issue:43504|issue:43720", "issue:43504|issue:43550", - "issue:43504|issue:44610", "issue:33666|issue:36010", + "issue:43504|issue:44610", "issue:43742|issue:44821", "issue:45823|issue:46050", "issue:43262|issue:43295", @@ -7003,226 +6935,226 @@ "issue:43720|issue:43927", "issue:43606|issue:43927" ], - "prepared_review_unit_hash": "ba0f109bd830122856edf37b496a29081f9b34ea64cb20ec149f3a8b2cfd67bb", + "prepared_review_unit_hash": "3c8e171f028b544ed6b1500c5bf520e77889959b48f87162912e819d481c11dc", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 32, - "serialized_chars": 36686, - "estimated_input_tokens": 9172, - "estimated_eval_tokens": 18600 + "serialized_chars": 36691, + "estimated_input_tokens": 9173, + "estimated_eval_tokens": 18602 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This is not a true duplicate cluster: the items span many unrelated transformer bugs (tokenizers, model loading, training, multimodal processors, device maps, quantization, etc.). All soft edges look like false positives rather than the same underlying bug/change.", - "confidence": 0.97, - "canonical_issue_reason": "No single canonical issue stands out because the set is heterogeneous and the proposed pairings do not describe the same root cause.", + "summary": "This cluster is mostly a set of false-positive soft matches across unrelated bugs/regressions (tokenization, model loading, training, generation, quantization, and device-mapping). There are no PRs in the input. If forced to anchor the cluster, issue:43927 is the most connected and concrete save/load regression, but it does not represent a true duplicate umbrella for the whole set.", + "confidence": 0.34, + "canonical_issue_reason": "issue:43927 has the highest soft-link connectivity among the candidates and is a concrete, actionable config/save-load bug around token IDs. It is the best anchor only in a weak clustering sense, not because the whole set is one duplicate.", "canonical_pr_reason": null, - "best_issue_reason": "If one issue must represent the cluster, issue 44843 is the strongest standalone candidate because it is concrete, reproducible, and clearly scoped to a specific regression (offline tokenizer loading).", + "best_issue_reason": "issue:43927 is the strongest overall issue candidate by centrality and clarity of repro; it is broad enough to serve as the cluster anchor, though the surrounding issues are not true duplicates.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:29159", "right": "issue:33357", "accept": false, - "reason": "Different subsystems and failure modes: tokenizer slow/fast parity vs a MacOS bus error on a CLIP model." + "reason": "Tokenizer inconsistency vs MacOS bus error in a CLIP model; different subsystems and failure modes." }, { "left": "issue:44843", "right": "issue:44991", "accept": false, - "reason": "Both involve tokenizer loading, but one is an offline hub access regression and the other is a specific tokenizer failure for a different model family." + "reason": "Both involve tokenizer loading, but one is offline Mistral regex patching and the other is a specific tokenizer load failure; not the same bug." }, { "left": "issue:43646", "right": "issue:43749", "accept": false, - "reason": "Distinct problems: custom model initialization breakage vs FSDP CPU RAM efficient loading regression." + "reason": "Custom model initialization breakage vs FSDP CPU RAM-efficient loading; unrelated code paths." }, { "left": "issue:43531", "right": "issue:43742", "accept": false, - "reason": "Qwen3-MoE sliding window behavior and MobileLLM key errors are unrelated bugs." + "reason": "Qwen3-MoE sliding_window behavior vs MobileLLM key error during load; different issues." }, { "left": "issue:43756", "right": "issue:46402", "accept": false, - "reason": "Different model-specific loading/architecture issues; no shared code-path is evident." + "reason": "Smollm3 RoPE layer selection bug vs Sam3 text encoder weights not loading; unrelated model internals." }, { "left": "issue:43688", "right": "issue:45440", "accept": false, - "reason": "Both are MoE-related, but they concern different implementations and different correctness defects." + "reason": "Aux-loss normalization bug vs DeepseekV3 implementation divergence; not the same underlying defect." }, { "left": "issue:43688", "right": "issue:45704", "accept": false, - "reason": "Aux-loss normalization in MoE models is unrelated to a T5 RMSNorm memory leak." + "reason": "Aux-loss normalization vs T5 apex memory leak; unrelated." }, { "left": "issue:45491", "right": "issue:46002", "accept": false, - "reason": "NaNs appear in different places with different triggers: sliding-window attention on GPU vs classification head init/checkout." + "reason": "Gemma3 sliding-window NaNs vs classification-head NaNs on init; different root causes." }, { "left": "issue:43653", "right": "issue:43927", "accept": false, - "reason": "BigBirdTokenizer special-token registration and DiaConfig token-ID persistence are separate issues." + "reason": "BigBirdTokenizer special-token registration issue vs DiaConfig token-ID persistence loss; both token-related but different mechanisms." }, { "left": "issue:43550", "right": "issue:43927", "accept": false, - "reason": "torch.compile + SDPA on Bamba is unrelated to config token-ID loss on save/load." + "reason": "Bamba torch.compile SDPA failure vs config token-ID persistence; unrelated code paths." }, { "left": "issue:43688", "right": "issue:45701", "accept": false, - "reason": "A model-loss normalization bug is not the same as a broad tokenization-version behavior change." + "reason": "Aux-loss normalization bug vs version-dependent tokenization changes; not the same bug." }, { "left": "issue:44464", "right": "issue:44977", "accept": false, - "reason": "Chunked generation with compiled forward and flash-attention generation regression are different code paths and symptoms." + "reason": "Compiled chunked generation inconsistency vs Qwen3.5 flash-attention generation failure; different generation paths." }, { "left": "issue:44479", "right": "issue:44625", "accept": false, - "reason": "Video input regression for Qwen VL models is unrelated to num_labels propagation in Qwen3.5." - }, - { - "left": "issue:43643", - "right": "issue:44561", - "accept": false, - "reason": "Both mention trust_remote_code, but one is missing fields in AutoConfig output and the other is removal of an FX helper breaking models." + "reason": "Video-input regression vs num_labels propagation bug; unrelated." }, { "left": "issue:43550", "right": "issue:44361", "accept": false, - "reason": "A compile/SDPA model-forward bug is not the same as an MLukeTokenizer AttributeError." + "reason": "torch.compile SDPA failure vs MLukeTokenizer AttributeError; different components." }, { - "left": "issue:41720", - "right": "issue:44464", + "left": "issue:43643", + "right": "issue:44561", + "accept": false, + "reason": "Missing fields from trust_remote_code config loading vs removal of is_torch_fx_available breaking trust_remote_code models; same area, different root cause." + }, + { + "left": "issue:41720", + "right": "issue:44464", "accept": false, - "reason": "Auto device mapping cuda assertions are unrelated to chunked generation discrepancies." + "reason": "Qwen3 auto device-map cuda assert vs chunked generation with compiled forward; unrelated." }, { "left": "issue:44977", "right": "issue:46209", "accept": false, - "reason": "Flash-attention generation failure and FP8 inference on B200 are different hardware/path-specific bugs." + "reason": "Qwen3.5 flash-attention generation issue vs FP8 inference failure on B200; different hardware/code paths." }, { "left": "issue:32090", "right": "issue:34567", "accept": false, - "reason": "Trainer NoneType broadcast error and TrainerState token counter not updating are unrelated Trainer bugs." + "reason": "Trainer broadcast TypeError vs num_input_tokens_seen not updating; both Trainer-related but not the same bug." }, { "left": "issue:43577", "right": "issue:45588", "accept": false, - "reason": "Blip2 dtype loading and flash_attention s_aux=None crash are separate issues." + "reason": "BLIP2 dtype propagation bug vs flash_attention s_aux=None crash; unrelated." }, { "left": "issue:43577", "right": "issue:44554", "accept": false, - "reason": "A dtype persistence issue does not match an MPS attention correctness bug." - }, - { - "left": "issue:43504", - "right": "issue:43720", - "accept": false, - "reason": "Legacy-field model loading and packed-weight unpacking during accelerate loading are different failure causes." + "reason": "BLIP2 dtype issue vs MPS attention correctness issue; different defects." }, { "left": "issue:44466", "right": "issue:46050", "accept": false, - "reason": "Serialization inconsistency of tied weights is not the same as sinq quantized model reload breakage." + "reason": "lm_head serialization inconsistency vs quantized save/load breakage; both serialization-related but not the same failure." }, { "left": "issue:32090", "right": "issue:33666", "accept": false, - "reason": "Trainer broadcast failure and Qwen2-VL multi-GPU training are unrelated." + "reason": "Trainer NoneType broadcast error vs Qwen2-VL multi-GPU training issue; unrelated." }, { "left": "issue:43504", - "right": "issue:43550", + "right": "issue:43720", "accept": false, - "reason": "These are different model-loading/compile bugs with no shared root cause." + "reason": "Beit legacy-field loading bug vs BitNet packed-weight unpacking during accelerate loading; both loading issues but different models and fixes." }, { "left": "issue:43504", - "right": "issue:44610", + "right": "issue:43550", "accept": false, - "reason": "Beit legacy-field loading and OmDet-Turbo processor size mismatch are unrelated." + "reason": "Beit pretrained-load issue vs Bamba torch.compile SDPA failure; unrelated." }, { "left": "issue:33666", "right": "issue:36010", "accept": false, - "reason": "Multi-GPU training for Qwen2-VL has no meaningful overlap with GenerationMixin import failures." + "reason": "Qwen2-VL multi-GPU training vs GenerationMixin import error; unrelated." + }, + { + "left": "issue:43504", + "right": "issue:44610", + "accept": false, + "reason": "Beit legacy-field load failure vs OmDet processor/model input-size mismatch; unrelated." }, { "left": "issue:43742", "right": "issue:44821", "accept": false, - "reason": "Key error loading one model is not the same as loading AutoImageProcessor from a URL." + "reason": "MobileLLM key error on load vs AutoImageProcessor URL loading; different components." }, { "left": "issue:45823", "right": "issue:46050", "accept": false, - "reason": "Device-mapped forward mismatch and quantized save/load corruption are different bugs." + "reason": "Gemma4 device-mismatch during forward vs sinq quantized save/load breakage; unrelated." }, { "left": "issue:43262", "right": "issue:43295", "accept": false, - "reason": "Audio processor chat-template sampling-rate defaulting and processor.tokenizer/image handling regressions are separate issues." + "reason": "Audio chat-template sample-rate default vs processor.tokenizer/image regression; both processor issues but not the same bug." }, { "left": "issue:43452", "right": "issue:44451", "accept": false, - "reason": "gguf_file-related from_pretrained breakage and inability to load a specific model are not the same defect." + "reason": "gguf_file loading failure vs ScandiBERT load failure; unrelated model-loading problems." }, { "left": "issue:43720", "right": "issue:43927", "accept": false, - "reason": "Packed-weight loading in BitNet and config token-ID loss in Dia are unrelated." + "reason": "BitNet packed-weight loading bug vs DiaConfig token-ID persistence loss; different code paths." }, { "left": "issue:43606", "right": "issue:43927", "accept": false, - "reason": "CPU offload device mismatch for bark-small is unrelated to DiaConfig token-ID persistence." + "reason": "CPU-offload device mismatch in bark-small vs config token-ID persistence; unrelated." } ] }, @@ -7319,8 +7251,8 @@ "issue:43334|issue:43572", "issue:43646|issue:44466", "issue:33357|issue:33666", - "issue:43257|issue:43299", "issue:43334|issue:45759", + "issue:43257|issue:43299", "issue:44589|issue:45520", "issue:43824|issue:44589", "issue:45229|issue:45823", @@ -7328,220 +7260,226 @@ "issue:43504|issue:43606", "issue:43572|issue:44291" ], - "prepared_review_unit_hash": "69c312aa1ab65211528509f9ef9734804ac9c78137ec8db2541f773341ad6165", + "prepared_review_unit_hash": "c02270b36815665c85dbff5cbf9a77bd9285b0b6741fc5b06755219b00919931", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 37120, - "estimated_input_tokens": 9280, - "estimated_eval_tokens": 18816 + "serialized_chars": 37111, + "estimated_input_tokens": 9278, + "estimated_eval_tokens": 18812 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These issues are a heterogeneous mix of unrelated tokenizer, configuration, model-loading, training, and runtime problems. None of the soft pairs look like the same underlying bug or change, so no duplicates should be merged here.", - "confidence": 0.96, - "canonical_issue_reason": "No single canonical issue: the set does not form one duplicate cluster and the pairs span distinct subsystems and failure modes.", + "summary": "This cluster is mostly false-positive similarity: the issues span unrelated model families and failure modes. The only broad commonality is \u201crecent Transformers/v5 regressions,\u201d but the concrete bugs differ too much to merge as duplicates.", + "confidence": 0.82, + "canonical_issue_reason": "If one issue has to anchor the cluster, `issue:43646` is the broadest umbrella (`Transformers 5.0.0 breaks custom model initialization`) and best matches the general v5-regression theme seen in a few nearby reports. Still, the cluster is not a true duplicate set.", "canonical_pr_reason": null, - "best_issue_reason": "No issue is a good global representative because the items are not variants of one bug; even the most similar titles refer to different code paths or models.", + "best_issue_reason": "`issue:43646` is the most representative issue because it is phrased at the right abstraction level and covers a general regression class rather than a single model-specific symptom.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43994", "right": "issue:46192", "accept": false, - "reason": "Different problems: SigLIP2 model/pipeline output correctness versus multimodal `return_tensors` handling." + "reason": "Different bugs: SigLIP2 inference correctness vs `return_tensors` being ignored when `text_kwargs` is passed." }, { "left": "issue:33666", "right": "issue:36331", "accept": false, - "reason": "Both involve training, but one is Qwen2-VL multi-GPU training and the other is a `compute_loss` signature break in custom trainers." + "reason": "Different areas and symptoms: multi-GPU training setup vs `compute_loss()` signature breakage in a custom trainer." }, { "left": "issue:43525", "right": "issue:43572", "accept": false, - "reason": "Similar symptom class, but different configs and fields: `Llama4Config.pad_token_id` vs `StableLmConfig.pad_token_idx`." + "reason": "Both are missing token-id config fields after v5, but they affect different model configs (`Llama4Config` vs `StableLmConfig`) and likely separate fixes." }, { "left": "issue:44849", "right": "issue:46209", "accept": false, - "reason": "Qwen3.5 hidden-state behavior is unrelated to FP8 inference failure on B200." + "reason": "Qwen3.5 hidden-state behavior vs FP8 inference failure on B200 are unrelated code paths." }, { "left": "issue:43278", "right": "issue:43688", "accept": false, - "reason": "Different bugs: embedding dtype drift during eval versus auxiliary-loss normalization in MoE models." + "reason": "Embedding dtype drift during evaluate is unrelated to MoE auxiliary-loss normalization." }, { "left": "issue:43720", "right": "issue:44361", "accept": false, - "reason": "Unrelated subsystems: BitNet accelerate loading versus tokenizer/task AttributeError for MLuke." + "reason": "Packed-weight loading bug in BitNet vs tokenizer/task AttributeError in MLuke are different subsystems." }, { "left": "issue:44625", "right": "issue:46002", "accept": false, - "reason": "`num_labels` propagation in Qwen3.5 is not the same as NaNs in classification heads after checkout/init." + "reason": "Config propagation for Qwen3.5 labels is not the same as NaNs in classification heads after checkout/init." }, { "left": "issue:43493", "right": "issue:43825", "accept": false, - "reason": "SigLIP2 implementation mismatch is unrelated to a pipeline error message about translation tasks." + "reason": "SigLIP2 implementation mismatch is unrelated to the pipeline error-message regression." }, { "left": "issue:28218", "right": "issue:33357", "accept": false, - "reason": "Tokenizer spacing regression and MacOS bus error are unrelated failures." + "reason": "Tokenizer spacing bug and MacOS CLIP bus error do not share the same underlying failure." }, { "left": "issue:39692", "right": "issue:43493", "accept": false, - "reason": "Both mention SigLIP2, but one is docs-example issues and the other is an implementation discrepancy; not the same bug." + "reason": "Docs-example mistakes are not the same bug as a SigLIP2 implementation discrepancy." }, { "left": "issue:44964", "right": "issue:45405", "accept": false, - "reason": "Model-loading failure for Phi-4 multimodal is unrelated to the PEFT version pin bump." + "reason": "Model loading failure and a dependency-version bump issue are unrelated." }, { "left": "issue:44849", "right": "issue:45357", "accept": false, - "reason": "Different Qwen3.5 regressions: hidden-states output versus incorrect visual encoder keys in save_pretrained." + "reason": "Different Qwen3.5 regressions: hidden states vs incorrect visual-encoder keys on save." }, { "left": "issue:44568", "right": "issue:45812", "accept": false, - "reason": "Both are tokenizer regressions, but they affect different models and different failure modes." + "reason": "Both affect tokenization, but one is about special tokens on a specific tokenizer and the other is a broad Granite token-ID regression." }, { "left": "issue:44464", "right": "issue:46424", "accept": false, - "reason": "Chunked generation with compiled forward is unrelated to cache length handling for quantized cache." + "reason": "Chunked-generation inconsistency with compiled forward is not the same as cache-length handling for quantized cache." }, { "left": "issue:45200", "right": "issue:45229", "accept": false, - "reason": "Gemma4 text-only mm_token_type_ids defaulting and Gemma4 multi-GPU OOM are different issues." + "reason": "Gemma4 text-only token-type defaults and multi-GPU OOM are different problems." }, { "left": "issue:43540", "right": "issue:44977", "accept": false, - "reason": "Qwen3OmniMoe video-input processing and Qwen3.5 flash-attention generation are unrelated." + "reason": "Video-input processing in Qwen3OmniMoe is unrelated to flash-attention generation behavior in Qwen3.5." }, { "left": "issue:43493", "right": "issue:43844", "accept": false, - "reason": "SigLIP2 implementation discrepancy versus gradient explosion in ZeRO-3 are not the same code-path problem." + "reason": "SigLIP2 model discrepancy and gradient blow-up under DeepSpeed/ZeRO-3 are different code paths." }, { "left": "issue:44488", "right": "issue:44843", "accept": false, - "reason": "Model load failure for `cjvt/sleng-bert` is not the same as the offline-mode hub call in `_patch_mistral_regex`." + "reason": "Loading a specific model and offline-mode breakage from `model_info()` are unrelated." }, { "left": "issue:39692", "right": "issue:43295", "accept": false, - "reason": "Documentation example errors and a regression in `processor.tokenizer`/image passing are distinct issues." + "reason": "Documentation example errors are not the same bug as the processor/tokenizer regression in v4.57.5." }, { "left": "issue:43278", "right": "issue:43844", "accept": false, - "reason": "Embedding dtype mismatch during evaluation is unrelated to ZeRO-3 gradient inflation." + "reason": "Training/eval dtype change is unrelated to ZeRO-3 gradient inflation." }, { "left": "issue:43334", "right": "issue:43572", "accept": false, - "reason": "Different missing-field regressions in different models: Qwen3-VL `pad_token_id` vs StableLm `pad_token_idx`." + "reason": "Both mention missing pad-token fields, but they are different configs and symptoms (`pad_token_id` vs `pad_token_idx`)." }, { "left": "issue:43646", "right": "issue:44466", "accept": false, - "reason": "Custom model initialization breakage is unrelated to device-dependent `lm_head.weight` serialization." + "reason": "Custom model initialization breakage is not the same as device-dependent `lm_head.weight` serialization." }, { "left": "issue:33357", "right": "issue:33666", "accept": false, - "reason": "MacOS bus error on CLIP loading is unrelated to Qwen2-VL multi-GPU training." + "reason": "MacOS CLIP bus error and Qwen2-VL multi-GPU training are unrelated." }, { - "left": "issue:43257", - "right": "issue:43299", + "left": "issue:43334", + "right": "issue:45759", "accept": false, - "reason": "Both mention Qwen3 MoE loading, but one is accelerate/deepspeed weight conversion and the other is a v5 Qwen3VL loading regression." + "reason": "Missing `pad_token_id` in Qwen3-VL config is unrelated to `from_config` not unwrapping `text_config`." }, { - "left": "issue:43334", - "right": "issue:45759", + "left": "issue:43257", + "right": "issue:43299", "accept": false, - "reason": "Missing `pad_token_id` in Qwen3-VL is not the same as failing to unwrap `text_config` in composite multimodal configs." + "reason": "Both involve Qwen3 MoE, but one is an accelerate/deepspeed weight-conversion bug and the other is a load-time regression in dev; not the same concrete fix." }, { "left": "issue:44589", "right": "issue:45520", "accept": false, - "reason": "Float8 storage deserialization and `flash_attn` import mapping on Python 3.13 are unrelated." + "reason": "Float8 storage lookup failure and missing `flash_attn` import key on Python 3.13 are unrelated." }, { "left": "issue:43824", "right": "issue:44589", "accept": false, - "reason": "Importing a missing Qwen2.5-VL class is unrelated to a Float8 storage lookup failure." + "reason": "Missing Qwen2.5-VL symbol export and Float8 storage lookup are separate failures." }, { "left": "issue:45229", "right": "issue:45823", "accept": false, - "reason": "Both are Gemma4, but OOM under multi-GPU inference and device mismatch under `device_map=\"auto\"` are different failures." + "reason": "Gemma4 OOM under multi-GPU inference is not the same as a PLE device-mismatch forward bug." }, { "left": "issue:43927", "right": "issue:44361", "accept": false, - "reason": "DiaConfig save/load token-ID loss is unrelated to MLuke tokenizer task errors." + "reason": "Dia save/load token-ID loss and MLuke tokenizer task AttributeError are different issues." }, { "left": "issue:43504", "right": "issue:43606", "accept": false, - "reason": "Legacy-field preset loading for BEiT and CPU-offload device mismatch for Bark are different bugs." + "reason": "Legacy-field loading failure in Beit semantic segmentation is unrelated to CPU-offload device mismatch in bark-small." + }, + { + "left": "issue:46424", + "right": "issue:44464", + "accept": false, + "reason": "Cache max-length being ignored for quantized cache is not the same bug as inconsistent chunked generation with compiled forward." }, { "left": "issue:43572", "right": "issue:44291", "accept": false, - "reason": "Missing `pad_token_idx` in StableLmConfig and an `init_empty_weights` `_is_hf_initialized` TypeError are different model-loading regressions." + "reason": "Missing `pad_token_idx` in one config and unexpected `_is_hf_initialized` during `init_empty_weights` are different regressions." } ] }, @@ -7616,10 +7554,10 @@ ], "soft_pairs": [ "issue:43278|issue:43701", - "issue:44493|issue:46285", "issue:43575|issue:45405", - "issue:43504|issue:43653", + "issue:44493|issue:46285", "issue:44792|issue:45405", + "issue:43504|issue:43653", "issue:43606|issue:44112", "issue:43994|issue:44493", "issue:43873|issue:46050", @@ -7632,8 +7570,8 @@ "issue:43844|issue:44964", "issue:28218|issue:34567", "issue:29159|issue:32090", - "issue:44554|issue:45446", "issue:45084|issue:45520", + "issue:44554|issue:45446", "issue:43582|issue:44977", "issue:15354|issue:38175", "issue:43540|issue:45200", @@ -7642,190 +7580,190 @@ "issue:45072|issue:45198", "issue:37428|issue:43957" ], - "prepared_review_unit_hash": "f3360c0e438f7552f80ad9a387970b0ac18664bb4cc595bd1b5a0bad49d26c90", + "prepared_review_unit_hash": "37099c17afa29e0f5b4fd755c2e7f7462bf987199ad1eb6c44ec6e6f99e17f49", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 26, - "serialized_chars": 35581, - "estimated_input_tokens": 8896, - "estimated_eval_tokens": 18048 + "serialized_chars": 35585, + "estimated_input_tokens": 8897, + "estimated_eval_tokens": 18050 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: it spans tokenizer behavior, multimodal/chat templating, model loading, attention/flash-attn, training, serving, and environment/version issues. None of the proposed soft pairs look like the same underlying bug, so all should be rejected.", - "confidence": 0.96, - "canonical_issue_reason": "Issue 43873 is the best broad representative: it is open, heavily discussed, has multiple inbound references, and describes a user-facing quantization/offloading failure that is more central than the other, mostly narrow or CI-specific reports.", + "summary": "This cluster is highly heterogeneous: it mixes unrelated tokenizer, training, serving, quantization, attention, and model-loading bugs. None of the proposed soft edges look like the same underlying issue, so all should be rejected. No PRs are present.", + "confidence": 0.94, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Issue 43873 is the strongest single issue to anchor triage because it has the most active discussion and a broad symptom statement, but the cluster overall is not a duplicate set.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43278", "right": "issue:43701", "accept": false, - "reason": "Different failures: dtype change during evaluate vs resume-from-checkpoint key mismatch." + "reason": "One is BF16/FP32 dtype drift in evaluation; the other is a resume_from_checkpoint key mismatch. Different code paths and failure modes." + }, + { + "left": "issue:43575", + "right": "issue:45405", + "accept": false, + "reason": "Tensor-parallel OOM while loading a large model is unrelated to a PEFT minimum-version packaging issue." }, { "left": "issue:44493", "right": "issue:46285", "accept": false, - "reason": "Unrelated symptoms: position-id warning vs CLIP class structure change." + "reason": "Unexpected position-id key handling is not the same as a CLIP text-model structure/API change." }, { - "left": "issue:43575", + "left": "issue:44792", "right": "issue:45405", "accept": false, - "reason": "Qwen2 tensor-parallel OOM and a PEFT version pin are unrelated." + "reason": "A failing Janus image-generation test is unrelated to the PEFT version bump problem." }, { "left": "issue:43504", "right": "issue:43653", "accept": false, - "reason": "Legacy-field model loading bug vs tokenizer special-token registration bug." - }, - { - "left": "issue:44792", - "right": "issue:45405", - "accept": false, - "reason": "Janus image-generation test failure is not the same as a released PEFT version mismatch." + "reason": "Legacy field loading in Beit segmentation and BigBirdTokenizer special-token registration are unrelated bugs." }, { "left": "issue:43606", "right": "issue:44112", "accept": false, - "reason": "CPU offload device mismatch and a GraniteSpeech CI stale test are different problems." + "reason": "Both mention device-related CI failures, but one is CPU offload mismatch and the other is a stale GraniteSpeech test; not the same bug." }, { "left": "issue:43994", "right": "issue:44493", "accept": false, - "reason": "SigLIP2 bad outputs and unexpected position-id keys do not share the same code path." + "reason": "SigLIP2 incorrect outputs and unexpected position-id key warnings are different model-level issues." }, { "left": "issue:43873", "right": "issue:46050", "accept": false, - "reason": "Both involve memory/quantization themes, but one is offloading behavior and the other is broken save/load for sinq quantization." + "reason": "Quantized offloading behavior and sinq save/load corruption are different quantization problems with different failure points." }, { "left": "issue:45588", "right": "issue:46050", "accept": false, - "reason": "Flash-attention s_aux=None crash is unrelated to sinq persistence breakage." + "reason": "A flash-attention integration crash on s_aux=None is unrelated to sinq serialization breakage." }, { "left": "issue:44423", "right": "issue:44734", "accept": false, - "reason": "Serve multimodal .to() crash and KV-cache continuation indexing are different serving bugs." + "reason": "Both affect serve, but one is multimodal continuous batching and the other is KV-cache continuation indexing; different concrete code paths." }, { "left": "issue:43065", "right": "issue:46369", "accept": false, - "reason": "Dummy Conv2d in Sam3PixelDecoder and Sam3 litetext memory overflow are distinct issues." + "reason": "Dummy Conv2d in Sam3PixelDecoder and Sam3-litetext memory overflow are distinct SAM3 issues." }, { "left": "issue:29942", "right": "issue:30333", "accept": false, - "reason": "Flash Attention 2 test failures and MLflow failure-reporting are unrelated CI/product bugs." + "reason": "Flash Attention 2 test failures and MLFlow job reporting on failure are unrelated subsystems." }, { "left": "issue:42490", "right": "issue:43696", "accept": false, - "reason": "Trailing-slash import warning is not the same as a CUDA OOM on GPT-oss-20b." + "reason": "Trailing-slash import warnings are unrelated to a CUDA OOM in GPT-oss-20b." }, { "left": "issue:45406", "right": "issue:46121", "accept": false, - "reason": "Gemma4Processor missing _tokenizer and rope-parameter JSON list handling are separate errors." + "reason": "Gemma4Processor missing _tokenizer and JSON-loaded list handling in rope validation are unrelated bugs." }, { "left": "issue:43844", "right": "issue:44964", "accept": false, - "reason": "ZeRO-3 gradient growth and Phi-4 multimodal model loading failure do not indicate the same defect." + "reason": "ZeRO-3 gradient growth during training is unrelated to loading Phi-4 multimodal models." }, { "left": "issue:28218", "right": "issue:34567", "accept": false, - "reason": "Tokenizer added-space behavior is unrelated to TrainerState token-count tracking." + "reason": "Tokenizer spacing after added tokens and TrainerState num_input_tokens_seen not updating are different tokenizer/trainer behaviors." }, { "left": "issue:29159", "right": "issue:32090", "accept": false, - "reason": "Slow/fast tokenizer inconsistency and _gpu_broadcast_one NoneType are unrelated subsystems." + "reason": "Slow-vs-fast tokenizer inconsistency is unrelated to a Trainer GPU broadcast NoneType error." }, { - "left": "issue:44554", - "right": "issue:45446", + "left": "issue:45084", + "right": "issue:45520", "accept": false, - "reason": "MPS attention correctness and AuxRequest PyTorch version checking are different code paths." + "reason": "Template-node compilation failure and a flash_attn import KeyError are separate import/compile problems." }, { - "left": "issue:45084", - "right": "issue:45520", + "left": "issue:44554", + "right": "issue:45446", "accept": false, - "reason": "Template-node compilation error is unrelated to a flash_attn import KeyError on Python 3.13." + "reason": "An attention correctness bug on MPS is unrelated to a flex_attention PyTorch version check import bug." }, { "left": "issue:43582", "right": "issue:44977", "accept": false, - "reason": "Apple Silicon warmup TypeError and Qwen3.5 flash-attention generation bug are not the same bug." + "reason": "Apple Silicon allocator TypeError and Qwen3.5 flash-attention generation issues are different failures." }, { "left": "issue:15354", "right": "issue:38175", "accept": false, - "reason": "TorchScript GeneratorExp support and SigLIP2 zero probabilities are wholly unrelated." + "reason": "TorchScript GeneratorExp export failure is unrelated to unexpected zero probabilities in SigLIP2." }, { "left": "issue:43540", "right": "issue:45200", "accept": false, - "reason": "Video-input processing failure and Gemma4 mm_token_type_ids defaults are different multimodal issues." + "reason": "Video-input processing in Qwen3OmniMoe and Gemma4 mm_token_type_ids defaults are different multimodal model issues." }, { "left": "issue:44521", "right": "issue:45290", "accept": false, - "reason": "All-zero assistant masks for multimodal inputs is not the same as tool-call messages with no content crashing." + "reason": "All-zero assistant masks for multimodal inputs and tool-call messages with no content are both chat-template related, but they are distinct edge cases." }, { "left": "issue:45200", "right": "issue:45823", "accept": false, - "reason": "Missing mm_token_type_ids defaults and device-map forward mismatch are separate Gemma4 problems." + "reason": "Missing mm_token_type_ids defaults in fine-tuning is unrelated to a device_map auto forward mismatch in Gemma4 PLE." }, { "left": "issue:45072", "right": "issue:45198", "accept": false, - "reason": "Bfloat16 dtype mismatch in inference and Wav2Vec2 save/tokenization failure are unrelated." + "reason": "bfloat16 dtype mismatches in inference and Wav2Vec2 save_pretrained/tokenization failures are unrelated." }, { "left": "issue:37428", "right": "issue:43957", "accept": false, - "reason": "Flash-support import-name error and meta-device model loading breakage are different loading/import issues." + "reason": "Flash-support import errors and meta-device loading regressions are separate model-loading issues." } ] }, @@ -7913,8 +7851,8 @@ "issue:44554|issue:45356", "issue:43572|issue:43957", "issue:45520|issue:46178", - "issue:44466|issue:45127", "issue:37428|issue:43824", + "issue:44466|issue:45127", "issue:43950|issue:46285", "issue:44898|issue:45412", "issue:43550|issue:43844", @@ -7930,14 +7868,14 @@ "issue:28218|issue:35141", "issue:38175|issue:43493" ], - "prepared_review_unit_hash": "f5836354460b9cfae4fe980ef14b5825c9327ffc3bd082fa93d065cd2ccea539", + "prepared_review_unit_hash": "802bc70d31a9cdd6cf52ade4a8c038a6bcd6264dba55fd36ed9ef14e36fd8592", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, @@ -7952,202 +7890,199 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly heterogeneous issue cluster. The only plausible duplicates are two small subclusters: SigLIP2 behavior/parity reports, and missing `pad_token_id` on new config classes. The rest are clearly different bugs, import errors, or unrelated model/processor problems.", - "confidence": 0.74, - "canonical_issue_reason": "`issue:43493` is the broadest, most root-cause-oriented report in the SigLIP2 pair; it can plausibly subsume the more specific zero-probability symptom in `issue:38175`.", + "summary": "This cluster is mostly heterogeneous. The closest items form a loose processor/tokenizer-regression subgroup, but none of the soft pairs look like the same concrete bug or change. Issue 43295 is the most central representative of that subgroup.", + "confidence": 0.88, + "canonical_issue_reason": "Issue 43295 is the broadest, most central item among the few related processor/tokenizer regression reports and has multiple soft links, making it the best cluster anchor.", "canonical_pr_reason": null, - "best_issue_reason": "`issue:43493` is the strongest representative of the only broad behavioral duplicate pair, with the clearest description of the underlying implementation discrepancy.", + "best_issue_reason": "43295 best represents the cluster\u2019s main recurring theme (processor/tokenizer regressions) and is the most general, well-scoped issue among the candidates.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43493", "right": "issue:43701", "accept": false, - "reason": "Different problems: SigLIP2 parity/behavior vs checkpoint resume key mismatch." + "reason": "Different problems: SigLIP2 implementation discrepancy vs. resume-from-checkpoint key mismatch." }, { "left": "issue:44186", "right": "issue:46192", "accept": false, - "reason": "Both involve tokenization/input handling, but the bugs are different and not the same code-path issue." + "reason": "Both involve tokenization/processor APIs, but the crashes/bugs are in different code paths and are not the same underlying issue." }, { "left": "issue:43232", "right": "issue:44849", "accept": false, - "reason": "Unrelated generation-state update bug vs Qwen3.5 hidden-state output bug." + "reason": "Generation kwargs update after sync_gpus is unrelated to the Qwen3.5 hidden-states bug." }, { "left": "issue:45823", "right": "issue:46192", "accept": false, - "reason": "Different subsystems: device-mapped Gemma4 forward mismatch vs `text_kwargs` ignoring `return_tensors`." + "reason": "Device-mapping forward mismatch is unrelated to silently ignored return_tensors handling." }, { "left": "issue:45381", "right": "issue:46209", "accept": false, - "reason": "Qwen2.5-VL video position-id bug is unrelated to FP8 inference failure on B200." + "reason": "Qwen2.5-VL video position-id bug and FP8 inference failure on B200 are different code-path failures." }, { "left": "issue:43295", "right": "issue:43450", "accept": false, - "reason": "Both are processor/multimodal issues, but they affect different code paths and symptoms." + "reason": "Both are processor-related, but one is about processor.tokenizer access/passing images and the other is batched video shape output; not the same bug." }, { "left": "issue:43295", "right": "issue:44610", "accept": false, - "reason": "Different processor regressions: `processor.tokenizer`/image passing vs OmDet-Turbo image-size mismatch." + "reason": "Different model-specific processor bugs: custom tokenizer/image handling vs. OmDet image size mismatch." }, { "left": "issue:44554", "right": "issue:46365", "accept": false, - "reason": "MPS attention correctness bug is unrelated to the Trainer import issue." + "reason": "MPS attention correctness and Trainer import failure are unrelated." }, { "left": "issue:39692", "right": "issue:46192", "accept": false, - "reason": "SigLIP2 doc/example errors are unrelated to silent `return_tensors` handling." + "reason": "Documentation/model mismatch and return_tensors being ignored are separate issues." }, { "left": "issue:43572", "right": "issue:44987", "accept": false, - "reason": "Missing `pad_token_idx` in StableLmConfig is not the same bug as the physical-intelligence/fast load failure." + "reason": "Missing pad_token_idx in a config and a model-loading failure for a specific repo are not the same bug." }, { "left": "issue:43295", "right": "issue:43873", "accept": false, - "reason": "Processor regression vs quantization offloading behavior are different issues." + "reason": "Processor/tokenizer regression is unrelated to quantization offloading behavior." }, { "left": "issue:44554", "right": "issue:45356", "accept": false, - "reason": "Attention correctness on MPS is unrelated to Kimi-K2.5 tokenizer regression." + "reason": "Attention on MPS and Kimi-K2.5 tokenizer regression are unrelated." }, { "left": "issue:43572", "right": "issue:43957", "accept": false, - "reason": "Different model-loading regressions with different failure modes." + "reason": "Both are loading/config regressions, but different models and failure modes." }, { "left": "issue:45520", "right": "issue:46178", "accept": false, - "reason": "Both are import/config errors, but they involve different backend mappings and unrelated symbols." + "reason": "Different import/backend lookup failures in separate modules." }, { - "left": "issue:44466", - "right": "issue:45127", + "left": "issue:37428", + "right": "issue:43824", "accept": false, - "reason": "One is serialization of tied weights; the other is LoRA merge collapse after vocab extension." + "reason": "Distinct import errors for different symbols/models." }, { - "left": "issue:37428", - "right": "issue:43824", + "left": "issue:44466", + "right": "issue:45127", "accept": false, - "reason": "Both are import errors, but for different missing names and different modules." + "reason": "Both touch tied-weight behavior, but one is serialization inconsistency and the other is LoRA merge collapse; not the same concrete issue." }, { "left": "issue:43950", "right": "issue:46285", "accept": false, - "reason": "Non-persistent buffer corruption on load is not the same as CLIP structure inconsistency." + "reason": "Non-persistent buffer corruption is unrelated to CLIPTextModel structure changes." }, { "left": "issue:44898", "right": "issue:45412", "accept": false, - "reason": "Perceiver resolution failure and RT-DETR memory release are unrelated." + "reason": "Perceiver resize failure and RT-DETR memory retention are unrelated." }, { "left": "issue:43550", "right": "issue:43844", "accept": false, - "reason": "torch.compile/SDPA failure is unrelated to ZeRO-3 gradient inflation." + "reason": "torch.compile/SDPA failure is unrelated to gradient blow-up under ZeRO-3." }, { "left": "issue:45200", "right": "issue:45325", "accept": false, - "reason": "Different multimodal bugs: Gemma4 token-type defaults vs Qwen2.5-VL rope/position-id scaling." + "reason": "Text-only mm_token_type_ids defaults and Qwen2.5-VL rope indexing are different bugs." }, { "left": "issue:43334", "right": "issue:43525", - "accept": true, - "reason": "Same underlying bug class: new config objects missing `pad_token_id`, causing AttributeError in the generation/loading path." + "accept": false, + "reason": "Similar symptom name, but different model families and code paths; not the same underlying fix." }, { "left": "issue:43450", "right": "issue:43575", "accept": false, - "reason": "Batched video processor shape bug is unrelated to tensor-parallel OOM." + "reason": "Video processor batching bug is unrelated to Qwen2-57B TP OOM." }, { "left": "issue:42915", "right": "issue:43824", "accept": false, - "reason": "FineGrainedFP8Config failure is unrelated to the Qwen2_5_VL import error." + "reason": "FP8 config failure for Qwen3Moe is unrelated to the Qwen2.5-VL import error." }, { "left": "issue:42490", "right": "issue:43493", "accept": false, - "reason": "Trailing-slash local path warning is unrelated to SigLIP2 implementation parity." + "reason": "Local-path warning behavior is unrelated to the SigLIP2 discrepancy." }, { "left": "issue:41720", "right": "issue:46209", "accept": false, - "reason": "Qwen3 auto device-mapping assert on A800 is unrelated to B200 FP8 inference failure." + "reason": "CUDA assert under auto device mapping and FP8 inference failure on B200 are different hardware/runtime issues." }, { "left": "issue:43824", "right": "issue:43975", "accept": false, - "reason": "ImportError for a missing class and tokenizer detokenization regression are different bugs." + "reason": "Importing a class and tokenizer detokenization are unrelated." }, { "left": "issue:44792", "right": "issue:45563", "accept": false, - "reason": "Failed janus image-generation test and stale warning logic are not the same issue." + "reason": "Failed image generation test and stale warning text are not the same bug." }, { "left": "issue:43388", "right": "issue:43526", "accept": false, - "reason": "Both mention labels, but one is metric gathering and the other is image processor label reduction." + "reason": "Different label-handling layers: metrics gathering vs. image processor reduction." }, { "left": "issue:28218", "right": "issue:35141", "accept": false, - "reason": "Tokenizer spacing bug and embedding reinitialization after resize are separate issues." + "reason": "Tokenizer spacing after added tokens is unrelated to embedding reinitialization after resize." }, { "left": "issue:38175", "right": "issue:43493", - "accept": true, - "reason": "Same SigLIP2 code-path problem: 38175 is a concrete symptom of the broader HF-vs-JAX implementation discrepancy reported in 43493." + "accept": false, + "reason": "Same model family, but one is zero probabilities and the other is a broader implementation discrepancy; not enough to treat as the same bug." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet and stays appropriately conservative. The accepted duplicate pairs are plausible and mergeable into single fixes: the SigLIP2 discrepancy pair (43493/38175) and the missing pad_token_id config pair (43334/43525). The rejected pairs are generally well-justified by distinct titles/symptoms/subsystems, and I don\u2019t see evidence of overclaiming." - }, - "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 }, @@ -8217,8 +8152,8 @@ "issue:43278|issue:43856", "issue:43526|issue:46192", "issue:43653|issue:44112", - "issue:45290|issue:46326", "issue:44442|issue:45198", + "issue:45290|issue:46326", "issue:38175|issue:45412", "issue:43262|issue:43450", "issue:45520|issue:46170", @@ -8231,8 +8166,8 @@ "issue:43827|issue:43901", "issue:43819|issue:45083", "issue:44857|issue:45593", - "issue:43611|issue:46402", "issue:43335|issue:43550", + "issue:43611|issue:46402", "issue:37428|issue:43425", "issue:43606|issue:45083", "issue:29942|issue:32090", @@ -8243,202 +8178,202 @@ "issue:43819|issue:44493", "issue:44855|issue:45083" ], - "prepared_review_unit_hash": "de7c78fe8cdce43b4d840a134d0ceee598269e5cffc20a89b98c089056027867", + "prepared_review_unit_hash": "ec29193c84294c7e88b4afd86360a9942a412df55eed712685de61f8e747aded", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 36338, - "estimated_input_tokens": 9085, - "estimated_eval_tokens": 18426 + "serialized_chars": 36344, + "estimated_input_tokens": 9086, + "estimated_eval_tokens": 18428 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose similarity bucket of mostly unrelated Transformers issues; I don\u2019t see any soft pair that looks like the same underlying bug or change.", - "confidence": 0.96, - "canonical_issue_reason": "No canonical issue: the items span distinct bugs across models, processors, tokenizers, training, CI, and import/runtime compatibility.", + "summary": "This cluster is a grab-bag of mostly unrelated issue reports. There are a few loose thematic overlaps (chat-template bugs, tokenizer/model-loading regressions, import-utils symbols), but none are the same underlying defect, so no soft-edge pair should be merged.", + "confidence": 0.91, + "canonical_issue_reason": "issue:44521 is the strongest issue anchor here: it is open, high-activity, and has the broadest concrete reproduction around multimodal apply_chat_template behavior.", "canonical_pr_reason": null, - "best_issue_reason": "No single issue is a good global representative for this cluster because the topics are too heterogeneous.", + "best_issue_reason": "issue:44521 is the best overall issue candidate because it is active, well-scoped, and central enough to serve as the representative issue for this mixed set.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43278", "right": "issue:43856", "accept": false, - "reason": "Different problems: dtype drift between train/eval vs MoE training memory usage." + "reason": "Different bugs: eval dtype drift vs MoE training memory usage." }, { "left": "issue:43526", "right": "issue:46192", "accept": false, - "reason": "Different processor bugs: label reduction vs `return_tensors` being ignored when `text_kwargs` is passed." + "reason": "Different processor bugs: label reduction vs return_tensors handling." }, { "left": "issue:43653", "right": "issue:44112", "accept": false, - "reason": "Unrelated: BigBirdTokenizer special-token registration vs a GraniteSpeech CI device-override test." + "reason": "Tokenizer special-token registration issue vs GraniteSpeech CI device-override test failure." }, { - "left": "issue:45290", - "right": "issue:46326", + "left": "issue:44442", + "right": "issue:45198", "accept": false, - "reason": "Both involve chat templating, but the concrete bugs differ: tool-call/empty-content crash vs `continue_final_message` on mistral_common tokenizers." + "reason": "Both audio-related, but different failure modes and different model/tokenizer stacks." }, { - "left": "issue:44442", - "right": "issue:45198", + "left": "issue:45290", + "right": "issue:46326", "accept": false, - "reason": "Different tokenizer failures for different models and code paths (`AutoTokenizer` load vs save/tokenization behavior)." + "reason": "Both touch apply_chat_template, but one is tool-call empty-content crashing and the other is mistral_common continue_final_message behavior." }, { "left": "issue:38175", "right": "issue:45412", "accept": false, - "reason": "Completely different areas: SIGLIP output probabilities vs RT-DETR memory release/GC." + "reason": "Different subsystems: unexpected probabilities vs RT-DETR memory release." }, { "left": "issue:43262", "right": "issue:43450", "accept": false, - "reason": "Different modality processors and bugs: audio chat-template sample-rate default vs batched video shape handling." + "reason": "Audio sampling-rate default bug vs video batch-shape bug." }, { "left": "issue:45520", "right": "issue:46170", "accept": false, - "reason": "Different environment/import failures: flash_attn lookup on Python 3.13 vs Windows ROCm/torch.distributed dependency issues." + "reason": "Unrelated platform/import problems: flash_attn lookup on Python 3.13 vs Windows ROCm torch.distributed dependency." }, { "left": "issue:45759", "right": "issue:46285", "accept": false, - "reason": "Both are model-structure regressions, but for different model families and different missing/unwrapped components." + "reason": "Both concern composite model structure, but the concrete failures differ: text_config unwrapping vs CLIP .text_model inconsistency." }, { "left": "issue:44514", "right": "issue:44521", "accept": false, - "reason": "Same general API, but different bugs: batched padding=False crash vs all-zero assistant masks for multimodal inputs." + "reason": "Both are VL/chat-template issues, but one is a padding=False batch crash and the other is all-zero assistant masks." }, { "left": "issue:44610", "right": "issue:44857", "accept": false, - "reason": "Different detection issues: wrong processor image size vs AMP/CUDA crash in a loss function." + "reason": "Different model and bug class: processor image size mismatch vs AMP loss crash." }, { "left": "issue:44561", "right": "issue:46234", "accept": false, - "reason": "Both mention missing availability helpers, but they are different symbols causing different breakages." + "reason": "Both mention import_utils availability helpers, but they are missing different symbols with different breakage paths." }, { "left": "issue:44206", "right": "issue:45584", "accept": false, - "reason": "Different speech bugs: unsupported `center` arg in a feature extractor vs empty-transcription generation failure." + "reason": "Feature-extractor arg regression vs Whisper empty-transcription generation failure." }, { "left": "issue:43819", "right": "issue:44112", "accept": false, - "reason": "Different code paths: DAC latent reconstruction mismatch vs GraniteSpeech stale CI test." + "reason": "Different defects: DAC latent parity vs GraniteSpeech CI device override." }, { "left": "issue:43827", "right": "issue:43901", "accept": false, - "reason": "Both are docs-related, but they refer to different pipeline APIs and different documentation mismatches." + "reason": "Both docs issues, but they target different pipeline APIs and different deprecations." }, { "left": "issue:43819", "right": "issue:45083", "accept": false, - "reason": "Different subsystems and symptoms: DAC latent behavior vs qwen3_omni_moe feature-length helper semantics." + "reason": "Unrelated audio/codebook latent mismatch vs helper length-calculation behavior." }, { "left": "issue:44857", "right": "issue:45593", "accept": false, - "reason": "Different detection-training bugs: AMP crash vs missing auxiliary losses when denoising is off." + "reason": "Different detection-model bugs: float16 AMP crash vs auxiliary-loss omission." }, { - "left": "issue:43611", - "right": "issue:46402", + "left": "issue:43335", + "right": "issue:43550", "accept": false, - "reason": "Different loading regressions affecting different models and versions." + "reason": "Config construction bug vs torch.compile/SDPA runtime incompatibility." }, { - "left": "issue:43335", - "right": "issue:43550", + "left": "issue:43611", + "right": "issue:46402", "accept": false, - "reason": "Unrelated model/config bug vs torch.compile/SDPA failure." + "reason": "Both affect model loading, but one is base_model_prefix handling and the other is missing Sam3 text-encoder weights." }, { "left": "issue:37428", "right": "issue:43425", "accept": false, - "reason": "Different compatibility issues: missing flash-attention helper import vs general Torch 2.10 incompatibility." + "reason": "Specific missing flash_attention_utils symbol vs broad Torch 2.10 incompatibility." }, { "left": "issue:43606", "right": "issue:45083", "accept": false, - "reason": "Different problems in different areas: CPU-offload device mismatch vs feature-extraction length helper behavior." + "reason": "CPU-offload device mismatch vs audio feature-length helper behavior; not the same bug." }, { "left": "issue:29942", "right": "issue:32090", "accept": false, - "reason": "Unrelated: Flash Attention 2 test failures vs Trainer `_gpu_broadcast_one` NoneType error." + "reason": "Flash Attention 2 test failures are unrelated to NoneType GPU broadcast errors in Trainer." }, { "left": "issue:44448", "right": "issue:44743", "accept": false, - "reason": "Different model behaviors: Pegasus output regression vs Qwen recurrent-state reset with cache." + "reason": "Different generation regressions: Pegasus v4/v5 output drift vs qwen3_5 recurrent state reset with cache." }, { "left": "issue:45200", "right": "issue:45588", "accept": false, - "reason": "Different multimodal issues: missing default `mm_token_type_ids` vs flash-attention `s_aux=None` crash." + "reason": "Different multimodal processor bugs: missing mm_token_type_ids default vs flash_attention.py s_aux=None crash." }, { "left": "issue:45127", "right": "issue:45245", "accept": false, - "reason": "Unrelated: LoRA merge/tied embeddings collapse vs category-count limit error." + "reason": "LoRA/tied-embedding collapse is unrelated to category-count overflow." }, { "left": "issue:43335", "right": "issue:43653", "accept": false, - "reason": "Different model/config bug vs tokenizer special-token registration issue." + "reason": "Sparse-layer config bug vs tokenizer special-token registration bug." }, { "left": "issue:43819", "right": "issue:44493", "accept": false, - "reason": "Different failures: DAC latent mismatch vs unexpected key in position-id state dicts." + "reason": "Different symptoms and code paths: DAC latent mismatch vs unexpected position-id keys across models." }, { "left": "issue:44855", "right": "issue:45083", "accept": false, - "reason": "Different issues entirely: Python 3.13 import/torch.jit parsing vs qwen3_omni_moe helper behavior." + "reason": "Python 3.13 decorator parsing/import error vs qwen3_omni_moe feature-length helper bug." } ] }, @@ -8518,12 +8453,12 @@ "issue:45290|issue:45381", "issue:43335|issue:43504", "issue:29127|issue:32090", - "issue:43278|issue:43825", "issue:45588|issue:45859", + "issue:43278|issue:43825", "issue:39692|issue:43526", "issue:45310|issue:45478", - "issue:45459|issue:46234", "issue:39692|issue:44493", + "issue:45459|issue:46234", "issue:43335|issue:45083", "issue:43720|issue:44112", "issue:44448|issue:44779", @@ -8544,229 +8479,223 @@ "issue:43701|issue:45083", "issue:29942|issue:35141" ], - "prepared_review_unit_hash": "be3f77fb01506dda01a53602ed0d11d8ebd4a78a581a33508732b40439543869", + "prepared_review_unit_hash": "ace45230a82d5a3e03ce3f2620a34902461580ac751fcf51d15de5f351044844", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 37049, - "estimated_input_tokens": 9263, - "estimated_eval_tokens": 18782 + "serialized_chars": 37048, + "estimated_input_tokens": 9262, + "estimated_eval_tokens": 18780 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly heterogeneous, with one likely duplicate pair centered on the Qwen3.5 MoE `from_pretrained` regression (45310/45478). The rest of the items appear to be unrelated bugs, feature regressions, or documentation/test issues.", - "confidence": 0.58, - "canonical_issue_reason": "Issue 45310 is the best canonical issue for the Qwen3.5 MoE `from_pretrained` bug because it is an open, concrete report and 45478 appears to describe the same underlying failure with essentially the same scope.", + "summary": "This cluster is mostly a set of unrelated issue reports. The only clear duplicate pair is the repeated Qwen3.5 MoE `from_pretrained` error report; the rest differ by model, subsystem, or failure mode and should stay separate.", + "confidence": 0.93, + "canonical_issue_reason": "issue:45310 is the earlier, broader report of the same Qwen3.5 MoE `from_pretrained` regression as issue:45478, so it is the best issue to keep as canonical among the candidates.", "canonical_pr_reason": null, - "best_issue_reason": "45310 is the best representative issue: it is open, specific, and the earlier of the two highly similar Qwen3.5 MoE `from_pretrained` reports.", + "best_issue_reason": "issue:45310 best represents the only true duplicate theme in this set: the Qwen3.5 MoE `from_pretrained` regression. It is the earlier report and captures the concrete failure more cleanly than the later repost.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:37428", "right": "issue:44351", "accept": false, - "reason": "Different missing-import regressions in different code paths (`_flash_supports_window_size` vs `HybridCache`)." + "reason": "Both are import errors, but they concern different missing symbols (`_flash_supports_window_size` vs `HybridCache`) and likely different code paths." }, { "left": "issue:43550", "right": "issue:44112", "accept": false, - "reason": "Bamba SDPA/torch.compile failure and GraniteSpeech CI flake are unrelated bugs." + "reason": "Different models and failures: torch.compile/SDPA behavior in Bamba versus a GraniteSpeech CI device-override issue." }, { "left": "issue:44206", "right": "issue:45823", "accept": false, - "reason": "Unsupported `center` arg in LasrFeatureExtractor is unrelated to Gemma4 device-map forward mismatch." + "reason": "One is a feature-extractor argument regression; the other is a Gemma4 device-mapping mismatch during forward. Not the same bug." }, { "left": "issue:45290", "right": "issue:45381", "accept": false, - "reason": "Chat-template tool-call crash and Qwen2.5-VL video position-id bug are different failures." + "reason": "Chat template tool-call crash and Qwen2.5-VL video position-id bug are separate failure modes." }, { "left": "issue:43335", "right": "issue:43504", "accept": false, - "reason": "SwitchTransformers sparse-layer config bug and BEiT legacy-field load failure are unrelated." + "reason": "Different architectures and symptoms: SwitchTransformers config construction versus BEiT pretrained preset loading." }, { "left": "issue:29127", "right": "issue:32090", "accept": false, - "reason": "LayoutLMv3 error-message clarity and Trainer `_gpu_broadcast_one` NoneType failure are unrelated." + "reason": "An error-message clarity issue in LayoutLMv3 is unrelated to the Trainer `_gpu_broadcast_one` TypeError." }, { - "left": "issue:43278", - "right": "issue:43825", + "left": "issue:45588", + "right": "issue:45859", "accept": false, - "reason": "BF16/FP32 dtype drift and pipeline translation-warning behavior are different issues." + "reason": "Both touch model internals, but one is a flash-attention crash on `s_aux=None` and the other is a CPU-offload incompatibility in rotary embeddings." }, { - "left": "issue:45588", - "right": "issue:45859", + "left": "issue:43278", + "right": "issue:43825", "accept": false, - "reason": "Flash-attention `s_aux=None` crash and Qwen3_5Moe CPU-offload incompatibility are distinct code paths." + "reason": "A dtype change between train/eval is unrelated to a pipeline warning about translation tasks in v5." }, { "left": "issue:39692", "right": "issue:43526", "accept": false, - "reason": "SigLIP2 docs/example issues do not match the BEiTImageProcessorFast `reduce_labels` bug." + "reason": "Documentation/example errors for SigLIP2 are unrelated to BEiT Fast label reduction returning only one label." }, { "left": "issue:45310", "right": "issue:45478", "accept": true, - "reason": "Same Qwen3.5 MoE `from_pretrained` regression with the same version scope and no meaningful distinguishing behavior." + "reason": "These are the same Qwen3.5 MoE `from_pretrained` error report; the titles and described failure match closely." }, { - "left": "issue:45459", - "right": "issue:46234", + "left": "issue:39692", + "right": "issue:44493", "accept": false, - "reason": "Tokenizer protobuf-error masking and missing `is_falcon_mamba_ssm_available` export are unrelated API issues." + "reason": "A SigLIP2 doc example problem is not the same as the broad positional-id warning regression." }, { - "left": "issue:39692", - "right": "issue:44493", + "left": "issue:45459", + "right": "issue:46234", "accept": false, - "reason": "SigLIP2 documentation problems and unexpected position-id keys are different bugs." + "reason": "Tokenizer protobuf error handling and a missing import-utils symbol are unrelated issues." }, { "left": "issue:43335", "right": "issue:45083", "accept": false, - "reason": "SwitchTransformers config bug and qwen3_omni_moe helper behavior are unrelated." + "reason": "SwitchTransformers sparse-layer construction and qwen3_omni_moe feature-length computation are different code paths." }, { "left": "issue:43720", "right": "issue:44112", "accept": false, - "reason": "Packed-weights loading bug and GraniteSpeech CI device-override flake are unrelated." + "reason": "Packed-weight loading in BitNet is unrelated to the GraniteSpeech stale device-override CI failure." }, { "left": "issue:44448", "right": "issue:44779", "accept": false, - "reason": "Pegasus output regression and DeepSeek tokenizer regression are distinct model-specific problems." - }, - { - "left": "issue:43335", - "right": "issue:44112", - "accept": false, - "reason": "SwitchTransformers config bug and GraniteSpeech CI failure are unrelated." + "reason": "Both are output regressions, but one is Pegasus generation drift and the other is DeepSeek tokenizer tokenization; not the same bug." }, { "left": "issue:43335", "right": "issue:45593", "accept": false, - "reason": "SwitchTransformers sparse-layer bug and D-FINE auxiliary-loss behavior are unrelated." + "reason": "A SwitchTransformers config bug is not the same as D-FINE missing auxiliary losses when denoising is off." }, { "left": "issue:43606", "right": "issue:44857", "accept": false, - "reason": "CPU offload device mismatch and LwDetr AMP crash are separate failures." + "reason": "Both are runtime failures, but one is CPU-offload device mismatch and the other is AMP/CUDA loss behavior in LwDetrImageLoss." }, { "left": "issue:42490", "right": "issue:43854", "accept": false, - "reason": "Local-path import warning and GLM-4.7 test loading issue are unrelated." + "reason": "A local-path trailing-slash warning is unrelated to a GLM-4.7-Flash load failure in tests." }, { "left": "issue:34689", "right": "issue:35141", "accept": false, - "reason": "Llama 3.2 Vision loading regression and embedding reinitialization after resize are different bugs." + "reason": "Llama 3.2 Vision loading breakage and embedding reinitialization on resize are different issues." }, { "left": "issue:44360", "right": "issue:44561", "accept": false, - "reason": "DSA indexer/ReLU discussion and `is_torch_fx_available` removal breakage are unrelated." + "reason": "A DSA indexer activation omission is unrelated to the removal of `is_torch_fx_available` breaking `trust_remote_code` models." }, { "left": "issue:38175", "right": "issue:44493", "accept": false, - "reason": "SigLIP2 zero-probability behavior and unexpected position-id keys are distinct bugs." + "reason": "Unexpected zero probabilities in SigLIP2 and unexpected key/position-id warnings are not the same underlying defect." }, { "left": "issue:44857", "right": "issue:45072", "accept": false, - "reason": "LwDetr AMP loss crash and dtype mismatches in SwitchTransformers/TimmWrapper inference are different issues." + "reason": "Both mention precision/device behavior, but they concern different models and different failure modes." }, { "left": "issue:45920", "right": "issue:46192", "accept": false, - "reason": "AutoTokenizer wrong-token-ID regression and `return_tensors` being ignored with `text_kwargs` are not the same bug." + "reason": "Wrong token IDs across multiple models is a tokenizer regression; silently ignoring `return_tensors` is a separate API bug." }, { "left": "issue:44448", "right": "issue:44977", "accept": false, - "reason": "Pegasus output regression and Qwen3.5 flash-attention generation issue are unrelated." + "reason": "Pegasus v4/v5 output drift and Qwen3.5 flash-attention generation failure are distinct model-specific problems." }, { "left": "issue:44977", "right": "issue:45229", "accept": false, - "reason": "Flash-attention generation bug and Gemma4 multi-GPU OOM are different problems." + "reason": "Flash-attention generation failure in Qwen3.5 is unrelated to Gemma4 multi-GPU OOM." }, { "left": "issue:28218", "right": "issue:32090", "accept": false, - "reason": "Tokenizer added-space behavior and Trainer NoneType broadcast error are unrelated." + "reason": "Tokenizer spacing after added tokens has no connection to the Trainer broadcast TypeError." }, { "left": "issue:40444", "right": "issue:43526", "accept": false, - "reason": "Qwen2.5-VL multi-image finetuning failure and BEiT reduce_labels bug are unrelated." + "reason": "Multi-image IterableDataset fine-tuning failure is unrelated to BEiT Fast label reduction behavior." }, { "left": "issue:43746", "right": "issue:46209", "accept": false, - "reason": "Local checkpoint PEFT loading and FP8 B200 inference failure are distinct." + "reason": "PEFT adapter loading from local checkpoints and FP8 inference on B200 are separate issues." }, { "left": "issue:43701", "right": "issue:45083", "accept": false, - "reason": "Resume-from-checkpoint key mismatch and qwen3_omni_moe helper behavior are unrelated." + "reason": "Checkpoint key mismatch and qwen3_omni_moe output-length helper behavior are different bugs." }, { "left": "issue:29942", "right": "issue:35141", "accept": false, - "reason": "Flash Attention 2 CI failures and embedding reinitialization after resize are different bugs." + "reason": "Flash Attention 2 test failures and embedding reinitialization after resize are unrelated." } ] }, "evaluator_result": { - "accept": false, - "feedback": "The cluster-level conclusion is too strong for the packet. The evidence supports that 45310 and 45478 are highly similar and likely about the same Qwen3.5 MoE `from_pretrained` bug, but the packet does not explicitly confirm them as duplicates. Phrases like \"only confirmed duplicate set,\" \"exact regression,\" and \"later duplicate\" overstate the evidence. The soft-edge rejections are generally conservative, but the duplicate verdict should be downgraded to likely/same-topic rather than confirmed." + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative: it only treats the exact Qwen3.5 MoE `from_pretrained` pair as a duplicate and keeps the other soft pairs separate based on different titles/failure modes. The canonical-issue choice is also supported by the earlier identical report. " }, "evaluator_used": true, - "retried": true, + "retried": false, "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null @@ -8815,6 +8744,7 @@ "issue:44779", "issue:44855", "issue:44857", + "issue:44871", "issue:44977", "issue:44987", "issue:45083", @@ -8836,15 +8766,15 @@ "issue:32090|issue:33357", "issue:44855|issue:46399", "issue:43550|issue:45083", - "issue:43927|issue:45698", "issue:42490|issue:43575", "issue:10000|issue:15354", + "issue:43927|issue:45698", "issue:44561|issue:44779", "issue:37428|issue:44987", "issue:43873|issue:45229", "issue:43278|issue:43329", - "issue:41720|issue:45229", "issue:44857|issue:45588", + "issue:41720|issue:45229", "issue:27957|issue:34689", "issue:44977|issue:45563", "issue:44568|issue:45491", @@ -8860,204 +8790,211 @@ "issue:45245|issue:45563", "issue:43653|issue:44743", "issue:44206|issue:44479", - "issue:32090|issue:33290" + "issue:32090|issue:33290", + "issue:44871|issue:44977" ], - "prepared_review_unit_hash": "08388b72805ea8acd4665019f04a9816810c2d5ec72aca918c34de2c015d2ff8", + "prepared_review_unit_hash": "abdbe7037d7b45237b9c7a8dc6b4f66123782354f9dd893356811d33e4da4fc8", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 28, - "serialized_chars": 35784, - "estimated_input_tokens": 8946, - "estimated_eval_tokens": 18148 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36700, + "estimated_input_tokens": 9175, + "estimated_eval_tokens": 18606 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is highly heterogeneous: the soft pairs share superficial vocabulary or broad subsystem overlap, but none look like the same underlying bug/change. No PRs are present, and I would reject every soft edge as a duplicate candidate.", - "confidence": 0.94, - "canonical_issue_reason": null, + "summary": "This cluster is a mixed bag of unrelated Transformers issues rather than a duplicate set. No soft-edge pair looks like the same underlying bug or change. The most broadly representative issue is the save/load custom-module regression (#45698).", + "confidence": 0.82, + "canonical_issue_reason": "#45698 is the clearest broad regression with concrete repro and multiple references: `from_pretrained` loading the wrong custom module after `save_pretrained`.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45588 is the strongest standalone issue: it has a concrete crash, a specific file/function (`integrations/flash_attention.py`), and a clear repro path. It is actionable, but it still does not unify the other items.", + "best_issue_reason": "#45698 is the strongest issue candidate overall because it is specific, actionable, and affects a common serialization/loading path.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:32090", "right": "issue:33357", "accept": false, - "reason": "Both are runtime failures, but one is a Trainer broadcast NoneType error and the other is a MacOS bus error on a community CLIP model; different code paths and symptoms." + "reason": "Different failures in different paths: Trainer GPU broadcast NoneType vs CLIP model bus error on macOS." }, { "left": "issue:44855", "right": "issue:46399", "accept": false, - "reason": "Both mention decorator/script parsing, but one is a Python 3.13 torch.jit import error and the other is a Qwen3.5 registration bug; not the same failure." + "reason": "Both mention parsing/registration problems, but they affect different models and unrelated code paths." }, { "left": "issue:43550", "right": "issue:45083", "accept": false, - "reason": "Different models and bugs: torch.compile/SDPA failure in Bamba vs a feature-length helper bug in qwen3_omni_moe." - }, - { - "left": "issue:43927", - "right": "issue:45698", - "accept": false, - "reason": "Both involve save/load, but one is custom token IDs being lost and the other is the wrong custom module being loaded; distinct serialization bugs." + "reason": "Unrelated bugs: torch.compile/SDPA failure in Bamba vs helper output-length logic in qwen3_omni_moe." }, { "left": "issue:42490", "right": "issue:43575", "accept": false, - "reason": "A trailing-slash import warning is unrelated to a tensor-parallel OOM when loading Qwen2-57B-A14B-Instruct." + "reason": "Import warning for local paths ending with slash is unrelated to Qwen2 TP OOM." }, { "left": "issue:10000", "right": "issue:15354", "accept": false, - "reason": "Completely different models and failures: DistilBertModel issue vs torch.jit.script GeneratorExp/export problem for ViT." + "reason": "Completely different models and issues: DistilBert error vs torch.jit GeneratorExp export failure." + }, + { + "left": "issue:43927", + "right": "issue:45698", + "accept": false, + "reason": "Both involve save/load, but one loses custom token IDs while the other loads the wrong custom module; different bugs." }, { "left": "issue:44561", "right": "issue:44779", "accept": false, - "reason": "One is a removed API breaking trust_remote_code models; the other is tokenizer output corruption in Deepseek. Related release era, not the same bug." + "reason": "Removal of `is_torch_fx_available` breaks remote-code models; Deepseek tokenizer regression is a separate tokenizer issue." }, { "left": "issue:37428", "right": "issue:44987", "accept": false, - "reason": "An import error from flash-attention utils and a model-loading failure for a specific repo are not the same code-path problem." + "reason": "One is a missing import symbol in flash attention utils; the other is a model-loading failure for a specific repo." }, { "left": "issue:43873", "right": "issue:45229", "accept": false, - "reason": "Quantization/offloading behavior and multi-GPU OOM are different resource-management issues with different likely fixes." + "reason": "Quantized offloading behavior and multi-GPU CUDA OOM are different problems." }, { "left": "issue:43278", "right": "issue:43329", "accept": false, - "reason": "BF16-to-FP32 dtype drift in eval is unrelated to undefined variables in a video token counting helper." + "reason": "BF16-to-FP32 dtype drift in eval is unrelated to the undefined-variable bug in multimodal token counting." }, { - "left": "issue:41720", - "right": "issue:45229", + "left": "issue:44857", + "right": "issue:45588", "accept": false, - "reason": "A cudaErrorAssert during auto device mapping is not the same as a multi-GPU OOM while loading Gemma4." + "reason": "Different crash sites: AMP image loss on CUDA vs flash_attention integration on sink-less models." }, { - "left": "issue:44857", - "right": "issue:45588", + "left": "issue:41720", + "right": "issue:45229", "accept": false, - "reason": "AMP/CUDA loss crash in LwDetrImageLoss is unrelated to an AttributeError in flash_attention handling of `s_aux=None`." + "reason": "Auto device-map assert on Qwen3 is not the same as Gemma4 multi-GPU OOM." }, { "left": "issue:27957", "right": "issue:34689", "accept": false, - "reason": "Both concern model loading/attention, but XLMRoberta Flash Attention 2 and Llama 3.2 90B Vision load break are distinct issues." + "reason": "Flash Attention 2 support issue and Llama 3.2 model loading regression are unrelated." }, { "left": "issue:44977", "right": "issue:45563", "accept": false, - "reason": "Flash-attention generation regression and a stale warning for num_return_sequences do not share the same bug surface." + "reason": "Generation bug with flash-attention is unrelated to a stale warning in paged generate()." }, { "left": "issue:44568", "right": "issue:45491", "accept": false, - "reason": "Tokenizer special-token insertion bug vs Gemma3 NaN embeddings on GPU; different layers of the stack." + "reason": "Tokenizer special-token handling and Gemma3 NaN embeddings are different failures." }, { "left": "issue:45372", "right": "issue:45588", "accept": false, - "reason": "Gemma4 processor loading fails because a dependency import is missing, while the other crashes inside flash_attention on `s_aux=None`; different root causes." + "reason": "Gemma4 processor import dependency error is not the same as flash_attention.py crashing on `s_aux=None`." }, { "left": "issue:43278", "right": "issue:43381", "accept": false, - "reason": "Eval-mode gradient checkpointing restriction is unrelated to embedding dtype changes." + "reason": "Dtype mismatch during eval is unrelated to the gradient-checkpointing eval-mode restriction." }, { "left": "issue:43412", "right": "issue:45588", "accept": false, - "reason": "Executorch export issues for gemma3n are a different path from a runtime flash_attention AttributeError." + "reason": "Executorch export failure and flash_attention `s_aux=None` crash are different code paths." }, { "left": "issue:45588", "right": "issue:45823", "accept": false, - "reason": "A flash_attention crash and a Gemma4 PLE device mismatch are separate bugs with different fixes." + "reason": "Both are Gemma4-related, but one is a flash-attention attribute crash and the other is a device-map forward mismatch." }, { "left": "issue:45372", "right": "issue:45406", "accept": false, - "reason": "Missing `ReasoningEffort` import during processor loading is unrelated to a missing `_tokenizer` attribute in serve." + "reason": "ImportError from `mistral_common` and missing `_tokenizer` on Gemma4Processor are different regressions." }, { "left": "issue:44448", "right": "issue:44466", "accept": false, - "reason": "Both are v5 regressions, but one is output divergence and the other is serialization of tied weights; not the same concrete defect." + "reason": "Sequence-output differences and inconsistent `lm_head.weight` serialization are not the same bug." }, { "left": "issue:43976", "right": "issue:45440", "accept": false, - "reason": "Python version compatibility and DeepseekV3MoE behavioral divergence are unrelated." + "reason": "Python-version compatibility failure and DeepseekV3MoE divergence are unrelated." }, { "left": "issue:45198", "right": "issue:45803", "accept": false, - "reason": "A save_pretrained/tokenization failure is not the same as a bare-except bug swallowing exceptions." + "reason": "Wav2Vec2 save/tokenization failure and Fuyu bare-except cleanup bug do not share the same underlying defect." }, { "left": "issue:25251", "right": "issue:32090", "accept": false, - "reason": "top_k output nesting in pipeline is unrelated to a Trainer GPU broadcast NoneType error." + "reason": "Pipeline `top_k` shape change is unrelated to the Trainer broadcast error." }, { "left": "issue:45245", "right": "issue:45563", "accept": false, - "reason": "Category-count limits and a stale warning in paged generate() are different issues." + "reason": "Category-count runtime limit and stale warning in paged generate are unrelated." }, { "left": "issue:43653", "right": "issue:44743", "accept": false, - "reason": "Tokenizer special-token registration and recurrent-state reset in a model file are unrelated." + "reason": "Tokenizer special-token registration bug and recurrent-state reset in Qwen3.5 are unrelated." }, { "left": "issue:44206", "right": "issue:44479", "accept": false, - "reason": "Unsupported feature arg in an extractor and a video-input regression in Qwen3.5 are not the same bug." + "reason": "Feature-extractor center-arg regression and video-input regression are different issues." }, { "left": "issue:32090", "right": "issue:33290", "accept": false, - "reason": "Unsupported NoneType passed to `_gpu_broadcast_one` is unrelated to an OOM caused by Adafactor in DeepSpeed." + "reason": "Unsupported NoneType broadcast and Adafactor/deepspeed OOM are unrelated training failures." + }, + { + "left": "issue:44871", + "right": "issue:44977", + "accept": false, + "reason": "Gemma eos_token mismatch is unrelated to Qwen3.5 flash-attention generation failure." } ] }, @@ -9103,6 +9040,7 @@ "issue:43653", "issue:43701", "issue:43756", + "issue:43761", "issue:43824", "issue:43873", "issue:43994", @@ -9114,8 +9052,6 @@ "issue:44462", "issue:44464", "issue:44625", - "issue:44871", - "issue:44977", "issue:45198", "issue:45372", "issue:45412", @@ -9126,25 +9062,25 @@ "issue:45593", "issue:45712", "issue:45834", + "issue:45859", "issue:46121", "issue:46369", "issue:46399" ], "soft_pairs": [ "issue:43065|issue:45412", - "issue:44871|issue:44977", "issue:44186|issue:45198", "issue:43611|issue:46121", "issue:42490|issue:43873", - "issue:33357|issue:39692", "issue:44361|issue:45198", + "issue:33357|issue:39692", "issue:29127|issue:34567", "issue:43065|issue:43994", "issue:45459|issue:45712", "issue:42915|issue:42994", "issue:43066|issue:44462", - "issue:29942|issue:30064", "issue:45372|issue:45459", + "issue:29942|issue:30064", "issue:45412|issue:46369", "issue:43278|issue:43575", "issue:43994|issue:45412", @@ -9163,234 +9099,235 @@ "issue:46121|issue:46399", "issue:43526|issue:44186", "issue:29127|issue:35141", - "issue:44625|issue:45584" + "issue:44625|issue:45584", + "issue:43761|issue:45859" ], - "prepared_review_unit_hash": "18c624fbb8ddcf6058e491a869c7ec5be175777bcfb4d96fbb1ca41c4b90089f", + "prepared_review_unit_hash": "18588924c3926efe7337de00b0e0cc79a046b9feab2e680f40e235031901e1d1", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 33, - "serialized_chars": 37492, - "estimated_input_tokens": 9373, - "estimated_eval_tokens": 19002 + "serialized_chars": 37493, + "estimated_input_tokens": 9374, + "estimated_eval_tokens": 19004 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is very heterogeneous; most soft pairs are only keyword-similar (often sharing a subsystem like tokenizers, quantization, or attention) but do not describe the same underlying bug. No PRs are present.", - "confidence": 0.69, - "canonical_issue_reason": "issue:44462 is the cleanest, most self-contained regression in the set (AutoTokenizer ignoring tokenizer.json), while most other issues are unrelated, model-specific bugs.", + "summary": "False-positive cluster: the issues span many unrelated models, tokenizers, and failure modes. None of the soft-edge pairs look like the same underlying bug or a mergeable duplicate PR case.", + "confidence": 0.97, + "canonical_issue_reason": "No safe canonical issue stands out; the set is heterogeneous and does not center on one underlying defect.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44462 has the clearest repro and narrow scope, making it the best representative anchor for this otherwise mixed cluster.", + "best_issue_reason": "No single issue is a good global representative because the cluster mixes unrelated bugs across different subsystems and model families.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43065", "right": "issue:45412", "accept": false, - "reason": "Different problems: a dummy Conv2d placeholder in Sam3PixelDecoder vs RT-DETR memory not being released." - }, - { - "left": "issue:44871", - "right": "issue:44977", - "accept": false, - "reason": "Same broad area only; eos_token_id mismatch in Gemma-3 is unrelated to Qwen3.5 flash-attention generation failure." + "reason": "Both mention model internals and memory, but one is a dummy Conv2d placeholder and the other is a garbage-collection memory leak; different bugs." }, { "left": "issue:44186", "right": "issue:45198", "accept": false, - "reason": "Different tokenizers and failure modes: LayoutLMv2 NER/padding crash vs Wav2Vec2 save/tokenization failure." + "reason": "Different tokenizer/model families and different failure paths; no shared code-path bug." }, { "left": "issue:43611", "right": "issue:46121", "accept": false, - "reason": "Unrelated code paths: base_model_prefix loading regression vs rope validation type error from JSON-loaded lists." + "reason": "Loading with base_model_prefix and rope-param dict conversion are unrelated issues." }, { "left": "issue:42490", "right": "issue:43873", "accept": false, - "reason": "Path-with-slash import warning and quantization offloading behavior are different bugs." + "reason": "Local-path import warning and quantization/offloading behavior are unrelated." }, { - "left": "issue:33357", - "right": "issue:39692", + "left": "issue:44361", + "right": "issue:45198", "accept": false, - "reason": "MacOS bus error with a community CLIP model is not the same as SigLIP2 documentation example errors." + "reason": "Both are tokenizer-related, but they affect different models with different symptoms and causes." }, { - "left": "issue:44361", - "right": "issue:45198", + "left": "issue:33357", + "right": "issue:39692", "accept": false, - "reason": "Different tokenizers and symptoms; MLuke AttributeError is not the same as Wav2Vec2 save/tokenization breakage." + "reason": "MacOS bus error with CLIP is unrelated to SigLIP2 documentation/example errors." }, { "left": "issue:29127", "right": "issue:34567", "accept": false, - "reason": "LayoutLMv3 error-message clarity and TrainerState token accounting are unrelated issues." + "reason": "LayoutLMv3 error handling and TrainerState token counting are unrelated." }, { "left": "issue:43065", "right": "issue:43994", "accept": false, - "reason": "Different model bugs: Sam3 pixel decoder placeholder vs SigLIP2 producing nonsensical outputs." + "reason": "Dummy Conv2d placeholder issue is not the same as quantized model saving failure." }, { "left": "issue:45459", "right": "issue:45712", "accept": false, - "reason": "Tokenizer exception masking and leftover dummy classes are separate maintenance bugs." + "reason": "Tokenizer exception masking is unrelated to leftover dummy classes." }, { "left": "issue:42915", "right": "issue:42994", "accept": false, - "reason": "Both mention quantization, but one is Qwen3Moe FP8 load failure and the other is quantized model saving; not the same code path." + "reason": "Both involve quantization-adjacent topics, but one is FP8 loading and the other is quantized saving; different code paths." }, { "left": "issue:43066", "right": "issue:44462", "accept": false, - "reason": "Tokenizer decoder-type migration bug is unrelated to AutoTokenizer ignoring tokenizer.json from the repo." + "reason": "Wrong decoder type in v5 and ignoring tokenizer.json are separate tokenizer-loading bugs." }, { - "left": "issue:29942", - "right": "issue:30064", + "left": "issue:45372", + "right": "issue:45459", "accept": false, - "reason": "Flash Attention 2 test failures and void segmentation-map processing are different subsystems." + "reason": "Gemma 4 processor import failure and protobuf-related tokenizer error masking are unrelated." }, { - "left": "issue:45372", - "right": "issue:45459", + "left": "issue:29942", + "right": "issue:30064", "accept": false, - "reason": "Gemma 4 processor import failure is unrelated to protobuf error masking in tokenizer decoding." + "reason": "Flash Attention 2 test failures and void segmentation-map processing are different subsystems." }, { "left": "issue:45412", "right": "issue:46369", "accept": false, - "reason": "Both are memory-related, but one is a leak on deletion and the other is an OOM during sam3-litetext usage." + "reason": "Memory leak on deletion is not the same as memory overflow when calling sam3-litetext." }, { "left": "issue:43278", "right": "issue:43575", "accept": false, - "reason": "BF16-to-FP32 embedding dtype change in eval is unrelated to tensor-parallel OOM on Qwen2-57B." + "reason": "Embedding dtype drift and tensor-parallel OOM are different problems." }, { "left": "issue:43994", "right": "issue:45412", "accept": false, - "reason": "SigLIP2 incorrect outputs and RT-DETR memory release are unrelated bugs." + "reason": "Quantized model saving failure is unrelated to model memory release on deletion." }, { "left": "issue:44265", "right": "issue:45198", "accept": false, - "reason": "torch.export failure with torch_compilable_check is not the same as Wav2Vec2 save_pretrained/tokenization breakage." + "reason": "torch.export failure from torch_compilable_check is unrelated to tokenizer/save_pretrained issues." }, { "left": "issue:43381", "right": "issue:43701", "accept": false, - "reason": "Gradient checkpointing in eval and resume_from_checkpoint key mismatch are different issues." + "reason": "Eval-mode gradient checkpointing and checkpoint resume key mismatch are different issues." }, { "left": "issue:43756", "right": "issue:45440", "accept": false, - "reason": "Different model-implementation divergences: Smollm3 RoPE-layer dropping vs DeepSeekV3MoE remote/native mismatch." + "reason": "Both concern model implementation discrepancies, but they are for different model families and different missing behavior." }, { "left": "issue:43653", "right": "issue:44442", "accept": false, - "reason": "BigBird tokenizer special-token registration bug is not the same as AutoTokenizer failing to load FastSpeech2ConformerTokenizer." + "reason": "BigBird special-token decode bug and AutoTokenizer loading failure are distinct tokenizer bugs." }, { "left": "issue:43421", "right": "issue:45491", "accept": false, - "reason": "Updating tokenizers post-processors at runtime is unrelated to Gemma3 NaN embeddings with sliding-window attention." + "reason": "Runtime post-processor updates and Gemma3 NaN embeddings are unrelated." }, { "left": "issue:27957", "right": "issue:29942", "accept": false, - "reason": "XLMRoberta Flash Attention 2 support and failing FA2 tests overlap only by topic, not by concrete bug." + "reason": "XLMRoberta Flash Attention 2 compatibility and a broader FA2 test failure are not clearly the same bug." }, { "left": "issue:29127", "right": "issue:33357", "accept": false, - "reason": "LayoutLMv3 error-message issue and CLIP bus error on MacOS are unrelated." + "reason": "LayoutLMv3 error messaging and CLIP MacOS bus error are unrelated." }, { "left": "issue:43824", "right": "issue:44279", "accept": false, - "reason": "Qwen2_5_VL import error is a specific symbol/version issue, while the other is a broad dependency problem." + "reason": "Missing Qwen2.5-VL import and a generic dependency issue do not match the same defect." }, { "left": "issue:43994", "right": "issue:46369", "accept": false, - "reason": "SigLIP2 wrong outputs and SAM3 memory overflow do not share the same failing code path." + "reason": "Quantized saving failure and SAM3 memory overflow are unrelated." }, { "left": "issue:45440", "right": "issue:45593", "accept": false, - "reason": "DeepSeekV3MoE divergence and D-FINE auxiliary-loss behavior are different model bugs." + "reason": "DeepSeekV3 implementation divergence and D-FINE auxiliary-loss behavior are different model bugs." }, { "left": "issue:43526", "right": "issue:45834", "accept": false, - "reason": "BeitImageProcessorFast label reduction and Kosmos2.5 long-OCR index error are unrelated image-processing failures." + "reason": "BeitImageProcessorFast label reduction and Kosmos2.5 OCR index error are different image-processing failures." }, { "left": "issue:44464", "right": "issue:45584", "accept": false, - "reason": "Compiled-forward chunked generation inconsistency is unrelated to Whisper empty-transcription failure after align_special_tokens." + "reason": "Chunked generation with compiled forward and Whisper empty-transcription failure are different generation paths." }, { "left": "issue:46121", "right": "issue:46399", "accept": false, - "reason": "Rope-validation list typing and apply_rotary_pos_emb registration errors are separate implementation defects." + "reason": "Rope-param dict conversion TypeError and missing decorator for rotary-pos-emb registration are unrelated." }, { "left": "issue:43526", "right": "issue:44186", "accept": false, - "reason": "ImageProcessor label handling and tokenizer crashes on NER inputs are different bug classes." + "reason": "BeitImageProcessorFast label reduction and LayoutLMv2Tokenizer NER/padding crashes are unrelated." }, { "left": "issue:29127", "right": "issue:35141", "accept": false, - "reason": "LayoutLMv3 error messaging and post-init embedding reinitialization are unrelated." + "reason": "LayoutLMv3 error messaging and embedding reinitialization after resize are different bugs." }, { "left": "issue:44625", "right": "issue:45584", "accept": false, "reason": "Qwen3.5 num_labels propagation and Whisper empty transcription are unrelated." + }, + { + "left": "issue:43761", + "right": "issue:45859", + "accept": false, + "reason": "CLIPVision hidden_states regression and Qwen3_5Moe CPU-offload incompatibility are different code paths." } ] }, @@ -9459,20 +9396,18 @@ "issue:45479", "issue:45491", "issue:45588", - "issue:45859", "issue:45907", "issue:46234" ], "soft_pairs": [ - "issue:43761|issue:45859", "issue:44442|issue:45072", "issue:43575|issue:43701", "issue:42915|issue:43931", "issue:42490|issue:43381", "issue:43575|issue:44387", "issue:43526|issue:44857", - "issue:44977|issue:45245", "issue:43334|issue:43531", + "issue:44977|issue:45245", "issue:43262|issue:44206", "issue:44464|issue:45491", "issue:27957|issue:29127", @@ -9480,8 +9415,8 @@ "issue:44743|issue:44811", "issue:27957|issue:33290", "issue:43646|issue:44077", - "issue:44062|issue:45907", "issue:33290|issue:34689", + "issue:44062|issue:45907", "issue:33290|issue:34567", "issue:33290|issue:35141", "issue:43381|issue:43873", @@ -9495,214 +9430,220 @@ "issue:34689|issue:43611", "issue:43720|issue:44442" ], - "prepared_review_unit_hash": "31f961d01637e5125352a4c445b9e3bf2cd415ea1069820fbfdcbf8f310526c6", + "prepared_review_unit_hash": "e0fd499cf4db813c151933d2712c5647c57a2a623de7505ae0864b2b3e23e002", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 30, - "serialized_chars": 36630, - "estimated_input_tokens": 9158, - "estimated_eval_tokens": 18572 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 35762, + "estimated_input_tokens": 8941, + "estimated_eval_tokens": 18138 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is heterogeneous: most candidates only share broad themes like loading, memory, or config regressions, but not the same concrete bug. I would keep them separate and use the most central memory/offload regression as the representative issue.", - "confidence": 0.91, - "canonical_issue_reason": "Issue 43873 is the most discussion-heavy and broadly phrased memory/offloading regression in the set, so it is the best representative if one issue must stand in for the cluster.", + "summary": "This is a mixed-topic similarity cluster, not a true duplicate set: it spans tokenizer/config API regressions, OOM/quantization issues, processor bugs, and unrelated model-loading failures. None of the soft pairs look like the same underlying bug/change closely enough to merge as duplicates. If one issue must serve as the anchor, the broad, high-activity quantization/offloading regression is the best representative.", + "confidence": 0.89, + "canonical_issue_reason": "Issue 43873 is the broadest and most active open thread in this set, with multiple inbound references and a large discussion around quantization/offloading memory behavior, so it is the best anchor if a single issue must represent the cluster.", "canonical_pr_reason": null, - "best_issue_reason": "43873 is the strongest global issue candidate because it has the most activity and sits closest to the recurring quantization/offloading/memory theme; however, the rest of the set is too diverse for true deduping.", + "best_issue_reason": "43873 is the most suitable representative issue because it is current, high-signal, and covers the widest user-facing regression surface in the cluster; however, it is not a duplicate of the other items.", "best_pr_reason": null, "soft_edge_verdicts": [ - { - "left": "issue:43761", - "right": "issue:45859", - "accept": false, - "reason": "Different bugs: CLIP hidden-states regression vs Qwen3 MoE CPU-offload incompatibility." - }, { "left": "issue:44442", "right": "issue:45072", "accept": false, - "reason": "Tokenizer loading failure vs bfloat16 dtype mismatch in inference; unrelated code paths." + "reason": "Tokenizer loading failure vs dtype-mismatch inference bug; different code paths and symptoms." }, { "left": "issue:43575", "right": "issue:43701", "accept": false, - "reason": "OOM while loading a large model with TP vs resume_from_checkpoint key mismatch." + "reason": "OOM during tensor-parallel model load vs resume_from_checkpoint key mismatch; unrelated bugs." }, { "left": "issue:42915", "right": "issue:43931", "accept": false, - "reason": "FineGrainedFP8Config failure vs Qwen3-VL weight-shape mismatch; different loading problems." + "reason": "FP8 config failure in Qwen3Moe vs Qwen3-VL weight-shape mismatch; different model families and failure modes." }, { "left": "issue:42490", "right": "issue:43381", "accept": false, - "reason": "Local-path import warning vs gradient-checkpointing-in-eval restriction; unrelated." + "reason": "Local-path import warning vs gradient-checkpointing eval-mode restriction; no shared code-path." }, { "left": "issue:43575", "right": "issue:44387", "accept": false, - "reason": "Both involve memory, but one is TP load OOM and the other is int4 reserved-memory regression." + "reason": "Both mention OOM, but one is TP load OOM and the other is int4 quantization memory growth; not the same bug." }, { "left": "issue:43526", "right": "issue:44857", "accept": false, - "reason": "ImageProcessor label reduction bug vs AMP/CUDA crash in loss computation." + "reason": "BeitImageProcessorFast label reduction bug vs LwDetrImageLoss AMP crash; unrelated subsystems." }, { - "left": "issue:44977", - "right": "issue:45245", + "left": "issue:43334", + "right": "issue:43531", "accept": false, - "reason": "Flash-attention generation bug vs category-count limit error; different failures." + "reason": "Missing pad_token_id on Qwen3-VL config vs sliding_window recurrent-state reset in Qwen3-MoE; different defects." }, { - "left": "issue:43334", - "right": "issue:43531", + "left": "issue:44977", + "right": "issue:45245", "accept": false, - "reason": "Missing pad_token_id on Qwen3-VL config vs sliding_window behavior in Qwen3-MoE." + "reason": "Flash-attention generation issue vs category-cardinality limit; not the same concrete bug." }, { "left": "issue:43262", "right": "issue:44206", "accept": false, - "reason": "Chat-template sample-rate default vs unsupported center arg in feature extraction." + "reason": "Audio chat-template sampling-rate default vs unsupported 'center' arg in feature extractor; different processor paths." }, { "left": "issue:44464", "right": "issue:45491", "accept": false, - "reason": "Compiled-forward chunked generation inconsistency vs NaNs from all-padding sliding-window attention." + "reason": "Compiled-forward chunked generation inconsistency vs NaNs from mixed-length sliding-window padding; different failure mechanisms." }, { "left": "issue:27957", "right": "issue:29127", "accept": false, - "reason": "Flash Attention 2 support request vs a LayoutLMv3 error-message clarity issue." + "reason": "FlashAttention 2 support question vs LayoutLMv3 error-message clarity; unrelated issues." }, { "left": "issue:43452", "right": "issue:45588", "accept": false, - "reason": "Tokenizer/model loading via gguf_file vs flash_attention sinkless AttributeError; different subsystems." + "reason": "gguf_file from_pretrained breakage vs flash_attention s_aux=None crash; different loaders and code paths." }, { "left": "issue:44743", "right": "issue:44811", "accept": false, - "reason": "Recurrent-state reset bug vs Whisper batch_decode skip_special_tokens handling." + "reason": "Cache/recurrent-state reset bug vs Whisper batch_decode skip_special_tokens bug; unrelated functionality." }, { "left": "issue:27957", "right": "issue:33290", "accept": false, - "reason": "Flash-attention model support vs deepspeed Adafactor OOM; not the same bug." + "reason": "Model/attention support request vs deepspeed Adafactor OOM; no shared underlying bug." }, { "left": "issue:43646", "right": "issue:44077", "accept": false, - "reason": "General custom-init regression vs patchtsmixer-specific post_init policy change." + "reason": "Broad v5 custom-init regression vs patchtsmixer-specific post_init policy change; too different and too model-specific." }, { - "left": "issue:44062", - "right": "issue:45907", + "left": "issue:33290", + "right": "issue:34689", "accept": false, - "reason": "AddedToken special-arg duplication vs list_repo_files revision forwarding." + "reason": "Optimizer OOM in deepspeed vs Llama 3.2 Vision loading breakage; unrelated failures." }, { - "left": "issue:33290", - "right": "issue:34689", + "left": "issue:44062", + "right": "issue:45907", "accept": false, - "reason": "Adafactor/deepspeed OOM vs Llama 3.2 Vision loading breakage; different root causes." + "reason": "AddedToken special-arg duplication vs list_repo_files revision forwarding; unrelated APIs." }, { "left": "issue:33290", "right": "issue:34567", "accept": false, - "reason": "Memory/OOM issue vs TrainerState token-count update bug." + "reason": "OOM in training/optimizer stack vs trainer token-count tracking not updating; different code paths." }, { "left": "issue:33290", "right": "issue:35141", "accept": false, - "reason": "OOM in deepspeed vs embedding reinitialization after resize_token_embeddings." + "reason": "OOM vs embedding reinitialization after resize; unrelated bugs." }, { "left": "issue:43381", "right": "issue:43873", "accept": false, - "reason": "Eval-mode checkpointing restriction vs quantization/offloading memory regression; related area but different bug." + "reason": "Gradient checkpointing eval-mode restriction vs quantization/offloading memory regression; different concrete problems." }, { "left": "issue:44442", "right": "issue:44898", "accept": false, - "reason": "Tokenizer load failure vs Perceiver image-classification interpolation failure." + "reason": "Tokenizer load failure vs Perceiver interpolation/non-default resolution failure; different model classes and symptoms." }, { "left": "issue:45003", "right": "issue:45440", "accept": false, - "reason": "Unsafe sys.modules access vs DeepseekV3 implementation divergence." + "reason": "Unsafe sys.modules access vs DeepSeekV3 implementation divergence; unrelated implementation issues." }, { "left": "issue:43723", "right": "issue:45479", "accept": false, - "reason": "Tokenizer loading in v5 vs degenerate zero-loss classification behavior." + "reason": "AutoTokenizer loading issue in v5 vs zero-loss sequence-classification regression; different failures." }, { "left": "issue:43761", "right": "issue:44079", "accept": false, - "reason": "CLIP hidden_states regression vs ModelOutput key assignment bug." - }, - { - "left": "issue:42490", - "right": "issue:44743", - "accept": false, - "reason": "Trailing-slash import warning vs recurrent-state reset in cached generation." - }, - { - "left": "issue:43316", - "right": "issue:43525", - "accept": false, - "reason": "Gemma3TextConfig API discrepancy vs Llama4Config missing pad_token_id; similar family, different attribute." + "reason": "CLIP hidden_states regression vs ModelOutput key assignment bug; not the same code-path." }, { "left": "issue:42175", "right": "issue:46234", "accept": false, - "reason": "Package extra/backend dependency issue vs missing import_utils symbol." + "reason": "Packaging/backends selection issue vs missing import_utils symbol; unrelated." }, { "left": "issue:34689", "right": "issue:43611", "accept": false, - "reason": "Model-loading regression for a specific model vs base_model_prefix-related v5 loading breakage." + "reason": "Specific model-loading failure vs base_model_prefix loading regression; same broad area, but not the same bug." }, { "left": "issue:43720", "right": "issue:44442", "accept": false, - "reason": "BitNet packed-weight unpacking bug vs tokenizer load failure; unrelated." + "reason": "Accelerate loading/unpacking bug for BitNet vs tokenizer loading failure; unrelated." + }, + { + "left": "issue:44743", + "right": "issue:44811", + "accept": false, + "reason": "Recurrent-state reset bug vs Whisper decoding option bug; unrelated." + }, + { + "left": "issue:43316", + "right": "issue:43525", + "accept": false, + "reason": "Both touch config API shape, but one is a broad Gemma3TextConfig discrepancy and the other is a specific Llama4 pad_token_id AttributeError; not clearly the same underlying bug." + }, + { + "left": "issue:44077", + "right": "issue:44079", + "accept": false, + "reason": "PatchTSMixer post_init policy issue vs ModelOutput key handling bug; unrelated." + }, + { + "left": "issue:42175", + "right": "issue:46234", + "accept": false, + "reason": "Package install backend omission vs missing utility export; different issues." } ] }, @@ -9726,8 +9667,6 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-10000-450", "nodes": [ - "issue:16998", - "issue:22355", "issue:25251", "issue:28282", "issue:42617", @@ -9744,6 +9683,7 @@ "issue:43643", "issue:43650", "issue:43701", + "issue:43742", "issue:43749", "issue:43761", "issue:43825", @@ -9777,216 +9717,216 @@ ], "soft_pairs": [ "issue:43232|issue:43388", - "issue:43931|issue:45216", "issue:43761|issue:44188", + "issue:43931|issue:45216", "issue:44077|issue:44479", - "issue:44464|issue:46421", "issue:44987|issue:46178", + "issue:44464|issue:46421", "issue:43493|issue:43827", "issue:44442|issue:45803", - "issue:42907|issue:45698", "issue:44479|issue:45290", "issue:43873|issue:44871", + "issue:42907|issue:45698", "issue:43749|issue:43931", "issue:43643|issue:44913", "issue:43316|issue:45759", - "issue:44190|issue:44496", "issue:44855|issue:45137", + "issue:44190|issue:44496", "issue:43650|issue:46202", "issue:45632|issue:46424", "issue:45803|issue:46365", "issue:25251|issue:28282", "issue:43295|issue:43825", "issue:43550|issue:43701", - "issue:43606|issue:43701", "issue:42617|issue:43575", + "issue:43606|issue:43701", "issue:43065|issue:44617", - "issue:16998|issue:22355" + "issue:43575|issue:43742" ], - "prepared_review_unit_hash": "2db86bf115ed068c7ff84163fa651cf060f9139a4ce10bdb9e4db8c91398913b", + "prepared_review_unit_hash": "28cac491a22c01e0e042c293f48473138f2105430c60a11e9db35ace58732b1a", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { - "node_count": 48, - "item_count": 48, + "node_count": 47, + "item_count": 47, "soft_pair_count": 26, - "serialized_chars": 35402, - "estimated_input_tokens": 8851, - "estimated_eval_tokens": 17958 + "serialized_chars": 34877, + "estimated_input_tokens": 8720, + "estimated_eval_tokens": 17696 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose basket of unrelated Transformers issues; none of the soft pairs look like true duplicates. The closest matches are still different bug classes or different code paths.", - "confidence": 0.95, - "canonical_issue_reason": "If one issue must serve as the cluster anchor, issue 43873 is the broadest and most central by activity/inbound refs, but it is not a duplicate of the others.", + "summary": "The cluster is heterogeneous: the soft pairs mostly connect unrelated issues that only share broad Transformers vocabulary (generation, loading, config, caching, offload, etc.). I found no true duplicate pair, so the safest interpretation is that this is not a real duplicate cluster.", + "confidence": 0.67, + "canonical_issue_reason": "Issue 45698 is the strongest standalone issue in this set: it is recent, active, open, and describes a concrete custom-module save/load regression with broad user impact. That said, the cluster itself is too mixed to indicate a true canonical duplicate.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 43873 is the best global issue anchor because it is active, broadly scoped, and has the strongest discussion/reference signal in this set.", + "best_issue_reason": "Issue 45698 is the best overall issue candidate for triage because it is the most actionable and currently active among the listed issues; however, it should not be treated as representative of the whole cluster beyond that.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43232", "right": "issue:43388", "accept": false, - "reason": "Generation kwarg syncing after sync_gpus vs metric gathering truncating tuple labels; different subsystems and failure modes." + "reason": "Generation kwargs handling after sync_gpus and gather_for_metrics label truncation are different bugs in different code paths." }, { - "left": "issue:43931", - "right": "issue:45216", + "left": "issue:43761", + "right": "issue:44188", "accept": false, - "reason": "Model weight-shape mismatch on load vs save_pretrained regression for Qwen3.5 checkpoints; related family, different bug." + "reason": "CLIP hidden_states regression and compile-time attention-kernel divergence are unrelated model/runtime problems." }, { - "left": "issue:43761", - "right": "issue:44188", + "left": "issue:43931", + "right": "issue:45216", "accept": false, - "reason": "CLIPVision hidden_states regression vs attention-kernel divergence under torch.compile; unrelated code paths." + "reason": "A Qwen3-VL shape-mismatch load failure is not the same as a Qwen3.5 save_pretrained checkpoint regression." }, { "left": "issue:44077", "right": "issue:44479", "accept": false, - "reason": "Optional post_init policy for patchtsmixer vs video-input regression in Qwen VL models; no shared concrete defect." + "reason": "PatchTSMixer post_init validation and Qwen video-input regression are different model-level issues." }, { - "left": "issue:44464", - "right": "issue:46421", + "left": "issue:44987", + "right": "issue:46178", "accept": false, - "reason": "Both mention chunked/compiled behavior, but one is wrong outputs and the other is spurious recompilation from lazy cache init." + "reason": "Loading a specific model failing under transformers>=5.1 and a backend mapping error for tensorflow_text are unrelated." }, { - "left": "issue:44987", - "right": "issue:46178", + "left": "issue:44464", + "right": "issue:46421", "accept": false, - "reason": "Loading a specific model under transformers>=5.1 vs backend mapping missing tensorflow_text; unrelated failures." + "reason": "Chunked generation inconsistency with compiled forward and cache lazy-init recompilation are only loosely related by compilation." }, { "left": "issue:43493", "right": "issue:43827", "accept": false, - "reason": "SigLIP2 implementation mismatch vs docs still referencing removed pipeline APIs; one is model parity, the other documentation." + "reason": "SigLIP2 implementation drift and v5 docs still referencing pipeline() are different product areas." }, { "left": "issue:44442", "right": "issue:45803", "accept": false, - "reason": "Tokenizer load failure for FastSpeech2ConformerTokenizer vs bare except swallowing errors in Fuyu batch conversion; unrelated bugs." - }, - { - "left": "issue:42907", - "right": "issue:45698", - "accept": false, - "reason": "Save/dequantization problem for specific models vs wrong custom module loaded after save/load; both custom-model related but not the same bug." + "reason": "AutoTokenizer failing on a custom tokenizer and a bare-except bug in Fuyu tensor conversion are distinct issues." }, { "left": "issue:44479", "right": "issue:45290", "accept": false, - "reason": "Video-input regression for several Qwen VL models vs chat-template crash on tool-call assistant messages; different APIs and code paths." + "reason": "A video-input regression and a chat-template crash on tool-call messages do not share the same root cause." }, { "left": "issue:43873", "right": "issue:44871", "accept": false, - "reason": "Quantization/offloading behavior vs Gemma eos_token_id config inconsistency; no duplicate evidence." + "reason": "Quantization/offloading behavior and Gemma eos_token_id config mismatch are unrelated." + }, + { + "left": "issue:42907", + "right": "issue:45698", + "accept": false, + "reason": "Saving dequantized model artifacts and loading the wrong custom module after save_pretrained are separate save/load failures." }, { "left": "issue:43749", "right": "issue:43931", "accept": false, - "reason": "Broken FSDP CPU RAM-efficient loading vs a model weight-shape mismatch at load time; different root causes." + "reason": "Broken FSDP CPU RAM efficient loading and a Qwen3-VL weight-shape mismatch are different loading paths and failures." }, { "left": "issue:43643", "right": "issue:44913", "accept": false, - "reason": "Missing fields from trust_remote_code AutoConfig result vs rotary_pct not persisting on reload; both config-related but distinct." + "reason": "Missing fields from trust_remote_code AutoConfig and rotary_pct not persisting on reload are different config serialization bugs." }, { "left": "issue:43316", "right": "issue:45759", "accept": false, - "reason": "Gemma3TextConfig API mismatch vs AutoModelForCausalLM.from_config not unwrapping text_config for Qwen multimodal configs; different model families and fixes." + "reason": "Gemma3TextConfig API discrepancy and AutoModelForCausalLM.from_config not unwrapping text_config are related only at a broad API level, not the same bug." }, { - "left": "issue:44190", - "right": "issue:44496", + "left": "issue:44855", + "right": "issue:45137", "accept": false, - "reason": "Local dataset loading failure in an example script vs unrecognized model_type in config.json; unrelated." + "reason": "A Python 3.13 import parsing error and a DeepSpeed ZeRO3 deque IndexError are completely unrelated." }, { - "left": "issue:44855", - "right": "issue:45137", + "left": "issue:44190", + "right": "issue:44496", "accept": false, - "reason": "Python 3.13 import parsing error from torch.jit.script comment placement vs DeepSpeed ZeRO3 deque underflow; unrelated." + "reason": "Local dataset loading in an example script and an unrecognized model/config error are different failures." }, { "left": "issue:43650", "right": "issue:46202", "accept": false, - "reason": "Placeholder issue vs Gemma pipeline roundtrip failure on a specific token; clearly unrelated." + "reason": "A placeholder-data issue and a Gemma pipeline roundtrip failure are not duplicates." }, { "left": "issue:45632", "right": "issue:46424", "accept": false, - "reason": "trust_remote_code cache-path collision for local models vs cache max_cache_length only honored for quantized cache; different cache layers." + "reason": "trust_remote_code cache-path collision and max_cache_length only applying to Quantized Cache are different cache-related behaviors." }, { "left": "issue:45803", "right": "issue:46365", "accept": false, - "reason": "Bare except in Fuyu tensor conversion vs importing Trainer broken until datasets is uninstalled; unrelated." + "reason": "A bare except swallowing errors and an import conflict involving Trainer/datasets are unrelated." }, { "left": "issue:25251", "right": "issue:28282", "accept": false, - "reason": "Pipeline top_k nesting bug vs AutoModel requiring PyTorch not installed; no overlap." + "reason": "Pipeline top_k output nesting and an AutoModel ImportError due to missing PyTorch are not the same underlying problem." }, { "left": "issue:43295", "right": "issue:43825", "accept": false, - "reason": "Custom code broke around processor.tokenizer/image passing vs v5 pipeline error-message wording; different regressions." + "reason": "A processor/tokenizer regression and a pipeline v5 error-message issue are different regression symptoms." }, { "left": "issue:43550", "right": "issue:43701", "accept": false, - "reason": "torch.compile + SDPA failure in Bamba vs resume_from_checkpoint key mismatch; unrelated." + "reason": "Bamba torch.compile SDPA failure and resume_from_checkpoint key mismatch share no concrete code path." }, { - "left": "issue:43606", - "right": "issue:43701", + "left": "issue:42617", + "right": "issue:43575", "accept": false, - "reason": "CPU offload device mismatch in bark-small vs checkpoint resume key mismatch; unrelated." + "reason": "3d_parallel.py runtime failure and Qwen2-57B tensor-parallel OOM are different large-model issues." }, { - "left": "issue:42617", - "right": "issue:43575", + "left": "issue:43606", + "right": "issue:43701", "accept": false, - "reason": "3d_parallel.py execution failure vs tensor-parallel OOM when loading a large Qwen2 model; not the same bug." + "reason": "CPU-offload device mismatch and checkpoint key mismatch are unrelated." }, { "left": "issue:43065", "right": "issue:44617", "accept": false, - "reason": "Dummy Conv2d in Sam3PixelDecoder vs Sam3Video CUDA OOM; same product area, but different concrete issues." + "reason": "Dummy Conv2d in Sam3PixelDecoder and Sam3Video CUDA OOM are not the same bug." }, { - "left": "issue:16998", - "right": "issue:22355", + "left": "issue:43575", + "right": "issue:43742", "accept": false, - "reason": "Model_max_length question for DeBERTa-V3 vs missing transformers.onnx import; unrelated." + "reason": "Tensor-parallel OOM for Qwen2-57B and a MobileLLM load KeyError are different model-loading failures." } ] }, @@ -10010,6 +9950,8 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-10000-450", "nodes": [ + "issue:16998", + "issue:22355", "issue:27957", "issue:32090", "issue:41950", @@ -10023,12 +9965,10 @@ "issue:43504", "issue:43540", "issue:43572", - "issue:43575", "issue:43646", "issue:43673", "issue:43688", "issue:43716", - "issue:43742", "issue:43761", "issue:44077", "issue:44162", @@ -10059,22 +9999,22 @@ "issue:46421" ], "soft_pairs": [ - "issue:43575|issue:43742", + "issue:16998|issue:22355", "issue:43540|issue:44514", "issue:27957|issue:32090", "issue:43316|issue:43716", "issue:44610|issue:46369", - "issue:43316|issue:44877", "issue:43295|issue:44871", "issue:44871|issue:45584", + "issue:43316|issue:44877", "issue:44162|issue:44792", "issue:43404|issue:43441", "issue:44077|issue:44534", "issue:45803|issue:46170", "issue:44871|issue:46050", "issue:41950|issue:43475", - "issue:43761|issue:44479", "issue:43452|issue:44496", + "issue:43761|issue:44479", "issue:43646|issue:44913", "issue:43388|issue:43688", "issue:45127|issue:45200", @@ -10086,200 +10026,197 @@ "issue:44734|issue:45406", "issue:45542|issue:46234" ], - "prepared_review_unit_hash": "a485d0044123188b4c3718aefe92f05643deeb5f249902b847776323186df492", + "prepared_review_unit_hash": "5df06c03705811b50b6e6ece28636d1d163983b827357881ebd8113345f94ea7", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 26, - "serialized_chars": 35010, - "estimated_input_tokens": 8753, - "estimated_eval_tokens": 17762 + "serialized_chars": 34984, + "estimated_input_tokens": 8746, + "estimated_eval_tokens": 17748 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "These issues are largely unrelated across different models and subsystems. The only plausible duplicate pair is the chunked-prefill cache regression in the generation path; the rest differ in concrete code path, model, or failure mode.", - "confidence": 0.83, - "canonical_issue_reason": "issue:46421 is the best canonical representative for the chunked-prefill cache regression because it names the underlying lazy-initialization problem directly; issue:43673 reads more like the narrower symptom report (missing cache during chunked prefill).", + "summary": "This cluster is mostly a loose semantic grouping of Transformers regressions and model-specific bugs rather than true duplicates. The only safe canonical choice is a broad v5 regression issue; the soft pairs are generally too different in model, code path, or symptom to merge.", + "confidence": 0.47, + "canonical_issue_reason": "issue:43646 is the broadest representative: a general Transformers v5 breakage around custom model initialization, which matches the cluster\u2019s overall regression theme better than the narrower model-specific reports.", "canonical_pr_reason": null, - "best_issue_reason": "issue:46421 is the strongest issue candidate overall among the soft-edge set: it is specific, actionable, and captures the shared generation/cache code path more clearly than the paired report.", + "best_issue_reason": "issue:43646 is the best umbrella issue for this set because it is broad, version-level, and likely to attract related reports; however, the cluster is heterogeneous and not a strong duplicate set.", "best_pr_reason": null, "soft_edge_verdicts": [ { - "left": "issue:43575", - "right": "issue:43742", + "left": "issue:16998", + "right": "issue:22355", "accept": false, - "reason": "Both are loading failures, but they involve different models and different root causes (TP OOM vs config key error)." + "reason": "Unrelated topics: DeBERTa model_max_length behavior vs an import error for transformers.onnx." }, { "left": "issue:43540", "right": "issue:44514", "accept": false, - "reason": "Both involve Qwen multimodal/video processing, but one is a video-input processing bug and the other is a batched apply_chat_template/padding bug." + "reason": "Both involve multimodal Qwen processing, but one is video-input processing in Qwen3OmniMoe and the other is batched chat template padding in Qwen2.5-VL; different code paths and bugs." }, { "left": "issue:27957", "right": "issue:32090", "accept": false, - "reason": "Unrelated subsystems: FlashAttention support for XLM-Roberta vs Trainer GPU broadcast TypeError." + "reason": "No shared underlying bug: FlashAttention support for XLM-Roberta vs a Trainer broadcast TypeError." }, { "left": "issue:43316", "right": "issue:43716", "accept": false, - "reason": "Different models and different issues: Gemma3TextConfig API mismatch vs Mistral-3 image preprocessor dtype mismatch." + "reason": "Different model families and failures: Gemma3TextConfig API mismatch vs Mistral-3 image preprocessor dtype mismatch." }, - { - "left": "issue:44610", - "right": "issue:46369", - "accept": false, - "reason": "Both are SAM/vision-related, but one is a processor size mismatch and the other is a memory overflow; not the same bug." - }, - { - "left": "issue:43316", - "right": "issue:44877", + { + "left": "issue:44610", + "right": "issue:46369", "accept": false, - "reason": "Both are config-related, but they concern different models and different failures (API discrepancy vs strict config loading)." + "reason": "Both are model-specific vision issues, but one is an input-size mismatch in OmDet-Turbo and the other is OOM in sam3-litetext; not the same fix." }, { "left": "issue:43295", "right": "issue:44871", "accept": false, - "reason": "A regression in processor.tokenizer/image handling is unrelated to the Gemma-3 eos_token_id config inconsistency." + "reason": "Processor/tokenizer regression in one issue vs eos_token_id config inconsistency in another; only loosely related to processors/configs." }, { "left": "issue:44871", "right": "issue:45584", "accept": false, - "reason": "Different models and symptoms: Gemma-3 EOS config mismatch vs Whisper empty transcription failure." + "reason": "Different bugs: Gemma-3 eos token configuration vs Whisper empty-transcription generation after special-token alignment." + }, + { + "left": "issue:43316", + "right": "issue:44877", + "accept": false, + "reason": "Both concern config loading, but Gemma3TextConfig API discrepancy and granite_speech strict-config loading failure are distinct issues." }, { "left": "issue:44162", "right": "issue:44792", "accept": false, - "reason": "ESM2 workflow breakage is unrelated to the janus image-generation test failure." + "reason": "Completely different models and symptoms: ESM2 breakage vs a failed janus image-generation test." }, { "left": "issue:43404", "right": "issue:43441", "accept": false, - "reason": "Same general Mistral family, but the concrete bugs differ: weight tying vs FlashAttention compatibility." + "reason": "Same broad model family, but different concrete failures: untied lm_head weights vs FlashAttention failure in Ministral-3." }, { "left": "issue:44077", "right": "issue:44534", "accept": false, - "reason": "Optional post_init on patchtsmixer is unrelated to non-persistent buffer initialization in v5." + "reason": "Optional post_init misuse in patchtsmixer is unrelated to non-persistent buffer junk values." }, { "left": "issue:45803", "right": "issue:46170", "accept": false, - "reason": "Different bugs and platforms: bare except in Fuyu tensor conversion vs Windows/ROCm torch.distributed import failure." + "reason": "Bare except swallowing interrupts in Fuyu conversion is unrelated to a Windows/ROCm torch.distributed import/type error." }, { "left": "issue:44871", "right": "issue:46050", "accept": false, - "reason": "Gemma-3 EOS token inconsistency is unrelated to quantized model save/load breakage after sinq." + "reason": "Different save/load regressions: Gemma-3 eos token mismatch vs sinq quantized model breaking after save_pretrained/from_pretrained." }, { "left": "issue:41950", "right": "issue:43475", "accept": false, - "reason": "Video-classification processor lookup and SAM3 video missing attribute are different code paths and failures." + "reason": "Both are video-adjacent, but one is pipeline processor selection and the other is a missing attribute in Sam3 video encoder output." }, { - "left": "issue:43761", - "right": "issue:44479", + "left": "issue:43452", + "right": "issue:44496", "accept": false, - "reason": "CLIPVision hidden_states regression is unrelated to Qwen VL video-input regression." + "reason": "Both are from_pretrained/config loading problems, but one is gguf/custom class loading and the other is an unrecognized model_type in config.json." }, { - "left": "issue:43452", - "right": "issue:44496", + "left": "issue:43761", + "right": "issue:44479", "accept": false, - "reason": "Both are model-loading issues, but one is gguf/from_pretrained handling and the other is a missing model_type in config.json." + "reason": "Different regressions: CLIPVisionModel hidden_states handling vs Qwen video-input regression across multiple model variants." }, { "left": "issue:43646", "right": "issue:44913", "accept": false, - "reason": "Transformers 5 custom model initialization breakage is broader and different from GPTNeoXConfig rotary_pct not persisting on reload." + "reason": "General custom model initialization breakage vs a specific GPTNeoXConfig rotary_pct persistence bug; not the same underlying defect." }, { "left": "issue:43388", "right": "issue:43688", "accept": false, - "reason": "Different training/eval bugs: gather_for_metrics tuple truncation vs auxiliary-loss normalization." + "reason": "Training metric gathering dropping tuple labels is unrelated to auxiliary-loss normalization." }, { "left": "issue:45127", "right": "issue:45200", "accept": false, - "reason": "LoRA merge collapse with tied embeddings is unrelated to Gemma 4 mm_token_type_ids defaults." + "reason": "LoRA merge with extended vocab collapsing models is unrelated to Gemma 4 mm_token_type_ids defaults." }, { "left": "issue:43673", "right": "issue:44743", "accept": false, - "reason": "Both mention cache/recurrent state, but one is a GenerationMixin chunked-prefill cache initialization issue and the other is a model-specific recurrent-state reset bug." + "reason": "Both mention cache/recurrent state, but one is missing cache in chunked_prefill and the other is recurrent states resetting in a specific modular Qwen3.5 file; different bugs." }, { "left": "issue:43572", "right": "issue:46178", "accept": false, - "reason": "StableLm pad_token_idx regression is unrelated to the tensorflow_text backend registration error." + "reason": "StableLmConfig pad_token_idx regression is unrelated to tensorflow_text backend mapping errors." }, { "left": "issue:44617", "right": "issue:45584", "accept": false, - "reason": "Sam3Video OOM and Whisper empty transcription failure are unrelated symptoms in different models." + "reason": "CUDA OOM in Sam3Video is unrelated to Whisper empty-transcription generation behavior." }, { "left": "issue:43504", "right": "issue:45698", "accept": false, - "reason": "Legacy-field preset loading and wrong custom-module selection after save_pretrained are both load-time issues, but they are different bugs." + "reason": "Legacy field blocks Beit preset loading vs wrong custom module loaded after save_pretrained; both are loading issues but not the same code path." }, { "left": "issue:43673", "right": "issue:46421", - "accept": true, - "reason": "Both describe the same chunked-prefill cache initialization regression; one reports a missing cache, the other the lazy-init/recompilation consequence, and one fix could plausibly address both." + "accept": false, + "reason": "Both involve chunked prefill caching, but one is a missing cache in v5.0.0 and the other is lazy-init causing spurious recompilation; related area, different bug." }, { "left": "issue:44734", "right": "issue:45406", "accept": false, - "reason": "KV-cache continuation tensor indexing and Gemma4Processor missing _tokenizer are unrelated serving bugs." + "reason": "KV-cache continuation tensor indexing crash in serve is unrelated to Gemma4Processor missing _tokenizer." }, { "left": "issue:45542", "right": "issue:46234", "accept": false, - "reason": "TensorBoard/TensorFlow backend detection and missing is_falcon_mamba_ssm_available are unrelated import-utils issues." + "reason": "TensorBoard/TensorFlow backend installation issue is unrelated to a missing import utility helper for falcon mamba ssm." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet and stays conservative on the soft-edge pairs. The only accepted duplicate pair, issue:43673 and issue:46421, is well supported by the shared chunked-prefill cache regression theme; the rest of the false verdicts are reasonably differentiated by model, subsystem, or failure mode. " - }, - "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 }, @@ -10347,14 +10284,14 @@ ], "soft_pairs": [ "issue:43525|issue:43531", - "issue:45381|issue:46369", "issue:43761|issue:44242", + "issue:45381|issue:46369", "issue:44792|issue:44977", "issue:45563|issue:45584", "issue:43931|issue:44060", "issue:44792|issue:45600", - "issue:44190|issue:44964", "issue:44534|issue:44913", + "issue:44190|issue:44964", "issue:44479|issue:44734", "issue:35141|issue:42490", "issue:44855|issue:45588", @@ -10375,202 +10312,202 @@ "issue:25251|issue:27957", "issue:45632|issue:46234" ], - "prepared_review_unit_hash": "756b80033388320a3318424fd8e44e8c06af536ad67239cd022469c56843969b", + "prepared_review_unit_hash": "805253767e368899cb6e8d0a4ce2db46a9761e97a7d4eef18d4aba13ee2aee88", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 28, - "serialized_chars": 35901, - "estimated_input_tokens": 8976, - "estimated_eval_tokens": 18208 + "serialized_chars": 35893, + "estimated_input_tokens": 8974, + "estimated_eval_tokens": 18204 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This set is a mixed bag of unrelated Transformers bugs and regressions. None of the soft-similarity pairs look like the same underlying issue/change, so no duplicates should be merged here. The most canonical standalone issue is the multimodal chat-template assistant-mask regression, which has clear repros and inbound references.", - "confidence": 0.84, - "canonical_issue_reason": "issue:44521 is the strongest standalone bug report: it is current, well-scoped, has multiple inbound references, and describes a concrete multimodal regression with clear user impact.", + "summary": "The items are mostly unrelated Transformers bug reports rather than one duplicate cluster. None of the soft pairs describe the same underlying bug or change closely enough to merge; they usually differ by model, code path, or bug type.", + "confidence": 0.93, + "canonical_issue_reason": "Issue 44521 is the best representative issue in this set: it is open, has the strongest cross-references, and describes a concrete multimodal regression with clear downstream impact.", "canonical_pr_reason": null, - "best_issue_reason": "issue:44521 is the best representative issue because it has the clearest bug description, the highest apparent external interest, and is actionable without needing broader context.", + "best_issue_reason": "Issue 44521 stands out as the strongest standalone bug report in the group, with multiple inbound references and active discussion, making it the most useful anchor issue here.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43525", "right": "issue:43531", "accept": false, - "reason": "Both involve Qwen3-family models, but one is a missing `pad_token_id` attribute and the other is a `sliding_window` behavior problem; different failure modes and code paths." + "reason": "Different models and different failure modes: missing pad_token_id on Llama4Config vs sliding_window behavior in Qwen3-MoE." }, { - "left": "issue:45381", - "right": "issue:46369", + "left": "issue:43761", + "right": "issue:44242", "accept": false, - "reason": "Qwen2.5-VL video `vision_position_ids` corruption and Sam3-litetext memory overflow are unrelated multimodal issues with different symptoms and fixes." + "reason": "Unrelated regressions: CLIPVision hidden_states handling vs MoE load-balancing loss when router logits are disabled." }, { - "left": "issue:43761", - "right": "issue:44242", + "left": "issue:45381", + "right": "issue:46369", "accept": false, - "reason": "CLIPVision hidden-states regression and MoE load-balancing loss omission are distinct model behaviors, not the same bug." + "reason": "Both are model-specific multimodal bugs, but one is a qwen2.5-vl video position-id regression and the other is Sam3-litetext memory overflow." }, { "left": "issue:44792", "right": "issue:44977", "accept": false, - "reason": "A janus image-generation test failure and Qwen3.5 flash-attention generation failure are different model-specific issues." + "reason": "Different code paths and symptoms: Janus image-generation test failure vs Qwen3.5 flash-attention generation issue." }, { "left": "issue:45563", "right": "issue:45584", "accept": false, - "reason": "A stale `num_return_sequences` warning in paged generate is unrelated to Whisper empty-transcription failure after token alignment." + "reason": "A stale warning about num_return_sequences is not the same as Whisper failing on empty transcription after token alignment." }, { "left": "issue:43931", "right": "issue:44060", "accept": false, - "reason": "Shape mismatch when loading Qwen3-VL weights and an incorrect tied-weights warning in Qwen3-Next are separate loading/config bugs." + "reason": "One is a weight-shape mismatch on load, the other is an incorrect tied-weights warning involving a different Qwen3-Next parameter." }, { "left": "issue:44792", "right": "issue:45600", "accept": false, - "reason": "A janus test failure and broken `auto_mappings.py` references to removed configs are not the same underlying problem." + "reason": "A failing Janus generation test and a broken auto_mappings reference to removed Sam3LiteText configs are distinct issues." }, { - "left": "issue:44190", - "right": "issue:44964", + "left": "issue:44534", + "right": "issue:44913", "accept": false, - "reason": "Local dataset loading in an example script and model loading for Phi-4 multimodal instruct are unrelated." + "reason": "Generic non-persistent buffer initialization bug vs a GPTNeoX config field not persisting on reload." }, { - "left": "issue:44534", - "right": "issue:44913", + "left": "issue:44190", + "right": "issue:44964", "accept": false, - "reason": "Non-persistent buffer junk initialization and GPTNeoX `rotary_pct` reset on reload affect different subsystems." + "reason": "Local dataset loading in a training script is unrelated to loading a specific multimodal model with latest transformers." }, { "left": "issue:44479", "right": "issue:44734", "accept": false, - "reason": "Both are generation-related, but one is a video-input regression in Qwen multimodal models and the other is a KV-cache continuation indexing crash in `transformers serve`." + "reason": "Video-input regression in Qwen multimodal models vs KV-cache continuation crash in transformers serve are different failures." }, { "left": "issue:35141", "right": "issue:42490", "accept": false, - "reason": "Output embedding reinitialization after resize and import warnings from trailing-slash local paths are unrelated bugs." + "reason": "Embedding reinitialization during post_init is unrelated to the import warning caused by a trailing slash in a local model path." }, { "left": "issue:44855", "right": "issue:45588", "accept": false, - "reason": "A Python 3.13 `torch.jit.script` parsing issue and a flash-attention `AttributeError` on `s_aux=None` are different failure modes." + "reason": "Python 3.13 import parsing failure from torch.jit.script formatting is unrelated to flash_attention.py crashing on s_aux=None." }, { "left": "issue:44617", "right": "issue:46369", "accept": false, - "reason": "Sam3Video OOM and Sam3-litetext memory overflow are both memory issues, but they concern different models and likely different causes." + "reason": "Both mention memory pressure, but they affect different models and likely different root causes: Sam3Video OOM vs Sam3-litetext memory overflow." }, { "left": "issue:43827", "right": "issue:43994", "accept": false, - "reason": "Docs still referencing removed `pipeline()` APIs is not the same as SigLIP2 producing nonsensical outputs through AutoModel/pipeline." + "reason": "Docs still referencing pipeline() after removals is not the same as a SigLIP2 pipeline producing nonsensical outputs." }, { "left": "issue:41720", "right": "issue:45245", "accept": false, - "reason": "Qwen3 auto device mapping CUDA assert and the 2^24 category limit error are unrelated problems." + "reason": "CUDA assert during Qwen3 auto device mapping is unrelated to the 2^24 category limit error." }, { "left": "issue:43065", "right": "issue:44610", "accept": false, - "reason": "A dummy `nn.Conv2d` in Sam3PixelDecoder and a processor/model input-size mismatch are different issues, even within Sam3." + "reason": "A dummy Conv2d in Sam3PixelDecoder is a model-internal placeholder issue, not the OmDet-Turbo processor/model input-size mismatch." }, { "left": "issue:44521", "right": "issue:45381", "accept": false, - "reason": "Both are multimodal regressions, but assistant-mask generation and video `vision_position_ids` are different pipeline stages and fixes." + "reason": "Both are multimodal, but one is all-zero assistant masks in apply_chat_template and the other is incorrect vision_position_ids in qwen2.5-vl video input processing." }, { "left": "issue:44464", "right": "issue:45563", "accept": false, - "reason": "Compiled-forward chunked generation inconsistency and a stale warning about `num_return_sequences` are not the same bug." + "reason": "Compiled-forward chunked generation inconsistency is a different bug from a stale generation warning." }, { "left": "issue:44933", "right": "issue:45362", "accept": false, - "reason": "A nonexistent import from `image_utils` and a Qwen3.5 chat crash are unrelated." + "reason": "A missing import in image_utils is unrelated to Qwen3.5-35B crashing in chat generation." }, { "left": "issue:43644", "right": "issue:44913", "accept": false, - "reason": "Global non-persistent buffer initialization junk and config reload losing `rotary_pct` are different classes of regression." + "reason": "A broad v5 buffer initialization regression is not the same as GPTNeoX rotary_pct being lost on config reload." }, { "left": "issue:43976", "right": "issue:44855", "accept": false, - "reason": "Python-version compatibility for Transformers 5.1.0 and a Python 3.13 decorator/comment parsing failure are not the same issue." + "reason": "General Python version incompatibility does not match a specific Python 3.13 IndentationError caused by a decorator/comment parsing edge case." }, { "left": "issue:43550", "right": "issue:44190", "accept": false, - "reason": "Bamba torch.compile/SDPA failure and local dataset loading in an example script are unrelated." + "reason": "torch.compile/SDPA failure in Bamba-9B-v2 is unrelated to local dataset loading in an image-classification training script." }, { "left": "issue:44077", "right": "issue:44877", "accept": false, - "reason": "Optional `post_init` on patchtsmixer and strict config rejection for granite_speech are distinct config/model-loading concerns." + "reason": "Optional post_init in patchtsmixer and strict config blocking granite_speech are different model/config validation issues." }, { "left": "issue:43540", "right": "issue:44734", "accept": false, - "reason": "Qwen3OmniMoe video-input validation and serve KV-cache indexing are different code paths." + "reason": "Processing video inputs in Qwen3OmniMoe is a different problem from server-side KV-cache continuation indexing." }, { "left": "issue:45127", "right": "issue:45357", "accept": false, - "reason": "LoRA merge collapse with extended vocab and incorrect visual encoder keys on save are separate serialization/training issues." + "reason": "LoRA merge collapse with extended vocab and incorrect visual encoder keys in save_pretrained are separate serialization/training bugs." }, { "left": "issue:44462", "right": "issue:45907", "accept": false, - "reason": "AutoTokenizer ignoring repository `tokenizer.json` and `list_repo_files` not forwarding `revision` are unrelated Hub behaviors." + "reason": "AutoTokenizer ignoring tokenizer.json is unrelated to list_repo_files failing to forward revision in CI." }, { "left": "issue:25251", "right": "issue:27957", "accept": false, - "reason": "Pipeline `top_k` nesting behavior and XLMRoberta Flash Attention 2 support are not the same bug or change." + "reason": "Pipeline top_k nesting behavior and XLMRoberta Flash Attention 2 support are unrelated." }, { "left": "issue:45632", "right": "issue:46234", "accept": false, - "reason": "A local-model cache path collision and a missing `is_falcon_mamba_ssm_available` import utility are unrelated." + "reason": "A cache-path collision for trust_remote_code is unrelated to a missing import-utils helper function." } ] }, @@ -10645,8 +10582,8 @@ ], "soft_pairs": [ "issue:44112|issue:45600", - "issue:44737|issue:45803", "issue:43335|issue:44190", + "issue:44737|issue:45803", "issue:43116|issue:44190", "issue:45276|issue:45584", "issue:43295|issue:45276", @@ -10674,208 +10611,208 @@ "issue:27957|issue:35141", "issue:43976|issue:44987" ], - "prepared_review_unit_hash": "bf4877cbe6ec624cd2224d7240f26e0682e6b132101ff1823774bf3d1a0edf95", + "prepared_review_unit_hash": "03a358bc17d297e717c933020a0bac8d2adf2eacf61b079c5f96688abbc1ccd0", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 36513, - "estimated_input_tokens": 9129, - "estimated_eval_tokens": 18514 + "serialized_chars": 36512, + "estimated_input_tokens": 9128, + "estimated_eval_tokens": 18512 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is heterogeneous: it mixes unrelated issues across model configs, import/export errors, cache behavior, pipelines, notebooks, and CI/test failures. None of the soft pairs look like the same underlying bug or change, so there is no meaningful canonical duplicate here.", - "confidence": 0.93, - "canonical_issue_reason": "No single canonical issue fits this cluster; the items are not duplicates and span unrelated subsystems.", + "summary": "The cluster is mostly heterogeneous: a few weakly similar import/cache/config issues, but no pair looks like the same underlying bug or change. I\u2019d use the resize-token-embeddings issue as the best representative of the only clearly coherent bug family here.", + "confidence": 0.9, + "canonical_issue_reason": "issue:35141 is the clearest, most general statement of a concrete bug in the resize_token_embeddings/post_init path; it best represents the only strong bug family in this set.", "canonical_pr_reason": null, - "best_issue_reason": "No single best issue is suitable because the cluster does not represent one bug family or one fixable change.", + "best_issue_reason": "issue:35141 is the most broadly applicable and well-scoped issue among the candidates, with a concrete reproducible failure and a model-agnostic fix path.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44112", "right": "issue:45600", "accept": false, - "reason": "Both mention CI-like failures, but one is a GraniteSpeech test issue and the other is a broken auto-mapping reference for Sam3LiteText; different bugs." + "reason": "Different failures: GraniteSpeech CI device-override test vs removed Sam3LiteText auto-mapping references." }, { - "left": "issue:44737", - "right": "issue:45803", + "left": "issue:43335", + "right": "issue:44190", "accept": false, - "reason": "XLNet CPU device placement and a bare-except bug in Fuyu tensor conversion are unrelated code paths." + "reason": "Unrelated code paths: SwitchTransformers sparse-layer config vs image-classification script local dataset loading." }, { - "left": "issue:43335", - "right": "issue:44190", + "left": "issue:44737", + "right": "issue:45803", "accept": false, - "reason": "SwitchTransformers config construction and an image-classification script failing on local datasets are different problems." + "reason": "Different bugs: XLNet CPU tensor allocation vs FuyuBatchFeature bare except swallowing errors." }, { "left": "issue:43116", "right": "issue:44190", "accept": false, - "reason": "Both are example/script bugs, but one is multi-label classification output handling and the other is local dataset loading in a different script." + "reason": "Both touch example/pipeline scripts, but one is multilabel classification output handling and the other is local dataset loading for image classification." }, { "left": "issue:45276", "right": "issue:45584", "accept": false, - "reason": "Gemma4 embedding resizing and Whisper empty transcription after token alignment are unrelated model behaviors." + "reason": "Unrelated model behaviors: Gemma4 embedding resize vs Whisper empty-transcription generation." }, { "left": "issue:43295", "right": "issue:45276", "accept": false, - "reason": "A processor/tokenizer regression and Gemma4 embedding resizing are different failures." + "reason": "Different regressions in different paths: processor/tokenizer image handling vs resize_token_embeddings output-embedding updates." }, { "left": "issue:44938", "right": "issue:45542", "accept": false, - "reason": "Python 3.14 import failure and missing TensorFlow backend under torch install are distinct dependency issues." + "reason": "Both are environment/import issues, but one is Python 3.14 load failure and the other is missing TensorFlow backend wiring." }, { "left": "issue:46082", "right": "issue:46121", "accept": false, - "reason": "Llama head_dim validation and rope-parameter list handling are unrelated config-validation bugs." + "reason": "Distinct config-validation bugs: Llama head_dim constraints vs rope-validation list typing." }, { "left": "issue:45632", "right": "issue:46421", "accept": false, - "reason": "Both involve cache, but one is a trust_remote_code path collision and the other is chunked-prefill recompilation from lazy cache init." + "reason": "Both mention cache, but one is trust_remote_code cache-path collision and the other is chunked-prefill recompilation behavior." }, { "left": "issue:45245", "right": "issue:45584", "accept": false, - "reason": "Category cardinality overflow and Whisper empty transcription are unrelated." + "reason": "Completely different symptoms and subsystems: category-limit runtime error vs Whisper empty transcription handling." }, { "left": "issue:36246", "right": "issue:41628", "accept": false, - "reason": "Both are import errors, but they concern different missing exports/classes and do not look like one shared bug." + "reason": "Both are import errors, but for different missing symbols and different modules." }, { "left": "issue:42175", "right": "issue:44279", "accept": false, - "reason": "TensorFlow backend packaging under pip extras and a generic dependency issue are not the same concrete problem." + "reason": "Dependency/backend packaging issue vs generic dependency issue; too broad and not the same defect." }, { "left": "issue:43404", "right": "issue:45216", "accept": false, - "reason": "Mistral3 lm_head tying and Qwen3.5 save_pretrained checkpoint correctness are different model-specific bugs." + "reason": "Different model-saving/training bugs: tied lm_head in Mistral3 vs Qwen3.5 checkpoint serialization." }, { "left": "issue:43650", "right": "issue:44220", "accept": false, - "reason": "A vague data request and an fbank feature extraction issue are unrelated." + "reason": "Malformed issue vs audio feature extraction bug; no shared underlying defect." }, { "left": "issue:44246", "right": "issue:45003", "accept": false, - "reason": "Slow import timing and unsafe sys.modules access are different import-related concerns, not the same bug." + "reason": "Import-time slowness vs unsafe sys.modules access are different problems in different code paths." }, { "left": "issue:36246", "right": "issue:38617", "accept": false, - "reason": "Two separate missing-import symbols from different modules; similar symptom, but not one underlying defect." + "reason": "Both are missing-export import errors, but they concern different symbols and modules." }, { "left": "issue:41950", "right": "issue:44016", "accept": false, - "reason": "A video-classification pipeline bug and a notebook syntax error are unrelated." + "reason": "Video pipeline processor-selection bug vs notebook syntax error; unrelated." }, { "left": "issue:44186", "right": "issue:44734", "accept": false, - "reason": "Tokenizer crashes on NER padding/truncation and serve KV-cache continuation indexing are different code paths." + "reason": "Tokenizer padding/NER crash vs serve KV-cache continuation indexing bug; different subsystems." }, { "left": "issue:43673", "right": "issue:46424", "accept": false, - "reason": "Both are cache-related, but one is missing cache in chunked_prefill and the other is max_cache_length handling for quantized cache only." + "reason": "Both cache-related, but one is missing GenerationMixin cache in chunked prefill and the other is quantized-cache max length handling." }, { "left": "issue:43479", "right": "issue:45375", "accept": false, - "reason": "Different multimodal config bugs: default subconfig initialization vs a missing field dropped by strict validation." + "reason": "Both are config-field issues, but they affect different models and different fields/serialization behavior." }, { "left": "issue:44683", "right": "issue:45446", "accept": false, - "reason": "Both touch flex_attention and torch version checks, but one is a compiled flex_attention failure on torch>=2.9 and the other is an AuxRequest import guard; not clearly the same fix." + "reason": "Related to flex_attention, but one is a compile failure on torch>=2.9 and the other is a version-check import guard bug." }, { "left": "issue:43643", "right": "issue:44496", "accept": false, - "reason": "trust_remote_code missing fields and unrecognized model_type are separate config-loading issues." + "reason": "Different auto-config problems: missing fields with trust_remote_code vs unrecognized model missing model_type." }, { "left": "issue:44964", "right": "issue:45600", "accept": false, - "reason": "Both are model-loading problems, but one is Phi-4-multimodal and the other is Sam3LiteText auto-mapping breakage." + "reason": "Distinct model-loading failures: Phi-4-multimodal regression vs stale Sam3LiteText auto-mapping references." }, { "left": "issue:43854", "right": "issue:44964", "accept": false, - "reason": "Different model-loading failures for different models and different failure modes." + "reason": "Different models and failure modes, both load issues but not the same bug." }, { "left": "issue:44016", "right": "issue:44220", "accept": false, - "reason": "Notebook syntax and audio feature extraction are unrelated." + "reason": "Notebook syntax error vs audio feature extraction bug; unrelated." }, { "left": "issue:44112", "right": "issue:44964", "accept": false, - "reason": "CI test flakiness in GraniteSpeech and Phi-4-multimodal loading are unrelated." + "reason": "CI test failure vs Phi-4-multimodal model loading regression; unrelated." }, { "left": "issue:46123", "right": "issue:46369", "accept": false, - "reason": "Partial-batch pipeline output dropping and memory overflow in sam3-litetext are different symptoms and code paths." + "reason": "MaskGenerationPipeline partial-batch bug vs sam3-litetext memory overflow; different symptoms and code paths." }, { "left": "issue:27957", "right": "issue:35141", "accept": false, - "reason": "Flash Attention support for XLMRoberta and embedding reinitialization after resize are unrelated bugs." + "reason": "Different areas: Flash Attention support for XLM-Roberta vs resize-token-embeddings reinitialization." }, { "left": "issue:43976", "right": "issue:44987", "accept": false, - "reason": "Python-version compatibility and loading a specific model on transformers>=5.1.0 are different issues." + "reason": "Python-version compatibility bug vs specific model load failure on 5.1.0; not the same underlying issue." } ] }, @@ -10948,11 +10885,11 @@ "issue:46424" ], "soft_pairs": [ - "issue:42907|issue:43441", "issue:43531|issue:45600", + "issue:42907|issue:43441", "issue:43526|issue:43901", - "issue:44792|issue:46424", "issue:43295|issue:45335", + "issue:44792|issue:46424", "issue:28282|issue:35141", "issue:44496|issue:45042", "issue:46082|issue:46399", @@ -10974,17 +10911,17 @@ "issue:44242|issue:45200", "issue:38175|issue:44617", "issue:44521|issue:45325", - "issue:43299|issue:45357", - "issue:44279|issue:45405" + "issue:44279|issue:45405", + "issue:43299|issue:45357" ], - "prepared_review_unit_hash": "1aa4d24ba2e5151a2a8559d0b05d52f1cad3b886e5d273a02cb6bf16ad43174b", + "prepared_review_unit_hash": "c41aab92e72b94ff3a60708eb49be9d7bc1c8bb43e2835b9fa4c031fc5e1d436", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 47, @@ -10999,186 +10936,180 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly heterogeneous and does not form one coherent duplicate set. The only strong duplicate-looking pair is the Qwen3.5 `save_pretrained` checkpoint serialization regression; the rest are distinct model/import/config/documentation issues across unrelated code paths.", - "confidence": 0.91, - "canonical_issue_reason": "Issue 45357 is the clearest statement of the underlying bug: Qwen3.5 `save_pretrained` is writing incorrect visual-encoder keys, which matches the broader checkpoint corruption report in 45216.", + "summary": "These items are mostly unrelated bug reports across different models, loaders, and utilities. None of the soft pairs look like the same underlying bug/change, so I rejected them all. No PRs are present in this cluster.", + "confidence": 0.93, + "canonical_issue_reason": "issue:44521 is the most active/open item in the set and has the strongest external signal (multiple inbound references and discussion), so it is the best representative issue for the cluster even though the cluster is not a true duplicate set.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45357 is the best representative of the only real duplicate pair because it pinpoints the concrete faulty save path and the incorrect serialized keys.", + "best_issue_reason": "issue:44521 is the best standalone issue to keep as the cluster representative because it is current, well-scoped, and has the most linkage/discussion among the listed issues.", "best_pr_reason": null, "soft_edge_verdicts": [ { - "left": "issue:42907", - "right": "issue:43441", + "left": "issue:43531", + "right": "issue:45600", "accept": false, - "reason": "Both involve Ministral/flash-attention, but one is about saving a dequantized model and the other is a runtime attention failure; different bugs." + "reason": "Different bugs: Qwen3-MoE sliding window behavior vs stale auto_mappings references. Same area of model support, but not the same code path." }, { - "left": "issue:43531", - "right": "issue:45600", + "left": "issue:42907", + "right": "issue:43441", "accept": false, - "reason": "Qwen3-MoE sliding-window behavior and a removed Sam3LiteText config reference are unrelated code paths." + "reason": "Both mention Ministral, but one is about saving dequantized weights and the other is a FlashAttention runtime failure. Not the same defect." }, { "left": "issue:43526", "right": "issue:43901", "accept": false, - "reason": "BeitImageProcessorFast label reduction and TextClassificationPipeline docs for `return_all_scores` are unrelated." + "reason": "BEIT processor label reduction bug vs docs mismatch for TextClassificationPipeline. Different subsystem and failure mode." }, { - "left": "issue:44792", - "right": "issue:46424", + "left": "issue:43295", + "right": "issue:45335", "accept": false, - "reason": "A janus image-generation test failure is not the same bug as cache-length handling for quantized cache." + "reason": "Both involve embeddings/tokenizer-related behavior, but one is a processor/tokenizer regression and the other is t5gemma decoder embedding resize. Not one underlying bug." }, { - "left": "issue:43295", - "right": "issue:45335", + "left": "issue:44792", + "right": "issue:46424", "accept": false, - "reason": "Custom processor/tokenizer handling regression and t5gemma decoder embedding resize are different failures." + "reason": "Janus image-generation test failure vs cache max length handling for quantized cache. Unrelated behaviors." }, { "left": "issue:28282", "right": "issue:35141", "accept": false, - "reason": "PyTorch import missing and embedding reinitialization on resize are completely different issues." + "reason": "AutoModel PyTorch import error vs post_init reinitializing output embeddings after resize. Different issue class and code path." }, { "left": "issue:44496", "right": "issue:45042", "accept": false, - "reason": "Unrecognized model config and incorrect torchvision dependency checks are unrelated." + "reason": "Olmo model config import failure vs PIL backend processors requiring torchvision. Separate problems." }, { "left": "issue:46082", "right": "issue:46399", "accept": false, - "reason": "LlamaConfig head_dim validation and Qwen3.5 rotary-pos-emb registration are different model/config bugs." + "reason": "LlamaConfig head_dim validation vs Qwen3.5 rotary kernel registration error. Different models and logic." }, { "left": "issue:43441", "right": "issue:45588", "accept": false, - "reason": "Both mention flash attention, but one is a Ministral model crash and the other is an `s_aux=None` sink-less-model edge case." + "reason": "Both touch FlashAttention, but one is a Ministral-specific failure and the other is an AttributeError for sink-less models in integration code. Not the same bug." }, { "left": "issue:45372", "right": "issue:46234", "accept": false, - "reason": "Missing `ReasoningEffort` import and missing `is_falcon_mamba_ssm_available` helper are unrelated import errors." + "reason": "Gemma 4 processor import breakage vs missing falcon_mamba availability helper. Unrelated imports." }, { "left": "issue:44016", "right": "issue:44062", "accept": false, - "reason": "Notebook syntax error and `AddedToken` keyword collision are unrelated." + "reason": "Notebook syntax error vs AddedToken keyword collision. No shared underlying code path." }, { "left": "issue:43299", "right": "issue:45216", "accept": false, - "reason": "Qwen3.5/Qwen3.6 model-loading failure and incorrect `save_pretrained` checkpoint output are different stages and bugs." + "reason": "Qwen3.5 loading failure vs incorrect save_pretrained checkpoint contents. Same family, but different operations and bugs." }, { "left": "issue:43901", "right": "issue:44496", "accept": false, - "reason": "Pipeline docs mismatch and model-type detection failure are not the same issue." + "reason": "Pipeline docs/behavior mismatch vs model config import error. Different issue type and scope." }, { "left": "issue:43824", "right": "issue:44933", "accept": false, - "reason": "Missing Qwen2.5-VL import and a nonexistent image_utils import are different import-path problems." + "reason": "Two import errors, but for different symbols/modules and different root causes." }, { "left": "issue:43673", "right": "issue:44336", "accept": false, - "reason": "Generation cache missing in chunked prefill and ANSI code emission in loading_report are unrelated." + "reason": "Generation cache regression vs ANSI color codes in loading_report. Completely unrelated." }, { "left": "issue:43381", "right": "issue:45593", "accept": false, - "reason": "Gradient checkpointing eval-mode restriction and D-FINE auxiliary-loss handling are unrelated training bugs." + "reason": "Gradient checkpointing eval-mode behavior vs D-FINE auxiliary loss omission. Different training semantics." }, { "left": "issue:43976", "right": "issue:44279", "accept": false, - "reason": "Python version compatibility breakage and a generic dependency issue are not the same concrete bug." + "reason": "Python version compatibility issue vs generic dependency problem. Too broad and not the same bug." }, { "left": "issue:44075", "right": "issue:44368", "accept": false, - "reason": "SGD optimizer args being ignored and a tie_word_embeddings warning are unrelated." + "reason": "SGD optimizer args ignored vs tie_word_embeddings warning in Qwen fine-tuning. Different feature areas." }, { "left": "issue:44496", "right": "issue:45759", "accept": false, - "reason": "Unrecognized model config and `from_config` not unwrapping `text_config` are different auto-model loading problems." + "reason": "Unrecognized model config vs AutoModelForCausalLM not unwrapping text_config for composite Qwen multimodal configs. Related to loading, but not the same defect." }, { "left": "issue:45335", "right": "issue:45584", "accept": false, - "reason": "t5gemma embedding resize and Whisper empty-transcription generation failure are unrelated." + "reason": "t5gemma embedding resize issue vs Whisper empty-transcription generation failure. Unrelated models and code paths." }, { "left": "issue:43650", "right": "issue:46209", "accept": false, - "reason": "A vague data request and an FP8 inference-on-B200 bug are unrelated." + "reason": "Placeholder issue title vs FP8 inference failure on B200. No overlap in bug report content." }, { "left": "issue:43577", "right": "issue:45216", "accept": false, - "reason": "Blip2 dtype propagation and Qwen3.5 checkpoint saving are different model behaviors." + "reason": "BLIP2 dtype propagation bug vs Qwen3.5 save_pretrained regression. Different models and operations." }, { "left": "issue:45103", "right": "issue:45372", "accept": false, - "reason": "Auto-docstring annotation handling and Gemma 4 processor import failure are unrelated." + "reason": "auto_docstring crash with future annotations vs Gemma 4 processor import failure. Different tooling and root cause." }, { "left": "issue:44242", "right": "issue:45200", "accept": false, - "reason": "MoE load-balancing loss logic and Gemma 4 multimodal token ids are unrelated." + "reason": "MoE load-balancing loss omission vs Gemma 4 mm_token_type_ids defaulting. Different model behavior." }, { "left": "issue:38175", "right": "issue:44617", "accept": false, - "reason": "SigLIP2 zero probabilities and Sam3Video OOM are unrelated model/runtime issues." + "reason": "SigLIP2 zero probabilities vs Sam3Video CUDA OOM. Not the same bug or subsystem." }, { "left": "issue:44521", "right": "issue:45325", "accept": false, - "reason": "All-zero assistant masks and Qwen2.5-VL rope-index scaling are different multimodal bugs." - }, - { - "left": "issue:43299", - "right": "issue:45357", - "accept": false, - "reason": "Qwen3.5 loading failure is not the same as incorrect `save_pretrained` serialization output." + "reason": "Both multimodal, but assistant-mask generation and Qwen2.5-VL rope-position scaling are distinct code paths and symptoms." }, { "left": "issue:44279", "right": "issue:45405", "accept": false, - "reason": "Generic dependency issue and a bumped-but-unreleased PEFT minimum version are different packaging problems." + "reason": "Generic transformer dependency issue vs unreleased MIN_PEFT_VERSION bump. Different release/dependency concerns." }, { - "left": "issue:45216", + "left": "issue:43299", "right": "issue:45357", - "accept": true, - "reason": "Both report the same Qwen3.5 `save_pretrained` regression producing incorrect checkpoint contents, specifically around serialized keys." + "accept": false, + "reason": "Both Qwen3.5 regressions, but loading breaks vs save_pretrained stores wrong visual encoder keys. Not the same concrete bug." } ] }, @@ -11260,8 +11191,8 @@ "issue:43479|issue:43525", "issue:44683|issue:45542", "issue:43976|issue:45405", - "issue:43299|issue:43854", "issue:44462|issue:45081", + "issue:43299|issue:43854", "issue:44297|issue:44462", "issue:43450|issue:44279", "issue:43650|issue:46132", @@ -11280,118 +11211,118 @@ "issue:44671|issue:44779", "issue:44016|issue:45823" ], - "prepared_review_unit_hash": "649fca8708ad4cb4775fb907b85ebf5a932b02321ce0f42fedaea65f59834b95", + "prepared_review_unit_hash": "d3b7d7781452c321599ef6d5c57ebe4a5b6e1b237a0f182defe7bf1e2a607c62", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 28, - "serialized_chars": 35383, - "estimated_input_tokens": 8846, - "estimated_eval_tokens": 17948 + "serialized_chars": 35380, + "estimated_input_tokens": 8845, + "estimated_eval_tokens": 17946 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a loose set of Transformers v5 regressions and model-specific bugs, not a set of true duplicates. I would use the broad remote_code regression report as the canonical issue, but no soft pair looks like the same concrete bug/change.", - "confidence": 0.89, - "canonical_issue_reason": "issue:45020 is the broadest and most umbrella-like report here, covering recent-version breakage for model loading/custom code and fitting the general regression theme of the cluster.", + "summary": "This cluster is heterogeneous: it mixes unrelated Transformers regressions across tokenizers, processors, configs, CI, docs, quantization, and deepspeed. None of the soft pairs look like true duplicates under a strict bug-level definition.", + "confidence": 0.91, + "canonical_issue_reason": "No clear canonical duplicate target; the items are mostly distinct regressions with only broad topical overlap.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45020 is the best single issue to anchor the cluster because it is the most general, well-scoped regression report and can subsume nearby v5 breakage discussions without overfitting to one model.", + "best_issue_reason": "If one issue must anchor the cluster, issue 45278 is the broadest umbrella-style regression report about imports breaking after an upgrade, but it still does not actually subsume the rest.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:41950", "right": "issue:45020", "accept": false, - "reason": "Different failures: video-classification pipeline processor lookup vs general remote_code loading regressions." + "reason": "Different failures: a video-classification processor lookup bug vs a broad remote_code regression." }, { "left": "issue:44351", "right": "issue:45278", "accept": false, - "reason": "Both are import-related, but one is HybridCache import and the other is many import errors after upgrade; not the same concrete bug." + "reason": "One is a specific missing HybridCache import; the other is a generic report of many import errors after upgrade." }, { "left": "issue:44779", "right": "issue:45812", "accept": false, - "reason": "Both affect tokenization, but they concern different models and different incorrect outputs, so they are not the same underlying tokenizer bug." + "reason": "Both are tokenizer regressions, but on different models and different concrete symptoms." }, { "left": "issue:43295", "right": "issue:44779", "accept": false, - "reason": "Processor/tokenizer API regression vs DeepSeek tokenization regression; different code paths and symptoms." + "reason": "Processor.tokenizer/custom image handling regression is not the same bug as Deepseek tokenizer misbehavior." }, { "left": "issue:39692", "right": "issue:43901", "accept": false, - "reason": "Both are docs issues, but one is a SigLIP2 example failure and the other is stale pipeline docs wording; unrelated." + "reason": "Both are documentation-related, but one is a SigLIP2 example error and the other is stale pipeline docs for return_all_scores." }, { "left": "issue:41950", "right": "issue:45823", "accept": false, - "reason": "Video pipeline processor bug and Gemma4 device-mapping mismatch are unrelated." + "reason": "Video-classification image-processor lookup and Gemma4 device-mismatch are unrelated bugs." }, { "left": "issue:43479", "right": "issue:43525", "accept": false, - "reason": "Different config classes and different initialization/attribute problems; not the same bug." + "reason": "Different config bugs: default multimodal sub-config initialization vs missing pad_token_id on Llama4Config." }, { "left": "issue:44683", "right": "issue:45542", "accept": false, - "reason": "Torch flex_attention compile failure vs TensorFlow/tensorboard backend error; no shared code path." + "reason": "Unrelated code paths: flex_attention compilation vs tensorboard/TensorFlow backend setup." }, { "left": "issue:43976", "right": "issue:45405", "accept": false, - "reason": "Python version support regression and PEFT version pin issue are unrelated." + "reason": "Python version compatibility failure is not the same as the PEFT minimum-version bump issue." }, { - "left": "issue:43299", - "right": "issue:43854", + "left": "issue:44462", + "right": "issue:45081", "accept": false, - "reason": "Both are model-loading issues, but they are model-specific regressions with different failure modes." + "reason": "Both involve tokenizer loading, but one ignores tokenizer.json while the other crashes in a Mistral regex patch." }, { - "left": "issue:44462", - "right": "issue:45081", + "left": "issue:43299", + "right": "issue:43854", "accept": false, - "reason": "Tokenizer repository loading vs Mistral regex patch crash are different tokenizer-loading bugs." + "reason": "Different model-loading failures for different models and different underlying causes." }, { "left": "issue:44297", "right": "issue:44462", "accept": false, - "reason": "Save-pretrained metadata mismatch is not the same as ignoring tokenizer.json on load." + "reason": "Related serialization/loading area, but tokenizer_class mismatch is not the same concrete bug as ignoring tokenizer.json." }, { "left": "issue:43450", "right": "issue:44279", "accept": false, - "reason": "Video processor shape bug and generic dependency issue do not match." + "reason": "Video processor shape bug and a dependency issue with transformers are not the same defect." }, { "left": "issue:43650", "right": "issue:46132", "accept": false, - "reason": "Unrelated: placeholder data issue vs behavior change in AttentionInterface.register." + "reason": "Completely unrelated issues: placeholder text vs AttentionInterface.register behavior change." }, { "left": "issue:45081", @@ -11403,73 +11334,79 @@ "left": "issue:44466", "right": "issue:45276", "accept": false, - "reason": "Serialization inconsistency for tied weights vs resize_token_embeddings not updating Gemma4 outputs; different code paths." + "reason": "lm_head.weight serialization inconsistency is different from Gemma4 resize_token_embeddings behavior." }, { "left": "issue:44016", "right": "issue:44291", "accept": false, - "reason": "Notebook syntax error and init_empty_weights argument regression are unrelated." + "reason": "Notebook syntax error and init_empty_weights argument handling are unrelated." }, { "left": "issue:43441", "right": "issue:45834", "accept": false, - "reason": "FlashAttention failure for Ministral-3 vs long-OCR index error in Kosmos2.5; different model-specific bugs." + "reason": "FlashAttention failure in Ministral-3 and long-ocr-input index error in Kosmos2.5 are different bugs." }, { "left": "issue:42994", "right": "issue:43317", "accept": false, - "reason": "Both involve quantization/offload, but one is saving and the other is loading with device_map; not the same bug." + "reason": "Both touch quantization/offload, but one is model saving and the other is loading dequantized models with device_map=auto." }, { "left": "issue:43653", "right": "issue:45593", "accept": false, - "reason": "Tokenizer special-token registration bug and D-FINE denoising-loss behavior are unrelated." + "reason": "BigBirdTokenizer special-token registration and D-FINE auxiliary-loss behavior are unrelated." }, { "left": "issue:44623", "right": "issue:45397", "accept": false, - "reason": "Missing processor save files vs Gemma-4 Zero3 loading are different issues." + "reason": "processor.save_pretrained missing files is not the same bug as gemma-4 zero3 from_pretrained failure." }, { "left": "issue:44623", "right": "issue:44749", "accept": false, - "reason": "Processor save-pretrained problem and data filtering slowdown are unrelated." + "reason": "A save_pretrained file issue is unrelated to a post-upgrade data filtering slowdown report." }, { "left": "issue:45600", "right": "issue:45712", "accept": false, - "reason": "Both are CI/repo hygiene issues, but one is stale auto_mappings entries and the other is leftover dummy classes; different fixes." + "reason": "Both are cleanup/CI issues, but they affect different files and different failure modes." }, { "left": "issue:33290", "right": "issue:45656", "accept": false, - "reason": "Both mention DeepSpeed/optimizer behavior, but OOM with Adafactor and optimizer.step being called twice are different concrete bugs." + "reason": "Deepspeed OOM with Adafactor and optimizer.step being called twice are related only at a high level, not the same concrete bug." }, { "left": "issue:44261", "right": "issue:44263", "accept": false, - "reason": "Different model internals: q_a_layernorm epsilon config vs torch.split return values in a separate indexer." + "reason": "A precision/config issue in MLA is unrelated to torch.split return values in GlmMoeDsaIndexer." }, { "left": "issue:44361", "right": "issue:45593", "accept": false, - "reason": "Tokenizer AttributeError on tasks and D-FINE loss behavior are unrelated." + "reason": "MLukeTokenizer AttributeError and D-FINE auxiliary-loss handling are unrelated." }, { "left": "issue:44671", "right": "issue:44779", "accept": false, - "reason": "Different model families and different failure modes: CamemBERT MLM predictions vs DeepSeek tokenization." + "reason": "Different model-specific tokenizer/masked-LM regressions with no shared concrete code path." + }, + { + "left": "issue:44016", + "right": "issue:45823", + "accept": false, + "reason": "Notebook syntax error and Gemma4 device mismatch are unrelated." } ] }, @@ -11493,10 +11430,12 @@ "model": "gpt-5.4-mini?service_tier=flex", "cluster_id": "cluster-10000-450", "nodes": [ + "issue:36246", "issue:41628", "issue:41950", "issue:42175", "issue:42491", + "issue:42757", "issue:42886", "issue:43262", "issue:43299", @@ -11527,7 +11466,6 @@ "issue:44521", "issue:44589", "issue:44683", - "issue:44734", "issue:44871", "issue:45003", "issue:45103", @@ -11538,13 +11476,12 @@ "issue:45563", "issue:45593", "issue:46132", - "issue:46399", - "issue:46424" + "issue:46399" ], "soft_pairs": [ "issue:43723|issue:44162", - "issue:43441|issue:44683", "issue:42175|issue:43976", + "issue:43441|issue:44683", "issue:43901|issue:43994", "issue:43262|issue:44521", "issue:43673|issue:43761", @@ -11559,214 +11496,208 @@ "issue:44188|issue:44336", "issue:45103|issue:46399", "issue:45561|issue:45563", - "issue:44336|issue:44373", "issue:44246|issue:44496", + "issue:44336|issue:44373", "issue:44295|issue:45397", "issue:44242|issue:45593", "issue:42491|issue:43299", "issue:43531|issue:45362", "issue:43526|issue:45003", "issue:41950|issue:44220", - "issue:42886|issue:46132", "issue:44246|issue:44871", - "issue:44734|issue:46424" + "issue:42886|issue:46132", + "issue:36246|issue:42757" ], - "prepared_review_unit_hash": "0813e514ba7a1ce29c1caab06c5158473d1c46bbfc6770d91ee3daa7b0827de0", + "prepared_review_unit_hash": "55689a5cf6bb13384d540b27b661672356698774438dce0d070242fd509f7c94", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 28, - "serialized_chars": 35380, - "estimated_input_tokens": 8845, - "estimated_eval_tokens": 17946 + "serialized_chars": 35454, + "estimated_input_tokens": 8864, + "estimated_eval_tokens": 17984 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "A heterogeneous set of Transformers regressions across tokenizer/config loading, multimodal processors, model families, attention/caching, and docs. They share a broad 'v5 broke something' theme, but most pairs are not the same underlying bug.", - "confidence": 0.91, - "canonical_issue_reason": "issue:43723 is the most general, user-facing loading regression here: it targets a core AutoTokenizer.from_pretrained path and is broad enough to represent nearby import/load failures better than the model-specific reports.", + "summary": "This is a broad cluster of mostly unrelated Transformers v5-era regression reports, import failures, and model-specific loading bugs. None of the soft pairs look like the same concrete underlying bug/change under a strict duplicate standard, so all are rejected. The strongest representative issue is the Qwen3 MoE/v5 loading regression report.", + "confidence": 0.33, + "canonical_issue_reason": "issue:43299 is the clearest, most concrete representative of the cluster\u2019s recurring v5 regression theme: a version-bounded model-loading failure with a detailed Qwen3 MoE repro and explicit hf4.x vs hf5.x breakage.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43723 is the best overall representative issue because it describes a common entrypoint failure in v5 and is broad, well-scoped, and close to several other loading-related reports in the cluster.", + "best_issue_reason": "issue:43299 best captures the dominant signal in this cluster: a specific, actionable Transformers v5 loading regression affecting Qwen3 MoE models, and it is closer to the cluster\u2019s main recurring pattern than the other, more isolated import/pipeline bugs.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43723", "right": "issue:44162", "accept": false, - "reason": "Both are loading-related, but one is a generic AutoTokenizer regression and the other is an ESM2-specific failure; different concrete bug surfaces." + "reason": "Both are load-time failures, but one is a generic AutoTokenizer regression and the other is an ESM2 model-specific breakage; different code paths." }, { - "left": "issue:43441", - "right": "issue:44683", + "left": "issue:42175", + "right": "issue:43976", "accept": false, - "reason": "Both involve attention/flash/flex kernels, but one is a Ministral-3 FlashAttention failure and the other is a torch>=2.9 compiled flex_attention regression; not the same bug." + "reason": "Packaging extras vs Python-version compatibility; unrelated issues." }, { - "left": "issue:42175", - "right": "issue:43976", + "left": "issue:43441", + "right": "issue:44683", "accept": false, - "reason": "Package/backend dependency issue vs Python version compatibility issue; different underlying causes." + "reason": "Both touch attention kernels, but one is Ministral FlashAttention in Transformers RC and the other is compiled flex_attention on torch>=2.9; not the same bug." }, { "left": "issue:43901", "right": "issue:43994", "accept": false, - "reason": "Docs mismatch for return_all_scores vs a SigLIP2 model output problem; unrelated." + "reason": "Docs mismatch vs incorrect model outputs; different underlying problems." }, { "left": "issue:43262", "right": "issue:44521", "accept": false, - "reason": "Both mention apply_chat_template, but one is about audio sampling rate defaults and the other about multimodal assistant masks; different code paths." + "reason": "Different apply_chat_template bugs: audio sampling-rate default vs multimodal assistant-mask generation." }, { "left": "issue:43673", "right": "issue:43761", "accept": false, - "reason": "Cache handling in generation vs CLIPVisionModel hidden_states output; unrelated regressions." + "reason": "Cache/chunked-prefill regression vs CLIP hidden_states regression; unrelated." }, { "left": "issue:44297", "right": "issue:45397", "accept": false, - "reason": "Tokenizer serialization metadata bug vs a gemma-4 from_pretrained/zero3 loading issue; not the same failure." + "reason": "Tokenizer save metadata mismatch vs DeepSpeed/from_pretrained loading failure; different paths." }, { "left": "issue:43824", "right": "issue:43976", "accept": false, - "reason": "Specific Qwen2.5-VL import error vs general Python support regression; different problems." + "reason": "Specific Qwen2.5-VL import error vs Python 3.9+ support issue; unrelated." }, { "left": "issue:43525", "right": "issue:43881", "accept": false, - "reason": "Llama4Config missing pad_token_id vs glm-4v-9b loading failure; different model-specific issues." + "reason": "Both are model-load failures, but they affect different model families and have different missing-field symptoms." }, { "left": "issue:43901", "right": "issue:44246", "accept": false, - "reason": "Documentation mismatch vs intermittent import slowdown; unrelated." + "reason": "Documentation issue vs intermittent import slowness; unrelated." }, { "left": "issue:43525", "right": "issue:43643", "accept": false, - "reason": "Both involve missing config fields, but they are different models and different manifestations; too broad to treat as the same bug." + "reason": "Different config-loading problems: missing pad_token_id for Llama4 vs missing fields with trust_remote_code." }, { "left": "issue:44589", "right": "issue:45362", "accept": false, - "reason": "Float8 storage deserialization issue vs Qwen3.5 chat crash; unrelated." + "reason": "Different error modes and model paths; not the same underlying bug." }, { "left": "issue:41628", "right": "issue:43976", "accept": false, - "reason": "AutoImageProcessor import failure vs Python version compatibility; no shared concrete bug." + "reason": "ImportError for AutoImageProcessor vs Python-version compatibility regression; unrelated." }, { "left": "issue:43784", "right": "issue:45542", "accept": false, - "reason": "Different backend/import regressions: sentence-transformers import NameError vs TensorFlow backend detection error." + "reason": "Both are import/backend issues, but they involve different packages and code paths." }, { "left": "issue:44188", "right": "issue:44336", "accept": false, - "reason": "Attention kernel divergence on torch.compile vs ANSI codes in loading_report; unrelated." + "reason": "Attention-kernel divergence vs ANSI formatting in loading_report; unrelated." }, { "left": "issue:45103", "right": "issue:46399", "accept": false, - "reason": "Auto-docstring annotation handling bug vs Qwen3.5 kernel registration crash; different subsystems." + "reason": "auto_docstring crash with future annotations vs Qwen3.5 kernel-registration ValueError; unrelated." }, { "left": "issue:45561", "right": "issue:45563", "accept": false, - "reason": "Test harness race condition vs stale generation warning; unrelated." + "reason": "pytest-xdist file race vs stale warning in paged generate; different bugs." }, { - "left": "issue:44336", - "right": "issue:44373", + "left": "issue:44246", + "right": "issue:44871", "accept": false, - "reason": "ANSI output formatting bug vs docstring text error; not the same bug." + "reason": "Import-performance issue vs Gemma eos_token_id config mismatch; unrelated." }, { - "left": "issue:44246", - "right": "issue:44496", + "left": "issue:44336", + "right": "issue:44373", "accept": false, - "reason": "Import performance issue vs unrecognized model/config problem; different symptoms and causes." + "reason": "ANSI code emission in loading_report vs a docstring typo; not the same issue." }, { "left": "issue:44295", "right": "issue:45397", "accept": false, - "reason": "Position_ids buffer access bug vs gemma-4 loading issue; unrelated." + "reason": "Buffer read error vs model loading with zero3; different code paths." }, { "left": "issue:44242", "right": "issue:45593", "accept": false, - "reason": "Both concern missing losses, but they are different models and different loss terms; too distinct to merge." + "reason": "Both concern optional-loss branches, but they are different models and different mechanisms; not the same concrete bug." }, { "left": "issue:42491", "right": "issue:43299", "accept": false, - "reason": "Both involve Qwen3 MoE models, but one is LoRA compatibility across branches and the other is a model loading regression; not the same concrete bug." + "reason": "Both mention Qwen3 MoE/v5 compatibility, but one is a LoRA portability problem and the other is a Qwen3VL MoE loading regression." }, { "left": "issue:43531", "right": "issue:45362", "accept": false, - "reason": "Qwen3-MoE sliding_window bug vs Qwen3.5 chat crash; related family, but different defects." + "reason": "Both are Qwen-related, but sliding_window behavior and chat crashes are distinct failures." }, { "left": "issue:43526", "right": "issue:45003", "accept": false, - "reason": "BeitImageProcessorFast label reduction bug vs unsafe sys.modules access in modeling_utils; unrelated." + "reason": "BeitImageProcessorFast label reduction vs unsafe sys.modules access; unrelated." }, { "left": "issue:41950", "right": "issue:44220", "accept": false, - "reason": "Video-classification processor mismatch vs audio fbank extraction issue; different modalities and code paths." + "reason": "Video-classification pipeline misrouting to image processors vs audio fbank extraction issue; different subsystems." }, { "left": "issue:42886", "right": "issue:46132", "accept": false, - "reason": "Offline tokenizer cache loading bug vs AttentionInterface.register behavior change; unrelated." - }, - { - "left": "issue:44246", - "right": "issue:44871", - "accept": false, - "reason": "Import slowdown vs Gemma-3 eos_token_id configuration mismatch; different bug classes." + "reason": "Offline tokenizer cache loading vs AttentionInterface.register semantics; unrelated." }, { - "left": "issue:44734", - "right": "issue:46424", + "left": "issue:36246", + "right": "issue:42757", "accept": false, - "reason": "Serve KV-continuation indexing crash vs cache max length configuration only applying to quantized cache; different cache issues." + "reason": "Two different import errors involving unrelated APIs/packages; not a duplicate." } ] }, @@ -11791,7 +11722,6 @@ "cluster_id": "cluster-10000-450", "nodes": [ "issue:36246", - "issue:42757", "issue:43257", "issue:43262", "issue:43352", @@ -11810,6 +11740,7 @@ "issue:44485", "issue:44496", "issue:44561", + "issue:44734", "issue:44749", "issue:44779", "issue:44857", @@ -11840,30 +11771,30 @@ "issue:46424" ], "soft_pairs": [ - "issue:36246|issue:42757", + "issue:44734|issue:46424", "issue:36246|issue:44351", "issue:44336|issue:44485", "issue:45003|issue:45561", - "issue:44336|issue:46424", "issue:44779|issue:45292", - "issue:43475|issue:46369", + "issue:44336|issue:46424", "issue:44242|issue:44857", + "issue:43475|issue:46369", "issue:44295|issue:44749", "issue:45859|issue:46399", "issue:46291|issue:46399", "issue:36246|issue:43596", "issue:44162|issue:45200", + "issue:43901|issue:45003", "issue:43352|issue:45250", "issue:43673|issue:44561", - "issue:43901|issue:45003", "issue:44496|issue:45003", "issue:43257|issue:45137", "issue:43262|issue:44246", "issue:45292|issue:45584", "issue:45412|issue:46123", "issue:45375|issue:45759", - "issue:44561|issue:46399", "issue:45292|issue:45701", + "issue:44561|issue:46399", "issue:44877|issue:45375", "issue:44987|issue:45459", "issue:44995|issue:45200", @@ -11871,214 +11802,214 @@ "issue:44479|issue:46170", "issue:45397|issue:45656" ], - "prepared_review_unit_hash": "dd0288808b2252f111977153a953e315918e453b0dbb14056bf41b331cb960b9", + "prepared_review_unit_hash": "5be94fde938ffc8d2b68019fb6dc065aa6efebcc6a498e2c29624f86199411bc", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36687, - "estimated_input_tokens": 9172, - "estimated_eval_tokens": 18600 + "serialized_chars": 36727, + "estimated_input_tokens": 9182, + "estimated_eval_tokens": 18620 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a similarity cluster, not a true duplicate set: the issues span unrelated import errors, model-specific bugs, config validation, performance, and docs. I did not find any soft pair that clearly describes the same underlying bug/change.", - "confidence": 0.34, - "canonical_issue_reason": "No strong canonical duplicate exists. If one representative is needed, issue 45003 is the broadest core-library bug and has the most central-looking weak similarity links, but the cluster is heterogeneous overall.", + "summary": "This set is highly heterogeneous and does not look like a true duplicate cluster. The only plausible representative issue is the broad Qwen video regression report, but even that is just a weak canonical choice.", + "confidence": 0.74, + "canonical_issue_reason": "Issue 44479 is the broadest and most concrete report here: a v5.3.0 regression affecting several Qwen VL/video variants with a clear symptom and scope.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45003 is the best single representative only because it is a concrete internal utility bug with broader applicability than the model-specific reports; however, it is not a true duplicate hub.", + "best_issue_reason": "44479 is the best single issue to represent the set because it is specific, user-facing, and spans multiple related model families, unlike the other narrower or vaguer reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { - "left": "issue:36246", - "right": "issue:42757", + "left": "issue:44734", + "right": "issue:46424", "accept": false, - "reason": "Both are import errors, but they involve different missing symbols from different packages and unrelated code paths." + "reason": "Different bugs: KV-cache continuation crash in serve responses vs cache-length config handling." }, { "left": "issue:36246", "right": "issue:44351", "accept": false, - "reason": "Both mention import failures in transformers, but the missing classes are different and the reports point to different modules." + "reason": "Both are import errors, but for unrelated symbols and model paths." }, { "left": "issue:44336", "right": "issue:44485", "accept": false, - "reason": "ANSI-code output handling and GLM-5 RoPE implementation are unrelated bugs." + "reason": "ANSI-color output bug in loading_report vs GLM-5 RoPE implementation discussion." }, { "left": "issue:45003", "right": "issue:45561", "accept": false, - "reason": "One is an unsafe sys.modules access in modeling_utils; the other is a pytest-xdist file race in testing utilities." - }, - { - "left": "issue:44336", - "right": "issue:46424", - "accept": false, - "reason": "Loading-report terminal coloring and cache-length behavior are unrelated." + "reason": "sys.modules access bug is unrelated to the xdist captured_info.txt race." }, { "left": "issue:44779", "right": "issue:45292", "accept": false, - "reason": "Tokenizer correctness and resize_token_embeddings output synchronization are different model behaviors." + "reason": "Tokenization regression and resize_token_embeddings propagation are different code paths." }, { - "left": "issue:43475", - "right": "issue:46369", + "left": "issue:44336", + "right": "issue:46424", "accept": false, - "reason": "Both are SAM3-related, but one is a missing attribute in vision encoder output and the other is a memory overflow problem." + "reason": "Console coloring in loading_report is unrelated to cache-length behavior." }, { "left": "issue:44242", "right": "issue:44857", "accept": false, - "reason": "Load-balancing loss computation and RT-DETR AMP crashes are unrelated." + "reason": "MoE load-balancing loss omission vs an AMP/CUDA crash in LwDetrImageLoss." + }, + { + "left": "issue:43475", + "right": "issue:46369", + "accept": false, + "reason": "SAM3 vision encoder attribute error vs memory overflow in sam3-litetext are different failures." }, { "left": "issue:44295", "right": "issue:44749", "accept": false, - "reason": "A position_ids buffer access error is not the same bug as a slowdown after upgrading transformers." + "reason": "Position_ids buffer access error vs a slow-data-processing regression after upgrade." }, { "left": "issue:45859", "right": "issue:46399", "accept": false, - "reason": "Both are Qwen3.5/MoE-related, but one is CPU-offload incompatibility in rotary embeddings and the other is a kernel registration error." + "reason": "CPU-offload incompatibility in Qwen3_5MoeTextRotaryEmbedding is distinct from a kernels decorator registration crash." }, { "left": "issue:46291", "right": "issue:46399", "accept": false, - "reason": "Kernels version/revision installation failure and apply_rotary_pos_emb registration failure are distinct issues." + "reason": "Kernels import/version requirement failure is not the same as the Qwen3.5 rotary-embedding registration error." }, { "left": "issue:36246", "right": "issue:43596", "accept": false, - "reason": "Qwen2.5-VL import failure and DeepSpeed ZeRO3 BertModel IndexError are unrelated." + "reason": "Qwen2.5-VL import error and DeepSpeed ZeRO3/BertModel index error are unrelated." }, { "left": "issue:44162", "right": "issue:45200", "accept": false, - "reason": "ESM2 breakage and Gemma 4 token-type defaults are different model-specific bugs." + "reason": "ESM2 model breakage and Gemma 4 token-type defaulting are different model-specific bugs." + }, + { + "left": "issue:43901", + "right": "issue:45003", + "accept": false, + "reason": "Docs mismatch for return_all_scores is unrelated to unsafe sys.modules access." }, { "left": "issue:43352", "right": "issue:45250", "accept": false, - "reason": "A model not supporting Flash Attention 2.0 yet is not the same as the generic Flash Attention 2.0 issue." + "reason": "Nemotron Flash Attention 2 unsupported is not the same as the vague generic Flash Attention 2 issue." }, { "left": "issue:43673", "right": "issue:44561", "accept": false, - "reason": "Chunked-prefill cache handling and removal of is_torch_fx_available affect different code paths." - }, - { - "left": "issue:43901", - "right": "issue:45003", - "accept": false, - "reason": "Docs for text classification pipeline and unsafe sys.modules access are unrelated." + "reason": "Chunked-prefill cache bug and removal of is_torch_fx_available affect different paths." }, { "left": "issue:44496", "right": "issue:45003", "accept": false, - "reason": "Unrecognized model config and sys.modules access are different failure modes." + "reason": "Unknown-model config loading failure is unrelated to sys.modules indexing." }, { "left": "issue:43257", "right": "issue:45137", "accept": false, - "reason": "Both involve DeepSpeed, but one is missing MOE weight conversion and the other is an empty-deque IndexError." + "reason": "Qwen3 MOE weight conversion issue and ZeRO3 deque underflow are different DeepSpeed failures." }, { "left": "issue:43262", "right": "issue:44246", "accept": false, - "reason": "Audio chat-template sampling rate defaults and import latency are unrelated." + "reason": "Audio processor sampling-rate default bug is unrelated to import slowdown." }, { "left": "issue:45292", "right": "issue:45584", "accept": false, - "reason": "Embedding resize propagation and Whisper empty transcription handling are different bugs." + "reason": "Embedding resize propagation and Whisper empty-transcription failure are different behaviors." }, { "left": "issue:45412", "right": "issue:46123", "accept": false, - "reason": "RT-DETR memory cleanup and mask pipeline batch-finalization are unrelated." + "reason": "RT-DETR memory leak and MaskGenerationPipeline batch-finalization bug are unrelated." }, { "left": "issue:45375", "right": "issue:45759", "accept": false, - "reason": "Both touch Qwen multimodal config loading, but one is strict-field dropping and the other is text_config unwrapping; they are not the same concrete bug." + "reason": "Missing deepstack_visual_indexes field and text_config unwrapping are separate Qwen config issues." }, { - "left": "issue:44561", - "right": "issue:46399", + "left": "issue:45292", + "right": "issue:45701", "accept": false, - "reason": "A removed helper breaking trust_remote_code is unrelated to a kernel registration error." + "reason": "Embedding resize bug and version-dependent tokenization change are not the same defect." }, { - "left": "issue:45292", - "right": "issue:45701", + "left": "issue:44561", + "right": "issue:46399", "accept": false, - "reason": "Embedding resize synchronization and tokenizer version drift are different problems." + "reason": "is_torch_fx_available removal breakage is unrelated to the Qwen3.5 apply_rotary_pos_emb registration crash." }, { "left": "issue:44877", "right": "issue:45375", "accept": false, - "reason": "Both mention strict config handling, but they affect different model configs and different symptoms." + "reason": "granite_speech strict-config loading and Qwen3_5MoeVisionConfig field loss are different config problems." }, { "left": "issue:44987", "right": "issue:45459", "accept": false, - "reason": "Loading physical-intelligence/fast and hidden tokenizer errors without protobuf are separate import/loading issues." + "reason": "Loading a specific model package failing is unrelated to tokenizer errors hidden by protobuf exception handling." }, { "left": "issue:44995", "right": "issue:45200", "accept": false, - "reason": "A stale indexer cache on second forward and Gemma 4 token-type defaults are unrelated." + "reason": "GlmMoeDsa stale cache on second forward and Gemma 4 mm_token_type_ids defaults are unrelated." }, { "left": "issue:44466", "right": "issue:45335", "accept": false, - "reason": "Serialization of tied lm_head weights and t5gemma embedding resize behavior are different bugs." + "reason": "lm_head.weight serialization inconsistency is distinct from resize_token_embeddings not updating decoder.embed_tokens." }, { "left": "issue:44479", "right": "issue:46170", "accept": false, - "reason": "Qwen VL video regression and Windows/ROCm torch.distributed import issues are unrelated." + "reason": "Qwen video-input regression and Windows ROCm distributed import failure are unrelated." }, { "left": "issue:45397", "right": "issue:45656", "accept": false, - "reason": "Gemma-4 ZeRO3 loading and double optimizer step with deepspeed are different DeepSpeed-related bugs." + "reason": "Gemma-4 ZeRO3 loading and duplicate optimizer-step behavior with DeepSpeed are different bugs." } ] }, @@ -12151,8 +12082,8 @@ "issue:46421" ], "soft_pairs": [ - "issue:45987|issue:46082", "issue:45137|issue:45600", + "issue:45987|issue:46082", "issue:43475|issue:43540", "issue:44483|issue:46202", "issue:43653|issue:44162", @@ -12182,214 +12113,214 @@ "issue:44038|issue:44623", "issue:43475|issue:43643" ], - "prepared_review_unit_hash": "66d33a4200971214ecd412efa7bfe321a85cf110a6625ff8c8d6f4dd349d7294", + "prepared_review_unit_hash": "89f0df0ec15f372b06c5b64fb02374d5c62db277ad340f276f5df4416c451507", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 30, - "serialized_chars": 35931, - "estimated_input_tokens": 8983, - "estimated_eval_tokens": 18222 + "serialized_chars": 35928, + "estimated_input_tokens": 8982, + "estimated_eval_tokens": 18220 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly a grab bag of unrelated closed issues; the soft pairs only show superficial embedding similarity, not the same underlying bug. No PRs are present.", - "confidence": 0.91, - "canonical_issue_reason": "issue:43646 is the broadest framework-level regression in the set ('Transformers 5.0.0 breaks custom model initialization') and is the closest thing to an umbrella issue; the others are mostly model-, tokenizer-, or subsystem-specific.", + "summary": "No soft pairs look like true duplicates; they mostly share a broad subsystem or model family but describe different concrete failures, so all soft edges are rejected. For the issue set, issue 46421 is the strongest canonical tracking candidate because it is specific, open, and directly describes a cache-lazy-init recompilation regression in chunked prefill.", + "confidence": 0.86, + "canonical_issue_reason": "Issue 46421 is the clearest representative of the cache/chunked-prefill regression family: it states a concrete symptom, a narrow code path, and an active bug surface.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43646 is the best representative issue for this cluster because it captures a general v5 breakage theme, while the rest are too fragmented to serve as a better canonical duplicate target.", + "best_issue_reason": "Issue 46421 is the most actionable issue in the set: it is open, technically specific, and tied to a reproducible performance regression rather than a broad or ambiguous report.", "best_pr_reason": null, "soft_edge_verdicts": [ { - "left": "issue:45987", - "right": "issue:46082", + "left": "issue:45137", + "right": "issue:45600", "accept": false, - "reason": "StaticCache return-type bug vs LlamaConfig head_dim validation; different code paths." + "reason": "DeepSpeed ZeRO3 deque exhaustion and removed auto_mappings config refs are unrelated failures in different code paths." }, { - "left": "issue:45137", - "right": "issue:45600", + "left": "issue:45987", + "right": "issue:46082", "accept": false, - "reason": "DeepSpeed ZeRO3 deque error vs stale auto_mappings references; unrelated failures." + "reason": "StaticCache return-type mismatch and LlamaConfig head_dim validation are different bugs." }, { "left": "issue:43475", "right": "issue:43540", "accept": false, - "reason": "SAM3 encoder missing attribute vs Qwen3Omni video input ValueError; not the same bug." + "reason": "SAM3 VisionEncoder attribute loss and Qwen3OmniMoe video-input ValueError are different model paths and symptoms." }, { "left": "issue:44483", "right": "issue:46202", "accept": false, - "reason": "Chat/completions request handling vs Gemma pipeline roundtrip; different behaviors." + "reason": "API request rejection and Gemma tokenizer roundtrip failure are unrelated." }, { "left": "issue:43653", "right": "issue:44162", "accept": false, - "reason": "Tokenizer special-token registration issue vs ESM2 regression; different subsystems." + "reason": "BigBirdTokenizer special-token registration and ESM2 breakage do not describe the same failure." }, { "left": "issue:42886", "right": "issue:45229", "accept": false, - "reason": "Offline cache tokenizer load failure vs Gemma4 multi-GPU CUDA OOM; unrelated." + "reason": "Offline tokenizer cache loading and Gemma4 multi-GPU CUDA OOM are unrelated bugs." }, { "left": "issue:44060", "right": "issue:44077", "accept": false, - "reason": "Tied-weights warning bug vs patchtsmixer post_init policy change; not the same problem." + "reason": "Tied-weights warning behavior and patchtsmixer post_init policy are different issues." }, { "left": "issue:44995", "right": "issue:45588", "accept": false, - "reason": "Stale cache on second forward vs flash_attention None-aux crash; different code paths." + "reason": "Stale indexer cache on second forward and flash_attention s_aux=None AttributeError are different code paths." }, { "left": "issue:44368", "right": "issue:44977", "accept": false, - "reason": "Tied-embeddings warning vs flash-attention generation issue; not the same bug." + "reason": "Tied embeddings warning and flash-attention generation failure are not the same concrete bug." }, { "left": "issue:43065", "right": "issue:43408", "accept": false, - "reason": "Dummy Conv2d in SAM3 decoder vs model-type warning for sam3_video/sam3_tracker; distinct issues." + "reason": "Both are SAM3-related, but one concerns a dummy Conv2d in the pixel decoder while the other is a model-type mismatch warning." }, { "left": "issue:44368", "right": "issue:45563", "accept": false, - "reason": "Different warnings in different generation/save paths; no shared failure mode." + "reason": "A tied-embeddings warning is unrelated to a stale num_return_sequences warning in paged generate()." }, { "left": "issue:44188", "right": "issue:46202", "accept": false, - "reason": "torch.compile kernel divergence vs Gemma pipeline token roundtrip; unrelated." + "reason": "Torch.compile kernel divergence and Gemma tokenizer roundtrip failure are unrelated." }, { "left": "issue:44877", "right": "issue:46170", "accept": false, - "reason": "Strict config loading failure vs Windows/ROCm import dependency error; different root causes." + "reason": "Strict granite_speech config loading and Windows ROCm import/type errors are different problems." }, { "left": "issue:43931", "right": "issue:44368", "accept": false, - "reason": "Weight-shape mismatch in Qwen3-VL-30B vs tied-embeddings warning; unrelated." + "reason": "Qwen3-VL weight-shape mismatch and tied-embeddings warnings are unrelated." }, { "left": "issue:45584", "right": "issue:46170", "accept": false, - "reason": "Whisper empty-transcription generation bug vs Windows ROCm import/type error; different areas." - }, - { - "left": "issue:44188", - "right": "issue:46421", - "accept": false, - "reason": "Attention-kernel divergence under compile vs cache lazy-init recompilation; different defects." + "reason": "Whisper empty-transcription generation failure and Windows ROCm import issues do not share the same code path." }, { "left": "issue:43065", "right": "issue:46123", "accept": false, - "reason": "SAM3 Conv2d placeholder vs MaskGenerationPipeline partial-batch dropping results; not the same bug." + "reason": "SAM3 pixel-decoder setup and MaskGenerationPipeline final-batch handling are unrelated." }, { "left": "issue:44297", "right": "issue:44623", "accept": false, - "reason": "tokenizer.save_pretrained metadata mismatch vs processor.save_pretrained missing files; same family only, not same bug." + "reason": "Tokenizer save_pretrained class mismatch and processor save_pretrained missing files are both save-pretrained issues, but not the same concrete failure." }, { "left": "issue:44361", "right": "issue:44749", "accept": false, - "reason": "MLukeTokenizer AttributeError vs global slowdown regression after upgrade; unrelated." + "reason": "MLukeTokenizer AttributeError and a performance regression after upgrade are different bugs." }, { "left": "issue:44869", "right": "issue:45356", "accept": false, - "reason": "Whisper timestamp decode crash vs Kimi-K2.5 tokenizer codec regression; different concrete failures." + "reason": "Whisper decode crash on trailing replacement characters and Kimi-K2.5 codec regression are different tokenizer issues." }, { "left": "issue:44623", "right": "issue:45656", "accept": false, - "reason": "Missing save_pretrained files vs optimizer step called twice; unrelated." + "reason": "Missing processor files and double optimizer stepping under deepspeed are unrelated." }, { "left": "issue:43646", "right": "issue:44162", "accept": false, - "reason": "Custom model initialization breakage vs ESM2-specific regression; too broad to be the same bug." + "reason": "Both mention broad v5 breakage, but custom model initialization and ESM2 are different code paths and symptoms." }, { "left": "issue:44749", "right": "issue:45656", "accept": false, - "reason": "Preprocessing slowdown vs double optimizer step under deepspeed; different behavior." + "reason": "A slowdown regression and optimizer.step being called twice are not the same bug." }, { "left": "issue:43701", "right": "issue:44749", "accept": false, - "reason": "Checkpoint key mismatch vs preprocessing slowdown; no shared underlying change." + "reason": "Checkpoint key mismatch and data-processing slowdown are unrelated." }, { "left": "issue:43232", "right": "issue:43673", "accept": false, - "reason": "Generation kwargs update after sync_gpus vs missing cache during chunked prefill; related area but distinct bug." + "reason": "Generation kwargs update after sync_gpus and missing cache during chunked_prefill are both generation/cache topics, but they are distinct failure modes." }, { "left": "issue:44871", "right": "issue:46132", "accept": false, - "reason": "Gemma EOS config inconsistency vs AttentionInterface.register behavior change; not the same issue." + "reason": "Gemma eos_token_id inconsistency and AttentionInterface.register behavior change are unrelated." }, { "left": "issue:45230", "right": "issue:45397", "accept": false, - "reason": "Generic bug report vs gemma-4 zero3 loading problem; insufficient overlap and different specifics." + "reason": "A generic bug report is not evidence of the same issue as gemma-4 zero3 from_pretrained." }, { "left": "issue:42994", "right": "issue:43482", "accept": false, - "reason": "Quantized model saving failure vs Qwen2.5-GGUF loading failure; different model-loading paths." + "reason": "Quantized model saving failure and Qwen2.5-GGUF loading failure are separate model I/O problems." }, { "left": "issue:44038", "right": "issue:44623", "accept": false, - "reason": "Qwen3-VL-Moe v5 bug vs processor.save_pretrained missing files; unrelated." + "reason": "Qwen3-VL-Moe/v5 breakage and processor save_pretrained missing files are different bugs." }, { "left": "issue:43475", "right": "issue:43643", "accept": false, - "reason": "SAM3 video encoder attribute error vs AutoConfig trust_remote_code field loss; different code paths." + "reason": "SAM3 video encoder missing fpn_position_embeddings and trust_remote_code config-field loss are unrelated." + }, + { + "left": "issue:44188", + "right": "issue:46421", + "accept": false, + "reason": "Torch.compile branching causing divergent attention kernels and cache lazy init causing recompilation are related broadly, but not the same concrete code-path problem." } ] }, @@ -12493,20 +12424,20 @@ "issue:42175|issue:44933", "issue:43377|issue:44242" ], - "prepared_review_unit_hash": "f6ec3ca0d9a8af03e4cbda2ed4e9b4e7a3e540074a99a999759be964bb136e40", + "prepared_review_unit_hash": "ced8dfcfe10754e79adb075424f0f1d34271a47c3d6205814059aa003cdb56bf", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 47, "item_count": 47, "soft_pair_count": 30, - "serialized_chars": 36057, + "serialized_chars": 36060, "estimated_input_tokens": 9015, "estimated_eval_tokens": 18286 }, @@ -12515,198 +12446,198 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a heterogeneous mix of unrelated issues. The only clear duplicate pair in the supplied soft edges is the Qwen3ForCausalLM VRAM leak report (42673/43704), with 42673 the best canonical issue. A few other pairs are superficially similar (imports, Qwen/MoE, FP8, masking) but describe different code paths or symptoms.", - "confidence": 0.9, - "canonical_issue_reason": "issue:42673 is the oldest and most-discussed instance of the only clearly duplicated bug in the set, and it has the stronger duplicate signal versus issue:43704.", + "summary": "This cluster is mostly heterogeneous; only the Qwen3ForCausalLM VRAM-leak reports (42673/43704) are a clear duplicate pair, while the rest are distinct import/config/runtime bugs.", + "confidence": 0.94, + "canonical_issue_reason": "issue:42673 is the best duplicate anchor because issue:43704 repeats the exact same VRAM-leak title, and 42673 has the fuller report with more discussion/references.", "canonical_pr_reason": null, - "best_issue_reason": "issue:42673 is the best representative of the duplicate cluster because it is the earliest, most active report of the same VRAM leak bug and has an exact-title duplicate.", + "best_issue_reason": "issue:42673 is the strongest representative issue overall: it describes a concrete reproducible bug, has the clearest duplicate sibling, and is better documented than the other candidates.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:28282", "right": "issue:30990", "accept": false, - "reason": "Different failures: missing PyTorch backend import vs Sentence Transformers hanging while loading." + "reason": "Both involve loading/runtime failures, but one is a missing PyTorch dependency error and the other is Sentence Transformers hanging; different bugs." }, { "left": "issue:43475", "right": "issue:45381", "accept": false, - "reason": "Both are multimodal bugs, but they affect different models and different failure modes." + "reason": "Both are multimodal/VL issues, but they fail on different model classes and different missing fields/position-id behavior." }, { "left": "issue:45406", "right": "issue:46202", "accept": false, - "reason": "Gemma tokenizer/processor problems, but one is a missing attribute and the other is a roundtrip tokenization issue." + "reason": "Gemma Processor attribute error and Gemma pipeline roundtrip tokenization failure are different code paths and symptoms." }, { "left": "issue:41628", "right": "issue:44933", "accept": false, - "reason": "Both are import/export problems, but they target different symbols and do not share the same code path." + "reason": "Both are import problems, but they concern different symbols and different modules; not the same underlying defect." }, { "left": "issue:42371", "right": "issue:45092", "accept": false, - "reason": "Packaging/backend install issue vs remote-code meta-init incompatibility; unrelated." + "reason": "TF32 API guidance and InternVL2 meta-init incompatibility are unrelated issues." }, { "left": "issue:45375", "right": "issue:45478", "accept": false, - "reason": "Related Qwen3.5 MoE area, but one is a dropped config field and the other is a from_pretrained load error." + "reason": "These are both Qwen3.5/MoE-related, but one is a missing vision-config field and the other is a generic from_pretrained error; not enough to merge." }, { "left": "issue:43723", "right": "issue:46326", "accept": false, - "reason": "Tokenizer loading regression vs continue_final_message behavior for mistral_common; different bugs." + "reason": "Tokenizer loading in v5 and mistral_common continue_final_message are different tokenizer bugs." }, { "left": "issue:42994", "right": "issue:43525", "accept": false, - "reason": "Quantized model saving failure vs missing config attribute; different problems." + "reason": "Quantized-model save failure and Llama4Config missing pad_token_id are unrelated failures." }, { "left": "issue:43784", "right": "issue:46399", "accept": false, - "reason": "Sentence-transformers import breakage vs Qwen3.5 kernel registration crash; unrelated." + "reason": "Import-time NameError in sentence-transformers and Qwen3.5 kernel-registration crash are different problems." }, { "left": "issue:45310", "right": "issue:45375", "accept": false, - "reason": "Same model family, but different bug types: load failure vs missing vision-config field." + "reason": "One is a Qwen3.5-MoE from_pretrained error; the other is a missing deepstack_visual_indexes field in Qwen3_5MoeVisionConfig." }, { "left": "issue:45081", "right": "issue:45103", "accept": false, - "reason": "Tokenizer regex crash vs auto-docstring crash from future annotations; unrelated." + "reason": "Tokenizer regex patch crash and auto-docstring annotation crash are unrelated attribute errors." }, { "left": "issue:43873", "right": "issue:44945", "accept": false, - "reason": "Quantization/offloading behavior vs pipeline-parallel output corruption; different code paths." + "reason": "Quantization offloading behavior and pipeline-parallel output corruption are different bugs." }, { "left": "issue:46077", "right": "issue:46123", "accept": false, - "reason": "Both involve masking, but one is Gemma4 attention masking and the other is pipeline batching state." + "reason": "Gemma4 attention-mask construction and MaskGenerationPipeline partial-batch handling are separate issues." }, { "left": "issue:42673", "right": "issue:43704", "accept": true, - "reason": "Exact same VRAM leak bug in Qwen3ForCausalLM multi-dataloader threads; 42673 is the stronger canonical report." + "reason": "Exact same Qwen3ForCausalLM VRAM-leak bug title and same failure mode; clearly duplicate reports." }, { "left": "issue:41628", "right": "issue:44351", "accept": false, - "reason": "Both are import errors, but they involve different missing symbols and different regressions." + "reason": "Both are import errors, but they are for different missing exports and different underlying causes." }, { "left": "issue:43232", "right": "issue:44188", "accept": false, - "reason": "Generation kwargs/sync_gpus bug vs torch.compile attention-kernel divergence; not the same issue." + "reason": "Generation kwarg sync behavior and torch.compile attention-kernel branching are different generation bugs." }, { "left": "issue:44964", "right": "issue:45137", "accept": false, - "reason": "Phi-4 multimodal load failure vs DeepSpeed ZeRO3 deque error; unrelated." + "reason": "Phi-4 multimodal loading and DeepSpeed ZeRO3 deque underflow are unrelated." }, { "left": "issue:44222", "right": "issue:45310", "accept": false, - "reason": "FP8 save_pretrained bug vs Qwen3.5 from_pretrained load error; different failures." + "reason": "FP8 save_pretrained on MoE and Qwen3.5 MoE from_pretrained failure are distinct operations and symptoms." }, { "left": "issue:44945", "right": "issue:45753", "accept": false, - "reason": "Pipeline-parallel correctness issue vs image-specific infinite loop; distinct symptoms and code paths." + "reason": "Pipeline-parallel output errors and Qwen3_5 image infinite-loop behavior are not the same bug." }, { "left": "issue:42994", "right": "issue:43824", "accept": false, - "reason": "Quantized saving vs missing model import; unrelated." + "reason": "Quantized model saving and missing Qwen2_5_VLForConditionalGeneration import are unrelated." }, { "left": "issue:44038", "right": "issue:45310", "accept": false, - "reason": "Both touch Qwen MoE and transformers v5, but they are different model variants and different failure modes." + "reason": "Qwen3-VL-Moe compatibility and Qwen3.5-MoE from_pretrained failure are different model/version problems." }, { "left": "issue:36010", "right": "issue:41628", "accept": false, - "reason": "Both are import errors, but they are for different classes and likely different API regressions." + "reason": "Missing GenerationMixin and missing AutoImageProcessor are separate import regressions." }, { "left": "issue:43232", "right": "issue:44792", "accept": false, - "reason": "Generation state bug vs janus image-generation test failure; unrelated." + "reason": "Generation kwargs after sync_gpus and janus image-generation test failure do not look like the same bug." }, { "left": "issue:42831", "right": "issue:44222", "accept": false, - "reason": "Both mention FP8, but one is an accuracy issue and the other is model saving." + "reason": "FineGrainedFP8 accuracy drift and FP8 save_pretrained failure are different FP8-related issues." }, { "left": "issue:42175", "right": "issue:45712", "accept": false, - "reason": "Backend-install packaging issue vs leftover dummy class leakage; different code paths." + "reason": "TensorFlow backend packaging and leftover dummy PT classes are unrelated packaging problems." }, { "left": "issue:44496", "right": "issue:45561", "accept": false, - "reason": "Model config loading failure vs xdist testing race; unrelated." + "reason": "Unrecognized model/config handling and pytest-xdist file-race in testing utils are unrelated." }, { "left": "issue:43526", "right": "issue:45561", "accept": false, - "reason": "Beit image-processor label bug vs testing race; unrelated." + "reason": "BeitImageProcessorFast label reduction bug and xdist race on captured_info.txt are unrelated." }, { "left": "issue:43504", "right": "issue:45561", "accept": false, - "reason": "Beit pretrained-loading legacy-field bug vs testing race; unrelated." + "reason": "Legacy-field pretrained loading for Beit and xdist testing race are unrelated." }, { "left": "issue:42175", "right": "issue:44933", "accept": false, - "reason": "Tensorflow-backend install issue vs image_utils import mismatch; unrelated." + "reason": "TensorFlow backend installation and image_utils import issue are not the same bug." }, { "left": "issue:43377", "right": "issue:44242", "accept": false, - "reason": "Different model behavior bugs: MIMI batching discrepancy vs MoE load-balancing loss omission." + "reason": "MIMI padding-mask batching bug and MoE load-balancing loss omission are different model behaviors." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is well grounded in the packet. The canonical choice of issue:42673 over issue:43704 is supported by the earlier creation date, higher discussion activity, and identical title. The soft-edge verdicts stay conservative overall and do not overstate duplicates where the packet only shows superficial similarity." + "feedback": "The summary is grounded in the packet and stays conservative. The only accepted soft edge, issue:42673 vs issue:43704, is well supported by the exact matching title and similar report content. The other pairwise rejections read appropriately cautious rather than overstated. No mergeability concerns apply because there are no PR pairs." }, "evaluator_used": true, "retried": false, @@ -12766,7 +12697,6 @@ "issue:45125", "issue:45137", "issue:45200", - "issue:45245", "issue:45310", "issue:45356", "issue:45405", @@ -12810,127 +12740,126 @@ "issue:42831|issue:44623", "issue:30333|issue:36331", "issue:30333|issue:35141", - "issue:22355|issue:28282", - "issue:45245|issue:46132" + "issue:22355|issue:28282" ], - "prepared_review_unit_hash": "d25ada5a1b30b975823c85751b13ded869c2afcf22d78031d8a602dc03134946", + "prepared_review_unit_hash": "63045dc476d79f5ef3d39f042f4afdbcd9115f3d56351fd0f6a0a4ff27da2289", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { - "node_count": 48, - "item_count": 48, - "soft_pair_count": 35, - "serialized_chars": 38199, - "estimated_input_tokens": 9550, - "estimated_eval_tokens": 19356 + "node_count": 47, + "item_count": 47, + "soft_pair_count": 34, + "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": "The soft pairs are mostly unrelated issues that only share broad themes like tokenization, save/load, or deepspeed. I would reject all proposed duplicates; there is no clear same-bug pair here.", - "confidence": 0.92, - "canonical_issue_reason": "If a single issue must represent the set, 42898 is the broadest and most generic tokenizer-regression report among the candidates, but the cluster is not actually a duplicate set.", + "summary": "The set is highly heterogeneous: most soft-similarity pairs share broad vocabulary or the same subsystem, but not the same concrete bug. I would not deduplicate any of the candidate pairs.", + "confidence": 0.64, + "canonical_issue_reason": "issue:42898 is the best cluster anchor because it is the clearest, most general v5 tokenizer-regression report and sits near several other tokenizer-related candidates, even though the set as a whole is not a true duplicate cluster.", "canonical_pr_reason": null, - "best_issue_reason": "42898 is the strongest general representative of the tokenizer-regression subgroup: concrete, user-visible, and broader than the more model-specific reports.", + "best_issue_reason": "issue:42898 is the most representative issue among these entries: it is specific, well-scoped, and broadly phrased compared with the more model-specific or unrelated reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:42898", "right": "issue:44568", "accept": false, - "reason": "Both are tokenizer regressions, but they describe different behaviors and different affected models." + "reason": "Both are tokenizer regressions in v5, but they describe different behaviors and different models; not the same bug." }, { "left": "issue:43976", "right": "issue:44933", "accept": false, - "reason": "Python-version incompatibility and a missing import are unrelated failures." + "reason": "Unrelated failures: Python compatibility vs a missing image_utils import." }, { "left": "issue:43278", "right": "issue:45137", "accept": false, - "reason": "Embedding dtype drift and a DeepSpeed deque IndexError are separate bugs." + "reason": "Different code paths: dtype mismatch in evaluation vs a DeepSpeed ZeRO3 deque error." }, { "left": "issue:30990", "right": "issue:35141", "accept": false, - "reason": "Loading hangs and embedding reinitialization are different code paths and symptoms." + "reason": "Stuck loading Sentence Transformers is unrelated to embedding reinitialization after resize." }, { "left": "issue:44623", "right": "issue:45310", "accept": false, - "reason": "Processor save-pretrained missing files is unrelated to Qwen3.5 MoE from_pretrained failure." + "reason": "Save-pretrained missing files and Qwen3.5 MoE from_pretrained errors are different problems." }, { "left": "issue:42222", "right": "issue:45200", "accept": false, - "reason": "Broken vitpose models and Gemma 4 multimodal token defaults are different issues." + "reason": "VitPose breakage and Gemma 4 mm_token_type_ids behavior are unrelated multimodal issues." }, { "left": "issue:30333", "right": "issue:33290", "accept": false, - "reason": "MLflow reporting after failure and Adafactor OOM are unrelated." + "reason": "MLFlow reporting on failed jobs is unrelated to an Adafactor/DeepSpeed OOM." }, { "left": "issue:43479", "right": "issue:44869", "accept": false, - "reason": "Config default initialization and Whisper timestamp decoding are unrelated." + "reason": "Different model components: multimodal config defaults vs Whisper timestamp decoding." }, { "left": "issue:43479", "right": "issue:44079", "accept": false, - "reason": "Multimodal config defaults and ModelOutput key assignment are different subsystems." + "reason": "Config initialization bug is unrelated to ModelOutput key assignment." }, { "left": "issue:42898", "right": "issue:45812", "accept": false, - "reason": "Both are v5 tokenizer regressions, but the affected models and concrete bugs differ." + "reason": "Both are tokenizer regressions, but they affect different tokenization behaviors and different models." }, { "left": "issue:43784", "right": "issue:43824", "accept": false, - "reason": "Importing sentence-transformers and missing Qwen2.5-VL export are unrelated import errors." + "reason": "Sentence-transformers import failure and missing Qwen2.5-VL export are different import errors." }, { "left": "issue:43723", "right": "issue:44568", "accept": false, - "reason": "Tokenizer loading failure and missing BOS/EOS addition are not the same bug." + "reason": "General tokenizer loading issue and a specific BOS/EOS regression are not the same underlying bug." }, { "left": "issue:42898", "right": "issue:44779", "accept": false, - "reason": "Different tokenizer regressions affecting different models and tokenization logic." + "reason": "Both mention v5 tokenizer regressions, but the affected tokenizer and failure mode differ." }, { "left": "issue:42898", "right": "issue:45356", "accept": false, - "reason": "Different tokenizer regression symptoms; one is generic behavior change, the other codec/regex handling." + "reason": "Different tokenizer regressions for different models; not one concrete duplicate." }, { "left": "issue:44016", "right": "issue:46231", "accept": false, - "reason": "Notebook syntax error and documentation parameter typo are unrelated." + "reason": "Notebook syntax error and documentation parameter mismatch are unrelated." }, { "left": "issue:44734", @@ -12942,103 +12871,85 @@ "left": "issue:43874", "right": "issue:45092", "accept": false, - "reason": "Missing image-processor method and remote-code meta-init incompatibility are different multimodal bugs." + "reason": "Different multimodal bugs: missing image patch method vs remote-code checkpoint meta-init incompatibility." }, { "left": "issue:44295", "right": "issue:45656", "accept": false, - "reason": "Position-id buffer access and duplicate optimizer steps are unrelated." + "reason": "Position_ids buffer read error is unrelated to duplicate optimizer stepping under DeepSpeed." }, { "left": "issue:44075", "right": "issue:46231", "accept": false, - "reason": "Optimizer argument handling and a docstring parameter-name mismatch are unrelated." + "reason": "SGD argument handling and a docs-only parameter name mismatch are unrelated." }, { "left": "issue:42898", "right": "issue:44448", "accept": false, - "reason": "Both mention v4-to-v5 output changes, but they are different tokenizer/output regressions." + "reason": "Both are v4-to-v5 regression reports, but for different models and different outputs." }, { "left": "issue:44060", "right": "issue:45125", "accept": false, - "reason": "Incorrect tied-weights warning and missing tensor-parallel plan are distinct Qwen issues." - }, - { - "left": "issue:45405", - "right": "issue:45712", - "accept": false, - "reason": "Dependency version bump and dummy-class cleanup are unrelated maintenance issues." - }, - { - "left": "issue:44222", - "right": "issue:44297", - "accept": false, - "reason": "FP8 save-pretrained and tokenizer save-pretrained bugs affect different components." - }, - { - "left": "issue:44933", - "right": "issue:45712", - "accept": false, - "reason": "Missing image_utils import and leftover dummy classes are not the same defect." - }, - { - "left": "issue:44297", - "right": "issue:45656", - "accept": false, - "reason": "Tokenizer metadata save bug and DeepSpeed optimizer stepping are unrelated." + "reason": "Incorrect tied-weights warning and missing tensor-parallel plan are separate Qwen3 issues." }, { - "left": "issue:44448", - "right": "issue:46132", + "left": "issue:45405", + "right": "issue:45712", "accept": false, - "reason": "Pegasus v4/v5 output regression and AttentionInterface.register behavior change are different bugs." + "reason": "PEFT version pinning and leftover dummy classes are unrelated repo-maintenance issues." }, { "left": "issue:44222", - "right": "issue:44623", + "right": "issue:44297", "accept": false, - "reason": "Model save-pretrained FP8 bug and processor save-pretrained missing files are not the same issue." + "reason": "Both involve save_pretrained, but one is FP8 MoE serialization and the other is tokenizer config metadata." }, { "left": "issue:27957", "right": "issue:30333", "accept": false, - "reason": "Flash Attention support and MLflow reporting after failures are unrelated." + "reason": "Flash Attention 2 on XLM-Roberta is unrelated to MLFlow reporting on failed training jobs." }, { "left": "issue:30333", "right": "issue:34689", "accept": false, - "reason": "Training-job reporting and Llama 3.2 vision model loading are unrelated." + "reason": "Training-job reporting and Llama 3.2 Vision model loading are unrelated." }, { "left": "issue:43479", "right": "issue:43644", "accept": false, - "reason": "Multimodal config defaults and junk-filled buffers are different v5 initialization bugs." + "reason": "Default sub-config initialization and non-persistent buffer filling are different issues." }, { "left": "issue:42831", "right": "issue:44623", "accept": false, - "reason": "FP8 accuracy issues and processor file-saving bugs are unrelated." + "reason": "FP8 accuracy concerns and processor save_pretrained missing files are unrelated." }, { - "left": "issue:22355", - "right": "issue:28282", + "left": "issue:30333", + "right": "issue:36331", "accept": false, - "reason": "Missing transformers.onnx and missing PyTorch are different optional-dependency import errors." + "reason": "MLFlow failure reporting is unrelated to the CustomTrainer signature breakage." }, { - "left": "issue:45245", - "right": "issue:46132", + "left": "issue:30333", + "right": "issue:35141", + "accept": false, + "reason": "Failed-job MLFlow reporting is unrelated to token-embedding resizing behavior." + }, + { + "left": "issue:22355", + "right": "issue:28282", "accept": false, - "reason": "Category-limit runtime error and registered-function behavior change are unrelated." + "reason": "Missing transformers.onnx module and missing PyTorch dependency are different import-time failures." } ] }, @@ -13066,7 +12977,6 @@ "issue:38175", "issue:41628", "issue:42757", - "issue:42831", "issue:42907", "issue:43122", "issue:43232", @@ -13096,6 +13006,7 @@ "issue:45071", "issue:45092", "issue:45103", + "issue:45245", "issue:45310", "issue:45357", "issue:45381", @@ -13104,6 +13015,7 @@ "issue:46002", "issue:46050", "issue:46082", + "issue:46132", "issue:46209", "issue:46305", "issue:46328", @@ -13112,6 +13024,7 @@ ], "soft_pairs": [ "issue:45071|issue:46082", + "issue:45245|issue:46132", "issue:43122|issue:46209", "issue:44297|issue:45310", "issue:41628|issue:42757", @@ -13138,211 +13051,210 @@ "issue:43761|issue:46082", "issue:43644|issue:46305", "issue:44829|issue:46002", - "issue:44261|issue:45092", - "issue:42831|issue:45310" + "issue:44261|issue:45092" ], - "prepared_review_unit_hash": "0fd8afdc594ca25355caf7b55b9a84e37acbd63ee89195963a967a8e057242a8", + "prepared_review_unit_hash": "cc33674823a871156e0cb2d8800fb182ade56d541d5f6029a1583574eb53bf7a", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { - "node_count": 47, - "item_count": 47, + "node_count": 48, + "item_count": 48, "soft_pair_count": 29, - "serialized_chars": 35560, - "estimated_input_tokens": 8890, - "estimated_eval_tokens": 18036 + "serialized_chars": 36136, + "estimated_input_tokens": 9034, + "estimated_eval_tokens": 18324 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Loose cluster of mostly unrelated Transformers issues: v5 regressions, tokenizer/config loading failures, cache/serving bugs, and a few model-specific defects. No soft pair appears to be the same underlying bug or change.", - "confidence": 0.91, - "canonical_issue_reason": "issue:45071 is the broadest and most representative item in the cluster: a clear Transformers v5 regression affecting core config behavior. It\u2019s concise, generic, and closest to the cluster\u2019s shared theme of v5 breakage.", + "summary": "The cluster is highly heterogeneous: most items are unrelated Transformers v5 regressions or model-specific bugs that only share broad themes like loading, saving, generation, or config handling. None of the soft pairs look like the same underlying issue, so I would not merge any of them.", + "confidence": 0.37, + "canonical_issue_reason": "If one issue must serve as the representative, issue 43723 is the broadest and most central v5 loading regression in this set, but the cluster does not really have a true duplicate canonical.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45071 is the best single issue representative because it describes a core framework regression rather than a model-specific symptom or a narrow edge case.", + "best_issue_reason": "Issue 43723 is the best overall representative because it is a concrete, user-facing tokenizer-loading regression on the common pretrained loading path; the other issues are mostly narrower, model-specific, or unrelated.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45071", "right": "issue:46082", "accept": false, - "reason": "Both are config-related v5 problems, but one is broad type-checking behavior and the other is LlamaConfig validation for head_dim; different code paths and fixes." + "reason": "Different bugs: PreTrainedConfig type-checking regression vs LlamaConfig head_dim validation error." + }, + { + "left": "issue:45245", + "right": "issue:46132", + "accept": false, + "reason": "Unrelated failures: category-count limit in tabular encoding vs AttentionInterface.register behavior change." }, { "left": "issue:43122", "right": "issue:46209", "accept": false, - "reason": "Different subsystems and failure modes: tokenizer output changes vs FP8 inference failure on B200." + "reason": "Different code paths and symptoms: tokenizer output changes vs FP8 inference failure on B200." }, { "left": "issue:44297", "right": "issue:45310", "accept": false, - "reason": "One is tokenizer save_pretrained metadata mismatch; the other is a Qwen3.5 from_pretrained model loading error. Related area, not the same bug." + "reason": "Both involve loading/saving, but the concrete bugs differ: tokenizer_class mismatch vs Qwen3.5 MoE from_pretrained failure." }, { "left": "issue:41628", "right": "issue:42757", "accept": false, - "reason": "Both are import errors, but for different symbols from different packages/dependencies; not the same underlying issue." + "reason": "Both are import errors, but for different symbols and different dependencies; not the same underlying breakage." }, { "left": "issue:43475", "right": "issue:46305", "accept": false, - "reason": "Different models and different bugs: missing vision encoder attribute vs a sliding-window mask off-by-one." + "reason": "Different model/components: SAM3 video encoder attribute error vs MoonshineStreaming sliding-window off-by-one." }, { "left": "issue:44190", "right": "issue:46082", "accept": false, - "reason": "Unrelated: local dataset loading in an example script vs LlamaConfig head_dim validation." + "reason": "Unrelated: local dataset loading script issue vs LlamaConfig configuration validation." }, { "left": "issue:43723", "right": "issue:44977", "accept": false, - "reason": "Tokenizer loading failure and flash-attention generation issue are unrelated." + "reason": "Different problems: tokenizer loading regression vs Qwen3.5 flash-attention generation behavior." }, { "left": "issue:44206", "right": "issue:45103", "accept": false, - "reason": "Different failure modes and code paths: feature extractor argument mismatch vs auto_docstring crash with future annotations." + "reason": "Different failure modes: unsupported center arg in feature extractor vs docstring processing crash with future annotations." }, { "left": "issue:45381", "right": "issue:46305", "accept": false, - "reason": "Both mention positional behavior, but they affect different models and different mechanisms; not the same bug." + "reason": "Both are index/position-related, but they affect different pipelines and tensors; not the same bug." }, { "left": "issue:43232", "right": "issue:46424", "accept": false, - "reason": "Both are cache/generation related, but one concerns updating model kwargs after sync_gpus and the other cache_length semantics for quantized cache." + "reason": "Generation kwargs update after sync_gpus is unrelated to cache max_length honoring." }, { "left": "issue:44734", "right": "issue:46421", "accept": false, - "reason": "Serving KV-cache indexing crash vs cache lazy init causing recompilation are separate issues." + "reason": "Both mention cache, but one is wrong tensor indexing during continuation and the other is lazy cache init causing recompilation." }, { "left": "issue:44222", "right": "issue:44484", "accept": false, - "reason": "Different save_pretrained topics: FP8 MoE serialization bug vs shard-size default question." + "reason": "Different save_pretrained concerns: FP8 moe serialization vs shard size default question." }, { "left": "issue:43232", "right": "issue:45834", "accept": false, - "reason": "Generation kwargs handling is unrelated to Kosmos2.5 long OCR index errors." + "reason": "Generation kwarg update bug is unrelated to Kosmos2.5 long OCR input indexing." }, { "left": "issue:44295", "right": "issue:44623", "accept": false, - "reason": "Buffer registration/read issue vs processor save_pretrained missing files; different components and symptoms." + "reason": "Position_ids buffer read error vs processor.save_pretrained missing files are different save/load issues." }, { "left": "issue:44263", "right": "issue:45092", "accept": false, - "reason": "Different model code paths: DSA indexer split behavior vs remote-code checkpoint meta-init incompatibility." + "reason": "torch.split return handling in a MoE indexer is unrelated to meta initialization compatibility for old InternVL2 checkpoints." }, { "left": "issue:44373", "right": "issue:44485", "accept": false, - "reason": "A docstring mismatch is not the same as a RoPE implementation bug." + "reason": "Wrong docstring is documentation-only; GLM-5 RoPE is a runtime implementation issue." }, { "left": "issue:43723", "right": "issue:44871", "accept": false, - "reason": "Tokenizer load regression and eos_token_id config mismatch are distinct problems." + "reason": "Tokenizer loading regression vs eos_token_id config mismatch; related area, but not the same underlying bug." }, { "left": "issue:44483", "right": "issue:45406", "accept": false, - "reason": "One is an API request acceptance issue, the other a processor attribute error in serve; not the same bug." + "reason": "Different server-side failures: request rejection in /v1/chat/completions vs missing _tokenizer on Gemma4Processor." }, { "left": "issue:44279", "right": "issue:46328", "accept": false, - "reason": "Generic dependency issue vs a performance bug in SequenceFeatureExtractor.pad." + "reason": "Dependency issue with transformers is unrelated to the numpy-to-list padding inefficiency." }, { "left": "issue:43723", "right": "issue:44991", "accept": false, - "reason": "Both are tokenizer-loading related, but they concern different models and likely different causes." + "reason": "Both are tokenizer-loading reports, but they do not identify the same concrete failure mode or fix." }, { "left": "issue:42907", "right": "issue:46050", "accept": false, - "reason": "Both involve save/load and quantization, but one is dequantized model saving and the other is sinq round-trip breakage; not the same concrete defect." + "reason": "Different quantization paths and symptoms: dequantized Ministral/Devstral save failure vs sinq save/load breakage." }, { "left": "issue:30990", "right": "issue:38175", "accept": false, - "reason": "Completely different models and symptoms: loading hangs vs zero probabilities." + "reason": "Completely different models and bugs: Sentence Transformers load hang vs SigLIP2 zero probabilities." }, { "left": "issue:44263", "right": "issue:44360", "accept": false, - "reason": "Same broad area (DSA indexer) but different concrete bugs: split return behavior vs missing ReLU." + "reason": "Same broad subsystem name, but different defects in the DSA indexer; not enough to merge." }, { "left": "issue:45092", "right": "issue:45357", "accept": false, - "reason": "Both are model-specific v5 compatibility problems, but they affect different checkpoints and different serialization/init paths." + "reason": "Old InternVL2 meta-init incompatibility is separate from Qwen3.5 save_pretrained visual key serialization." }, { "left": "issue:43761", "right": "issue:46082", "accept": false, - "reason": "Vision model hidden_states output bug vs LlamaConfig head_dim validation are unrelated." + "reason": "CLIPVisionModel hidden_states regression is unrelated to LlamaConfig head_dim validation." }, { "left": "issue:43644", "right": "issue:46305", "accept": false, - "reason": "Non-persistent buffer initialization corruption is unrelated to sliding-window masking." + "reason": "Non-persistent buffer junk fill is unrelated to streaming mask off-by-one behavior." }, { "left": "issue:44829", "right": "issue:46002", "accept": false, - "reason": "Different training pathologies with different likely causes; not the same bug." + "reason": "Different training stability problems with different triggers and model paths." }, { "left": "issue:44261", "right": "issue:45092", "accept": false, - "reason": "RoPE/layernorm precision issue vs remote-code checkpoint meta-init incompatibility are unrelated." - }, - { - "left": "issue:42831", - "right": "issue:45310", - "accept": false, - "reason": "FP8 accuracy degradation is unrelated to Qwen3.5 from_pretrained loading failure." + "reason": "MLA layernorm epsilon precision issue is unrelated to InternVL2 meta-init checkpoint compatibility." } ] }, @@ -13390,7 +13302,6 @@ "issue:43756", "issue:43784", "issue:43824", - "issue:43937", "issue:44038", "issue:44060", "issue:44117", @@ -13406,6 +13317,7 @@ "issue:44913", "issue:44945", "issue:45278", + "issue:45310", "issue:45593", "issue:45753", "issue:46002", @@ -13416,6 +13328,7 @@ "issue:46402" ], "soft_pairs": [ + "issue:42831|issue:45310", "issue:43479|issue:44913", "issue:41628|issue:43824", "issue:43408|issue:46402", @@ -13426,10 +13339,10 @@ "issue:42757|issue:45278", "issue:27957|issue:30990", "issue:42886|issue:43066", - "issue:42886|issue:42994", "issue:44368|issue:46305", - "issue:42831|issue:45753", "issue:22355|issue:28218", + "issue:42886|issue:42994", + "issue:42831|issue:45753", "issue:43408|issue:46369", "issue:43232|issue:45593", "issue:43576|issue:43784", @@ -13444,217 +13357,216 @@ "issue:42907|issue:44945", "issue:44188|issue:44373", "issue:43122|issue:46326", - "issue:41628|issue:45278", - "issue:42907|issue:43937" + "issue:41628|issue:45278" ], - "prepared_review_unit_hash": "6804ae1a4f0db8c3be9bb9383fb06d40d48c0cc9a1b98ec2405f4cb9ae4b1398", + "prepared_review_unit_hash": "505cb83f8d686abd8587ff79dddcba2f005c10348a54c751585cacad16978f3e", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36184, - "estimated_input_tokens": 9046, - "estimated_eval_tokens": 18348 + "serialized_chars": 36196, + "estimated_input_tokens": 9049, + "estimated_eval_tokens": 18354 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose semantic grouping of mostly unrelated Transformers issues (imports, tokenizers, model-specific regressions, and training/runtime bugs). None of the proposed soft pairs look like the same underlying bug/change, so all are rejected. The broad import-regression report is the best representative issue, but only as an umbrella, not a duplicate target for the others.", - "confidence": 0.84, - "canonical_issue_reason": "issue:45278 is the broadest and most representative umbrella for the cluster\u2019s recurring import-regression theme after upgrade, even though it does not duplicate the other issues.", + "summary": "This cluster is mostly a grab bag of distinct Transformers regressions and API issues (imports, tokenizers, config reloads, model-specific bugs). The apparent similarities are broad subsystem overlaps, not the same concrete defect, so no soft-edge duplicate merges are justified.", + "confidence": 0.9, + "canonical_issue_reason": "issue:45278 is the broadest umbrella in the set for version-upgrade import regressions, but the cluster is too heterogeneous for a true duplicate canonical.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45278 is the most suitable cluster representative because it captures a general user-facing regression pattern and can contextualize several related import complaints.", + "best_issue_reason": "issue:45278 is the closest cluster anchor because it is the most general and representative of the import/API-regression subset, even though it does not actually subsume the other reports.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:42831", + "right": "issue:45310", + "accept": false, + "reason": "Different bugs: FP8 accuracy drift vs a Qwen3.5-Moe from_pretrained loading error." + }, { "left": "issue:43479", "right": "issue:44913", "accept": false, - "reason": "Both concern config handling, but they hit different models and different fields; not the same bug." + "reason": "Different config problems: multimodal default sub-config initialization vs GPTNeoX rotary_pct not persisting on reload." }, { "left": "issue:41628", "right": "issue:43824", "accept": false, - "reason": "Both are import errors, but for different missing symbols and different code paths." + "reason": "Both are import failures, but for different symbols/models and different code paths." }, { "left": "issue:43408", "right": "issue:46402", "accept": false, - "reason": "Same sam3 family, but one is a wrong model-type warning and the other is missing text-encoder weights." + "reason": "Different SAM3 issues: a model-type warning vs missing text-encoder weight loading." }, { "left": "issue:42831", "right": "issue:44038", "accept": false, - "reason": "Different model-specific regressions: FP8 accuracy vs Qwen3-VL-Moe / Transformers 5 compatibility." + "reason": "Unrelated: FP8 accuracy issue vs a Qwen3-VL-Moe transformer's from_pretrained bug." }, { "left": "issue:44263", "right": "issue:44373", "accept": false, - "reason": "Unrelated problems: tensor split return values versus a docstring issue." + "reason": "Different scope: torch.split behavior in an indexer vs a docstring mistake for position_ids." }, { "left": "issue:39401", "right": "issue:44568", "accept": false, - "reason": "Both are tokenizer bugs, but they affect different tokenizers and different behaviors (offset mapping vs BOS/EOS insertion)." + "reason": "Both are tokenization-related, but one is wrong offset mappings and the other is missing BOS/EOS insertion." }, { "left": "issue:22355", "right": "issue:27957", "accept": false, - "reason": "ONNX import failure and Flash Attention model support are unrelated." + "reason": "Different areas: missing transformers.onnx module vs XLMRoberta Flash Attention 2 support." }, { "left": "issue:42757", "right": "issue:45278", "accept": false, - "reason": "There is broad import-error overlap, but one is a specific missing hub export and the other is a general post-upgrade import regression; not the same concrete bug." + "reason": "45278 is a broad import-regression umbrella, but this is a specific huggingface_hub symbol import breakage." }, { "left": "issue:27957", "right": "issue:30990", "accept": false, - "reason": "Different subsystems and symptoms: attention backend support vs loading hang." + "reason": "XLMRoberta Flash Attention 2 support is unrelated to Sentence Transformers hanging on load." }, { "left": "issue:42886", "right": "issue:43066", "accept": false, - "reason": "Offline cache loading failure is unrelated to wrong tokenizer decoder type in v5." + "reason": "Different tokenizer bugs: offline cache loading vs wrong decoder type in v5." }, { - "left": "issue:42886", - "right": "issue:42994", + "left": "issue:44368", + "right": "issue:46305", "accept": false, - "reason": "Tokenizer cache/offline behavior and quantized model saving are different bugs." + "reason": "A warning about tie_word_embeddings is unrelated to the MoonshineStreaming off-by-one mask bug." }, { - "left": "issue:44368", - "right": "issue:46305", + "left": "issue:22355", + "right": "issue:28218", "accept": false, - "reason": "A warning about tie_word_embeddings is unrelated to a sliding-window mask off-by-one error." + "reason": "onnx import failure and tokenizer added-space behavior are unrelated bugs." }, { - "left": "issue:42831", - "right": "issue:45753", + "left": "issue:42886", + "right": "issue:42994", "accept": false, - "reason": "Accuracy issue in FP8 is not the same as an infinite-loop multimodal inference bug." + "reason": "Offline tokenizer cache loading and quantized model saving failure are different code paths." }, { - "left": "issue:22355", - "right": "issue:28218", + "left": "issue:42831", + "right": "issue:45753", "accept": false, - "reason": "Missing transformers.onnx module and extra-space tokenization behavior are unrelated." + "reason": "Different failures: accuracy regression in FineGrainedFP8 vs Qwen3_5 infinite loop on a specific image." }, { "left": "issue:43408", "right": "issue:46369", "accept": false, - "reason": "Both touch sam3, but one is a model-type mismatch warning and the other is a memory overflow." + "reason": "Both mention SAM3, but one is a warning about model type and the other is a memory overflow in a different model variant." }, { "left": "issue:43232", "right": "issue:45593", "accept": false, - "reason": "Generation kwargs after sync_gpus and D-FINE auxiliary-loss logic are different code paths." + "reason": "Generation kwargs after sync_gpus is unrelated to D-FINE auxiliary-loss behavior." }, { "left": "issue:43576", "right": "issue:43784", "accept": false, - "reason": "A broken env CLI command is not the same as a NameError during sentence-transformers import." + "reason": "A broken transformers env command is unrelated to a sentence-transformers import NameError." }, { "left": "issue:43646", "right": "issue:46002", "accept": false, - "reason": "Custom model initialization breakage and NaNs in classification heads are distinct failures." + "reason": "Custom model initialization breakage and NaNs in classification heads are different symptoms and likely different fixes." }, { "left": "issue:43756", "right": "issue:44060", "accept": false, - "reason": "Different model-specific issues: RoPE layer retention vs tied-weights warning." + "reason": "Different model-specific regressions: Smollm3 RoPE layer handling vs Qwen3-Next tied-weights warning." }, { "left": "issue:43550", "right": "issue:46132", "accept": false, - "reason": "torch.compile SDPA failure is unrelated to AttentionInterface.register changing function behavior." + "reason": "torch.compile SDPA failure is not the same as AttentionInterface.register changing behavior." }, { "left": "issue:44295", "right": "issue:44297", "accept": false, - "reason": "Reading a registered buffer and tokenizer save metadata mismatch are different bugs." + "reason": "These are separate tokenizer/position-id issues: buffer access error vs tokenizer_class mismatch on save." }, { "left": "issue:44117", "right": "issue:44704", "accept": false, - "reason": "One is None handling in tokenizer mapping; the other is missing kwargs propagation in AutoProcessor." + "reason": "TOKENIZER_MAPPING_NAMES returning None is a different failure from AutoProcessor not forwarding kwargs." }, { "left": "issue:30990", "right": "issue:33357", "accept": false, - "reason": "Loading stuck and a MacOS bus error are different symptoms with different likely causes." + "reason": "Sentence Transformers loading hangs and a MacOS bus error with CLIP are distinct runtime failures." }, { "left": "issue:42371", "right": "issue:44263", "accept": false, - "reason": "TF32 configuration guidance is unrelated to torch.split return values." + "reason": "TF32 settings guidance and torch.split return values in GlmMoeDsaIndexer are unrelated." }, { "left": "issue:43122", "right": "issue:44521", "accept": false, - "reason": "Different tokenizer/template bugs: tokenization drift vs assistant mask generation for multimodal inputs." + "reason": "Different tokenizer behavior: version-dependent tokenization drift vs multimodal assistant_masks all-zero." }, { "left": "issue:42907", "right": "issue:44945", "accept": false, - "reason": "Quantized model saving failure and pipeline-parallel wrong output are unrelated runtime issues." + "reason": "Quantized model saving failure and incorrect pipeline-parallel LLM output are separate bugs." }, { "left": "issue:44188", "right": "issue:44373", "accept": false, - "reason": "Compiler-dependent attention-kernel divergence is not the same as a docstring problem." + "reason": "torch.compile attention-kernel divergence is unrelated to a docstring error." }, { "left": "issue:43122", "right": "issue:46326", "accept": false, - "reason": "Both involve tokenization, but one is tokenization drift across versions and the other is continue_final_message for mistral_common tokenizers." + "reason": "Tokenization drift across versions and continue_final_message for mistral_common tokenizers are different defects." }, { "left": "issue:41628", "right": "issue:45278", "accept": false, - "reason": "Both mention import errors, but they are not the same concrete missing import or code path." - }, - { - "left": "issue:42907", - "right": "issue:43937", - "accept": false, - "reason": "Saving dequantized models and invalid GenerationConfig are different failure modes." + "reason": "Both are import problems, but 45278 is a broad report of many import errors while 41628 is a specific missing AutoImageProcessor symbol." } ] }, @@ -13697,6 +13609,7 @@ "issue:43493", "issue:43784", "issue:43824", + "issue:43937", "issue:43976", "issue:43992", "issue:44038", @@ -13727,11 +13640,12 @@ "issue:45753" ], "soft_pairs": [ + "issue:42907|issue:43937", "issue:39401|issue:44488", "issue:44261|issue:44373", "issue:43976|issue:44945", - "issue:44704|issue:45290", "issue:43784|issue:45600", + "issue:44704|issue:45290", "issue:42907|issue:44964", "issue:39401|issue:44521", "issue:43116|issue:45200", @@ -13757,214 +13671,214 @@ "issue:22355|issue:29127", "issue:43992|issue:44704" ], - "prepared_review_unit_hash": "0cd1e82f7b2798d8db01e6155f1f0ceafb3286877dcd29c680dfb6fcaddf4a2d", + "prepared_review_unit_hash": "61f86d457a6a738d8d5600be048587a42e89c5ef077b34a937c47a9d1741eb7f", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { - "node_count": 47, - "item_count": 47, - "soft_pair_count": 29, - "serialized_chars": 35476, - "estimated_input_tokens": 8869, - "estimated_eval_tokens": 17994 + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36310, + "estimated_input_tokens": 9078, + "estimated_eval_tokens": 18412 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The set is mostly a collection of unrelated issues across tokenizer regressions, model-loading bugs, training/inference problems, and packaging/import errors. I found no strong duplicate pair with enough evidence to merge; the best anchor issue is the broad tokenization-regression report in 43122.", - "confidence": 0.84, - "canonical_issue_reason": "Issue 43122 is the clearest broad regression report: it defines a version boundary, a reproducible tokenizer behavior change, and is more general than the model-specific or symptom-specific reports around it.", + "summary": "Most candidate edges are not true duplicates; they connect issues from different subsystems, models, or failure modes. I reject all soft edges. If one issue has to represent the cluster, issue 43122 is the broadest tokenizer-regression report, but it is not a duplicate of the others.", + "confidence": 0.66, + "canonical_issue_reason": "Issue 43122 is the broadest, most representative regression report in this set (tokenization changes across versions), while most other issues are narrower, model-specific, or unrelated.", "canonical_pr_reason": null, - "best_issue_reason": "43122 is the strongest standalone issue to use as an anchor because it is concrete, versioned, and likely to attract follow-up analysis without being overly niche.", + "best_issue_reason": "Issue 43122 best serves as the cluster\u2019s representative because it captures a generic tokenizer-regression pattern and has more cross-references than the other candidates.", "best_pr_reason": null, "soft_edge_verdicts": [ + { + "left": "issue:42907", + "right": "issue:43937", + "accept": false, + "reason": "Different failure modes: dequantized save/load behavior vs GenerationConfig validation." + }, { "left": "issue:39401", "right": "issue:44488", "accept": false, - "reason": "Both are about Qwen-related text behavior, but one is an offset-mapping bug and the other is a model-loading failure; different code paths." + "reason": "Qwen3 tokenizer offset mapping bug and a separate model-loading failure are unrelated." }, { "left": "issue:44261", "right": "issue:44373", "accept": false, - "reason": "One is a numerical/config bug in MLA layernorm, the other is a docstring issue for position_ids; not the same defect." + "reason": "LayerNorm epsilon precision issue and a docstring problem are not the same bug." }, { "left": "issue:43976", "right": "issue:44945", "accept": false, - "reason": "Python compatibility/install failure vs incorrect output under pipeline parallelism; unrelated symptoms and fixes." + "reason": "Python-version compatibility and pipeline-parallelism output corruption are distinct issues." }, { - "left": "issue:44704", - "right": "issue:45290", + "left": "issue:43784", + "right": "issue:45600", "accept": false, - "reason": "AutoProcessor kwargs forwarding and chat-template tool-call handling are distinct APIs with different failure modes." + "reason": "Import-time NameError in sentence-transformers is unrelated to stale auto_mappings references." }, { - "left": "issue:43784", - "right": "issue:45600", + "left": "issue:44704", + "right": "issue:45290", "accept": false, - "reason": "Import-time NameError in sentence-transformers integration vs CI break from stale auto-mapping references; not the same bug." + "reason": "cached_file kwargs forwarding and chat-template tool-call crashes are different code paths." }, { "left": "issue:42907", "right": "issue:44964", "accept": false, - "reason": "Save/dequantization failure for Ministral/Devstral is unrelated to loading Phi-4 multimodal." + "reason": "Dequantized save_pretrained failure and Phi-4 multimodal loading failure are unrelated." }, { "left": "issue:39401", "right": "issue:44521", "accept": false, - "reason": "Both touch Qwen3.5 tokenization/chat templating, but offset_mapping and assistant_masks are different behaviors and code paths." + "reason": "Offset mapping correctness and multimodal assistant_masks generation are different tokenizer behaviors." }, { "left": "issue:43116", "right": "issue:45200", "accept": false, - "reason": "Example script multi-label output bug vs Gemma 4 multimodal token-type defaulting; unrelated functionality." + "reason": "Example-script empty outputs and mm_token_type_ids defaults are not the same underlying bug." }, { "left": "issue:44038", "right": "issue:44297", "accept": false, - "reason": "Qwen3-VL-Moe loading bug and tokenizer_config save-pretrained mismatch are separate issues." + "reason": "Qwen3-VL-Moe issue and tokenizer_class serialization mismatch are separate problems." }, { "left": "issue:43122", "right": "issue:44514", "accept": false, - "reason": "Versioned tokenizer regression vs batched chat-template crash with padding=False; different failure surfaces." + "reason": "Both are tokenizer-related, but one is a version-wide regression and the other is a batched chat-template crash." }, { "left": "issue:44625", "right": "issue:44829", "accept": false, - "reason": "Config propagation for Qwen3.5 labels is unrelated to flash_attention_3 training degeneration." + "reason": "num_labels propagation and flash_attention_3 training degeneration are unrelated." }, { "left": "issue:44360", "right": "issue:44485", "accept": false, - "reason": "A missing ReLU in a DSA indexer is unrelated to a GLM-5 RoPE implementation discussion." + "reason": "DSA indexer activation bug and GLM-5 RoPE implementation issue are not duplicates." }, { "left": "issue:44038", "right": "issue:44295", "accept": false, - "reason": "Model-load compatibility issue vs a position_ids buffer-read error; no shared concrete bug path." + "reason": "Qwen3-VL-Moe loading behavior and position_ids buffer access are unrelated." }, { "left": "issue:43408", "right": "issue:43475", "accept": false, - "reason": "Both involve SAM 3 Video, but one is a config-type warning and the other is a missing attribute in encoder output; too different to merge." + "reason": "Model-type warning and missing fpn_position_embeddings attribute are separate SAM3 issues." }, { "left": "issue:42907", "right": "issue:43452", "accept": false, - "reason": "Dequantized save failure and gguf_file load failure are different serialization paths." + "reason": "Dequantized save/load behavior and gguf_file loading failures do not share the same bug." }, { "left": "issue:43122", "right": "issue:44779", "accept": false, - "reason": "Both mention tokenizer regressions in v5, but Deepseek-specific incorrect outputs are not the same concrete bug as the generic slow/fast-tokenizer inconsistency." + "reason": "Both mention tokenizer regressions, but they affect different models and likely different code paths." }, { "left": "issue:44038", "right": "issue:44222", "accept": false, - "reason": "Qwen3-VL-Moe loading regression and FP8 save_pretrained bug are unrelated model lifecycle failures." + "reason": "Qwen3-VL-Moe load issue and FP8 save_pretrained moe issue are different concrete failures." }, { "left": "issue:45216", "right": "issue:45478", "accept": false, - "reason": "Both concern Qwen3.5 and 5.4.0, but save_pretrained corruption and from_pretrained errors are different code paths and not clearly the same defect." + "reason": "Related model/version, but save_pretrained corruption and from_pretrained load error are not clearly the same bug." }, { "left": "issue:43824", "right": "issue:44351", "accept": false, - "reason": "Both are import errors, but they involve different missing symbols from different modules and likely different causes." + "reason": "Missing Qwen2_5_VLForConditionalGeneration import and missing HybridCache import are unrelated import errors." }, { "left": "issue:36296", "right": "issue:43116", "accept": false, - "reason": "Tensor-parallel training bug is unrelated to example-script multi-label classification behavior." + "reason": "Tensor parallel training bug and multi-label classification example behavior are unrelated." }, { "left": "issue:42757", "right": "issue:44351", "accept": false, - "reason": "Missing huggingface_hub symbol and missing transformers symbol are separate dependency/API breakages." + "reason": "huggingface_hub import failure and transformers import failure are different missing-symbol problems." }, { "left": "issue:22355", "right": "issue:29159", "accept": false, - "reason": "Module import failure for transformers.onnx is unrelated to tokenizer slow/fast inconsistency." + "reason": "Missing transformers.onnx module and slow/fast tokenizer inconsistency are unrelated." }, { "left": "issue:43278", "right": "issue:44945", "accept": false, - "reason": "Dtype drift between training and evaluate is unrelated to incorrect outputs under pipeline parallelism." + "reason": "Embedding dtype drift in evaluate and pipeline-parallelism output errors are not the same issue." }, { "left": "issue:42175", "right": "issue:44938", "accept": false, - "reason": "Packaging extras missing TensorFlow and Python 3.14 load failure are different environment issues." + "reason": "Package-extras backend omission and Python 3.14 load failure are unrelated." }, { "left": "issue:43493", "right": "issue:45753", "accept": false, - "reason": "SigLIP2 implementation mismatch and Qwen3_5 infinite-loop-on-image are different model-specific problems." + "reason": "SigLIP2 implementation discrepancy and Qwen3_5 infinite-loop behavior are different problems." }, { "left": "issue:43066", "right": "issue:44779", "accept": false, - "reason": "Wrong tokenizer decoder type and Deepseek tokenization regression may both touch tokenizers, but they are not the same concrete failure." + "reason": "Both are tokenizer regressions in v5, but the reported symptoms and models differ too much to treat as one bug." }, { "left": "issue:42673", "right": "issue:45072", "accept": false, - "reason": "VRAM leak in multi-threaded Qwen3ForCausalLM use is unrelated to dtype mismatches in bfloat16 inference." + "reason": "VRAM leak in dataloader-thread usage and bfloat16 dtype mismatches are unrelated." }, { "left": "issue:22355", "right": "issue:29127", "accept": false, - "reason": "Import error for transformers.onnx and a LayoutLMv3 error-message clarity issue are unrelated." - }, - { - "left": "issue:43278", - "right": "issue:44945", - "accept": false, - "reason": "Embedding dtype change and pipeline-parallel wrong output are distinct training/inference bugs." + "reason": "Missing module import and LayoutLMv3 error-message clarity are not duplicates." }, { "left": "issue:43992", "right": "issue:44704", "accept": false, - "reason": "Missing embed_tokens.weight on load is unrelated to AutoProcessor kwargs forwarding." + "reason": "Missing embed_tokens.weight on load and cached_file kwarg forwarding are different issues." } ] }, @@ -14069,214 +13983,214 @@ "issue:44117|issue:45081", "issue:43122|issue:46298" ], - "prepared_review_unit_hash": "8e7ad74bfe1e0056e9cdc5fa01c3c248cf6b3886ec704be0788872e4d99cdba2", + "prepared_review_unit_hash": "c3334f1916fd2b48bc42a1bc07591c99501a68ccad829ca96ec1978705874c1d", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 30, - "serialized_chars": 36562, - "estimated_input_tokens": 9141, - "estimated_eval_tokens": 18538 + "serialized_chars": 36566, + "estimated_input_tokens": 9142, + "estimated_eval_tokens": 18540 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is a grab-bag of mostly unrelated Transformers bugs/regressions. The broadest representative is the v5 import-regression report, while the soft edge candidates are mostly similar only at the subsystem/model-family level and should remain separate.", - "confidence": 0.78, - "canonical_issue_reason": "issue:45278 is the most general and representative bug report in this set: a broad post-upgrade import-regression affecting multiple symbols, rather than a narrow model- or feature-specific failure.", + "summary": "The set is a heterogeneous collection of Transformers v5 regression issues across unrelated subsystems (tokenizers, model loading, training, attention, docs, etc.). The soft links are mostly superficial title similarity rather than true duplicates. No PRs were provided.", + "confidence": 0.87, + "canonical_issue_reason": "issue:45278 is the broadest, most representative regression report in the set: it captures a general v5 upgrade breakage theme and is more suitable as an umbrella than the narrower model-specific bugs.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45278 best captures the cluster\u2019s dominant theme of v5 upgrade breakage and is broad enough to stand in for the general regression class.", + "best_issue_reason": "issue:45278 is the best single issue representative because it is high-level, version-focused, and likely the easiest to route or cluster against compared with the many narrow, model-specific reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:38617", "right": "issue:45278", "accept": false, - "reason": "Both are import errors after a version upgrade, but 38617 is a specific missing symbol while 45278 is a broader multi-import regression; they do not look like the same concrete bug." + "reason": "Both are import-related, but 45278 is a broad upgrade regression while 38617 is a specific missing symbol ImportError; not the same concrete bug." }, { "left": "issue:43937", "right": "issue:44485", "accept": false, - "reason": "Both mention GLM-5, but one is a GenerationConfig validation failure and the other is a RoPE implementation discussion; different failures and code paths." + "reason": "Same model family (GLM-5) but different failures: invalid GenerationConfig vs RoPE implementation discussion." }, { "left": "issue:43611", "right": "issue:46002", "accept": false, - "reason": "Loading with base_model_prefix and NaNs in classification heads are unrelated symptoms with different root causes." + "reason": "Unrelated bugs: model loading with base_model_prefix vs NaNs in classification heads." }, { "left": "issue:44188", "right": "issue:46328", "accept": false, - "reason": "Attention-kernel divergence under torch.compile and a padding performance issue are unrelated." + "reason": "Different subsystems entirely: attention kernel divergence vs SequenceFeatureExtractor.pad performance." }, { "left": "issue:44355", "right": "issue:46231", "accept": false, - "reason": "Compiled Python-file errors and a documentation parameter-name mismatch are not the same underlying bug." + "reason": "Compiled Python file errors are unrelated to the set_training() docs parameter mismatch." }, { "left": "issue:43116", "right": "issue:43493", "accept": false, - "reason": "Example-script multi-label output failure and SigLIP2 implementation discrepancy are different model/problem domains." + "reason": "Example script multi-label empty results and SigLIP2 implementation discrepancy are unrelated." }, { "left": "issue:42222", "right": "issue:43441", "accept": false, - "reason": "VitPose loading breakage and Ministral FlashAttention failure are separate model-specific regressions." + "reason": "Both are model-specific issues, but vitpose broken imports and Ministral-3 flash-attention failures are different bugs." }, { "left": "issue:42371", "right": "issue:43874", "accept": false, - "reason": "TF32 API guidance and a missing image-processor method are unrelated." + "reason": "TF32 API settings and a missing image-patch helper in GLM46V are unrelated changes." }, { "left": "issue:44075", "right": "issue:44936", "accept": false, - "reason": "Optimizer-argument handling and trainer.evaluate() failing after train() are distinct training bugs." + "reason": "Optimizer argument handling is unrelated to trainer.evaluate() failing after trainer.train()." }, { "left": "issue:45357", "right": "issue:45478", "accept": false, - "reason": "Both concern Qwen3.5/MoE, but one is save_pretrained key serialization and the other is from_pretrained loading failure; not the same concrete defect." + "reason": "Same model family, but save_pretrained key serialization and from_pretrained loading failure are not the same concrete bug." }, { "left": "issue:44038", "right": "issue:45478", "accept": false, - "reason": "Related model family, but these are different Qwen variants and different failure modes." + "reason": "Both mention Qwen3/Qwen3.5 MoE, but they concern different model variants and different failure modes." }, { "left": "issue:43976", "right": "issue:45446", "accept": false, - "reason": "Python-version compatibility and an incorrect PyTorch version check are unrelated." + "reason": "Python version compatibility and a PyTorch version check bug are unrelated." }, { "left": "issue:43116", "right": "issue:43825", "accept": false, - "reason": "An example-script classification result bug is unrelated to a pipeline error-message/docs issue." + "reason": "Classification example output bug and pipeline removal/error-message issue are different problems." }, { "left": "issue:45083", "right": "issue:45446", "accept": false, - "reason": "Feature-extraction length behavior and a flex_attention import/version check are unrelated." + "reason": "Feature-extraction length behavior in qwen3_omni_moe is unrelated to the flex_attention import/version check." }, { "left": "issue:43064", "right": "issue:43673", "accept": false, - "reason": "Optimizer-state corruption under FSDP2/PEFT and missing GenerationMixin cache in chunked prefill are different code paths." + "reason": "Training/FSDP optimizer-state corruption and GenerationMixin cache missing in chunked_prefill are unrelated." }, { "left": "issue:33453", "right": "issue:39401", "accept": false, - "reason": "Both are tokenizer-related regressions, but one is loading and the other is offset_mapping correctness; not the same bug." + "reason": "Both are tokenizer regressions, but loading regression and wrong offset_mapping are distinct bugs." }, { "left": "issue:43937", "right": "issue:45563", "accept": false, - "reason": "GenerationConfig invalidity and a stale warning about num_return_sequences are unrelated." + "reason": "GLM-5 invalid GenerationConfig and a stale warning for num_return_sequences are unrelated." }, { "left": "issue:39401", "right": "issue:44779", "accept": false, - "reason": "Different tokenizer regressions in different model families; no sign of the same underlying defect." + "reason": "Both are tokenizer correctness issues, but they affect different tokenizers and different failure modes." }, { "left": "issue:44117", "right": "issue:45588", "accept": false, - "reason": "Tokenizer mapping fallback handling and a flash_attention crash on s_aux=None are unrelated." + "reason": "TOKENIZER_MAPPING_NAMES returning None is unrelated to the flash_attention crash on s_aux=None." }, { "left": "issue:44380", "right": "issue:46077", "accept": false, - "reason": "Both involve attention behavior, but one is backend scaling under SDPA/FlashAttention and the other is Gemma4 bidirectional masking; different concrete bugs." + "reason": "Both involve attention behavior, but GPT2 backend scaling and Gemma4 bidirectional-mask construction are different bugs." }, { "left": "issue:42673", "right": "issue:43906", "accept": false, - "reason": "VRAM leakage in multi-threaded dataloader use and an isolated reproduction of another issue are not enough to treat as the same bug." + "reason": "VRAM leak in Qwen3ForCausalLM and a reproduction of another issue are not the same underlying bug." }, { "left": "issue:43827", "right": "issue:45446", "accept": false, - "reason": "Outdated docs referencing pipeline() and a PyTorch version check bug are unrelated." + "reason": "Docs still referencing pipeline() and a flex_attention PyTorch version check are unrelated." }, { "left": "issue:43450", "right": "issue:43986", "accept": false, - "reason": "Batched video processor shape handling and a missing torchvision crash are distinct video-processing issues." + "reason": "Video processor batched-shape bug and AutoProcessor crash without torchvision are different issues." }, { "left": "issue:43673", "right": "issue:46328", "accept": false, - "reason": "Generation cache missing in chunked prefill and a padding performance inefficiency are unrelated." + "reason": "Generation cache handling and SequenceFeatureExtractor.pad performance are unrelated." }, { "left": "issue:44291", "right": "issue:45071", "accept": false, - "reason": "init_empty_weights constructor-arg breakage and PretrainedConfig type-checking regression are both v5 regressions but not the same code-path problem." + "reason": "Both are model-loading/config-related regressions, but unexpected _is_hf_initialized argument and PretrainedConfig type checking are distinct problems." }, { "left": "issue:44466", "right": "issue:46002", "accept": false, - "reason": "lm_head.weight serialization inconsistency and NaNs in classification heads are unrelated." + "reason": "lm_head serialization inconsistency and NaNs in classification heads are unrelated." }, { "left": "issue:43937", "right": "issue:44977", "accept": false, - "reason": "GenerationConfig validation and a flash-attention generation failure in a different model are separate issues." + "reason": "GLM-5 invalid GenerationConfig and Qwen3.5 flash-attention generation failure are different bugs." }, { "left": "issue:45405", "right": "issue:45446", "accept": false, - "reason": "A dependency version pinning problem and an import-version check bug are unrelated." + "reason": "Package version bumping for PEFT and a PyTorch version check in flex_attention are unrelated." }, { "left": "issue:44117", "right": "issue:45081", "accept": false, - "reason": "Tokenizer mapping None-handling and a Mistral regex patch crash are both tokenizer-loading adjacent, but they are different failures and should not be merged." + "reason": "Tokenizer mapping fallback bug and Mistral regex patch crash are unrelated." }, { "left": "issue:43122", "right": "issue:46298", "accept": false, - "reason": "Tokenizer output changes and a noisy logging message are unrelated." + "reason": "Tokenizer behavior change across versions and a noisy logging message are not the same issue." } ] }, @@ -14357,8 +14271,8 @@ "issue:42222|issue:46178", "issue:44829|issue:45216", "issue:42617|issue:45341", - "issue:45362|issue:46328", "issue:42673|issue:43582", + "issue:45362|issue:46328", "issue:43408|issue:43742", "issue:39401|issue:43421", "issue:42490|issue:42673", @@ -14382,20 +14296,20 @@ "issue:44263|issue:44485", "issue:45071|issue:45542" ], - "prepared_review_unit_hash": "28929648be1f9614d2836d7689ec2cadbe958fd16249b2ce19a36f11804e1e48", + "prepared_review_unit_hash": "5206a7682bbcdcc2d031ebac25621e2a3f0dcc6acf586bb8bf9e69d42b191901", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 31, - "serialized_chars": 36530, + "serialized_chars": 36531, "estimated_input_tokens": 9133, "estimated_eval_tokens": 18522 }, @@ -14404,208 +14318,205 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly false positives across unrelated regressions, import errors, model-specific bugs, and docs/perf issues. Only the tokenizer pair 43066/43122 looks like the same v5 tokenizer regression.", - "confidence": 0.79, - "canonical_issue_reason": "Issue 43066 is the clearest root-cause report for the tokenizer v5 regression; 43122 reads like the same symptom reported from a different angle.", + "summary": "This is a loose thematic cluster of Transformers issues, not a true duplicate set: it mixes import/API regressions, tokenizer behavior changes, model-loading failures, backend/install problems, precision warnings, and testing utility bugs. I would not merge any of the soft edges.", + "confidence": 0.93, + "canonical_issue_reason": "No single canonical issue is safe to choose; the cluster is heterogeneous and the items are not the same underlying bug.", "canonical_pr_reason": null, - "best_issue_reason": "43066 is the best single issue to keep: it is specific, actionable, and can subsume the duplicate symptom in 43122.", + "best_issue_reason": "If forced, issue 45278 is the broadest representative of the import-regression subtheme, but it still does not represent most of the cluster.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44410", "right": "issue:45310", "accept": false, - "reason": "Different model families and failures: missing projections in qwen3next vs Qwen3.5 MoE from_pretrained error." + "reason": "Different Qwen bugs: missing projections in one model vs a from_pretrained loading error in another." }, { "left": "issue:45341", "right": "issue:45561", "accept": false, - "reason": "Both mention testing utils, but one is a vague bug report and the other is a specific xdist race on captured_info.txt; not clearly the same bug." + "reason": "Different test-utils problems: a generic bug in testing_utils.py vs a pytest-xdist file race." }, { "left": "issue:38617", "right": "issue:41628", "accept": false, - "reason": "Both are import errors, but for different symbols and code paths." + "reason": "Both are import errors, but they concern different missing symbols and different code paths." }, { "left": "issue:45341", "right": "issue:46328", "accept": false, - "reason": "Testing utils bug vs a padding performance issue in SequenceFeatureExtractor; unrelated." + "reason": "One is a correctness/race issue in test helpers; the other is a performance inefficiency in feature padding." }, { "left": "issue:42222", "right": "issue:46178", "accept": false, - "reason": "VitPose model breakage and tensorflow_text backend mapping are unrelated." + "reason": "Unrelated: vitpose model breakage vs a TensorFlow backend mapping error." }, { "left": "issue:44829", "right": "issue:45216", "accept": false, - "reason": "Flash-attention training degeneration and Qwen3.5 save_pretrained checkpoint corruption are different bugs." + "reason": "Different regressions: FA3 training degeneracy vs Qwen3.5 save_pretrained checkpoint corruption." }, { "left": "issue:42617", "right": "issue:45341", "accept": false, - "reason": "A script launch failure is unrelated to a testing_utils bug." + "reason": "A script execution problem is not the same bug as a testing_utils.py issue." }, { - "left": "issue:45362", - "right": "issue:46328", + "left": "issue:42673", + "right": "issue:43582", "accept": false, - "reason": "Qwen3.5 chat crash vs padding performance; no shared codepath." + "reason": "Different failures: VRAM leak under threaded loading vs a TypeError on Apple Silicon warmup." }, { - "left": "issue:42673", - "right": "issue:43582", + "left": "issue:45362", + "right": "issue:46328", "accept": false, - "reason": "VRAM leak in dataloader threads and an Apple Silicon warmup TypeError are separate issues." + "reason": "Qwen chat crash and feature-extractor padding inefficiency are unrelated." }, { "left": "issue:43408", "right": "issue:43742", "accept": false, - "reason": "Different loading warnings/errors for different model types." + "reason": "A model-type warning is not the same as a key error when loading a specific checkpoint." }, { "left": "issue:39401", "right": "issue:43421", "accept": false, - "reason": "Tokenizer offset mapping bug vs runtime special-token post-processor update; related area but not the same defect." + "reason": "Both touch tokenization, but offset_mapping is a different bug from runtime post-processor updates." }, { "left": "issue:42490", "right": "issue:42673", "accept": false, - "reason": "Local path slash import warning and VRAM leak are unrelated." + "reason": "A local-path import warning is unrelated to a multi-thread VRAM leak." }, { "left": "issue:44877", "right": "issue:46231", "accept": false, - "reason": "Strict config loading vs a documentation parameter-name mismatch." + "reason": "Config loading strictness and a documentation parameter typo are different issues." }, { "left": "issue:42371", "right": "issue:44261", "accept": false, - "reason": "TF32 API warning and MLA rms_norm_eps precision issue are unrelated." + "reason": "TF32 behavior guidance is unrelated to the MLA rms_norm_eps precision bug." }, { "left": "issue:44484", "right": "issue:45310", "accept": false, - "reason": "Shard-size question vs Qwen3.5 MoE loading error; different concerns." + "reason": "A question about save_pretrained shard size is not the same as a Qwen3.5 from_pretrained failure." }, { "left": "issue:38617", "right": "issue:44220", "accept": false, - "reason": "ImportError for a missing config_utils symbol vs an fbank feature-extraction bug." + "reason": "ImportError for a missing symbol and an _torch_extract_fbank_features bug are different code paths." }, { "left": "issue:43937", "right": "issue:44496", "accept": false, - "reason": "GenerationConfig validation failure and unrecognized model loading are different issues." + "reason": "GenerationConfig validation failure is distinct from an unrecognized-model config.json error." }, { "left": "issue:44933", "right": "issue:45278", "accept": false, - "reason": "A specific missing image_utils import is not the same as a broad many-import-errors regression." + "reason": "Related theme of imports, but not the same concrete missing-symbol bug." }, { "left": "issue:43066", "right": "issue:43122", - "accept": true, - "reason": "Both point to the same v5 tokenizer regression: changed decoder/type behavior leading to different tokenization." + "accept": false, + "reason": "Both are tokenizer regressions, but one is decoder type and the other is output tokenization differences." }, { "left": "issue:42222", "right": "issue:45542", "accept": false, - "reason": "VitPose model breakage and missing TensorFlow backend after tensorboard install are unrelated." + "reason": "Different backend problems: model breakage vs TensorFlow being missing despite tensorboard being installed." }, { "left": "issue:44410", "right": "issue:45478", "accept": false, - "reason": "Same model family, but different concrete failures: missing projections vs Qwen3.5 MoE from_pretrained error." + "reason": "Different Qwen3.5 failures with different symptoms and likely different code paths." }, { "left": "issue:44483", "right": "issue:45464", "accept": false, - "reason": "Different chat/completions API failures and likely different request/streaming paths." + "reason": "Both involve chat/completions, but one is request validation and the other is streaming inference failure." }, { "left": "issue:42371", "right": "issue:42831", "accept": false, - "reason": "TF32 behavior warning and FineGrainedFP8 accuracy regression are unrelated." + "reason": "TF32 API guidance is unrelated to FineGrainedFP8 accuracy degradation." }, { "left": "issue:43012", "right": "issue:43577", "accept": false, - "reason": "Both involve dtype/precision, but one is a compile warning and the other is incorrect model dtype propagation on load." + "reason": "A compile-time float32 warning is not the same as model dtype actually remaining float32 on load." }, { "left": "issue:44075", "right": "issue:45656", "accept": false, - "reason": "Optimizer args ignored and optimizer step called twice are distinct optimizer bugs." + "reason": "SGD argument handling and double optimizer stepping under deepspeed are different bugs." }, { "left": "issue:42175", "right": "issue:42222", "accept": false, - "reason": "TensorFlow backend installation issue vs broken VitPose models; unrelated." + "reason": "Backend dependency installation and vitpose model breakage are unrelated." }, { "left": "issue:43097", "right": "issue:44493", "accept": false, - "reason": "Removed tie_embeddings API and unexpected key warnings from position ids are different regressions." + "reason": "Removed API warning and unexpected position-id keys are different regressions." }, { "left": "issue:43673", "right": "issue:45341", "accept": false, - "reason": "Generation cache missing in chunked_prefill is unrelated to a testing_utils bug." + "reason": "Chunked-prefill cache handling is unrelated to the testing_utils bug." }, { "left": "issue:41628", "right": "issue:45071", "accept": false, - "reason": "Import error for AutoImageProcessor vs PretrainedConfig type-checking regression; not the same bug." + "reason": "An import error for AutoImageProcessor is not the same as a PreTrainedConfig type-checking regression." }, { "left": "issue:44263", "right": "issue:44485", "accept": false, - "reason": "torch.split return-value issue in GlmMoeDsaIndexer vs GLM-5 RoPE implementation; different codepaths." + "reason": "These are separate GLM-5 issues: torch.split return handling vs RoPE implementation details." }, { "left": "issue:45071", "right": "issue:45542", "accept": false, - "reason": "PretrainedConfig type-checking regression and undefined tf backend error are unrelated." + "reason": "Type-checking behavior and TensorFlow backend installation are unrelated." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The clustering is mostly conservative and grounded in the packet. The only accepted pair, 43066/43122, is plausibly the same tokenizer-v5 regression based on the titles, and the other pair judgments stay on the cautious side of false unless there is stronger body-level evidence not shown here." - }, - "evaluator_used": true, + "evaluator_result": null, + "evaluator_used": false, "retried": false, - "accepted_nontrivial_soft_edge": true, + "accepted_nontrivial_soft_edge": false, "error_kind": null, "error_message": null }, @@ -14679,8 +14590,8 @@ "issue:44360|issue:46172", "issue:43441|issue:44829", "issue:43931|issue:44038", - "issue:43335|issue:45071", "issue:43097|issue:43452", + "issue:43335|issue:45071", "issue:45003|issue:45341", "issue:43673|issue:46123", "issue:43986|issue:45341", @@ -14690,29 +14601,29 @@ "issue:43986|issue:44484", "issue:43874|issue:44263", "issue:43986|issue:45372", - "issue:45440|issue:46172", "issue:43531|issue:45341", + "issue:45440|issue:46172", "issue:43097|issue:46050", - "issue:44261|issue:44485", "issue:43097|issue:45588", - "issue:43064|issue:44315", + "issue:44261|issue:44485", "issue:38617|issue:42757", + "issue:43064|issue:44315", "issue:44062|issue:44117", "issue:44410|issue:45375", - "issue:44038|issue:44410", "issue:43502|issue:43992", + "issue:44038|issue:44410", "issue:42890|issue:46298", "issue:38617|issue:44351", "issue:43097|issue:43873" ], - "prepared_review_unit_hash": "153be6571f80e530131df75892e1404c769ee256f50b6b994c369558a13da3b9", + "prepared_review_unit_hash": "a3064feab6b6b2bb8d58fba6a334efbcd8f87897c6297c714f750f37fd467f34", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 47, @@ -14727,210 +14638,210 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly soft-similarity false positives: the paired issues generally share broad themes like FlashAttention, import errors, or model-loading failures, but not the same concrete bug or code path. I do not see a true duplicate pair here.", - "confidence": 0.93, - "canonical_issue_reason": "No single issue is a clear canonical duplicate target: the items span unrelated bugs across different models and subsystems. If a representative issue is needed, it should be a broad, active bug report rather than one of the narrow model-specific cases.", + "summary": "This cluster is mostly a grab-bag of unrelated Transformers regressions: import/API breakages, model-specific loading mismatches, FlashAttention issues, and a few training/pipeline bugs. I don\u2019t see any soft pair that looks like the same concrete underlying bug, so all candidate duplicates are rejected.", + "confidence": 0.79, + "canonical_issue_reason": "No single issue is a strong canonical duplicate for the cluster; if one must be chosen, issue 43873 is the least-bad representative because it is the broadest active bug report with the most discussion/inbound references.", "canonical_pr_reason": null, - "best_issue_reason": "issue:43873 is the strongest standalone issue in the set: it is open, has substantial discussion/inbound references, and describes a substantive, broadly relevant bug rather than a narrow one-off failure.", + "best_issue_reason": "Issue 43873 is the best representative by scope and activity, but only weakly: it covers a broad quantization/offloading regression and has the strongest signal of being a hub among these otherwise unrelated reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:27957", "right": "issue:43352", "accept": false, - "reason": "Both mention Flash Attention 2, but they concern different models and likely different model-compatibility checks." + "reason": "Both are FlashAttention-related, but they involve different models and different failure conditions; not the same code-path bug." }, { "left": "issue:43874", "right": "issue:44261", "accept": false, - "reason": "Different problems: a missing image-patch method in GLM46V versus an MLA layernorm config precision issue." + "reason": "Different bugs in different components: image patch counting vs MLA q_a_layernorm config precision." }, { "left": "issue:44484", "right": "issue:45478", "accept": false, - "reason": "Unrelated: shard-size documentation/question versus a Qwen3.5 MoE loading error." + "reason": "Unrelated symptoms: shard-size question vs Qwen3.5 MoE from_pretrained load error." }, { "left": "issue:44263", "right": "issue:46172", "accept": false, - "reason": "Different code paths and models: GlmMoeDsaIndexer split handling versus DeepSeek comb transpose logic." + "reason": "Different model internals and different failure modes; no shared underlying defect." }, { "left": "issue:43931", "right": "issue:44410", "accept": false, - "reason": "Both are Qwen-family model bugs, but one is a weight-shape mismatch and the other is missing projections in layer 0." + "reason": "Both concern Qwen-family models, but one is a weight-shape mismatch and the other is missing projections in a different model; not the same bug." }, { "left": "issue:44360", "right": "issue:46172", "accept": false, - "reason": "DSA indexer ReLU concern is not the same bug as the DeepSeek-v4 comb dtype/transpose issue." + "reason": "DSA ReLU concern is unrelated to the DeepSeek comb transpose issue." }, { "left": "issue:43441", "right": "issue:44829", "accept": false, - "reason": "Both involve FlashAttention, but one is a model-specific failure and the other is degenerate training with flash_attention_3." + "reason": "Both mention FlashAttention, but one is model support/loading and the other is degenerate training with flash_attention_3; different problems." }, { "left": "issue:43931", "right": "issue:44038", "accept": false, - "reason": "Same general Qwen/VL area, but different failures: shape mismatch versus a Transformers v5 compatibility bug." + "reason": "Same general model family, but different concrete breakages: Qwen3-VL weight mismatch vs Qwen3-VL-MoE import/load failure." }, { - "left": "issue:43335", - "right": "issue:45071", + "left": "issue:43097", + "right": "issue:43452", "accept": false, - "reason": "SwitchTransformers sparse-layer construction and PretrainedConfig type checking are unrelated bugs." + "reason": "Different regressions: a removed config field vs gguf_file handling in tokenizer/model loading." }, { - "left": "issue:43097", - "right": "issue:43452", + "left": "issue:43335", + "right": "issue:45071", "accept": false, - "reason": "An API-removal/documentation issue is not the same as gguf_file breaking tokenizer/model loading." + "reason": "SwitchTransformers layer creation bug is unrelated to PreTrainedConfig type checking." }, { "left": "issue:45003", "right": "issue:45341", "accept": false, - "reason": "Different files and symptoms: unsafe sys.modules access versus a testing_utils bug." + "reason": "Unsafe sys.modules access is unrelated to the testing_utils bug." }, { "left": "issue:43673", "right": "issue:46123", "accept": false, - "reason": "Both affect generation/pipeline behavior, but the cache-missing chunked_prefill bug is distinct from the final-batch is_last bug." + "reason": "Generation cache handling and MaskGenerationPipeline final-batch handling are different code paths." }, { "left": "issue:43986", "right": "issue:45341", "accept": false, - "reason": "A video AutoProcessor import crash is unrelated to the testing_utils issue." + "reason": "Video AutoProcessor/torchvision crash is unrelated to the testing_utils issue." }, { "left": "issue:44933", "right": "issue:46231", "accept": false, - "reason": "Nonexistent image_utils import and a doc parameter-name mismatch are different classes of issues." + "reason": "Nonexistent import from image_utils is a code bug; set_training docs parameter mismatch is documentation-only." }, { "left": "issue:45753", "right": "issue:46369", "accept": false, - "reason": "Different models and failures: Qwen3_5 infinite loop on one image versus SAM3-lite memory overflow." + "reason": "Different multimodal failures with different models and symptoms; no shared underlying bug." }, { "left": "issue:44117", "right": "issue:44987", "accept": false, - "reason": "Tokenizer-mapping None handling is not the same as the physical-intelligence/fast loading failure." + "reason": "Tokenizer mapping None-handling is a different failure than loading physical-intelligence/fast." }, { "left": "issue:43986", "right": "issue:44484", "accept": false, - "reason": "Completely different topics: torchvision-dependent processor loading versus shard-size behavior." + "reason": "Completely different topics: missing torchvision vs shard-size behavior." }, { "left": "issue:43874", "right": "issue:44263", "accept": false, - "reason": "Different GLM46V vs GlmMoeDsaIndexer failures; same broad area only by naming." + "reason": "Unrelated GLM46V image patch API bug vs GlmMoeDsaIndexer split semantics." }, { "left": "issue:43986", "right": "issue:45372", "accept": false, - "reason": "Both are loading/import failures, but they involve different optional dependencies and different processors/models." + "reason": "Missing torchvision crash is unrelated to the mistral_common ReasoningEffort import error." + }, + { + "left": "issue:43531", + "right": "issue:45341", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior has no connection to the testing_utils bug." }, { "left": "issue:45440", "right": "issue:46172", "accept": false, - "reason": "Native DeepseekV3 divergence is a model-implementation mismatch, not the DeepSeek-v4 comb transpose bug." + "reason": "DeepSeekV3MoE divergence from remote implementation is a broader model parity issue, not the same as the comb transpose bug." }, { - "left": "issue:43531", - "right": "issue:45341", + "left": "issue:43097", + "right": "issue:46050", "accept": false, - "reason": "Qwen3-MoE sliding_window behavior is unrelated to the testing_utils bug." + "reason": "Both are save/load related, but one is an API removal and the other is a quantized model persistence regression." }, { "left": "issue:43097", - "right": "issue:46050", + "right": "issue:45588", "accept": false, - "reason": "A v5 API removal issue and a quantization save/load regression are different bugs." + "reason": "Removed tie_embeddings API and flash_attention s_aux=None crash are unrelated bugs." }, { "left": "issue:44261", "right": "issue:44485", "accept": false, - "reason": "Both are model-specific implementation discussions, but the missing rms_norm_eps bug is distinct from GLM-5 RoPE implementation." + "reason": "RMS norm epsilon precision discussion and GLM-5 RoPE implementation are different model math issues with no clear shared defect." }, { - "left": "issue:43097", - "right": "issue:45588", + "left": "issue:38617", + "right": "issue:42757", "accept": false, - "reason": "Tie-embeddings removal and flash_attention.py crashing on sink-less models are unrelated." + "reason": "Both are import errors, but they involve different missing symbols from different modules and do not look like the same bug." }, { "left": "issue:43064", "right": "issue:44315", "accept": false, - "reason": "FSDP2/PEFT optimizer-state corruption is a different trainer bug from Liger Kernel not being applied with model_init." - }, - { - "left": "issue:38617", - "right": "issue:42757", - "accept": false, - "reason": "Both are ImportError reports, but they are for different missing symbols in different packages." + "reason": "Optimizer-state corruption under FSDP2/PEFT is unrelated to Liger Kernel not being applied with model_init." }, { "left": "issue:44062", "right": "issue:44117", "accept": false, - "reason": "AddedToken special-argument conflict and tokenizer-mapping None handling are separate tokenizer bugs." + "reason": "AddedToken argument duplication and tokenizer mapping None are distinct tokenizer-layer bugs." }, { "left": "issue:44410", "right": "issue:45375", "accept": false, - "reason": "Different Qwen3-family config/model issues: missing projections versus a silently dropped config field." + "reason": "Missing attn projections in qwen3next and dropped deepstack_visual_indexes in Qwen3_5MoeVisionConfig are different config/model issues." }, { - "left": "issue:44038", - "right": "issue:44410", + "left": "issue:43502", + "right": "issue:43992", "accept": false, - "reason": "Both involve Qwen3 variants, but one is a Transformers v5 loading break and the other is missing layer projections." + "reason": "local_files_only network leakage and missing embed_tokens.weight are unrelated loading bugs." }, { - "left": "issue:43502", - "right": "issue:43992", + "left": "issue:44038", + "right": "issue:44410", "accept": false, - "reason": "local_files_only network leakage and missing embed_tokens.weight are different from load-path bugs." + "reason": "Both touch Qwen3 variants, but one is a Qwen3-VL-Moe from_pretrained issue and the other is missing projections in qwen3next; not the same code path." }, { "left": "issue:42890", "right": "issue:46298", "accept": false, - "reason": "Test flakiness from missing set_seed() is unrelated to a noisy logging message." + "reason": "Test seeding failures and noisy logging are unrelated." }, { "left": "issue:38617", "right": "issue:44351", "accept": false, - "reason": "Two different import failures; they do not point to the same missing symbol or code path." + "reason": "Different missing imports from different APIs; same symptom type, not the same underlying bug." }, { - "left": "issue:43097", + "left": "issue:43064", "right": "issue:43873", "accept": false, - "reason": "Tie-embeddings removal and quantization offloading problems are different subsystems and bugs." + "reason": "FSDP2/PEFT optimizer-state corruption and quantization offloading are different training-system issues." } ] }, @@ -15037,226 +14948,226 @@ "issue:45600|issue:46298", "issue:44038|issue:44230" ], - "prepared_review_unit_hash": "c132f81ef0451e713aba86b97a079d4487e77844d69ddb1229c8902e47073b7b", + "prepared_review_unit_hash": "f478899ba0bb9a519c3823b9164d8a5cf1857ac2711d799affcdf3e77ebd17b3", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 32, - "serialized_chars": 37093, - "estimated_input_tokens": 9274, - "estimated_eval_tokens": 18804 + "serialized_chars": 37089, + "estimated_input_tokens": 9273, + "estimated_eval_tokens": 18802 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The soft pairs are overwhelmingly false positives: they span unrelated import errors, model-specific config regressions, backend/compile issues, and save/load behaviors. I don\u2019t see any pair that clearly shares the same underlying bug or change.", - "confidence": 0.93, - "canonical_issue_reason": "Issue #43824 is the cleanest high-signal regression report here: a specific ImportError, a clear user-facing failure mode, and multiple inbound references. It is more concrete and representative than the other candidates.", + "summary": "No soft pair looks like the same underlying bug/change, so I would not merge any of these issues as duplicates. The cluster is mostly a mix of unrelated import errors, model-specific regressions, and a few attention/compile/config bugs.", + "confidence": 0.84, + "canonical_issue_reason": "Issue 45103 is the cleanest standalone bug report: it has a precise crash condition, a clear subsystem, and a concrete reproduction path, making it the best canonical issue among this set.", "canonical_pr_reason": null, - "best_issue_reason": "#43824 is the strongest single issue to anchor this set because it has a precise failure, visible impact, and enough discussion/references to be actionable.", + "best_issue_reason": "45103 is the most suitable representative issue because it is narrowly scoped, actionable, and currently open, while the rest are largely unrelated single-bug reports.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44704", "right": "issue:45103", "accept": false, - "reason": "Different failures: kwargs propagation into cached_file vs an auto-docstring crash under future annotations." + "reason": "Different subsystems and failures: AutoProcessor kwargs propagation vs auto_docstring crash with future annotations." }, { "left": "issue:44380", "right": "issue:46421", "accept": false, - "reason": "Both touch model/runtime behavior, but one is SDPA/FlashAttention scaling semantics and the other is cache lazy initialization/recompilation." + "reason": "One is GPT2 attention scaling ignored in SDPA/FlashAttention; the other is lazy cache init causing recompilation with chunked prefill." }, { "left": "issue:43986", "right": "issue:44117", "accept": false, - "reason": "One is an AutoProcessor crash without torchvision; the other is TOKENIZER_MAPPING_NAMES returning None. Different code paths." + "reason": "Video AutoProcessor/torchvision import crash is unrelated to TOKENIZER_MAPPING_NAMES returning None." }, { "left": "issue:43824", "right": "issue:43986", "accept": false, - "reason": "Both are import/loading-related, but one is a missing model class export and the other is a video-model AutoProcessor crash tied to torchvision." + "reason": "Qwen2_5_VL class import failure is distinct from AutoProcessor failing without torchvision." }, { "left": "issue:44938", "right": "issue:46178", "accept": false, - "reason": "Different platform/backend issues: Python 3.14 import failure vs an invalid BACKENDS_MAPPING entry for tensorflow_text." + "reason": "Python 3.14 import/load failure is unrelated to a backend mapping error for tensorflow_text." }, { "left": "issue:43066", "right": "issue:43906", "accept": false, - "reason": "No evidence they share the same bug; 43906 is only a reproduction issue, not the same tokenizer-decoder regression." + "reason": "Wrong tokenizer decoder type is not the same as an isolated reproduction of another issue." }, { "left": "issue:43208", "right": "issue:43550", "accept": false, - "reason": "xLSTM training bugs and Bamba torch.compile/SDPA failure are unrelated model-specific issues." + "reason": "xLSTM training blockers and Bamba SDPA compile failure are different models and code paths." }, { "left": "issue:43992", "right": "issue:44484", "accept": false, - "reason": "Missing embed_tokens weights during from_pretrained is unrelated to the max_shard_size default in save_pretrained." + "reason": "Missing embed_tokens weight on UMT5Encoder is unrelated to max_shard_size behavior in save_pretrained()." }, { "left": "issue:43012", "right": "issue:43550", "accept": false, - "reason": "A PyTorch precision warning on bfloat16 is not the same as a compile-time SDPA failure." + "reason": "A PyTorch precision warning during compile is not the same bug as Bamba SDPA compile failure." }, { "left": "issue:43012", "right": "issue:44368", "accept": false, - "reason": "A warning about float32 precision is unrelated to tie_word_embeddings/tie_embeddings warnings in Qwen3.5 fine-tuning." + "reason": "Float32 warning on compile and tie_word_embeddings warning are unrelated configuration/runtime behaviors." }, { "left": "issue:43825", "right": "issue:44655", "accept": false, - "reason": "One is a misleading pipeline task message; the other is failure to save Pipeline objects. Different bugs." + "reason": "Translation-task pipeline error is unrelated to saving Pipeline objects with save_pretrained()." }, { "left": "issue:45712", "right": "issue:46298", "accept": false, - "reason": "Left is leftover dummy class cleanup; right is noisy logging. Same area of hygiene, not same defect." + "reason": "Leftover dummy classes failing repo checks is not the same as a noisy logging message." }, { "left": "issue:42907", "right": "issue:44655", "accept": false, - "reason": "Saving dequantized Ministrals/Devstrals and saving Pipeline objects are different save_pretrained failures." + "reason": "Dequantized model save failure is different from generic pipeline serialization failure." }, { "left": "issue:42907", "right": "issue:45753", "accept": false, - "reason": "A dequantized save failure is unrelated to an infinite loop on a specific image input." + "reason": "Model save failure and Qwen3_5 image infinite loop are unrelated." }, { "left": "issue:43097", "right": "issue:46298", "accept": false, - "reason": "Removed tie_embeddings_and_encoder_decoder behavior is a functional API change, not a generic noisy logging issue." + "reason": "Removed tie_embeddings_and_encoder_decoder is a behavior/regression issue, not a logging-noise issue." }, { "left": "issue:43352", "right": "issue:44380", "accept": false, - "reason": "Nemotron FlashAttention-2 support and GPT2 attention scaling under SDPA/FlashAttention are different backend/model problems." + "reason": "Unsupported Flash Attention 2.0 in NemotronH is unrelated to GPT2 attention scaling being ignored." }, { "left": "issue:43352", "right": "issue:43976", "accept": false, - "reason": "FlashAttention support for a model is unrelated to Python version compatibility in Transformers 5.1.0." + "reason": "Flash Attention support in a model is unrelated to a Python version compatibility problem." }, { "left": "issue:44162", "right": "issue:46172", "accept": false, - "reason": "ESM2 breakage and a deepseek-v4 tensor expression bug are different model-specific regressions." + "reason": "ESM2 breakage and a DeepSeek tensor transpose issue do not share the same failure mode." }, { "left": "issue:44380", "right": "issue:45479", "accept": false, - "reason": "Attention scaling under backend dispatch is unrelated to a zero-loss classification bug." + "reason": "Attention scaling config loss is unrelated to a degenerate zero-loss classification bug." }, { "left": "issue:44380", "right": "issue:46132", "accept": false, - "reason": "GPT2 scaling under SDPA/FlashAttention is not the same as AttentionInterface.register changing registered-function behavior." + "reason": "GPT2 backend scaling bug is distinct from AttentionInterface.register changing function behavior." }, { "left": "issue:43064", "right": "issue:44380", "accept": false, - "reason": "FSDP2/PEFT optimizer-state corruption is unrelated to attention scaling being ignored under backend kernels." + "reason": "FSDP2/PEFT optimizer-state corruption is unrelated to GPT2 attention scaling being ignored." }, { "left": "issue:44483", "right": "issue:44514", "accept": false, - "reason": "Chat/completions request rejection and Qwen2_5_VLProcessor batching crash are different API surfaces and failures." + "reason": "Chat completions API rejection and Qwen2_5_VLProcessor batch-padding crash are different endpoints/code paths." }, { "left": "issue:44222", "right": "issue:45478", "accept": false, - "reason": "Both mention Qwen3.5 MoE, but one is FP8 save_pretrained and the other is from_pretrained loading; not the same concrete bug." + "reason": "FP8 save_pretrained behavior is different from a Qwen3.5 MoE from_pretrained error." }, { "left": "issue:36683", "right": "issue:43582", "accept": false, - "reason": "Gemma3Config missing vocab_size and Apple Silicon allocator warmup TypeError are unrelated." + "reason": "Gemma3Config missing vocab_size and Apple Silicon caching_allocator_warmup TypeError are unrelated." }, { "left": "issue:43208", "right": "issue:43526", "accept": false, - "reason": "xLSTM training blockers and BeitImageProcessorFast reduce_labels returning one label are unrelated." + "reason": "xLSTM training bugs do not match BeitImageProcessorFast reduce_labels behavior." }, { "left": "issue:44936", "right": "issue:46365", "accept": false, - "reason": "trainer.evaluate() after train() failing and inability to import Trainer with datasets installed are different trainer/import issues." + "reason": "trainer.evaluate() failing after train is not the same as failing to import Trainer until datasets is uninstalled." }, { "left": "issue:42490", "right": "issue:43012", "accept": false, - "reason": "Trailing-slash local model path import warning is unrelated to a bfloat16 precision warning." + "reason": "Trailing-slash local model import warning is unrelated to a bfloat16 compile warning." }, { "left": "issue:43012", "right": "issue:44683", "accept": false, - "reason": "Precision warning on bfloat16 and compiled flex_attention failing on torch>=2.9 are different bugs." + "reason": "General compile warning and compiled flex_attention failure on newer torch versions are different bugs." }, { "left": "issue:42757", "right": "issue:43824", "accept": false, - "reason": "Missing is_offline_mode and missing Qwen2_5_VLForConditionalGeneration are unrelated import errors." + "reason": "Missing is_offline_mode import from huggingface_hub is unrelated to missing Qwen2_5_VLForConditionalGeneration export." }, { "left": "issue:38617", "right": "issue:43502", "accept": false, - "reason": "layer_type_validation import removal and API requests despite local_files_only=True are different regressions." + "reason": "layer_type_validation import breakage is unrelated to local_files_only still allowing API requests." }, { "left": "issue:45600", "right": "issue:46298", "accept": false, - "reason": "Broken auto_mappings CI references and noisy logging message are not the same issue." + "reason": "Broken auto_mappings references and a noisy logging message are not the same issue." }, { "left": "issue:44038", "right": "issue:44230", "accept": false, - "reason": "Qwen3-VL-Moe bug in Transformers 5.0 and FP8 inference support for Qwen3.5 MoE are different requests/failures." + "reason": "Generic Qwen3-VL-Moe breakage and fp8 inference support are related in model family only, not the same concrete bug." } ] }, @@ -15335,10 +15246,10 @@ "issue:43618|issue:43761", "issue:44483|issue:45290", "issue:43299|issue:43867", - "issue:43756|issue:44485", "issue:36683|issue:43475", "issue:42994|issue:44655", "issue:43066|issue:46209", + "issue:43756|issue:44485", "issue:45276|issue:46077", "issue:43901|issue:44655", "issue:36010|issue:42757", @@ -15364,232 +15275,232 @@ "issue:43867|issue:46202", "issue:43867|issue:45375" ], - "prepared_review_unit_hash": "49e00c70d9d71d2393921f79d237f13bd3eeb6af3403af66826955288f8c5c60", + "prepared_review_unit_hash": "79ce4c8f80d2e8929e55232626f12177563a029b38a6b0a1861fe26298e59010", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 33, - "serialized_chars": 37169, - "estimated_input_tokens": 9293, - "estimated_eval_tokens": 18842 + "serialized_chars": 37164, + "estimated_input_tokens": 9291, + "estimated_eval_tokens": 18838 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "The cluster is mostly heterogeneous: many items are Transformer v5 regressions, but they touch different models, configs, and code paths. I did not find any soft pair that looks like the same underlying bug/change.", - "confidence": 0.29, - "canonical_issue_reason": "Issue 45698 is the broadest and most actively discussed remaining issue, centered on save/load behavior for custom modules after `save_pretrained`/`from_pretrained`. It is the best anchor only for that subset, not for the whole mixed cluster.", + "summary": "The cluster is a loose mix of Transformers v5 regressions and unrelated import/load/save bugs. None of the soft pairs are the same concrete defect, so they should stay separate rather than be collapsed into one duplicate set.", + "confidence": 0.92, + "canonical_issue_reason": "No true canonical issue: the items are heterogeneous. If forced to pick one umbrella, issue 45278 is the broadest import-regression bucket and is the most reusable triage anchor among the set.", "canonical_pr_reason": null, - "best_issue_reason": "45698 is the strongest standalone issue in the set: open, high discussion activity, and a concrete reproducible save/load regression with clear user impact.", + "best_issue_reason": "Issue 45278 is the best representative only as a broad umbrella for import failures after upgrading; the rest of the cluster is too diverse to treat as one bug.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45698", "right": "issue:46172", "accept": false, - "reason": "Different bugs: custom module reload after saving vs a DeepSeek-V4 tensor transpose/dtype issue." + "reason": "Save/load custom-module corruption and a deepseek-v4 tensor transpose bug are different code paths and symptoms." }, { "left": "issue:43618", "right": "issue:44380", "accept": false, - "reason": "Both mention attention, but one is missing CLIPOutput attentions and the other ignores GPT-2 attention scaling under SDPA/FlashAttention; different code paths." + "reason": "Missing CLIP attentions and ignored GPT2 attention scaling under SDPA/FlashAttention are unrelated regressions." }, { "left": "issue:43618", "right": "issue:43761", "accept": false, - "reason": "Same model family, but separate regressions: CLIPOutput.attentions not assigned vs CLIPVisionModel.hidden_states becoming None." + "reason": "Both involve CLIP/v5 output behavior, but one drops attentions while the other drops hidden_states; different concrete defects." }, { "left": "issue:44483", "right": "issue:45290", "accept": false, - "reason": "API request validation failure vs chat template tokenization crash; not the same bug." + "reason": "An API request rejection and a chat-template crash on tool-call messages are different layers and bugs." }, { "left": "issue:43299", "right": "issue:43867", "accept": false, - "reason": "Qwen3VL MoE loading failure and a generic sorted-state_dict load error are different root causes." - }, - { - "left": "issue:43756", - "right": "issue:44485", - "accept": false, - "reason": "Smollm3 RoPE layer-count mismatch and GLM-5 RoPE implementation discussion are not the same concrete defect." + "reason": "Qwen3VL MoE loading breakage and a state_dict sorting load error are not the same failure mode." }, { "left": "issue:36683", "right": "issue:43475", "accept": false, - "reason": "Different missing attributes on different objects/models: `Gemma3Config.vocab_size` vs `Sam3VisionEncoderOutput.fpn_position_embeddings`." + "reason": "Missing Gemma3Config vocab_size and missing Sam3VisionEncoderOutput fpn_position_embeddings are unrelated attribute errors." }, { "left": "issue:42994", "right": "issue:44655", "accept": false, - "reason": "Quantized model saving failure and pipeline `save_pretrained` failure are related only broadly to saving, but not the same bug." + "reason": "Quantized model saving failure and Pipeline save_pretrained failure are both save-related but not the same bug." }, { "left": "issue:43066", "right": "issue:46209", "accept": false, - "reason": "Tokenizer decoder-type regression vs FP8 inference failure on B200; unrelated symptoms and subsystems." + "reason": "Wrong tokenizer decoder type and FP8 inference failure on B200 are unrelated issues." + }, + { + "left": "issue:43756", + "right": "issue:44485", + "accept": false, + "reason": "Different RoPE-related model bugs in different models; not the same concrete implementation defect." }, { "left": "issue:45276", "right": "issue:46077", "accept": false, - "reason": "Both are Gemma4 issues, but one is resize_token_embeddings propagation and the other is bidirectional attention-mask construction." + "reason": "Gemma4 embedding resize not propagating and Gemma4 attention-mask construction are distinct code paths." }, { "left": "issue:43901", "right": "issue:44655", "accept": false, - "reason": "Documentation mismatch for `return_all_scores` is not the same as a runtime pipeline serialization bug." + "reason": "Docs wording for return_all_scores and Pipeline serialization are unrelated." }, { "left": "issue:36010", "right": "issue:42757", "accept": false, - "reason": "Both are import errors, but for different symbols from different packages (`transformers.generation` vs `huggingface_hub`)." + "reason": "Both are import errors, but they concern different symbols and different dependency breakages." }, { "left": "issue:43611", "right": "issue:44655", "accept": false, - "reason": "Different loading/saving problems: `base_model_prefix` handling vs pipeline serialization." + "reason": "Base_model_prefix loading regression and Pipeline save_pretrained failure are not the same bug." }, { "left": "issue:43352", "right": "issue:43937", "accept": false, - "reason": "Flash Attention support gating for Nemotron is unrelated to GLM-5 generation-config validation." + "reason": "FlashAttention support limitation and invalid GenerationConfig are unrelated model/config failures." }, { "left": "issue:36683", "right": "issue:44077", "accept": false, - "reason": "Missing config attribute vs a restriction on optional `post_init`; not the same failure." + "reason": "A missing config attribute and a deprecation/allowlist issue in patchtsmixer are different bugs." }, { "left": "issue:43867", "right": "issue:45216", "accept": false, - "reason": "Both involve model loading/saving, but one is state_dict ordering and the other is incorrect Qwen3.5 checkpoint contents after `save_pretrained`." + "reason": "State_dict ordering causing load failure is not the same as an incorrect checkpoint produced by save_pretrained." }, { "left": "issue:44625", "right": "issue:46077", "accept": false, - "reason": "`num_labels` propagation in Qwen3.5 is unrelated to Gemma4 attention-mask behavior." + "reason": "Qwen3.5 num_labels propagation and Gemma4 bidirectional-attention masking are unrelated." }, { "left": "issue:43976", "right": "issue:44938", "accept": false, - "reason": "Different Python-version compatibility bugs on different interpreter versions and with different causes." + "reason": "These are both version-compatibility problems, but for different Python versions and different failures." }, { "left": "issue:45584", "right": "issue:46365", "accept": false, - "reason": "Whisper generation after special-token alignment is unrelated to `Trainer` import being blocked by `datasets`." + "reason": "Whisper empty-transcription generation and a Trainer import conflict with datasets are unrelated." }, { "left": "issue:42175", "right": "issue:46365", "accept": false, - "reason": "Dependency/backend installation problem vs import conflict with `datasets`; not the same underlying bug." + "reason": "Missing TensorFlow backend in an install extra is unrelated to the Trainer import conflict." }, { "left": "issue:43352", "right": "issue:45593", "accept": false, - "reason": "Flash Attention unsupported warning and missing auxiliary losses in D-FINE are unrelated." + "reason": "No FlashAttention 2 support and missing auxiliary losses are unrelated model bugs." }, { "left": "issue:46077", "right": "issue:46082", "accept": false, - "reason": "Gemma4 attention-mask bug vs LlamaConfig head-dim validation are different model/config problems." + "reason": "Gemma4 causal-mask behavior and LlamaConfig head_dim validation are different configuration/architecture issues." }, { "left": "issue:36683", "right": "issue:43525", "accept": false, - "reason": "Both are missing config attributes, but they affect different models and different fields; too broad to merge." + "reason": "Two missing config attributes on different model families are similar in shape but not the same underlying bug." }, { "left": "issue:43618", "right": "issue:45216", "accept": false, - "reason": "CLIP attentions missing and Qwen3.5 saved checkpoint corruption are unrelated." + "reason": "Missing CLIP attentions and incorrect Qwen3.5 saved checkpoints are unrelated regressions." }, { "left": "issue:44077", "right": "issue:46365", "accept": false, - "reason": "`patchtsmixer` init-policy change is unrelated to a `Trainer` import conflict caused by `datasets`." + "reason": "patchtsmixer post_init policy and the datasets-induced Trainer import error are unrelated." }, { "left": "issue:36683", "right": "issue:44242", "accept": false, - "reason": "Missing `vocab_size` and missing load-balancing loss are different bugs in different subsystems." + "reason": "A missing Gemma3 config attribute and a missing load-balancing loss are different defects." }, { "left": "issue:44075", "right": "issue:46365", "accept": false, - "reason": "Optimizer argument handling has nothing to do with `Trainer` import resolution." + "reason": "SGD arguments being ignored and a Trainer import conflict are unrelated." }, { "left": "issue:43992", "right": "issue:45310", "accept": false, - "reason": "Different models and different errors: UMT5 missing weights vs Qwen3.5 MoE `from_pretrained` failure." + "reason": "UMT5Encoder missing embed_tokens.weight and Qwen3.5 MoE from_pretrained failure are different loading bugs." }, { "left": "issue:43064", "right": "issue:44492", "accept": false, - "reason": "Distributed optimizer-state bug and a typo in cache strategies are unrelated." + "reason": "A distributed optimizer-state bug and a typo in cache strategies are not the same issue." }, { "left": "issue:43010", "right": "issue:45278", "accept": false, - "reason": "`@torch.no_grad` decoration for cache updates is not the same as broad import failures after upgrade." + "reason": "Cache update no_grad behavior and broad import errors after upgrade do not share the same code path." }, { "left": "issue:44355", "right": "issue:44855", "accept": false, - "reason": "Compiled-file import errors and a Python 3.13 decorator/comment parsing regression are different root causes." + "reason": "Compiled-file execution errors and a Python 3.13 JIT parsing/indentation problem are distinct." }, { "left": "issue:43867", "right": "issue:46202", "accept": false, - "reason": "Generic sorted-state_dict load failure vs Gemma pipeline roundtrip on the underscore token are unrelated." + "reason": "State_dict-sorting load failure and Gemma pipeline roundtrip failure on \u2581 are different serialization bugs." }, { "left": "issue:43867", "right": "issue:45375", "accept": false, - "reason": "State_dict ordering issue and missing `deepstack_visual_indexes` config field are different bugs." + "reason": "Load failure from sorted state_dict and a silently dropped config field are unrelated." } ] }, @@ -15664,8 +15575,8 @@ ], "soft_pairs": [ "issue:41669|issue:44246", - "issue:44938|issue:45071", "issue:36683|issue:43064", + "issue:44938|issue:45071", "issue:42890|issue:45561", "issue:43010|issue:45850", "issue:44625|issue:46365", @@ -15692,230 +15603,240 @@ "issue:45125|issue:45478", "issue:44355|issue:44749", "issue:30333|issue:33453", + "issue:45125|issue:45375", "issue:43010|issue:43232", - "issue:45125|issue:45375" + "issue:43502|issue:44222" ], - "prepared_review_unit_hash": "c02bf1e946eac8e1896227e8dae61b7a841a41067e6edafb5af497350b3c06a9", + "prepared_review_unit_hash": "dc5256ed377d879792e3997c42dece091188d2f610b88d6750e03b21b525b25f", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 31, - "serialized_chars": 36541, - "estimated_input_tokens": 9136, - "estimated_eval_tokens": 18528 + "soft_pair_count": 32, + "serialized_chars": 36840, + "estimated_input_tokens": 9210, + "estimated_eval_tokens": 18676 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a loose mix of unrelated Transformers v5 regression reports. The soft-edge pairs share broad themes like import/loading, quantization, tensor parallelism, or test failures, but none look like the same concrete bug or change.", - "confidence": 0.95, - "canonical_issue_reason": "issue:42548 is the strongest representative because it describes a broad, user-facing core import/export break in the package, which is more central than the narrower model- or test-specific reports here.", + "summary": "Mostly a heterogeneous mix of unrelated Transformers issues. The only plausible duplicate thread is import-time slowdown/import-surface regression (41669 and 44246); the rest do not look like the same underlying bug.", + "confidence": 0.68, + "canonical_issue_reason": "44246 is the best available anchor because it captures the broad user-facing symptom in the only plausible shared theme (slow imports), though the cluster is not truly coherent.", "canonical_pr_reason": null, - "best_issue_reason": "issue:42548 is the best single issue to anchor the cluster since it reflects a high-impact API regression and is easy to recognize as a canonical core-package failure mode.", + "best_issue_reason": "44246 is the most representative issue overall: it is broad, user-facing, and closest to the shared import-performance theme; the other issues are mostly unrelated point bugs.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:41669", "right": "issue:44246", - "accept": false, - "reason": "Both concern import-related performance, but one is about `import *` slowing import time and the other about intermittent slow `import transformers`; not the same bug." + "accept": true, + "reason": "Same import-performance regression theme: 41669 gives a likely root cause (`import *` in models), while 44246 reports the user-visible slowdown." }, { - "left": "issue:44938", - "right": "issue:45071", + "left": "issue:36683", + "right": "issue:43064", "accept": false, - "reason": "Python 3.14 load failure and `PretrainedConfig` type-checking break are different compatibility issues." + "reason": "Different failures: missing `vocab_size` in Gemma3 config vs distributed optimizer-state corruption in FSDP2/PEFT." }, { - "left": "issue:36683", - "right": "issue:43064", + "left": "issue:44938", + "right": "issue:45071", "accept": false, - "reason": "A missing `vocab_size` on `Gemma3Config` and a FSDP2/PEFT optimizer-state mismatch are unrelated problems." + "reason": "Both are version-regression reports, but one is Python 3.14 import failure and the other is `PretrainedConfig` type-checking breakage; different bugs." }, { "left": "issue:42890", "right": "issue:45561", "accept": false, - "reason": "Seed-dependent integration flakiness and a pytest-xdist race in testing utils are different test failures." + "reason": "Both are flaky test issues, but one is missing `set_seed()` in SamHQ tests and the other is an xdist race on `captured_info.txt`." }, { "left": "issue:43010", "right": "issue:45850", "accept": false, - "reason": "Decorating cache/layer `update()` with `torch.no_grad` is a specific autograd fix; \u201cbreaks too many models\u201d is far broader and not the same defect." + "reason": "`@torch.no_grad` on cache/layer updates is a specific API bug; 45850 is a broad complaint about many models breaking in v5." }, { "left": "issue:44625", "right": "issue:46365", "accept": false, - "reason": "Missing `num_labels` propagation in Qwen3.5 is a model config bug; inability to import `Trainer` until uninstalling `datasets` is an import dependency issue." + "reason": "Unrelated: Qwen3.5 `num_labels` propagation vs inability to import `Trainer` until uninstalling `datasets`." }, { "left": "issue:43992", "right": "issue:45478", "accept": false, - "reason": "`UMT5Encoder` missing `embed_tokens.weight` and Qwen3.5 MoE `from_pretrained` error are distinct model-loading failures." + "reason": "Both involve `from_pretrained`, but the concrete problems differ: missing `embed_tokens.weight` vs Qwen3.5 MoE load error." }, { "left": "issue:43867", "right": "issue:45357", "accept": false, - "reason": "A load error when state dict keys are sorted and incorrect visual encoder keys saved by Qwen3.5 are not the same bug." + "reason": "Different save/load paths and model families: sorted state_dict loading error vs incorrect visual-encoder keys on save." }, { "left": "issue:43010", "right": "issue:43976", "accept": false, - "reason": "`torch.no_grad` on cache updates and Python-version support in 5.1.0 are unrelated." + "reason": "No relation beyond being v5-related regressions; one is cache/layer update semantics, the other is a Python-version compatibility failure." }, { "left": "issue:36296", "right": "issue:43575", "accept": false, - "reason": "Tensor-parallel training bug and OOM when loading a specific Qwen2 model with TP are related at most broadly, but not the same concrete issue." + "reason": "Tensor-parallel training bug vs OOM while loading a specific Qwen2 model with TP; related area, but not the same concrete bug." }, { "left": "issue:43618", "right": "issue:44077", "accept": false, - "reason": "Missing `CLIPOutput.attentions` assignment and `patchtsmixer` post_init policy are separate model API changes." + "reason": "Unrelated model/API issues: CLIP output attentions assignment vs patchtsmixer `post_init` allowance." }, { "left": "issue:28282", "right": "issue:33453", "accept": false, - "reason": "PyTorch-missing import error for `AutoModel` and tokenizer-loading regression are different code paths." + "reason": "Different subsystems and symptoms: missing PyTorch import for AutoModel vs tokenizer-loading regression." }, { "left": "issue:43618", "right": "issue:46285", "accept": false, - "reason": "Both mention CLIP-related structures, but one is an output field regression and the other is inconsistent model class structure; not the same bug." + "reason": "Both touch CLIP, but one is missing attentions in output and the other is an inconsistent internal structure change; not the same bug." }, { "left": "issue:43618", "right": "issue:44857", "accept": false, - "reason": "CLIP attentions regression and `LwDetrImageLoss` AMP crash are unrelated model bugs." + "reason": "CLIP attentions regression and LwDetr float16 AMP crash are unrelated." }, { "left": "issue:44360", "right": "issue:45440", "accept": false, - "reason": "A discussion about a DSA indexer activation function and a DeepSeekV3 implementation divergence are different topics." + "reason": "No shared concrete code-path: DSA indexer activation issue vs DeepseekV3MoE divergence from remote implementation." }, { "left": "issue:43502", "right": "issue:44062", "accept": false, - "reason": "`local_files_only=True` still making network requests and `AddedToken()` duplicate `special` kwarg are unrelated tokenization/loading issues." + "reason": "Different bug classes: local_files_only still making network requests vs AddedToken constructor argument collision." }, { "left": "issue:42890", "right": "issue:44964", "accept": false, - "reason": "Missing `set_seed()` in tests and failure to load a specific multimodal model are unrelated." + "reason": "Test flakiness in SamHQ vs failure loading a Phi multimodal model; not the same underlying issue." }, { "left": "issue:42890", "right": "issue:44112", "accept": false, - "reason": "A flaky test due to missing seeds and a CI stale device-override failure are different test issues." + "reason": "Both are CI/test-related, but one is missing seeding and the other is a stale device override in GraniteSpeech." }, { "left": "issue:42890", "right": "issue:45600", "accept": false, - "reason": "Seed-related integration test instability and auto_mappings references to removed configs are unrelated." + "reason": "Unrelated test flake vs auto_mappings.py referencing removed Sam3LiteText configs." }, { "left": "issue:43723", "right": "issue:44556", "accept": false, - "reason": "Tokenizer loading regression and checkpoint reload failure after upgrade are different loading problems." + "reason": "Both are loading regressions, but one is tokenizer loading in v5 and the other is checkpoint reload incompatibility after upgrading from v4.57." }, { "left": "issue:44387", "right": "issue:44556", "accept": false, - "reason": "Int4 quantization memory growth/OOM and checkpoint reload failure are not the same bug." + "reason": "OOM under int4 quantization vs reload failure after version upgrade; different mechanisms and fixes." }, { "left": "issue:44222", "right": "issue:45125", "accept": false, - "reason": "FP8 `save_pretrained` behavior for MoE and missing `_tp_plan` for tensor parallelism are separate features/bugs." + "reason": "Different Qwen3.5 MoE issues: FP8 save_pretrained bug vs missing tensor-parallel plan." }, { "left": "issue:43010", "right": "issue:43065", "accept": false, - "reason": "`torch.no_grad` decoration for cache updates and a dummy `nn.Conv2d` in Sam3PixelDecoder are unrelated." + "reason": "No relation: cache/layer no_grad decoration vs dummy Conv2d in Sam3PixelDecoder." }, { "left": "issue:43576", "right": "issue:46132", "accept": false, - "reason": "A broken `transformers env` command and `AttentionInterface.register` behavior change are distinct issues." + "reason": "`transformers env` command breakage is unrelated to AttentionInterface.register behavior changes." }, { "left": "issue:43502", "right": "issue:44485", "accept": false, - "reason": "Unexpected network requests with `local_files_only=True` and GLM-5 RoPE implementation are unrelated." + "reason": "Local-files-only network bug vs GLM-5 RoPE implementation discussion; different problems." }, { "left": "issue:42548", "right": "issue:44351", "accept": false, - "reason": "Both are import failures, but one is `PreTrainedModel` export/import and the other is `HybridCache`; different symbols and likely different causes." + "reason": "Both are import-name errors from `transformers`, but they concern different missing symbols and do not look like the same concrete defect." }, { "left": "issue:45125", "right": "issue:45478", "accept": false, - "reason": "Missing `_tp_plan` and Qwen3.5 MoE `from_pretrained` failure are related only by model family, not the same concrete bug." + "reason": "Both concern Qwen3.5 MoE, but one is missing `_tp_plan` and the other is a `from_pretrained` error in newer transformers." }, { "left": "issue:44355", "right": "issue:44749", "accept": false, - "reason": "Compiled Python file errors and a slowdown when filtering data after upgrading are not the same issue." + "reason": "Compiled-Python-file errors vs performance slowdown after upgrading from 4.57.3 to 5.3.0 are unrelated." }, { "left": "issue:30333", "right": "issue:33453", "accept": false, - "reason": "MLflow reporting on failed training jobs and tokenizer loading regression are unrelated." + "reason": "MLFlow reporting on failed training jobs has nothing to do with tokenizer loading regression." + }, + { + "left": "issue:45125", + "right": "issue:45375", + "accept": false, + "reason": "Both are Qwen3.5-related, but one is tensor parallelism setup and the other is a missing vision-config field silently dropped by strict validation." }, { "left": "issue:43010", "right": "issue:43232", "accept": false, - "reason": "Cache/layer update autograd decoration and `_update_model_kwargs_for_generation` after `sync_gpus` are different generation-path bugs." + "reason": "No shared bug: cache/layer update decoration vs generation state sync after `sync_gpus`." }, { - "left": "issue:45125", - "right": "issue:45375", + "left": "issue:43502", + "right": "issue:44222", "accept": false, - "reason": "Missing tensor-parallel plan and a missing `deepstack_visual_indexes` field dropped by strict config handling are different config/model issues." + "reason": "Different issues: unwanted API requests during offline loading vs FP8 MoE save_pretrained behavior." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. It correctly treats 44246 as the best anchor for the only clearly related thread (import-time slowdown, with 41669), and the pair verdicts avoid overstating weaker similarities. No mergeability issue since there are no PR pairs." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -15983,11 +15904,10 @@ ], "soft_pairs": [ "issue:30990|issue:31795", - "issue:43502|issue:44222", "issue:43299|issue:44556", "issue:43502|issue:44961", - "issue:30333|issue:36296", "issue:42831|issue:43874", + "issue:30333|issue:36296", "issue:44355|issue:45071", "issue:43827|issue:44961", "issue:44355|issue:44655", @@ -16002,9 +15922,9 @@ "issue:44222|issue:44230", "issue:43576|issue:44938", "issue:41628|issue:42548", - "issue:46143|issue:46144", "issue:44908|issue:44961", "issue:44485|issue:44961", + "issue:46143|issue:46144", "issue:36296|issue:43901", "issue:44315|issue:46209", "issue:36296|issue:43526", @@ -16016,232 +15936,226 @@ "issue:44230|issue:46095", "issue:44393|issue:45071" ], - "prepared_review_unit_hash": "1153933e6143ed68598457792aa529f41851a6def46cf4d1a4cc52fdd092e015", + "prepared_review_unit_hash": "ebd052d6e9a63d460f3df0529544a4047502c3fcc16261c22f9d2c3e942e584e", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, - "soft_pair_count": 33, - "serialized_chars": 36712, - "estimated_input_tokens": 9178, - "estimated_eval_tokens": 18612 + "soft_pair_count": 32, + "serialized_chars": 36415, + "estimated_input_tokens": 9104, + "estimated_eval_tokens": 18464 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is mostly a grab bag of unrelated Transformers issues, with a few broad v5-regression reports mixed in. None of the soft pairs look like true duplicates under a strict same-bug/same-change standard.", - "confidence": 0.58, - "canonical_issue_reason": "issue:45850 is the broadest umbrella in the set: it explicitly summarizes the general v5 breakage theme and can absorb the various model-loading/regression complaints better than the narrower, model-specific reports.", + "summary": "The set is heterogeneous: a few pairs share broad themes like v5 regressions, loading failures, or import errors, but they describe different concrete bugs and should not be deduplicated. No pull requests are present.", + "confidence": 0.86, + "canonical_issue_reason": "No single issue is a strong canonical duplicate target here; the items span unrelated problems across loading, docs, FP8, import/export, CLI, and tensor-parallel behavior.", "canonical_pr_reason": null, - "best_issue_reason": "issue:45850 is the best global representative for the cluster because it is the only high-level issue that spans the repeated 'v5 breaks X' pattern rather than a single model or subsystem.", + "best_issue_reason": "issue:45850 is the broadest umbrella-style report, but it is too vague to be a reliable dedup target.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:30990", "right": "issue:31795", "accept": false, - "reason": "Sentence Transformers loading hangs and forward-docs confusion are unrelated problems." - }, - { - "left": "issue:43502", - "right": "issue:44222", - "accept": false, - "reason": "local_files_only network leakage and FP8 moe save_pretrained are different code paths." + "reason": "One is a Sentence Transformers load hang; the other is a docs wording complaint for model.forward." }, { "left": "issue:43299", "right": "issue:44556", "accept": false, - "reason": "Both are load regressions, but they affect different models and likely different failure points." + "reason": "Both involve version-related loading regressions, but for different models and likely different failure modes." }, { "left": "issue:43502", "right": "issue:44961", "accept": false, - "reason": "API requests despite local_files_only and a unrelated racoon-named issue are not the same bug." + "reason": "local_files_only API leakage is unrelated to the unrelated 'racoon' report." }, { - "left": "issue:30333", - "right": "issue:36296", + "left": "issue:42831", + "right": "issue:43874", "accept": false, - "reason": "MLflow reporting on failure and tensor-parallel training are unrelated." + "reason": "FP8 accuracy regression and a missing image-processor method are different bugs." }, { - "left": "issue:42831", - "right": "issue:43874", + "left": "issue:30333", + "right": "issue:36296", "accept": false, - "reason": "FineGrainedFP8 accuracy and a missing image-patch helper in GLM46V are distinct defects." + "reason": "MLFlow reporting on failed jobs is unrelated to tensor-parallel training." }, { "left": "issue:44355", "right": "issue:45071", "accept": false, - "reason": "Compiled Python file errors and PretrainedConfig type checking are different regressions." + "reason": "Compiled-file loading errors are not the same as PretrainedConfig type-checking breakage." }, { "left": "issue:43827", "right": "issue:44961", "accept": false, - "reason": "Pipeline docs cleanup and the racoon issue share no underlying change." + "reason": "Documentation references to removed pipeline() APIs are unrelated to 'racoon'." }, { "left": "issue:44355", "right": "issue:44655", "accept": false, - "reason": "Compiled-file handling and Pipeline.save_pretrained are separate failures." + "reason": "Compiled Python file errors and save_pretrained pipeline issues affect different code paths." }, { "left": "issue:44315", "right": "issue:44360", "accept": false, - "reason": "Liger Kernel application via model_init and a DSA ReLU discussion are unrelated." + "reason": "Liger Kernel not applied with model_init is unrelated to the DSA/ReLU discussion." }, { "left": "issue:33453", "right": "issue:34689", "accept": false, - "reason": "Tokenizer-loading regression and Llama 3.2 vision model loading are not the same concrete bug." + "reason": "Tokenizer loading regression and Llama 3.2 vision model loading failure are different concrete load bugs." }, { "left": "issue:27957", "right": "issue:33453", "accept": false, - "reason": "Flash Attention 2 support for XLMRoberta is unrelated to tokenizer loading regression." + "reason": "Flash Attention 2 support for XLM-Roberta is unrelated to tokenizer loading regression." }, { "left": "issue:42757", "right": "issue:43576", "accept": false, - "reason": "missing is_offline_mode import and a broken env command are different issues." + "reason": "Missing huggingface_hub import and the v5 env CLI breakage are not the same issue." }, { "left": "issue:43576", "right": "issue:45071", "accept": false, - "reason": "The env CLI breakage and PretrainedConfig type checking have no common code-path." + "reason": "Broken env command and PretrainedConfig type checking are separate regressions." }, { "left": "issue:43576", "right": "issue:43824", "accept": false, - "reason": "Broken env command and missing Qwen2.5-VL export are unrelated." + "reason": "CLI env failure is unrelated to the missing Qwen2_5_VL export." }, { "left": "issue:31795", "right": "issue:38617", "accept": false, - "reason": "Documentation ambiguity and a missing symbol import are not duplicates." + "reason": "A docs confusion issue does not match an ImportError for layer_type_validation." }, { "left": "issue:44961", "right": "issue:45468", "accept": false, - "reason": "Racoon and Gemma-4 audio positional encoding are unrelated." + "reason": "The placeholder 'racoon' report does not match the Gemma-4 audio positional encoding bug." }, { "left": "issue:44222", "right": "issue:44230", "accept": false, - "reason": "FP8 save_pretrained for moe and FP8 inference support are different features/failures." + "reason": "FP8 save_pretrained and FP8 inference support are different failures, not one code-path bug." }, { "left": "issue:43576", "right": "issue:44938", "accept": false, - "reason": "env CLI breakage and Python 3.14 load failure are distinct regressions." + "reason": "Broken transformers env CLI and Python 3.14 import/load failure are distinct problems." }, { "left": "issue:41628", "right": "issue:42548", "accept": false, - "reason": "Missing AutoImageProcessor and missing PreTrainedModel exports are similar in shape but not clearly the same underlying bug." - }, - { - "left": "issue:46143", - "right": "issue:46144", - "accept": false, - "reason": "One is kwargs not being forwarded, the other is the RoFormer attention implementation not using the attention interface." + "reason": "Different missing exports: AutoImageProcessor versus PreTrainedModel." }, { "left": "issue:44908", "right": "issue:44961", "accept": false, - "reason": "inverse_sqrt scheduler kwargs handling and racoon are unrelated." + "reason": "Scheduler kwargs handling is unrelated to the 'racoon' issue." }, { "left": "issue:44485", "right": "issue:44961", "accept": false, - "reason": "GLM-5 RoPE implementation and racoon have no shared failure mode." + "reason": "GLM-5 RoPE implementation discussion is unrelated to the 'racoon' report." + }, + { + "left": "issue:46143", + "right": "issue:46144", + "accept": false, + "reason": "Both concern RoFormer, but one is about kwargs propagation and the other about attention-interface usage; not the same concrete bug." }, { "left": "issue:36296", "right": "issue:43901", "accept": false, - "reason": "tensor-parallel training and TextClassificationPipeline docs mismatch are unrelated." + "reason": "Tensor-parallel training and pipeline docs behavior are unrelated." }, { "left": "issue:44315", "right": "issue:46209", "accept": false, - "reason": "Liger Kernel not applied with model_init and FP8 inference on B200 are different bugs." + "reason": "Liger Kernel application in model_init is unrelated to FP8 inference on B200." }, { "left": "issue:36296", "right": "issue:43526", "accept": false, - "reason": "tensor-parallel training and BeitImageProcessorFast reduce_labels are unrelated." + "reason": "Tensor-parallel training bug and BeitImageProcessorFast label reduction are different code paths." }, { "left": "issue:43673", "right": "issue:44492", "accept": false, - "reason": "chunked_prefill cache handling and a cache-strategies typo are not the same issue." + "reason": "Generation cache missing during chunked_prefill is not a typo in cache strategy docs." }, { "left": "issue:45362", "right": "issue:45850", "accept": false, - "reason": "A single Qwen3.5-35B chat crash is narrower than the broad v5 breakage umbrella." + "reason": "A specific Qwen3.5-35B chat crash is not the same bug as a broad 'many models break in v5' complaint." }, { "left": "issue:44623", "right": "issue:45230", "accept": false, - "reason": "processor.save_pretrained missing files and a generic bug report title are not enough to match as duplicates." + "reason": "Missing files in processor.save_pretrained is unrelated to the placeholder bug report." }, { "left": "issue:44273", "right": "issue:45542", "accept": false, - "reason": "lazy loading and a TensorFlow backend installation error are different problems." + "reason": "Lazy loading failures and the TensorFlow/tensorboard backend error are different issues." }, { "left": "issue:44038", "right": "issue:45125", "accept": false, - "reason": "generic Qwen3-VL-Moe breakage and missing _tp_plan for tensor parallelism are not the same concrete defect." + "reason": "Qwen3-VL MoE breakage and missing _tp_plan for tensor parallelism are separate model-support gaps." }, { "left": "issue:44230", "right": "issue:46095", "accept": false, - "reason": "FP8 inference support and a DeepSeek weight-conversion question are unrelated." + "reason": "FP8 inference support and a weight-conversion-script question are not the same underlying bug." }, { "left": "issue:44393", "right": "issue:45071", "accept": false, - "reason": "Qwen3-VL bounding-box output issues and PretrainedConfig type checking do not share a code path." + "reason": "Qwen3-VL 2D bounding-box output problems are unrelated to PretrainedConfig type-checking regressions." } ] }, @@ -16358,20 +16272,20 @@ "issue:41669|issue:43704", "issue:31795|issue:34689" ], - "prepared_review_unit_hash": "8b6312b4a650ed82ca56e8b52b95e28da0dc0d58394fca4cd0ad22e9ef59e618", + "prepared_review_unit_hash": "42adaef6177f8e02448c17a48b8f60049262999e74cb4d1b61b20971c3fa16ca", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 48, "item_count": 48, "soft_pair_count": 42, - "serialized_chars": 39546, + "serialized_chars": 39548, "estimated_input_tokens": 9887, "estimated_eval_tokens": 20030 }, @@ -16380,234 +16294,234 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a mixed set of Transformers v5-era regressions and adjacent docs/perf/model-specific bugs. The items are thematically related by version upgrades, loading/import behavior, and model-specific breakages, but there are no strong duplicate pairs with the evidence provided.", - "confidence": 0.71, - "canonical_issue_reason": "Issue 45278 is the best representative of the cluster because it states a concrete, broad v5 upgrade regression with many import failures, which matches the cluster\u2019s dominant theme better than the narrower or more vague reports.", + "summary": "The soft pairs are mostly false positives: they share broad themes like v5 regressions, import/loading problems, or model-family-specific bugs, but not the same concrete underlying defect. The closest umbrella issue is the general v5 import/loading breakage report, but there is no strong duplicate cluster here.", + "confidence": 0.89, + "canonical_issue_reason": "issue:45278 is the broadest concrete regression report in the set (many import errors after upgrading 4.57.0 -> 5.5.0), so it is the closest umbrella candidate for the import/load-related subset, though it is not a true duplicate of most others.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45278 is the most suitable global representative: it is concrete, user-facing, and broadly captures the upgrade breakage pattern without being as vague as the more generic complaints.", + "best_issue_reason": "issue:45278 is the most representative and actionable issue for this cluster because it clearly describes a widespread upgrade regression affecting imports/loading, while the other issues are mostly narrower model-specific bugs or unrelated regressions.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44273", "right": "issue:45003", "accept": false, - "reason": "Both touch lazy-loading/import behavior, but one is about lazy loading not working and the other about unsafe sys.modules access/performance; not the same bug." + "reason": "Both involve import/lazy-loading behavior, but one is a generic lazy-loading complaint and the other is a specific sys.modules access concern; not clearly the same bug." }, { "left": "issue:43576", "right": "issue:44556", "accept": false, - "reason": "One is a broken `transformers env` command, the other is checkpoint reload failure after upgrading; different surfaces and code paths." + "reason": "Both are v5 breakage reports, but one is about the `transformers env` command and the other about checkpoint reload compatibility; different code paths." }, { "left": "issue:44393", "right": "issue:45753", "accept": false, - "reason": "Both are Qwen3-VL problems, but one is 2D box hallucination/error and the other is an infinite loop on a specific image; too different to merge." + "reason": "Same model family, but one is a 2D bounding-box output issue and the other is an infinite-loop on a specific image; different failures." }, { "left": "issue:43525", "right": "issue:45850", "accept": false, - "reason": "A missing `pad_token_id` in `Llama4Config` is a specific config bug; the other is a broad complaint that v5 breaks too many models." + "reason": "One is a specific `Llama4Config` attribute error; the other is a vague general complaint about many models breaking in 5.0." }, { "left": "issue:41669", "right": "issue:43673", "accept": false, - "reason": "Removing `import *` for import-time cost is unrelated to missing `GenerationMixin` cache during chunked prefill." + "reason": "One concerns expensive `import *` usage and import-time overhead; the other is a GenerationMixin cache problem during chunked prefill." }, { "left": "issue:44556", "right": "issue:45092", "accept": false, - "reason": "Both are upgrade incompatibilities, but they affect different models and failure modes; not the same concrete loading bug." + "reason": "Both mention v5 reload/incompatibility after upgrades, but they concern different model families and likely different underlying breakages." }, { "left": "issue:43572", "right": "issue:45850", "accept": false, - "reason": "A missing `pad_token_idx` in one config is not the same as a broad statement that many models break in v5." + "reason": "Specific missing `pad_token_idx` in StableLmConfig is not the same as a broad claim that 5.0 breaks many models." }, { "left": "issue:44492", "right": "issue:45850", "accept": false, - "reason": "A typo in cache strategies is a narrow docs/code issue, not the same as a general v5 breakage report." + "reason": "A cache-strategy typo is unrelated to the generic v5 model-breakage complaint." }, { "left": "issue:45278", "right": "issue:45850", "accept": false, - "reason": "Both are broad regression reports, but they are not the same underlying failure; one is import errors on upgrade, the other is a generic complaint about many models." + "reason": "Both are broad v5 regression reports, but one is specifically about import errors after upgrade and the other is an unspecific blanket complaint; not the same defect." }, { "left": "issue:42371", "right": "issue:44492", "accept": false, - "reason": "TF32 API settings and a cache-strategy typo are unrelated changes." + "reason": "TF32 API guidance and a cache-strategy typo are unrelated." }, { "left": "issue:41669", "right": "issue:46218", "accept": false, - "reason": "One is about import-star cleanup and import-time cost; the other is a missing `Qwen3_5ForCausalLM` import error." + "reason": "One is about import-time cost from `import *`; the other is a missing `Qwen3_5ForCausalLM` import, which is a separate symbol/export issue." }, { "left": "issue:42890", "right": "issue:44908", "accept": false, - "reason": "Missing `set_seed()` in tests is unrelated to scheduler kwargs being ignored." + "reason": "A flaky test due to missing `set_seed()` is unrelated to scheduler kwargs handling." }, { "left": "issue:43408", "right": "issue:44315", "accept": false, - "reason": "A model-type warning for SAM3 is unrelated to Liger Kernel not being applied during `model_init`." + "reason": "Model-type warning for sam3_tracker/sam3_video and Liger Kernel not applied with `model_init` are unrelated." }, { "left": "issue:44297", "right": "issue:45230", "accept": false, - "reason": "Tokenizer class mismatch on save is a specific serialization bug; the other is an unspecified generic bug report." + "reason": "Tokenizer save metadata mismatch and a generic bug report do not describe the same bug." }, { "left": "issue:44246", "right": "issue:44273", "accept": false, - "reason": "Import slowness and lazy-loading failure are related at a high level, but not clearly the same defect." + "reason": "Slow imports and broken lazy loading may be related at a high level, but the reports do not identify the same concrete defect." }, { "left": "issue:43352", "right": "issue:44393", "accept": false, - "reason": "Flash Attention 2 support gating for Nemotron is unrelated to Qwen3-VL bounding-box errors." + "reason": "Flash Attention 2 support for Nemotron and Qwen3-VL bounding-box errors are unrelated." }, { "left": "issue:44295", "right": "issue:45230", "accept": false, - "reason": "Reading a buffer-registered `position_ids` and a generic bug report do not describe the same issue." + "reason": "A buffer/`position_ids` access error is not the same as a generic bug report." }, { "left": "issue:43576", "right": "issue:44908", "accept": false, - "reason": "A broken env command and an inverse_sqrt scheduler kwargs bug are unrelated." + "reason": "The `transformers env` command and inverse_sqrt scheduler kwargs are different areas of the codebase." }, { "left": "issue:44704", "right": "issue:44908", "accept": false, - "reason": "`AutoProcessor.from_pretrained` kwargs propagation and scheduler kwargs handling are different code paths." + "reason": "AutoProcessor kwarg forwarding and scheduler kwargs are unrelated parameter-passing bugs." }, { "left": "issue:43976", "right": "issue:44393", "accept": false, - "reason": "Python version support breakage is unrelated to Qwen3-VL 2D box output errors." + "reason": "Python-version compatibility and Qwen3-VL bounding-box output are unrelated issues." }, { "left": "issue:44908", "right": "issue:45071", "accept": false, - "reason": "Ignored scheduler kwargs and PretrainedConfig type checking are unrelated regressions." + "reason": "Scheduler kwargs handling and `PretrainedConfig` type-checking are different bugs." }, { "left": "issue:41669", "right": "issue:44351", "accept": false, - "reason": "Import-star cleanup/import cost is not the same as failing to import `HybridCache`." + "reason": "Import performance from `import *` is unrelated to failing to import `HybridCache`." }, { "left": "issue:44492", "right": "issue:46328", "accept": false, - "reason": "A cache-strategy typo and a SequenceFeatureExtractor.pad performance problem are unrelated." + "reason": "A typo in cache strategies is unrelated to `SequenceFeatureExtractor.pad` converting arrays inefficiently." }, { "left": "issue:41669", "right": "issue:45341", "accept": false, - "reason": "Model import cleanup/perf is unrelated to a bug in `testing_utils.py`." + "reason": "Import-star cleanup/performance and a small testing_utils bug are unrelated." }, { "left": "issue:43976", "right": "issue:45850", "accept": false, - "reason": "Python compatibility breakage is not the same as the vague claim that too many models break." + "reason": "Python version support and generic v5 model breakage are not the same underlying bug." }, { "left": "issue:44393", "right": "issue:45478", "accept": false, - "reason": "Different Qwen model families and different symptoms; not the same bug." + "reason": "Both concern Qwen-related models, but one is a VL bounding-box issue and the other is a Qwen3.5 MoE `from_pretrained` failure." }, { "left": "issue:44077", "right": "issue:44861", "accept": false, - "reason": "A `post_init` policy issue in patchtsmixer is unrelated to `_get_tied_weight_keys` crashing on a list." + "reason": "A `patchtsmixer` initialization policy issue is unrelated to `_get_tied_weight_keys` crashing on a list." }, { "left": "issue:43976", "right": "issue:46218", "accept": false, - "reason": "Python version incompatibility and a missing Qwen3_5 import are different problems." + "reason": "Python-version compatibility and a missing `Qwen3_5ForCausalLM` import are unrelated." }, { "left": "issue:43827", "right": "issue:44908", "accept": false, - "reason": "Docs still referencing `pipeline()` after removals is unrelated to scheduler kwargs." + "reason": "Docs referencing removed pipelines and scheduler kwargs are unrelated." }, { "left": "issue:31795", "right": "issue:33453", "accept": false, - "reason": "Documentation confusion about forward args is unrelated to a tokenizer loading regression." + "reason": "Documentation confusion in `model.forward` is unrelated to tokenizer loading regression." }, { "left": "issue:44315", "right": "issue:44829", "accept": false, - "reason": "Liger Kernel not being applied and flash_attention_3 causing degenerate training are different implementation issues." + "reason": "Liger Kernel not applied with `model_init` and degenerate training under flash_attention_3 are different training/model-init problems." }, { "left": "issue:43576", "right": "issue:44273", "accept": false, - "reason": "A broken env command and lazy loading not working are both v5-adjacent, but not the same specific bug." + "reason": "`transformers env` being broken and lazy loading not working are both broad runtime complaints, but not the same specific bug." }, { "left": "issue:44655", "right": "issue:44861", "accept": false, - "reason": "Saving Pipeline objects and tied-weight key extraction are unrelated failures." + "reason": "Saving Pipeline objects and `_get_tied_weight_keys` crashing are unrelated failures." }, { "left": "issue:45230", "right": "issue:45310", "accept": false, - "reason": "A generic bug report and a specific Qwen3.5 MoE from_pretrained failure are not the same issue." + "reason": "A generic bug report and a specific Qwen3.5 MoE `from_pretrained` error are not duplicates." }, { "left": "issue:44360", "right": "issue:45468", "accept": false, - "reason": "DSA indexer ReLU discussion is unrelated to Gemma-4 audio positional encoding." + "reason": "A DSA indexer activation issue is unrelated to Gemma4AudioRelPositionalEncoding." }, { "left": "issue:44273", "right": "issue:44908", "accept": false, - "reason": "Lazy-loading issues and scheduler kwargs handling are unrelated." + "reason": "Lazy loading problems and inverse_sqrt scheduler kwargs are unrelated." }, { "left": "issue:43576", "right": "issue:46218", "accept": false, - "reason": "A broken env command is unrelated to a Qwen3_5 import failure." + "reason": "Broken env command and missing Qwen3_5 import are different issues." }, { "left": "issue:43519", @@ -16619,25 +16533,25 @@ "left": "issue:46018", "right": "issue:46172", "accept": false, - "reason": "Both are DeepSeek-V4 related, but one reports gating semantics and the other a tensor transpose/casting issue; not clearly the same bug." + "reason": "DeepSeek-V4 shared expert gating and a DeepSeek-V4 transpose/dtype bug are different concrete defects." }, { "left": "issue:43519", "right": "issue:44908", "accept": false, - "reason": "A multimodal timestamp bug and a scheduler kwargs bug are unrelated." + "reason": "Timestamp calculation in a processor and scheduler kwargs handling are unrelated." }, { "left": "issue:41669", "right": "issue:43704", "accept": false, - "reason": "Import-star cleanup/import time and a VRAM leak in dataloader threads are different problems." + "reason": "Import-star cleanup/performance and a VRAM leak in multi-threaded dataloaders are unrelated." }, { "left": "issue:31795", "right": "issue:34689", "accept": false, - "reason": "Docs wording around `input_ids`/`past_key_values` is unrelated to a model-loading regression for Llama 3.2 Vision." + "reason": "A documentation issue and a model-loading regression for Llama 3.2 Vision are not the same bug." } ] }, @@ -16750,20 +16664,20 @@ "issue:37934|issue:43668", "issue:42936|issue:43668" ], - "prepared_review_unit_hash": "62348282138d5ec4e8049bcb03c84c756f1e0ebd35842297ba63c46edc6c2764", + "prepared_review_unit_hash": "41043ad1267420e234cb6cfd9c1b4bf3355c2f12583e8f29ffad08d8e038b73b", "component_budget": { "node_count": 450, "item_count": 450, "soft_pair_count": 1664, - "serialized_chars": 861347, - "estimated_input_tokens": 215337, - "estimated_eval_tokens": 430930 + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 }, "budget": { "node_count": 44, "item_count": 44, "soft_pair_count": 42, - "serialized_chars": 37123, + "serialized_chars": 37124, "estimated_input_tokens": 9281, "estimated_eval_tokens": 18818 }, @@ -16772,264 +16686,264 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Loose cluster of mostly unrelated Transformers bug reports, with a weak theme around model/tokenizer loading regressions. None of the soft pairs are true duplicates under a strict triage standard.", - "confidence": 0.41, - "canonical_issue_reason": "issue:33453 is the clearest concrete regression report in the tokenizer-loading family and is more representative than the broader or more vague items.", + "summary": "The cluster is highly heterogeneous: most items are model-specific regressions, tokenizer/config issues, or unrelated API/docs bugs. The soft-similarity pairs mostly share only a broad subsystem, not the same concrete defect, so I reject all soft edges.", + "confidence": 0.69, + "canonical_issue_reason": "Issue 45850 is the broadest umbrella in the set and best matches the apparent recurring theme of Transformers v5 regressions affecting multiple models, but the cluster is not truly duplicate-heavy.", "canonical_pr_reason": null, - "best_issue_reason": "issue:33453 is the strongest standalone issue to anchor this cluster: it is specific, user-facing, and closest to the small tokenizer-regression subtheme here.", + "best_issue_reason": "Issue 45850 is the best overall representative because it captures the widest common theme across the set (version-5 breakages), even though it is too broad to be a precise duplicate target.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:31515", "right": "issue:31795", "accept": false, - "reason": "Different bugs: slow checkpoint loading vs documentation confusion about input_ids/past_key_values." + "reason": "Different problems: checkpoint loading speed versus documentation wording for forward inputs." }, { "left": "issue:43704", "right": "issue:44485", "accept": false, - "reason": "Different code paths: VRAM leak in dataloader-thread usage vs GLM-5 RoPE implementation question." + "reason": "Different model/code paths: VRAM leak in dataloader threads versus GLM-5 RoPE implementation." }, { "left": "issue:44485", "right": "issue:45468", "accept": false, - "reason": "Both concern model internals, but the specific bugs are unrelated: RoPE math vs Gemma-4 audio positional encoding." + "reason": "Both are model-specific bugs, but they affect different models and distinct logic." }, { "left": "issue:43298", "right": "issue:46095", "accept": false, - "reason": "Feature/support request mismatch: continuous batching for audio models vs DeepSeek-V4 weight conversion tooling." + "reason": "Continuous batching audio support and DeepSeek-V4 weight conversion are unrelated." }, { "left": "issue:45850", "right": "issue:46218", "accept": false, - "reason": "Too broad vs specific import failure; not the same underlying bug." + "reason": "One is a broad version-5 complaint; the other is an import error for a specific class." }, { "left": "issue:43502", "right": "issue:43519", "accept": false, - "reason": "Local-files-only network leakage vs incorrect timestamp calculation in a processor." + "reason": "Local-files-only network requests and Qwen3VL timestamp math are different bugs." }, { "left": "issue:42371", "right": "issue:43704", "accept": false, - "reason": "Unrelated areas: TF32 API behavior vs VRAM leak in dataloader-threaded inference." + "reason": "TF32 settings guidance and a VRAM leak are unrelated." }, { "left": "issue:43519", "right": "issue:44623", "accept": false, - "reason": "Different processor bugs: timestamp calculation vs missing files from save_pretrained." + "reason": "Timestamp calculation and missing save_pretrained files are different code paths." }, { "left": "issue:43519", "right": "issue:43704", "accept": false, - "reason": "Different failures in different code paths: processor timestamp math vs multi-thread VRAM leak." + "reason": "No concrete overlap beyond both being runtime bugs." }, { "left": "issue:43606", "right": "issue:45468", "accept": false, - "reason": "CPU-offload device mismatch is not the same bug as Gemma-4 audio positional encoding." + "reason": "CPU offload device mismatch and Gemma-4 positional encoding are unrelated." }, { "left": "issue:44263", "right": "issue:45468", "accept": false, - "reason": "Different model logic: torch.split return handling vs Gemma-4 audio positional encoding." + "reason": "torch.split return handling in GLM-MoE is not the same as Gemma-4 audio positional encoding." }, { "left": "issue:44261", "right": "issue:45468", "accept": false, - "reason": "RMS norm epsilon precision issue is unrelated to Gemma-4 audio positional encoding." + "reason": "Different model internals and different failure modes." }, { "left": "issue:44351", "right": "issue:46218", "accept": false, - "reason": "Importing HybridCache and importing Qwen3_5ForCausalLM are separate import-path failures." + "reason": "Both are import errors, but for different symbols and likely different fixes." }, { "left": "issue:43352", "right": "issue:46218", "accept": false, - "reason": "Flash-Attention support gating for Nemotron is unrelated to a Qwen3_5 module import error." + "reason": "Flash Attention support gating and a module import failure are unrelated." }, { "left": "issue:42913", "right": "issue:43066", "accept": false, - "reason": "Related tokenizer-v5 theme, but one is a behavior regression and the other is a wrong decoder type; not clearly the same bug." + "reason": "Both concern tokenizer v5 behavior, but one is a broad behavior change and the other is a decoder-type mapping bug." }, { "left": "issue:44908", "right": "issue:44998", "accept": false, - "reason": "Inverse-sqrt scheduler kwargs bug has nothing to do with the unrelated unemployment issue." + "reason": "Scheduler kwargs bug versus an unrelated placeholder issue." }, { "left": "issue:44492", "right": "issue:44998", "accept": false, - "reason": "Typo in cache strategies is unrelated to the unemployment placeholder issue." + "reason": "A typo in cache strategies is unrelated to the other issue." }, { "left": "issue:44998", "right": "issue:45850", "accept": false, - "reason": "Unemployment placeholder vs broad model-breakage complaint; no shared code path." + "reason": "One is not a technical bug; the other is a broad Transformers regression report." }, { "left": "issue:43618", "right": "issue:44998", "accept": false, - "reason": "CLIPOutput attentions regression is unrelated to the unemployment placeholder issue." + "reason": "CLIPOutput attentions regression and an unrelated placeholder issue." }, { "left": "issue:44861", "right": "issue:45290", "accept": false, - "reason": "Tied-weights utility crash vs chat-template crash with tool calls; different functions and symptoms." + "reason": "Tied-weights crash and chat-template tool-call crash are different code paths." }, { "left": "issue:44998", "right": "issue:45341", "accept": false, - "reason": "Unemployment placeholder vs a testing_utils bug; unrelated." + "reason": "Unrelated issues; no shared defect." }, { "left": "issue:43525", "right": "issue:44861", "accept": false, - "reason": "Missing pad_token_id in Llama4Config is unrelated to _get_tied_weight_keys crashing on list input." + "reason": "Missing config attribute in Llama4 versus tied-weights list handling are unrelated." }, { "left": "issue:43572", "right": "issue:44861", "accept": false, - "reason": "StableLmConfig pad_token_idx regression is unrelated to tied-weight key handling." + "reason": "StableLm config padding field regression is not the same bug as tied-weights handling." }, { "left": "issue:31515", "right": "issue:38617", "accept": false, - "reason": "Checkpoint loading slowdown and missing layer_type_validation import are unrelated issues." + "reason": "Slow checkpoint loading and an ImportError for layer_type_validation are unrelated." }, { "left": "issue:44230", "right": "issue:44623", "accept": false, - "reason": "FP8 inference support request is unrelated to processor.save_pretrained missing files." + "reason": "FP8 support request and processor.save_pretrained missing files are different issues." }, { "left": "issue:30333", "right": "issue:31515", "accept": false, - "reason": "MLFlow status reporting on failed training jobs is unrelated to checkpoint loading speed." + "reason": "MLFlow reporting on failed training jobs is unrelated to checkpoint loading speed." }, { "left": "issue:31515", "right": "issue:36296", "accept": false, - "reason": "Checkpoint loading slowdown vs tensor-parallel training bug; different failure modes." + "reason": "Different symptoms and different subsystems." }, { "left": "issue:44230", "right": "issue:45310", "accept": false, - "reason": "FP8 support request is not the same as the Qwen3.5 MoE from_pretrained loading error." + "reason": "FP8 feature support request versus from_pretrained error in Qwen3.5 MoE are not the same defect." }, { "left": "issue:31515", "right": "issue:33453", "accept": false, - "reason": "Both mention loading, but one is a performance complaint about checkpoints and the other is a tokenizer-loading regression." + "reason": "Checkpoint loading performance and tokenizer loading regression are different problems." }, { "left": "issue:43298", "right": "issue:44230", "accept": false, - "reason": "Continuous batching for audio models and FP8 inference support are distinct feature requests." + "reason": "Audio continuous batching support and FP8 inference support are unrelated." }, { "left": "issue:43717", "right": "issue:46144", "accept": false, - "reason": "init_weights behavior in Mamba models is unrelated to RoFormer attention-interface usage." + "reason": "Mamba init_weights and RoFormer attention interface are separate model implementations." }, { "left": "issue:42936", "right": "issue:46018", "accept": false, - "reason": "Mask2former ignore_value handling and DeepSeek-V4 shared-expert gating are different model-specific bugs." + "reason": "Different model-specific bugs with no shared concrete code path." }, { "left": "issue:43668", "right": "issue:44031", "accept": false, - "reason": "Incorrect type hint for norm_eps is unrelated to tokenizer regex warning noise." + "reason": "Incorrect type hint and tokenizer regex warning are unrelated." }, { "left": "issue:42936", "right": "issue:43240", "accept": false, - "reason": "Mask2former ignore_value usage and kwargs not passed to loss calculation are separate issues." + "reason": "One is a missing use of ignore_value; the other is kwargs not passed into loss calculation." }, { "left": "issue:43240", "right": "issue:46144", "accept": false, - "reason": "Loss-function kwargs plumbing is unrelated to RoFormer attention-interface implementation." + "reason": "Loss kwargs handling and RoFormer attention interface are unrelated." }, { "left": "issue:42936", "right": "issue:44031", "accept": false, - "reason": "Different problem areas: segmentation ignore_value vs tokenizer regex warnings." + "reason": "Different bugs: ignore_value usage versus regex warning." }, { "left": "issue:42936", "right": "issue:46144", "accept": false, - "reason": "Different model bugs with no shared concrete code path." + "reason": "Different model internals and failure modes." }, { "left": "issue:42936", "right": "issue:43813", "accept": false, - "reason": "Mask2former ignore_value bug is unrelated to a quantization-operation typo." + "reason": "A model-loss bug and a typo in quantization operation naming are unrelated." }, { "left": "issue:43240", "right": "issue:43668", "accept": false, - "reason": "Loss kwargs propagation and a config type-hint typo are unrelated." + "reason": "Loss kwargs propagation and a config type-hint issue are unrelated." }, { "left": "issue:43668", "right": "issue:43813", "accept": false, - "reason": "Norm_eps type hint and quantization-operation typo are different, isolated fixes." + "reason": "Type-hint correction and typo in cache conversion naming are not the same bug." }, { "left": "issue:37934", "right": "issue:43668", "accept": false, - "reason": "Llama4TextL2Norm/RMS-norm question is unrelated to ModernBERTConfig type-hint correctness." + "reason": "RMSNorm question for Llama4 and a ModernBERT type-hint bug are unrelated." }, { "left": "issue:42936", "right": "issue:43668", "accept": false, - "reason": "Ignore_value handling and norm_eps typing are unrelated model/config bugs." + "reason": "Different model-specific issues with no concrete overlap." } ] }, @@ -17062,12 +16976,12 @@ "issue:10519|issue:11005", "issue:11005|issue:11307" ], - "prepared_review_unit_hash": "ff007f9e3a3f74e44d0abc898d94eb227657988274d7e73c30b9c000c749dd0e", + "prepared_review_unit_hash": "e3740ca505d98751732cb5dfb22c3cef79ead20a31a24fb521d5b38b2628895a", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 2877, + "serialized_chars": 2878, "estimated_input_tokens": 720, "estimated_eval_tokens": 1696 }, @@ -17075,7 +16989,7 @@ "node_count": 3, "item_count": 3, "soft_pair_count": 3, - "serialized_chars": 2877, + "serialized_chars": 2878, "estimated_input_tokens": 720, "estimated_eval_tokens": 1696 }, @@ -17084,30 +16998,30 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster does not appear to contain duplicates: the three issues are about unrelated topics (truncation behavior, ReduceLROnPlateau-like scheduling, and Wav2Vec2 word time offsets).", + "summary": "These three issues are not duplicates of one another; they request unrelated features in different parts of the library.", "confidence": 0.98, - "canonical_issue_reason": "No clear canonical issue: the items describe different feature requests in different parts of the library, so none is a true representative of the others.", + "canonical_issue_reason": "No single canonical issue: the items cover unrelated requests (token truncation behavior, ReduceLROnPlateau-like scheduling, and Wav2Vec2 word offsets).", "canonical_pr_reason": null, - "best_issue_reason": "If one must be selected, issue 11307 is the most discussion-heavy, but it is still not a good cluster representative because it is unrelated to the other two.", + "best_issue_reason": "issue:10519 is the most concrete and narrowly scoped request, but it should not be treated as a canonical duplicate for the others.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:10519", "right": "issue:11307", "accept": false, - "reason": "Different requests: token truncation direction vs Wav2Vec2 word boundary offsets; no same underlying bug/change." + "reason": "Different topics: truncation from the beginning vs. Wav2Vec2 word start/end offsets." }, { "left": "issue:10519", "right": "issue:11005", "accept": false, - "reason": "Completely different features: truncation option vs ReduceLROnPlateau-like functionality." + "reason": "Different topics: truncation behavior vs. ReduceLROnPlateau-like learning-rate scheduling." }, { "left": "issue:11005", "right": "issue:11307", "accept": false, - "reason": "Unrelated topics: learning-rate scheduling behavior vs audio token/word offset extraction." + "reason": "Different topics: scheduler functionality vs. extracting Wav2Vec2 word timing offsets." } ] }, @@ -17139,7 +17053,7 @@ "issue:13244|issue:16618", "issue:15980|issue:16618" ], - "prepared_review_unit_hash": "a7f570ce78e4ef1966634cd978ab3fcd91a368937819f00523bd91b16cc1eec0", + "prepared_review_unit_hash": "ff92f07787566a7c34c16ba09c17d7b146ba179e536a63c41cc087e29de0c599", "component_budget": { "node_count": 3, "item_count": 3, @@ -17161,24 +17075,24 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The cluster is not a true duplicate set: one issue is about tokenization behavior, one about private-model auth/download errors, and one about a general model-loading failure. The closest canonical issue is the model-loading report, but the others are distinct bugs.", - "confidence": 0.91, - "canonical_issue_reason": "Issue 16618 ('Can't load the model for 'bert-base-uncased'.') is the broadest and most central model-loading failure in the set, so it is the best anchor for the cluster even though the other issues are not true duplicates of it.", + "summary": "These issues do not look like true duplicates: one is about Tapas tokenization mismatch, one about an incorrect auth-related error when downloading private models, and one about a generic failure to load bert-base-uncased. The only mild overlap is around model-loading UX, not the same underlying bug.", + "confidence": 0.31, + "canonical_issue_reason": "No strong canonical issue exists because the topics differ. If one must be chosen, #16618 is the broadest model-loading report and the closest to the other two by title, but it is still not the same bug as the auth error or Tapas tokenization issue.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 16618 is the best representative because it concerns a concrete loading failure and is the most likely duplicate target among these reports; the others are narrower and clearly about different code paths.", + "best_issue_reason": "#16618 is the best representative only in the sense that it is the most general and closest to the shared 'model loading' wording; however, it should not be treated as a true duplicate target for the other issues.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:13244", "right": "issue:16618", "accept": false, - "reason": "Tokenization mismatch vs. model loading failure are different underlying bugs and different code paths." + "reason": "Tapas tokenization mismatch and bert-base-uncased loading failure are unrelated code paths; one concerns tokenizer parity, the other a model load error." }, { "left": "issue:15980", "right": "issue:16618", "accept": false, - "reason": "Private-model download/auth error is not the same as failing to load bert-base-uncased; they may both involve loading, but the failure mode and fix are different." + "reason": "#15980 is specifically about the wrong error when a private model download is attempted without login, while #16618 is a generic load failure for a public model; same broad area, different underlying bug." } ] }, @@ -17208,12 +17122,12 @@ "soft_pairs": [ "issue:26350|issue:33260" ], - "prepared_review_unit_hash": "5a4d3ca3ecf13c79859842add75908b045e1024341ef657119dffc2fd50219f9", + "prepared_review_unit_hash": "8c45da95a32f4cf0d9ad84a91386f4dfd723ffda7cd05e0da5c978614dd51fce", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1590, + "serialized_chars": 1589, "estimated_input_tokens": 398, "estimated_eval_tokens": 1052 }, @@ -17221,7 +17135,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1590, + "serialized_chars": 1589, "estimated_input_tokens": 398, "estimated_eval_tokens": 1052 }, @@ -17230,18 +17144,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two issues are both broad community-contribution requests, but they target different features: Flash Attention 2 support versus GGUF support. They do not describe the same underlying bug or change, so they should not be clustered as duplicates.", - "confidence": 0.97, - "canonical_issue_reason": null, + "summary": "These are two separate community feature requests: 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 or code path.", + "confidence": 0.98, + "canonical_issue_reason": "Issue 26350 is the stronger anchor only by activity/engagement, but it is not a duplicate of 33260 because the requested features are different.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "26350 has much higher discussion activity and is the more established tracker, so it is the best representative issue here even though the requests are unrelated.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:26350", "right": "issue:33260", "accept": false, - "reason": "Different feature requests in different code paths (Flash Attention 2 vs GGUF). The shared wording about adding support for more architectures is too generic to indicate the same underlying change." + "reason": "Different feature requests (Flash Attention 2 vs GGUF support); shared wording about 'more architectures' is too broad to indicate the same underlying bug or change." } ] }, @@ -17271,12 +17185,12 @@ "soft_pairs": [ "issue:26413|issue:31501" ], - "prepared_review_unit_hash": "7f63e11b860fea25c0fe90c33eeafbc80aaeb7fd31a1d2c4f8c69a9f291e694a", + "prepared_review_unit_hash": "37670e0b969fa8bf6c1a2ee26fa2a530c83bd1951e908505673cc72dc9a44a50", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1594, + "serialized_chars": 1593, "estimated_input_tokens": 399, "estimated_eval_tokens": 1054 }, @@ -17284,7 +17198,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1594, + "serialized_chars": 1593, "estimated_input_tokens": 399, "estimated_eval_tokens": 1054 }, @@ -17293,18 +17207,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two issues both concern training flow failures, but they describe different symptoms and code paths: one is a checkpoint-resume iterator crash, the other is a training job that never starts for FineWeb SLM. I would not treat them as duplicates.", - "confidence": 0.48, - "canonical_issue_reason": "Issue 26413 is the more specific and actionable report, centered on a concrete `resume_from_checkpoint` failure and a clear error message.", + "summary": "These two issues both concern training startup failures, but they do not clearly describe the same underlying bug. The checkpoint-resume issue has a concrete error path; the FineWeb training issue is broader and could have many causes.", + "confidence": 0.76, + "canonical_issue_reason": "issue #26413 is the more specific, technically grounded report with a clear failure signature (`resume_from_checkpoint` / empty `epoch_iterator`), making it the better anchor for this cluster.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 26413 is the best representative because it names a precise failure mode and likely narrower root cause than the broad 'training doesn't start' report.", + "best_issue_reason": "issue #26413 is the best representative because it states an exact error condition and code path, while #31501 is a high-level symptom report without enough evidence of the same root cause.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:26413", "right": "issue:31501", "accept": false, - "reason": "The reports do not describe the same underlying bug: one is a resume/checkpoint epoch-iterator crash, the other is a broader training-start failure on FineWeb. Shared training context alone is not enough to merge them." + "reason": "Both are training-related, but one is a specific checkpoint-resume exception and the other is a generic \u201ctraining doesn\u2019t start\u201d report. There is not enough evidence they describe the same underlying bug." } ] }, @@ -17363,184 +17277,184 @@ "issue:44370|issue:46226", "issue:46222|issue:46226" ], - "prepared_review_unit_hash": "5dda4b39e45831de65a7d22e4605e8c18380a20be289630bf13240754dd3bdb1", + "prepared_review_unit_hash": "17c198444253ef129a9b2afc134de84cbdaac3f35c23cd662abb3b2fe589547e", "component_budget": { "node_count": 8, "item_count": 8, "soft_pair_count": 24, - "serialized_chars": 13524, - "estimated_input_tokens": 3381, - "estimated_eval_tokens": 7018 + "serialized_chars": 13530, + "estimated_input_tokens": 3383, + "estimated_eval_tokens": 7022 }, "budget": { "node_count": 8, "item_count": 8, "soft_pair_count": 24, - "serialized_chars": 13524, - "estimated_input_tokens": 3381, - "estimated_eval_tokens": 7018 + "serialized_chars": 13530, + "estimated_input_tokens": 3383, + "estimated_eval_tokens": 7022 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is mostly a set of distinct i18n translation-request issues for different languages, so they should not be merged across languages. The only true duplicates are the near-identical generic placeholder issues ([i18n-]...) and the identical Thai pair.", - "confidence": 0.97, - "canonical_issue_reason": "No single canonical issue for the whole cluster: most items are separate language-specific translation requests. The generic placeholder issues form a smaller duplicate subgroup only.", + "summary": "This cluster is mostly a set of similarly worded i18n doc-translation requests, but they are not the same underlying issue across languages. The only clear duplicate-like pair is the exact Thai duplicate 46222/46226; the rest look like separate translation requests despite high textual similarity from the shared template.", + "confidence": 0.67, + "canonical_issue_reason": "issue:46222 is the strongest duplicate anchor because it is text-identical to issue:46226, making it the clearest representative of the only confirmed duplicate subcluster.", "canonical_pr_reason": null, - "best_issue_reason": "43834 is the best representative of the generic placeholder duplicate subgroup because it is one of the earliest near-identical template issues, but it is not a global canonical for the full cluster.", + "best_issue_reason": "issue:46222 is the best overall issue representative because it cleanly matches an exact duplicate (46226) and has the strongest evidence of being the canonical copy among the issues here.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:27088", "right": "issue:44370", "accept": false, - "reason": "Different target languages; same translation-request template, but not the same underlying issue." + "reason": "Different language translation requests; same template wording but not the same underlying work." }, { "left": "issue:27088", "right": "issue:44365", "accept": false, - "reason": "Different target languages; not the same underlying translation request." + "reason": "Different language translation requests; similarity is from the shared i18n template, not a duplicate." }, { "left": "issue:27088", "right": "issue:43834", "accept": false, - "reason": "Turkish request vs generic placeholder/template issue; not the same concrete issue." + "reason": "Turkish vs a separate placeholder translation issue; not enough evidence of the same concrete request." }, { "left": "issue:27088", "right": "issue:38435", "accept": false, - "reason": "Different target languages; these are separate translation requests." + "reason": "Different languages (Turkish vs Romanian); these are separate translation tasks." }, { "left": "issue:27088", "right": "issue:28936", "accept": false, - "reason": "Turkish vs Spanish translation request; distinct issues." + "reason": "Different languages (Turkish vs Spanish); not a duplicate." }, { "left": "issue:28936", "right": "issue:43834", "accept": false, - "reason": "Spanish request vs generic placeholder/template issue; not the same underlying issue." + "reason": "Spanish translation request versus a separate placeholder translation issue; high similarity is just template overlap." }, { "left": "issue:28936", "right": "issue:44365", "accept": false, - "reason": "Different target languages; separate translation requests." + "reason": "Different translation requests; not the same underlying issue." }, { "left": "issue:28936", "right": "issue:38435", "accept": false, - "reason": "Spanish vs Romanian translation request; distinct issues." + "reason": "Spanish vs Romanian translation requests are distinct." }, { "left": "issue:28936", "right": "issue:44370", "accept": false, - "reason": "Different target languages; not duplicates." + "reason": "Shared template only; no evidence they refer to the same language/change." }, { "left": "issue:38435", "right": "issue:44370", "accept": false, - "reason": "Romanian request vs generic placeholder/template issue; not the same issue." + "reason": "Romanian translation request versus a separate placeholder issue; not clearly the same work." }, { "left": "issue:38435", "right": "issue:44365", "accept": false, - "reason": "Romanian request vs generic placeholder/template issue; not duplicates." + "reason": "Same broad i18n area, but different translation requests." }, { "left": "issue:38435", "right": "issue:43834", "accept": false, - "reason": "Romanian request vs generic placeholder/template issue; distinct translation tasks." + "reason": "Romanian vs another placeholder translation issue; similarity is not sufficient to call a duplicate." }, { "left": "issue:38435", "right": "issue:46222", "accept": false, - "reason": "Different target languages/content; not the same underlying request." + "reason": "Different language requests; not the same underlying change." }, { "left": "issue:38435", "right": "issue:46226", "accept": false, - "reason": "Different target languages/content; not duplicates." + "reason": "Different language requests; not a duplicate." }, { "left": "issue:43834", "right": "issue:44370", - "accept": true, - "reason": "Near-identical generic placeholder/template issue; same underlying translation-request stub." + "accept": false, + "reason": "Near-identical template text, but no concrete evidence they are the same translation request." }, { "left": "issue:43834", "right": "issue:44365", - "accept": true, - "reason": "Near-identical generic placeholder/template issue; same underlying translation-request stub." + "accept": false, + "reason": "Looks like separate placeholder issues rather than one underlying task." }, { "left": "issue:43834", "right": "issue:46222", "accept": false, - "reason": "Generic placeholder/template issue vs different Thai-language issue; not the same underlying task." + "reason": "Different language/issue context; not enough to merge." }, { "left": "issue:43834", "right": "issue:46226", "accept": false, - "reason": "Generic placeholder/template issue vs different Thai-language issue; not duplicates." + "reason": "Different language/issue context; not a confirmed duplicate." }, { "left": "issue:44365", "right": "issue:44370", - "accept": true, - "reason": "Near-identical generic placeholder/template issue; same underlying translation-request stub." + "accept": false, + "reason": "Very similar template wording, but no explicit evidence of the same underlying issue." }, { "left": "issue:44365", "right": "issue:46222", "accept": false, - "reason": "Generic placeholder/template issue vs different Thai-language issue; distinct issues." + "reason": "Different translation requests; do not merge on template similarity alone." }, { "left": "issue:44365", "right": "issue:46226", "accept": false, - "reason": "Generic placeholder/template issue vs different Thai-language issue; not duplicates." + "reason": "Different translation requests; not the same bug/change." }, { "left": "issue:44370", "right": "issue:46222", "accept": false, - "reason": "Generic placeholder/template issue vs different Thai-language issue; not the same underlying request." + "reason": "Different translation requests; shared template text is not enough." }, { "left": "issue:44370", "right": "issue:46226", "accept": false, - "reason": "Generic placeholder/template issue vs different Thai-language issue; not duplicates." + "reason": "Different translation requests; not a duplicate." }, { "left": "issue:46222", "right": "issue:46226", "accept": true, - "reason": "Identical Thai text and same body shape; true duplicate issue." + "reason": "These are text-identical duplicate Thai translation issues with no distinguishing evidence." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is conservative and grounded in the packet. It correctly treats the language-specific translation requests as distinct issues, and it only groups the clearly near-identical generic placeholder issues plus the identical Thai pair as duplicates. The canonical-issue conclusion is appropriately cautious, and the soft-edge false/true verdicts do not overstate evidence." + "feedback": "The summary is grounded in the packet and stays conservative: the only fully supported duplicate pair is issue:46222/issue:46226, which have identical titles and maximal similarity, while the other high-similarity pairs are reasonably treated as separate translation requests because the titles indicate different languages or placeholder variants. No overreach on mergeability or PR pairing." }, "evaluator_used": true, "retried": false, @@ -17616,12 +17530,12 @@ "issue:43837|issue:45177", "issue:45322|issue:46133" ], - "prepared_review_unit_hash": "cab8c2e546cd855918ab27f1f2a13c38f758e61ca5a26d9ce99d569fcc5bad97", + "prepared_review_unit_hash": "3868bfb8e2cf6ceb881c130fe300e5589df9acbf7e016de416c1d232464df89c", "component_budget": { "node_count": 16, "item_count": 16, "soft_pair_count": 36, - "serialized_chars": 21659, + "serialized_chars": 21660, "estimated_input_tokens": 5415, "estimated_eval_tokens": 11086 }, @@ -17629,7 +17543,7 @@ "node_count": 16, "item_count": 16, "soft_pair_count": 36, - "serialized_chars": 21659, + "serialized_chars": 21660, "estimated_input_tokens": 5415, "estimated_eval_tokens": 11086 }, @@ -17638,228 +17552,228 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is mostly a set of distinct model-request issues with similar wording, not true duplicates. The overlap is broad template similarity around \"add/support model X\" rather than the same underlying bug or change.", + "summary": "This cluster is a set of independent model-addition/support requests across different model families and tasks (BERT variants, detection, audio/TTS/ASR, OCR, vision-language). They share the broad pattern of \"add model support,\" but none appear to be the same underlying change.", "confidence": 0.92, - "canonical_issue_reason": "No single canonical issue: the items request different models, architectures, and modalities, so they are not duplicates of one underlying ticket.", + "canonical_issue_reason": "issue:43671 is the most central and broadly connected request in the set, with the strongest soft links to many of the other model-support issues. It best represents the cluster theme, but it is still a specific Qwen3-TTS request rather than a duplicate of the others.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 43837 is a clear, moderately discussed support request and a reasonable representative of the cluster, but it is still not a duplicate target for the others.", + "best_issue_reason": "issue:43671 is the best representative issue for this cluster because it sits at the center of the similarity graph and reflects the common model-support request pattern most clearly. It should not be treated as a duplicate target for the other items.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:36817", "right": "issue:41211", "accept": false, - "reason": "Different models (EuroBert vs DEIMv2); only generic request-template similarity." + "reason": "Different model families: EuroBert config vs DEIMv2 support." }, { "left": "issue:36817", "right": "issue:43671", "accept": false, - "reason": "Different model families and modalities; not the same change." + "reason": "BERT config request vs Qwen3-TTS audio support; different task and code path." }, { "left": "issue:36817", "right": "issue:36879", "accept": false, - "reason": "EuroBert vs RF-DETR are unrelated model requests." + "reason": "EuroBert vs RF-DETR are unrelated models." }, { "left": "issue:36879", "right": "issue:41211", "accept": false, - "reason": "Different models and likely different integration paths." + "reason": "RF-DETR detection support is unrelated to DEIMv2." }, { "left": "issue:36879", "right": "issue:43909", "accept": false, - "reason": "RF-DETR and LFM2.5 Audio are unrelated requests." + "reason": "RF-DETR detection vs LFM2.5 audio support are different changes." }, { "left": "issue:36879", "right": "issue:44121", "accept": false, - "reason": "Object detection model vs sparse transformer request; not same underlying task." + "reason": "Different model families and implementation paths." }, { "left": "issue:36879", "right": "issue:43837", "accept": false, - "reason": "Different model requests (RF-DETR vs Qwen3-ASR)." + "reason": "RF-DETR vs Qwen3-ASR are different modalities and models." }, { "left": "issue:36879", "right": "issue:43671", "accept": false, - "reason": "RF-DETR and Qwen3-TTS are different modalities and code paths." + "reason": "RF-DETR and Qwen3-TTS are unrelated requests." }, { "left": "issue:37015", "right": "issue:43837", "accept": false, - "reason": "NeoBERT and Qwen3-ASR are unrelated model additions." + "reason": "NeoBERT support is unrelated to Qwen3-ASR." }, { "left": "issue:37015", "right": "issue:44121", "accept": false, - "reason": "Different model families; only broad 'add model' similarity." + "reason": "NeoBERT and weight-sparse transformer are different model additions." }, { "left": "issue:37015", "right": "issue:43909", "accept": false, - "reason": "NeoBERT vs LFM2.5 Audio are separate requests." + "reason": "NeoBERT vs audio model support are different changes." }, { "left": "issue:36879", "right": "issue:37015", "accept": false, - "reason": "RF-DETR and NeoBERT are distinct models with no shared concrete fix." + "reason": "RF-DETR and NeoBERT are distinct model requests." }, { "left": "issue:37015", "right": "issue:42503", "accept": false, - "reason": "NeoBERT and ModernVBERT are different model integrations." + "reason": "NeoBERT and ModernVBERT are different model families." }, { "left": "issue:41115", "right": "issue:43837", "accept": false, - "reason": "MiniCPM3 vs Qwen3-ASR are different model requests." + "reason": "MiniCPM3 support is separate from Qwen3-ASR support." }, { "left": "issue:41115", "right": "issue:44121", "accept": false, - "reason": "MiniCPM3 and weight-sparse transformer are not the same underlying change." + "reason": "MiniCPM3 and weight-sparse transformer are not the same code change." }, { "left": "issue:41115", "right": "issue:42503", "accept": false, - "reason": "MiniCPM3 and ModernVBERT are separate model-add tickets." + "reason": "MiniCPM3 vs ModernVBERT are different model requests." }, { "left": "issue:41115", "right": "issue:45027", "accept": false, - "reason": "MiniCPM3 and Voxtral TTS are different model targets." + "reason": "MiniCPM3 and Voxtral TTS are unrelated models." }, { "left": "issue:41211", "right": "issue:43671", "accept": false, - "reason": "DEIMv2 and Qwen3-TTS are unrelated additions." + "reason": "DEIMv2 and Qwen3-TTS are different models and tasks." }, { "left": "issue:42503", "right": "issue:43837", "accept": false, - "reason": "ModernVBERT and Qwen3-ASR are different model families and tasks." + "reason": "ModernVBERT models and Qwen3-ASR are unrelated." }, { "left": "issue:42503", "right": "issue:43671", "accept": false, - "reason": "ModernVBERT and Qwen3-TTS are not the same code-path change." + "reason": "ModernVBERT and Qwen3-TTS are different support requests." }, { "left": "issue:42503", "right": "issue:42738", "accept": false, - "reason": "ModernVBERT vs BERT-like models with RoPE is a broad thematic overlap, not a duplicate." + "reason": "Related BERT wording, but not the same underlying change: model addition vs RoPE variant support." }, { "left": "issue:42738", "right": "issue:45027", "accept": false, - "reason": "RoPE support request vs Voxtral TTS support request are different changes." + "reason": "BERT-like RoPE support is unrelated to Voxtral TTS." }, { "left": "issue:43671", "right": "issue:43837", "accept": false, - "reason": "Both are Qwen3 speech-related, but TTS and ASR are different concrete model integrations." + "reason": "Qwen3-TTS and Qwen3-ASR are different tasks and would not be one merged PR." }, { "left": "issue:43671", "right": "issue:45027", "accept": false, - "reason": "Qwen3-TTS and Voxtral TTS are different models and implementations." + "reason": "Both are TTS-related in title only; they target different model families and code paths." }, { "left": "issue:43671", "right": "issue:45177", "accept": false, - "reason": "Different models and domains; no shared duplicate bug." + "reason": "TTS support and OCR2 support are unrelated." }, { "left": "issue:43671", "right": "issue:44121", "accept": false, - "reason": "TTS support request vs sparse transformer request are unrelated." + "reason": "Qwen3-TTS support is unrelated to the sparse transformer request." }, { "left": "issue:43671", "right": "issue:43909", "accept": false, - "reason": "Qwen3-TTS and LFM2.5 Audio are different audio-model requests." + "reason": "Qwen3-TTS and LFM2.5 Audio are distinct model additions." }, { "left": "issue:43837", "right": "issue:43909", "accept": false, - "reason": "ASR support and LFM2.5 Audio are different model additions." + "reason": "ASR support vs audio model support are different requests." }, { "left": "issue:43837", "right": "issue:45322", "accept": false, - "reason": "Qwen3-ASR and EUPE target different model families and tasks." + "reason": "Qwen3-ASR and EUPE are different model families and tasks." }, { "left": "issue:43837", "right": "issue:44121", "accept": false, - "reason": "ASR support request vs weight-sparse transformer request are unrelated." + "reason": "ASR support is unrelated to the weight-sparse transformer request." }, { "left": "issue:43909", "right": "issue:44205", "accept": false, - "reason": "LFM2.5 Audio and SAM3-LiteText are different model requests." + "reason": "Audio model support vs SAM3-LiteText are unrelated." }, { "left": "issue:44121", "right": "issue:45322", "accept": false, - "reason": "Sparse transformer and EUPE are different concrete model changes." + "reason": "Weight-sparse transformer and EUPE are different underlying changes." }, { "left": "issue:43837", "right": "issue:45027", "accept": false, - "reason": "Qwen3-ASR and Voxtral TTS are distinct speech-model integrations." + "reason": "Qwen3-ASR and Voxtral TTS are different model support changes." }, { "left": "issue:37015", "right": "issue:45027", "accept": false, - "reason": "NeoBERT and Voxtral TTS are unrelated requests." + "reason": "NeoBERT and Voxtral TTS are unrelated model requests." }, { "left": "issue:43837", "right": "issue:45177", "accept": false, - "reason": "ASR support and DeepSeek-OCR2 are different model additions." + "reason": "ASR support and DeepSeek-OCR2 support are different tasks." }, { "left": "issue:45322", "right": "issue:46133", "accept": false, - "reason": "EUPE and TIPSv2 are both vision-language related, but they are different models and changes, not the same fix." + "reason": "EUPE and TIPSv2 are distinct vision-language model requests." } ] }, @@ -17889,12 +17803,12 @@ "soft_pairs": [ "issue:41084|issue:43037" ], - "prepared_review_unit_hash": "3fb562e696d96d146870bfd902d591beefcb11bc0f24d0d1a7d033da25125203", + "prepared_review_unit_hash": "afcf19f5cfa2617fa95ec40963375d01fc2748d28373b1e14f13e5d9b068fa8f", "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 }, @@ -17902,7 +17816,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1479, + "serialized_chars": 1480, "estimated_input_tokens": 370, "estimated_eval_tokens": 996 }, @@ -17911,18 +17825,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two issues are not duplicates: one requests block decoding behavior, while the other asks for DeepSeek v3.2 model support. They describe different features/problems with no shared targets or code path.", - "confidence": 0.96, + "summary": "These two issues are not duplicates: one asks for Set Block Decoding, the other requests DeepSeek v3.2 support. They describe different user-facing problems and have no shared targets or code-path evidence.", + "confidence": 0.98, "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "No canonical issue stands out because the issues are about unrelated functionality rather than the same underlying bug or feature request.", + "best_issue_reason": "Issue #43037 is the more concrete and actionable request, but it is still unrelated to #41084 rather than a duplicate of it.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:41084", "right": "issue:43037", "accept": false, - "reason": "Different requests: block decoding vs DeepSeek v3.2 support. No evidence they refer to the same bug, change, or code path." + "reason": "Different topics entirely: block decoding vs model support for DeepSeek v3.2. No shared filenames, targets, or code path evidence." } ] }, @@ -17952,7 +17866,7 @@ "soft_pairs": [ "issue:43598|issue:43600" ], - "prepared_review_unit_hash": "5e286194a90ffcbc2b4a7f2c542f0e1916bd06993b08db088c7b714030ebb4f4", + "prepared_review_unit_hash": "ee7f3caea1d265221a4c99a923b80b6e92f37127fcfe2ab261200f352e17fbe9", "component_budget": { "node_count": 2, "item_count": 2, @@ -17974,18 +17888,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two issues are related to Trainer extensibility, but they describe different problems: one is about `remove_unused_column` behavior, the other about FLOPS estimation for custom trainers. They should not be merged as duplicates.", - "confidence": 0.92, - "canonical_issue_reason": "Issue #43598 is the more concrete and narrowly scoped request, centered on a specific Trainer behavior (`remove_unused_column`) and its customizability.", + "summary": "These are related only at a very high level (Trainer customization), but they describe different problems: one is about `remove_unused_column` behavior, the other about FLOPS estimation for custom trainers. I would not treat them as duplicates.", + "confidence": 0.94, + "canonical_issue_reason": "Issue 43598 is the more general Trainer customizability request and is the better representative if one issue must be kept.", "canonical_pr_reason": null, - "best_issue_reason": "Issue #43598 is the best issue to keep as the representative item because it has a clear, actionable change request tied to one Trainer code path.", + "best_issue_reason": "Issue 43598 is the better global anchor for this cluster because it centers on Trainer behavior customization rather than a narrower utility concern.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43598", "right": "issue:43600", "accept": false, - "reason": "They concern different Trainer behaviors: column removal customizability vs FLOPS estimation for custom trainers. Shared theme is extensibility, but not the same underlying bug or change." + "reason": "They concern different code paths and outcomes: `remove_unused_column` customizability versus FLOPS estimation for custom trainers. Shared Trainer context is too broad to imply the same underlying bug." } ] }, @@ -18015,12 +17929,12 @@ "soft_pairs": [ "issue:43684|issue:44405" ], - "prepared_review_unit_hash": "e9206a51b1e18b30d13ed162778a8992d3e4c12ce37a4939cd43171183491bb1", + "prepared_review_unit_hash": "888e9bf61323e8c0a8cdd3ba357fe6d5dc49ce63bf5a15a6f5b244489aa8219b", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1612, + "serialized_chars": 1611, "estimated_input_tokens": 403, "estimated_eval_tokens": 1062 }, @@ -18028,7 +17942,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1612, + "serialized_chars": 1611, "estimated_input_tokens": 403, "estimated_eval_tokens": 1062 }, @@ -18037,18 +17951,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are related Qwen feature requests, but they are not duplicates: one asks for registering Qwen3-Omni with AutoModel/AutoModelForConditionalGeneration, while the other requests AutoModelForSequenceClassification support for Qwen3.5 (Qwen3_5Config). They target different model families and different auto-class APIs.", - "confidence": 0.96, - "canonical_issue_reason": "No single canonical duplicate here; the requests are about different Qwen variants and different auto-model classes.", + "summary": "These are related only at a very broad level (Qwen model support), but they describe different model families and different auto-class registration gaps, so they should not be deduplicated.", + "confidence": 0.89, + "canonical_issue_reason": "Issue 43684 is the broader, more central request: it asks for model registration in AutoModel and AutoModelForConditionalGeneration, which is the more general missing-support problem.", "canonical_pr_reason": null, - "best_issue_reason": "Issue #43684 is the broader, more foundational registration request and is the better representative of this small cluster\u2019s theme.", + "best_issue_reason": "Issue 43684 is the best representative because it covers the core registration gap in the auto-model machinery, whereas 44405 is a narrower sequence-classification-specific request for a different config family.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43684", "right": "issue:44405", "accept": false, - "reason": "Different underlying requests: Qwen3-Omni registration for AutoModel/AutoModelForConditionalGeneration vs. Qwen3.5 sequence classification support. Different model/config targets and different code paths." + "reason": "Different underlying changes: one requests Qwen3-Omni registration for generic/conditional-generation auto classes, the other requests AutoModelForSequenceClassification support for Qwen3.5. Same broad area, but not the same bug or mergeable fix." } ] }, @@ -18085,12 +17999,12 @@ "issue:43845|issue:43859", "issue:43846|issue:43859" ], - "prepared_review_unit_hash": "2adaf9613d744269dfa53ce4f61ddd96931983a92b51391988eb966ca4e92992", + "prepared_review_unit_hash": "0092da70d7afde1e86b61af67a0bfb8ef34dbc43c9f41841072d29b6b41776b0", "component_budget": { "node_count": 4, "item_count": 4, "soft_pair_count": 6, - "serialized_chars": 4266, + "serialized_chars": 4267, "estimated_input_tokens": 1067, "estimated_eval_tokens": 2390 }, @@ -18098,7 +18012,7 @@ "node_count": 4, "item_count": 4, "soft_pair_count": 6, - "serialized_chars": 4266, + "serialized_chars": 4267, "estimated_input_tokens": 1067, "estimated_eval_tokens": 2390 }, @@ -18107,58 +18021,55 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Three near-identical closed issues titled \"huggingface\" look like duplicates of the same report; the \"Fett\" issue appears unrelated. I\u2019d pick the most active \"huggingface\" issue as the canonical representative.", - "confidence": 0.82, - "canonical_issue_reason": "Issue 43859 is the best representative among the duplicate-looking \"huggingface\" issues because it shares the same title and has the most discussion activity.", + "summary": "All four items are low-information closed issues. Three share the exact title \"huggingface \" but there is no body/context to confirm they describe the same underlying bug; the other issue titled \"Fett\" looks unrelated. I would keep one of the repeated \"huggingface \" reports as the representative issue, but I would not merge any pair based on the available evidence.", + "confidence": 0.89, + "canonical_issue_reason": "issue:43845 is the best representative of the repeated \"huggingface \" reports because it is the earliest of the near-identical titles and has no less evidence than the later duplicates.", "canonical_pr_reason": null, - "best_issue_reason": "43859 is the strongest issue candidate: same apparent topic as the others and slightly more activity, making it the best cluster representative.", + "best_issue_reason": "issue:43845 is the strongest global issue candidate: it matches the repeated title cluster, is earliest among the duplicates, and there is no evidence that another issue is a better source report.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:43835", "right": "issue:43845", "accept": false, - "reason": "Different titles and no evidence they describe the same bug or request." + "reason": "Different titles and no body details. \"Fett\" has no visible relation to the repeated \"huggingface \" reports." }, { "left": "issue:43835", "right": "issue:43846", "accept": false, - "reason": "Different titles and no concrete shared problem shown." + "reason": "Different titles and no substantive description to indicate the same underlying bug." }, { "left": "issue:43835", "right": "issue:43859", "accept": false, - "reason": "Different title; no sign this is the same underlying issue." + "reason": "Different titles and no evidence tying this issue to the \"huggingface \" reports." }, { "left": "issue:43845", "right": "issue:43846", - "accept": true, - "reason": "Same title, created back-to-back, and no distinguishing evidence suggests they are separate reports." + "accept": false, + "reason": "Titles match, but the issues are too empty to confirm they report the same concrete problem; identical phrasing alone is not enough." }, { "left": "issue:43845", "right": "issue:43859", - "accept": true, - "reason": "Identical title and very similar metadata strongly suggest duplicate reports of the same issue." + "accept": false, + "reason": "Same generic title, but there is no body or technical context to prove these are duplicate reports of one bug." }, { "left": "issue:43846", "right": "issue:43859", - "accept": true, - "reason": "Identical title and near-identical timing make them likely duplicates of the same report." + "accept": false, + "reason": "Same generic title, but the available evidence is insufficient to establish they describe the same underlying issue." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary is grounded in the packet: 43845/43846/43859 share the same title and close creation times, while 43835 differs in title and has no evidence tying it to the others. The canonical choice of 43859 is conservative and supported by its slightly higher discussion activity. No PR-related claims were made." - }, - "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 }, @@ -18192,7 +18103,7 @@ "issue:44572|issue:44937", "issue:44573|issue:44937" ], - "prepared_review_unit_hash": "b9bd213bee093e8f6842b80ef5768a39bc5aeee011b5b22cfea493178d62ee43", + "prepared_review_unit_hash": "7ade19427739b93dd009c2ff6f807100171492bf62a5d0bd034c9ce677180429", "component_budget": { "node_count": 5, "item_count": 5, @@ -18214,66 +18125,66 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster looks like unrelated low-information/spam issues rather than a duplicate set. One item is a documentation suggestion, one mentions \"Google LLC\", and the rest are generic spam placeholders.", - "confidence": 0.91, - "canonical_issue_reason": "No clear canonical issue: the items do not describe the same underlying bug or change, so there is no valid representative duplicate to merge around.", + "summary": "This cluster looks like unrelated low-information/spam issues rather than true duplicates. There is no evidence of a shared bug or request, and the only somewhat descriptive title is still too vague to anchor a canonical duplicate.", + "confidence": 0.93, + "canonical_issue_reason": "No strong canonical issue: the items have different titles, no bodies or linked references, and no shared technical details. They do not appear to describe the same underlying problem.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 44418 is the only semantically readable title, but it still does not match the other spam-like items and is not a good duplicate representative.", + "best_issue_reason": "If forced to pick a representative, issue 44418 is the least opaque because it has the only non-spam-like title, but it is still not a convincing canonical duplicate target.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:44371", "right": "issue:44572", "accept": false, - "reason": "Both are spam-labeled placeholders, but there is no evidence they report the same underlying issue or change." + "reason": "Both are titled \"\" with no body or links, so there is no concrete evidence they report the same issue." }, { "left": "issue:44371", "right": "issue:44573", "accept": false, - "reason": "Identical placeholder-style titles are not enough to treat them as duplicates; no shared bug/change is described." + "reason": "Same generic spam title, but no technical content or references to show a shared underlying bug." }, { "left": "issue:44371", "right": "issue:44418", "accept": false, - "reason": "A spam placeholder and a documentation enhancement suggestion are clearly different kinds of issues." + "reason": "One is spam-like and the other is a generic documentation suggestion; they do not look like the same report." }, { "left": "issue:44371", "right": "issue:44937", "accept": false, - "reason": "No sign these describe the same problem; one is a spam placeholder and the other is a distinct titled item." + "reason": "The titles are unrelated and there is no content suggesting a common root cause." }, { "left": "issue:44418", "right": "issue:44572", "accept": false, - "reason": "Documentation enhancement and spam placeholder are unrelated; no shared underlying change is evident." + "reason": "A vague documentation suggestion does not match a spam placeholder issue." }, { "left": "issue:44418", "right": "issue:44573", "accept": false, - "reason": "Documentation enhancement and spam placeholder are unrelated; not the same issue." + "reason": "No shared subject matter beyond being low-information issues; not the same underlying bug or change." }, { "left": "issue:44572", "right": "issue:44573", "accept": false, - "reason": "Same placeholder title alone is insufficient; there is no content tying them to the same underlying bug or request." + "reason": "Identical spam titles are not enough to merge them as duplicates without any substantive body or evidence." }, { "left": "issue:44572", "right": "issue:44937", "accept": false, - "reason": "No concrete overlap is shown between a spam placeholder and the \"Google LLC\" issue." + "reason": "The titles are unrelated and the available metadata gives no indication of the same problem." }, { "left": "issue:44573", "right": "issue:44937", "accept": false, - "reason": "No concrete overlap is shown between a spam placeholder and the \"Google LLC\" issue." + "reason": "No shared technical details, targets, or references; these are not the same issue." } ] }, @@ -18303,12 +18214,12 @@ "soft_pairs": [ "issue:45145|issue:45151" ], - "prepared_review_unit_hash": "f12390b74d023fc2eb88cd190687a802b139dcb41ffd1658646cd52a2c218826", + "prepared_review_unit_hash": "ec34912142b4604103177982616d4fa4e7c58c20934f235dbb2601b3e38fc1a5", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1646, + "serialized_chars": 1645, "estimated_input_tokens": 412, "estimated_eval_tokens": 1080 }, @@ -18316,7 +18227,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1646, + "serialized_chars": 1645, "estimated_input_tokens": 412, "estimated_eval_tokens": 1080 }, @@ -18325,28 +18236,28 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two closed issues have the same title and very similar body size, so they are likely duplicate or near-duplicate reports. #45145 is the better canonical issue because it was created first and already has an inbound reference.", - "confidence": 0.84, - "canonical_issue_reason": "#45145 is earlier and has an inbound reference, so it is the more established duplicate target.", + "summary": "The two closed issues appear likely to be duplicates: they share the same title and very similar body sizes, and #45145 has the only inbound reference, so it is the better canonical representative.", + "confidence": 0.87, + "canonical_issue_reason": "#45145 appears to be the better canonical issue because it has an inbound reference, making it slightly better connected in the discussion trail.", "canonical_pr_reason": null, - "best_issue_reason": "#45145 is the best single issue to keep because it appears to be the earliest and most representative report of the pair.", + "best_issue_reason": "#45145 is the strongest issue representative because it appears to be the more connected copy of the same likely duplicate report.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45145", "right": "issue:45151", - "accept": false, - "reason": "They look likely related or duplicate, but the packet only provides soft similarity plus basic metadata; that is not enough to confidently treat them as the same underlying issue." + "accept": true, + "reason": "Identical titles and very similar body lengths make these issues appear to describe the same likely report, though the evidence is limited." } ] }, "evaluator_result": { "accept": false, - "feedback": "The packet supports a likely duplicate/near-duplicate relationship: the titles match exactly, the bodies are similar in size, and #45145 is earlier with one inbound reference. However, the soft-edge verdict is too strong for the evidence available. The packet only provides soft_similarity plus metadata, so the conclusion should stay conservative (e.g., 'likely duplicate' or 'probably same report'), not 'same underlying issue rather than merely related reports.'" + "feedback": "The pairing is plausible, but the writeup is a bit too strong for the evidence available. The packet only shows identical titles, similar body lengths, and a soft_similarity signal; it does not establish that these are definitely duplicates. Please soften the language to \u2018likely\u2019/\u2018appear to be\u2019 throughout, especially in the soft-edge reason and confidence." }, "evaluator_used": true, "retried": true, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -18423,7 +18334,7 @@ "soft_pairs": [ "issue:45672|issue:45674" ], - "prepared_review_unit_hash": "2b455ed2231a85a0e27d0db2c41d37847a11bb50729e400c6936a236e129b06e", + "prepared_review_unit_hash": "d22306be406110f45c523b58ae1363b61120985338f421e14dc925759f2a4732", "component_budget": { "node_count": 2, "item_count": 2, @@ -18445,18 +18356,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both items are about 4-bit quantized inference, but they describe different regressions in different code paths: one is a Gemma4/audio-module uint8 vs. torch.finfo TypeError, the other is a BitsAndBytesConfig skip-modules regression that removes the default lm_head exclusion and triggers an AssertionError.", - "confidence": 0.95, - "canonical_issue_reason": null, + "summary": "These are related 4-bit inference issues, but they describe different bugs: one is a Gemma4/audio-module uint8 dtype failure in `torch.finfo()`, the other is a `BitsAndBytesConfig.llm_int8_skip_modules` regression that drops the default `lm_head` exclusion and triggers an assertion. They should not be deduplicated.", + "confidence": 0.91, + "canonical_issue_reason": "Issue #45674 is the more general and actionable report: it names the configuration regression, the affected setting, and the concrete 4-bit inference failure path.", "canonical_pr_reason": null, - "best_issue_reason": "Issue 45674 is the cleaner standalone report: it names the configuration knob, the exact regression, and the resulting AssertionError, making it the better representative issue in this small cluster.", + "best_issue_reason": "Issue #45674 is the best representative of the cluster because it describes a central library-side bug rather than a model-specific symptom.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45672", "right": "issue:45674", "accept": false, - "reason": "Not the same underlying bug: 45672 concerns Gemma4/audio-module uint8 weights hitting torch.finfo during NF4 inference, while 45674 concerns llm_int8_skip_modules clearing the default lm_head exclusion and causing an AssertionError. They share only the broad 4-bit inference context." + "reason": "Different underlying bugs and code paths. #45672 is Gemma4/audio-module-specific `torch.finfo()` TypeError on uint8 weights; #45674 is a `BitsAndBytesConfig` regression where `llm_int8_skip_modules` clears the default `lm_head` exclusion and causes an `AssertionError` in 4-bit inference." } ] }, @@ -18486,7 +18397,7 @@ "soft_pairs": [ "issue:45762|issue:45763" ], - "prepared_review_unit_hash": "558f54fd9b8d868ea55224a139067871178efe8bb04ebe6b9a4dc069ea63d630", + "prepared_review_unit_hash": "bfbb7a22b10662deda6f06fa65fa922505f2f0dcd0cbe7881d8d2718c7b2406d", "component_budget": { "node_count": 2, "item_count": 2, @@ -18508,18 +18419,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two issues are unrelated closed items with only superficial title similarity (personal names + dates). There is no evidence they describe the same bug or request, so they should not be clustered as duplicates.", - "confidence": 0.97, - "canonical_issue_reason": null, + "summary": "The two closed issues are not duplicates; they are unrelated placeholder/personal-name titles with no shared bug report context.", + "confidence": 0.98, + "canonical_issue_reason": "No canonical issue: the issues do not describe the same problem or request, so neither can serve as a representative duplicate target.", "canonical_pr_reason": null, - "best_issue_reason": null, + "best_issue_reason": "Neither issue is a good global representative; both are standalone, unrelated entries with no actionable problem statement.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45762", "right": "issue:45763", "accept": false, - "reason": "Different named subjects and different dates; no shared problem statement, target, or code context. The similarity is incidental, not duplicate evidence." + "reason": "Titles only share a surname-like token and both are bare name/date strings; there is no evidence they report the same underlying issue." } ] }, @@ -18549,40 +18460,40 @@ "soft_pairs": [ "issue:45764|issue:45765" ], - "prepared_review_unit_hash": "c9f858fcd367c816893b2f046e70e47f7abc68c12a559cfb4c641720cd8aa929", + "prepared_review_unit_hash": "5c956266222604b78382b0c18326765328a9b9a0b956c2cd79323214457cc791", "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": "These two closed issues are just separate personal-name/date entries with no shared bug context, so they should not be clustered as duplicates.", - "confidence": 0.98, - "canonical_issue_reason": "Issue 45764 is the earliest-created item and, given the lack of any substantive body or context in either issue, is the most reasonable placeholder canonical issue.", + "summary": "The two issues are not convincing duplicates: they are just two different personal-name/date titles with no body, no links, and no shared technical context. The soft similarity signal is too weak to treat them as the same underlying issue.", + "confidence": 0.96, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Issue 45764 is the least arbitrary representative for the same reason: earliest created and equally sparse as the other issue.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45764", "right": "issue:45765", "accept": false, - "reason": "The titles refer to different people and dates, with no shared problem statement, code path, or explicit linkage; they do not appear to describe the same underlying issue." + "reason": "Different names and dates, no description or evidence of a shared bug/report; not the same underlying issue." } ] }, @@ -18612,40 +18523,40 @@ "soft_pairs": [ "issue:45810|issue:46295" ], - "prepared_review_unit_hash": "014f8f99a0ecbf77a1aff23c8ef51296065ea325282fed5ddeaea0682ab9172e", + "prepared_review_unit_hash": "da00cc043a3158f7fd01531d3ec003a10d2f5e0b4ac107f1930e188cf5e09d38", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1541, - "estimated_input_tokens": 386, - "estimated_eval_tokens": 1028 + "serialized_chars": 1540, + "estimated_input_tokens": 385, + "estimated_eval_tokens": 1026 }, "budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 1541, - "estimated_input_tokens": 386, - "estimated_eval_tokens": 1028 + "serialized_chars": 1540, + "estimated_input_tokens": 385, + "estimated_eval_tokens": 1026 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two issues are not duplicates: one requests a new Qwen3_5 token-classification/value-model class, while the other asks for FP8 training support. They concern different features and code paths.", - "confidence": 0.96, - "canonical_issue_reason": "Issue #45810 is the more concrete, implementation-specific request; it names a missing model class and a clear target use case.", + "summary": "These two issues are not duplicates: one requests a new Qwen token-classification model class, while the other asks for FP8 training support.", + "confidence": 0.97, + "canonical_issue_reason": null, "canonical_pr_reason": null, - "best_issue_reason": "Issue #45810 is the best representative issue because it is narrowly scoped and directly tied to an actionable model-class addition.", + "best_issue_reason": null, "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45810", "right": "issue:46295", "accept": false, - "reason": "These are unrelated feature requests: one is for adding Qwen3_5ForTokenClassification, the other is for FP8 training support. They do not describe the same underlying bug or change." + "reason": "Different feature requests in different areas: model class addition vs FP8 training support. They do not describe the same underlying bug or change." } ] }, @@ -18685,92 +18596,92 @@ "issue:45871|issue:45873", "issue:45872|issue:45873" ], - "prepared_review_unit_hash": "12b022c514266ee11c7a182d270a830930e482c70066c872d3b2e9c473dd12d2", + "prepared_review_unit_hash": "3fd48f9b1e16ab7fbc31db8c262ebaef503f8fe96e9b9af2d69cff3719906124", "component_budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 8, - "serialized_chars": 5724, - "estimated_input_tokens": 1431, - "estimated_eval_tokens": 3118 + "serialized_chars": 5727, + "estimated_input_tokens": 1432, + "estimated_eval_tokens": 3120 }, "budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 8, - "serialized_chars": 5724, - "estimated_input_tokens": 1431, - "estimated_eval_tokens": 3118 + "serialized_chars": 5727, + "estimated_input_tokens": 1432, + "estimated_eval_tokens": 3120 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The cluster appears to split into two highly similar issue pairs\u201445870/45871 and 45872/45873\u2014plus one unrelated dataset promotion issue (45869). Based on the available metadata, I would not call any pair proven exact duplicates; the evidence is suggestive but not conclusive.", - "confidence": 0.84, - "canonical_issue_reason": "No single issue cleanly represents the whole cluster because it contains at least two distinct topic groups and an unrelated dataset post. If a representative is needed, 45870 is the best anchor for the first DataForge example pair because it is the earlier of the two nearly identical reports.", + "summary": "The cluster most likely contains two strong duplicate pairs\u201445870/45871 and 45872/45873\u2014plus one unrelated dataset post (45869). The evidence supports likely duplicates, but not confirmed exact duplication from the packet alone.", + "confidence": 0.9, + "canonical_issue_reason": "No single canonical issue fits all five items because the cluster splits into two separate likely-duplicate pairs with different concrete topics, plus one unrelated issue.", "canonical_pr_reason": null, - "best_issue_reason": "45870 is the strongest overall issue representative only as the earliest, cleanest member of the most tightly matched pair. It is not a perfect global representative because 45872/45873 cover a different DataForge topic and 45869 is unrelated.", + "best_issue_reason": "45870 is the earliest clean representative of the 'Example: Financial time series prediction with DataForge' pair, but it only represents that pair and not the separate scraping-service pair.", "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "issue:45869", "right": "issue:45870", "accept": false, - "reason": "Dataset promotion vs DataForge prediction example; different topics." + "reason": "Dataset bundle promo vs. financial prediction example: related finance wording, but different underlying request." }, { "left": "issue:45869", "right": "issue:45871", "accept": false, - "reason": "Dataset promotion vs DataForge prediction example; different topics." + "reason": "Dataset bundle promo vs. financial prediction example; not the same issue." }, { "left": "issue:45870", "right": "issue:45871", - "accept": false, - "reason": "Same title and similar metadata, but the packet does not prove an exact duplicate or same underlying issue." + "accept": true, + "reason": "Same title and matching body length with very strong similarity; this is a likely duplicate pair, though not proven exact from metadata alone." }, { "left": "issue:45870", "right": "issue:45872", "accept": false, - "reason": "Both mention finance/DataForge, but one is a prediction example and the other is a scraping-service topic; distinct requests." + "reason": "Prediction example vs. scraping-service integration are different concrete asks." }, { "left": "issue:45870", "right": "issue:45873", "accept": false, - "reason": "Both mention finance/DataForge, but one is a prediction example and the other is a scraping-service topic; distinct requests." + "reason": "Prediction example vs. scraping-service integration are different concrete asks." }, { "left": "issue:45871", "right": "issue:45872", "accept": false, - "reason": "Prediction example vs scraping-service topic; not the same underlying issue." + "reason": "Different underlying topics despite broad DataForge/finance overlap." }, { "left": "issue:45871", "right": "issue:45873", "accept": false, - "reason": "Prediction example vs scraping-service topic; not the same underlying issue." + "reason": "Different underlying topics despite broad DataForge/finance overlap." }, { "left": "issue:45872", "right": "issue:45873", - "accept": false, - "reason": "Same title and similar metadata, but the packet does not prove an exact duplicate or same underlying issue." + "accept": true, + "reason": "Same title and matching body length with very strong similarity; this is a likely duplicate pair, though not proven exact from metadata alone." } ] }, "evaluator_result": { "accept": false, - "feedback": "The summary overstates the evidence. The packet only provides titles, body lengths, and soft-similarity scores; it does not prove any pair is an exact duplicate. In particular, calling 45870/45871 and 45872/45873 \"clear\" or \"exact\" duplicates is stronger than the available metadata supports, and every pair in the packet has deterministic_accept=false. A more conservative summary would say the cluster appears to split into two highly similar topic pairs plus an unrelated dataset post, without asserting exact duplication." + "feedback": "The analysis is too strong for the evidence in the packet. We only have titles, body lengths, and soft-similarity signals; there is no body text proving exact duplication. In particular, calling 45870/45871 and 45872/45873 \"exact duplicate\" pairs overstates what is grounded here. A more conservative summary would say they are likely duplicates or strong matches, not confirmed exact duplicates. " }, "evaluator_used": true, "retried": true, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -18806,8 +18717,8 @@ ], "soft_pairs": [ "pull_request:39895|pull_request:43424", - "pull_request:41855|pull_request:43424", "pull_request:41855|pull_request:43823", + "pull_request:41855|pull_request:43424", "pull_request:42668|pull_request:43823", "pull_request:43251|pull_request:43823", "pull_request:43251|pull_request:44827", @@ -18828,12 +18739,12 @@ "pull_request:44827|pull_request:46048", "pull_request:43251|pull_request:43254" ], - "prepared_review_unit_hash": "cba00d6424d124a7562e958998ca7cc20b0987b39b7d0ac50401fffcbeed37d2", + "prepared_review_unit_hash": "66c7f538f489f1a16daabb1d18d018c2f853d2d72de246a5d90ffc4651784c72", "component_budget": { "node_count": 16, "item_count": 16, "soft_pair_count": 22, - "serialized_chars": 43088, + "serialized_chars": 43086, "estimated_input_tokens": 10772, "estimated_eval_tokens": 21800 }, @@ -18841,7 +18752,7 @@ "node_count": 16, "item_count": 16, "soft_pair_count": 22, - "serialized_chars": 43088, + "serialized_chars": 43086, "estimated_input_tokens": 10772, "estimated_eval_tokens": 21800 }, @@ -18850,150 +18761,150 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Only one true duplicate pair appears here: the two `fixed_cross_entropy` PRs both address issue #43240 by adding supported kwargs to the same loss call. The rest are distinct model additions, refactors, or unrelated fixes that merely share broad subsystem similarity.", - "confidence": 0.96, + "summary": "Only one true duplicate pair stands out: two PRs that both extend `fixed_cross_entropy` to forward `weight` and `label_smoothing` into `nn.functional.cross_entropy`. The rest are unrelated changes that only overlap at a broad subsystem level or via generic soft similarity.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #43251 is the strongest canonical representative: it has the explicit issue target (#43240) and the fuller implementation of the same `fixed_cross_entropy` fix.", + "canonical_pr_reason": "PR 43251 is the cleaner canonical representative of the shared loss-function fix: it has the explicit issue target (43240) and directly patches `fixed_cross_entropy` to pass the missing kwargs through.", "best_issue_reason": null, - "best_pr_reason": "PR #43251 is the best overall representative because it directly fixes the tracked issue and cleanly captures the underlying change in one place.", + "best_pr_reason": "PR 43251 is the best overall representative of the cluster because it is the only clearly grounded bug-fix PR here, with a concrete issue link and a narrowly scoped implementation in the core loss helper.", "soft_edge_verdicts": [ { "left": "pull_request:39895", "right": "pull_request:43424", "accept": false, - "reason": "Videoprism model addition vs executorch dynamic-shape export test; unrelated changes." + "reason": "Different changes: VideoPrism model/docs scaffolding vs an ExecuTorch dynamic-shape test." }, { "left": "pull_request:41855", - "right": "pull_request:43424", + "right": "pull_request:43823", "accept": false, - "reason": "Mistral tokenizer method additions vs executorch export tests; different code paths and goals." + "reason": "Mistral tokenizer missing methods vs adding MobileLLM; unrelated model areas and code paths." }, { "left": "pull_request:41855", - "right": "pull_request:43823", + "right": "pull_request:43424", "accept": false, - "reason": "Tokenizer API additions vs adding MobileLLM model support; not the same underlying change." + "reason": "Tokenizer method additions vs ExecuTorch export test; no shared bug or change." }, { "left": "pull_request:42668", "right": "pull_request:43823", "accept": false, - "reason": "Processor auto-loading robustness work vs new MobileLLM model files; broad shared area only." + "reason": "Processor auto-loading robustness changes vs MobileLLM model addition; unrelated." }, { "left": "pull_request:43251", "right": "pull_request:43823", "accept": false, - "reason": "Loss-function kwargs fix vs new model implementation; unrelated." + "reason": "Loss-function kwargs fix vs MobileLLM model addition; different concrete changes." }, { "left": "pull_request:43251", "right": "pull_request:44827", "accept": false, - "reason": "Cross-entropy kwargs fix vs Mistral4 test/model tweaks; no shared bug or code path." + "reason": "Loss-function kwargs fix vs Mistral4 test fixes; unrelated code paths." }, { "left": "pull_request:43424", "right": "pull_request:44827", "accept": false, - "reason": "Executorch dynamic-shape test vs Mistral4 test fixes; both are tests, but not the same issue." + "reason": "ExecuTorch dynamic-shape test vs Mistral4 test fixes; not the same underlying change." }, { "left": "pull_request:43424", "right": "pull_request:43823", "accept": false, - "reason": "Executorch export test vs MobileLLM model addition; unrelated." + "reason": "ExecuTorch export test vs MobileLLM model addition; unrelated." }, { "left": "pull_request:43823", "right": "pull_request:45655", "accept": false, - "reason": "Model addition vs config-resolution ordering fix; distinct changes." + "reason": "MobileLLM model addition vs config-resolution ordering fix; different targets." }, { "left": "pull_request:43823", "right": "pull_request:43995", "accept": false, - "reason": "MobileLLM implementation vs Falcon refactor for standardized output capture; different models and purposes." + "reason": "MobileLLM model addition vs Falcon output-interface refactor; unrelated." }, { "left": "pull_request:43995", "right": "pull_request:45213", "accept": false, - "reason": "Falcon model refactor vs model-creation skill/docs; not the same code-path problem." + "reason": "Falcon refactor vs model-creation skill draft; not the same bug or feature." }, { "left": "pull_request:43995", "right": "pull_request:44827", "accept": false, - "reason": "Falcon output-interface refactor vs Mistral4 test fixes; unrelated." + "reason": "Falcon output-interface refactor vs Mistral4 test fix; unrelated." }, { "left": "pull_request:44827", "right": "pull_request:46344", "accept": false, - "reason": "Mistral4 test fixes vs VitPose graph-break fix; different model subsystems and bugs." + "reason": "Mistral4 test fixes vs VitPose flip_back graph-break fix; different models and issues." }, { "left": "pull_request:44827", "right": "pull_request:45170", "accept": false, - "reason": "Mistral4 test fixes vs layrnorm/layernorm rename sweep; unrelated." + "reason": "Mistral4 test fixes vs layernorm renaming/mapping fixes; not the same change." }, { "left": "pull_request:44827", "right": "pull_request:46267", "accept": false, - "reason": "Mistral4 test fixes vs chat-response parsing overhaul; different feature areas." + "reason": "Mistral4 test fixes vs chat-response parsing updates; unrelated." }, { "left": "pull_request:44827", "right": "pull_request:46367", "accept": false, - "reason": "Mistral4 test fixes vs deepseek/xpu kernel integration; different model and kernel work." + "reason": "Mistral4 test fixes vs finegrained FP8 kernel rename/update; unrelated." }, { "left": "pull_request:44827", "right": "pull_request:45213", "accept": false, - "reason": "Mistral4 test fixes vs model-creation skill content; completely unrelated." + "reason": "Mistral4 test fixes vs model-creation skill draft; unrelated." }, { "left": "pull_request:43823", "right": "pull_request:45213", "accept": false, - "reason": "New MobileLLM model support vs repository skill/docs change; not the same artifact." + "reason": "MobileLLM model addition vs model-creation skill draft; unrelated artifacts." }, { "left": "pull_request:44827", "right": "pull_request:45655", "accept": false, - "reason": "Mistral4 test fixes vs config-resolution order fix; different problems." + "reason": "Mistral4 test fixes vs config-resolution ordering fix; unrelated." }, { "left": "pull_request:46048", "right": "pull_request:46367", "accept": false, - "reason": "TDT loss kernel integration vs finegrained-fp8/xpu kernel integration; same general integration area, but different kernels and code paths." + "reason": "TDT loss kernel integration vs finegrained FP8 kernel API rename; different kernels and purposes." }, { "left": "pull_request:44827", "right": "pull_request:46048", "accept": false, - "reason": "Mistral4 test fixes vs TDT loss kernel addition; unrelated." + "reason": "Mistral4 test fixes vs TDT loss kernel integration; unrelated." }, { "left": "pull_request:43251", "right": "pull_request:43254", "accept": true, - "reason": "Both PRs patch the same `fixed_cross_entropy` function to pass through supported `cross_entropy` kwargs for issue #43240, and could plausibly be merged into one change." + "reason": "Both PRs fix the same `fixed_cross_entropy` bug by forwarding supported kwargs (`weight`, `label_smoothing`) into `nn.functional.cross_entropy`, and both point at issue 43240." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet. The only clearly mergeable duplicate pair is the two `fixed_cross_entropy` PRs (#43251 and #43254), which both target issue #43240 and patch the same loss helper in compatible ways. The other pair verdicts are conservative and consistent with the listed filenames/diffs, and no unsupported broad claims stand out." + "feedback": "The summary is grounded in the packet, and the lone accepted pair is a conservative call: PRs 43251 and 43254 make the same narrow `fixed_cross_entropy` kwargs-forwarding change in the same file and share issue target 43240, so they look mergeable into one bugfix PR. The rejected soft pairs are also conservatively framed as unrelated or merely broadly similar." }, "evaluator_used": true, "retried": false, @@ -19022,7 +18933,7 @@ "pull_request:40546|pull_request:42781", "pull_request:42781|pull_request:46331" ], - "prepared_review_unit_hash": "ad88523de97426c411f1255c18c2f8fadd252b7f4ff61517a621acdd53719b03", + "prepared_review_unit_hash": "506819152e21cdc077232213ab320c8eba4f72998d0c73feb12eca92731b379a", "component_budget": { "node_count": 3, "item_count": 3, @@ -19044,24 +18955,24 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster contains three distinct open PRs for speech-model additions: a VibeVoice implementation, a VibeVoice Realtime extension, and a Parakeet-RNNT addition. They overlap in auto-mapping/doc plumbing, but they are not the same underlying change.", - "confidence": 0.92, + "summary": "These are three distinct feature PRs: an initial VibeVoice integration, a later VibeVoice Realtime extension, and an unrelated Parakeet-RNNT integration. They share some auto-registry and speech-model scaffolding, but they do not look like the same concrete change.", + "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #40546 is the cleanest canonical representative: it is the original VibeVoice implementation, has substantial review activity, and is more central than the later draft realtime extension.", + "canonical_pr_reason": "PR 40546 is the clearest canonical PR for the VibeVoice cluster because it is the original, broad VibeVoice implementation and docs addition; the others are separate follow-on or unrelated model integrations rather than the same change.", "best_issue_reason": null, - "best_pr_reason": "PR #40546 is the best single PR for the cluster because it is the core VibeVoice add-on; #42781 is a different realtime feature set, and #46331 is for a different model family entirely.", + "best_pr_reason": "PR 40546 best represents the underlying VibeVoice feature introduction. PR 42781 is a later realtime/acoustic-tokenizer extension, and PR 46331 is for a different model family (Parakeet-RNNT).", "soft_edge_verdicts": [ { "left": "pull_request:40546", "right": "pull_request:42781", "accept": false, - "reason": "Both are VibeVoice-related and share doc/auto-registration files, but #42781 adds VibeVoice Realtime with extra acoustic-tokenizer and generation changes, so it is a separate feature rather than the same concrete fix." + "reason": "Both are about VibeVoice, but 42781 is a distinct follow-up adding realtime and acoustic-tokenizer support plus additional generation/runtime changes. It is not the same concrete fix/feature as the original VibeVoice implementation and would not plausibly be merged as a single duplicate PR." }, { "left": "pull_request:42781", "right": "pull_request:46331", "accept": false, - "reason": "These target different model families (VibeVoice Realtime vs Parakeet-RNNT). Shared auto-mapping files are broad plumbing overlap, not evidence of the same underlying change." + "reason": "These are different model integrations: VibeVoice Realtime versus Parakeet-RNNT. Shared auto-mapping edits are generic registry updates, not evidence of the same underlying bug or change." } ] }, @@ -19091,7 +19002,7 @@ "soft_pairs": [ "pull_request:40755|pull_request:41224" ], - "prepared_review_unit_hash": "12ca58dec53934a4d49a2c62113833ca2dc03c6bc3a753c2f0576a43c38991ce", + "prepared_review_unit_hash": "313538dd2696a58deaa39e47334fe6003ca3e7adab0fa25912113659a43c82b9", "component_budget": { "node_count": 2, "item_count": 2, @@ -19113,18 +19024,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two pull requests are unrelated: one adds TimesFM covariate forecasting support, while the other adds DINOv3 image classification support. They touch different models, code paths, and docs, so they should not be clustered as duplicates.", + "summary": "These two pull requests are unrelated: one adds covariate forecasting support for TimesFM, while the other adds DINOv3 image-classification support. They do not describe the same bug, change, or code path.", "confidence": 0.99, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No canonical PR: the PRs target different model families and implement unrelated features.", "best_issue_reason": null, - "best_pr_reason": null, + "best_pr_reason": "Neither PR is a good global representative for the cluster because there is no duplicate relationship to collapse.", "soft_edge_verdicts": [ { "left": "pull_request:40755", "right": "pull_request:41224", "accept": false, - "reason": "Different underlying changes for different model families: TimesFM forecasting with covariates vs DINOv3 image classification support. No shared bug, feature, or mergeable code-path overlap beyond generic docs/tests/model plumbing." + "reason": "Different subsystems and features: TimesFM covariate forecasting vs. DINOv3 image classification. They do not fix or implement the same underlying change." } ] }, @@ -19219,12 +19130,12 @@ "pull_request:41797|pull_request:45186", "pull_request:45186|pull_request:45630" ], - "prepared_review_unit_hash": "45fb9d56dd0cee1594598486fd6dd977b9667ade073c9aabed567d7e5691ee23", + "prepared_review_unit_hash": "1758fc83c7883da502f043db78a9be497b6b6f78c94810db2edcb33ca13c32c0", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 2, - "serialized_chars": 10435, + "serialized_chars": 10433, "estimated_input_tokens": 2609, "estimated_eval_tokens": 5474 }, @@ -19232,7 +19143,7 @@ "node_count": 3, "item_count": 3, "soft_pair_count": 2, - "serialized_chars": 10435, + "serialized_chars": 10433, "estimated_input_tokens": 2609, "estimated_eval_tokens": 5474 }, @@ -19241,24 +19152,24 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Three open PRs add unrelated new models (DeepSeek OCR, Isaac, Kimi2-6). They overlap on onboarding boilerplate, but not on the same underlying change or code path.", + "summary": "Three unrelated model-addition PRs were grouped by generic Hugging Face model-registration/doc changes, but they target different models and codepaths, so they are not duplicates.", "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: these are distinct model additions for different model families and implementation paths, not duplicates.", + "canonical_pr_reason": "No canonical PR: each PR introduces a different model package and docs (DeepSeek OCR, Isaac, Kimi2-6), and the shared auto-registration/doc edits are boilerplate rather than the same underlying change.", "best_issue_reason": null, - "best_pr_reason": "No single best PR for deduping this cluster; the shared files are mostly registration/docs boilerplate, while the substantive model code differs across all three PRs.", + "best_pr_reason": "pull_request:45630 is the strongest standalone candidate because it has the most discussion/review activity and the most complete model integration surface, but it is still not merge-equivalent to the others.", "soft_edge_verdicts": [ { "left": "pull_request:41797", "right": "pull_request:45186", "accept": false, - "reason": "Different model additions (DeepSeek OCR vs Isaac). Shared auto-mapping/docs boilerplate is not enough to indicate the same underlying change." + "reason": "Different model additions (DeepSeek OCR vs Isaac) with separate model directories and docs; shared registration updates are generic boilerplate, not the same fix/change." }, { "left": "pull_request:45186", "right": "pull_request:45630", "accept": false, - "reason": "Different model families and code paths (Isaac vs Kimi2-6). They share onboarding scaffolding, but not a concrete fix or mergeable duplicate PR." + "reason": "Different model additions (Isaac vs Kimi2-6); despite overlapping auto-mapping/doc files, they introduce distinct model implementations and should not be merged as one PR." } ] }, @@ -19288,7 +19199,7 @@ "soft_pairs": [ "pull_request:42975|pull_request:42976" ], - "prepared_review_unit_hash": "1f2ddb817b71bde1ce4fd72d2f6dfd96701b01e18f923bf17d8218858bb6677a", + "prepared_review_unit_hash": "ed8da67cb21ef2260ba97a52f108240dea1b860e1414c41db26da9f797c6830f", "component_budget": { "node_count": 2, "item_count": 2, @@ -19310,18 +19221,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are two separate GitHub Actions maintenance PRs: one upgrades general workflow actions for Node 24 compatibility, the other bumps docker/build-push-action to v6 in docker-related workflows. They overlap only at the workflow-infrastructure level, not the same concrete change.", - "confidence": 0.97, + "summary": "Two separate GitHub Actions maintenance PRs: one bumps core workflow actions for Node 24 compatibility, the other updates Docker build actions and a few related workflow actions to newer versions. They overlap in files but address different action families and different upgrade goals.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: 42975 and 42976 touch different action families and solve different upgrade tasks, so they should not be deduplicated.", + "canonical_pr_reason": "PR 42975 is the more clearly scoped representative of the cluster because its title and diff directly target a specific compatibility issue across many workflows; PR 42976 is a separate generic action-version refresh focused on Docker-related actions.", "best_issue_reason": null, - "best_pr_reason": "PR 42975 is the broader workflow-actions upgrade and is the more representative umbrella change, but it is still not a duplicate of 42976.", + "best_pr_reason": "PR 42975 is the best representative PR for this cluster because it has the clearest, most concrete problem statement and broad workflow impact, while still being distinct from the Docker action upgrades in PR 42976.", "soft_edge_verdicts": [ { "left": "pull_request:42975", "right": "pull_request:42976", "accept": false, - "reason": "Different upgrades and different action ecosystems: Node 24 compatibility bumps core workflow actions, while the other only updates docker/build-push-action to v6. They are not the same underlying PR." + "reason": "Do not merge as duplicates: PR 42975 upgrades checkout/cache/upload-artifact for Node 24 compatibility, while PR 42976 upgrades docker/build-push-action and other workflow actions to latest versions. They overlap only in general CI/workflow maintenance, not the same underlying code-path change." } ] }, @@ -19363,94 +19274,94 @@ "pull_request:43451|pull_request:46411", "pull_request:45490|pull_request:46411" ], - "prepared_review_unit_hash": "332ef72d80aab67c606a602127a60e24a1da61e5466869ae28c5c1c41baf4388", + "prepared_review_unit_hash": "e16c1d768cfbdd61b5dc731a854a114d5deec2e2fed90db57ab261baacba004e", "component_budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 10, - "serialized_chars": 17822, - "estimated_input_tokens": 4456, - "estimated_eval_tokens": 9168 + "serialized_chars": 17815, + "estimated_input_tokens": 4454, + "estimated_eval_tokens": 9164 }, "budget": { "node_count": 5, "item_count": 5, "soft_pair_count": 10, - "serialized_chars": 17822, - "estimated_input_tokens": 4456, - "estimated_eval_tokens": 9168 + "serialized_chars": 17815, + "estimated_input_tokens": 4454, + "estimated_eval_tokens": 9164 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "All five PRs are separate model-addition efforts for different model families (ViT NEPA, Molmo, Molmo2, CTSM, MiniMax M2.7). They share a common scaffold of docs/auto-registration changes, but they do not look like duplicates of the same underlying bug or feature.", + "summary": "All items are separate model-addition PRs for different architectures (ViT NEPA, Molmo, Molmo2, CTSM, MiniMax M2.7). The overlap is only at the template level of adding model docs/registry/tests, not the same underlying bug or change.", "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical PR stands out for deduplication because each PR adds a different model/package and touches different model-specific code paths.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "No PR is a good global canonical duplicate target; the PRs are unrelated model integrations rather than alternative fixes for one change.", + "best_pr_reason": "PR 43451 is the most substantial and actively reviewed model-addition PR, but it still represents a distinct Molmo2 integration rather than a duplicate of any other PR here.", "soft_edge_verdicts": [ { "left": "pull_request:42978", "right": "pull_request:43448", "accept": false, - "reason": "Different models and code paths: ViT NEPA vs Molmo. Shared docs/auto-registration patterns are not enough to call them the same change." + "reason": "Different models and code paths: ViT NEPA vs Molmo. Shared auto/docs plumbing is generic model-integration work, not the same change." }, { "left": "pull_request:42978", "right": "pull_request:43451", "accept": false, - "reason": "Different model families: ViT NEPA vs Molmo2. They add separate implementations, not the same underlying feature." + "reason": "Different model families: ViT NEPA vs Molmo2. Similar file patterns, but no shared concrete bug or feature." }, { "left": "pull_request:42978", "right": "pull_request:45490", "accept": false, - "reason": "ViT NEPA and CTSM are unrelated model additions, despite both updating docs and auto mappings." + "reason": "ViT NEPA and CTSM are unrelated architectures (vision-language vs time-series); only the repo integration pattern overlaps." }, { "left": "pull_request:42978", "right": "pull_request:46411", "accept": false, - "reason": "ViT NEPA and MiniMax M2.7 are clearly different model integrations with no shared concrete bug/change." + "reason": "ViT NEPA model addition is unrelated to the MiniMax demo/code drop; they do not fix or add the same thing." }, { "left": "pull_request:43448", "right": "pull_request:43451", "accept": false, - "reason": "Molmo and Molmo2 are related by name, but they are distinct model releases with different files and capabilities, so not the same change." + "reason": "Molmo and Molmo2 are related by name, but they are distinct model ports with different files and capabilities; not the same underlying change." }, { "left": "pull_request:43448", "right": "pull_request:45490", "accept": false, - "reason": "Molmo and CTSM are different model additions; overlap is only in generic Transformers plumbing." + "reason": "Molmo vs CTSM are completely different domains and implementations; shared top-level registry/doc edits are incidental." }, { "left": "pull_request:43448", "right": "pull_request:46411", "accept": false, - "reason": "Molmo and MiniMax M2.7 target different model families and code paths." + "reason": "Molmo model integration and MiniMax demo/code are unrelated; no common code-path problem or mergeable duplicate change." }, { "left": "pull_request:43451", "right": "pull_request:45490", "accept": false, - "reason": "Molmo2 and CTSM are separate model integrations, not one shared fix or feature." + "reason": "Molmo2 and CTSM address different model stacks; overlap is only in generic auto-registration/docs." }, { "left": "pull_request:43451", "right": "pull_request:46411", "accept": false, - "reason": "Molmo2 and MiniMax M2.7 are unrelated model additions." + "reason": "Molmo2 integration and MiniMax demo code are unrelated changes, not the same feature or bug fix." }, { "left": "pull_request:45490", "right": "pull_request:46411", "accept": false, - "reason": "CTSM and MiniMax M2.7 are distinct model-specific PRs with no evidence of a common underlying issue." + "reason": "CTSM model integration and a MiniMax demo file have no shared concrete code-path issue; they are separate additions." } ] }, @@ -19487,70 +19398,70 @@ "pull_request:43665|pull_request:46380", "pull_request:44815|pull_request:46380" ], - "prepared_review_unit_hash": "192b572cc904d08ee5acfb45771970515bf3889215407ad45f7c07a0f9333565", + "prepared_review_unit_hash": "5430e326dbfd546fd09438bef8ab2986e6a2900789ee99abff1b7f9229008ebf", "component_budget": { "node_count": 4, "item_count": 4, "soft_pair_count": 6, - "serialized_chars": 11145, - "estimated_input_tokens": 2787, - "estimated_eval_tokens": 5830 + "serialized_chars": 11142, + "estimated_input_tokens": 2786, + "estimated_eval_tokens": 5828 }, "budget": { "node_count": 4, "item_count": 4, "soft_pair_count": 6, - "serialized_chars": 11145, - "estimated_input_tokens": 2787, - "estimated_eval_tokens": 5830 + "serialized_chars": 11142, + "estimated_input_tokens": 2786, + "estimated_eval_tokens": 5828 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These four PRs are not duplicates of one another: they target unrelated code paths (Whisper decoding, test-only edits, FP8 dequantization, and generation config validation). No soft-edge pair looks like the same underlying fix or change.", - "confidence": 0.96, + "summary": "These are four mostly unrelated pull requests that happen to have similar \u201cfix\u201d titles and some superficial overlap in touched areas/tests. They do not look like duplicate implementations of the same bug or change.", + "confidence": 0.89, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #46380 is the strongest representative of the cluster because it contains a concrete runtime bug fix in core generation config validation, with broader but coherent follow-up test coverage.", + "canonical_pr_reason": "PR #46380 is the strongest canonical representative: it fixes a concrete runtime bug in generation config validation, has a clear code-path change, and includes targeted follow-up tests.", "best_issue_reason": null, - "best_pr_reason": "PR #46380 is the best standalone PR here: it fixes a clear validation/type-conversion bug in an actual code path rather than only adjusting tests or touching an unrelated subsystem.", + "best_pr_reason": "PR #46380 is the best overall PR in this set because it addresses a specific correctness issue in core generation behavior rather than test-only edits or an unrelated subsystem change.", "soft_edge_verdicts": [ { "left": "pull_request:43291", "right": "pull_request:43665", "accept": false, - "reason": "Whisper tokenizer decode logic and test-only CLIP/ViT edits are unrelated; they do not look like the same bug or change." + "reason": "Both are test-oriented and share a generic \u201cfix\u201d theme, but they target different models/code paths: Whisper tokenizer decoding vs. CLIP/ViT test edits. No same underlying bug." }, { "left": "pull_request:43291", "right": "pull_request:44815", "accept": false, - "reason": "Whisper decoding and FP8 dequantization are different subsystems with different failure modes and code paths." + "reason": "Whisper decoding tests/code and FP8 dequantization/model loading are unrelated problems in different subsystems." }, { "left": "pull_request:43291", "right": "pull_request:46380", "accept": false, - "reason": "Both touch a Whisper file, but one fixes timestamp decoding while the other fixes generation watermarking config validation; not the same underlying bug." + "reason": "Whisper tokenizer decoding changes and generation watermarking validation are distinct code paths and bugs; only a broad \u201cfix\u201d label overlaps." }, { "left": "pull_request:43665", "right": "pull_request:44815", "accept": false, - "reason": "A test-only PR with placeholder asserts is not the same change as an FP8 dequantization fix." + "reason": "One PR is test assertions in CLIP/ViT, the other is FP8 dequant/model-loading logic; they are not the same change or bug." }, { "left": "pull_request:43665", "right": "pull_request:46380", "accept": false, - "reason": "Test placeholder edits across CLIP/ViT are unrelated to generation config validation and watermarking handling." + "reason": "CLIP/ViT test-only edits do not match the generation-config runtime fix in PR #46380." }, { "left": "pull_request:44815", "right": "pull_request:46380", "accept": false, - "reason": "FP8 dequant/padding changes and generation config validation are different concrete problems; shared broad ML subsystem is not enough." + "reason": "FP8 dequantization/padding and generation watermarking validation are different concrete bugs in different areas; they could not plausibly be merged as one fix." } ] }, @@ -19580,12 +19491,12 @@ "soft_pairs": [ "pull_request:43488|pull_request:43532" ], - "prepared_review_unit_hash": "c01b27e2d94d5d4b2b724689706b57ff6c1179903409e61d8392f7bb75d6a5d7", + "prepared_review_unit_hash": "68d63d27bc514066a7aa832f6bba3ec69f9236e2e33d4abf4b07a6dd2e598090", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4562, + "serialized_chars": 4563, "estimated_input_tokens": 1141, "estimated_eval_tokens": 2538 }, @@ -19593,7 +19504,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4562, + "serialized_chars": 4563, "estimated_input_tokens": 1141, "estimated_eval_tokens": 2538 }, @@ -19602,18 +19513,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two PRs are not duplicates: one is a small repo-bot/setup/version sanity check, while the other rewires CI/comment workflows and testing utilities. They only overlap on a shared Vit test file, which is not enough to indicate the same underlying change.", - "confidence": 0.97, + "summary": "These two pull requests are not duplicates: 43488 is about setup/versioning and a test dependency change, while 43532 is about workflow/CI behavior and related utility changes. They only overlap on a test file and do not appear to fix the same underlying code-path problem.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #43532 is the better representative of the cluster because it contains the broader, concrete code/workflow change set; PR #43488 is a narrow bot-check/style test PR and does not share the same change intent.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR #43532 is the most substantial and generalizable artifact here, touching CI and test infrastructure across multiple files. PR #43488 is too specialized and unrelated to serve as the cluster representative.", + "best_pr_reason": "PR 43532 is the more substantial and self-contained change set, but it still addresses a different area than 43488 and should not be treated as the same fix.", "soft_edge_verdicts": [ { "left": "pull_request:43488", "right": "pull_request:43532", "accept": false, - "reason": "Different goals and code paths: #43488 changes packaging/versioning for a bot/check, while #43532 updates CI/workflow and testing utility behavior. Shared Vit test file alone is not enough to merge them as one PR." + "reason": "Shared vit test files are not enough here: 43488 changes packaging/version metadata and adds pytest-random-order, while 43532 rewrites CI/workflow behavior and notification utilities. Different code paths and different goals." } ] }, @@ -19643,40 +19554,40 @@ "soft_pairs": [ "pull_request:43612|pull_request:43613" ], - "prepared_review_unit_hash": "279d5dc56023b89d5195bc4accce091ce0af63f160fd24985be800c27bf972d9", + "prepared_review_unit_hash": "0775009ddf58f9203a2a77d244d3e358ef8f4e51bc1d8f89cccbbc19d0c57c5c", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 6083, - "estimated_input_tokens": 1521, - "estimated_eval_tokens": 3298 + "serialized_chars": 6080, + "estimated_input_tokens": 1520, + "estimated_eval_tokens": 3296 }, "budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 6083, - "estimated_input_tokens": 1521, - "estimated_eval_tokens": 3298 + "serialized_chars": 6080, + "estimated_input_tokens": 1520, + "estimated_eval_tokens": 3296 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two separate PRs add different new segmentation pipelines with similar scaffolding, but they target different tasks/models and are not duplicates.", - "confidence": 0.98, + "summary": "Two open PRs add similarly named but distinct segmentation pipelines: one for promptable concept segmentation and one for promptable visual segmentation. They share some scaffolding, but they target different features/models and should not be treated as duplicates.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #43613 is the broader/more representative pipeline addition in this cluster: it adds Promptable Visual Segmentation and wires in support across multiple related model docs/tests, making it the stronger canonical example of the shared pipeline-creation pattern.", + "canonical_pr_reason": "PR 43613 is the broader and more complete addition, covering the visual segmentation pipeline with more model integrations and test coverage.", "best_issue_reason": null, - "best_pr_reason": "PR #43613 is the best representative of the cluster because it is a complete, self-contained pipeline feature addition with substantial integration work; however, it should not subsume PR #43612 since that PR adds a different pipeline.", + "best_pr_reason": "PR 43613 is the best representative of the cluster because it is the more substantial, end-to-end pipeline implementation.", "soft_edge_verdicts": [ { "left": "pull_request:43612", "right": "pull_request:43613", "accept": false, - "reason": "Both PRs follow the same template, but they implement different concrete features: Promptable Concept Segmentation vs Promptable Visual Segmentation. They touch different task pages, pipeline modules, and model/test sets, so they are not the same underlying change and should not be merged as duplicates." + "reason": "Reject: these are different features, not the same underlying change. One adds Promptable Concept Segmentation, the other Promptable Visual Segmentation, with different docs, pipeline modules, and model/test coverage. Shared boilerplate does not make them mergeable into one PR." } ] }, @@ -19706,7 +19617,7 @@ "soft_pairs": [ "pull_request:43651|pull_request:43663" ], - "prepared_review_unit_hash": "3965e6b52e223fce6934c3136e678633f3f28406edd910dde0b4dfc2a1c10641", + "prepared_review_unit_hash": "de73c04e8345a45e7c0285b0fb51338c270535040863c5047fc1792ce0ed0dd3", "component_budget": { "node_count": 2, "item_count": 2, @@ -19728,18 +19639,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two PRs both touch `trainer.py`, but they fix different trainer extension points: one changes gradient-accumulation loss scaling, the other refactors signature-column filtering for custom trainers/PEFT. They are not duplicates and should not be merged together.", + "summary": "These two PRs both touch `trainer.py` and add extensibility hooks, but they address different code paths: gradient-accumulation loss scaling vs. signature-based column filtering. They are related in theme, not duplicates.", "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: the changes target different code paths and different customization hooks in `Trainer`, so they are not the same underlying fix.", + "canonical_pr_reason": "No single canonical PR: 43651 and 43663 change different Trainer behaviors and would not be merged as one duplicate fix.", "best_issue_reason": null, - "best_pr_reason": "PR #43651 is the more self-contained behavior change and directly addresses a concrete training-step loss-scaling issue; PR #43663 is a separate refactor/extension point for column filtering.", + "best_pr_reason": "PR 43651 is the better representative of the cluster because it makes a concrete training-step behavior change with a clear API hook; 43663 is a separate hook for input-column filtering.", "soft_edge_verdicts": [ { "left": "pull_request:43651", "right": "pull_request:43663", "accept": false, - "reason": "Different problems and merge targets: #43651 changes loss normalization during training, while #43663 adds a signature-column helper for data filtering. They are not the same bug or change." + "reason": "Reject: both are Trainer extensibility changes, but they fix different problems in different code paths. 43651 changes gradient-accumulation loss normalization; 43663 changes how signature columns are derived for filtering." } ] }, @@ -19769,7 +19680,7 @@ "soft_pairs": [ "pull_request:43656|pull_request:43842" ], - "prepared_review_unit_hash": "524ef028715d04e10b5efcfa22239f8dbb320321e204f10a1ecb11c7e2bd7e60", + "prepared_review_unit_hash": "b9723602bddeacfebe346049128f67c1783a4fe8a1c5c276403557d66b5e418e", "component_budget": { "node_count": 2, "item_count": 2, @@ -19791,28 +19702,25 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs target the same import-time NameError in `src/transformers/cli/serve.py` around `TypeAdapter`/type annotations; PR 43842 is the cleaner, more targeted fix, while PR 43656 is a broader annotation-runtime cleanup that appears to address the same bug.", - "confidence": 0.84, + "summary": "These PRs touch the same file but address different TypeAdapter/runtime typing problems in `src/transformers/cli/serve.py`. They should stay separate; PR 43842 is the more focused standalone fix for the missing-pydantic case.", + "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 43842 is the best canonical PR because it directly fixes the reported `TypeAdapter` NameError with the smallest, most specific change (guarding the import under `TYPE_CHECKING` and quoting the annotation).", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR 43842 is the strongest representative of the cluster since it cleanly addresses the concrete CLI runtime error without mixing in broader typing refactors.", + "best_pr_reason": "43842 is the cleaner, more targeted fix: it explicitly guards the `TypeAdapter` import behind `TYPE_CHECKING` and quotes the annotation, matching the stated missing-pydantic NameError.", "soft_edge_verdicts": [ { "left": "pull_request:43656", "right": "pull_request:43842", - "accept": true, - "reason": "Same underlying code-path problem in `cli/serve.py`: runtime evaluation of annotations causing a `TypeAdapter`-related NameError. The two PRs are plausible alternative fixes for the same bug and could have been merged into one change." + "accept": false, + "reason": "Reject: both touch `serve.py`, but they fix different concrete issues. 43656 is a broader annotation/future-import cleanup, while 43842 specifically fixes `TypeAdapter` when pydantic is absent." } ] }, - "evaluator_result": { - "accept": true, - "feedback": "The summary stays within the packet: both PRs touch `src/transformers/cli/serve.py`, both are about annotation/runtime `TypeAdapter` NameError fixes, and PR 43842 is the more targeted change. The soft-edge acceptance is conservative enough because the two diffs look compatible and plausibly mergeable into one bugfix PR." - }, - "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 }, @@ -19997,12 +19905,12 @@ "pull_request:44044|pull_request:44129", "pull_request:44044|pull_request:44154" ], - "prepared_review_unit_hash": "b6ca91714e984d79dc5a72a2ecd33ecdb505f9e987d29483288b6738db6d1771", + "prepared_review_unit_hash": "351319817012d0d54a2c857aebe183700fdb17fa867f17c6cf597052a9dbc88b", "component_budget": { "node_count": 28, "item_count": 28, "soft_pair_count": 78, - "serialized_chars": 84894, + "serialized_chars": 84896, "estimated_input_tokens": 21224, "estimated_eval_tokens": 42704 }, @@ -20010,369 +19918,321 @@ "node_count": 27, "item_count": 27, "soft_pair_count": 72, - "serialized_chars": 75088, - "estimated_input_tokens": 18772, - "estimated_eval_tokens": 37800 + "serialized_chars": 75092, + "estimated_input_tokens": 18773, + "estimated_eval_tokens": 37802 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "Mostly a model-by-model output-tracing refactor sweep, not a single duplicate bug. The only clear duplicate groups are the GPT-Neo pair and the GPT-J/CodeGen family; the rest are parallel refactors in different subsystems.", - "confidence": 0.91, + "summary": "This is mostly a theme cluster of standardized output-tracing refactors across many different models. Only the GPT-J/CodeGen and GPT-Neo pairs look like true duplicates; the rest are separate model-specific changes that merely share the same refactor pattern.", + "confidence": 0.85, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 44066 is the strongest representative: it explicitly targets the shared output-tracing refactor and covers the central GPT-J/CodeGen code path that later follow-up PRs overlap with.", + "canonical_pr_reason": "PR 44722 is the best anchor for the only real duplicate subcluster: it covers the same GPT-J/CodeGen output-tracing refactor as the overlapping GPT-J PRs, with the clearest and broadest concrete code-path overlap.", "best_issue_reason": null, - "best_pr_reason": "PR 44066 is the best single PR to represent this cluster because it is the broadest concrete implementation in the repeated GPT-J/CodeGen tracing-refactor family and has the clearest issue linkage.", + "best_pr_reason": "PR 44722 is the strongest single representative in this cluster because it reflects the shared GPT-J/CodeGen decorator migration most directly and appears to subsume the closely related GPT-J variants.", "soft_edge_verdicts": [ { "left": "pull_request:43996", "right": "pull_request:44085", "accept": false, - "reason": "Different model/code path; same issue theme is not enough to make these duplicates." + "reason": "Different model families; same refactor theme only." }, { "left": "pull_request:43996", "right": "pull_request:44044", "accept": false, - "reason": "Both are output-tracing refactors, but they touch unrelated model implementations." + "reason": "CVT/FNet output tracing vs DeBERTa output tracing are unrelated code paths." }, { "left": "pull_request:44066", "right": "pull_request:44085", "accept": true, - "reason": "Same GPT-J output-tracing refactor family, with overlapping gptj/modeling_gptj.py changes." + "reason": "Both target the GPT-J output-tracing refactor in the same file family; they look like overlapping versions of the same change." }, { "left": "pull_request:44007", "right": "pull_request:44072", "accept": false, - "reason": "ResNet and EfficientNet are separate code paths; these are parallel refactors, not the same fix." + "reason": "RegNet/ResNet refactors and EfficientNet refactors are separate model changes." }, { "left": "pull_request:44072", "right": "pull_request:44722", "accept": false, - "reason": "Different models and different tracing code paths; shared theme alone is insufficient." + "reason": "EfficientNet and GPT-J are unrelated model code paths." }, { "left": "pull_request:44066", "right": "pull_request:44072", "accept": false, - "reason": "GPT-J/CodeGen vs EfficientNet are unrelated subsystems." + "reason": "GPT-J/CodeGen output tracing is unrelated to EfficientNet output tracing." }, { "left": "pull_request:44013", "right": "pull_request:44044", "accept": false, - "reason": "MobileNetV2 and DeBERTa are unrelated implementations." + "reason": "MobileNetV2 and DeBERTa are unrelated models; only the refactor pattern matches." }, { "left": "pull_request:44066", "right": "pull_request:44086", "accept": false, - "reason": "Different model families and different concrete code paths." + "reason": "GPT-J/CodeGen and MGP-STR are different code paths." }, { "left": "pull_request:44066", "right": "pull_request:44071", "accept": false, - "reason": "GPT-J/CodeGen output tracing and MPT output tracing are separate model-specific changes." + "reason": "GPT-J/CodeGen and MPT are separate model implementations." }, { "left": "pull_request:44018", "right": "pull_request:44068", "accept": true, - "reason": "Both are GPT-Neo output-tracing refactors in the same code path." + "reason": "Both are GPT-Neo output-tracing refactors against the same model/file path, so they appear to be the same underlying change." }, { "left": "pull_request:44018", "right": "pull_request:44066", "accept": false, - "reason": "GPT-Neo is a different model/code path from GPT-J/CodeGen." + "reason": "GPT-Neo and GPT-J are different models; same umbrella refactor is not enough." }, { "left": "pull_request:44066", "right": "pull_request:44068", "accept": false, - "reason": "GPT-J/CodeGen and GPT-Neo are distinct implementations." + "reason": "GPT-J/CodeGen vs GPT-Neo are distinct model code paths." }, { "left": "pull_request:44019", "right": "pull_request:44722", "accept": false, - "reason": "ResNet and GPT-J/CodeGen are unrelated subsystems." + "reason": "ResNet output tracing and GPT-J/CodeGen output tracing are unrelated changes." }, { "left": "pull_request:44086", "right": "pull_request:44722", "accept": false, - "reason": "Different model code paths; both mention output tracing but do not fix the same bug." + "reason": "MGP-STR and GPT-J touch different model implementations." }, { "left": "pull_request:44019", "right": "pull_request:44085", "accept": false, - "reason": "ResNet and GPT-J are different concrete implementations." + "reason": "ResNet and GPT-J are different concrete model paths." }, { "left": "pull_request:44018", "right": "pull_request:44019", "accept": false, - "reason": "GPT-Neo and ResNet are unrelated; shared issue target does not imply duplication." + "reason": "GPT-Neo and ResNet are different models; not a duplicate change." }, { "left": "pull_request:44019", "right": "pull_request:44071", "accept": false, - "reason": "Different model families and no shared code path." + "reason": "ResNet and MPT are unrelated implementations." }, { "left": "pull_request:44068", "right": "pull_request:44116", "accept": false, - "reason": "GPT-Neo and Flaubert/XLM are unrelated concrete changes." + "reason": "GPT-Neo and Flaubert are separate model-specific refactors." }, { "left": "pull_request:44018", "right": "pull_request:44116", "accept": false, - "reason": "Different model families and different output paths." + "reason": "GPT-Neo and Flaubert do not share the same code path." }, { "left": "pull_request:44068", "right": "pull_request:44722", "accept": false, - "reason": "GPT-Neo and GPT-J/CodeGen are separate model implementations." + "reason": "GPT-Neo output tracing is unrelated to GPT-J/CodeGen output tracing." }, { "left": "pull_request:44071", "right": "pull_request:44722", "accept": false, - "reason": "MPT and GPT-J/CodeGen are unrelated code paths." + "reason": "MPT and GPT-J/CodeGen are different model implementations." }, { - "left": "pull_request:44059", - "right": "pull_request:44086", + "left": "pull_request:44018", + "right": "pull_request:44085", "accept": false, - "reason": "GPT2 and MGP-STR are different model families." + "reason": "GPT-Neo and GPT-J are different models." }, { - "left": "pull_request:44085", + "left": "pull_request:44018", "right": "pull_request:44086", "accept": false, - "reason": "GPT-J and MGP-STR are unrelated implementations." + "reason": "GPT-Neo and MGP-STR are unrelated code paths." }, { "left": "pull_request:44098", "right": "pull_request:44101", "accept": false, - "reason": "ViLT and Flaubert/XLM are different model-specific output changes." + "reason": "ViLT and Flaubert are separate model-specific changes." }, { "left": "pull_request:44059", - "right": "pull_request:44722", + "right": "pull_request:44086", "accept": false, - "reason": "GPT2 and GPT-J/CodeGen are distinct code paths." + "reason": "GPT-2 and MGP-STR are unrelated concrete implementations." }, { - "left": "pull_request:44017", - "right": "pull_request:44722", + "left": "pull_request:44085", + "right": "pull_request:44086", "accept": false, - "reason": "SegFormer and GPT-J/CodeGen are unrelated subsystems." + "reason": "GPT-J and MGP-STR are different models." }, { - "left": "pull_request:44017", + "left": "pull_request:44098", "right": "pull_request:44161", "accept": false, - "reason": "SegFormer and LongT5 are separate model implementations." + "reason": "ViLT and LongT5 are unrelated model paths." }, { - "left": "pull_request:44017", - "right": "pull_request:44073", + "left": "pull_request:44073", + "right": "pull_request:44101", "accept": false, - "reason": "SegFormer and VisualBert are unrelated code paths." + "reason": "VisualBert and Flaubert are different models." }, { - "left": "pull_request:44068", + "left": "pull_request:44059", "right": "pull_request:44071", "accept": false, - "reason": "GPT-Neo and MPT are different model families." + "reason": "GPT-2 and MPT are unrelated code paths." }, { "left": "pull_request:44071", - "right": "pull_request:44086", - "accept": false, - "reason": "MPT and MGP-STR are unrelated fixes." - }, - { - "left": "pull_request:43997", - "right": "pull_request:44073", + "right": "pull_request:44076", "accept": false, - "reason": "RegNet and VisualBert are different model implementations." + "reason": "MPT and ImageGPT are separate model refactors." }, { - "left": "pull_request:44056", - "right": "pull_request:44101", + "left": "pull_request:44074", + "right": "pull_request:44161", "accept": false, - "reason": "MPNet and Flaubert/XLM are unrelated code paths." + "reason": "TextNet and LongT5 are different model implementations." }, { - "left": "pull_request:44059", - "right": "pull_request:44068", + "left": "pull_request:44071", + "right": "pull_request:44129", "accept": false, - "reason": "GPT2 and GPT-Neo are different model-specific output tracing changes." + "reason": "MPT and SpeechT5 are unrelated concrete changes." }, { - "left": "pull_request:44101", - "right": "pull_request:44161", + "left": "pull_request:44010", + "right": "pull_request:44073", "accept": false, - "reason": "Flaubert/XLM and LongT5 are separate subsystems." + "reason": "SqueezeBert and VisualBert are different models." }, { - "left": "pull_request:44017", + "left": "pull_request:44010", "right": "pull_request:44018", "accept": false, - "reason": "SegFormer and GPT-Neo are unrelated models." + "reason": "SqueezeBert and GPT-Neo do not share the same code path." }, { "left": "pull_request:43997", "right": "pull_request:44056", "accept": false, - "reason": "RegNet and MPNet are separate model families." + "reason": "RegNet and MPNet are separate model-specific refactors." }, { "left": "pull_request:44010", "right": "pull_request:44071", "accept": false, - "reason": "SqueezeBert and MPT are unrelated concrete code paths." - }, - { - "left": "pull_request:44074", - "right": "pull_request:44101", - "accept": false, - "reason": "TextNet and Flaubert/XLM are different model implementations." + "reason": "SqueezeBert and MPT are unrelated implementations." }, { "left": "pull_request:43997", - "right": "pull_request:44018", - "accept": false, - "reason": "RegNet and GPT-Neo are unrelated changes." - }, - { - "left": "pull_request:44056", - "right": "pull_request:44161", + "right": "pull_request:44074", "accept": false, - "reason": "MPNet and LongT5 are different model-specific refactors." + "reason": "RegNet and TextNet are different models." }, { "left": "pull_request:43997", "right": "pull_request:44161", "accept": false, - "reason": "RegNet and LongT5 are unrelated code paths." - }, - { - "left": "pull_request:44074", - "right": "pull_request:44161", - "accept": false, - "reason": "TextNet and LongT5 are separate models." - }, - { - "left": "pull_request:44018", - "right": "pull_request:44161", - "accept": false, - "reason": "GPT-Neo and LongT5 are unrelated." - }, - { - "left": "pull_request:43997", - "right": "pull_request:44071", - "accept": false, - "reason": "RegNet and MPT are different model families." - }, - { - "left": "pull_request:44010", - "right": "pull_request:44073", - "accept": false, - "reason": "SqueezeBert and VisualBert are unrelated implementations." - }, - { - "left": "pull_request:43997", - "right": "pull_request:44074", - "accept": false, - "reason": "RegNet and TextNet are separate model-specific changes." + "reason": "RegNet and LongT5 are unrelated changes." }, { - "left": "pull_request:43996", + "left": "pull_request:44056", "right": "pull_request:44161", "accept": false, - "reason": "FNet and LongT5 are unrelated code paths." + "reason": "MPNet and LongT5 are different model paths." }, { "left": "pull_request:43996", "right": "pull_request:44017", "accept": false, - "reason": "FNet and SegFormer are different model implementations." + "reason": "CVT/FNet refactors and Segformer refactors are separate model-specific changes." }, { "left": "pull_request:44010", "right": "pull_request:44044", "accept": false, - "reason": "SqueezeBert and DeBERTa v2 are unrelated model code paths." + "reason": "SqueezeBert and DeBERTa are unrelated models." }, { "left": "pull_request:44044", "right": "pull_request:44074", "accept": false, - "reason": "DeBERTa v2 and TextNet are separate fixes." + "reason": "DeBERTa and TextNet are different code paths." }, { "left": "pull_request:44066", "right": "pull_request:44722", "accept": true, - "reason": "Same GPT-J/CodeGen output-tracing refactor, with overlapping model file changes." + "reason": "Same GPT-J/CodeGen output-tracing change; 44722 looks like the consolidated version of the overlapping refactor." }, { "left": "pull_request:44085", "right": "pull_request:44722", "accept": true, - "reason": "Both target the same GPT-J output-tracing code path and would plausibly merge into one PR." + "reason": "Both touch the same GPT-J output-tracing refactor in the same file area, so they are effectively duplicates." }, { "left": "pull_request:44085", "right": "pull_request:44129", "accept": false, - "reason": "GPT-J and SpeechT5 are unrelated model-specific changes." + "reason": "GPT-J and SpeechT5 are unrelated model changes." }, { "left": "pull_request:43996", "right": "pull_request:44007", "accept": false, - "reason": "FNet and ResNet are different concrete code paths." + "reason": "CVT/FNet and RegNet/ResNet are different model-specific refactors." }, { "left": "pull_request:43996", "right": "pull_request:44129", "accept": false, - "reason": "FNet and SpeechT5 are unrelated models." + "reason": "CVT/FNet and SpeechT5 are unrelated code paths." }, { "left": "pull_request:44044", "right": "pull_request:44129", "accept": false, - "reason": "DeBERTa v2 and SpeechT5 are separate model implementations." + "reason": "DeBERTa and SpeechT5 are different models." }, { "left": "pull_request:44044", "right": "pull_request:44154", "accept": false, - "reason": "DeBERTa v2 and ViTS are unrelated fixes." + "reason": "DeBERTa and VITS are separate model implementations." } ] }, "evaluator_result": { "accept": true, - "feedback": "The summary is grounded in the packet: the accepted soft edges stay within the same concrete GPT-Neo or GPT-J/CodeGen code paths, and the rejected edges correctly avoid conflating unrelated model implementations that only share the high-level output-tracing theme. The chosen representative PR (44066) is also defensible from the evidence because it is issue-linked and central to the GPT-J/CodeGen family." + "feedback": "Grounded and conservative overall. The accepted soft edges stay within the same model/file families with clear diff overlap (GPT-J/CodeGen and GPT-Neo), while cross-family pairs are appropriately rejected. The summary does not overstate broader duplication beyond what the packet supports." }, "evaluator_used": true, "retried": false, @@ -20410,12 +20270,12 @@ "pull_request:43997|pull_request:44010", "pull_request:44010|pull_request:44129" ], - "prepared_review_unit_hash": "4e9450537d6878aec8f5053c6bc2749a287e08f8ae519e3db67976210ab49ed1", + "prepared_review_unit_hash": "52d22dd3050bb234fe5a8981fa793614fa7b9e6244a6f815e5f50b8f63728e30", "component_budget": { "node_count": 28, "item_count": 28, "soft_pair_count": 78, - "serialized_chars": 84894, + "serialized_chars": 84896, "estimated_input_tokens": 21224, "estimated_eval_tokens": 42704 }, @@ -20423,7 +20283,7 @@ "node_count": 8, "item_count": 8, "soft_pair_count": 6, - "serialized_chars": 17127, + "serialized_chars": 17125, "estimated_input_tokens": 4282, "estimated_eval_tokens": 8820 }, @@ -20432,48 +20292,48 @@ "aggressively_trimmed": false, "split": true, "analyst_result": { - "summary": "This cluster is a set of model-specific refactors to standardized output tracing/capture_outputs across different architectures. They are related by theme, but they are not duplicates of the same underlying bug or change because each PR touches a different model code path.", - "confidence": 0.89, + "summary": "All items are model-specific PRs that apply the same output-tracing refactor pattern to different code paths, not duplicates of one another. The cluster is best represented by the SpeechT5 PR because it is the most concrete and has an explicit issue target.", + "confidence": 0.92, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 44129 is the strongest representative: it has an explicit issue target, a clear refactor title, and a more substantive diff showing the standardized output-capture pattern applied to SpeechT5.", + "canonical_pr_reason": "PR 44129 is the strongest representative: it has an explicit issue target, a clear standardized output-capture refactor, and enough context to describe the cluster\u2019s common theme.", "best_issue_reason": null, - "best_pr_reason": "PR 44129 is the best standalone PR in the set because it is the most clearly scoped and explicitly linked to an issue, with a concrete model-specific implementation change rather than only a generic decorator migration.", + "best_pr_reason": "44129 is the most suitable standalone representative because it is self-contained, clearly scoped to one model, and anchored to a linked issue, whereas the others are similar but equally model-specific refactors.", "soft_edge_verdicts": [ { "left": "pull_request:44024", "right": "pull_request:44076", "accept": false, - "reason": "Both are standardized-output refactors, but they affect different models (FocalNet vs ImageGPT) and different forward paths; this is a shared pattern, not the same change." + "reason": "Both are output-tracing refactors, but they touch different model implementations (FocalNet vs ImageGPT) and different code paths; not the same underlying change." }, { "left": "pull_request:44073", "right": "pull_request:44074", "accept": false, - "reason": "VisualBert and TextNet are separate model implementations with different output-tracing code paths, so these are not the same concrete fix." + "reason": "Same refactor pattern, but VisualBert and TextNet are separate model-specific implementations; these are parallel changes, not duplicate PRs." }, { "left": "pull_request:44056", "right": "pull_request:44073", "accept": false, - "reason": "MPNet and VisualBert are unrelated model-specific refactors; similar decorator usage does not make them duplicates." + "reason": "MPNet and VisualBert are distinct model code paths; the similarity is architectural, not a shared concrete bug or change." }, { "left": "pull_request:44056", "right": "pull_request:44074", "accept": false, - "reason": "These PRs touch different models and different tracing logic, so they are not mergeable as one duplicate change." + "reason": "Both are standardized-output refactors, but they are separate model migrations in different files with no shared concrete fix to merge into one PR." }, { "left": "pull_request:43997", "right": "pull_request:44010", "accept": false, - "reason": "RegNet and SqueezeBert are distinct architectures; the PRs share a migration theme but not the same underlying bug or code path." + "reason": "RegNet and SqueezeBert are unrelated model implementations; the PRs share a refactor theme only, not the same underlying issue." }, { "left": "pull_request:44010", "right": "pull_request:44129", "accept": false, - "reason": "SqueezeBert and SpeechT5 are separate model refactors; the shared output-capture pattern is too broad to treat as one duplicate PR." + "reason": "SqueezeBert and SpeechT5 modify different model-specific forward/output plumbing, so they are not duplicate fixes even though both standardize output capture." } ] }, @@ -20552,268 +20412,274 @@ "pull_request:44028|pull_request:44030", "pull_request:44029|pull_request:44030" ], - "prepared_review_unit_hash": "e22d35c90497aa6473076010e9c989e6d8a6e46eac7b99bc5152c4c477dbecc0", + "prepared_review_unit_hash": "12ae77e0d182a15f237b4e8ee98ea2c820851f0c9d062c0473b053e6b1f68a91", "component_budget": { "node_count": 13, "item_count": 13, "soft_pair_count": 39, - "serialized_chars": 39774, - "estimated_input_tokens": 9944, - "estimated_eval_tokens": 20144 + "serialized_chars": 39788, + "estimated_input_tokens": 9947, + "estimated_eval_tokens": 20150 }, "budget": { "node_count": 13, "item_count": 13, "soft_pair_count": 39, - "serialized_chars": 39774, - "estimated_input_tokens": 9944, - "estimated_eval_tokens": 20144 + "serialized_chars": 39788, + "estimated_input_tokens": 9947, + "estimated_eval_tokens": 20150 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "This cluster is a set of similarly named but distinct model-specific output-tracing refactors, not duplicate bugfixes or one mergeable PR.", - "confidence": 0.96, + "summary": "This cluster is a set of similar-looking but model-specific output-tracing refactors across unrelated code paths. They share a theme and some naming, but not the same underlying bug/change, so none of the soft edges should be merged as duplicates.", + "confidence": 0.93, "canonical_issue_reason": null, - "canonical_pr_reason": "No single canonical duplicate: each PR edits a different model path. If one representative is needed, 44025 is the broadest example because it spans two related files, but it does not subsume the others.", + "canonical_pr_reason": "PR 44025 is a good cluster representative: it clearly shows the core refactor pattern (adding `can_return_tuple` and simplifying forward output handling) and spans two related files, making it a strong exemplar of the shared theme.", "best_issue_reason": null, - "best_pr_reason": "44025 is the best anchor for the cluster because it is the most representative multi-file output-tracing refactor, but it is still model-specific rather than a duplicate of the rest.", + "best_pr_reason": "PR 44025 is the best overall representative for the cluster because its diff is straightforward, clearly documents the refactor intent, and captures the common output-tracing change pattern without being overly specialized to a single code path.", "soft_edge_verdicts": [ { "left": "pull_request:43998", "right": "pull_request:44002", "accept": false, - "reason": "Different model code paths (`timm_backbone` vs `upernet`); same refactor theme only." + "reason": "Different model families and different files; same refactor theme only, not the same concrete bug or change." }, { "left": "pull_request:43998", "right": "pull_request:44000", "accept": false, - "reason": "Different models (`timm_backbone` vs `vision_text_dual_encoder`); not the same underlying change." + "reason": "Both are output-tracing refactors, but they target different architectures and code paths." }, { "left": "pull_request:43998", "right": "pull_request:44001", "accept": false, - "reason": "Separate model implementations (`timm_backbone` vs `univnet`); too broad to treat as duplicates." + "reason": "Unrelated model implementations; similar decorator/import cleanup is not enough to treat as one change." }, { "left": "pull_request:43998", "right": "pull_request:44003", "accept": false, - "reason": "`timm_backbone` and `falcon_mamba/mamba` are unrelated forward paths; same pattern, different change." + "reason": "Timm backbone vs Mamba/Falcon Mamba: different model behavior and separate forward paths." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44004", + "accept": false, + "reason": "CodeGen\u2019s forward/cache handling is a different change from timm_backbone output tracing." }, { "left": "pull_request:43998", "right": "pull_request:44029", "accept": false, - "reason": "Different architectures (`timm_backbone` vs `rwkv`); not one mergeable fix." + "reason": "These are distinct model stacks; similarity is limited to the generic output-tracing refactor pattern." }, { "left": "pull_request:43999", "right": "pull_request:44001", "accept": false, - "reason": "`mobilenet_v1` and `univnet` are distinct model paths; same output-tracing refactor theme only." + "reason": "MobileNetV1 and UnivNet are unrelated models; not the same underlying fix." }, { "left": "pull_request:43999", "right": "pull_request:44002", "accept": false, - "reason": "Different files and models (`mobilenet_v1` vs `upernet`); not the same concrete bug/change." + "reason": "Different model code paths and different output types; only broad refactor overlap." }, { "left": "pull_request:43999", "right": "pull_request:44025", "accept": false, - "reason": "`mobilenet_v1` and `depth_anything` refactor different model forwards; too different to merge as duplicates." + "reason": "MobileNetV1 vs Depth Anything: same repo-wide refactor style, but not a shared bug/change." }, { "left": "pull_request:43999", "right": "pull_request:44026", "accept": false, - "reason": "`mobilenet_v1` vs `vision_encoder_decoder`; same subsystem theme, different code path." + "reason": "Separate model wrappers with different forward signatures and paths." }, { "left": "pull_request:43999", "right": "pull_request:44027", "accept": false, - "reason": "`mobilenet_v1` and `speech_encoder_decoder` are separate model wrappers; not one underlying change." + "reason": "MobileNetV1 and Speech Encoder-Decoder are unrelated; no shared concrete defect." }, { "left": "pull_request:44000", "right": "pull_request:44001", "accept": false, - "reason": "Different model families (`vision_text_dual_encoder` vs `univnet`); same refactor pattern, not duplicate." + "reason": "Vision-Text Dual Encoder and UnivNet are different model families; similarity is only in the refactor pattern." }, { "left": "pull_request:44000", "right": "pull_request:44002", "accept": false, - "reason": "`vision_text_dual_encoder` and `upernet` touch different forward paths; no shared concrete fix." + "reason": "Different output-tracing changes in distinct models, not one mergeable fix." }, { "left": "pull_request:44000", "right": "pull_request:44003", "accept": false, - "reason": "`vision_text_dual_encoder` vs `falcon_mamba/mamba`; unrelated model implementations." + "reason": "Vision-Text Dual Encoder and Mamba/Falcon Mamba touch different code paths and APIs." }, { "left": "pull_request:44000", "right": "pull_request:44025", "accept": false, - "reason": "`vision_text_dual_encoder` and `depth_anything` are different model-specific refactors; not mergeable as one PR." + "reason": "These are separate model implementations; the common theme is too broad to duplicate." }, { "left": "pull_request:44001", "right": "pull_request:44002", "accept": false, - "reason": "`univnet` and `upernet` are different code paths; same output-tracing theme only." + "reason": "UnivNet and UperNet are unrelated; same cleanup style does not imply same underlying issue." }, { "left": "pull_request:44001", "right": "pull_request:44003", "accept": false, - "reason": "Different model implementations (`univnet` vs `falcon_mamba/mamba`); not the same bug/change." + "reason": "Different model stacks and semantics; not plausibly one combined PR." }, { "left": "pull_request:44001", "right": "pull_request:44025", "accept": false, - "reason": "`univnet` and `depth_anything` refactor distinct forwards; too broad to combine." + "reason": "Depth Anything vs UnivNet: no shared code path or concrete bug." }, { - "left": "pull_request:44002", - "right": "pull_request:44029", + "left": "pull_request:44001", + "right": "pull_request:44026", "accept": false, - "reason": "`upernet` and `rwkv` are unrelated model paths; similar refactor, not duplicate." + "reason": "Unrelated model wrappers; similar decorator changes only." }, { "left": "pull_request:44002", "right": "pull_request:44003", "accept": false, - "reason": "`upernet` vs `falcon_mamba/mamba`; separate model-specific changes." + "reason": "UperNet and Mamba/Falcon Mamba are different subsystems with different output handling." }, { - "left": "pull_request:44003", + "left": "pull_request:44002", "right": "pull_request:44004", "accept": false, - "reason": "`falcon_mamba/mamba` and `codegen` target different code paths; not the same concrete fix." + "reason": "UperNet segmentation output refactor is not the same as CodeGen cache/output typing changes." }, { - "left": "pull_request:43998", + "left": "pull_request:44003", "right": "pull_request:44004", "accept": false, - "reason": "`timm_backbone` vs `codegen`; same refactor motif but unrelated implementations." + "reason": "Mamba/Falcon Mamba and CodeGen have different forward behavior and problem scope." }, { - "left": "pull_request:44001", + "left": "pull_request:43998", "right": "pull_request:44004", "accept": false, - "reason": "`univnet` and `codegen` are unrelated forward paths; not one underlying change." + "reason": "Timm backbone tracing cleanup and CodeGen output refactor are unrelated concrete changes." }, { - "left": "pull_request:44002", + "left": "pull_request:44001", "right": "pull_request:44004", "accept": false, - "reason": "`upernet` and `codegen` touch different model internals; not a duplicate pair." + "reason": "UnivNet and CodeGen do not share the same code path or defect." }, { - "left": "pull_request:44000", + "left": "pull_request:44002", "right": "pull_request:44004", "accept": false, - "reason": "`vision_text_dual_encoder` vs `codegen`; similar output-tracing refactor, different model change." + "reason": "Different model families and different return semantics." }, { "left": "pull_request:44025", "right": "pull_request:44026", "accept": false, - "reason": "Different wrappers (`depth_anything` vs `vision_encoder_decoder`); not the same code-path problem." + "reason": "Depth Anything and Vision/Encoder-Decoder are distinct model wrappers; same refactor theme only." }, { "left": "pull_request:44025", "right": "pull_request:44027", "accept": false, - "reason": "`depth_anything` and `speech_encoder_decoder` are separate model-specific refactors." + "reason": "Depth Anything and Speech Encoder-Decoder are not the same underlying bug/change." }, { "left": "pull_request:44025", "right": "pull_request:44028", "accept": false, - "reason": "`depth_anything` vs `superpoint`; same theme, different concrete model path." + "reason": "Depth estimation vs SuperPoint keypoint model: unrelated implementation paths." }, { "left": "pull_request:44025", "right": "pull_request:44029", "accept": false, - "reason": "`depth_anything` and `rwkv` are unrelated implementations; not mergeable as one PR." + "reason": "Depth Anything and Mamba/Falcon Mamba are different subsystems; no shared fix." }, { "left": "pull_request:44025", "right": "pull_request:44030", "accept": false, - "reason": "`depth_anything` vs `dpr`; different model code paths and scope." + "reason": "Depth Anything and DPR touch different model behaviors and output contracts." }, { "left": "pull_request:44026", "right": "pull_request:44027", "accept": false, - "reason": "`vision_encoder_decoder` and `speech_encoder_decoder` are different wrappers; similar refactor only." + "reason": "Vision Encoder-Decoder and Speech Encoder-Decoder are separate wrappers; similarity is broad architectural pattern only." }, { "left": "pull_request:44026", "right": "pull_request:44028", "accept": false, - "reason": "`vision_encoder_decoder` vs `superpoint`; not the same underlying bug/change." + "reason": "Different tasks and code paths; not the same concrete issue." }, { "left": "pull_request:44026", "right": "pull_request:44029", "accept": false, - "reason": "`vision_encoder_decoder` and `rwkv` are unrelated model paths." + "reason": "Distinct model families; no evidence they fix the same bug." }, { "left": "pull_request:44026", "right": "pull_request:44030", "accept": false, - "reason": "`vision_encoder_decoder` vs `dpr`; distinct model-specific changes." + "reason": "Vision/Speech encoder-decoder wrappers are unrelated to DPR's output handling." }, { "left": "pull_request:44027", "right": "pull_request:44028", "accept": false, - "reason": "`speech_encoder_decoder` and `superpoint` are separate implementations; not a duplicate fix." + "reason": "Speech Encoder-Decoder and SuperPoint are unrelated models with different forward contracts." }, { "left": "pull_request:44027", "right": "pull_request:44029", "accept": false, - "reason": "`speech_encoder_decoder` vs `rwkv`; same refactor pattern, different concrete change." + "reason": "Different model families and output mechanisms." }, { "left": "pull_request:44027", "right": "pull_request:44030", "accept": false, - "reason": "`speech_encoder_decoder` and `dpr` are unrelated model paths." + "reason": "Speech Encoder-Decoder and DPR are separate code paths; not a duplicate." }, { "left": "pull_request:44028", "right": "pull_request:44029", "accept": false, - "reason": "`superpoint` and `rwkv` are different code paths; too dissimilar for a duplicate." + "reason": "SuperPoint and Mamba/Falcon Mamba are unrelated; same refactor motif is insufficient." }, { "left": "pull_request:44028", "right": "pull_request:44030", "accept": false, - "reason": "`superpoint` vs `dpr`; distinct model-specific refactors." + "reason": "SuperPoint keypoint model vs DPR retrieval model: different concrete changes." }, { "left": "pull_request:44029", "right": "pull_request:44030", "accept": false, - "reason": "`rwkv` and `dpr` are unrelated implementations; not one mergeable PR." + "reason": "Mamba/Falcon Mamba and DPR do not share the same underlying bug or change." } ] }, @@ -20843,12 +20709,12 @@ "soft_pairs": [ "pull_request:44662|pull_request:45401" ], - "prepared_review_unit_hash": "2c7fcb183c37c13bdd48e87f62b3e17b1ca8c00b65f28318eed786226d496963", + "prepared_review_unit_hash": "15e041515752d0ca8f0dc56774432caae3973f9bc02c45f63f795910be52f14b", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 6552, + "serialized_chars": 6551, "estimated_input_tokens": 1638, "estimated_eval_tokens": 3532 }, @@ -20856,7 +20722,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 6552, + "serialized_chars": 6551, "estimated_input_tokens": 1638, "estimated_eval_tokens": 3532 }, @@ -20865,18 +20731,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These PRs are both new model integrations, but for different models and different code paths: PenguinVL vs Voxtral TTS. The overlap is only in generic Transformer scaffolding (docs, auto registries, model init), so they are not duplicates.", - "confidence": 0.98, + "summary": "Two unrelated model-addition PRs: one adds PenguinVL, the other adds Voxtral TTS support. They share only broad Transformers integration files and docs plumbing, not the same underlying feature or bug.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical PR: 44662 adds PenguinVL while 45401 adds Voxtral TTS, so they do not represent the same underlying change.", + "canonical_pr_reason": "PR #45401 is the more self-contained representative: it targets a specific tracked issue, has a focused scope around Voxtral TTS support, and includes the full implementation/testing stack for that model.", "best_issue_reason": null, - "best_pr_reason": "No single best PR for duplication purposes; both are standalone model-addition PRs and neither subsumes the other.", + "best_pr_reason": "PR #45401 is the best cluster representative because it is the more explicit, issue-backed feature addition and is easier to interpret as the cluster's exemplar model-support PR.", "soft_edge_verdicts": [ { "left": "pull_request:44662", "right": "pull_request:45401", "accept": false, - "reason": "Different target models and implementations; shared files are broad registry/docs updates, not the same concrete bug or change." + "reason": "Different models and code paths: PenguinVL vs Voxtral TTS. Shared files are only common Transformers registration/documentation surfaces, which is too broad to indicate the same underlying change." } ] }, @@ -20910,12 +20776,12 @@ "pull_request:45363|pull_request:46339", "pull_request:45666|pull_request:46339" ], - "prepared_review_unit_hash": "b817b71c7caaaa5f0f85555ee5731d2d1e7279c2e02596b1b2b651a32ed6beb0", + "prepared_review_unit_hash": "f290302ae2b505321b825f2508a04d3f5f0eb1744efeafc904120166a57c02b4", "component_budget": { "node_count": 4, "item_count": 4, "soft_pair_count": 3, - "serialized_chars": 9633, + "serialized_chars": 9634, "estimated_input_tokens": 2409, "estimated_eval_tokens": 5074 }, @@ -20923,7 +20789,7 @@ "node_count": 4, "item_count": 4, "soft_pair_count": 3, - "serialized_chars": 9633, + "serialized_chars": 9634, "estimated_input_tokens": 2409, "estimated_eval_tokens": 5074 }, @@ -20932,37 +20798,40 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "All items are related to the same kernel/module fusion area, but they look like successive iterations rather than duplicate PRs. PR 44979 introduces the base Module Fusion API, 45363 adds n-to-1 fusion via KernelConfig, 45666 extends that path, and 46339 further extends/simplifies the same feature set with additional loading/refactor changes.", - "confidence": 0.86, + "summary": "This cluster is mostly a chain of related kernel-fusion PRs. The latest PR, 46339, looks like the best canonical PR for the concrete n-to-1 KernelConfig fusion feature, while 44979 is a separate lower-level module-fusion API and should not be merged with the later KernelConfig work.", + "confidence": 0.87, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46339 is the broadest and most recent implementation in this line, touching the core loading and hub-kernel code paths with follow-up simplification/refactoring on top of the KernelConfig fusion work.", + "canonical_pr_reason": "PR 46339 is the most complete and latest version of the same n-to-1 KernelConfig fusion feature, adding the broader integration/simplification work that builds on the earlier PRs.", "best_issue_reason": null, - "best_pr_reason": "PR 46339 best represents the cluster because it is the latest, most complete kernel-fusion iteration and subsumes the earlier n-to-1 fusion direction more than the others.", + "best_pr_reason": "46339 is the best standalone representative because it appears to subsume the earlier KernelConfig fusion iterations and is the most polished implementation of the shared code-path change.", "soft_edge_verdicts": [ { "left": "pull_request:44979", "right": "pull_request:45363", "accept": false, - "reason": "Related feature area, but 44979 is the foundational Module Fusion API while 45363 is a separate n-to-1 KernelConfig implementation; they are not the same concrete change." + "reason": "44979 introduces a foundational Module Fusion API, while 45363 adds a specific KernelConfig-based n-to-1 kernel fusion integration; related, but not the same concrete change." }, { "left": "pull_request:45363", "right": "pull_request:46339", - "accept": false, - "reason": "46339 appears to be a later extension/simplification of the same fusion machinery, not the same single PR-level change; the shared files alone are not enough to treat them as duplicates." + "accept": true, + "reason": "Both target the same concrete n-to-1 kernel fusion via KernelConfig path, touch overlapping integration files, and 46339 reads like a follow-up simplification/extension of 45363." }, { "left": "pull_request:45666", "right": "pull_request:46339", - "accept": false, - "reason": "Both work on n-to-1 KernelConfig fusion, but 46339 is a follow-up refactor/extension with additional loading changes, so they are consecutive iterations rather than one duplicate fix." + "accept": true, + "reason": "These are the same KernelConfig-based n-to-1 fusion feature area, with 46339 appearing to extend and simplify the exact mechanism introduced in 45666." } ] }, - "evaluator_result": null, - "evaluator_used": false, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet: 44979 is clearly the separate Module Fusion API PR, while 45363/45666/46339 all center on KernelConfig-based n-to-1 kernel fusion and share overlapping files/titles. The soft-edge verdicts are conservative and do not overstate the evidence." + }, + "evaluator_used": true, "retried": false, - "accepted_nontrivial_soft_edge": false, + "accepted_nontrivial_soft_edge": true, "error_kind": null, "error_message": null }, @@ -20985,7 +20854,7 @@ "soft_pairs": [ "pull_request:45086|pull_request:45317" ], - "prepared_review_unit_hash": "e0939d904095dbc9451ca32b12fa6e427575b43a194b432b70c79a1b6dd3805c", + "prepared_review_unit_hash": "61cbe46094d9d2184a0bec3bb0f07407b7bd7d745ab33d022c2c7c9d62954757", "component_budget": { "node_count": 2, "item_count": 2, @@ -21007,18 +20876,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both pull requests fix the same AttributeError in `_patch_mistral_regex` by switching from `tokenizer.backend_tokenizer.pre_tokenizer` to `tokenizer.pre_tokenizer`; 45317 is the more complete version because it includes tests.", + "summary": "Both pull requests target the same Mistral regex patch bug in `tokenization_utils_tokenizers.py`; the later PR adds tests and is the stronger canonical candidate.", "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45317 is the better canonical PR because it contains the same core fix as 45086 plus test coverage, making it the more complete representative of this change.", + "canonical_pr_reason": "No issue node is present. Between the PRs, #45317 is the better canonical PR because it fixes the same AttributeError but also adds test coverage, making it the more complete change.", "best_issue_reason": null, - "best_pr_reason": "PR 45317 is the strongest overall PR candidate because it directly addresses the bug, includes validation in tests, and appears to subsume the earlier simpler patch.", + "best_pr_reason": "PR #45317 is the best overall representative: it addresses the same concrete `_patch_mistral_regex` AttributeError and includes validation tests, while #45086 is a narrower earlier version of the same fix.", "soft_edge_verdicts": [ { "left": "pull_request:45086", "right": "pull_request:45317", "accept": true, - "reason": "Same concrete bug fix in the same code path, with identical key edits in `tokenization_utils_tokenizers.py`; 45317 simply adds test coverage, so these are duplicate/overlapping implementations of the same change." + "reason": "Same bug, same code path, same file, and nearly identical fix; #45317 is essentially an expanded version of #45086 with tests, so they are duplicate/mergeable." } ] }, @@ -21048,7 +20917,7 @@ "soft_pairs": [ "pull_request:45115|pull_request:45569" ], - "prepared_review_unit_hash": "b457cb0bb457a684fea8123e943e10cf7cdd77edeabbeb9968695466a06e42e8", + "prepared_review_unit_hash": "25ca93cd0c9d3a9f49c7d80a6c84dcda1e42b265617ee910f4bbbe9c8bfb6254", "component_budget": { "node_count": 2, "item_count": 2, @@ -21070,18 +20939,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These PRs are related NemotronH architecture work, but they target different changes: one is a conversion-mapping/refactor PR for inheriting GraniteMoeHybrid-compatible weights, while the other is a broader split into NemotronH dense/sparse variants with docs and auto-mapping updates. They are not the same fix and should not be merged as duplicates.", + "summary": "These are related NemotronH model work items, but they are not the same fix: one is a narrower refactor/conversion-mapping update, while the other splits NemotronH into dense/sparse variants with broader API, docs, and test changes.", "confidence": 0.91, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45569 is the better representative of the cluster because it is the broader, more complete NemotronH rework and includes the model split plus supporting documentation and registration changes.", + "canonical_pr_reason": "PR 45569 is the better canonical PR because it is the broader, more complete model-structure change: it introduces the dense/sparse variants, updates auto/model registrations, adds docs, and includes tests. PR 45115 is narrower and mainly refactors inheritance/conversion mappings.", "best_issue_reason": null, - "best_pr_reason": "PR 45569 best captures the main final direction of the NemotronH changes; PR 45115 is a narrower conversion-mapping refactor that overlaps only partially on shared model files.", + "best_pr_reason": "PR 45569 is the best representative of the cluster since it addresses the main architectural change and touches the full user-facing surface. PR 45115 looks like a partial supporting refactor rather than the primary end-state implementation.", "soft_edge_verdicts": [ { "left": "pull_request:45115", "right": "pull_request:45569", "accept": false, - "reason": "They share NemotronH files, but the concrete changes differ: 45115 focuses on checkpoint conversion and GraniteMoeHybrid-compatible renaming, while 45569 introduces dense/sparse model variants, docs, and auto/model registration updates. Same subsystem, not the same underlying change." + "reason": "They overlap on NemotronH internals, but 45115 is about inheritance/conversion mapping cleanup for NemotronH, while 45569 is a larger split into NemotronH dense/sparse variants with docs, auto-mapping, and tests. Not the same concrete change and not safely mergeable into one PR." } ] }, @@ -21111,7 +20980,7 @@ "soft_pairs": [ "pull_request:45332|pull_request:45333" ], - "prepared_review_unit_hash": "67b4f5d135617a858ff7d516494b441bc7a55ca90a373be98f82dc878aef120d", + "prepared_review_unit_hash": "fbef38615b0bdc887bdabbda27a14d8ef4bbd53418cdad7f8b46a2de1920bc78", "component_budget": { "node_count": 2, "item_count": 2, @@ -21133,18 +21002,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two open PRs in the same heterogeneity feature area: one is a broad implementation of heterogeneous model support, while the other is a narrower per-layer configuration PR. They overlap in config scaffolding but do not look like duplicate fixes/changes.", - "confidence": 0.89, + "summary": "Two related but distinct PRs for heterogeneous model support: one adds per-layer config infrastructure, the other adds heterogeneous modeling/cache changes. They overlap on shared config plumbing but are not the same concrete change.", + "confidence": 0.84, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #45332 is the broader umbrella change: it adds heterogeneous model support across config, modeling, cache handling, and tests. It subsumes the per-layer config groundwork and is the better representative of the cluster.", + "canonical_pr_reason": "PR #45332 is the broader umbrella change and includes the modeling-side integration; #45333 is a narrower config-only foundation. If picking one representative, #45332 is the better canonical PR, but they are not duplicates.", "best_issue_reason": null, - "best_pr_reason": "PR #45332 is the most complete and central implementation in this cluster, so it is the best candidate to keep as the primary PR for the feature.", + "best_pr_reason": "#45332 is the best representative of the cluster because it spans the larger end-to-end heterogeneous model support surface, while #45333 is a prerequisite/config subset.", "soft_edge_verdicts": [ { "left": "pull_request:45332", "right": "pull_request:45333", "accept": false, - "reason": "Related feature work, but not the same concrete change. #45333 is configuration-only groundwork, while #45332 extends into modeling and cache behavior; they could coexist as layered PRs rather than being merged as duplicates." + "reason": "They share heterogeneous-config plumbing, but one is a broad modeling/cache/masking PR and the other is a config-only PR. Different concrete code-paths and not safe to treat as the same change." } ] }, @@ -21174,12 +21043,12 @@ "soft_pairs": [ "pull_request:45415|pull_request:45425" ], - "prepared_review_unit_hash": "c56d93992db94261afe194e1b37407cc1516061a53ec316522767714e9cbf8bf", + "prepared_review_unit_hash": "19e3bcce600d7e3f35500917b22fadb3a3354b17d5227fe61d35da67962e0bee", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 5513, + "serialized_chars": 5514, "estimated_input_tokens": 1379, "estimated_eval_tokens": 3014 }, @@ -21187,7 +21056,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 5513, + "serialized_chars": 5514, "estimated_input_tokens": 1379, "estimated_eval_tokens": 3014 }, @@ -21196,18 +21065,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These are both typing-related PRs, but they address different changes: one is a broad type-checking sweep across many files, while the other adds typing helper protocols and targeted fixes around `modeling_utils`/flash-attention integration.", - "confidence": 0.72, + "summary": "Both items are typing-related PRs, but they address different scopes: #45415 is a broad repo-wide type-checking sweep, while #45425 is a more targeted typing/refactor PR centered on modeling_utils and shared typing protocols. They are not duplicates.", + "confidence": 0.94, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45425 is the better representative because it is a focused typing-infrastructure change with a clearer, narrower scope. PR 45415 is a broad repo-wide typing pass that bundles many unrelated file edits.", + "canonical_pr_reason": "No clear duplicate PR in this cluster. If one must be selected, #45425 is the better representative because it is a focused, self-contained typing change, whereas #45415 is a wide-ranging type-checking cleanup across many files.", "best_issue_reason": null, - "best_pr_reason": "PR 45425 is the most suitable single representative for the cluster because its change set is more concrete and self-contained, whereas PR 45415 is a wide typing sweep with many touched files and mixed cleanup work.", + "best_pr_reason": "#45425 is the best representative of the cluster because it is narrowly scoped and technically coherent; #45415 is broader and mixes many unrelated typing edits, making it a poorer canonical stand-in.", "soft_edge_verdicts": [ { "left": "pull_request:45415", "right": "pull_request:45425", "accept": false, - "reason": "Both are about typing, but they do not appear to fix the same concrete bug or implement the same change. 45415 is a broad type-checking enablement across many modules; 45425 adds specific typing protocols and related fixes in a smaller area. They would not plausibly be merged as one PR." + "reason": "Shared typing-related files are not enough here: #45415 is a broad typing sweep, while #45425 introduces specific typing protocols and annotations. They do not look like the same underlying fix and would not plausibly merge into one PR." } ] }, @@ -21237,12 +21106,12 @@ "soft_pairs": [ "pull_request:45513|pull_request:46084" ], - "prepared_review_unit_hash": "baf48fe75836276b9d91b2731ff9080f9950162981d3a1fee2fc960d2d071a62", + "prepared_review_unit_hash": "8d6dd30b13f1bae327f83e28b182395920e3236527105fc96895cef4ab218543", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4857, + "serialized_chars": 4858, "estimated_input_tokens": 1215, "estimated_eval_tokens": 2686 }, @@ -21250,7 +21119,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4857, + "serialized_chars": 4858, "estimated_input_tokens": 1215, "estimated_eval_tokens": 2686 }, @@ -21259,18 +21128,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These two PRs both touch cached multi-token forward paths, but they target different model families and different bugs. They are superficially similar yet not the same underlying change, so the soft edge should be rejected.", - "confidence": 0.97, + "summary": "Both PRs address cache-enabled forward paths with seq_len > 1, but they target different model families and distinct code paths, so they do not look like the same underlying fix.", + "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical duplicate PR: #45513 is a broad Qwen3.5/Qwen3-next/OLMo-hybrid cache fix, while #46084 is a focused Mamba2 seq_len>1 cache-path fix for a different code path.", + "canonical_pr_reason": "PR 46084 is the cleaner representative: it is narrowly scoped to Mamba2 cache handling and explicitly tied to issue #46032.", "best_issue_reason": null, - "best_pr_reason": "#46084 is the cleaner standalone PR: it is tightly scoped, explicitly tied to issue #46032, and fixes one concrete cache/seq_len bug in Mamba2.", + "best_pr_reason": "PR 46084 is the best single PR to represent this cluster because it is tightly scoped, issue-linked, and more clearly describes one concrete bug fix.", "soft_edge_verdicts": [ { "left": "pull_request:45513", "right": "pull_request:46084", "accept": false, - "reason": "Both involve use_cache and multi-token cached forwarding, but they address different architectures (Qwen/OLMo hybrid vs Mamba2) and different implementation bugs. They do not appear to be the same fix and would not plausibly merge into one PR." + "reason": "Different architectures and different implementations: Qwen/OLMo linear-attention cached forward vs Mamba2 CUDA-kernel cache handling. Same broad theme, not the same concrete bug." } ] }, @@ -21300,12 +21169,12 @@ "soft_pairs": [ "pull_request:45548|pull_request:45662" ], - "prepared_review_unit_hash": "5bf18bcf0dea43cc3a836c3a5fee2b5261e2e8d474c8f499127efd0318ee9799", + "prepared_review_unit_hash": "6c3cb2761dce33a74873dcb9c6035c69d7d54407de6a5a7aa75e73798fdcb76f", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4508, + "serialized_chars": 4507, "estimated_input_tokens": 1127, "estimated_eval_tokens": 2510 }, @@ -21313,7 +21182,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4508, + "serialized_chars": 4507, "estimated_input_tokens": 1127, "estimated_eval_tokens": 2510 }, @@ -21322,18 +21191,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two PRs both involve expert parallelism, but they address different concrete bugs: one fixes EP + ZeRO-3 loading under accelerate launch, the other fixes EP + FSDP2 expert parameters being overwritten by rank-0 broadcast. They share a broad subsystem and one file, but not the same underlying change.", - "confidence": 0.89, + "summary": "These are related by expert-parallel loading infrastructure, but they fix different bugs: PR 45548 is about preserving EP/DeepSpeed ZeRO-3 loading metadata through accelerate launch, while PR 45662 targets expert parameters being overwritten during FSDP2 loading. Not a duplicate pair.", + "confidence": 0.92, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #45662 is the better canonical representative for the cluster because it fixes the more explicit EP runtime corruption bug and includes the broader code-path changes across loading and MoE/tensor-parallel integration.", + "canonical_pr_reason": "PR 45662 is the better cluster representative because it addresses a concrete expert-parameter corruption bug and includes the broader set of loading-path changes around that failure mode.", "best_issue_reason": null, - "best_pr_reason": "PR #45662 is the strongest standalone candidate: it targets a concrete failure mode (experts overwritten by rank-0 broadcast) with a coherent fix across the relevant loading and distributed parameter paths.", + "best_pr_reason": "PR 45662 is the strongest overall representative: its title and diff describe a specific, high-signal EP loading bug with direct code-path fixes, whereas 45548 is narrower and centered on config serialization/launch behavior.", "soft_edge_verdicts": [ { "left": "pull_request:45548", "right": "pull_request:45662", "accept": false, - "reason": "Different bugs and different fix paths: #45548 is about EP + DeepSpeed ZeRO-3 loading via accelerate launch, while #45662 is about EP + FSDP2 rank-0 broadcast overwriting expert parameters. Shared EP context and one shared file are not enough to treat them as the same PR-worthy change." + "reason": "Shared expert-parallel context and one common file are not enough; 45548 fixes config serialization/launch handling for ZeRO-3, while 45662 fixes expert tensors being overwritten in FSDP2 loading. Different underlying bugs and not plausibly one merged PR." } ] }, @@ -21363,12 +21232,12 @@ "soft_pairs": [ "pull_request:45639|pull_request:45645" ], - "prepared_review_unit_hash": "c1b0045dc92905bd5ad14ac59dd0c3c91f7fa498441ec9f94954f5ee11da0ef5", + "prepared_review_unit_hash": "b401dbd0659916b5294fc7147db4101cd24cf9e6aa07295e15cf0f3377db2caf", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4590, + "serialized_chars": 4589, "estimated_input_tokens": 1148, "estimated_eval_tokens": 2552 }, @@ -21376,7 +21245,7 @@ "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4590, + "serialized_chars": 4589, "estimated_input_tokens": 1148, "estimated_eval_tokens": 2552 }, @@ -21385,18 +21254,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs address the same underlying xdist collision in captured_info debug artifacts; #45645 is the broader, more complete follow-up that also updates CI log collection, while #45639 is the smaller core fix in testing_utils.", - "confidence": 0.93, + "summary": "Both PRs address the same xdist collision problem around captured_info output in testing_utils; PR 45645 appears to be the broader, more complete version with CI log aggregation added.", + "confidence": 0.79, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #45645 is the better canonical PR because it covers the core xdist-safe captured_info fix plus the downstream workflow/log-handling changes needed to preserve debug output in CI.", + "canonical_pr_reason": "PR 45645 is the better canonical PR because it covers the core xdist-safe output fix plus the downstream CI/workflow handling needed to read multiple captured_info files.", "best_issue_reason": null, - "best_pr_reason": "PR #45645 is the best overall representative since it fully addresses the artifact-collision problem and includes the CI-side adjustments, not just the library-side change.", + "best_pr_reason": "PR 45645 is the strongest representative: it addresses the same underlying captured_info xdist collision and also preserves/debug-prints the resulting artifacts in CI.", "soft_edge_verdicts": [ { "left": "pull_request:45639", "right": "pull_request:45645", "accept": true, - "reason": "Both PRs fix the same concrete bug: xdist workers colliding on captured_info debug output. #45645 appears to extend the same fix with CI artifact aggregation, so they are plausibly mergeable as one change." + "reason": "Same underlying bug: xdist workers colliding on captured_info output for patched testing logs. PR 45645 extends the same fix into CI artifact collection, so these are plausibly one combined change." } ] }, @@ -21428,46 +21297,46 @@ "pull_request:45670|pull_request:46320", "pull_request:46320|pull_request:46401" ], - "prepared_review_unit_hash": "42790d45d71d8710b14213bb15d55f197d452de65d29dbf9f053fb8de94907fc", + "prepared_review_unit_hash": "7307da740cca9642070b10d873f943f0e2e69e50b2858d5ec6f3993400c75195", "component_budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 2, - "serialized_chars": 7028, - "estimated_input_tokens": 1757, - "estimated_eval_tokens": 3770 + "serialized_chars": 7029, + "estimated_input_tokens": 1758, + "estimated_eval_tokens": 3772 }, "budget": { "node_count": 3, "item_count": 3, "soft_pair_count": 2, - "serialized_chars": 7028, - "estimated_input_tokens": 1757, - "estimated_eval_tokens": 3770 + "serialized_chars": 7029, + "estimated_input_tokens": 1758, + "estimated_eval_tokens": 3772 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "These PRs form a loose topic cluster around model/auto-model support and attention dispatch, but they are not duplicates of the same underlying change. The only potentially similar pair is the RecurrentGemma and BLIP-2-family SDPA/attn dispatch work, which targets different model code paths and would not plausibly merge into one PR. The GLM ASR auto-mapping PR is unrelated.", - "confidence": 0.84, + "summary": "The cluster is mostly about model-specific attn_implementation dispatch support, plus one unrelated AutoModel registration nit. The RecurrentGemma and BLIP-2 PRs are similar in theme but target different code paths and should not be treated as duplicates.", + "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": "PR #46401 is the best canonical representative only in the sense that it is the clearest, most self-contained implementation PR in the cluster: it adds SDPA dispatch for the BLIP-2 family with code changes and tests. However, it is not a duplicate of the RecurrentGemma or GLM ASR PRs.", + "canonical_pr_reason": "PR 46320 is the best representative of the cluster\u2019s main theme: a focused model-specific attn_implementation dispatch implementation for RecurrentGemma, with supporting conversion and tests.", "best_issue_reason": null, - "best_pr_reason": "PR #46401 is the strongest standalone PR here because it has the clearest concrete code-path fix, spans the main implementation plus tests/docs, and is more substantial than the small auto-model mapping tweak in #45670 or the narrower RecurrentGemma dispatch change in #46320.", + "best_pr_reason": "PR 46320 is the strongest canonical PR because it is the clearest, most self-contained example of the attention-dispatch feature work in this cluster.", "soft_edge_verdicts": [ { "left": "pull_request:45670", "right": "pull_request:46320", "accept": false, - "reason": "Different underlying changes: auto-model registry mapping for glmasr vs RecurrentGemma attention-dispatch support. Same broad Transformers area, but not the same bug or mergeable change." + "reason": "Different changes: AutoModel mapping for glmasr versus RecurrentGemma attention-dispatch support. They do not fix the same code-path problem." }, { "left": "pull_request:46320", "right": "pull_request:46401", "accept": false, - "reason": "Both mention attn_implementation dispatch, but they fix different model families and code paths (RecurrentGemma vs BLIP-2/InstructBLIP). They are related themes, not duplicates of one concrete change." + "reason": "Both add attn_implementation dispatch, but for different model families and different implementations. They are similar feature work, not the same concrete bug/change, and would not naturally merge into one PR." } ] }, @@ -21497,40 +21366,40 @@ "soft_pairs": [ "pull_request:45720|pull_request:46259" ], - "prepared_review_unit_hash": "4c1941fb403da8d2182f6de834754e58e4c48a81daee9362ab79ad8b4ab63d70", + "prepared_review_unit_hash": "b326c2c80e43af865f995d624df965cc2cce76de41b05bb9696d24425e0c1268", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 7189, - "estimated_input_tokens": 1798, - "estimated_eval_tokens": 3852 + "serialized_chars": 7188, + "estimated_input_tokens": 1797, + "estimated_eval_tokens": 3850 }, "budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 7189, - "estimated_input_tokens": 1798, - "estimated_eval_tokens": 3852 + "serialized_chars": 7188, + "estimated_input_tokens": 1797, + "estimated_eval_tokens": 3850 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Two unrelated linter/modeling-rule PRs: one implements TRF003 for `return_dict`/tuple-return handling in model forwards, the other adds TRF016 for preprocessing flag consumption. They share some touched model files but not the same underlying bug or change.", - "confidence": 0.96, + "summary": "Two unrelated open PRs were clustered only by soft similarity. PR #45720 implements TRF003 tuple/return_dict handling across model forwards, while PR #46259 adds a different modeling rule about do_* flags in image/video processors. They do not look like the same underlying change.", + "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "No canonical PR as a duplicate target: the PRs address different lint rules and different code paths, so they should not be merged or treated as one change.", "best_issue_reason": null, - "best_pr_reason": "No single PR is a good canonical representative because these are distinct rule additions/fixes affecting different code paths and different linter behavior.", + "best_pr_reason": "PR #45720 is the cleaner representative of the cluster if one must be chosen, because it is a focused code-path fix across model forwards; PR #46259 is a broader, different rule addition spanning processors and docs.", "soft_edge_verdicts": [ { "left": "pull_request:45720", "right": "pull_request:46259", "accept": false, - "reason": "Different changes: TRF003 modifies model forward signatures/tuple-return behavior, while TRF016 adds a rule about dead `do_*` preprocessing flags. Shared model filenames are incidental and do not indicate the same fix." + "reason": "Different underlying tasks: tuple/return_dict forward cleanup vs do_* flag consumption in image/video processors. Shared filenames are incidental and not evidence of the same fix." } ] }, @@ -21626,40 +21495,40 @@ "soft_pairs": [ "pull_request:45932|pull_request:46100" ], - "prepared_review_unit_hash": "26f25861a59b30af689700ec2c1e11a92e1884bdb364ae72c64eea37f6b79dc0", + "prepared_review_unit_hash": "701eba2c86a20efc1e8a909c4ba2bc1bf500df8228552da7ef88b42407d7c6ab", "component_budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4300, - "estimated_input_tokens": 1075, - "estimated_eval_tokens": 2406 + "serialized_chars": 4301, + "estimated_input_tokens": 1076, + "estimated_eval_tokens": 2408 }, "budget": { "node_count": 2, "item_count": 2, "soft_pair_count": 1, - "serialized_chars": 4300, - "estimated_input_tokens": 1075, - "estimated_eval_tokens": 2406 + "serialized_chars": 4301, + "estimated_input_tokens": 1076, + "estimated_eval_tokens": 2408 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two PRs are not duplicates: one is a broad code/config/workflow change across multiple model files, while the other is a test-only PR adding deliberately failing assertions in ViT tests. They do not address the same underlying bug or change.", - "confidence": 0.96, + "summary": "These are not duplicates. PR #45932 is a broad model/config/workflow change, while PR #46100 is a tiny test-only change that intentionally adds failing assertions and targets a different code path. They should not be clustered as the same underlying change.", + "confidence": 0.98, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 45932 is the substantive cluster member, touching actual model/config/workflow code across multiple files; PR 46100 is only a test-migration placeholder with failing assertions and is not the same change.", + "canonical_pr_reason": "PR #45932 is the better canonical PR because it contains the substantive code changes across model implementations and workflow/secrets plumbing, whereas PR #46100 is just a test-migration stub with deliberate failures.", "best_issue_reason": null, - "best_pr_reason": "PR 45932 is the best representative because it is the real implementation change; PR 46100 is clearly test-only and not a candidate for duplicate consolidation with it.", + "best_pr_reason": "PR #45932 is the most representative and materially meaningful change in the set; PR #46100 is a narrow test-only edit and is not a good canonical representative.", "soft_edge_verdicts": [ { "left": "pull_request:45932", "right": "pull_request:46100", "accept": false, - "reason": "Reject: 45932 changes model/config/workflow plumbing across several files, while 46100 only edits ViT tests with dummy failing assertions. They do not fix the same code-path problem and would not plausibly be merged as one PR." + "reason": "Different purpose and code paths: #45932 changes multiple model implementations and CI secrets, while #46100 only edits a ViT test file and adds failing assertions. They do not appear to fix the same bug or implement the same feature." } ] }, @@ -21692,7 +21561,7 @@ "pull_request:46073|pull_request:46196", "pull_request:46154|pull_request:46196" ], - "prepared_review_unit_hash": "e254aee5bacdb3f280e0af5528a85fc8262fef2b4e50552dccdef2fa7eb1417e", + "prepared_review_unit_hash": "f605969488e985877dc33290309e829e6f53dd1992380ea53f75f1975535c6e4", "component_budget": { "node_count": 3, "item_count": 3, @@ -21714,30 +21583,30 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Three separate model-specific PRs add SDPA dispatch support in different code paths (CTRL, T5, and wav2vec2/SeamlessM4T variants). They share a theme but do not look mergeable into one single duplicate PR.", - "confidence": 0.94, + "summary": "These are three separate PRs adding SDPA dispatch support to different model families (CTRL, T5, and wav2vec2/SeamlessM4T). They share a broad theme but not the same concrete bug or code path, so they should not be clustered as duplicates.", + "confidence": 0.96, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46196 is the broadest representative of the cluster: it covers multiple model families and the same SDPA-dispatch theme, making it the best single PR to anchor the cluster.", + "canonical_pr_reason": "No PR is a true duplicate of the others. If a representative must be chosen, 46154 is the cleanest single-model PR and the most self-contained of the set.", "best_issue_reason": null, - "best_pr_reason": "PR 46196 is the best representative PR because it spans the widest set of affected models and most clearly captures the shared feature direction across the cluster.", + "best_pr_reason": "46154 is the most focused and least broad: it targets one model family (T5) with a clear SDPA dispatch change, making it the best representative of the cluster, though not a duplicate of the others.", "soft_edge_verdicts": [ { "left": "pull_request:46073", "right": "pull_request:46154", "accept": false, - "reason": "Both add SDPA support, but for different architectures (CTRL vs T5) with different model code paths and tests; not the same underlying change." + "reason": "Both are SDPA-related, but they modify different model implementations (CTRL vs T5) and different code paths. Same theme, not the same underlying fix." }, { "left": "pull_request:46073", "right": "pull_request:46196", "accept": false, - "reason": "These target different model families and distinct attention implementations; they are related feature work, not duplicate fixes or one mergeable PR." + "reason": "CTRL SDPA support is unrelated to the wav2vec2_bert / wav2vec2_conformer / SeamlessM4T changes. The PRs touch different architectures and cannot plausibly be merged as one fix." }, { "left": "pull_request:46154", "right": "pull_request:46196", "accept": false, - "reason": "T5 SDPA dispatch and wav2vec2/SeamlessM4T SDPA dispatch are separate model-specific changes with different affected files and logic." + "reason": "T5 dispatch support and the multi-model wav2vec2/SeamlessM4T dispatch work address different model-specific implementations, so they are not duplicates." } ] }, @@ -21767,7 +21636,7 @@ "soft_pairs": [ "pull_request:46126|pull_request:46269" ], - "prepared_review_unit_hash": "fdd46eda2eb3f503daa9baf11ace80366f82182947d87d64b848c884ab035226", + "prepared_review_unit_hash": "41468e419c5c5cb67fcf6a981df82e5667fca5841c45d993f3c5e15f21d04f15", "component_budget": { "node_count": 2, "item_count": 2, @@ -21789,18 +21658,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both items are PRs in the distributed stack, but they are not the same change. PR 46126 is a focused EP/Trainer integration patch, while PR 46269 is a broad distributed branch base that spans many subsystems.", - "confidence": 0.92, + "summary": "Two pull requests overlap on distributed/tensor-parallel files, but they are not the same change: PR 46126 is a focused EP + Trainer integration patch, while PR 46269 is a broad distributed branch base touching many unrelated areas. This is not a duplicate pair.", + "confidence": 0.95, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46126 is the more focused, concrete change set and better represents a single underlying fix/feature than the broad base branch in PR 46269.", + "canonical_pr_reason": "PR 46126 is the more focused, self-contained change set centered on EP and Trainer integration; PR 46269 is a large umbrella branch with many unrelated distributed updates, so it is a poorer canonical duplicate target.", "best_issue_reason": null, - "best_pr_reason": "PR 46126 is the best representative PR for triage because it targets a specific EP routing/Trainer integration path and is narrowly scoped.", + "best_pr_reason": "PR 46126 is the best representative of a concrete code-path change. PR 46269 is too broad and mixed to serve as a clean duplicate anchor.", "soft_edge_verdicts": [ { "left": "pull_request:46126", "right": "pull_request:46269", "accept": false, - "reason": "Reject: the overlap is only superficial. 46126 implements EP/Trainer integration around tensor-parallel routing, while 46269 is a large distributed branch base touching many unrelated files; they do not appear to fix the same concrete code-path problem and would not plausibly merge into one PR." + "reason": "Shared filenames and soft similarity are not enough here: 46126 implements a specific EP/Trainer integration, while 46269 is a large distributed branch base spanning many subsystems. They do not look like the same underlying bug or mergeable single PR." } ] }, @@ -21830,7 +21699,7 @@ "soft_pairs": [ "pull_request:46152|pull_request:46207" ], - "prepared_review_unit_hash": "30e14149d8fdf8fd0be3feddf93f0b290fe8c1dd1fd0b044d4c6d997713ff1b2", + "prepared_review_unit_hash": "b06bec4a63189d5d5108cb1eec88b0f437d7b18ec7e8b55b375ab80503d5f269", "component_budget": { "node_count": 2, "item_count": 2, @@ -21852,18 +21721,18 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "Both PRs are about removing brittle source-file open() usage, but they fix different code paths in different modules: one in import_utils for resolving package traversables, the other in modeling_utils for attention-implementation detection. They are thematically related but not duplicates.", - "confidence": 0.94, + "summary": "These PRs are related only at a high level: both stop using direct file-open source inspection, but they fix different code paths in different modules with different mechanisms. They should not be merged as duplicates.", + "confidence": 0.89, "canonical_issue_reason": null, - "canonical_pr_reason": "PR 46207 is the more concrete and self-contained fix: it updates the _can_set_attn_implementation path in modeling_utils to use inspect.getsource-style source inspection instead of file open(), with tests. It targets a single behavior change and is clearly scoped.", + "canonical_pr_reason": "PR 46207 is the better representative of the cluster because it targets one concrete source-reading check in `modeling_utils.py` and includes tests, whereas PR 46152 is a broader import-utils refactor with a different implementation strategy.", "best_issue_reason": null, - "best_pr_reason": "PR 46207 is the best representative of the cluster because it addresses one specific bug-prone source-reading path directly. PR 46152 is broader and centered on a utility helper in import_utils, so it is less canonical for this cluster.", + "best_pr_reason": "PR 46207 is the strongest standalone candidate: it addresses a specific bug-prone source inspection path and is narrowly scoped enough to evaluate on its own.", "soft_edge_verdicts": [ { "left": "pull_request:46152", "right": "pull_request:46207", "accept": false, - "reason": "Same broad theme (avoid open() on source), but they modify different modules and different concrete mechanisms. 46152 adds a traversable resolver in import_utils; 46207 changes model capability detection in modeling_utils. Not the same underlying bug or mergeable into one PR." + "reason": "Different underlying changes: 46152 refactors internal source-path resolution in `import_utils.py`, while 46207 changes `_can_set_attn_implementation` in `modeling_utils.py` from file-open to `inspect.getsource`. Same broad theme, but not the same concrete bug or patch." } ] }, @@ -21898,12 +21767,12 @@ "pull_request:46362|pull_request:46368", "pull_request:46362|pull_request:46366" ], - "prepared_review_unit_hash": "538fa3eed621a1be5726697d03b63ee2c8c18f113f0df68c3690d958ab26e4d7", + "prepared_review_unit_hash": "daf9670b7a981f8840e7eed5a9545e310174299c0652fcb79e09e0a428fa375b", "component_budget": { "node_count": 4, "item_count": 4, "soft_pair_count": 4, - "serialized_chars": 8228, + "serialized_chars": 8227, "estimated_input_tokens": 2057, "estimated_eval_tokens": 4370 }, @@ -21911,7 +21780,7 @@ "node_count": 4, "item_count": 4, "soft_pair_count": 4, - "serialized_chars": 8228, + "serialized_chars": 8227, "estimated_input_tokens": 2057, "estimated_eval_tokens": 4370 }, @@ -21920,36 +21789,36 @@ "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The cluster is superficially similar because all four PRs touch model test files and three of them are test skips, but they address different models and different concrete behaviors. There is no true duplicate pair here.", - "confidence": 0.95, + "summary": "These PRs are mostly unrelated model-specific test adjustments. The only close soft link is between the two DeepSeek V4 PRs, but they fix different problems: one skips an incompatible quant-cache test, the other adds XPU/CUDA backend selection for distributed testing. I would not merge any of the soft pairs as duplicates.", + "confidence": 0.92, "canonical_issue_reason": null, - "canonical_pr_reason": null, + "canonical_pr_reason": "PR 46366 is the most substantive standalone change in the set: it addresses distributed-test backend selection and is more representative of an actual code-path fix than the model-specific skip-only PRs.", "best_issue_reason": null, - "best_pr_reason": "PR 46366 is the most substantive and self-contained change in the set: it adds backend detection and a clearer error path for the deepseek_v4 distributed test. Even so, it is not a duplicate of the other PRs.", + "best_pr_reason": "PR 46366 is the strongest standalone candidate to keep because it fixes environment/backend handling rather than merely skipping a failing test, and it is not tied to a single model-specific incompatibility.", "soft_edge_verdicts": [ { "left": "pull_request:46337", "right": "pull_request:46368", "accept": false, - "reason": "Both are skip-style test changes, but for different architectures and different model-specific failures: cohere2_moe test sizing/offload behavior vs nemotron_h quant-cache incompatibility." + "reason": "Both are model-test maintenance PRs, but they target different models and different failure modes: Cohere2 MoE test setup vs NemotronH quant-cache skipping." }, { "left": "pull_request:46337", "right": "pull_request:46362", "accept": false, - "reason": "cohere2_moe adjusts test configuration to fit accelerate splitting, while deepseek_v4 skips a quant-cache test; these are unrelated code paths and bugs." + "reason": "Different models and different changes. One adjusts Cohere2 MoE test configuration; the other skips a DeepSeek V4 quant-cache test." }, { "left": "pull_request:46362", "right": "pull_request:46368", "accept": false, - "reason": "They share the same test name pattern, but the failures are model-specific quant-cache incompatibilities in different models, so they are not the same underlying change." + "reason": "Superficially similar skips, but they are for different model code paths and different cache implementations; not the same underlying bug/change." }, { "left": "pull_request:46362", "right": "pull_request:46366", "accept": false, - "reason": "One PR skips a quant-cache test; the other adds CUDA/XPU backend selection for distributed execution. Different functionality, not one mergeable fix." + "reason": "Same file, but the changes are unrelated: one skips a quant-cache test, the other adds XPU fallback logic for a distributed worker." } ] }, @@ -21971,48 +21840,56 @@ "source": "llm", "cache_hit": false, "model": "gpt-5.4-mini?service_tier=flex", - "cluster_id": "cluster-46376-2", + "cluster_id": "cluster-46376-3", "nodes": [ "pull_request:46376", - "pull_request:46417" + "pull_request:46417", + "pull_request:46425" ], "soft_pairs": [ - "pull_request:46376|pull_request:46417" + "pull_request:46376|pull_request:46417", + "pull_request:46376|pull_request:46425" ], - "prepared_review_unit_hash": "082c3754f1ae3cf547f9c1eb05f06f159c9f86af014f648bb0cc3ed501665595", + "prepared_review_unit_hash": "18e3dad1947182c71e5c4dd657bec9ca5de2262d5b8e1c24c9c5a379958ead4d", "component_budget": { - "node_count": 2, - "item_count": 2, - "soft_pair_count": 1, - "serialized_chars": 5960, - "estimated_input_tokens": 1490, - "estimated_eval_tokens": 3236 + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 9319, + "estimated_input_tokens": 2330, + "estimated_eval_tokens": 4916 }, "budget": { - "node_count": 2, - "item_count": 2, - "soft_pair_count": 1, - "serialized_chars": 5960, - "estimated_input_tokens": 1490, - "estimated_eval_tokens": 3236 + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 9319, + "estimated_input_tokens": 2330, + "estimated_eval_tokens": 4916 }, "overflow_policy": "truncate_then_skip", "trimmed": false, "aggressively_trimmed": false, "split": false, "analyst_result": { - "summary": "The two PRs are unrelated: one adds multimodal support to continuous batching, the other adds a new HunYuan VL model integration and docs. No duplicate/merge candidate here.", + "summary": "This cluster groups three unrelated PRs: multimodal continuous batching infrastructure, a new HunYuan VL model integration, and RADIO model support. They touch different codepaths and represent separate feature additions, not duplicate fixes.", "confidence": 0.97, "canonical_issue_reason": null, - "canonical_pr_reason": "No canonical duplicate PR exists in this cluster; the PRs target different code paths and feature areas.", + "canonical_pr_reason": null, "best_issue_reason": null, - "best_pr_reason": "PR 46417 is the clearer standalone representative because it is a complete model integration with tests/docs, but it is not a duplicate of 46376.", + "best_pr_reason": null, "soft_edge_verdicts": [ { "left": "pull_request:46376", "right": "pull_request:46417", "accept": false, - "reason": "Different underlying changes: continuous batching multimodal/cache/scheduler work vs a HunYuan VL model add-on with docs, auto-mappings, and tests. No shared concrete bug or fix path, so they should not be merged." + "reason": "Different changesets and goals: continuous batching multimodal support vs. adding a new HunYuan VL model package and docs. No shared concrete bug or mergeable duplicate implementation." + }, + { + "left": "pull_request:46376", + "right": "pull_request:46425", + "accept": false, + "reason": "These are unrelated feature PRs: one modifies continuous batching internals, the other adds RADIO model support across model files and auto mappings. They do not fix the same underlying problem." } ] }, diff --git a/analysis/current/manifest.json b/analysis/current/manifest.json index 49e4a14f23cf5a9a2cec242274483810c7fca981..747ae15037201a21cfc9bf523707c0aa0c44247c 100644 --- a/analysis/current/manifest.json +++ b/analysis/current/manifest.json @@ -1,8 +1,8 @@ { - "analysis_id": "hybrid-model-20260604t180024z", + "analysis_id": "hybrid-model-20260605t000034z", "archived_artifacts": { - "hybrid": "snapshots/20260604T180024Z/analysis-runs/hybrid-model-20260604t180024z/analysis-report-hybrid.json", - "hybrid_reviews": "snapshots/20260604T180024Z/analysis-runs/hybrid-model-20260604t180024z/analysis-report-hybrid.llm-reviews.json" + "hybrid": "snapshots/20260605T000034Z/analysis-runs/hybrid-model-20260605t000034z/analysis-report-hybrid.json", + "hybrid_reviews": "snapshots/20260605T000034Z/analysis-runs/hybrid-model-20260605t000034z/analysis-report-hybrid.llm-reviews.json" }, "artifacts": { "hybrid": "analysis/current/analysis-report-hybrid.json", @@ -10,9 +10,9 @@ }, "channel": "canonical", "model": null, - "published_at": "2026-06-04T18:11:32Z", + "published_at": "2026-06-05T00:09:30Z", "repo": "huggingface/transformers", "schema_version": 1, - "snapshot_id": "20260604T180024Z", + "snapshot_id": "20260605T000034Z", "variant": "hybrid" } diff --git a/snapshots/20260605T000034Z/analysis-runs/hybrid-model-20260605t000034z/analysis-report-hybrid.json b/snapshots/20260605T000034Z/analysis-runs/hybrid-model-20260605t000034z/analysis-report-hybrid.json new file mode 100644 index 0000000000000000000000000000000000000000..463256e9b94f2be0a873f7e44e891d1b279da243 --- /dev/null +++ b/snapshots/20260605T000034Z/analysis-runs/hybrid-model-20260605t000034z/analysis-report-hybrid.json @@ -0,0 +1,2403 @@ +{ + "schema_version": "1.0", + "repo": "huggingface/transformers", + "snapshot_id": "20260605T000034Z", + "generated_at": "2026-06-05T00:09:29Z", + "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-45363-3", + "summary": "Cluster of 3 related pull requests linked by soft_similarity.", + "status": "open", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 46339, + "issue_numbers": [], + "pr_numbers": [ + 45363, + 45666, + 46339 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 45363, + "right_pr_number": 45666, + "code_similarity": 0.392, + "size_similarity": 0.298, + "file_overlap": 0.571, + "area_overlap": 0.329, + "patch_similarity": 0.306, + "shared_filenames": [ + "fused_qwen_example.py", + "src/transformers/integrations/__init__.py", + "src/transformers/integrations/hub_kernels.py", + "src/transformers/modeling_utils.py" + ], + "shared_file_areas": [ + { + "filename": "fused_qwen_example.py", + "left_ranges": [ + [ + 1, + 30 + ] + ], + "right_ranges": [ + [ + 1, + 49 + ] + ] + }, + { + "filename": "src/transformers/integrations/__init__.py", + "left_ranges": [ + [ + 67, + 81 + ], + [ + 226, + 240 + ] + ], + "right_ranges": [ + [ + 67, + 73 + ], + [ + 222, + 228 + ] + ] + }, + { + "filename": "src/transformers/integrations/hub_kernels.py", + "left_ranges": [ + [ + 11, + 17 + ], + [ + 22, + 35 + ], + [ + 505, + 716 + ] + ], + "right_ranges": [ + [ + 11, + 53 + ], + [ + 286, + 295 + ], + [ + 521, + 1077 + ] + ] + }, + { + "filename": "src/transformers/modeling_utils.py", + "left_ranges": [ + [ + 3738, + 3747 + ] + ], + "right_ranges": [ + [ + 1330, + 1338 + ], + [ + 3763, + 3789 + ], + [ + 4223, + 4234 + ], + [ + 4628, + 4641 + ] + ] + } + ] + }, + { + "left_pr_number": 45363, + "right_pr_number": 46339, + "code_similarity": 0.506, + "size_similarity": 0.925, + "file_overlap": 0.286, + "area_overlap": 0.545, + "patch_similarity": 0.297, + "shared_filenames": [ + "src/transformers/integrations/hub_kernels.py", + "src/transformers/modeling_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/integrations/hub_kernels.py", + "left_ranges": [ + [ + 11, + 17 + ], + [ + 22, + 35 + ], + [ + 505, + 716 + ] + ], + "right_ranges": [ + [ + 14, + 52 + ], + [ + 285, + 294 + ], + [ + 521, + 685 + ] + ] + }, + { + "filename": "src/transformers/modeling_utils.py", + "left_ranges": [ + [ + 3738, + 3747 + ] + ], + "right_ranges": [ + [ + 1346, + 1354 + ], + [ + 3785, + 3811 + ], + [ + 4245, + 4256 + ], + [ + 4650, + 4663 + ] + ] + } + ] + }, + { + "left_pr_number": 45666, + "right_pr_number": 46339, + "code_similarity": 0.399, + "size_similarity": 0.322, + "file_overlap": 0.667, + "area_overlap": 0.247, + "patch_similarity": 0.318, + "shared_filenames": [ + "src/transformers/core_model_loading.py", + "src/transformers/integrations/hub_kernels.py", + "src/transformers/modeling_utils.py", + "tests/kernels/test_kernels.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/core_model_loading.py", + "left_ranges": [ + [ + 19, + 25 + ], + [ + 80, + 86 + ] + ], + "right_ranges": [ + [ + 19, + 25 + ], + [ + 80, + 86 + ] + ] + }, + { + "filename": "src/transformers/integrations/hub_kernels.py", + "left_ranges": [ + [ + 11, + 53 + ], + [ + 286, + 295 + ], + [ + 521, + 1077 + ] + ], + "right_ranges": [ + [ + 14, + 52 + ], + [ + 285, + 294 + ], + [ + 521, + 685 + ] + ] + }, + { + "filename": "src/transformers/modeling_utils.py", + "left_ranges": [ + [ + 1330, + 1338 + ], + [ + 3763, + 3789 + ], + [ + 4223, + 4234 + ], + [ + 4628, + 4641 + ] + ], + "right_ranges": [ + [ + 1346, + 1354 + ], + [ + 3785, + 3811 + ], + [ + 4245, + 4256 + ], + [ + 4650, + 4663 + ] + ] + }, + { + "filename": "tests/kernels/test_kernels.py", + "left_ranges": [ + [ + 19, + 54 + ], + [ + 228, + 281 + ], + [ + 294, + 317 + ], + [ + 605, + 995 + ] + ], + "right_ranges": [ + [ + 21, + 28 + ], + [ + 33, + 39 + ], + [ + 90, + 103 + ], + [ + 217, + 323 + ], + [ + 336, + 359 + ] + ] + } + ] + } + ] + }, + { + "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-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": 0.797, + "size_similarity": 0.885, + "file_overlap": 0.75, + "area_overlap": 0.831, + "patch_similarity": 0.693, + "shared_filenames": [ + "docs/source/en/_toctree.yml", + "docs/source/en/model_doc/minicpm3.md", + "src/transformers/models/auto/modeling_auto.py", + "src/transformers/models/minicpm3/__init__.py", + "src/transformers/models/minicpm3/configuration_minicpm3.py", + "src/transformers/models/minicpm3/modeling_minicpm3.py", + "src/transformers/models/minicpm3/modular_minicpm3.py", + "tests/models/minicpm3/__init__.py", + "tests/models/minicpm3/test_modeling_minicpm3.py" + ], + "shared_file_areas": [ + { + "filename": "docs/source/en/_toctree.yml", + "left_ranges": [ + [ + 713, + 720 + ] + ], + "right_ranges": [ + [ + 730, + 737 + ] + ] + }, + { + "filename": "docs/source/en/model_doc/minicpm3.md", + "left_ranges": [ + [ + 1, + 67 + ] + ], + "right_ranges": [ + [ + 1, + 45 + ] + ] + }, + { + "filename": "src/transformers/models/auto/modeling_auto.py", + "left_ranges": [ + [ + 285, + 291 + ], + [ + 700, + 706 + ], + [ + 1298, + 1304 + ] + ], + "right_ranges": [ + [ + 720, + 726 + ], + [ + 1331, + 1337 + ] + ] + }, + { + "filename": "src/transformers/models/minicpm3/__init__.py", + "left_ranges": [ + [ + 1, + 29 + ] + ], + "right_ranges": [ + [ + 1, + 29 + ] + ] + }, + { + "filename": "src/transformers/models/minicpm3/configuration_minicpm3.py", + "left_ranges": [ + [ + 1, + 141 + ] + ], + "right_ranges": [ + [ + 1, + 126 + ] + ] + }, + { + "filename": "src/transformers/models/minicpm3/modeling_minicpm3.py", + "left_ranges": [ + [ + 1, + 544 + ] + ], + "right_ranges": [ + [ + 1, + 522 + ] + ] + }, + { + "filename": "src/transformers/models/minicpm3/modular_minicpm3.py", + "left_ranges": [ + [ + 1, + 444 + ] + ], + "right_ranges": [ + [ + 1, + 342 + ] + ] + }, + { + "filename": "tests/models/minicpm3/test_modeling_minicpm3.py", + "left_ranges": [ + [ + 1, + 109 + ] + ], + "right_ranges": [ + [ + 1, + 136 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-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-45561-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #45561.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 45561, + "canonical_pr_number": 45645, + "issue_numbers": [ + 45561 + ], + "pr_numbers": [ + 45639, + 45645 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 45639, + "right_pr_number": 45645, + "code_similarity": 0.494, + "size_similarity": 0.619, + "file_overlap": 0.5, + "area_overlap": 0.434, + "patch_similarity": 0.451, + "shared_filenames": [ + "src/transformers/testing_utils.py", + "tests/utils/test_testing_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/testing_utils.py", + "left_ranges": [ + [ + 3595, + 3602 + ], + [ + 3819, + 3853 + ] + ], + "right_ranges": [ + [ + 3525, + 3558 + ], + [ + 3782, + 3788 + ] + ] + }, + { + "filename": "tests/utils/test_testing_utils.py", + "left_ranges": [ + [ + 1, + 86 + ] + ], + "right_ranges": [ + [ + 1, + 114 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-45081-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #45081.", + "status": "open", + "confidence": 0.8, + "canonical_issue_number": 45081, + "canonical_pr_number": 45317, + "issue_numbers": [ + 45081 + ], + "pr_numbers": [ + 45086, + 45317 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target", + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 45086, + "right_pr_number": 45317, + "code_similarity": 0.589, + "size_similarity": 0.34, + "file_overlap": 0.5, + "area_overlap": 1.0, + "patch_similarity": 0.136, + "shared_filenames": [ + "src/transformers/tokenization_utils_tokenizers.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/tokenization_utils_tokenizers.py", + "left_ranges": [ + [ + 1360, + 1370 + ], + [ + 1374, + 1380 + ] + ], + "right_ranges": [ + [ + 1360, + 1370 + ], + [ + 1374, + 1380 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-43698-3", + "summary": "Cluster of 1 issues and 2 PRs centered on issue #43698.", + "status": "open", + "confidence": 0.75, + "canonical_issue_number": 43698, + "canonical_pr_number": 43779, + "issue_numbers": [ + 43698 + ], + "pr_numbers": [ + 43779, + 43816 + ], + "evidence_types": [ + "closing_reference", + "shared_issue_target" + ], + "pr_comparisons": [ + { + "left_pr_number": 43779, + "right_pr_number": 43816, + "code_similarity": 0.418, + "size_similarity": 0.538, + "file_overlap": 1.0, + "area_overlap": 0.02, + "patch_similarity": 0.02, + "shared_filenames": [ + "src/transformers/integrations/integration_utils.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/integrations/integration_utils.py", + "left_ranges": [ + [ + 2227, + 2241 + ], + [ + 2303, + 2309 + ] + ], + "right_ranges": [ + [ + 2278, + 2291 + ], + [ + 2309, + 2322 + ] + ] + } + ] + } + ] + }, + { + "cluster_id": "cluster-44002-2", + "summary": "Cluster of 2 related pull requests linked by soft_similarity.", + "status": "open", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 44002, + "issue_numbers": [], + "pr_numbers": [ + 44002, + 44029 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 44002, + "right_pr_number": 44029, + "code_similarity": 0.135, + "size_similarity": 0.43, + "file_overlap": 0.0, + "area_overlap": 0.0, + "patch_similarity": 0.329, + "shared_filenames": [], + "shared_file_areas": [] + } + ] + }, + { + "cluster_id": "cluster-44018-2", + "summary": "Cluster of 2 related pull requests linked by soft_similarity.", + "status": "open", + "confidence": 0.5, + "canonical_issue_number": null, + "canonical_pr_number": 44068, + "issue_numbers": [], + "pr_numbers": [ + 44018, + 44068 + ], + "evidence_types": [ + "soft_similarity" + ], + "pr_comparisons": [ + { + "left_pr_number": 44018, + "right_pr_number": 44068, + "code_similarity": 0.766, + "size_similarity": 0.939, + "file_overlap": 1.0, + "area_overlap": 0.425, + "patch_similarity": 0.866, + "shared_filenames": [ + "src/transformers/models/gpt_neo/modeling_gpt_neo.py", + "tests/models/gpt_neo/test_modeling_gpt_neo.py" + ], + "shared_file_areas": [ + { + "filename": "src/transformers/models/gpt_neo/modeling_gpt_neo.py", + "left_ranges": [ + [ + 26, + 46 + ], + [ + 138, + 143 + ], + [ + 182, + 187 + ], + [ + 283, + 295 + ], + [ + 327, + 341 + ], + [ + 348, + 354 + ], + [ + 360, + 369 + ], + [ + 399, + 406 + ], + [ + 411, + 419 + ], + [ + 428, + 433 + ], + [ + 472, + 492 + ], + [ + 507, + 513 + ], + [ + 519, + 528 + ], + [ + 541, + 559 + ], + [ + 562, + 567 + ], + [ + 595, + 601 + ], + [ + 607, + 614 + ], + [ + 627, + 643 + ], + [ + 685, + 690 + ], + [ + 708, + 714 + ], + [ + 720, + 727 + ], + [ + 740, + 757 + ], + [ + 761, + 766 + ], + [ + 780, + 786 + ], + [ + 791, + 798 + ], + [ + 807, + 822 + ], + [ + 840, + 845 + ] + ], + "right_ranges": [ + [ + 26, + 31 + ], + [ + 34, + 44 + ], + [ + 136, + 141 + ], + [ + 180, + 185 + ], + [ + 281, + 293 + ], + [ + 325, + 339 + ], + [ + 346, + 352 + ], + [ + 358, + 367 + ], + [ + 397, + 404 + ], + [ + 409, + 417 + ], + [ + 426, + 434 + ], + [ + 463, + 483 + ], + [ + 498, + 504 + ], + [ + 510, + 519 + ], + [ + 532, + 550 + ], + [ + 553, + 564 + ], + [ + 586, + 592 + ], + [ + 598, + 605 + ], + [ + 618, + 634 + ], + [ + 676, + 688 + ], + [ + 699, + 705 + ], + [ + 711, + 718 + ], + [ + 731, + 748 + ], + [ + 752, + 762 + ], + [ + 771, + 777 + ], + [ + 782, + 789 + ], + [ + 798, + 813 + ], + [ + 831, + 836 + ] + ] + }, + { + "filename": "tests/models/gpt_neo/test_modeling_gpt_neo.py", + "left_ranges": [ + [ + 458, + 464 + ] + ], + "right_ranges": [ + [ + 458, + 464 + ] + ] + } + ] + } + ] + } + ], + "duplicate_issues": [], + "duplicate_prs": [ + { + "cluster_id": "cluster-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-44002-2", + "canonical_pr_number": 44002, + "duplicate_pr_numbers": [ + 44029 + ], + "target_issue_number": null, + "reason": "PRs in cluster-44002-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-44018-2", + "canonical_pr_number": 44068, + "duplicate_pr_numbers": [ + 44018 + ], + "target_issue_number": null, + "reason": "PRs in cluster-44018-2 are treated as duplicates because they share soft_similarity evidence." + }, + { + "cluster_id": "cluster-45081-3", + "canonical_pr_number": 45317, + "duplicate_pr_numbers": [ + 45086 + ], + "target_issue_number": 45081, + "reason": "PRs in cluster-45081-3 are treated as duplicates because they converge on issue #45081 with closing_reference, shared_issue_target, soft_similarity evidence." + }, + { + "cluster_id": "cluster-45363-3", + "canonical_pr_number": 46339, + "duplicate_pr_numbers": [ + 45363, + 45666 + ], + "target_issue_number": null, + "reason": "PRs in cluster-45363-3 are treated as duplicates because they share 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." + } + ], + "best_issue": { + "cluster_id": "cluster-41211-3", + "issue_number": 41211, + "reason": "Issue #41211 is the strongest global issue candidate because it is open, belongs to a cluster with 3 artifacts, and carries 6 discussion comments plus 6 inbound references.", + "score": 148.0 + }, + "best_pr": { + "cluster_id": "cluster-43979-11", + "pr_number": 44007, + "reason": "PR #44007 is the strongest global PR candidate because it is open, belongs to a cluster with 11 artifacts, links to 1 issues, and carries 2 review/discussion events.", + "score": 238.886 + } +} diff --git a/snapshots/20260605T000034Z/analysis-runs/hybrid-model-20260605t000034z/analysis-report-hybrid.llm-reviews.json b/snapshots/20260605T000034Z/analysis-runs/hybrid-model-20260605t000034z/analysis-report-hybrid.llm-reviews.json new file mode 100644 index 0000000000000000000000000000000000000000..e73f5fc11bc4a04eb772b2f241defd7c1caa3be7 --- /dev/null +++ b/snapshots/20260605T000034Z/analysis-runs/hybrid-model-20260605t000034z/analysis-report-hybrid.llm-reviews.json @@ -0,0 +1,21904 @@ +{ + "schema_version": "1.0", + "repo": "huggingface/transformers", + "snapshot_id": "20260605T000034Z", + "generated_at": "2026-06-05T00:09:29Z", + "model": "gpt-5.4-mini?service_tier=flex", + "reviews": [ + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 1, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "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:43296", + "issue:43425", + "issue:43541", + "issue:43572", + "issue:43596", + "issue:43638", + "issue:43716", + "issue:43827", + "issue:43828", + "issue:43866", + "issue:43975", + "issue:44292", + "issue:44322", + "issue:44457", + "issue:44509", + "issue:44512", + "issue:44560", + "issue:44805", + "issue:44918", + "issue:45081", + "issue:45161", + "issue:45237", + "issue:45464", + "issue:45507", + "issue:45685", + "issue:45800", + "issue:45823", + "issue:45859", + "issue:45874", + "issue:45901", + "issue:45902", + "issue:46015", + "issue:46165", + "issue:46294", + "issue:46304", + "issue:46402" + ], + "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:44918|issue:46165", + "issue:44805|issue:45823", + "issue:43638|issue:44805", + "issue:43638|issue:46165", + "issue:43596|issue:44805", + "issue:44805|issue:46294", + "issue:44805|issue:45161", + "issue:44560|issue:46165", + "issue:43596|issue:46165", + "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:45874|issue:46294", + "issue:39290|issue:45081", + "issue:45823|issue:45874", + "issue:43425|issue:44292", + "issue:34928|issue:41720", + "issue:43828|issue:45161", + "issue:43541|issue:43716", + "issue:43828|issue:44292", + "issue:43716|issue:45685", + "issue:44918|issue:45161", + "issue:44292|issue:44918", + "issue:45507|issue:45685", + "issue:43716|issue:44292", + "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:43866|issue:46402", + "issue:45800|issue:46304", + "issue:43296|issue:46304", + "issue:41762|issue:43596", + "issue:39290|issue:41720" + ], + "prepared_review_unit_hash": "50bfea56a64ba632f40d890e975a7f65fd2cc664722718084808f796ad965790", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 64, + "serialized_chars": 46875, + "estimated_input_tokens": 11719, + "estimated_eval_tokens": 23694 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a mixed cluster of unrelated, model-specific runtime/docs issues. Only one clear duplicate pair stands out: the two ZeRO-3/BERT IndexError reports. The rest share broad subsystems (DeepSpeed, Qwen, docs, dtype/device issues) but not the same concrete bug.", + "confidence": 0.77, + "canonical_issue_reason": "issue:43596 is the best canonical issue for the only clear duplicate pair: it predates issue:43638 and describes the same ZeRO-3/BERT IndexError failure more cleanly and broadly.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43596 is also the best overall issue to keep: earliest, concise, and the clearest representative of the shared concrete failure.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44322", + "right": "issue:45464", + "accept": false, + "reason": "Different failures: missing config attribute vs streaming chat/completions error." + }, + { + "left": "issue:45237", + "right": "issue:45685", + "accept": false, + "reason": "Different hardware/backend problems on different models; only broad runtime similarity." + }, + { + "left": "issue:43716", + "right": "issue:45237", + "accept": false, + "reason": "Image preprocessor dtype mismatch is not the same as AMD GPU execution failure." + }, + { + "left": "issue:43828", + "right": "issue:45237", + "accept": false, + "reason": "Both are dtype/runtime issues, but on different models and code paths." + }, + { + "left": "issue:45237", + "right": "issue:45507", + "accept": false, + "reason": "Different model/backend failures; not the same concrete bug." + }, + { + "left": "issue:44292", + "right": "issue:45237", + "accept": false, + "reason": "Different Qwen/GPT-OSS inference failures with no concrete shared code path." + }, + { + "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": "Export-to-ONNX bug vs FP8 model failure; different issues." + }, + { + "left": "issue:35545", + "right": "issue:41720", + "accept": false, + "reason": "Different model/runtime failures; no shared concrete bug." + }, + { + "left": "issue:45901", + "right": "issue:46015", + "accept": false, + "reason": "Table QA crash and outdated RAG docs are unrelated." + }, + { + "left": "issue:44509", + "right": "issue:44512", + "accept": false, + "reason": "Both are v5 docs cleanup items, but they reference different removed docs content." + }, + { + "left": "issue:35545", + "right": "issue:36010", + "accept": false, + "reason": "ONNX export error and GenerationMixin import error are different failures." + }, + { + "left": "issue:44918", + "right": "issue:46165", + "accept": false, + "reason": "TRL SFT embedding unpacking and training-resume/trust-remote-code are different problems." + }, + { + "left": "issue:44805", + "right": "issue:45823", + "accept": false, + "reason": "Mask/index shape mismatch is not the same as Gemma4 device-map mismatch." + }, + { + "left": "issue:43638", + "right": "issue:44805", + "accept": false, + "reason": "Both are shape/index errors, but the model/setup and failure path differ." + }, + { + "left": "issue:43596", + "right": "issue:44805", + "accept": false, + "reason": "Different error signatures and code paths; not the same bug." + }, + { + "left": "issue:44805", + "right": "issue:46294", + "accept": false, + "reason": "Mask shape mismatch is unrelated to video token/template mismatch." + }, + { + "left": "issue:44805", + "right": "issue:45161", + "accept": false, + "reason": "Different model features and failure mechanisms." + }, + { + "left": "issue:44560", + "right": "issue:46165", + "accept": false, + "reason": "Video apply_chat_template failure and training resume issue are unrelated." + }, + { + "left": "issue:43596", + "right": "issue:46165", + "accept": false, + "reason": "Different training/setup issue; no concrete overlap beyond general model loading." + }, + { + "left": "issue:43572", + "right": "issue:46165", + "accept": false, + "reason": "StableLmConfig pad token regression is unrelated to training-resume behavior." + }, + { + "left": "issue:34634", + "right": "issue:35545", + "accept": false, + "reason": "BarkProcessor voice preset bug and ONNX export error are unrelated." + }, + { + "left": "issue:30277", + "right": "issue:43638", + "accept": false, + "reason": "Both mention DeepSpeed, but the failures are different and model-specific." + }, + { + "left": "issue:43596", + "right": "issue:43638", + "accept": true, + "reason": "Same ZeRO-3/BERT IndexError during model init/loading; same concrete failure path, with 43596 being the cleaner canonical report." + }, + { + "left": "issue:43039", + "right": "issue:45161", + "accept": false, + "reason": "Liger cross_entropy routing and GPT-OSS tensor-parallel failure are different bugs." + }, + { + "left": "issue:43975", + "right": "issue:44457", + "accept": false, + "reason": "Detokenization regression and LoRA merge/save/load inconsistency are different issues." + }, + { + "left": "issue:44292", + "right": "issue:44457", + "accept": false, + "reason": "Qwen-3 NVFP4 runtime failure and LoRA save/load mismatch are unrelated." + }, + { + "left": "issue:34928", + "right": "issue:44805", + "accept": false, + "reason": "Activation-checkpoint/FSDP tensor-size mismatch is a different bug from mask/index errors." + }, + { + "left": "issue:24643", + "right": "issue:30277", + "accept": false, + "reason": "Different DeepSpeed failures: 2-D weight error vs collective mismatch." + }, + { + "left": "issue:45859", + "right": "issue:45902", + "accept": false, + "reason": "Same Qwen3.5-MoE rotary-embedding area, but CPU-offload incompatibility and meta\u2192to_empty uninitialized-memory are not clearly the same bug." + }, + { + "left": "issue:43638", + "right": "issue:45161", + "accept": false, + "reason": "BERT/ZeRO-3 loading issue and GPT-OSS tensor-parallel issue are unrelated." + }, + { + "left": "issue:34928", + "right": "issue:36331", + "accept": false, + "reason": "Activation checkpointing FSDP mismatch and custom compute_loss signature error are unrelated." + }, + { + "left": "issue:43596", + "right": "issue:45161", + "accept": false, + "reason": "Different models and failure mechanisms; no shared bug." + }, + { + "left": "issue:43716", + "right": "issue:45507", + "accept": false, + "reason": "Different model families and distinct runtime errors." + }, + { + "left": "issue:45874", + "right": "issue:46294", + "accept": false, + "reason": "Input-embeds reverse lookup cost and video token/template mismatch are unrelated." + }, + { + "left": "issue:39290", + "right": "issue:45081", + "accept": false, + "reason": "Gemma3 vLLM config regression and Mistral regex patch crash are different bugs." + }, + { + "left": "issue:45823", + "right": "issue:45874", + "accept": false, + "reason": "Device-map mismatch and expensive reverse embedding lookup are different forward-path issues." + }, + { + "left": "issue:43425", + "right": "issue:44292", + "accept": false, + "reason": "Torch/version compatibility issue and Qwen NVFP4 runtime failure are unrelated." + }, + { + "left": "issue:34928", + "right": "issue:41720", + "accept": false, + "reason": "FSDP activation-checkpoint mismatch and Qwen3 auto device-map assert are different bugs." + }, + { + "left": "issue:43828", + "right": "issue:45161", + "accept": false, + "reason": "Phi dtype/autocast issue and GPT-OSS tensor-parallel failure are unrelated." + }, + { + "left": "issue:30277", + "right": "issue:36331", + "accept": false, + "reason": "DeepSpeed collective mismatch and compute_loss signature mismatch are different code paths." + }, + { + "left": "issue:34928", + "right": "issue:36010", + "accept": false, + "reason": "FSDP recomputed-size mismatch and GenerationMixin import error are unrelated." + }, + { + "left": "issue:45237", + "right": "issue:46015", + "accept": false, + "reason": "GPU/backend runtime failure and outdated RAG docs are unrelated." + }, + { + "left": "issue:43039", + "right": "issue:44805", + "accept": false, + "reason": "Liger kernel cross-entropy routing and tensor mask shape mismatch are different issues." + }, + { + "left": "issue:43866", + "right": "issue:46402", + "accept": false, + "reason": "Corrupted checkpoint and missing text-encoder weights are not the same bug." + }, + { + "left": "issue:45800", + "right": "issue:46304", + "accept": false, + "reason": "General torch/transformers incompatibility and USE_DISTRIBUTED=0 regression are different compatibility issues." + }, + { + "left": "issue:43296", + "right": "issue:46304", + "accept": false, + "reason": "PaddleOCR-VL/vLLM load failure and distributed-build regression are unrelated." + }, + { + "left": "issue:41762", + "right": "issue:43596", + "accept": false, + "reason": "Same broad ZeRO-3 IndexError theme, but different model family and report context; not clearly the same bug." + }, + { + "left": "issue:39290", + "right": "issue:41720", + "accept": false, + "reason": "Gemma3 vLLM config regression and Qwen3 auto device-map assert are unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The only accepted duplicate pair, issue:43596 vs issue:43638, is well supported by the shared ZeRO-3/BERT IndexError wording, and the remaining soft-edge rejections do not overclaim sameness." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 2, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:36010", + "issue:39290", + "issue:41093", + "issue:41762", + "issue:43039", + "issue:43296", + "issue:43366", + "issue:43404", + "issue:43425", + "issue:43531", + "issue:43541", + "issue:43572", + "issue:43596", + "issue:43638", + "issue:43792", + "issue:43828", + "issue:43854", + "issue:43866", + "issue:43872", + "issue:43901", + "issue:43975", + "issue:44291", + "issue:44292", + "issue:44387", + "issue:44512", + "issue:44560", + "issue:44589", + "issue:44805", + "issue:44841", + "issue:44863", + "issue:44918", + "issue:45005", + "issue:45020", + "issue:45070", + "issue:45084", + "issue:45161", + "issue:45237", + "issue:45362", + "issue:45464", + "issue:45685", + "issue:45800", + "issue:45812", + "issue:45823", + "issue:45901", + "issue:45920", + "issue:46015", + "issue:46294", + "issue:46304" + ], + "soft_pairs": [ + "issue:44589|issue:46304", + "issue:45823|issue:46294", + "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:45020|issue:46304", + "issue:43866|issue:44863", + "issue:43872|issue:46304", + "issue:44292|issue:46015", + "issue:39290|issue:43531", + "issue:41093|issue:43638", + "issue:43792|issue:45901", + "issue:43854|issue:43866", + "issue:43425|issue:45070", + "issue: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:43541|issue:45685", + "issue:43404|issue:43866", + "issue:43541|issue:43828", + "issue:44841|issue:45084", + "issue:36010|issue:39290", + "issue:45800|issue:45823", + "issue:41093|issue:41762", + "issue:43296|issue:43366", + "issue:43572|issue:45070", + "issue:44805|issue:44918", + "issue:44387|issue:45005", + "issue:43541|issue:45161", + "issue:43828|issue:45464", + "issue:44291|issue:45070", + "issue:43366|issue:45464", + "issue:43531|issue:45070" + ], + "prepared_review_unit_hash": "1309680ea656fd2dc8d79781e0ee3f14e32da7737c8a10662811bad205f7887b", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 41, + "serialized_chars": 39796, + "estimated_input_tokens": 9949, + "estimated_eval_tokens": 20154 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a heterogeneous cluster of mostly unrelated Transformers issues. The soft edges are largely superficial similarities (same model family, same generic error class, or same subsystem) rather than the same underlying bug/change, so none should be merged as duplicates.", + "confidence": 0.79, + "canonical_issue_reason": "If one issue must be treated as the representative, issue #45812 is the broadest v5 regression here because it affects multiple model families and has the clearest cross-cutting user impact. That said, the cluster does not form a single duplicate family.", + "canonical_pr_reason": null, + "best_issue_reason": "#45812 is the best overall issue representative because it is the widest-scope regression in the set and most plausibly central to the cluster theme, even though it should not absorb the others as duplicates.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44589", + "right": "issue:46304", + "accept": false, + "reason": "Different failures: float8 storage lookup vs distributed-pytorch support. Same broad v5 timing, but not the same bug." + }, + { + "left": "issue:45823", + "right": "issue:46294", + "accept": false, + "reason": "Both involve Qwen3.5/Gemma4-era multimodal behavior, but one is a PLE device-mismatch forward bug and the other is a video chat-template/token mismatch." + }, + { + "left": "issue:43901", + "right": "issue:44512", + "accept": false, + "reason": "Both are docs-related, but they concern different API surfaces and different kinds of stale documentation." + }, + { + "left": "issue:45237", + "right": "issue:45901", + "accept": false, + "reason": "Both crashy, but on different tasks/models and with different root symptoms; too generic to be the same bug." + }, + { + "left": "issue:44841", + "right": "issue:45823", + "accept": false, + "reason": "Different model families and failures: Voxtral processor issue versus Gemma4 PLE device mismatch." + }, + { + "left": "issue:43039", + "right": "issue:44560", + "accept": false, + "reason": "Liger cross-entropy invocation and Qwen3.5 video StopIteration are unrelated beyond both being runtime errors." + }, + { + "left": "issue:43975", + "right": "issue:46015", + "accept": false, + "reason": "DeepSeek detokenization regression is a model-specific bug; the other is generic outdated RAG examples docs work." + }, + { + "left": "issue:45812", + "right": "issue:45920", + "accept": false, + "reason": "Both are tokenizer regressions, but they affect different model sets and likely different code paths; not enough to unify." + }, + { + "left": "issue:44863", + "right": "issue:46015", + "accept": false, + "reason": "NemotronH checkpoint loading and RAG docs are entirely unrelated." + }, + { + "left": "issue:45020", + "right": "issue:46304", + "accept": false, + "reason": "Remote-code breakage and distributed build/runtime support are distinct compatibility issues." + }, + { + "left": "issue:43866", + "right": "issue:44863", + "accept": false, + "reason": "Corrupted Ovis2 checkpoint versus NemotronH load failure are different model/loading problems." + }, + { + "left": "issue:43872", + "right": "issue:46304", + "accept": false, + "reason": "bitsandbytes initializer incompatibility is unrelated to USE_DISTRIBUTED=0 PyTorch support." + }, + { + "left": "issue:44292", + "right": "issue:46015", + "accept": false, + "reason": "Qwen-3-8B-NVFP4 runtime failure and RAG docs are unrelated." + }, + { + "left": "issue:39290", + "right": "issue:43531", + "accept": false, + "reason": "Both mention sliding_window, but one is a missing config attribute under vLLM and the other is a Qwen3-MoE sliding_window behavior issue." + }, + { + "left": "issue:41093", + "right": "issue:43638", + "accept": false, + "reason": "Both are mask/index errors with ZeRO-3 context, but one is a shape mismatch and the other is out-of-bounds during BertModel init; likely different causes." + }, + { + "left": "issue:43792", + "right": "issue:45901", + "accept": false, + "reason": "Whisper-large-v2 runtime failure and table-question-answering crash do not point to the same code path." + }, + { + "left": "issue:43854", + "right": "issue:43866", + "accept": false, + "reason": "Model loading in unit tests versus a corrupted checkpoint are different failure modes." + }, + { + "left": "issue:43425", + "right": "issue:45070", + "accept": false, + "reason": "Torch 2.10 incompatibility and a v5.4.0 pydantic PretrainedConfig regression are distinct versioning issues." + }, + { + "left": "issue:43596", + "right": "issue:45800", + "accept": false, + "reason": "Both mention ZeRO-3-style init/version problems, but one is BertModel index failure and the other is torch 2.4.1 vs transformers 5.8.0 incompatibility." + }, + { + "left": "issue:43975", + "right": "issue:45901", + "accept": false, + "reason": "Different model/task contexts and different symptom classes; no shared bug evidence." + }, + { + "left": "issue:43792", + "right": "issue:45070", + "accept": false, + "reason": "Whisper loading/runtime issue versus pydantic PretrainedConfig regression are not the same problem." + }, + { + "left": "issue:45362", + "right": "issue:45464", + "accept": false, + "reason": "Both involve Qwen3.5, but one is a chat crash and the other is a streaming inference API failure; not enough to merge." + }, + { + "left": "issue:44292", + "right": "issue:45901", + "accept": false, + "reason": "Different model/application areas and no concrete shared failure mechanism." + }, + { + "left": "issue:39290", + "right": "issue:45823", + "accept": false, + "reason": "Shared version era only; Gemma3 sliding_window_pattern attribute error is unrelated to Gemma4 PLE device placement." + }, + { + "left": "issue:45464", + "right": "issue:45685", + "accept": false, + "reason": "Streaming chat/completions failure and MPS histogram int error are unrelated." + }, + { + "left": "issue:43541", + "right": "issue:45685", + "accept": false, + "reason": "Torch dynamo grouped_mm tracing issue versus MPS int histogram error are different backends and code paths." + }, + { + "left": "issue:43404", + "right": "issue:43866", + "accept": false, + "reason": "Tied lm_head weights in Mistral3 is a model implementation bug; Ovis2 corrupted checkpoint is a load/data issue." + }, + { + "left": "issue:43541", + "right": "issue:43828", + "accept": false, + "reason": "Both are MoE-related, but one is a grouped_mm tracing failure and the other is an autocast dtype mismatch; different underlying bugs." + }, + { + "left": "issue:44841", + "right": "issue:45084", + "accept": false, + "reason": "Processor failure for Voxtral and template-node compilation errors are unrelated." + }, + { + "left": "issue:36010", + "right": "issue:39290", + "accept": false, + "reason": "ImportError for GenerationMixin and Gemma3 sliding_window_pattern error are separate API regressions." + }, + { + "left": "issue:45800", + "right": "issue:45823", + "accept": false, + "reason": "Torch/transformers version incompatibility and Gemma4 device mismatch are distinct bugs." + }, + { + "left": "issue:41093", + "right": "issue:41762", + "accept": false, + "reason": "Both are index errors with ZeRO-3, but one is a mask/tensor length mismatch and the other is zero-sized tensor access while loading Gemma3ForConditionalGeneration." + }, + { + "left": "issue:43296", + "right": "issue:43366", + "accept": false, + "reason": "PaddleOCR-VL loading in vLLM and GGUF gpt-oss support are different model-compatibility requests." + }, + { + "left": "issue:43572", + "right": "issue:45070", + "accept": false, + "reason": "Missing pad_token_idx in StableLmConfig is a model config regression; pydantic PretrainedConfig breakage is a framework-level API issue." + }, + { + "left": "issue:44805", + "right": "issue:44918", + "accept": false, + "reason": "Both are tensor/embedding shape problems, but one is a mask indexing mismatch and the other is unpacking Qwen3.5 input embeddings with TRL SFT trainer." + }, + { + "left": "issue:44387", + "right": "issue:45005", + "accept": false, + "reason": "Int4 OOM regression and tied-weights translation-model issues are different change sets and failure modes." + }, + { + "left": "issue:43541", + "right": "issue:45161", + "accept": false, + "reason": "Grouped_mm tracing crash and GPT-OSS only-TP-not-working are separate MoE/tracing versus parallelism issues." + }, + { + "left": "issue:43828", + "right": "issue:45464", + "accept": false, + "reason": "Autocast dtype mismatch for Phi-tiny-MoE and Qwen3.5 streaming API failure are unrelated." + }, + { + "left": "issue:44291", + "right": "issue:45070", + "accept": false, + "reason": "init_empty_weights/_is_hf_initialized TypeError and pydantic PretrainedConfig regression are related only by broad v5 breakage, not by code path." + }, + { + "left": "issue:43366", + "right": "issue:45464", + "accept": false, + "reason": "GGUF gpt-oss support and Qwen3.5 streaming failure are different features and bugs." + }, + { + "left": "issue:43531", + "right": "issue:45070", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior and pydantic model-field regression are not the same issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 3, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:36331", + "issue:38175", + "issue:39186", + "issue:39692", + "issue:40990", + "issue:41093", + "issue:41762", + "issue:42915", + "issue:43054", + "issue:43404", + "issue:43541", + "issue:43596", + "issue:43644", + "issue:43645", + "issue:43716", + "issue:43828", + "issue:43866", + "issue:43872", + "issue:43873", + "issue:43950", + "issue:43975", + "issue:43994", + "issue:44292", + "issue:44360", + "issue:44403", + "issue:44479", + "issue:44492", + "issue:44512", + "issue:44530", + "issue:44534", + "issue:44589", + "issue:44863", + "issue:44918", + "issue:44928", + "issue:44960", + "issue:45005", + "issue:45020", + "issue:45399", + "issue:45507", + "issue:45542", + "issue:45632", + "issue:45698", + "issue:45701", + "issue:45759", + "issue:45800", + "issue:45902", + "issue:46294" + ], + "soft_pairs": [ + "issue:43716|issue:46294", + "issue:45020|issue:45800", + "issue:43541|issue:44589", + "issue:44479|issue:46294", + "issue:41093|issue:43596", + "issue:43054|issue:43994", + "issue:45759|issue:45902", + "issue:45542|issue:45800", + "issue:39186|issue:44918", + "issue:43716|issue:44918", + "issue:40990|issue:44960", + "issue:45632|issue:45698", + "issue:43645|issue:45399", + "issue:43404|issue:45005", + "issue:42915|issue:45399", + "issue:44530|issue:45005", + "issue:43873|issue:45005", + "issue: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:39692|issue:43054", + "issue:43644|issue:43950", + "issue:43828|issue:44928", + "issue:36331|issue:41762", + "issue:43872|issue:44589", + "issue:43950|issue:45902", + "issue:43054|issue:45701", + "issue:43866|issue:44403", + "issue:38175|issue:43054" + ], + "prepared_review_unit_hash": "d63fea1f3dee2113b0e6c092a44ac5c63fd28c3373d7891f3f00f12e64d6b20f", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 33, + "serialized_chars": 36897, + "estimated_input_tokens": 9225, + "estimated_eval_tokens": 18706 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Only the non-persistent-buffer corruption chain looks like a clear duplicate set. The SigLIP2, Qwen video, custom-module caching, and other same-subsystem pairs are too weakly evidenced from titles alone to merge conservatively.", + "confidence": 0.84, + "canonical_issue_reason": "issue:43950 is the clearest duplicate hub: it states the non-persistent-buffer corruption regression directly and sits between the exact/near-exact duplicate phrasing in issue:43644 and issue:44534.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43950 is the best global issue candidate because it most precisely captures a concrete regression with the strongest duplicate evidence and a clear underlying fix target.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43716", + "right": "issue:46294", + "accept": false, + "reason": "Different model families and failure modes; only a broad multimodal theme overlaps." + }, + { + "left": "issue:45020", + "right": "issue:45800", + "accept": false, + "reason": "One is a remote_code regression; the other is a torch/transformers version incompatibility." + }, + { + "left": "issue:43541", + "right": "issue:44589", + "accept": false, + "reason": "Different runtime failures in unrelated code paths." + }, + { + "left": "issue:44479", + "right": "issue:46294", + "accept": false, + "reason": "Both involve Qwen video inputs, but the packet does not show the same concrete bug or a duplicate chain." + }, + { + "left": "issue:41093", + "right": "issue:43596", + "accept": false, + "reason": "Both are index errors, but the shapes, triggers, and model contexts differ." + }, + { + "left": "issue:43054", + "right": "issue:43994", + "accept": true, + "reason": "Both report the same SigLIP2 bad-output regression: degraded/nonsensical text-embedding behavior for google/siglip2-base-patch16-224." + }, + { + "left": "issue:45759", + "right": "issue:45902", + "accept": false, + "reason": "Different composite-config loading bug vs buffer-materialization bug." + }, + { + "left": "issue:45542", + "right": "issue:45800", + "accept": false, + "reason": "TensorBoard/TensorFlow backend issue is unrelated to version compatibility." + }, + { + "left": "issue:39186", + "right": "issue:44918", + "accept": false, + "reason": "Different model setups and different concrete failures." + }, + { + "left": "issue:43716", + "right": "issue:44918", + "accept": false, + "reason": "Different model families and unrelated dtype vs embedding-unpack failures." + }, + { + "left": "issue:40990", + "right": "issue:44960", + "accept": false, + "reason": "Completely different models and symptoms; only the general training/eval context overlaps." + }, + { + "left": "issue:45632", + "right": "issue:45698", + "accept": false, + "reason": "Cache-path collision and wrong custom module after save/load are related, but not the same bug from the packet alone." + }, + { + "left": "issue:43645", + "right": "issue:45399", + "accept": false, + "reason": "Notebook model initialization regression is unrelated to flash-attn fallback gating." + }, + { + "left": "issue:43404", + "right": "issue:45005", + "accept": false, + "reason": "Both mention tied weights, but the concrete model classes and bugs differ." + }, + { + "left": "issue:42915", + "right": "issue:45399", + "accept": false, + "reason": "FP8 config failure is unrelated to flash-attn fallback checks." + }, + { + "left": "issue:44530", + "right": "issue:45005", + "accept": false, + "reason": "PagedAttention cache crash is not the same as a tied-weights loading regression." + }, + { + "left": "issue:43873", + "right": "issue:45005", + "accept": false, + "reason": "Quantization/offloading behavior is a different problem from tied-weight loading." + }, + { + "left": "issue:44863", + "right": "issue:45005", + "accept": false, + "reason": "NemotronH checkpoint loading is a separate model-specific bug." + }, + { + "left": "issue:44360", + "right": "issue:44512", + "accept": false, + "reason": "Indexer discussion is unrelated to a docs wording issue." + }, + { + "left": "issue:44918", + "right": "issue:45507", + "accept": false, + "reason": "Different models and failures; no shared code-path bug is evident." + }, + { + "left": "issue:44492", + "right": "issue:44512", + "accept": false, + "reason": "Separate documentation/wording issues." + }, + { + "left": "issue:43975", + "right": "issue:44292", + "accept": false, + "reason": "Different models and symptoms; no concrete shared bug is shown." + }, + { + "left": "issue:44292", + "right": "issue:45507", + "accept": false, + "reason": "Different model-loading/runtime issues." + }, + { + "left": "issue:43950", + "right": "issue:44534", + "accept": true, + "reason": "Same non-persistent-buffer corruption regression, described with different wording." + }, + { + "left": "issue:39692", + "right": "issue:43054", + "accept": false, + "reason": "Docs-example errors and SigLIP2 embedding-quality regressions are related only thematically." + }, + { + "left": "issue:43644", + "right": "issue:43950", + "accept": true, + "reason": "Exact same non-persistent-buffer junk/corruption regression." + }, + { + "left": "issue:43828", + "right": "issue:44928", + "accept": false, + "reason": "Both involve numeric instability, but the trigger and failure mechanism differ." + }, + { + "left": "issue:36331", + "right": "issue:41762", + "accept": false, + "reason": "Trainer API signature mismatch and ZeRO-3 loading error are unrelated." + }, + { + "left": "issue:43872", + "right": "issue:44589", + "accept": false, + "reason": "Different quantization/storage incompatibilities." + }, + { + "left": "issue:43950", + "right": "issue:45902", + "accept": false, + "reason": "Both touch non-persistent buffers, but the triggering paths differ enough to keep separate." + }, + { + "left": "issue:43054", + "right": "issue:45701", + "accept": false, + "reason": "SigLIP2 output quality and tokenization-version changes are distinct issues." + }, + { + "left": "issue:43866", + "right": "issue:44403", + "accept": false, + "reason": "Checkpoint corruption is unrelated to loading noise/warnings." + }, + { + "left": "issue:38175", + "right": "issue:43054", + "accept": false, + "reason": "Same model family, but the packet does not demonstrate the same underlying bug." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The summary is mostly grounded, but a few soft-edge accepts overstate the evidence. In particular, `issue:45632` vs `issue:45698` is not clearly the same bug from the packet alone (cache-path collision vs wrong custom module after save/load), and `issue:44479` vs `issue:46294` is also only title-level similarity rather than a demonstrated duplicate chain. `issue:38175` vs `issue:43054` is plausible, but still weakly supported. Because the instructions call for conservative soft-edge verdicts, I can\u2019t accept this as-is." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 4, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:41553", + "issue:41762", + "issue:42491", + "issue:42915", + "issue:42947", + "issue:43054", + "issue:43257", + "issue:43404", + "issue:43493", + "issue:43643", + "issue:43742", + "issue:43782", + "issue:43792", + "issue:43824", + "issue:43827", + "issue:43828", + "issue:43856", + "issue:43872", + "issue:43881", + "issue:43883", + "issue:43950", + "issue:43975", + "issue:44220", + "issue:44291", + "issue:44368", + "issue:44387", + "issue:44403", + "issue:44451", + "issue:44488", + "issue:44509", + "issue:44661", + "issue:44910", + "issue:44960", + "issue:44991", + "issue:45020", + "issue:45081", + "issue:45305", + "issue:45356", + "issue:45399", + "issue:45520", + "issue:45663", + "issue:45685", + "issue:45715", + "issue:45823", + "issue:45859", + "issue:45902", + "issue:46315" + ], + "soft_pairs": [ + "issue:41553|issue:43883", + "issue:45663|issue:45902", + "issue:44910|issue:45859", + "issue:42947|issue:44387", + "issue:43054|issue:43493", + "issue:41553|issue:43881", + "issue:44403|issue:44451", + "issue:43975|issue:45520", + "issue:44403|issue:44488", + "issue:43643|issue:45020", + "issue:42947|issue:43856", + "issue:41762|issue:42915", + "issue:45685|issue:45823", + "issue:43782|issue:44960", + "issue:43824|issue:43872", + "issue:44991|issue:45399", + "issue:43881|issue:43883", + "issue:45020|issue:45823", + "issue:42491|issue:44368", + "issue:44661|issue:45020", + "issue:43828|issue:45823", + "issue:45081|issue:45399", + "issue:45715|issue:46315", + "issue:45020|issue:45356", + "issue:43742|issue:45399", + "issue:43404|issue:43950", + "issue:43827|issue:44509", + "issue:43872|issue:44291", + "issue:42915|issue:43872", + "issue:43792|issue:44220", + "issue:43257|issue:45305" + ], + "prepared_review_unit_hash": "d84ed3afc9f7cb676581f522bf8a3bc3d04106e9783cc33999f079ab2edc2605", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 31, + "serialized_chars": 36795, + "estimated_input_tokens": 9199, + "estimated_eval_tokens": 18654 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most items are separate model-specific loading or training regressions, not duplicates. Only two pairs look like clear duplicates: the v5 pipeline-docs cleanup and the `_is_hf_initialized` constructor incompatibility.", + "confidence": 0.76, + "canonical_issue_reason": "Issue 45020 is the broadest representative of the cluster\u2019s main theme: recent Transformers v5 loading regressions affecting remote-code/model loading across multiple models.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45020 is the strongest single anchor for triage because it is cross-cutting, clearly describes a user-facing regression, and is broad enough to represent the surrounding loading-breakage reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:41553", + "right": "issue:43883", + "accept": false, + "reason": "Different models and different failure modes: tokenizer error messaging vs missing tied-weights attribute on Molmo." + }, + { + "left": "issue:45663", + "right": "issue:45902", + "accept": false, + "reason": "Different models and code paths: Gemma4 FSDP2 shared_kv_states bug vs Qwen3_5MoE rotary embedding materialization bug." + }, + { + "left": "issue:44910", + "right": "issue:45859", + "accept": false, + "reason": "Both are Qwen3-family issues, but one is Flash Attention position_ids handling and the other is CPU-offload compatibility in rotary embedding." + }, + { + "left": "issue:42947", + "right": "issue:44387", + "accept": false, + "reason": "Different regressions: PEFT LoRA gradient checkpointing vs int4 quantization memory growth." + }, + { + "left": "issue:43054", + "right": "issue:43493", + "accept": false, + "reason": "Related to SigLIP2, but one is a quality complaint and the other is an HF-vs-JAX implementation discrepancy; not clearly the same bug." + }, + { + "left": "issue:41553", + "right": "issue:43881", + "accept": false, + "reason": "Different model-specific loading problems with no shared code-path evidence." + }, + { + "left": "issue:44403", + "right": "issue:44451", + "accept": false, + "reason": "One is a noisy-warning complaint, the other is a concrete model load failure." + }, + { + "left": "issue:43975", + "right": "issue:45520", + "accept": false, + "reason": "Different subsystems: tokenizer detokenization regression vs import_utils flash-attn KeyError." + }, + { + "left": "issue:44403", + "right": "issue:44488", + "accept": false, + "reason": "Same general loading area, but one is noise and the other is a model-specific load failure." + }, + { + "left": "issue:43643", + "right": "issue:45020", + "accept": false, + "reason": "Both involve remote-code loading, but 43643 is a specific missing-fields bug in AutoConfig while 45020 is a broader compatibility report across models." + }, + { + "left": "issue:42947", + "right": "issue:43856", + "accept": false, + "reason": "Different problems: LoRA/gradient-checkpointing behavior vs Qwen3 MoE training memory efficiency." + }, + { + "left": "issue:41762", + "right": "issue:42915", + "accept": false, + "reason": "Both are model-loading issues, but Gemma3 ZeRO-3 index error and Qwen3 MoE FP8 failure are unrelated code paths." + }, + { + "left": "issue:45685", + "right": "issue:45823", + "accept": false, + "reason": "Different device/runtime issues: MPS histogram op limitation vs Gemma4 PLE device mismatch." + }, + { + "left": "issue:43782", + "right": "issue:44960", + "accept": false, + "reason": "Different models and symptoms; no evidence they share the same underlying bug." + }, + { + "left": "issue:43824", + "right": "issue:43872", + "accept": false, + "reason": "Different incompatibilities: missing Qwen2_5_VL import vs bitsandbytes `_is_hf_initialized` constructor error." + }, + { + "left": "issue:44991", + "right": "issue:45399", + "accept": false, + "reason": "Tokenizer load failure and flash-attn fallback logic are unrelated." + }, + { + "left": "issue:43881", + "right": "issue:43883", + "accept": false, + "reason": "Different model failures: GLM-4V loading vs Molmo tied-weights attribute error." + }, + { + "left": "issue:45020", + "right": "issue:45823", + "accept": false, + "reason": "Remote-code loading regressions and Gemma4 PLE device placement are unrelated." + }, + { + "left": "issue:42491", + "right": "issue:44368", + "accept": false, + "reason": "Both mention Qwen MoE/LoRA, but one is a version-compatibility failure and the other is a tied-embeddings warning." + }, + { + "left": "issue:44661", + "right": "issue:45020", + "accept": false, + "reason": "A tokenizer-mapping edge case is not the same bug as remote-code breakage." + }, + { + "left": "issue:43828", + "right": "issue:45823", + "accept": false, + "reason": "Different runtime failures: autocast dtype mismatch vs device_map placement mismatch." + }, + { + "left": "issue:45081", + "right": "issue:45399", + "accept": false, + "reason": "Tokenizer regex patch crash and flash-attn fallback checks are separate issues." + }, + { + "left": "issue:45715", + "right": "issue:46315", + "accept": false, + "reason": "Both are tokenizer performance regressions, but they affect opposite conversion directions and different hot paths." + }, + { + "left": "issue:45020", + "right": "issue:45356", + "accept": false, + "reason": "General remote-code breakage is not the same as the Kimi-K2.5 tokenizer codec regression." + }, + { + "left": "issue:43742", + "right": "issue:45399", + "accept": false, + "reason": "KeyError loading MobileLLM and flash-attn fallback gating are unrelated." + }, + { + "left": "issue:43404", + "right": "issue:43950", + "accept": false, + "reason": "lm_head tying in Mistral3 is unrelated to silent corruption of non-persistent buffers." + }, + { + "left": "issue:43827", + "right": "issue:44509", + "accept": true, + "reason": "Same docs cleanup: both note pipeline tasks removed in v5 and still referenced in documentation." + }, + { + "left": "issue:43872", + "right": "issue:44291", + "accept": true, + "reason": "Same underlying `_is_hf_initialized` incompatibility during loading/init_empty_weights, described from two angles." + }, + { + "left": "issue:42915", + "right": "issue:43872", + "accept": false, + "reason": "Different Qwen/DeepSpeed and bitsandbytes incompatibility bugs." + }, + { + "left": "issue:43792", + "right": "issue:44220", + "accept": false, + "reason": "The Whisper load failure is too generic to confirm it is the same bug as the fbank-feature helper issue." + }, + { + "left": "issue:43257", + "right": "issue:45305", + "accept": false, + "reason": "Both involve DeepSpeed/Qwen3-ish training, but one is weight conversion and the other is gradient averaging with GAS." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well-grounded in the packet and stays conservative on the soft-edge calls. The two accepted pairs are plausibly the same bugfix/doc cleanup, and the rejected pairs are mostly justified by model- or code-path-specific differences. No clear overstatement of evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 5, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:24643", + "issue:30064", + "issue:36010", + "issue:36331", + "issue:38175", + "issue:41093", + "issue:41553", + "issue:42617", + "issue:42915", + "issue:42947", + "issue:43366", + "issue:43531", + "issue:43550", + "issue:43632", + "issue:43638", + "issue:43645", + "issue:43646", + "issue:43824", + "issue:43828", + "issue:43854", + "issue:43940", + "issue:43950", + "issue:43957", + "issue:43994", + "issue:44155", + "issue:44479", + "issue:44488", + "issue:44530", + "issue:44560", + "issue:44589", + "issue:44661", + "issue:44805", + "issue:44863", + "issue:44910", + "issue:44918", + "issue:44928", + "issue:44991", + "issue:45313", + "issue:45325", + "issue:45584", + "issue:45588", + "issue:45663", + "issue:45685", + "issue:45823", + "issue:45834", + "issue:45902", + "issue:45923", + "issue:46402" + ], + "soft_pairs": [ + "issue:36331|issue:41093", + "issue:42947|issue:45923", + "issue:41553|issue:44991", + "issue:43940|issue:45313", + "issue:44928|issue:45588", + "issue:43854|issue:44863", + "issue:41093|issue:44805", + "issue:44479|issue:44560", + "issue:44863|issue:46402", + "issue:41553|issue:44488", + "issue:24643|issue:30064", + "issue:44530|issue:44863", + "issue:45325|issue:45902", + "issue:43366|issue:43828", + "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:42915|issue:43957", + "issue:43366|issue:44589", + "issue:43550|issue:44910", + "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" + ], + "prepared_review_unit_hash": "5d256cfc11c36bff7172663c9ddf9a61ab39c79142c828b71679e26358c3b64e", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 37486, + "estimated_input_tokens": 9372, + "estimated_eval_tokens": 19000 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a grab bag of unrelated Transformers bug reports, with a few clear duplicate-like pairs around the same model or error path. I\u2019d keep the more general custom-model init issue as the canonical issue.", + "confidence": 0.81, + "canonical_issue_reason": "issue:43646 is the broad, general statement of the custom-model initialization regression; issue:43645 looks like a notebook-specific variant of the same problem.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43646 is the cleanest representative of the duplicate pair because it describes the underlying regression without the extra notebook-specific constraint.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:36331", + "right": "issue:41093", + "accept": false, + "reason": "Both are training-time errors, but the compute_loss kwarg mismatch and the mask/tensor shape mismatch are different failures and different code paths." + }, + { + "left": "issue:42947", + "right": "issue:45923", + "accept": false, + "reason": "Both mention gradient checkpointing, but one is about PEFT LoRA behavior and the other is a missing flag on a trust_remote_code model; not the same bug." + }, + { + "left": "issue:41553", + "right": "issue:44991", + "accept": false, + "reason": "Both concern tokenizer loading, but they affect different models and one is an error-message issue while the other is a hard load failure." + }, + { + "left": "issue:43940", + "right": "issue:45313", + "accept": false, + "reason": "Both are ZeRO-3 weight-loading problems, but they target different Qwen model families and appear to be separate model-specific failures." + }, + { + "left": "issue:44928", + "right": "issue:45588", + "accept": false, + "reason": "One is a Qwen3.5 RLHF gradient explosion from 3D position_ids; the other is a flash-attention AttributeError on sink-less models. Different issues." + }, + { + "left": "issue:43854", + "right": "issue:44863", + "accept": false, + "reason": "Different models and different loading failures; too broad to treat as the same underlying bug." + }, + { + "left": "issue:41093", + "right": "issue:44805", + "accept": true, + "reason": "These describe the same mask-vs-indexed-tensor shape mismatch IndexError, with essentially the same error signature." + }, + { + "left": "issue:44479", + "right": "issue:44560", + "accept": true, + "reason": "Both are Qwen3-VL family video-input regressions in the same release line, and the StopIteration looks like the same broken video path." + }, + { + "left": "issue:44863", + "right": "issue:46402", + "accept": false, + "reason": "Different architectures and different loading failures; no strong evidence of a shared root cause." + }, + { + "left": "issue:41553", + "right": "issue:44488", + "accept": false, + "reason": "Both are loading-related, but one is Voxtral AutoTokenizer messaging and the other is a separate tokenizer load failure for a different model." + }, + { + "left": "issue:24643", + "right": "issue:30064", + "accept": false, + "reason": "Completely different subsystems: DeepSpeed training vs image processor segmentation maps." + }, + { + "left": "issue:44530", + "right": "issue:44863", + "accept": false, + "reason": "Qwen3.5 PagedAttentionCache/linear_attention crash is unrelated to NemotronH checkpoint loading." + }, + { + "left": "issue:45325", + "right": "issue:45902", + "accept": false, + "reason": "Both touch rotary/position handling, but one is Qwen2.5-VL video still-image scaling and the other is a meta/to_empty initialization memory bug." + }, + { + "left": "issue:43366", + "right": "issue:43828", + "accept": false, + "reason": "Different models and different failure modes: GGUF architecture support vs autocast dtype mismatch." + }, + { + "left": "issue:45663", + "right": "issue:45823", + "accept": false, + "reason": "Same model family, but one is an FSDP2 KeyError in attention and the other is a device-map forward mismatch; different code paths." + }, + { + "left": "issue:43366", + "right": "issue:43531", + "accept": false, + "reason": "No overlap beyond broad model-family similarity; gpt-oss GGUF support is unrelated to Qwen3-MoE sliding_window behavior." + }, + { + "left": "issue:44155", + "right": "issue:45584", + "accept": false, + "reason": "AudioFlamingo3 batched inference leakage and Whisper empty-transcription generation are unrelated bugs." + }, + { + "left": "issue:43950", + "right": "issue:45834", + "accept": false, + "reason": "Non-persistent buffer corruption during from_pretrained is unrelated to Kosmos2.5 OCR index errors." + }, + { + "left": "issue:44560", + "right": "issue:45685", + "accept": false, + "reason": "Video-input regression and an MPS histogram-not-implemented MoE issue are unrelated." + }, + { + "left": "issue:43824", + "right": "issue:43957", + "accept": false, + "reason": "Import/export of a specific class and meta-device loading breakage are different issues." + }, + { + "left": "issue:42915", + "right": "issue:43957", + "accept": false, + "reason": "FineGrainedFP8Config failure for Qwen3Moe is unrelated to meta-device model loading." + }, + { + "left": "issue:43366", + "right": "issue:44589", + "accept": false, + "reason": "GGUF support for gpt-oss and Float8 storage lookup failure are different problems." + }, + { + "left": "issue:43550", + "right": "issue:44910", + "accept": false, + "reason": "Different code paths: torch.compile/SDPA on Bamba versus flash-attention with 3D position_ids on Qwen3.5." + }, + { + "left": "issue:43645", + "right": "issue:43646", + "accept": true, + "reason": "Same custom model initialization regression; one report is just notebook-specific while the other is the general failure." + }, + { + "left": "issue:44560", + "right": "issue:44918", + "accept": false, + "reason": "Both involve Qwen3.5/VL embeddings, but one is a video StopIteration regression and the other is TRL SFT unpacking failure." + }, + { + "left": "issue:42617", + "right": "issue:43366", + "accept": false, + "reason": "A script execution problem and GGUF architecture support are not the same bug." + }, + { + "left": "issue:36010", + "right": "issue:42915", + "accept": false, + "reason": "GenerationMixin import failure and Qwen3Moe FP8 config failure are unrelated." + }, + { + "left": "issue:44661", + "right": "issue:45823", + "accept": false, + "reason": "Tokenizer-mapping add-new-model-like failure is unrelated to Gemma4 PLE device mismatch." + }, + { + "left": "issue:44560", + "right": "issue:44805", + "accept": false, + "reason": "Both are IndexError-shaped titles, but one is a Qwen3-VL video issue and the other is a generic mask/tensor mismatch; not enough to merge." + }, + { + "left": "issue:43632", + "right": "issue:45834", + "accept": false, + "reason": "_is_hf_initialized regression and Kosmos2.5 OCR long-input indexing are unrelated." + }, + { + "left": "issue:38175", + "right": "issue:43994", + "accept": true, + "reason": "Same model and same symptom class: incorrect/zeroed outputs from siglip2-base-patch16-224, likely the same underlying regression." + }, + { + "left": "issue:43638", + "right": "issue:44661", + "accept": false, + "reason": "DeepSpeed zero3 with a non-pretrained Bert model is unrelated to tokenizer mapping support failures." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays mostly conservative. The accepted soft-edge pairs are plausibly supported by the titles alone, and the rejected pairs are not overclaimed. The canonical custom-model init choice also looks well supported." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 6, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:38175", + "issue:39692", + "issue:41720", + "issue:42491", + "issue:42915", + "issue:43257", + "issue:43317", + "issue:43425", + "issue:43596", + "issue:43644", + "issue:43645", + "issue:43708", + "issue:43720", + "issue:43854", + "issue:43856", + "issue:43873", + "issue:43931", + "issue:43957", + "issue:43994", + "issue:44075", + "issue:44155", + "issue:44164", + "issue:44291", + "issue:44387", + "issue:44451", + "issue:44488", + "issue:44534", + "issue:44661", + "issue:44756", + "issue:44805", + "issue:44877", + "issue:44910", + "issue:44960", + "issue:44977", + "issue:44991", + "issue:45030", + "issue:45276", + "issue:45292", + "issue:45313", + "issue:45335", + "issue:45357", + "issue:45362", + "issue:45406", + "issue:45488", + "issue:45588", + "issue:45812", + "issue:46050" + ], + "soft_pairs": [ + "issue:44877|issue:45030", + "issue:43596|issue:44661", + "issue:44661|issue:44805", + "issue:43317|issue:44756", + "issue:38175|issue:39692", + "issue:43644|issue:44534", + "issue:42491|issue:43720", + "issue:41720|issue:42915", + "issue:44960|issue:44977", + "issue:42915|issue:44756", + "issue:44451|issue:44488", + "issue:45313|issue:45357", + "issue:42491|issue:45276", + "issue:43708|issue:44075", + "issue:44488|issue:44991", + "issue:43644|issue:43645", + "issue:39692|issue:43994", + "issue:45488|issue:45812", + "issue:43856|issue:43873", + "issue:43257|issue:43854", + "issue:43854|issue:43931", + "issue:43856|issue:44387", + "issue:44960|issue:46050", + "issue:45362|issue:45406", + "issue:42491|issue:45292", + "issue:42491|issue:43931", + "issue:44164|issue:44291", + "issue:44910|issue:45588", + "issue:45276|issue:45335", + "issue:41720|issue:44155", + "issue:43425|issue:43957" + ], + "prepared_review_unit_hash": "b8a6bf62ad043f7ef0acf1ff79816fba6d788c8e8d9db5de4b2b14fe060856a0", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 31, + "serialized_chars": 36396, + "estimated_input_tokens": 9099, + "estimated_eval_tokens": 18454 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose set of Transformers v5-era regression reports around loading, quantization, offload, and model-specific configuration, with only a few clear duplicate pairs. The strongest duplicate signal is the non-persistent buffer corruption report, while most other pairs are only thematically similar and should stay separate.", + "confidence": 0.61, + "canonical_issue_reason": "issue:43644 is the cleanest anchor: it describes a concrete Transformers 5 regression about non-persistent buffers being filled with junk, and issue:44534 is a near-exact duplicate with the same symptom.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43644 is the most representative and actionable issue in the set, with the clearest duplicate relationship and a concrete library-wide v5 regression rather than a model-specific one.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44877", + "right": "issue:45030", + "accept": false, + "reason": "Both are config-loading breakages, but they affect different models and appear to stem from different validation paths." + }, + { + "left": "issue:43596", + "right": "issue:44661", + "accept": false, + "reason": "One is a DeepSpeed/BertModel init failure; the other is an add-new-model-like tokenizer mapping issue." + }, + { + "left": "issue:44661", + "right": "issue:44805", + "accept": false, + "reason": "Tokenizer mapping failure and mask-shape IndexError are unrelated code paths." + }, + { + "left": "issue:43317", + "right": "issue:44756", + "accept": false, + "reason": "Both mention loading/offload behavior, but one is dequantized gpu+cpu offload and the other is mmap/OOM on Strix Halo." + }, + { + "left": "issue:38175", + "right": "issue:39692", + "accept": false, + "reason": "Same model family, but one is zero probabilities and the other is a docs-example mix of model/processor mismatch and quantization errors." + }, + { + "left": "issue:43644", + "right": "issue:44534", + "accept": true, + "reason": "They describe the same Transformers 5 bug: non-persistent buffers getting corrupted/junk values." + }, + { + "left": "issue:42491", + "right": "issue:43720", + "accept": false, + "reason": "LoRA/Qwen3 MoE version incompatibility and BitNet packed-weight loading are different failures." + }, + { + "left": "issue:41720", + "right": "issue:42915", + "accept": false, + "reason": "Both are Qwen3-related, but one is a device-map CUDA assert and the other is a FineGrainedFP8Config failure." + }, + { + "left": "issue:44960", + "right": "issue:44977", + "accept": false, + "reason": "Different models and different failure modes; too broad to be the same bug." + }, + { + "left": "issue:42915", + "right": "issue:44756", + "accept": false, + "reason": "Qwen3Moe FP8 loading and Strix Halo mmap/OOM are unrelated." + }, + { + "left": "issue:44451", + "right": "issue:44488", + "accept": false, + "reason": "Both are load failures for different models, but there is no concrete evidence they share the same underlying bug." + }, + { + "left": "issue:45313", + "right": "issue:45357", + "accept": false, + "reason": "DeepSpeed ZeRO-3 weight loading and incorrect visual encoder save keys are different issues." + }, + { + "left": "issue:42491", + "right": "issue:45276", + "accept": false, + "reason": "Qwen3 MoE LoRA incompatibility is not the same as resize_token_embeddings propagation in Gemma4." + }, + { + "left": "issue:43708", + "right": "issue:44075", + "accept": false, + "reason": "Trainer checkpoint-resume math and SGD args not being used are unrelated." + }, + { + "left": "issue:44488", + "right": "issue:44991", + "accept": false, + "reason": "Both are tokenizer/model load complaints, but for different models and without a shared concrete failure path." + }, + { + "left": "issue:43644", + "right": "issue:43645", + "accept": false, + "reason": "Non-persistent buffer corruption and Jupyter custom-model initialization are distinct regressions." + }, + { + "left": "issue:39692", + "right": "issue:43994", + "accept": false, + "reason": "Both involve SigLIP2, but one is a docs-example/mismatch report and the other is a nonsensical-results inference report; not clearly the same bug." + }, + { + "left": "issue:45488", + "right": "issue:45812", + "accept": false, + "reason": "Tokenizer behavior regressions, but they affect different tokenizer stacks and model families." + }, + { + "left": "issue:43856", + "right": "issue:43873", + "accept": false, + "reason": "Both mention memory/offloading around Qwen3 MoE, but one is training memory inefficiency and the other is quantization offload behavior." + }, + { + "left": "issue:43257", + "right": "issue:43854", + "accept": false, + "reason": "Qwen3 MoE weight conversion under accelerate+deepspeed is unrelated to GLM-4.7-Flash loading in tests." + }, + { + "left": "issue:43854", + "right": "issue:43931", + "accept": false, + "reason": "Different model-loading failures with different symptoms and code paths." + }, + { + "left": "issue:43856", + "right": "issue:44387", + "accept": false, + "reason": "Both are memory-related, but one is training inefficiency and the other is int4 quantization reserved-memory/OOM regression." + }, + { + "left": "issue:44960", + "right": "issue:46050", + "accept": false, + "reason": "GLM5 issue and quantized save/load breakage are separate problems." + }, + { + "left": "issue:45362", + "right": "issue:45406", + "accept": false, + "reason": "Qwen3.5 chat crash and Gemma4Processor serve crash are unrelated." + }, + { + "left": "issue:42491", + "right": "issue:45292", + "accept": false, + "reason": "LoRA/hf4.x versus resize_token_embeddings propagation are different bugs." + }, + { + "left": "issue:42491", + "right": "issue:43931", + "accept": false, + "reason": "Qwen3 MoE version incompatibility is not the same as Qwen3-VL shape mismatch." + }, + { + "left": "issue:44164", + "right": "issue:44291", + "accept": false, + "reason": "save/from_pretrained extra_state handling and init_empty_weights argument validation are different loading issues." + }, + { + "left": "issue:44910", + "right": "issue:45588", + "accept": false, + "reason": "Both touch flash_attention, but one is a packed-sequence position_ids crash and the other is an s_aux=None AttributeError." + }, + { + "left": "issue:45276", + "right": "issue:45335", + "accept": true, + "reason": "Both point to the same resize_token_embeddings propagation bug across model-specific embedding/output modules." + }, + { + "left": "issue:41720", + "right": "issue:44155", + "accept": false, + "reason": "CUDA assert during Qwen3 auto device map is unrelated to AudioFlamingo3 batched inference token/embedding leakage." + }, + { + "left": "issue:43425", + "right": "issue:43957", + "accept": false, + "reason": "Torch 2.10 incompatibility and meta-device loading regressions are different problems." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The analysis is grounded in the packet and stays conservative on the soft edges. The two accepted duplicate pairs are well supported by the issue titles/body themes, and the rejected pairs do not overclaim similarity. The summary is appropriately cautious overall." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 7, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:36010", + "issue:39404", + "issue:39692", + "issue:42947", + "issue:43335", + "issue:43381", + "issue:43404", + "issue:43425", + "issue:43454", + "issue:43611", + "issue:43645", + "issue:43646", + "issue:43708", + "issue:43828", + "issue:43854", + "issue:43873", + "issue:43957", + "issue:44155", + "issue:44164", + "issue:44291", + "issue:44292", + "issue:44387", + "issue:44402", + "issue:44451", + "issue:44479", + "issue:44488", + "issue:44589", + "issue:44683", + "issue:44756", + "issue:44877", + "issue:44912", + "issue:44928", + "issue:44960", + "issue:45216", + "issue:45276", + "issue:45292", + "issue:45335", + "issue:45588", + "issue:45663", + "issue:45684", + "issue:45698", + "issue:45701", + "issue:45704", + "issue:45759", + "issue:45812", + "issue:45823", + "issue:45859", + "issue:46315" + ], + "soft_pairs": [ + "issue:43404|issue:43454", + "issue:39692|issue:44960", + "issue:44402|issue:44488", + "issue:42947|issue:45704", + "issue:36010|issue:39404", + "issue:43873|issue:44387", + "issue:45663|issue:45759", + "issue:44291|issue:44756", + "issue:45663|issue:45859", + "issue:44155|issue:45823", + "issue:44292|issue:44912", + "issue:44387|issue:44756", + "issue:43335|issue:45663", + "issue:43957|issue:44756", + "issue:42947|issue:43381", + "issue:45276|issue:45292", + "issue:44164|issue:44756", + "issue:43381|issue:44928", + "issue:43854|issue:45216", + "issue:44479|issue:44877", + "issue:45684|issue:45698", + "issue:45588|issue:45663", + "issue:43646|issue:43708", + "issue:43645|issue:44164", + "issue:45292|issue:45335", + "issue:45812|issue:46315", + "issue:43828|issue:43957", + "issue:43425|issue:43828", + "issue:44402|issue:44451", + "issue:44589|issue:44756", + "issue:39404|issue:45701", + "issue:44683|issue:45823", + "issue:43611|issue:43645", + "issue:43425|issue:45823" + ], + "prepared_review_unit_hash": "2d2fd9a038855907456d2c3c9611352f94442864a49190ca8f782db10c7aba68", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 34, + "serialized_chars": 37967, + "estimated_input_tokens": 9492, + "estimated_eval_tokens": 19240 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab-bag of unrelated Transformers bug reports. The only clear duplicate family here is the resize_token_embeddings regression around downstream embeddings; most other pairs are only loosely related by subsystem and should stay separate.", + "confidence": 0.64, + "canonical_issue_reason": "issue:45292 is the most general, model-agnostic report of the resize_token_embeddings/output_embeddings regression, so it works best as the cluster anchor.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45292 is the best duplicate target because its title captures the shared API-level bug without narrowing to one model family.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43404", + "right": "issue:43454", + "accept": false, + "reason": "Both mention missing weight tying/generation quality, but they are different model-specific failures and not the same concrete bug." + }, + { + "left": "issue:39692", + "right": "issue:44960", + "accept": false, + "reason": "Unrelated topics: SigLIP2 documentation example errors vs a generic GLM5 issue." + }, + { + "left": "issue:44402", + "right": "issue:44488", + "accept": false, + "reason": "Both are model-loading complaints, but the underlying failures appear different (vocab mismatch vs a specific model load regression)." + }, + { + "left": "issue:42947", + "right": "issue:45704", + "accept": false, + "reason": "Different bugs: LoRA gradient checkpointing behavior vs a T5 RMSNorm memory leak." + }, + { + "left": "issue:36010", + "right": "issue:39404", + "accept": false, + "reason": "ImportError for GenerationMixin is unrelated to Whisper pipeline return_language behavior." + }, + { + "left": "issue:43873", + "right": "issue:44387", + "accept": false, + "reason": "Both involve quantization, but one is offloading behavior and the other is CUDA memory/OOM regression." + }, + { + "left": "issue:45663", + "right": "issue:45759", + "accept": false, + "reason": "Different code paths: Gemma4 FSDP attention state vs Qwen multimodal config unwrapping." + }, + { + "left": "issue:44291", + "right": "issue:44756", + "accept": false, + "reason": "init_empty_weights/_is_hf_initialized is unrelated to mmap-related OOM on Strix Halo." + }, + { + "left": "issue:45663", + "right": "issue:45859", + "accept": false, + "reason": "Different model families and failures: Gemma4 FSDP attention vs Qwen3_5Moe CPU offload compatibility." + }, + { + "left": "issue:44155", + "right": "issue:45823", + "accept": false, + "reason": "AudioFlamingo3 batch leakage and Gemma4 device mismatch are unrelated bugs." + }, + { + "left": "issue:44292", + "right": "issue:44912", + "accept": false, + "reason": "Both are quantization load issues, but the concrete models and failure modes differ." + }, + { + "left": "issue:44387", + "right": "issue:44756", + "accept": false, + "reason": "OOM under int4 quantization is not the same as disabling mmap to avoid OOM on Strix Halo." + }, + { + "left": "issue:43335", + "right": "issue:45663", + "accept": false, + "reason": "Sparse-layer construction in SwitchTransformers is unrelated to Gemma4 FSDP attention state rebuilding." + }, + { + "left": "issue:43957", + "right": "issue:44756", + "accept": false, + "reason": "meta-device model loading breakage is a different issue from Strix Halo mmap OOM." + }, + { + "left": "issue:42947", + "right": "issue:43381", + "accept": false, + "reason": "Both mention gradient checkpointing, but one is ineffective under LoRA and the other is an eval-mode restriction." + }, + { + "left": "issue:45276", + "right": "issue:45292", + "accept": true, + "reason": "Same underlying regression: resize_token_embeddings does not propagate to downstream output embeddings; 45276 is the Gemma4-specific instance." + }, + { + "left": "issue:44164", + "right": "issue:44756", + "accept": false, + "reason": "save/from_pretrained extra_state handling is unrelated to mmap/OOM behavior." + }, + { + "left": "issue:43381", + "right": "issue:44928", + "accept": false, + "reason": "Gradient checkpointing in eval mode is unrelated to RLHF NaN explosions from 3D position_ids and SDPA fallback." + }, + { + "left": "issue:43854", + "right": "issue:45216", + "accept": false, + "reason": "Different save/load regressions affecting different models and checkpoints." + }, + { + "left": "issue:44479", + "right": "issue:44877", + "accept": false, + "reason": "Video input regression in Qwen variants is unrelated to granite_speech config loading." + }, + { + "left": "issue:45684", + "right": "issue:45698", + "accept": false, + "reason": "Both involve save_pretrained/custom modules, but one is permissions propagation and the other is loading the wrong module." + }, + { + "left": "issue:45588", + "right": "issue:45663", + "accept": false, + "reason": "Flash-attention sink-less-model crash is unrelated to Gemma4 FSDP shared_kv_states." + }, + { + "left": "issue:43646", + "right": "issue:43708", + "accept": false, + "reason": "Custom model initialization failure and checkpoint-resume max_steps miscalculation are different bugs." + }, + { + "left": "issue:43645", + "right": "issue:44164", + "accept": false, + "reason": "Notebook custom-model initialization and extra_state save/load handling are unrelated." + }, + { + "left": "issue:45292", + "right": "issue:45335", + "accept": true, + "reason": "Same API regression class: resize_token_embeddings fails to update the relevant downstream embeddings, with different model-specific manifestations." + }, + { + "left": "issue:45812", + "right": "issue:46315", + "accept": false, + "reason": "Token-ID correctness regression and tokenization performance regression are different bugs." + }, + { + "left": "issue:43828", + "right": "issue:43957", + "accept": false, + "reason": "Autocast dtype mismatch in Phi-tiny-MoE is unrelated to meta-device model loading breakage." + }, + { + "left": "issue:43425", + "right": "issue:43828", + "accept": false, + "reason": "Torch 2.10 compatibility and autocast dtype mismatch are different failure classes." + }, + { + "left": "issue:44402", + "right": "issue:44451", + "accept": false, + "reason": "Tokenizer vocab-size mismatch and a specific model loading failure are not the same bug." + }, + { + "left": "issue:44589", + "right": "issue:44756", + "accept": false, + "reason": "Missing Float8 storage type and mmap-related OOM are unrelated." + }, + { + "left": "issue:39404", + "right": "issue:45701", + "accept": false, + "reason": "Whisper pipeline return_language regression is unrelated to tokenizer-version changes." + }, + { + "left": "issue:44683", + "right": "issue:45823", + "accept": false, + "reason": "Compiled flex_attention on torch>=2.9 and Gemma4 device mismatch are unrelated." + }, + { + "left": "issue:43611", + "right": "issue:43645", + "accept": false, + "reason": "Both concern Transformers 5 custom-model behavior, but they describe different regressions and code paths." + }, + { + "left": "issue:43425", + "right": "issue:45823", + "accept": false, + "reason": "Torch version incompatibility is unrelated to Gemma4 PLE device placement." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet: the cluster is mostly heterogeneous, and the only clearly supported duplicate family is the resize_token_embeddings/output_embeddings regression centered on issue:45292 with the Gemma4 and t5gemma variants. The rejected soft-edge pairs are phrased conservatively and do not overclaim sameness; the accepted pairs within the resize_token_embeddings family look like legitimate single-bug variants." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 8, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:32090", + "issue:36032", + "issue:39404", + "issue:40990", + "issue:42915", + "issue:43299", + "issue:43421", + "issue:43425", + "issue:43452", + "issue:43582", + "issue:43596", + "issue:43611", + "issue:43632", + "issue:43643", + "issue:43644", + "issue:43716", + "issue:43856", + "issue:43901", + "issue:43906", + "issue:43931", + "issue:43994", + "issue:44079", + "issue:44164", + "issue:44265", + "issue:44479", + "issue:44488", + "issue:44509", + "issue:44530", + "issue:44589", + "issue:44661", + "issue:44671", + "issue:44683", + "issue:44737", + "issue:44756", + "issue:44805", + "issue:44898", + "issue:44912", + "issue:44918", + "issue:44987", + "issue:44991", + "issue:45081", + "issue:45084", + "issue:45759", + "issue:45859", + "issue:46050", + "issue:46285", + "issue:46326", + "issue:46402" + ], + "soft_pairs": [ + "issue:43425|issue:43716", + "issue:44530|issue:45084", + "issue:43906|issue:44530", + "issue:39404|issue:44912", + "issue:44265|issue:44898", + "issue:44479|issue:44805", + "issue:43906|issue:46326", + "issue:44589|issue:45084", + "issue:46050|issue:46326", + "issue:43901|issue:44509", + "issue:43425|issue:44918", + "issue:43299|issue:43931", + "issue:43994|issue:44079", + "issue:36032|issue:43452", + "issue:44164|issue:45759", + "issue:44671|issue:44991", + "issue:43596|issue:44683", + "issue:44991|issue:45081", + "issue:43856|issue:44756", + "issue:44164|issue:44479", + "issue:39404|issue:44671", + "issue:44737|issue:45859", + "issue:43582|issue:43632", + "issue:44488|issue:45081", + "issue:44661|issue:44683", + "issue:44488|issue:44987", + "issue:32090|issue:39404", + "issue:43611|issue:43644", + "issue:43716|issue:45084", + "issue:46285|issue:46402", + "issue:40990|issue:43421", + "issue:36032|issue:42915", + "issue:43643|issue:44530" + ], + "prepared_review_unit_hash": "358b24bdc4efbbfae0ea3c8f533959461009987a0cf97b7b3b5eb7edf929b93e", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 33, + "serialized_chars": 37710, + "estimated_input_tokens": 9428, + "estimated_eval_tokens": 19112 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items span many unrelated Transformers regressions; none of the soft pairs look like the same underlying bug/change, so all soft edges are rejected. The most general issue in the set is the v5 model-loading regression around `base_model_prefix`, but there is no strong single duplicate cluster overall.", + "confidence": 0.81, + "canonical_issue_reason": "Issue 43611 is the broadest, most user-facing loading regression in the set and is the closest thing to a canonical issue, but the overall cluster is heterogeneous.", + "canonical_pr_reason": null, + "best_issue_reason": "43611 best represents the recurring v5 model-loading breakages because it is broad, concrete, and likely to subsume adjacent loading regressions better than the more model-specific reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43425", + "right": "issue:43716", + "accept": false, + "reason": "Torch version incompatibility vs. Qwen3-Mistral image preprocessor/model dtype mismatch; different failure modes and code paths." + }, + { + "left": "issue:44530", + "right": "issue:45084", + "accept": false, + "reason": "PagedAttentionCache runtime crash on Qwen3.5 vs tokenizer template compilation error; unrelated subsystems." + }, + { + "left": "issue:43906", + "right": "issue:44530", + "accept": false, + "reason": "Isolated reproduction of another issue vs Qwen3.5 cache crash; no shared bug surface." + }, + { + "left": "issue:39404", + "right": "issue:44912", + "accept": false, + "reason": "Whisper pipeline `return_language` regression vs MXFP4 quantization fallback; different model/features." + }, + { + "left": "issue:44265", + "right": "issue:44898", + "accept": false, + "reason": "`torch.export` failure around `torch_compilable_check` vs Perceiver interpolation/image-size issue; distinct problems." + }, + { + "left": "issue:44479", + "right": "issue:44805", + "accept": false, + "reason": "Video-input regression in VL models vs mask-shape index error; not the same underlying defect." + }, + { + "left": "issue:43906", + "right": "issue:46326", + "accept": false, + "reason": "Reproduction of issue 38071 vs mistral_common tokenizer `continue_final_message` bug; different tokenizer behavior." + }, + { + "left": "issue:44589", + "right": "issue:45084", + "accept": false, + "reason": "Float8 storage lookup error vs template compilation error; unrelated runtime failures." + }, + { + "left": "issue:46050", + "right": "issue:46326", + "accept": false, + "reason": "Quantization save/load corruption vs tokenizer final-message handling; different code paths." + }, + { + "left": "issue:43901", + "right": "issue:44509", + "accept": false, + "reason": "Docs mismatch for `return_all_scores` vs docs mentioning removed pipeline tasks; documentation-only but not the same issue." + }, + { + "left": "issue:43425", + "right": "issue:44918", + "accept": false, + "reason": "Torch incompatibility vs TRL SFT unpacking embeddings; different components and failure signatures." + }, + { + "left": "issue:43299", + "right": "issue:43931", + "accept": false, + "reason": "Qwen3VL MoE loading regression vs Qwen3-VL weight shape mismatch; related model family but different concrete bug." + }, + { + "left": "issue:43994", + "right": "issue:44079", + "accept": false, + "reason": "SigLIP2 nonsensical outputs vs `ModelOutput` key assignment bug; one is inference quality, the other data-structure behavior." + }, + { + "left": "issue:36032", + "right": "issue:43452", + "accept": false, + "reason": "Tokenizer special-token method conflict vs gguf file loading breakage; different tokenizer/load paths." + }, + { + "left": "issue:44164", + "right": "issue:45759", + "accept": false, + "reason": "`extra_state` save/load handling vs composite Qwen config `text_config` unwrapping; both loading-related but not the same defect." + }, + { + "left": "issue:44671", + "right": "issue:44991", + "accept": false, + "reason": "CamemBERT masked-LM prediction regression vs tokenizer load failure for a specific model; unrelated model-specific issues." + }, + { + "left": "issue:43596", + "right": "issue:44683", + "accept": false, + "reason": "DeepSpeed ZeRO3/BertModel indexing error vs compiled flex_attention on newer torch; different stack and root cause." + }, + { + "left": "issue:44991", + "right": "issue:45081", + "accept": false, + "reason": "Tokenizer load failure for EMBEDDIA/est-roberta vs Mistral regex patch crash; same broad area, different bugs." + }, + { + "left": "issue:43856", + "right": "issue:44756", + "accept": false, + "reason": "Qwen3 MoE training memory inefficiency vs mmap OOM on Strix Halo; not the same underlying issue." + }, + { + "left": "issue:44164", + "right": "issue:44479", + "accept": false, + "reason": "Save/from_pretrained extra_state handling vs VL video-input regression; no shared code path." + }, + { + "left": "issue:39404", + "right": "issue:44671", + "accept": false, + "reason": "Whisper pipeline language return regression vs CamemBERT masked LM prediction issue; unrelated models and symptoms." + }, + { + "left": "issue:44737", + "right": "issue:45859", + "accept": false, + "reason": "XLNet CPU arange/device bug vs Qwen3_5Moe CPU offload incompatibility; both device-related but different code paths." + }, + { + "left": "issue:43582", + "right": "issue:43632", + "accept": false, + "reason": "Apple Silicon allocator warmup TypeError vs `_is_hf_initialized` flag regression; different internal failures." + }, + { + "left": "issue:44488", + "right": "issue:45081", + "accept": false, + "reason": "Loading `cjvt/sleng-bert` failure vs Mistral regex patch crash; distinct tokenizer-loading problems." + }, + { + "left": "issue:44661", + "right": "issue:44683", + "accept": false, + "reason": "`add-new-model-like` generator logic vs compiled flex_attention; unrelated feature/tooling areas." + }, + { + "left": "issue:44488", + "right": "issue:44987", + "accept": false, + "reason": "Model load failure for `cjvt/sleng-bert` vs `physical-intelligence/fast` load regression; different affected models and likely causes." + }, + { + "left": "issue:32090", + "right": "issue:39404", + "accept": false, + "reason": "Trainer `_gpu_broadcast_one` NoneType TypeError vs Whisper `return_language` pipeline regression; unrelated behavior." + }, + { + "left": "issue:43611", + "right": "issue:43644", + "accept": false, + "reason": "`base_model_prefix` loading regression vs non-persistent buffer junk initialization; both v5-related but different bugs." + }, + { + "left": "issue:43716", + "right": "issue:45084", + "accept": false, + "reason": "Image preprocessor/model dtype mismatch vs tokenizer template compilation error; no shared underlying change." + }, + { + "left": "issue:46285", + "right": "issue:46402", + "accept": false, + "reason": "CLIP text-model structure inconsistency vs Sam3 text encoder weights not loading; separate model architecture/loading issues." + }, + { + "left": "issue:40990", + "right": "issue:43421", + "accept": false, + "reason": "High perplexity on gpt-oss-20b vs runtime special-token/post-processor update behavior; unrelated problems." + }, + { + "left": "issue:36032", + "right": "issue:42915", + "accept": false, + "reason": "Tokenizer method name conflict vs Qwen3Moe FineGrainedFP8Config failure; different tokenizer/model-loading paths." + }, + { + "left": "issue:43643", + "right": "issue:44530", + "accept": false, + "reason": "Missing fields from `trust_remote_code=True` config objects vs Qwen3.5 cache crash; different APIs and failure modes." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 9, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:34567", + "issue:36032", + "issue:40444", + "issue:42175", + "issue:42915", + "issue:43089", + "issue:43262", + "issue:43317", + "issue:43322", + "issue:43377", + "issue:43404", + "issue:43452", + "issue:43482", + "issue:43540", + "issue:43541", + "issue:43644", + "issue:43645", + "issue:43646", + "issue:43746", + "issue:43819", + "issue:43883", + "issue:43994", + "issue:44206", + "issue:44220", + "issue:44265", + "issue:44291", + "issue:44403", + "issue:44451", + "issue:44464", + "issue:44466", + "issue:44488", + "issue:44530", + "issue:44534", + "issue:44610", + "issue:44683", + "issue:44811", + "issue:44849", + "issue:44987", + "issue:44991", + "issue:45081", + "issue:45084", + "issue:45325", + "issue:45507", + "issue:45584", + "issue:45987", + "issue:46178", + "issue:46315", + "issue:46326" + ], + "soft_pairs": [ + "issue:34567|issue:40444", + "issue:40444|issue:46326", + "issue:44451|issue:44987", + "issue:43541|issue:45084", + "issue:44811|issue:45584", + "issue:42175|issue:46178", + "issue:42915|issue:44530", + "issue:43644|issue:43646", + "issue:44206|issue:44220", + "issue:43452|issue:43482", + "issue:43883|issue:44488", + "issue:36032|issue:43645", + "issue:43317|issue:43482", + "issue:44464|issue:45987", + "issue:45084|issue:45507", + "issue:44987|issue:44991", + "issue:44849|issue:45325", + "issue:43746|issue:46326", + "issue:43262|issue:43377", + "issue:43645|issue:44534", + "issue:43646|issue:44534", + "issue:44403|issue:45081", + "issue:43404|issue:44466", + "issue:42915|issue:45084", + "issue:43883|issue:44991", + "issue:43089|issue:45987", + "issue:43994|issue:44610", + "issue:43322|issue:43540", + "issue:43377|issue:43819", + "issue:44451|issue:45081", + "issue:44987|issue:45081", + "issue:44265|issue:44610", + "issue:44991|issue:46315", + "issue:42915|issue:44291", + "issue:44220|issue:44683" + ], + "prepared_review_unit_hash": "dd254cdcf3bf9ad46bb27d4dda0415656d71964f39f0bbe813097e96d1f6c69d", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 35, + "serialized_chars": 38320, + "estimated_input_tokens": 9580, + "estimated_eval_tokens": 19416 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a loose set of mostly unrelated Transformers regressions and model-loading issues. The soft pairs generally share only broad subsystem similarity, not the same concrete bug or fixable code path, so they should be rejected as duplicates.", + "confidence": 0.68, + "canonical_issue_reason": "issue:43646 is the broadest representative in the set: it describes a general Transformers 5.0.0 custom model initialization regression, which is closer to the recurring v5 initialization/load failures than the more specific or unrelated issues nearby.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43646 is the best issue candidate because it is the most general and reusable description of the recurring v5 initialization breakage, while the other candidates are narrower symptoms (Jupyter-only, non-persistent buffers, tokenizer/model-specific loading failures).", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:34567", + "right": "issue:40444", + "accept": false, + "reason": "Unrelated bugs: token-count tracking in TrainerState vs multimodal IterableDataset loading." + }, + { + "left": "issue:40444", + "right": "issue:46326", + "accept": false, + "reason": "Both mention model-processing failures, but one is Qwen2.5-VL multi-image dataset handling and the other is mistral_common continue_final_message behavior." + }, + { + "left": "issue:44451", + "right": "issue:44987", + "accept": false, + "reason": "Different model-loading failures for different checkpoints; no evidence of the same underlying bug." + }, + { + "left": "issue:43541", + "right": "issue:45084", + "accept": false, + "reason": "One is a grouped_mm tracing failure in Mixtral; the other is a template compilation error. Different code paths." + }, + { + "left": "issue:44811", + "right": "issue:45584", + "accept": false, + "reason": "Both involve Whisper generation/token handling, but one is batch_decode skip_special_tokens and the other is empty transcription after align_special_tokens." + }, + { + "left": "issue:42175", + "right": "issue:46178", + "accept": false, + "reason": "Backend packaging vs backend registry validation are related at a high level, but not the same defect." + }, + { + "left": "issue:42915", + "right": "issue:44530", + "accept": false, + "reason": "Different model families and failures: FineGrainedFP8Config on Qwen3Moe vs PagedAttentionCache linear_attention crash on Qwen3.5." + }, + { + "left": "issue:43644", + "right": "issue:43646", + "accept": false, + "reason": "Both are Transformers 5 regressions, but one is non-persistent buffer initialization and the other is custom model initialization." + }, + { + "left": "issue:44206", + "right": "issue:44220", + "accept": false, + "reason": "Similar audio preprocessing area, but the failures differ: unsupported center arg vs _torch_extract_fbank_features behavior." + }, + { + "left": "issue:43452", + "right": "issue:43482", + "accept": false, + "reason": "Both are gguf-related loading problems, but one targets AutoTokenizer/AutoModelForCausalLM with gguf_file and the other is Qwen2.5-GGUF under v5." + }, + { + "left": "issue:43883", + "right": "issue:44488", + "accept": false, + "reason": "Same broad loader regression theme, but one is missing all_tied_weights_keys and the other is a specific model load failure." + }, + { + "left": "issue:36032", + "right": "issue:43645", + "accept": false, + "reason": "Tokenizer class/method conflict vs notebook custom model initialization; unrelated." + }, + { + "left": "issue:43317", + "right": "issue:43482", + "accept": false, + "reason": "Device-map offload/dequantized model loading vs GGUF model loading; different failure surfaces." + }, + { + "left": "issue:44464", + "right": "issue:45987", + "accept": false, + "reason": "Chunked generation inconsistencies vs StaticCache.get_seq_length return type; not the same bug." + }, + { + "left": "issue:45084", + "right": "issue:45507", + "accept": false, + "reason": "Template compilation error vs GraniteMoEHybrid invalid method call; different underlying issues." + }, + { + "left": "issue:44987", + "right": "issue:44991", + "accept": false, + "reason": "Both are v5 loading regressions, but they affect different models and likely different loader/tokenizer paths." + }, + { + "left": "issue:44849", + "right": "issue:45325", + "accept": false, + "reason": "Different Qwen3.5 bugs: hidden-state output vs rope index scaling for Qwen2.5-VL." + }, + { + "left": "issue:43746", + "right": "issue:46326", + "accept": false, + "reason": "PEFT/local-checkpoint loading for GraniteSpeech is unrelated to mistral_common continue_final_message handling." + }, + { + "left": "issue:43262", + "right": "issue:43377", + "accept": false, + "reason": "Audio sampling-rate defaulting vs MIMI encoder padding-mask correctness; different components and symptoms." + }, + { + "left": "issue:43645", + "right": "issue:44534", + "accept": false, + "reason": "Notebook/custom model initialization bug vs non-persistent buffer junk; related only by being v5 regressions." + }, + { + "left": "issue:43646", + "right": "issue:44534", + "accept": false, + "reason": "General custom initialization regression vs buffer initialization corruption; not the same code path." + }, + { + "left": "issue:44403", + "right": "issue:45081", + "accept": false, + "reason": "Unnecessary loading noise vs Mistral regex patch backend_tokenizer AttributeError; unrelated." + }, + { + "left": "issue:43404", + "right": "issue:44466", + "accept": false, + "reason": "lm_head tying bug in Mistral3 vs device-dependent lm_head serialization inconsistency; related around weights, but not the same defect." + }, + { + "left": "issue:42915", + "right": "issue:45084", + "accept": false, + "reason": "Qwen3Moe FP8 config failure and template compilation failure are unrelated." + }, + { + "left": "issue:43883", + "right": "issue:44991", + "accept": false, + "reason": "Different loading failures for different models and causes; no shared concrete bug." + }, + { + "left": "issue:43089", + "right": "issue:45987", + "accept": false, + "reason": "Generation performance overhead vs cache length return type contract; different problem classes." + }, + { + "left": "issue:43994", + "right": "issue:44610", + "accept": false, + "reason": "SigLIP2 incorrect outputs via AutoModel/pipeline vs OmDet-Turbo processor/model resolution mismatch; not the same issue." + }, + { + "left": "issue:43322", + "right": "issue:43540", + "accept": false, + "reason": "Segfault on Llava Next loading vs Qwen3OmniMoe video-input ValueError; both multimodal, but distinct failures." + }, + { + "left": "issue:43377", + "right": "issue:43819", + "accept": false, + "reason": "MIMI batched-vs-single padding-mask mismatch vs DAC.from_latents missing STE mismatch; unrelated audio-model correctness bugs." + }, + { + "left": "issue:44451", + "right": "issue:45081", + "accept": false, + "reason": "Different model-specific loading failure vs Mistral tokenizer regex patch crash." + }, + { + "left": "issue:44987", + "right": "issue:45081", + "accept": false, + "reason": "Model loading regression for physical-intelligence/fast vs Mistral tokenizer regex backend issue; different subsystems." + }, + { + "left": "issue:44265", + "right": "issue:44610", + "accept": false, + "reason": "torch.export/exportability with torch_compilable_check vs processor size mismatch; not the same defect." + }, + { + "left": "issue:44991", + "right": "issue:46315", + "accept": false, + "reason": "Tokenizer loading failure for a specific model vs slow convert_tokens_to_ids regression; different functions and symptoms." + }, + { + "left": "issue:42915", + "right": "issue:44291", + "accept": false, + "reason": "FineGrainedFP8Config on Qwen3Moe vs init_empty_weights/_is_hf_initialized argument mismatch; unrelated initialization paths." + }, + { + "left": "issue:44220", + "right": "issue:44683", + "accept": false, + "reason": "Audio feature extraction issue vs compiled flex_attention on torch>=2.9; 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": 19, + "review_unit_index": 10, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:33666", + "issue:35707", + "issue:36032", + "issue:40444", + "issue:40990", + "issue:41720", + "issue:42915", + "issue:43296", + "issue:43381", + "issue:43404", + "issue:43421", + "issue:43425", + "issue:43482", + "issue:43525", + "issue:43531", + "issue:43611", + "issue:43632", + "issue:43742", + "issue:43792", + "issue:44062", + "issue:44079", + "issue:44291", + "issue:44403", + "issue:44462", + "issue:44479", + "issue:44534", + "issue:44568", + "issue:44589", + "issue:44683", + "issue:44756", + "issue:44811", + "issue:44829", + "issue:44877", + "issue:44987", + "issue:44991", + "issue:45020", + "issue:45127", + "issue:45216", + "issue:45325", + "issue:45357", + "issue:45479", + "issue:45538", + "issue:45663", + "issue:45812", + "issue:45923", + "issue:46209", + "issue:46326" + ], + "soft_pairs": [ + "issue:45216|issue:45357", + "issue:36032|issue:43742", + "issue:44062|issue:44987", + "issue:46209|issue:46326", + "issue:43632|issue:44756", + "issue:42915|issue:44589", + "issue:44479|issue:45325", + "issue:36032|issue:41720", + "issue:43404|issue:45127", + "issue:43531|issue:44877", + "issue:40444|issue:45325", + "issue:43296|issue:43531", + "issue:43611|issue:44534", + "issue:43425|issue:44403", + "issue:40990|issue:44811", + "issue:43792|issue:44589", + "issue:43632|issue:44534", + "issue:44079|issue:45663", + "issue:44403|issue:44991", + "issue:33666|issue:40444", + "issue:43421|issue:44568", + "issue:44683|issue:45020", + "issue:43381|issue:45923", + "issue:44291|issue:44589", + "issue:35707|issue:40444", + "issue:44462|issue:45812", + "issue:43482|issue:45538", + "issue:44829|issue:45479", + "issue:43482|issue:43525" + ], + "prepared_review_unit_hash": "624248e23865daf1e07182c9a36a9e271ad65da8ec7db4a6240d8068ca30b4ca", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 35802, + "estimated_input_tokens": 8951, + "estimated_eval_tokens": 18158 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: most items are distinct regressions or feature requests across tokenizers, Qwen/VL, FP8, checkpointing, and loading paths. There is no strong duplicate set, so the best single issue is only an umbrella representative, not a true canonical duplicate.", + "confidence": 0.31, + "canonical_issue_reason": "issue:45020 is the broadest umbrella report in the set, covering recent-version breakages in model loading/remote code, but it still does not subsume the many unrelated specific bugs here.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45020 is the best representative because it captures the widest cross-cutting regression theme among these otherwise unrelated issues.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45216", + "right": "issue:45357", + "accept": false, + "reason": "Both are Qwen3.5 save_pretrained regressions, but they describe different failures: overall checkpoint corruption vs incorrect visual-encoder keys. Not the same concrete bug." + }, + { + "left": "issue:36032", + "right": "issue:43742", + "accept": false, + "reason": "Tokenizer constructor conflict in T5 vs a KeyError while loading MobileLLM-125M. Different subsystems and symptoms." + }, + { + "left": "issue:44062", + "right": "issue:44987", + "accept": false, + "reason": "AddedToken special-argument duplication is a tokenizer API bug; loading physical-intelligence/fast is a model loading regression. No shared concrete code path." + }, + { + "left": "issue:46209", + "right": "issue:46326", + "accept": false, + "reason": "FP8 inference failure on B200 and continue_final_message tokenizer behavior are unrelated." + }, + { + "left": "issue:43632", + "right": "issue:44756", + "accept": false, + "reason": "_is_hf_initialized init regression and Strix Halo mmap OOM are unrelated bugs." + }, + { + "left": "issue:42915", + "right": "issue:44589", + "accept": false, + "reason": "Both mention FP8, but one is a Qwen3MoE FineGrainedFP8Config failure and the other is a missing Float8 storage type error. Different failure modes." + }, + { + "left": "issue:44479", + "right": "issue:45325", + "accept": false, + "reason": "Both involve Qwen VL media handling, but one is a broad video-input regression while the other is a specific still-image temporal position_id scaling bug." + }, + { + "left": "issue:36032", + "right": "issue:41720", + "accept": false, + "reason": "T5 tokenizer load error and Qwen3 auto device mapping cuda assert are unrelated." + }, + { + "left": "issue:43404", + "right": "issue:45127", + "accept": false, + "reason": "Mistral3 lm_head tying bug and LoRA merge collapse with extended vocab are different issues, even if both touch tied weights." + }, + { + "left": "issue:43531", + "right": "issue:44877", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior and granite_speech strict-config loading are unrelated." + }, + { + "left": "issue:40444", + "right": "issue:45325", + "accept": false, + "reason": "Both are Qwen2.5-VL-related, but one is training with multiple images per prompt and the other is a still-image position-id scaling regression." + }, + { + "left": "issue:43296", + "right": "issue:43531", + "accept": false, + "reason": "PaddleOCR-VL loading in vLLM and Qwen3-MoE sliding_window are different model/load-path bugs." + }, + { + "left": "issue:43611", + "right": "issue:44534", + "accept": false, + "reason": "base_model_prefix loading breakage and non-persistent buffer junk are unrelated implementation issues." + }, + { + "left": "issue:43425", + "right": "issue:44403", + "accept": false, + "reason": "Torch 2.10 incompatibility is a compatibility issue; unnecessary load noise is a separate UX/logging bug." + }, + { + "left": "issue:40990", + "right": "issue:44811", + "accept": false, + "reason": "High perplexity on gpt-oss-20b and Whisper batch_decode skip_special_tokens behavior are unrelated." + }, + { + "left": "issue:43792", + "right": "issue:44589", + "accept": false, + "reason": "Whisper-large-v2 runtime failure and missing Float8 storage type are unrelated." + }, + { + "left": "issue:43632", + "right": "issue:44534", + "accept": false, + "reason": "Different regressions: init flag handling vs buffer initialization." + }, + { + "left": "issue:44079", + "right": "issue:45663", + "accept": false, + "reason": "ModelOutput key assignment bug and Gemma-4 FSDP2 shared_kv_states KeyError are unrelated." + }, + { + "left": "issue:44403", + "right": "issue:44991", + "accept": false, + "reason": "Loading noise and tokenizer loading regression for EMBEDDIA/est-roberta are different issues." + }, + { + "left": "issue:33666", + "right": "issue:40444", + "accept": false, + "reason": "Both are Qwen-VL training reports, but multi-GPU training and iterable-dataset multi-image failure are not the same bug." + }, + { + "left": "issue:43421", + "right": "issue:44568", + "accept": false, + "reason": "Runtime special-token post-processor update bug and mdeberta add_special_tokens BOS/EOS regression are related in theme but not the same concrete failure." + }, + { + "left": "issue:44683", + "right": "issue:45020", + "accept": false, + "reason": "Torch flex_attention compilation failure and remote_code loading regressions are unrelated." + }, + { + "left": "issue:43381", + "right": "issue:45923", + "accept": false, + "reason": "Eval-mode gradient checkpointing restriction and missing supports_gradient_checkpointing on a trust_remote_code model are different code paths." + }, + { + "left": "issue:44291", + "right": "issue:44589", + "accept": false, + "reason": "Unexpected _is_hf_initialized argument during init_empty_weights and missing Float8 storage type are separate loading/runtime issues." + }, + { + "left": "issue:35707", + "right": "issue:40444", + "accept": false, + "reason": "Progressive generation with inputs_embeds/past_key_values and Qwen2.5-VL multi-image finetuning failure are unrelated." + }, + { + "left": "issue:44462", + "right": "issue:45812", + "accept": false, + "reason": "Both concern AutoTokenizer behavior, but one ignores tokenizer.json from the repo and the other produces wrong Granite token IDs. Too little evidence of the same underlying bug." + }, + { + "left": "issue:43482", + "right": "issue:45538", + "accept": false, + "reason": "Qwen2.5-GGUF loading with transformers v5 and CLIPTokenizer max_length semantics 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": 19, + "review_unit_index": 11, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:40444", + "issue:41553", + "issue:41720", + "issue:42175", + "issue:42915", + "issue:43262", + "issue:43296", + "issue:43317", + "issue:43377", + "issue:43454", + "issue:43482", + "issue:43531", + "issue:43582", + "issue:43606", + "issue:43611", + "issue:43632", + "issue:43644", + "issue:43646", + "issue:43701", + "issue:43746", + "issue:43749", + "issue:43792", + "issue:43818", + "issue:43856", + "issue:43873", + "issue:43957", + "issue:43994", + "issue:44079", + "issue:44265", + "issue:44291", + "issue:44466", + "issue:44479", + "issue:44534", + "issue:44877", + "issue:44898", + "issue:45072", + "issue:45081", + "issue:45325", + "issue:45357", + "issue:45381", + "issue:45491", + "issue:45542", + "issue:45701", + "issue:45800", + "issue:45823", + "issue:45834", + "issue:46402" + ], + "soft_pairs": [ + "issue:43317|issue:43856", + "issue:42175|issue:45542", + "issue:43818|issue:46402", + "issue:41553|issue:43582", + "issue:44079|issue:44534", + "issue:43818|issue:45357", + "issue:43531|issue:44291", + "issue:41720|issue:45081", + "issue:43377|issue:45491", + "issue:40444|issue:45381", + "issue:43644|issue:44079", + "issue:44265|issue:45072", + "issue:43749|issue:44079", + "issue:45325|issue:45381", + "issue:43611|issue:43646", + "issue:42175|issue:43296", + "issue:43632|issue:45072", + "issue:43792|issue:43957", + "issue:43994|issue:44898", + "issue:45325|issue:45834", + "issue:43454|issue:44466", + "issue:42915|issue:43482", + "issue:43701|issue:45325", + "issue:44479|issue:45823", + "issue:44291|issue:44877", + "issue:43873|issue:45701", + "issue:43606|issue:44265", + "issue:43296|issue:45800", + "issue:43262|issue:43746" + ], + "prepared_review_unit_hash": "578ee650938b455b86ab8e7a7ee2321bd80b2f8aa788c28ca3afb72bbfded906", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 35885, + "estimated_input_tokens": 8972, + "estimated_eval_tokens": 18200 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly separate bug reports, not true duplicates. The only clear common theme is a broad set of Transformers v5 loading/init regressions, but the proposed soft pairs each point to different code paths or different model families, so none should be merged.", + "confidence": 0.86, + "canonical_issue_reason": "Issue 43611 is the broadest, most representative loading regression in the set; it captures a core Transformers v5 breakage affecting model loading, whereas the others are narrower variants or unrelated bugs.", + "canonical_pr_reason": null, + "best_issue_reason": "43611 is the best single issue to anchor a broader v5 initialization/loading regression cluster because it is general, concrete, and likely relevant to multiple downstream failures.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43317", + "right": "issue:43856", + "accept": false, + "reason": "Both mention device/offload behavior, but one is a dequantized model load failure and the other is Qwen3 MoE training memory usage; different bug classes." + }, + { + "left": "issue:42175", + "right": "issue:45542", + "accept": false, + "reason": "Both involve TensorFlow/backend packaging, but one is a missing dependency in the torch extra and the other is a tensorboard-without-TF runtime backend error." + }, + { + "left": "issue:43818", + "right": "issue:46402", + "accept": false, + "reason": "Different models and failures: Video-LLaVA video_tower wiring vs Sam3Model text encoder weights not loading." + }, + { + "left": "issue:41553", + "right": "issue:43582", + "accept": false, + "reason": "AutoTokenizer/Voxtral error messaging and Apple Silicon caching_allocator_warmup TypeError are unrelated." + }, + { + "left": "issue:44079", + "right": "issue:44534", + "accept": false, + "reason": "ModelOutput key assignment bug and non-persistent buffer junk are distinct internal behaviors." + }, + { + "left": "issue:43818", + "right": "issue:45357", + "accept": false, + "reason": "Video-LLaVA tower/weight-sharing bug is unrelated to Qwen3.5 save_pretrained key serialization." + }, + { + "left": "issue:43531", + "right": "issue:44291", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior and init_empty_weights/_is_hf_initialized TypeError are different regressions." + }, + { + "left": "issue:41720", + "right": "issue:45081", + "accept": false, + "reason": "Qwen3 auto device mapping cuda assert and Mistral regex patching crash are unrelated code paths." + }, + { + "left": "issue:43377", + "right": "issue:45491", + "accept": false, + "reason": "MIMI padding-mask batching bug and Gemma3 NaN embeddings from sliding-window all-padding windows are different issues." + }, + { + "left": "issue:40444", + "right": "issue:45381", + "accept": false, + "reason": "IterableDataset multi-image finetuning failure is not the same as Qwen2.5-VL video vision_position_ids regression." + }, + { + "left": "issue:43644", + "right": "issue:44079", + "accept": false, + "reason": "Both touch model internals, but one is non-persistent buffer serialization and the other is ModelOutput key handling." + }, + { + "left": "issue:44265", + "right": "issue:45072", + "accept": false, + "reason": "torch.export with torch_compilable_check is a different failure mode from bf16 dtype mismatches in inference." + }, + { + "left": "issue:43749", + "right": "issue:44079", + "accept": false, + "reason": "FSDP CPU RAM efficient loading and ModelOutput key assignment are unrelated." + }, + { + "left": "issue:45325", + "right": "issue:45381", + "accept": false, + "reason": "Both are Qwen2.5-VL position-id issues, but one is still-image RoPE scaling and the other is video input vision_position_ids; not the same bug." + }, + { + "left": "issue:43611", + "right": "issue:43646", + "accept": false, + "reason": "Both are Transformers v5 regressions, but base_model_prefix loading failure and custom model initialization breakage are distinct concrete problems." + }, + { + "left": "issue:42175", + "right": "issue:43296", + "accept": false, + "reason": "TensorFlow packaging/backend issue and PaddleOCR-VL loading failure in vLLM are unrelated." + }, + { + "left": "issue:43632", + "right": "issue:45072", + "accept": false, + "reason": "_is_hf_initialized flag breakage and bfloat16 dtype mismatch are different failure modes." + }, + { + "left": "issue:43792", + "right": "issue:43957", + "accept": false, + "reason": "Whisper can't run and torch.device('meta') loading failures affect different model-loading paths." + }, + { + "left": "issue:43994", + "right": "issue:44898", + "accept": false, + "reason": "Siglip2 nonsensical outputs and Perceiver non-default resolution failure are different vision-model bugs." + }, + { + "left": "issue:45325", + "right": "issue:45834", + "accept": false, + "reason": "Both are long-context/video-style positional issues, but one is Qwen2.5-VL RoPE scaling and the other is Kosmos2.5 OCR index overflow." + }, + { + "left": "issue:43454", + "right": "issue:44466", + "accept": false, + "reason": "AyaVision lm_head tying and inconsistent lm_head serialization are related only loosely; they are not the same concrete bug." + }, + { + "left": "issue:42915", + "right": "issue:43482", + "accept": false, + "reason": "Qwen3Moe FP8 failure and Qwen2.5-GGUF loading under v5 are unrelated." + }, + { + "left": "issue:43701", + "right": "issue:45325", + "accept": false, + "reason": "resume_from_checkpoint key mismatch is unrelated to Qwen2.5-VL position-id scaling." + }, + { + "left": "issue:44479", + "right": "issue:45823", + "accept": false, + "reason": "Qwen video-input regression and Gemma4 PLE device mismatch are different model-specific failures." + }, + { + "left": "issue:44291", + "right": "issue:44877", + "accept": false, + "reason": "init_empty_weights TypeError and strict config blocking granite_speech config are different issues." + }, + { + "left": "issue:43873", + "right": "issue:45701", + "accept": false, + "reason": "Quantization/offloading behavior and tokenization changes across versions are not duplicates." + }, + { + "left": "issue:43606", + "right": "issue:44265", + "accept": false, + "reason": "CPU offload device mismatch and torch.export/torch_compilable_check failure are unrelated." + }, + { + "left": "issue:43296", + "right": "issue:45800", + "accept": false, + "reason": "PaddleOCR-VL load failure in vLLM and torch 2.4.1/transformers 5.8 incompatibility are different problems." + }, + { + "left": "issue:43262", + "right": "issue:43746", + "accept": false, + "reason": "Audio chat-template sample-rate defaulting and GraniteSpeech PEFT checkpoint loading 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": 19, + "review_unit_index": 12, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:34634", + "issue:35707", + "issue:39692", + "issue:40444", + "issue:40990", + "issue:41720", + "issue:42617", + "issue:42915", + "issue:42947", + "issue:43257", + "issue:43317", + "issue:43322", + "issue:43377", + "issue:43454", + "issue:43526", + "issue:43550", + "issue:43572", + "issue:43606", + "issue:43720", + "issue:43749", + "issue:43819", + "issue:43873", + "issue:43881", + "issue:43883", + "issue:44186", + "issue:44265", + "issue:44466", + "issue:44488", + "issue:44534", + "issue:44568", + "issue:44610", + "issue:44898", + "issue:44977", + "issue:44991", + "issue:45072", + "issue:45325", + "issue:45357", + "issue:45584", + "issue:45701", + "issue:45759", + "issue:45823", + "issue:45920", + "issue:45923", + "issue:46050", + "issue:46291", + "issue:46305", + "issue:46315" + ], + "soft_pairs": [ + "issue:40990|issue:45584", + "issue:34634|issue:35707", + "issue:35707|issue:41720", + "issue:44568|issue:45920", + "issue:44186|issue:44265", + "issue:41720|issue:45823", + "issue:43606|issue:43819", + "issue:43454|issue:43550", + "issue:39692|issue:40444", + "issue:43881|issue:44488", + "issue:46305|issue:46315", + "issue:43377|issue:44610", + "issue:43454|issue:45072", + "issue:43257|issue:45923", + "issue:43572|issue:45759", + "issue:43317|issue:43873", + "issue:40990|issue:42947", + "issue:43377|issue:44898", + "issue:43322|issue:46050", + "issue:43454|issue:43720", + "issue:41720|issue:43322", + "issue:45325|issue:46315", + "issue:43749|issue:44534", + "issue:44991|issue:46291", + "issue:42617|issue:42915", + "issue:41720|issue:44977", + "issue:44466|issue:45357", + "issue:43526|issue:43883", + "issue:40990|issue:45701", + "issue:43881|issue:44991" + ], + "prepared_review_unit_hash": "a95972fe420ce92760dbf0d0f1010eb0b6a5d4ba370778af94ce55603fe55465", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 36234, + "estimated_input_tokens": 9059, + "estimated_eval_tokens": 18374 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No strong duplicate cluster emerges here. The soft links are mostly broad thematic overlaps (device_map/offloading, tokenizer regressions, serialization, generation) rather than the same underlying bug. No PRs are present.", + "confidence": 0.94, + "canonical_issue_reason": "issue:41720 is the clearest concrete runtime regression in the set and the most central anchor for the device-mapping/offloading subgroup, even though the neighboring reports are not true duplicates.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:41720 is the best single issue to anchor triage because it is specific, reproducible, and has the strongest thematic overlap with several other candidates, but it should not absorb the others as duplicates.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:40990", + "right": "issue:45584", + "accept": false, + "reason": "Different bugs: high perplexity on WikiText-2 vs Whisper empty-transcription generation after token alignment." + }, + { + "left": "issue:34634", + "right": "issue:35707", + "accept": false, + "reason": "Processor voice_preset bug and progressive generation with inputs_embeds/past_key_values are unrelated code paths." + }, + { + "left": "issue:35707", + "right": "issue:41720", + "accept": false, + "reason": "Generation cache behavior and Qwen3 auto device-map CUDA assert are different failures." + }, + { + "left": "issue:44568", + "right": "issue:45920", + "accept": false, + "reason": "One is a missing BOS/EOS insertion bug for a specific tokenizer; the other is a broad wrong-token-IDs regression across multiple models." + }, + { + "left": "issue:44186", + "right": "issue:44265", + "accept": false, + "reason": "Tokenizer NER padding/truncation crash is not the same as torch.export.export failing on torch_compilable_check." + }, + { + "left": "issue:41720", + "right": "issue:45823", + "accept": false, + "reason": "Both involve device_map=auto, but they affect different models and fail differently (CUDA assert vs device mismatch in forward)." + }, + { + "left": "issue:43606", + "right": "issue:43819", + "accept": false, + "reason": "Bark CPU offload mismatch and DAC.from_latents missing STE are separate model/runtime bugs." + }, + { + "left": "issue:43454", + "right": "issue:43550", + "accept": false, + "reason": "Weight tying/garbage text generation is unrelated to torch.compile + SDPA failures." + }, + { + "left": "issue:39692", + "right": "issue:40444", + "accept": false, + "reason": "Documentation example errors and multi-image IterableDataset finetuning failure are different problems." + }, + { + "left": "issue:43881", + "right": "issue:44488", + "accept": false, + "reason": "Different models with different load failures; no shared concrete code-path." + }, + { + "left": "issue:46305", + "right": "issue:46315", + "accept": false, + "reason": "Sliding-window off-by-one in masking is unrelated to tokenizer conversion performance regression." + }, + { + "left": "issue:43377", + "right": "issue:44610", + "accept": false, + "reason": "Missing padding-mask support in MIMI encoder and OmDet processor/model size mismatch are not the same bug." + }, + { + "left": "issue:43454", + "right": "issue:45072", + "accept": false, + "reason": "lm_head tying/garbage generation is unrelated to bfloat16 dtype mismatches in inference." + }, + { + "left": "issue:43257", + "right": "issue:45923", + "accept": false, + "reason": "MOE weight conversion with accelerate+deepspeed and a missing supports_gradient_checkpointing flag are unrelated." + }, + { + "left": "issue:43572", + "right": "issue:45759", + "accept": false, + "reason": "Missing pad_token_idx in StableLmConfig is unrelated to unwrap-text_config behavior for composite Qwen configs." + }, + { + "left": "issue:43317", + "right": "issue:43873", + "accept": false, + "reason": "Closely themed around offloading/quantization, but not the same concrete failure; 43873 is a broad umbrella report." + }, + { + "left": "issue:40990", + "right": "issue:42947", + "accept": false, + "reason": "Perplexity regression on GPT OSS is unrelated to gradient checkpointing not taking effect with PEFT LoRA." + }, + { + "left": "issue:43377", + "right": "issue:44898", + "accept": false, + "reason": "Batched-padding discrepancy in MIMI and Perceiver non-default-resolution failure are different issues." + }, + { + "left": "issue:43322", + "right": "issue:46050", + "accept": false, + "reason": "Segmentation fault when loading Llava Next and sinq save/load corruption are distinct load-path bugs." + }, + { + "left": "issue:43454", + "right": "issue:43720", + "accept": false, + "reason": "Weight tying problem and BitNet packed-weight unpacking during accelerate loading are unrelated." + }, + { + "left": "issue:41720", + "right": "issue:43322", + "accept": false, + "reason": "Both are model-loading/runtime failures, but the failure modes and affected models are different." + }, + { + "left": "issue:45325", + "right": "issue:46315", + "accept": false, + "reason": "Qwen2.5-VL rope-index scaling bug is unrelated to tokenizer conversion speed regression." + }, + { + "left": "issue:43749", + "right": "issue:44534", + "accept": false, + "reason": "FSDP CPU RAM efficient loading and junk-filled non-persistent buffers are separate regressions." + }, + { + "left": "issue:44991", + "right": "issue:46291", + "accept": false, + "reason": "Tokenizer loading failure and kernels v0.15 import/version-revision failure do not share the same bug." + }, + { + "left": "issue:42617", + "right": "issue:42915", + "accept": false, + "reason": "3d_parallel.py execution failure and FineGrainedFP8Config failure are different issues." + }, + { + "left": "issue:41720", + "right": "issue:44977", + "accept": false, + "reason": "Same model family, but different symptoms and likely different code paths (CUDA assert vs flash-attention generation issue)." + }, + { + "left": "issue:44466", + "right": "issue:45357", + "accept": false, + "reason": "Both concern saving/serialization, but one is tied lm_head.weight behavior and the other is incorrect visual encoder keys." + }, + { + "left": "issue:43526", + "right": "issue:43883", + "accept": false, + "reason": "BeitImageProcessorFast label reduction bug and Molmo missing all_tied_weights_keys are unrelated." + }, + { + "left": "issue:40990", + "right": "issue:45701", + "accept": false, + "reason": "Perplexity regression is not the same as version-dependent tokenization changes." + }, + { + "left": "issue:43881", + "right": "issue:44991", + "accept": false, + "reason": "Different models and different failure surfaces: model loading vs tokenizer loading." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 13, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:36010", + "issue:36331", + "issue:41720", + "issue:42175", + "issue:42617", + "issue:42915", + "issue:43065", + "issue:43295", + "issue:43316", + "issue:43317", + "issue:43404", + "issue:43425", + "issue:43450", + "issue:43525", + "issue:43531", + "issue:43550", + "issue:43572", + "issue:43577", + "issue:43606", + "issue:43708", + "issue:43720", + "issue:43873", + "issue:44008", + "issue:44186", + "issue:44387", + "issue:44423", + "issue:44610", + "issue:44671", + "issue:44743", + "issue:44756", + "issue:44877", + "issue:44898", + "issue:44913", + "issue:44991", + "issue:45072", + "issue:45081", + "issue:45161", + "issue:45357", + "issue:45406", + "issue:45632", + "issue:45684", + "issue:45701", + "issue:45704", + "issue:45812", + "issue:45823", + "issue:46291", + "issue:46402" + ], + "soft_pairs": [ + "issue:44387|issue:45704", + "issue:36331|issue:42617", + "issue:43065|issue:46402", + "issue:44186|issue:45072", + "issue:43577|issue:43708", + "issue:44008|issue:45823", + "issue:44877|issue:44991", + "issue:36331|issue:42915", + "issue:43450|issue:45701", + "issue:43873|issue:44756", + "issue:43550|issue:43720", + "issue:43404|issue:45357", + "issue:42175|issue:46291", + "issue:41720|issue:42617", + "issue:44423|issue:45406", + "issue:43550|issue:43606", + "issue:44610|issue:44898", + "issue:44186|issue:44610", + "issue:42915|issue:43317", + "issue:43295|issue:45701", + "issue:45632|issue:45684", + "issue:43425|issue:46291", + "issue:36010|issue:42617", + "issue:41720|issue:45161", + "issue:43525|issue:46291", + "issue:45081|issue:45812", + "issue:44671|issue:44743", + "issue:43531|issue:43572", + "issue:43316|issue:44913" + ], + "prepared_review_unit_hash": "a6b9e66582ea78f13181a010ad9bd07284df534df5e2bd50641768b19fcc7645", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 35945, + "estimated_input_tokens": 8987, + "estimated_eval_tokens": 18230 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items span many unrelated regressions across different subsystems. None of the soft pairs look like the same underlying bug or change, so I would reject all duplicate merges. If an anchor is needed, issue 43873 is the broadest discussion-heavy issue, but it is not a true duplicate match for the others.", + "confidence": 0.93, + "canonical_issue_reason": "Issue 43873 is the broadest and most actively discussed item here, with multiple inbound references and a recurring offloading/quantization theme, so it is the best available cluster anchor.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43873 is the most suitable representative issue because it has the widest scope and the strongest evidence of ongoing community impact, even though this set does not contain a clear duplicate group.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44387", + "right": "issue:45704", + "accept": false, + "reason": "Different problems: int4 quantization OOM vs T5/apex RMSNorm memory leak." + }, + { + "left": "issue:36331", + "right": "issue:42617", + "accept": false, + "reason": "Trainer custom compute_loss keyword mismatch is unrelated to failing to run a 3d_parallel script." + }, + { + "left": "issue:43065", + "right": "issue:46402", + "accept": false, + "reason": "Different SAM3 bugs: dummy Conv2d vs text encoder weights not loading." + }, + { + "left": "issue:44186", + "right": "issue:45072", + "accept": false, + "reason": "Tokenizer crash on NER/padding is unrelated to dtype mismatch in bfloat16 inference." + }, + { + "left": "issue:43577", + "right": "issue:43708", + "accept": false, + "reason": "BLIP2 dtype propagation bug is unrelated to Trainer resume/max_steps recalculation." + }, + { + "left": "issue:44008", + "right": "issue:45823", + "accept": false, + "reason": "Different Gemma models and failures: variable-name collision in forward vs PLE device mismatch." + }, + { + "left": "issue:44877", + "right": "issue:44991", + "accept": false, + "reason": "Strict config loading for granite_speech is unrelated to tokenizer loading for EMBEDDIA/est-roberta." + }, + { + "left": "issue:36331", + "right": "issue:42915", + "accept": false, + "reason": "CustomTrainer loss signature mismatch is unrelated to Qwen3MoE FP8 failure." + }, + { + "left": "issue:43450", + "right": "issue:45701", + "accept": false, + "reason": "Video processor batched-shape bug is unrelated to a broad tokenization version-change report." + }, + { + "left": "issue:43873", + "right": "issue:44756", + "accept": false, + "reason": "Both involve loading/offloading, but one is quantization offload behavior and the other is mmap OOM on Strix Halo; not the same bug." + }, + { + "left": "issue:43550", + "right": "issue:43720", + "accept": false, + "reason": "Both are BitNet/SDPA vs packed-weight-unpacking issues; different failure paths and fixes." + }, + { + "left": "issue:43404", + "right": "issue:45357", + "accept": false, + "reason": "Mistral3 lm_head tying bug is unrelated to Qwen3.5 visual encoder key saving." + }, + { + "left": "issue:42175", + "right": "issue:46291", + "accept": false, + "reason": "TensorFlow backend packaging issue is unrelated to kernels version/revision import failure in evaluate." + }, + { + "left": "issue:41720", + "right": "issue:42617", + "accept": false, + "reason": "Qwen3 auto device-map cuda assert is unrelated to a script execution failure." + }, + { + "left": "issue:44423", + "right": "issue:45406", + "accept": false, + "reason": "Continuous-batching multimodal crash is unrelated to Gemma4Processor missing _tokenizer in serve." + }, + { + "left": "issue:43550", + "right": "issue:43606", + "accept": false, + "reason": "Different subsystems and symptoms: torch.compile+SDPA in Bamba vs CPU offload device mismatch in bark-small." + }, + { + "left": "issue:44610", + "right": "issue:44898", + "accept": false, + "reason": "OmDet processor output size bug is different from Perceiver non-default resolution failure." + }, + { + "left": "issue:44186", + "right": "issue:44610", + "accept": false, + "reason": "Tokenizer/NER padding crash and image processor resolution mismatch are unrelated." + }, + { + "left": "issue:42915", + "right": "issue:43317", + "accept": false, + "reason": "Qwen3Moe FP8 failure is unrelated to dequantized model loading with auto device_map offload." + }, + { + "left": "issue:43295", + "right": "issue:45701", + "accept": false, + "reason": "Processor/tokenizer regression is unrelated to a general tokenization-version report." + }, + { + "left": "issue:45632", + "right": "issue:45684", + "accept": false, + "reason": "Both touch save/load paths, but cache-path collisions and read-only permission propagation are different bugs." + }, + { + "left": "issue:43425", + "right": "issue:46291", + "accept": false, + "reason": "Torch 2.10 incompatibility is unrelated to kernels version/revision requirements." + }, + { + "left": "issue:36010", + "right": "issue:42617", + "accept": false, + "reason": "GenerationMixin import error is unrelated to failing to run 3d_parallel.py." + }, + { + "left": "issue:41720", + "right": "issue:45161", + "accept": false, + "reason": "Both mention model dispatching, but one is a cuda assert on Qwen3 and the other is only-TP not working on GPT-OSS MoE." + }, + { + "left": "issue:43525", + "right": "issue:46291", + "accept": false, + "reason": "Missing pad_token_id in Llama4Config is unrelated to kernels import/version requirements." + }, + { + "left": "issue:45081", + "right": "issue:45812", + "accept": false, + "reason": "Mistral regex patch crash is unrelated to Granite token ID regression." + }, + { + "left": "issue:44671", + "right": "issue:44743", + "accept": false, + "reason": "CamemBERT masked-LM predictions and Qwen3.5 recurrent-state handling are different code paths and failures." + }, + { + "left": "issue:43531", + "right": "issue:43572", + "accept": false, + "reason": "Qwen3-MoE sliding-window behavior is unrelated to StableLmConfig missing pad_token_idx." + }, + { + "left": "issue:43316", + "right": "issue:44913", + "accept": false, + "reason": "Gemma3TextConfig API mismatch is unrelated to GPTNeoX rotary_pct reload regression." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 14, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:29159", + "issue:30064", + "issue:33357", + "issue:34567", + "issue:34634", + "issue:35707", + "issue:36032", + "issue:36331", + "issue:41720", + "issue:42490", + "issue:42915", + "issue:43322", + "issue:43329", + "issue:43388", + "issue:43421", + "issue:43526", + "issue:43531", + "issue:43540", + "issue:43550", + "issue:43606", + "issue:43653", + "issue:43720", + "issue:43746", + "issue:43782", + "issue:43873", + "issue:43881", + "issue:44186", + "issue:44361", + "issue:44451", + "issue:44492", + "issue:44509", + "issue:44514", + "issue:44530", + "issue:44545", + "issue:44589", + "issue:44625", + "issue:44991", + "issue:45059", + "issue:45357", + "issue:45406", + "issue:45464", + "issue:45479", + "issue:45701", + "issue:45704", + "issue:45759", + "issue:45812", + "issue:46291", + "issue:46402" + ], + "soft_pairs": [ + "issue:43540|issue:43782", + "issue:42915|issue:43322", + "issue:44186|issue:44361", + "issue:43421|issue:45812", + "issue:36331|issue:41720", + "issue:43526|issue:43881", + "issue:29159|issue:36032", + "issue:33357|issue:35707", + "issue:43388|issue:45059", + "issue:30064|issue:34634", + "issue:43550|issue:43653", + "issue:43606|issue:43720", + "issue:45406|issue:45464", + "issue:43873|issue:45704", + "issue:41720|issue:43746", + "issue:44530|issue:46291", + "issue:43421|issue:45701", + "issue:45357|issue:46402", + "issue:43606|issue:43653", + "issue:44492|issue:44509", + "issue:43531|issue:45759", + "issue:34634|issue:41720", + "issue:42490|issue:43329", + "issue:34567|issue:35707", + "issue:44514|issue:44545", + "issue:44625|issue:45479", + "issue:44451|issue:44991", + "issue:43531|issue:44589" + ], + "prepared_review_unit_hash": "f6cc990cdb85435f074956b0b2f28365e4cd453d282d782424a3909e51fe0248", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36229, + "estimated_input_tokens": 9058, + "estimated_eval_tokens": 18372 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Only one soft edge looks like a true duplicate: the two Qwen2_5_VLProcessor.apply_chat_template crash reports. The rest are distinct bugs across different components and should stay separate.", + "confidence": 0.96, + "canonical_issue_reason": "issue:44514 is the cleanest canonical anchor for the batched apply_chat_template crash: it was filed slightly earlier and describes the failure mode directly. issue:44545 is essentially the same report with a near-identical title.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44514 best represents the duplicated Qwen2_5_VLProcessor.padding=False crash because it is specific, actionable, and the duplicate pair maps to the same underlying code path.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43540", + "right": "issue:43782", + "accept": false, + "reason": "Different models and different failure modes: Qwen3OmniMoe video-input ValueError vs Qwen3VL from_pretrained weight_only load error." + }, + { + "left": "issue:42915", + "right": "issue:43322", + "accept": false, + "reason": "Different multimodal model families and distinct load-time errors; no shared code-path evidence." + }, + { + "left": "issue:44186", + "right": "issue:44361", + "accept": false, + "reason": "Both are tokenizer-related but the bugs differ: NER/batched padding crash in LayoutLMv2Tokenizer vs AttributeError in MLukeTokenizer." + }, + { + "left": "issue:43421", + "right": "issue:45812", + "accept": false, + "reason": "Both involve tokenizers post-processing, but one is runtime special-token updates while the other is a Granite AutoTokenizer regression producing wrong IDs." + }, + { + "left": "issue:36331", + "right": "issue:41720", + "accept": false, + "reason": "Unrelated areas: trainer compute_loss signature mismatch vs Qwen3 auto device mapping CUDA assert." + }, + { + "left": "issue:43526", + "right": "issue:43881", + "accept": false, + "reason": "Different bugs in different models: BeitImageProcessorFast label reduction vs glm-4v-9b loading failure." + }, + { + "left": "issue:29159", + "right": "issue:36032", + "accept": false, + "reason": "Tokenizer inconsistency vs T5Tokenizer class/method conflict; these are not the same bug." + }, + { + "left": "issue:33357", + "right": "issue:35707", + "accept": false, + "reason": "A MacOS bus error on a CLIP community model is unrelated to generation with inputs_embeds and past_key_values." + }, + { + "left": "issue:43388", + "right": "issue:45059", + "accept": false, + "reason": "Different subsystems and symptoms: gather_for_metrics label truncation vs SAM3 text/bbox behavior." + }, + { + "left": "issue:30064", + "right": "issue:34634", + "accept": false, + "reason": "Both affect processors, but one is void segmentation-map handling and the other is BarkProcessor voice_preset support." + }, + { + "left": "issue:43550", + "right": "issue:43653", + "accept": false, + "reason": "Different code paths: torch.compile+SDPA failure in Bamba vs tokenizer mask-token registration/empty decode issue in BigBirdTokenizer." + }, + { + "left": "issue:43606", + "right": "issue:43720", + "accept": false, + "reason": "Both are CI/loading failures, but one is CPU offload device mismatch and the other is packed-weight unpacking in accelerate loading." + }, + { + "left": "issue:45406", + "right": "issue:45464", + "accept": false, + "reason": "Different crash surfaces: Gemma4Processor missing _tokenizer in serve vs chat/completions streaming failure on Qwen3.5-0.8B." + }, + { + "left": "issue:43873", + "right": "issue:45704", + "accept": false, + "reason": "Quantization/offloading behavior is broad, but these are unrelated specific bugs: offloading with quantization vs T5 using apex.FusedRMSNorm memory leak." + }, + { + "left": "issue:41720", + "right": "issue:43746", + "accept": false, + "reason": "Different model families and causes: Qwen3 auto device mapping assert vs GraniteSpeech PEFT local checkpoint loading." + }, + { + "left": "issue:44530", + "right": "issue:46291", + "accept": false, + "reason": "PagedAttentionCache linear_attention group error is unrelated to kernels import/version requirements in evaluate." + }, + { + "left": "issue:43421", + "right": "issue:45701", + "accept": false, + "reason": "Both involve tokenization behavior broadly, but one is runtime post-processor updates and the other is a version-dependent tokenization difference report." + }, + { + "left": "issue:45357", + "right": "issue:46402", + "accept": false, + "reason": "Both concern Qwen3.5 checkpoints, but one is save_pretrained visual encoder keys and the other is loading text encoder weights; different operations and failures." + }, + { + "left": "issue:43606", + "right": "issue:43653", + "accept": false, + "reason": "No shared underlying bug: CPU offload device mismatch vs BigBirdTokenizer special-token registration." + }, + { + "left": "issue:44492", + "right": "issue:44509", + "accept": false, + "reason": "A typo in cache strategies is unrelated to documentation still mentioning removed pipeline tasks." + }, + { + "left": "issue:43531", + "right": "issue:45759", + "accept": false, + "reason": "Different Qwen MoE configuration issues: sliding_window support vs from_config not unwrapping text_config." + }, + { + "left": "issue:34634", + "right": "issue:41720", + "accept": false, + "reason": "Different symptoms and model families: BarkProcessor voice_preset vs Qwen3 auto device mapping assert." + }, + { + "left": "issue:42490", + "right": "issue:43329", + "accept": false, + "reason": "Local-path trailing-slash import warning is unrelated to a video-branch undefined-variable bug in multimodal token counting." + }, + { + "left": "issue:34567", + "right": "issue:35707", + "accept": false, + "reason": "TrainerState token counting is unrelated to progressive generation with inputs_embeds/past_key_values." + }, + { + "left": "issue:44514", + "right": "issue:44545", + "accept": true, + "reason": "These are the same underlying crash report: Qwen2_5_VLProcessor.apply_chat_template fails on batched input when padding=False." + }, + { + "left": "issue:44625", + "right": "issue:45479", + "accept": false, + "reason": "Different classification bugs: num_labels propagation in Qwen3.5 vs degenerate zero loss across sequence-classification models." + }, + { + "left": "issue:44451", + "right": "issue:44991", + "accept": false, + "reason": "Both are tokenizer loading regressions, but they affect different checkpoints and likely different root causes." + }, + { + "left": "issue:43531", + "right": "issue:44589", + "accept": false, + "reason": "Different model/config problems: Qwen3-MoE sliding_window vs missing Float8 storage type during loading." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The only accepted soft edge, issue:44514 \u2194 issue:44545, is a clear duplicate by title and timing. The remaining pairs are reasonably rejected as distinct bugs based on their titles and scopes, without overstating shared root cause evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 15, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:30064", + "issue:36010", + "issue:39692", + "issue:41720", + "issue:42886", + "issue:42915", + "issue:43381", + "issue:43421", + "issue:43531", + "issue:43540", + "issue:43596", + "issue:43645", + "issue:43653", + "issue:43746", + "issue:43782", + "issue:43844", + "issue:43873", + "issue:43927", + "issue:43994", + "issue:44062", + "issue:44077", + "issue:44186", + "issue:44220", + "issue:44368", + "issue:44451", + "issue:44610", + "issue:44625", + "issue:44671", + "issue:44843", + "issue:44898", + "issue:44918", + "issue:44936", + "issue:44987", + "issue:45042", + "issue:45072", + "issue:45081", + "issue:45356", + "issue:45357", + "issue:45381", + "issue:45588", + "issue:45663", + "issue:45704", + "issue:45759", + "issue:45859", + "issue:45907", + "issue:45923", + "issue:46050", + "issue:46294" + ], + "soft_pairs": [ + "issue:44898|issue:45072", + "issue:44077|issue:44625", + "issue:45759|issue:45859", + "issue:43421|issue:43927", + "issue:45588|issue:45759", + "issue:44062|issue:44451", + "issue:44368|issue:45357", + "issue:44987|issue:45042", + "issue:44918|issue:44936", + "issue:43782|issue:45381", + "issue:39692|issue:44610", + "issue:43746|issue:46050", + "issue:42886|issue:44843", + "issue:45704|issue:45923", + "issue:43381|issue:43844", + "issue:43421|issue:43653", + "issue:45663|issue:45923", + "issue:45704|issue:45859", + "issue:41720|issue:45704", + "issue:43531|issue:43645", + "issue:43994|issue:44671", + "issue:43421|issue:44186", + "issue:43746|issue:43873", + "issue:45081|issue:45356", + "issue:39692|issue:43873", + "issue:43540|issue:46294", + "issue:43596|issue:44220", + "issue:42915|issue:43746", + "issue:45356|issue:45907", + "issue:30064|issue:36010" + ], + "prepared_review_unit_hash": "bf9844fa89c752951063787a1e740edb163382279cc2e30139c9bf4e35e187cf", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 37078, + "estimated_input_tokens": 9270, + "estimated_eval_tokens": 18796 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly heterogeneous, with one strong duplicate pair around offline/Mistral-regex tokenizer loading (42886 \u2194 44843). A second Mistral-regex pair (45081 \u2194 45356) is related but too different in failure mode to treat as the same bug.", + "confidence": 0.84, + "canonical_issue_reason": "44843 is the best canonical issue for the only clearly duplicate subgroup: it names the concrete HF_HUB_OFFLINE regression in _patch_mistral_regex and matches 42886 closely.", + "canonical_pr_reason": null, + "best_issue_reason": "44843 is the strongest standalone issue here because it captures the specific offline tokenizer regression and serves as the clearest representative of the duplicate subgroup.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44898", + "right": "issue:45072", + "accept": false, + "reason": "Different models and failure modes: image-size/interpolation behavior vs dtype mismatch in bfloat16." + }, + { + "left": "issue:44077", + "right": "issue:44625", + "accept": false, + "reason": "Unrelated config issues: optional post_init validation vs num_labels propagation." + }, + { + "left": "issue:45759", + "right": "issue:45859", + "accept": false, + "reason": "Different code paths: composite config unwrapping vs CPU-offload incompatibility in rotary embedding." + }, + { + "left": "issue:43421", + "right": "issue:43927", + "accept": false, + "reason": "Both tokenizer-related, but one is runtime post-processor updates and the other is save/load losing custom token IDs." + }, + { + "left": "issue:45588", + "right": "issue:45759", + "accept": false, + "reason": "Flash-attention sinkless-model crash is unrelated to Qwen config unwrapping." + }, + { + "left": "issue:44062", + "right": "issue:44451", + "accept": false, + "reason": "AddedToken keyword conflict is not the same as failing to load a specific model." + }, + { + "left": "issue:44368", + "right": "issue:45357", + "accept": false, + "reason": "Different regressions: LoRA warning about tie_word_embeddings vs incorrect visual encoder keys on save_pretrained." + }, + { + "left": "issue:44987", + "right": "issue:45042", + "accept": false, + "reason": "Loading a specific model failing is not the same bug as PIL image processors requiring torchvision." + }, + { + "left": "issue:44918", + "right": "issue:44936", + "accept": false, + "reason": "Different trainer failures: embedding unpacking in SFT vs evaluate() after train()." + }, + { + "left": "issue:43782", + "right": "issue:45381", + "accept": false, + "reason": "Different multimodal bugs: weight_only loading error vs wrong video vision_position_ids." + }, + { + "left": "issue:39692", + "right": "issue:44610", + "accept": false, + "reason": "Both mention processor/model mismatch, but they involve different models and different failures." + }, + { + "left": "issue:43746", + "right": "issue:46050", + "accept": false, + "reason": "Local checkpoint loading with PEFT adapters is not the same as quantized-model save/load corruption." + }, + { + "left": "issue:42886", + "right": "issue:44843", + "accept": true, + "reason": "Same offline tokenizer regression: HF_HUB_OFFLINE breaks tokenizer loading via the unconditional model_info call in the Mistral regex path." + }, + { + "left": "issue:45704", + "right": "issue:45923", + "accept": false, + "reason": "Apex RMSNorm memory leak and missing supports_gradient_checkpointing are unrelated bugs." + }, + { + "left": "issue:43381", + "right": "issue:43844", + "accept": false, + "reason": "Eval-mode checkpointing restriction is a different issue from abnormal gradient growth under ZeRO-3." + }, + { + "left": "issue:43421", + "right": "issue:43653", + "accept": false, + "reason": "Special-token post-processor updating and BigBird mask-token registration are related, but not the same underlying bug." + }, + { + "left": "issue:45663", + "right": "issue:45923", + "accept": false, + "reason": "FSDP2 shared_kv_states KeyError is unrelated to the gradient-checkpointing capability flag." + }, + { + "left": "issue:45704", + "right": "issue:45859", + "accept": false, + "reason": "Apex RMSNorm memory leak and CPU-offload incompatibility in Qwen rotary embedding are unrelated." + }, + { + "left": "issue:41720", + "right": "issue:45704", + "accept": false, + "reason": "Auto device mapping cudaErrorAssert and an RMSNorm memory leak are different bugs." + }, + { + "left": "issue:43531", + "right": "issue:43645", + "accept": false, + "reason": "Sliding-window behavior in Qwen3-MoE is unrelated to Jupyter custom-model initialization." + }, + { + "left": "issue:43994", + "right": "issue:44671", + "accept": false, + "reason": "Both are incorrect-output reports, but on different models and mechanisms." + }, + { + "left": "issue:43421", + "right": "issue:44186", + "accept": false, + "reason": "Token post-processor updates and LayoutLMv2 NER/padding crashes are distinct tokenizer issues." + }, + { + "left": "issue:43746", + "right": "issue:43873", + "accept": false, + "reason": "PEFT local checkpoint loading is unrelated to quantization/offloading behavior." + }, + { + "left": "issue:45081", + "right": "issue:45356", + "accept": false, + "reason": "Both touch the Mistral-regex/tokenizer path, but the actual failures differ: backend_tokenizer AttributeError crash vs Kimi-K2.5 codec/warning regression." + }, + { + "left": "issue:39692", + "right": "issue:43873", + "accept": false, + "reason": "SigLIP2 doc/example problems are unrelated to quantization/offloading not working as expected." + }, + { + "left": "issue:43540", + "right": "issue:46294", + "accept": false, + "reason": "Both are video-input issues, but they affect different Qwen families and different failure points." + }, + { + "left": "issue:43596", + "right": "issue:44220", + "accept": false, + "reason": "Deepspeed zero3/BertModel indexing and _torch_extract_fbank_features() are unrelated." + }, + { + "left": "issue:42915", + "right": "issue:43746", + "accept": false, + "reason": "Qwen3Moe FP8 configuration failure is unrelated to GraniteSpeech PEFT checkpoint loading." + }, + { + "left": "issue:45356", + "right": "issue:45907", + "accept": false, + "reason": "Kimi-K2.5 tokenizer regression is unrelated to list_repo_files not forwarding revision." + }, + { + "left": "issue:30064", + "right": "issue:36010", + "accept": false, + "reason": "Void segmentation map processing and a GenerationMixin import error are completely different." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The clustering is mostly grounded, and 42886\u219444843 is a defensible duplicate link. But the accept on 45081\u219445356 is too strong for the packet: the titles show related Mistral/fix_mistral_regex code paths, yet one is a backend_tokenizer AttributeError crash and the other is a Kimi-K2.5 codec/warning regression. That is similar enough to group softly, but not enough evidence to call it a true duplicate with confidence. The overall summary should stay more conservative." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 16, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "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:43618", + "issue:43643", + "issue:43653", + "issue:43697", + "issue:43716", + "issue:43720", + "issue:43756", + "issue:43818", + "issue:43819", + "issue:43950", + "issue:44008", + "issue:44062", + "issue:44186", + "issue:44220", + "issue:44448", + "issue:44488", + "issue:44661", + "issue:44671", + "issue:44811", + "issue:44821", + "issue:44843", + "issue:44877", + "issue:44898", + "issue:44991", + "issue:45072", + "issue:45081", + "issue:45084", + "issue:45183", + "issue:45381", + "issue:45479", + "issue:45685", + "issue:45834", + "issue:45987", + "issue:46032" + ], + "soft_pairs": [ + "issue:43720|issue:43819", + "issue:43653|issue:43720", + "issue:44220|issue:44661", + "issue:43550|issue:45072", + "issue:43540|issue:44008", + "issue:43295|issue:45834", + "issue:43618|issue:43697", + "issue:44062|issue:44488", + "issue:43550|issue:43819", + "issue:45183|issue:45479", + "issue:44186|issue:44898", + "issue:43322|issue:45685", + "issue:43716|issue:44008", + "issue:43653|issue:43819", + "issue:43531|issue:43643", + "issue:43388|issue:46032", + "issue:43335|issue:43756", + "issue:43950|issue:45987", + "issue:43643|issue:44877", + "issue:44843|issue:45081", + "issue:30064|issue:36331", + "issue:44448|issue:44671", + "issue:41720|issue:44991", + "issue:43540|issue:45381", + "issue:43421|issue:44811", + "issue:44821|issue:45084", + "issue:43577|issue:43818", + "issue:39692|issue:45834" + ], + "prepared_review_unit_hash": "6cf40e4c047972f2bde3ad60fcac2c15679e27eb684011772e0d36491a59fe25", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 35960, + "estimated_input_tokens": 8990, + "estimated_eval_tokens": 18236 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab-bag of unrelated bug reports spanning tokenizers, multimodal models, cache/dtype issues, and model-specific regressions. None of the soft pairs look like the same underlying bug or change, so I would not merge any of them into a duplicate cluster.", + "confidence": 0.98, + "canonical_issue_reason": "No single canonical issue fits this cluster: the items target different models, different code paths, and different failure modes.", + "canonical_pr_reason": null, + "best_issue_reason": "No global best issue; the reports are heterogeneous and none is a clear umbrella for the rest.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43720", + "right": "issue:43819", + "accept": false, + "reason": "BitNet packed-weight loading during accelerate vs DAC latent/STE mismatch; different models and different code paths." + }, + { + "left": "issue:43653", + "right": "issue:43720", + "accept": false, + "reason": "BigBirdTokenizer special-token decode bug is unrelated to BitNet packed-weight unpacking in accelerate loading." + }, + { + "left": "issue:44220", + "right": "issue:44661", + "accept": false, + "reason": "Audio fbank feature extraction vs add-new-model-like tokenizer mapping; unrelated failures." + }, + { + "left": "issue:43550", + "right": "issue:45072", + "accept": false, + "reason": "One is Bamba torch.compile with SDPA, the other is SwitchTransformers/TimmWrapperModel bfloat16 dtype mismatch; not the same bug." + }, + { + "left": "issue:43540", + "right": "issue:44008", + "accept": false, + "reason": "Qwen3OmniMoe video-input ValueError and Gemma3n audio_mel_mask name collision are different model-specific issues." + }, + { + "left": "issue:43295", + "right": "issue:45834", + "accept": false, + "reason": "processor.tokenizer regression in v4.57.5 is unrelated to Kosmos2.5 OCR index handling." + }, + { + "left": "issue:43618", + "right": "issue:43697", + "accept": false, + "reason": "Missing CLIPOutput attentions and RTDetrV2 output drift in v5 are different regressions." + }, + { + "left": "issue:44062", + "right": "issue:44488", + "accept": false, + "reason": "AddedToken special-arg duplication is unrelated to loading cjvt/sleng-bert." + }, + { + "left": "issue:43550", + "right": "issue:43819", + "accept": false, + "reason": "Bamba SDPA/torch.compile failure and DAC from_latents mismatch are unrelated." + }, + { + "left": "issue:45183", + "right": "issue:45479", + "accept": false, + "reason": "Input validation XOR messaging is unrelated to single-label-classification zero-loss behavior." + }, + { + "left": "issue:44186", + "right": "issue:44898", + "accept": false, + "reason": "LayoutLMv2Tokenizer NER/padding crash and Perceiver resize/interpolate issue are different subsystems." + }, + { + "left": "issue:43322", + "right": "issue:45685", + "accept": false, + "reason": "Llava Next loading segfault and MPS histogram int error do not share a concrete code-path." + }, + { + "left": "issue:43716", + "right": "issue:44008", + "accept": false, + "reason": "Mistral-3 image-preprocessor dtype mismatch and Gemma3n variable collision are separate bugs." + }, + { + "left": "issue:43653", + "right": "issue:43819", + "accept": false, + "reason": "BigBirdTokenizer mask-token decode bug and DAC latent/forward mismatch are unrelated." + }, + { + "left": "issue:43531", + "right": "issue:43643", + "accept": false, + "reason": "Qwen3-MoE sliding_window bug and trust_remote_code missing-fields regression are different issues." + }, + { + "left": "issue:43388", + "right": "issue:46032", + "accept": false, + "reason": "gather_for_metrics tuple-label trimming and Mamba2Mixer use_cache corruption are not the same underlying defect." + }, + { + "left": "issue:43335", + "right": "issue:43756", + "accept": false, + "reason": "Sparse-layer creation with zero layers and Smollm3 RoPE-layer dropping are distinct model-configuration bugs." + }, + { + "left": "issue:43950", + "right": "issue:45987", + "accept": false, + "reason": "Non-persistent buffer corruption on from_pretrained is unrelated to StaticCache.get_seq_length returning a tensor." + }, + { + "left": "issue:43643", + "right": "issue:44877", + "accept": false, + "reason": "Missing fields with trust_remote_code and strict config blocking granite_speech are separate config-loading problems." + }, + { + "left": "issue:44843", + "right": "issue:45081", + "accept": false, + "reason": "Both touch Mistral regex patching, but one is offline model_info calls and the other is backend_tokenizer attribute access; different failure points and fixes." + }, + { + "left": "issue:30064", + "right": "issue:36331", + "accept": false, + "reason": "Void segmentation maps in Image Processor vs CustomTrainer.compute_loss kwarg change are unrelated." + }, + { + "left": "issue:44448", + "right": "issue:44671", + "accept": false, + "reason": "Both mention v4/v5 output differences, but they concern different models and different prediction paths." + }, + { + "left": "issue:41720", + "right": "issue:44991", + "accept": false, + "reason": "Qwen3 auto-device-map CUDA assert and est-roberta tokenizer loading failure are unrelated." + }, + { + "left": "issue:43540", + "right": "issue:45381", + "accept": false, + "reason": "Qwen3OmniMoe video-input ValueError and Qwen2.5-VL vision_position_ids bug are model-specific and not the same fix." + }, + { + "left": "issue:43421", + "right": "issue:44811", + "accept": false, + "reason": "Runtime post-processor updates for special tokens and Whisper batch_decode skip_special_tokens behavior are different tokenizer features." + }, + { + "left": "issue:44821", + "right": "issue:45084", + "accept": false, + "reason": "AutoImageProcessor URL loading and template-node compilation errors are unrelated." + }, + { + "left": "issue:43577", + "right": "issue:43818", + "accept": false, + "reason": "BLIP2 dtype propagation and Video-LLaVA temporal-attention/weight-sharing issues are different model families and bugs." + }, + { + "left": "issue:39692", + "right": "issue:45834", + "accept": false, + "reason": "SigLIP2 doc example errors and Kosmos2.5 OCR index error are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 17, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:28218", + "issue:29159", + "issue:34634", + "issue:36010", + "issue:36331", + "issue:42175", + "issue:43262", + "issue:43577", + "issue:43582", + "issue:43606", + "issue:43644", + "issue:43720", + "issue:43749", + "issue:43782", + "issue:43819", + "issue:43825", + "issue:43827", + "issue:43906", + "issue:43994", + "issue:44186", + "issue:44291", + "issue:44442", + "issue:44479", + "issue:44485", + "issue:44509", + "issue:44561", + "issue:44610", + "issue:44671", + "issue:44805", + "issue:44857", + "issue:44936", + "issue:44977", + "issue:45003", + "issue:45020", + "issue:45072", + "issue:45081", + "issue:45325", + "issue:45372", + "issue:45406", + "issue:45584", + "issue:45701", + "issue:45834", + "issue:45987", + "issue:46192", + "issue:46209", + "issue:46285", + "issue:46326" + ], + "soft_pairs": [ + "issue:43782|issue:44977", + "issue:45584|issue:45834", + "issue:44186|issue:44442", + "issue:43644|issue:43749", + "issue:43606|issue:45072", + "issue:44479|issue:45834", + "issue:43825|issue:43994", + "issue:34634|issue:36010", + "issue:44805|issue:44936", + "issue:43577|issue:43582", + "issue:44485|issue:44509", + "issue:45987|issue:46192", + "issue:28218|issue:29159", + "issue:44561|issue:45020", + "issue:43720|issue:45072", + "issue:44610|issue:45072", + "issue:44671|issue:45701", + "issue:43782|issue:45325", + "issue:43825|issue:43827", + "issue:43582|issue:44857", + "issue:43906|issue:46209", + "issue:43749|issue:44291", + "issue:45081|issue:45406", + "issue:42175|issue:45081", + "issue:45372|issue:46326", + "issue:43262|issue:43819", + "issue:34634|issue:36331", + "issue:45003|issue:46285" + ], + "prepared_review_unit_hash": "f7b90a0e6c326257821ca03dfbb61cd3df486e89da0fe5d01c7f4fa68fc29141", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 35685, + "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 highly heterogeneous: most items are unrelated point bugs across tokenizers, model loading, training, vision/audio processors, and docs. All soft-similarity pairs look like false positives, so no merges are warranted.", + "confidence": 0.79, + "canonical_issue_reason": "issue:45020 is the broadest open-ended regression report in the set ('Recent transformers versions break models using `remote_code`'), so it is the closest thing to an umbrella issue; however, the cluster overall is mixed and it does not actually subsume the other reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45020 is the best single representative because it is the most generic version-regression report and could attract related breakages; the other issues are narrower and largely unrelated.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43782", + "right": "issue:44977", + "accept": false, + "reason": "Different failures: Qwen3VL `weight_only=True` loading error vs Qwen3.5 flash-attention generation bug." + }, + { + "left": "issue:45584", + "right": "issue:45834", + "accept": false, + "reason": "Whisper empty-transcription bug after `align_special_tokens` is unrelated to Kosmos2.5 long-OCR index errors." + }, + { + "left": "issue:44186", + "right": "issue:44442", + "accept": false, + "reason": "LayoutLMv2 tokenizer NER/padding crash and AutoTokenizer failing to load a different tokenizer are separate code paths." + }, + { + "left": "issue:43644", + "right": "issue:43749", + "accept": false, + "reason": "Non-persistent buffer junk initialization is not the same as FSDP CPU RAM efficient loading being broken." + }, + { + "left": "issue:43606", + "right": "issue:45072", + "accept": false, + "reason": "CPU-offload device mismatch for Bark is a different bug from dtype mismatches in SwitchTransformers/TimmWrapperModel." + }, + { + "left": "issue:44479", + "right": "issue:45834", + "accept": false, + "reason": "Qwen VL video-input regression and Kosmos2.5 OCR index error are unrelated model-specific failures." + }, + { + "left": "issue:43825", + "right": "issue:43994", + "accept": false, + "reason": "Wrong pipeline error messaging about removed translation tasks is not the same as SigLIP2 producing bad outputs." + }, + { + "left": "issue:34634", + "right": "issue:36010", + "accept": false, + "reason": "BarkProcessor `voice_preset` issue and `GenerationMixin` import failure are unrelated." + }, + { + "left": "issue:44805", + "right": "issue:44936", + "accept": false, + "reason": "Mask shape/indexing error is a different failure than `trainer.evaluate()` breaking after `train()`." + }, + { + "left": "issue:43577", + "right": "issue:43582", + "accept": false, + "reason": "BLIP2 dtype staying float32 and Apple Silicon warmup `TypeError` are unrelated problems." + }, + { + "left": "issue:44485", + "right": "issue:44509", + "accept": false, + "reason": "GLM-5 RoPE discussion and docs still mentioning removed pipeline tasks are separate issues." + }, + { + "left": "issue:45987", + "right": "issue:46192", + "accept": false, + "reason": "`StaticCache.get_seq_length()` returning a tensor and `return_tensors` being ignored are different API bugs." + }, + { + "left": "issue:28218", + "right": "issue:29159", + "accept": false, + "reason": "Extra space after added token and slow/fast tokenizer inconsistency are related only at a very high level, not the same concrete bug." + }, + { + "left": "issue:44561", + "right": "issue:45020", + "accept": false, + "reason": "Both mention remote-code breakage, but one is specifically `is_torch_fx_available` removal and the other is a broader version-regression report." + }, + { + "left": "issue:43720", + "right": "issue:45072", + "accept": false, + "reason": "Packed-weight unpacking during accelerate loading is not the same as bf16 dtype mismatches." + }, + { + "left": "issue:44610", + "right": "issue:45072", + "accept": false, + "reason": "OmDet-Turbo processor producing 640px inputs is unrelated to SwitchTransformers/TimmWrapperModel dtype mismatch." + }, + { + "left": "issue:44671", + "right": "issue:45701", + "accept": false, + "reason": "CamemBERT masked-LM prediction regression and general tokenization changes are not the same underlying bug." + }, + { + "left": "issue:43782", + "right": "issue:45325", + "accept": false, + "reason": "Qwen3VL `weight_only` load failure and Qwen2.5-VL RoPE scaling bug affect different models and code paths." + }, + { + "left": "issue:43825", + "right": "issue:43827", + "accept": false, + "reason": "An incorrect runtime error message and stale docs references are separate artifacts, not one duplicate bug." + }, + { + "left": "issue:43582", + "right": "issue:44857", + "accept": false, + "reason": "Apple Silicon warmup `TypeError` and LwDetrImageLoss AMP/CUDA crash are unrelated." + }, + { + "left": "issue:43906", + "right": "issue:46209", + "accept": false, + "reason": "An isolated reproduction of an earlier issue and an FP8 B200 inference failure do not describe the same bug." + }, + { + "left": "issue:43749", + "right": "issue:44291", + "accept": false, + "reason": "FSDP CPU RAM loading failure and `init_empty_weights` unexpected `_is_hf_initialized` argument are distinct initialization/loading regressions." + }, + { + "left": "issue:45081", + "right": "issue:45406", + "accept": false, + "reason": "Mistral tokenizer regex patch crash and Gemma4Processor missing `_tokenizer` are different failures." + }, + { + "left": "issue:42175", + "right": "issue:45081", + "accept": false, + "reason": "Package install backend omission and Mistral tokenizer regex crash are unrelated." + }, + { + "left": "issue:45372", + "right": "issue:46326", + "accept": false, + "reason": "`ReasoningEffort` import failure in Gemma 4 processor loading and `continue_final_message` behavior for mistral_common tokenizers are different bugs." + }, + { + "left": "issue:43262", + "right": "issue:43819", + "accept": false, + "reason": "Audio chat-template sampling-rate defaulting and DAC `from_latents` mismatch are unrelated." + }, + { + "left": "issue:34634", + "right": "issue:36331", + "accept": false, + "reason": "BarkProcessor voice preset and `CustomTrainer.compute_loss()` signature mismatch are unrelated." + }, + { + "left": "issue:45003", + "right": "issue:46285", + "accept": false, + "reason": "Unsafe `sys.modules` access in `modeling_utils` and CLIPTextModel structure inconsistency are different code-path regressions." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 18, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:36010", + "issue:36331", + "issue:38175", + "issue:39692", + "issue:42490", + "issue:43257", + "issue:43262", + "issue:43329", + "issue:43452", + "issue:43454", + "issue:43526", + "issue:43531", + "issue:43550", + "issue:43577", + "issue:43638", + "issue:43645", + "issue:43650", + "issue:43653", + "issue:43696", + "issue:43697", + "issue:43720", + "issue:43742", + "issue:43746", + "issue:43819", + "issue:43854", + "issue:43906", + "issue:43927", + "issue:44060", + "issue:44368", + "issue:44387", + "issue:44554", + "issue:44589", + "issue:44610", + "issue:44617", + "issue:44811", + "issue:44912", + "issue:44936", + "issue:44977", + "issue:44991", + "issue:45020", + "issue:45042", + "issue:45084", + "issue:45459", + "issue:45538", + "issue:45600", + "issue:45803", + "issue:46421", + "issue:46424" + ], + "soft_pairs": [ + "issue:43653|issue:44912", + "issue:43526|issue:43697", + "issue:43742|issue:45020", + "issue:44060|issue:44368", + "issue:45084|issue:45538", + "issue:43526|issue:44554", + "issue:43650|issue:43906", + "issue:43329|issue:45600", + "issue:44589|issue:45042", + "issue:43452|issue:44991", + "issue:43746|issue:44811", + "issue:43638|issue:44936", + "issue:43550|issue:44912", + "issue:44387|issue:44617", + "issue:43819|issue:43927", + "issue:43257|issue:43329", + "issue:46421|issue:46424", + "issue:36010|issue:43645", + "issue:43854|issue:44977", + "issue:43720|issue:44610", + "issue:38175|issue:44610", + "issue:39692|issue:44912", + "issue:36010|issue:36331", + "issue:43531|issue:45042", + "issue:45459|issue:45803", + "issue:43454|issue:43577", + "issue:42490|issue:43257", + "issue:43262|issue:44811", + "issue:43257|issue:43696" + ], + "prepared_review_unit_hash": "102ba651a4c4e03f09d06d01e4cbef642b08cfa43b40d1bef95b3f2780e90c84", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36606, + "estimated_input_tokens": 9152, + "estimated_eval_tokens": 18560 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set does not form one duplicate cluster; it contains many unrelated issues across tokenizers, loading, quantization, cache, and multimodal processors. All soft-similarity pairs look like false positives and should be kept separate.", + "confidence": 0.97, + "canonical_issue_reason": "No single issue is a good canonical representative because the items are thematically unrelated rather than duplicates of one underlying bug.", + "canonical_pr_reason": null, + "best_issue_reason": "None of the issues is suitable as a global canonical issue for this set; the strongest matches still describe different code paths or different products/models.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43653", + "right": "issue:44912", + "accept": false, + "reason": "Different bugs: BigBirdTokenizer special-token registration vs GPT-OSS/gguf quantization loading." + }, + { + "left": "issue:43526", + "right": "issue:43697", + "accept": false, + "reason": "Different subsystems and symptoms: BEiT label reduction vs RTDetrV2 output drift." + }, + { + "left": "issue:43742", + "right": "issue:45020", + "accept": false, + "reason": "One is a MobileLLM key error; the other is a broad remote_code regression affecting many models." + }, + { + "left": "issue:44060", + "right": "issue:44368", + "accept": false, + "reason": "Both mention Qwen warnings, but one is a tied-weights bug and the other is a fine-tuning config warning; not the same defect." + }, + { + "left": "issue:45084", + "right": "issue:45538", + "accept": false, + "reason": "Unrelated: template compilation error vs CLIPTokenizer max length constant." + }, + { + "left": "issue:43526", + "right": "issue:44554", + "accept": false, + "reason": "Unrelated vision preprocessing bug vs MPS attention correctness issue." + }, + { + "left": "issue:43650", + "right": "issue:43906", + "accept": false, + "reason": "No meaningful relation; placeholder issue text vs a reproduction of a specific upstream bug." + }, + { + "left": "issue:43329", + "right": "issue:45600", + "accept": false, + "reason": "Different areas: multimodal token counting vs auto_mappings/config cleanup." + }, + { + "left": "issue:44589", + "right": "issue:45042", + "accept": false, + "reason": "Storage type import error is unrelated to PIL/image-processor torchvision dependency handling." + }, + { + "left": "issue:43452", + "right": "issue:44991", + "accept": false, + "reason": "Both involve loading, but one is gguf_file support and the other is a tokenizer regression for a specific model; not the same code path." + }, + { + "left": "issue:43746", + "right": "issue:44811", + "accept": false, + "reason": "Different models and APIs: GraniteSpeech PEFT checkpoint loading vs Whisper batch_decode token handling." + }, + { + "left": "issue:43638", + "right": "issue:44936", + "accept": false, + "reason": "Different training/evaluation failures with different root causes; not a duplicate." + }, + { + "left": "issue:43550", + "right": "issue:44912", + "accept": false, + "reason": "Different model/runtime issues: torch.compile+SDPA vs gguf/MXFP4 quantized loading." + }, + { + "left": "issue:44387", + "right": "issue:44617", + "accept": false, + "reason": "Both can OOM, but one is a Transformers 5.x int4 memory regression and the other is a model-specific runtime OOM." + }, + { + "left": "issue:43819", + "right": "issue:43927", + "accept": false, + "reason": "Different failures: DAC latent reconstruction mismatch vs DiaConfig save/load token ID loss." + }, + { + "left": "issue:43257", + "right": "issue:43329", + "accept": false, + "reason": "Completely different areas: MoE weight conversion/loading vs multimodal token counting video path." + }, + { + "left": "issue:46421", + "right": "issue:46424", + "accept": false, + "reason": "Related cache area, but one is lazy init causing recompilation and the other is max_cache_length handling for quantized cache; not the same bug." + }, + { + "left": "issue:36010", + "right": "issue:43645", + "accept": false, + "reason": "Import error for GenerationMixin is unrelated to Jupyter custom-model initialization in Transformers 5.0.0." + }, + { + "left": "issue:43854", + "right": "issue:44977", + "accept": false, + "reason": "Different model-specific generation/load problems; no shared concrete code-path defect." + }, + { + "left": "issue:43720", + "right": "issue:44610", + "accept": false, + "reason": "Unrelated: packed-weight unloading in accelerate vs image processor input-size mismatch." + }, + { + "left": "issue:38175", + "right": "issue:44610", + "accept": false, + "reason": "Different model families and failure modes; zero probabilities vs wrong processor size." + }, + { + "left": "issue:39692", + "right": "issue:44912", + "accept": false, + "reason": "Documentation/example errors are unrelated to the gguf/MXFP4 loading regression." + }, + { + "left": "issue:36010", + "right": "issue:36331", + "accept": false, + "reason": "Different APIs and failures: import path breakage vs Trainer custom-loss signature change." + }, + { + "left": "issue:43531", + "right": "issue:45042", + "accept": false, + "reason": "Qwen3-MoE sliding window behavior is unrelated to PIL image-processor torchvision requirements." + }, + { + "left": "issue:45459", + "right": "issue:45803", + "accept": false, + "reason": "Both are exception-handling issues, but they affect different code paths and different error types; not the same bug." + }, + { + "left": "issue:43454", + "right": "issue:43577", + "accept": false, + "reason": "Different model-loading/data-type problems: lm_head tying vs dtype propagation to qformer." + }, + { + "left": "issue:42490", + "right": "issue:43257", + "accept": false, + "reason": "Trailing-slash path warning is unrelated to Qwen3 MOE weight conversion with accelerate+deepspeed." + }, + { + "left": "issue:43262", + "right": "issue:44811", + "accept": false, + "reason": "Different processor APIs and symptoms: audio chat-template sample-rate default vs Whisper decode token filtering." + }, + { + "left": "issue:43257", + "right": "issue:43696", + "accept": false, + "reason": "MoE weight conversion/loading is unrelated to GPT-oss CUDA OOM." + } + ] + }, + "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": 19, + "review_unit_index": 19, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:24643", + "issue:28282", + "issue:29159", + "issue:33357", + "issue:34567", + "issue:34689", + "issue:40990", + "issue:41720", + "issue:42175", + "issue:42915", + "issue:43257", + "issue:43317", + "issue:43329", + "issue:43450", + "issue:43452", + "issue:43540", + "issue:43582", + "issue:43606", + "issue:43643", + "issue:43697", + "issue:43742", + "issue:43749", + "issue:43824", + "issue:43927", + "issue:43931", + "issue:44062", + "issue:44291", + "issue:44479", + "issue:44560", + "issue:44568", + "issue:44756", + "issue:44821", + "issue:44849", + "issue:44912", + "issue:44987", + "issue:44991", + "issue:45042", + "issue:45137", + "issue:45356", + "issue:45357", + "issue:45381", + "issue:45405", + "issue:45488", + "issue:45538", + "issue:45701", + "issue:45759", + "issue:45859", + "issue:45907" + ], + "soft_pairs": [ + "issue:43643|issue:45759", + "issue:43329|issue:44479", + "issue:43329|issue:43450", + "issue:43606|issue:45859", + "issue:34567|issue:40990", + "issue:43749|issue:44756", + "issue:44568|issue:45356", + "issue:29159|issue:34567", + "issue:43257|issue:43931", + "issue:42915|issue:43742", + "issue:44291|issue:45042", + "issue:42915|issue:44821", + "issue:44987|issue:45538", + "issue:43452|issue:43582", + "issue:43697|issue:45357", + "issue:43927|issue:44912", + "issue:42915|issue:45538", + "issue:44849|issue:45701", + "issue:44568|issue:45488", + "issue:41720|issue:43540", + "issue:33357|issue:34567", + "issue:42175|issue:43824", + "issue:44479|issue:45381", + "issue:43317|issue:45538", + "issue:43540|issue:44479", + "issue:44062|issue:44991", + "issue:43540|issue:44560", + "issue:28282|issue:34689", + "issue:45137|issue:45405", + "issue:24643|issue:29159", + "issue:43742|issue:44291", + "issue:44987|issue:45907" + ], + "prepared_review_unit_hash": "fa24daff671f1bba2faeecf50b14c9b9edb5f34004add87710825838738a0d4c", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 37328, + "estimated_input_tokens": 9332, + "estimated_eval_tokens": 18920 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes unrelated tokenizer, video/VLM, loading, offload, and training-state bugs. I found no soft pair that clearly represents the same underlying fix, so all pairs should be kept separate.", + "confidence": 0.24, + "canonical_issue_reason": "Issue 45701 is the broadest, most umbrella-like report in the set and has inbound references, but there is no truly strong canonical duplicate cluster here because the topics are too mixed.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45701 is the closest thing to a representative issue for the cluster because it is the most general regression report and could absorb adjacent tokenization/version-related complaints, though it does not cover the whole set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43643", + "right": "issue:45759", + "accept": false, + "reason": "Both are config/model-construction complaints, but one is AutoConfig trust_remote_code metadata loss and the other is AutoModelForCausalLM.from_config not unwrapping text_config. Different code paths and failure modes." + }, + { + "left": "issue:43329", + "right": "issue:44479", + "accept": false, + "reason": "Both involve video/VLM regressions, but 43329 is an undefined-name bug in a helper path while 44479 is a broader v5.3.0 video-input regression across several Qwen variants. Not the same concrete bug." + }, + { + "left": "issue:43329", + "right": "issue:43450", + "accept": false, + "reason": "One is token counting with undefined video-branch symbols; the other is batched video processor shape handling. Related subsystem, but different underlying defects." + }, + { + "left": "issue:43606", + "right": "issue:45859", + "accept": false, + "reason": "Both mention CPU offload, but they hit different models and different code paths: Bark device mismatch versus Qwen3.5 MoE rotary embedding compatibility." + }, + { + "left": "issue:34567", + "right": "issue:40990", + "accept": false, + "reason": "TrainerState token accounting is unrelated to perplexity on gpt-oss-20b." + }, + { + "left": "issue:43749", + "right": "issue:44756", + "accept": false, + "reason": "Both are memory/loading related, but FSDP CPU RAM-efficient loading and Strix Halo mmap OOM are distinct problems." + }, + { + "left": "issue:44568", + "right": "issue:45356", + "accept": false, + "reason": "Both are tokenizer regressions, but one is add_special_tokens not adding BOS/EOS and the other is codec/warning behavior for Kimi-K2.5. Different bugs." + }, + { + "left": "issue:29159", + "right": "issue:34567", + "accept": false, + "reason": "Tokenizer slow/fast inconsistency is unrelated to TrainerState num_input_tokens_seen not updating." + }, + { + "left": "issue:43257", + "right": "issue:43931", + "accept": false, + "reason": "Both are Qwen3 loading issues, but one is deepspeed/accelerate MoE weight conversion and the other is a Qwen3-VL shape mismatch. Not the same fix." + }, + { + "left": "issue:42915", + "right": "issue:43742", + "accept": false, + "reason": "Qwen3Moe FP8 failure and MobileLLM key error are unrelated loading failures." + }, + { + "left": "issue:44291", + "right": "issue:45042", + "accept": false, + "reason": "init_empty_weights/_is_hf_initialized and PIL backend torchvision dependency are different code paths with no shared concrete bug." + }, + { + "left": "issue:42915", + "right": "issue:44821", + "accept": false, + "reason": "FP8 Qwen3Moe loading and AutoImageProcessor-from-URL are unrelated." + }, + { + "left": "issue:44987", + "right": "issue:45538", + "accept": false, + "reason": "Loading physical-intelligence/fast on v5.1.0 and CLIPTokenizer model_max_length are separate issues." + }, + { + "left": "issue:43452", + "right": "issue:43582", + "accept": false, + "reason": "gguf_file tokenizer/model loading breakage and AppleSilicon caching_allocator_warmup TypeError are unrelated." + }, + { + "left": "issue:43697", + "right": "issue:45357", + "accept": false, + "reason": "RTDetrV2 output changes and Qwen3.5 save_pretrained visual-key regression are different model families and different failure modes." + }, + { + "left": "issue:43927", + "right": "issue:44912", + "accept": false, + "reason": "DiaConfig save/load token ID loss is unrelated to MXFP4 quantization fallback on git-oss-20b." + }, + { + "left": "issue:42915", + "right": "issue:45538", + "accept": false, + "reason": "Qwen3Moe FP8 loading and CLIPTokenizer max length are not the same bug." + }, + { + "left": "issue:44849", + "right": "issue:45701", + "accept": false, + "reason": "Qwen3.5 output_hidden_states regression is a model-forward bug, while 45701 is a generic tokenization/version-change report." + }, + { + "left": "issue:44568", + "right": "issue:45488", + "accept": false, + "reason": "Both are tokenizer regressions, but one concerns special-token insertion and the other concerns overriding tokenizer.json pre-tokenizer settings. Different code paths." + }, + { + "left": "issue:41720", + "right": "issue:43540", + "accept": false, + "reason": "cudaErrorAssert on auto device mapping for Qwen3 and video-processing ValueError in Qwen3OmniMoe are distinct issues." + }, + { + "left": "issue:33357", + "right": "issue:34567", + "accept": false, + "reason": "MacOS bus error with community CLIP and TrainerState token tracking are unrelated." + }, + { + "left": "issue:42175", + "right": "issue:43824", + "accept": false, + "reason": "TensorFlow backend packaging and missing Qwen2.5-VL import are different concerns." + }, + { + "left": "issue:44479", + "right": "issue:45381", + "accept": false, + "reason": "Both are video-input regressions, but 45381 is specifically wrong vision_position_ids in qwen2.5-vl, while 44479 is a broader multi-model regression. Too little evidence for one bug." + }, + { + "left": "issue:43317", + "right": "issue:45538", + "accept": false, + "reason": "Dequantized model loading with device_map=auto/offload and CLIPTokenizer model_max_length are unrelated." + }, + { + "left": "issue:43540", + "right": "issue:44479", + "accept": false, + "reason": "Qwen3OmniMoe video ValueError and the broader v5.3.0 video-input regression are not clearly the same underlying defect." + }, + { + "left": "issue:44062", + "right": "issue:44991", + "accept": false, + "reason": "AddedToken special-argument conflict and tokenizer loading for EMBEDDIA/est-roberta are different tokenizer bugs." + }, + { + "left": "issue:43540", + "right": "issue:44560", + "accept": false, + "reason": "Both are Qwen video-related, but one is a processing ValueError and the other is a StopIteration in qwen3-vl embedding. Different concrete failures." + }, + { + "left": "issue:28282", + "right": "issue:34689", + "accept": false, + "reason": "Missing PyTorch import and Llama 3.2 Vision model loading breakage are unrelated load-time errors." + }, + { + "left": "issue:45137", + "right": "issue:45405", + "accept": false, + "reason": "DeepSpeed ZeRO3 deque underflow and a PEFT version bump are not the same issue." + }, + { + "left": "issue:24643", + "right": "issue:29159", + "accept": false, + "reason": "DeepSpeed weight dimensionality error and tokenizer inconsistency are unrelated." + }, + { + "left": "issue:43742", + "right": "issue:44291", + "accept": false, + "reason": "Key error loading MobileLLM and init_empty_weights unexpected argument are separate model-loading failures." + }, + { + "left": "issue:44987", + "right": "issue:45907", + "accept": false, + "reason": "Loading physical-intelligence/fast and list_repo_files not forwarding revision are unrelated regressions." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 20, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:27957", + "issue:28282", + "issue:29127", + "issue:34567", + "issue:36331", + "issue:38175", + "issue:42175", + "issue:42915", + "issue:43257", + "issue:43404", + "issue:43452", + "issue:43493", + "issue:43526", + "issue:43531", + "issue:43577", + "issue:43582", + "issue:43643", + "issue:43742", + "issue:43749", + "issue:43756", + "issue:43782", + "issue:43906", + "issue:43994", + "issue:44062", + "issue:44291", + "issue:44488", + "issue:44589", + "issue:44737", + "issue:44792", + "issue:44821", + "issue:44898", + "issue:44913", + "issue:44964", + "issue:45042", + "issue:45072", + "issue:45081", + "issue:45325", + "issue:45357", + "issue:45375", + "issue:45381", + "issue:45405", + "issue:45479", + "issue:45588", + "issue:45600", + "issue:45920", + "issue:45987", + "issue:46121", + "issue:46399" + ], + "soft_pairs": [ + "issue:43577|issue:45072", + "issue:28282|issue:29127", + "issue:42915|issue:43749", + "issue:38175|issue:44898", + "issue:42175|issue:44488", + "issue:43257|issue:43756", + "issue:43582|issue:43782", + "issue:44737|issue:45588", + "issue:43493|issue:43994", + "issue:44291|issue:44821", + "issue:34567|issue:36331", + "issue:43404|issue:45987", + "issue:43526|issue:45479", + "issue:44062|issue:45081", + "issue:43906|issue:45381", + "issue:27957|issue:28282", + "issue:43756|issue:46399", + "issue:44913|issue:46121", + "issue:44792|issue:44964", + "issue:43756|issue:45325", + "issue:45042|issue:45081", + "issue:43452|issue:44488", + "issue:43742|issue:44589", + "issue:45357|issue:45920", + "issue:43531|issue:44821", + "issue:45405|issue:45600", + "issue:43643|issue:45375" + ], + "prepared_review_unit_hash": "618bee7731f22f39421c60b7c68da478310cbf9bc4bfed4afd0506c08d6b7471", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 27, + "serialized_chars": 36188, + "estimated_input_tokens": 9047, + "estimated_eval_tokens": 18350 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly heterogeneous: the issues span unrelated models, loaders, dtype paths, tokenizer bugs, and config serialization problems. Only the SigLIP2 pair (43493/43994) looks like a plausible duplicate; the rest are distinct reports and should not be merged.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43577", + "right": "issue:45072", + "accept": false, + "reason": "Both mention dtype mismatches, but they affect different models and code paths (BLIP2 load vs SwitchTransformers/TimmWrapper inference)." + }, + { + "left": "issue:28282", + "right": "issue:29127", + "accept": false, + "reason": "One is an ImportError about missing PyTorch; the other is a layoutlmv3 error-message clarity issue. No shared bug." + }, + { + "left": "issue:42915", + "right": "issue:43749", + "accept": false, + "reason": "Qwen3Moe FP8 failure and FSDP CPU RAM efficient loading are different mechanisms and different failure modes." + }, + { + "left": "issue:38175", + "right": "issue:44898", + "accept": false, + "reason": "Both are image-model correctness reports, but SigLIP2 zero probabilities and Perceiver non-default-resolution failure are distinct bugs." + }, + { + "left": "issue:42175", + "right": "issue:44488", + "accept": false, + "reason": "Packaging/backend dependency coverage is unrelated to a model-loading failure for cjvt/sleng-bert." + }, + { + "left": "issue:43257", + "right": "issue:43756", + "accept": false, + "reason": "Different Qwen3/Smollm3 bugs: weight conversion under accelerate+deepspeed vs RoPE layer dropping." + }, + { + "left": "issue:43582", + "right": "issue:43782", + "accept": false, + "reason": "AppleSilicon allocator TypeError and Qwen3VL weight_only loading error are unrelated." + }, + { + "left": "issue:44737", + "right": "issue:45588", + "accept": false, + "reason": "XLNet CPU-device placement bug and flash_attention sink-less crash are different code paths." + }, + { + "left": "issue:43493", + "right": "issue:43994", + "accept": true, + "reason": "Both describe the same SigLIP2 model producing incorrect outputs; 43994 looks like a user-facing manifestation of the implementation discrepancy in 43493." + }, + { + "left": "issue:44291", + "right": "issue:44821", + "accept": false, + "reason": "Loading with init_empty_weights and AutoImageProcessor-from-URL failure are unrelated." + }, + { + "left": "issue:34567", + "right": "issue:36331", + "accept": false, + "reason": "TrainerState token counting and CustomTrainer.compute_loss signature mismatch are different trainer bugs." + }, + { + "left": "issue:43404", + "right": "issue:45987", + "accept": false, + "reason": "Mistral3 lm_head tying and StaticCache.get_seq_length return type are unrelated." + }, + { + "left": "issue:43526", + "right": "issue:45479", + "accept": false, + "reason": "BeitImageProcessorFast label reduction bug and sequence-classification zero-loss bug are different subsystems and symptoms." + }, + { + "left": "issue:44062", + "right": "issue:45081", + "accept": false, + "reason": "Tokenizer AddedToken argument duplication and Mistral regex patch backend_tokenizer crash are unrelated tokenizer bugs." + }, + { + "left": "issue:43906", + "right": "issue:45381", + "accept": false, + "reason": "An isolated reproduction of an earlier issue and a Qwen2.5-VL video position-id regression are not the same bug." + }, + { + "left": "issue:27957", + "right": "issue:28282", + "accept": false, + "reason": "XLMRoberta Flash Attention 2 support and missing PyTorch environment detection are unrelated." + }, + { + "left": "issue:43756", + "right": "issue:46399", + "accept": false, + "reason": "Both involve RoPE-adjacent models, but Smollm3 RoPE-layer loss and Qwen3.5 kernel registration failure are different code paths." + }, + { + "left": "issue:44913", + "right": "issue:46121", + "accept": false, + "reason": "GPTNeoX rotary_pct reload regression and JSON-list handling in rope validation are separate config bugs." + }, + { + "left": "issue:44792", + "right": "issue:44964", + "accept": false, + "reason": "Janus image-generation test failure and Phi-4 multimodal load failure are different model-specific problems." + }, + { + "left": "issue:43756", + "right": "issue:45325", + "accept": false, + "reason": "Smollm3 RoPE-layer dropping and Qwen2.5-VL still-image temporal position scaling are unrelated." + }, + { + "left": "issue:45042", + "right": "issue:45081", + "accept": false, + "reason": "A PIL-backend torchvision dependency regression is unrelated to the Mistral tokenizer regex crash." + }, + { + "left": "issue:43452", + "right": "issue:44488", + "accept": false, + "reason": "gguf_file loading breakage and inability to load cjvt/sleng-bert are both loading issues, but not the same concrete failure." + }, + { + "left": "issue:43742", + "right": "issue:44589", + "accept": false, + "reason": "MobileLLM-125M key error and missing Float8 storage type are unrelated load/runtime issues." + }, + { + "left": "issue:45357", + "right": "issue:45920", + "accept": false, + "reason": "Qwen3.5 save_pretrained visual-key regression and AutoTokenizer wrong token IDs across multiple models are different regressions." + }, + { + "left": "issue:43531", + "right": "issue:44821", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior and AutoImageProcessor URL loading are unrelated." + }, + { + "left": "issue:45405", + "right": "issue:45600", + "accept": false, + "reason": "A PEFT version bump problem and auto_mappings referencing removed configs are separate CI/package issues." + }, + { + "left": "issue:43643", + "right": "issue:45375", + "accept": false, + "reason": "Generic trust_remote_code config field loss and Qwen3_5MoeVisionConfig missing deepstack_visual_indexes are both config issues, but not the same bug." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The analysis is grounded in the packet and the conclusions are conservative overall. The one accepted pair, 43493/43994, is a plausible duplicate about SigLIP2 incorrect outputs, while the rest are reasonably rejected as different models, code paths, or bug classes. No soft-edge verdict looks overstated based on the provided titles and evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 21, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:24643", + "issue:27957", + "issue:28282", + "issue:29127", + "issue:30064", + "issue:32090", + "issue:33357", + "issue:33666", + "issue:38175", + "issue:41762", + "issue:43295", + "issue:43452", + "issue:43577", + "issue:43653", + "issue:43688", + "issue:43697", + "issue:43720", + "issue:43749", + "issue:43819", + "issue:43906", + "issue:44361", + "issue:44442", + "issue:44464", + "issue:44466", + "issue:44484", + "issue:44493", + "issue:44514", + "issue:44568", + "issue:44625", + "issue:44704", + "issue:44743", + "issue:44811", + "issue:44843", + "issue:44877", + "issue:44898", + "issue:44936", + "issue:45005", + "issue:45216", + "issue:45290", + "issue:45356", + "issue:45372", + "issue:45381", + "issue:45542", + "issue:45823", + "issue:46050", + "issue:46170", + "issue:46178", + "issue:46202" + ], + "soft_pairs": [ + "issue:44625|issue:44877", + "issue:43720|issue:44898", + "issue:24643|issue:29127", + "issue:43697|issue:45216", + "issue:41762|issue:44936", + "issue:44514|issue:45381", + "issue:44493|issue:46050", + "issue:44466|issue:45005", + "issue:43452|issue:43577", + "issue:43906|issue:46202", + "issue:43452|issue:44843", + "issue:30064|issue:33666", + "issue:43688|issue:43749", + "issue:43819|issue:44811", + "issue:45372|issue:46170", + "issue:28282|issue:32090", + "issue:44936|issue:45823", + "issue:28282|issue:33357", + "issue:43295|issue:45356", + "issue:44361|issue:44442", + "issue:43653|issue:44361", + "issue:44484|issue:44704", + "issue:38175|issue:43295", + "issue:45542|issue:46178", + "issue:24643|issue:27957", + "issue:43653|issue:44568", + "issue:44514|issue:45290", + "issue:43577|issue:43697", + "issue:44464|issue:44743" + ], + "prepared_review_unit_hash": "ddb3e76b0adf2c727e6872060f515c70de4e711429a863445b2412e82387fa25", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36685, + "estimated_input_tokens": 9172, + "estimated_eval_tokens": 18600 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most items are unrelated bug reports across tokenizer, processor, training, quantization, and serialization paths. Only one soft duplicate looks plausible: the tied-weights save/serialization regression pair (44466 \u2194 45005).", + "confidence": 0.87, + "canonical_issue_reason": "Issue 45005 is the best umbrella report: it describes the broader v5 tied-weights serialization regression, which fits the more specific lm_head.weight device-dependent serialization symptom in 44466.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45005 is the strongest cluster representative because it is the broadest and most actionable issue about the underlying tied-weights/save_pretrained bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44625", + "right": "issue:44877", + "accept": false, + "reason": "Both are config/loading regressions, but they affect different models and different failure modes; not the same underlying bug." + }, + { + "left": "issue:43720", + "right": "issue:44898", + "accept": false, + "reason": "Unrelated code paths: BitNet accelerate weight unpacking vs Perceiver positional interpolation." + }, + { + "left": "issue:24643", + "right": "issue:29127", + "accept": false, + "reason": "DeepSpeed training weight-shape error and a LayoutLMv3 error-message clarity issue are not the same bug." + }, + { + "left": "issue:43697", + "right": "issue:45216", + "accept": false, + "reason": "Both are v5 regressions, but one is RTDetrV2 output drift and the other is Qwen3.5 checkpoint saving; different code paths." + }, + { + "left": "issue:41762", + "right": "issue:44936", + "accept": false, + "reason": "Gemma3 ZeRO-3 loading failure and trainer.evaluate-after-train behavior are different problems." + }, + { + "left": "issue:44514", + "right": "issue:45381", + "accept": false, + "reason": "Same model family, but one is chat-template batching and the other is video vision_position_ids; not the same concrete bug." + }, + { + "left": "issue:44493", + "right": "issue:46050", + "accept": false, + "reason": "Unexpected position-id keys vs sinq quantized model save/load failure are unrelated." + }, + { + "left": "issue:44466", + "right": "issue:45005", + "accept": true, + "reason": "Both report the v5 tied-weights serialization/save_pretrained regression; 44466 is a more specific manifestation of the broader bug in 45005." + }, + { + "left": "issue:43452", + "right": "issue:43577", + "accept": false, + "reason": "Tokenizer/AutoModel loading with gguf_file is unrelated to Blip2 dtype propagation." + }, + { + "left": "issue:43906", + "right": "issue:46202", + "accept": false, + "reason": "A reproduction of an older issue and a Gemma pipeline roundtrip bug are not the same underlying defect." + }, + { + "left": "issue:43452", + "right": "issue:44843", + "accept": false, + "reason": "Both involve AutoTokenizer loading, but the gguf_file breakage and offline model_info call are distinct code paths." + }, + { + "left": "issue:30064", + "right": "issue:33666", + "accept": false, + "reason": "Void segmentation map processing is unrelated to Qwen2-VL multi-GPU training." + }, + { + "left": "issue:43688", + "right": "issue:43749", + "accept": false, + "reason": "Auxiliary-loss normalization and FSDP CPU RAM efficient loading are different issues." + }, + { + "left": "issue:43819", + "right": "issue:44811", + "accept": false, + "reason": "DAC latent decoding mismatch and Whisper batch_decode skip_special_tokens handling are unrelated." + }, + { + "left": "issue:45372", + "right": "issue:46170", + "accept": false, + "reason": "Gemma 4 processor import failure and Windows/ROCm torch.distributed import issues are different backend problems." + }, + { + "left": "issue:28282", + "right": "issue:32090", + "accept": false, + "reason": "Missing PyTorch import and Trainer NoneType broadcast are unrelated errors." + }, + { + "left": "issue:44936", + "right": "issue:45823", + "accept": false, + "reason": "Trainer lifecycle failure and Gemma4 device_map forward mismatch are different code paths." + }, + { + "left": "issue:28282", + "right": "issue:33357", + "accept": false, + "reason": "General PyTorch-import error and a MacOS bus error with CLIP are not the same bug." + }, + { + "left": "issue:43295", + "right": "issue:45356", + "accept": false, + "reason": "Both are regressions, but one is custom processor/tokenizer behavior and the other is Kimi-K2.5 codec/regex handling." + }, + { + "left": "issue:44361", + "right": "issue:44442", + "accept": false, + "reason": "Different tokenizer failures on different models; no shared underlying defect is evident." + }, + { + "left": "issue:43653", + "right": "issue:44361", + "accept": false, + "reason": "BigBird special-token registration and MLukeTokenizer task AttributeError are unrelated." + }, + { + "left": "issue:44484", + "right": "issue:44704", + "accept": false, + "reason": "save_pretrained shard-size question and AutoProcessor kwargs propagation are not the same issue." + }, + { + "left": "issue:38175", + "right": "issue:43295", + "accept": false, + "reason": "Siglip2 zero probabilities and processor.tokenizer regression are unrelated." + }, + { + "left": "issue:45542", + "right": "issue:46178", + "accept": false, + "reason": "Both touch backend detection, but they are different missing-backend/registration failures." + }, + { + "left": "issue:24643", + "right": "issue:27957", + "accept": false, + "reason": "DeepSpeed training weight-shape error and XLMRoberta Flash Attention 2 support are different issues." + }, + { + "left": "issue:43653", + "right": "issue:44568", + "accept": false, + "reason": "Both involve token special handling, but the BigBird mask-token bug and mDeBERTa BOS/EOS regression are distinct." + }, + { + "left": "issue:44514", + "right": "issue:45290", + "accept": false, + "reason": "Both are chat/template related, but one is batched input padding and the other is tool-call assistant messages; different bugs." + }, + { + "left": "issue:43577", + "right": "issue:43697", + "accept": false, + "reason": "Blip2 dtype propagation and RTDetrV2 output differences are unrelated." + }, + { + "left": "issue:44464", + "right": "issue:44743", + "accept": false, + "reason": "Both involve generation/cache behavior, but one is chunked generation with compiled forward and the other is recurrent-state reset in modular_qwen3_5; not enough to call the same bug." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The cluster summary is well grounded in the packet, and the only accepted soft edge (44466 \u2194 45005) is conservative and plausibly the same tied-weights/save_pretrained regression. The rejected pairs are also appropriately cautious given the titles alone." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 22, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:29159", + "issue:32090", + "issue:33357", + "issue:33666", + "issue:34567", + "issue:36010", + "issue:41720", + "issue:43262", + "issue:43295", + "issue:43452", + "issue:43504", + "issue:43531", + "issue:43550", + "issue:43577", + "issue:43606", + "issue:43643", + "issue:43646", + "issue:43653", + "issue:43688", + "issue:43720", + "issue:43742", + "issue:43749", + "issue:43756", + "issue:43927", + "issue:44361", + "issue:44451", + "issue:44464", + "issue:44466", + "issue:44479", + "issue:44554", + "issue:44561", + "issue:44610", + "issue:44625", + "issue:44821", + "issue:44843", + "issue:44977", + "issue:44991", + "issue:45440", + "issue:45491", + "issue:45588", + "issue:45701", + "issue:45704", + "issue:45823", + "issue:46002", + "issue:46050", + "issue:46209", + "issue:46402" + ], + "soft_pairs": [ + "issue:29159|issue:33357", + "issue:44843|issue:44991", + "issue:43646|issue:43749", + "issue:43531|issue:43742", + "issue:43756|issue:46402", + "issue:43688|issue:45440", + "issue:43688|issue:45704", + "issue:45491|issue:46002", + "issue:43653|issue:43927", + "issue:43550|issue:43927", + "issue:43688|issue:45701", + "issue:44464|issue:44977", + "issue:44479|issue:44625", + "issue:43550|issue:44361", + "issue:43643|issue:44561", + "issue:41720|issue:44464", + "issue:44977|issue:46209", + "issue:32090|issue:34567", + "issue:43577|issue:45588", + "issue:43577|issue:44554", + "issue:44466|issue:46050", + "issue:32090|issue:33666", + "issue:43504|issue:43720", + "issue:43504|issue:43550", + "issue:33666|issue:36010", + "issue:43504|issue:44610", + "issue:43742|issue:44821", + "issue:45823|issue:46050", + "issue:43262|issue:43295", + "issue:43452|issue:44451", + "issue:43720|issue:43927", + "issue:43606|issue:43927" + ], + "prepared_review_unit_hash": "3c8e171f028b544ed6b1500c5bf520e77889959b48f87162912e819d481c11dc", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 32, + "serialized_chars": 36691, + "estimated_input_tokens": 9173, + "estimated_eval_tokens": 18602 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a set of false-positive soft matches across unrelated bugs/regressions (tokenization, model loading, training, generation, quantization, and device-mapping). There are no PRs in the input. If forced to anchor the cluster, issue:43927 is the most connected and concrete save/load regression, but it does not represent a true duplicate umbrella for the whole set.", + "confidence": 0.34, + "canonical_issue_reason": "issue:43927 has the highest soft-link connectivity among the candidates and is a concrete, actionable config/save-load bug around token IDs. It is the best anchor only in a weak clustering sense, not because the whole set is one duplicate.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43927 is the strongest overall issue candidate by centrality and clarity of repro; it is broad enough to serve as the cluster anchor, though the surrounding issues are not true duplicates.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:29159", + "right": "issue:33357", + "accept": false, + "reason": "Tokenizer inconsistency vs MacOS bus error in a CLIP model; different subsystems and failure modes." + }, + { + "left": "issue:44843", + "right": "issue:44991", + "accept": false, + "reason": "Both involve tokenizer loading, but one is offline Mistral regex patching and the other is a specific tokenizer load failure; not the same bug." + }, + { + "left": "issue:43646", + "right": "issue:43749", + "accept": false, + "reason": "Custom model initialization breakage vs FSDP CPU RAM-efficient loading; unrelated code paths." + }, + { + "left": "issue:43531", + "right": "issue:43742", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior vs MobileLLM key error during load; different issues." + }, + { + "left": "issue:43756", + "right": "issue:46402", + "accept": false, + "reason": "Smollm3 RoPE layer selection bug vs Sam3 text encoder weights not loading; unrelated model internals." + }, + { + "left": "issue:43688", + "right": "issue:45440", + "accept": false, + "reason": "Aux-loss normalization bug vs DeepseekV3 implementation divergence; not the same underlying defect." + }, + { + "left": "issue:43688", + "right": "issue:45704", + "accept": false, + "reason": "Aux-loss normalization vs T5 apex memory leak; unrelated." + }, + { + "left": "issue:45491", + "right": "issue:46002", + "accept": false, + "reason": "Gemma3 sliding-window NaNs vs classification-head NaNs on init; different root causes." + }, + { + "left": "issue:43653", + "right": "issue:43927", + "accept": false, + "reason": "BigBirdTokenizer special-token registration issue vs DiaConfig token-ID persistence loss; both token-related but different mechanisms." + }, + { + "left": "issue:43550", + "right": "issue:43927", + "accept": false, + "reason": "Bamba torch.compile SDPA failure vs config token-ID persistence; unrelated code paths." + }, + { + "left": "issue:43688", + "right": "issue:45701", + "accept": false, + "reason": "Aux-loss normalization bug vs version-dependent tokenization changes; not the same bug." + }, + { + "left": "issue:44464", + "right": "issue:44977", + "accept": false, + "reason": "Compiled chunked generation inconsistency vs Qwen3.5 flash-attention generation failure; different generation paths." + }, + { + "left": "issue:44479", + "right": "issue:44625", + "accept": false, + "reason": "Video-input regression vs num_labels propagation bug; unrelated." + }, + { + "left": "issue:43550", + "right": "issue:44361", + "accept": false, + "reason": "torch.compile SDPA failure vs MLukeTokenizer AttributeError; different components." + }, + { + "left": "issue:43643", + "right": "issue:44561", + "accept": false, + "reason": "Missing fields from trust_remote_code config loading vs removal of is_torch_fx_available breaking trust_remote_code models; same area, different root cause." + }, + { + "left": "issue:41720", + "right": "issue:44464", + "accept": false, + "reason": "Qwen3 auto device-map cuda assert vs chunked generation with compiled forward; unrelated." + }, + { + "left": "issue:44977", + "right": "issue:46209", + "accept": false, + "reason": "Qwen3.5 flash-attention generation issue vs FP8 inference failure on B200; different hardware/code paths." + }, + { + "left": "issue:32090", + "right": "issue:34567", + "accept": false, + "reason": "Trainer broadcast TypeError vs num_input_tokens_seen not updating; both Trainer-related but not the same bug." + }, + { + "left": "issue:43577", + "right": "issue:45588", + "accept": false, + "reason": "BLIP2 dtype propagation bug vs flash_attention s_aux=None crash; unrelated." + }, + { + "left": "issue:43577", + "right": "issue:44554", + "accept": false, + "reason": "BLIP2 dtype issue vs MPS attention correctness issue; different defects." + }, + { + "left": "issue:44466", + "right": "issue:46050", + "accept": false, + "reason": "lm_head serialization inconsistency vs quantized save/load breakage; both serialization-related but not the same failure." + }, + { + "left": "issue:32090", + "right": "issue:33666", + "accept": false, + "reason": "Trainer NoneType broadcast error vs Qwen2-VL multi-GPU training issue; unrelated." + }, + { + "left": "issue:43504", + "right": "issue:43720", + "accept": false, + "reason": "Beit legacy-field loading bug vs BitNet packed-weight unpacking during accelerate loading; both loading issues but different models and fixes." + }, + { + "left": "issue:43504", + "right": "issue:43550", + "accept": false, + "reason": "Beit pretrained-load issue vs Bamba torch.compile SDPA failure; unrelated." + }, + { + "left": "issue:33666", + "right": "issue:36010", + "accept": false, + "reason": "Qwen2-VL multi-GPU training vs GenerationMixin import error; unrelated." + }, + { + "left": "issue:43504", + "right": "issue:44610", + "accept": false, + "reason": "Beit legacy-field load failure vs OmDet processor/model input-size mismatch; unrelated." + }, + { + "left": "issue:43742", + "right": "issue:44821", + "accept": false, + "reason": "MobileLLM key error on load vs AutoImageProcessor URL loading; different components." + }, + { + "left": "issue:45823", + "right": "issue:46050", + "accept": false, + "reason": "Gemma4 device-mismatch during forward vs sinq quantized save/load breakage; unrelated." + }, + { + "left": "issue:43262", + "right": "issue:43295", + "accept": false, + "reason": "Audio chat-template sample-rate default vs processor.tokenizer/image regression; both processor issues but not the same bug." + }, + { + "left": "issue:43452", + "right": "issue:44451", + "accept": false, + "reason": "gguf_file loading failure vs ScandiBERT load failure; unrelated model-loading problems." + }, + { + "left": "issue:43720", + "right": "issue:43927", + "accept": false, + "reason": "BitNet packed-weight loading bug vs DiaConfig token-ID persistence loss; different code paths." + }, + { + "left": "issue:43606", + "right": "issue:43927", + "accept": false, + "reason": "CPU-offload device mismatch in bark-small vs config token-ID persistence; 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": 19, + "review_unit_index": 23, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:28218", + "issue:33357", + "issue:33666", + "issue:36331", + "issue:39692", + "issue:43257", + "issue:43278", + "issue:43295", + "issue:43299", + "issue:43334", + "issue:43493", + "issue:43504", + "issue:43525", + "issue:43540", + "issue:43572", + "issue:43606", + "issue:43646", + "issue:43688", + "issue:43720", + "issue:43824", + "issue:43825", + "issue:43844", + "issue:43927", + "issue:43994", + "issue:44291", + "issue:44361", + "issue:44464", + "issue:44466", + "issue:44488", + "issue:44568", + "issue:44589", + "issue:44625", + "issue:44843", + "issue:44849", + "issue:44964", + "issue:44977", + "issue:45200", + "issue:45229", + "issue:45357", + "issue:45405", + "issue:45520", + "issue:45759", + "issue:45812", + "issue:45823", + "issue:46002", + "issue:46192", + "issue:46209", + "issue:46424" + ], + "soft_pairs": [ + "issue:43994|issue:46192", + "issue:33666|issue:36331", + "issue:43525|issue:43572", + "issue:44849|issue:46209", + "issue:43278|issue:43688", + "issue:43720|issue:44361", + "issue:44625|issue:46002", + "issue:43493|issue:43825", + "issue:28218|issue:33357", + "issue:39692|issue:43493", + "issue:44964|issue:45405", + "issue:44849|issue:45357", + "issue:44568|issue:45812", + "issue:44464|issue:46424", + "issue:45200|issue:45229", + "issue:43540|issue:44977", + "issue:43493|issue:43844", + "issue:44488|issue:44843", + "issue:39692|issue:43295", + "issue:43278|issue:43844", + "issue:43334|issue:43572", + "issue:43646|issue:44466", + "issue:33357|issue:33666", + "issue:43334|issue:45759", + "issue:43257|issue:43299", + "issue:44589|issue:45520", + "issue:43824|issue:44589", + "issue:45229|issue:45823", + "issue:43927|issue:44361", + "issue:43504|issue:43606", + "issue:43572|issue:44291" + ], + "prepared_review_unit_hash": "c02270b36815665c85dbff5cbf9a77bd9285b0b6741fc5b06755219b00919931", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 37111, + "estimated_input_tokens": 9278, + "estimated_eval_tokens": 18812 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly false-positive similarity: the issues span unrelated model families and failure modes. The only broad commonality is \u201crecent Transformers/v5 regressions,\u201d but the concrete bugs differ too much to merge as duplicates.", + "confidence": 0.82, + "canonical_issue_reason": "If one issue has to anchor the cluster, `issue:43646` is the broadest umbrella (`Transformers 5.0.0 breaks custom model initialization`) and best matches the general v5-regression theme seen in a few nearby reports. Still, the cluster is not a true duplicate set.", + "canonical_pr_reason": null, + "best_issue_reason": "`issue:43646` is the most representative issue because it is phrased at the right abstraction level and covers a general regression class rather than a single model-specific symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43994", + "right": "issue:46192", + "accept": false, + "reason": "Different bugs: SigLIP2 inference correctness vs `return_tensors` being ignored when `text_kwargs` is passed." + }, + { + "left": "issue:33666", + "right": "issue:36331", + "accept": false, + "reason": "Different areas and symptoms: multi-GPU training setup vs `compute_loss()` signature breakage in a custom trainer." + }, + { + "left": "issue:43525", + "right": "issue:43572", + "accept": false, + "reason": "Both are missing token-id config fields after v5, but they affect different model configs (`Llama4Config` vs `StableLmConfig`) and likely separate fixes." + }, + { + "left": "issue:44849", + "right": "issue:46209", + "accept": false, + "reason": "Qwen3.5 hidden-state behavior vs FP8 inference failure on B200 are unrelated code paths." + }, + { + "left": "issue:43278", + "right": "issue:43688", + "accept": false, + "reason": "Embedding dtype drift during evaluate is unrelated to MoE auxiliary-loss normalization." + }, + { + "left": "issue:43720", + "right": "issue:44361", + "accept": false, + "reason": "Packed-weight loading bug in BitNet vs tokenizer/task AttributeError in MLuke are different subsystems." + }, + { + "left": "issue:44625", + "right": "issue:46002", + "accept": false, + "reason": "Config propagation for Qwen3.5 labels is not the same as NaNs in classification heads after checkout/init." + }, + { + "left": "issue:43493", + "right": "issue:43825", + "accept": false, + "reason": "SigLIP2 implementation mismatch is unrelated to the pipeline error-message regression." + }, + { + "left": "issue:28218", + "right": "issue:33357", + "accept": false, + "reason": "Tokenizer spacing bug and MacOS CLIP bus error do not share the same underlying failure." + }, + { + "left": "issue:39692", + "right": "issue:43493", + "accept": false, + "reason": "Docs-example mistakes are not the same bug as a SigLIP2 implementation discrepancy." + }, + { + "left": "issue:44964", + "right": "issue:45405", + "accept": false, + "reason": "Model loading failure and a dependency-version bump issue are unrelated." + }, + { + "left": "issue:44849", + "right": "issue:45357", + "accept": false, + "reason": "Different Qwen3.5 regressions: hidden states vs incorrect visual-encoder keys on save." + }, + { + "left": "issue:44568", + "right": "issue:45812", + "accept": false, + "reason": "Both affect tokenization, but one is about special tokens on a specific tokenizer and the other is a broad Granite token-ID regression." + }, + { + "left": "issue:44464", + "right": "issue:46424", + "accept": false, + "reason": "Chunked-generation inconsistency with compiled forward is not the same as cache-length handling for quantized cache." + }, + { + "left": "issue:45200", + "right": "issue:45229", + "accept": false, + "reason": "Gemma4 text-only token-type defaults and multi-GPU OOM are different problems." + }, + { + "left": "issue:43540", + "right": "issue:44977", + "accept": false, + "reason": "Video-input processing in Qwen3OmniMoe is unrelated to flash-attention generation behavior in Qwen3.5." + }, + { + "left": "issue:43493", + "right": "issue:43844", + "accept": false, + "reason": "SigLIP2 model discrepancy and gradient blow-up under DeepSpeed/ZeRO-3 are different code paths." + }, + { + "left": "issue:44488", + "right": "issue:44843", + "accept": false, + "reason": "Loading a specific model and offline-mode breakage from `model_info()` are unrelated." + }, + { + "left": "issue:39692", + "right": "issue:43295", + "accept": false, + "reason": "Documentation example errors are not the same bug as the processor/tokenizer regression in v4.57.5." + }, + { + "left": "issue:43278", + "right": "issue:43844", + "accept": false, + "reason": "Training/eval dtype change is unrelated to ZeRO-3 gradient inflation." + }, + { + "left": "issue:43334", + "right": "issue:43572", + "accept": false, + "reason": "Both mention missing pad-token fields, but they are different configs and symptoms (`pad_token_id` vs `pad_token_idx`)." + }, + { + "left": "issue:43646", + "right": "issue:44466", + "accept": false, + "reason": "Custom model initialization breakage is not the same as device-dependent `lm_head.weight` serialization." + }, + { + "left": "issue:33357", + "right": "issue:33666", + "accept": false, + "reason": "MacOS CLIP bus error and Qwen2-VL multi-GPU training are unrelated." + }, + { + "left": "issue:43334", + "right": "issue:45759", + "accept": false, + "reason": "Missing `pad_token_id` in Qwen3-VL config is unrelated to `from_config` not unwrapping `text_config`." + }, + { + "left": "issue:43257", + "right": "issue:43299", + "accept": false, + "reason": "Both involve Qwen3 MoE, but one is an accelerate/deepspeed weight-conversion bug and the other is a load-time regression in dev; not the same concrete fix." + }, + { + "left": "issue:44589", + "right": "issue:45520", + "accept": false, + "reason": "Float8 storage lookup failure and missing `flash_attn` import key on Python 3.13 are unrelated." + }, + { + "left": "issue:43824", + "right": "issue:44589", + "accept": false, + "reason": "Missing Qwen2.5-VL symbol export and Float8 storage lookup are separate failures." + }, + { + "left": "issue:45229", + "right": "issue:45823", + "accept": false, + "reason": "Gemma4 OOM under multi-GPU inference is not the same as a PLE device-mismatch forward bug." + }, + { + "left": "issue:43927", + "right": "issue:44361", + "accept": false, + "reason": "Dia save/load token-ID loss and MLuke tokenizer task AttributeError are different issues." + }, + { + "left": "issue:43504", + "right": "issue:43606", + "accept": false, + "reason": "Legacy-field loading failure in Beit semantic segmentation is unrelated to CPU-offload device mismatch in bark-small." + }, + { + "left": "issue:46424", + "right": "issue:44464", + "accept": false, + "reason": "Cache max-length being ignored for quantized cache is not the same bug as inconsistent chunked generation with compiled forward." + }, + { + "left": "issue:43572", + "right": "issue:44291", + "accept": false, + "reason": "Missing `pad_token_idx` in one config and unexpected `_is_hf_initialized` during `init_empty_weights` are different regressions." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 24, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:15354", + "issue:28218", + "issue:29159", + "issue:29942", + "issue:30333", + "issue:32090", + "issue:34567", + "issue:37428", + "issue:38175", + "issue:42490", + "issue:43065", + "issue:43278", + "issue:43504", + "issue:43540", + "issue:43575", + "issue:43582", + "issue:43606", + "issue:43653", + "issue:43696", + "issue:43701", + "issue:43844", + "issue:43873", + "issue:43957", + "issue:43994", + "issue:44112", + "issue:44423", + "issue:44493", + "issue:44521", + "issue:44554", + "issue:44734", + "issue:44792", + "issue:44964", + "issue:44977", + "issue:45072", + "issue:45084", + "issue:45198", + "issue:45200", + "issue:45290", + "issue:45405", + "issue:45406", + "issue:45446", + "issue:45520", + "issue:45588", + "issue:45823", + "issue:46050", + "issue:46121", + "issue:46285", + "issue:46369" + ], + "soft_pairs": [ + "issue:43278|issue:43701", + "issue:43575|issue:45405", + "issue:44493|issue:46285", + "issue:44792|issue:45405", + "issue:43504|issue:43653", + "issue:43606|issue:44112", + "issue:43994|issue:44493", + "issue:43873|issue:46050", + "issue:45588|issue:46050", + "issue:44423|issue:44734", + "issue:43065|issue:46369", + "issue:29942|issue:30333", + "issue:42490|issue:43696", + "issue:45406|issue:46121", + "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:15354|issue:38175", + "issue:43540|issue:45200", + "issue:44521|issue:45290", + "issue:45200|issue:45823", + "issue:45072|issue:45198", + "issue:37428|issue:43957" + ], + "prepared_review_unit_hash": "37099c17afa29e0f5b4fd755c2e7f7462bf987199ad1eb6c44ec6e6f99e17f49", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 26, + "serialized_chars": 35585, + "estimated_input_tokens": 8897, + "estimated_eval_tokens": 18050 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is highly heterogeneous: it mixes unrelated tokenizer, training, serving, quantization, attention, and model-loading bugs. None of the proposed soft edges look like the same underlying issue, so all should be rejected. No PRs are present.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43278", + "right": "issue:43701", + "accept": false, + "reason": "One is BF16/FP32 dtype drift in evaluation; the other is a resume_from_checkpoint key mismatch. Different code paths and failure modes." + }, + { + "left": "issue:43575", + "right": "issue:45405", + "accept": false, + "reason": "Tensor-parallel OOM while loading a large model is unrelated to a PEFT minimum-version packaging issue." + }, + { + "left": "issue:44493", + "right": "issue:46285", + "accept": false, + "reason": "Unexpected position-id key handling is not the same as a CLIP text-model structure/API change." + }, + { + "left": "issue:44792", + "right": "issue:45405", + "accept": false, + "reason": "A failing Janus image-generation test is unrelated to the PEFT version bump problem." + }, + { + "left": "issue:43504", + "right": "issue:43653", + "accept": false, + "reason": "Legacy field loading in Beit segmentation and BigBirdTokenizer special-token registration are unrelated bugs." + }, + { + "left": "issue:43606", + "right": "issue:44112", + "accept": false, + "reason": "Both mention device-related CI failures, but one is CPU offload mismatch and the other is a stale GraniteSpeech test; not the same bug." + }, + { + "left": "issue:43994", + "right": "issue:44493", + "accept": false, + "reason": "SigLIP2 incorrect outputs and unexpected position-id key warnings are different model-level issues." + }, + { + "left": "issue:43873", + "right": "issue:46050", + "accept": false, + "reason": "Quantized offloading behavior and sinq save/load corruption are different quantization problems with different failure points." + }, + { + "left": "issue:45588", + "right": "issue:46050", + "accept": false, + "reason": "A flash-attention integration crash on s_aux=None is unrelated to sinq serialization breakage." + }, + { + "left": "issue:44423", + "right": "issue:44734", + "accept": false, + "reason": "Both affect serve, but one is multimodal continuous batching and the other is KV-cache continuation indexing; different concrete code paths." + }, + { + "left": "issue:43065", + "right": "issue:46369", + "accept": false, + "reason": "Dummy Conv2d in Sam3PixelDecoder and Sam3-litetext memory overflow are distinct SAM3 issues." + }, + { + "left": "issue:29942", + "right": "issue:30333", + "accept": false, + "reason": "Flash Attention 2 test failures and MLFlow job reporting on failure are unrelated subsystems." + }, + { + "left": "issue:42490", + "right": "issue:43696", + "accept": false, + "reason": "Trailing-slash import warnings are unrelated to a CUDA OOM in GPT-oss-20b." + }, + { + "left": "issue:45406", + "right": "issue:46121", + "accept": false, + "reason": "Gemma4Processor missing _tokenizer and JSON-loaded list handling in rope validation are unrelated bugs." + }, + { + "left": "issue:43844", + "right": "issue:44964", + "accept": false, + "reason": "ZeRO-3 gradient growth during training is unrelated to loading Phi-4 multimodal models." + }, + { + "left": "issue:28218", + "right": "issue:34567", + "accept": false, + "reason": "Tokenizer spacing after added tokens and TrainerState num_input_tokens_seen not updating are different tokenizer/trainer behaviors." + }, + { + "left": "issue:29159", + "right": "issue:32090", + "accept": false, + "reason": "Slow-vs-fast tokenizer inconsistency is unrelated to a Trainer GPU broadcast NoneType error." + }, + { + "left": "issue:45084", + "right": "issue:45520", + "accept": false, + "reason": "Template-node compilation failure and a flash_attn import KeyError are separate import/compile problems." + }, + { + "left": "issue:44554", + "right": "issue:45446", + "accept": false, + "reason": "An attention correctness bug on MPS is unrelated to a flex_attention PyTorch version check import bug." + }, + { + "left": "issue:43582", + "right": "issue:44977", + "accept": false, + "reason": "Apple Silicon allocator TypeError and Qwen3.5 flash-attention generation issues are different failures." + }, + { + "left": "issue:15354", + "right": "issue:38175", + "accept": false, + "reason": "TorchScript GeneratorExp export failure is unrelated to unexpected zero probabilities in SigLIP2." + }, + { + "left": "issue:43540", + "right": "issue:45200", + "accept": false, + "reason": "Video-input processing in Qwen3OmniMoe and Gemma4 mm_token_type_ids defaults are different multimodal model issues." + }, + { + "left": "issue:44521", + "right": "issue:45290", + "accept": false, + "reason": "All-zero assistant masks for multimodal inputs and tool-call messages with no content are both chat-template related, but they are distinct edge cases." + }, + { + "left": "issue:45200", + "right": "issue:45823", + "accept": false, + "reason": "Missing mm_token_type_ids defaults in fine-tuning is unrelated to a device_map auto forward mismatch in Gemma4 PLE." + }, + { + "left": "issue:45072", + "right": "issue:45198", + "accept": false, + "reason": "bfloat16 dtype mismatches in inference and Wav2Vec2 save_pretrained/tokenization failures are unrelated." + }, + { + "left": "issue:37428", + "right": "issue:43957", + "accept": false, + "reason": "Flash-support import errors and meta-device loading regressions are separate model-loading 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": 19, + "review_unit_index": 25, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:28218", + "issue:35141", + "issue:37428", + "issue:38175", + "issue:39692", + "issue:41720", + "issue:42490", + "issue:42915", + "issue:43232", + "issue:43295", + "issue:43334", + "issue:43388", + "issue:43450", + "issue:43493", + "issue:43525", + "issue:43526", + "issue:43550", + "issue:43572", + "issue:43575", + "issue:43701", + "issue:43824", + "issue:43844", + "issue:43873", + "issue:43950", + "issue:43957", + "issue:43975", + "issue:44186", + "issue:44466", + "issue:44554", + "issue:44610", + "issue:44792", + "issue:44849", + "issue:44898", + "issue:44987", + "issue:45127", + "issue:45200", + "issue:45325", + "issue:45356", + "issue:45381", + "issue:45412", + "issue:45520", + "issue:45563", + "issue:45823", + "issue:46178", + "issue:46192", + "issue:46209", + "issue:46285", + "issue:46365" + ], + "soft_pairs": [ + "issue:43493|issue:43701", + "issue:44186|issue:46192", + "issue:43232|issue:44849", + "issue:45823|issue:46192", + "issue:45381|issue:46209", + "issue:43295|issue:43450", + "issue:43295|issue:44610", + "issue:44554|issue:46365", + "issue:39692|issue:46192", + "issue:43572|issue:44987", + "issue:43295|issue:43873", + "issue:44554|issue:45356", + "issue:43572|issue:43957", + "issue:45520|issue:46178", + "issue:37428|issue:43824", + "issue:44466|issue:45127", + "issue:43950|issue:46285", + "issue:44898|issue:45412", + "issue:43550|issue:43844", + "issue:45200|issue:45325", + "issue:43334|issue:43525", + "issue:43450|issue:43575", + "issue:42915|issue:43824", + "issue:42490|issue:43493", + "issue:41720|issue:46209", + "issue:43824|issue:43975", + "issue:44792|issue:45563", + "issue:43388|issue:43526", + "issue:28218|issue:35141", + "issue:38175|issue:43493" + ], + "prepared_review_unit_hash": "802bc70d31a9cdd6cf52ade4a8c038a6bcd6264dba55fd36ed9ef14e36fd8592", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 37015, + "estimated_input_tokens": 9254, + "estimated_eval_tokens": 18764 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous. The closest items form a loose processor/tokenizer-regression subgroup, but none of the soft pairs look like the same concrete bug or change. Issue 43295 is the most central representative of that subgroup.", + "confidence": 0.88, + "canonical_issue_reason": "Issue 43295 is the broadest, most central item among the few related processor/tokenizer regression reports and has multiple soft links, making it the best cluster anchor.", + "canonical_pr_reason": null, + "best_issue_reason": "43295 best represents the cluster\u2019s main recurring theme (processor/tokenizer regressions) and is the most general, well-scoped issue among the candidates.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43493", + "right": "issue:43701", + "accept": false, + "reason": "Different problems: SigLIP2 implementation discrepancy vs. resume-from-checkpoint key mismatch." + }, + { + "left": "issue:44186", + "right": "issue:46192", + "accept": false, + "reason": "Both involve tokenization/processor APIs, but the crashes/bugs are in different code paths and are not the same underlying issue." + }, + { + "left": "issue:43232", + "right": "issue:44849", + "accept": false, + "reason": "Generation kwargs update after sync_gpus is unrelated to the Qwen3.5 hidden-states bug." + }, + { + "left": "issue:45823", + "right": "issue:46192", + "accept": false, + "reason": "Device-mapping forward mismatch is unrelated to silently ignored return_tensors handling." + }, + { + "left": "issue:45381", + "right": "issue:46209", + "accept": false, + "reason": "Qwen2.5-VL video position-id bug and FP8 inference failure on B200 are different code-path failures." + }, + { + "left": "issue:43295", + "right": "issue:43450", + "accept": false, + "reason": "Both are processor-related, but one is about processor.tokenizer access/passing images and the other is batched video shape output; not the same bug." + }, + { + "left": "issue:43295", + "right": "issue:44610", + "accept": false, + "reason": "Different model-specific processor bugs: custom tokenizer/image handling vs. OmDet image size mismatch." + }, + { + "left": "issue:44554", + "right": "issue:46365", + "accept": false, + "reason": "MPS attention correctness and Trainer import failure are unrelated." + }, + { + "left": "issue:39692", + "right": "issue:46192", + "accept": false, + "reason": "Documentation/model mismatch and return_tensors being ignored are separate issues." + }, + { + "left": "issue:43572", + "right": "issue:44987", + "accept": false, + "reason": "Missing pad_token_idx in a config and a model-loading failure for a specific repo are not the same bug." + }, + { + "left": "issue:43295", + "right": "issue:43873", + "accept": false, + "reason": "Processor/tokenizer regression is unrelated to quantization offloading behavior." + }, + { + "left": "issue:44554", + "right": "issue:45356", + "accept": false, + "reason": "Attention on MPS and Kimi-K2.5 tokenizer regression are unrelated." + }, + { + "left": "issue:43572", + "right": "issue:43957", + "accept": false, + "reason": "Both are loading/config regressions, but different models and failure modes." + }, + { + "left": "issue:45520", + "right": "issue:46178", + "accept": false, + "reason": "Different import/backend lookup failures in separate modules." + }, + { + "left": "issue:37428", + "right": "issue:43824", + "accept": false, + "reason": "Distinct import errors for different symbols/models." + }, + { + "left": "issue:44466", + "right": "issue:45127", + "accept": false, + "reason": "Both touch tied-weight behavior, but one is serialization inconsistency and the other is LoRA merge collapse; not the same concrete issue." + }, + { + "left": "issue:43950", + "right": "issue:46285", + "accept": false, + "reason": "Non-persistent buffer corruption is unrelated to CLIPTextModel structure changes." + }, + { + "left": "issue:44898", + "right": "issue:45412", + "accept": false, + "reason": "Perceiver resize failure and RT-DETR memory retention are unrelated." + }, + { + "left": "issue:43550", + "right": "issue:43844", + "accept": false, + "reason": "torch.compile/SDPA failure is unrelated to gradient blow-up under ZeRO-3." + }, + { + "left": "issue:45200", + "right": "issue:45325", + "accept": false, + "reason": "Text-only mm_token_type_ids defaults and Qwen2.5-VL rope indexing are different bugs." + }, + { + "left": "issue:43334", + "right": "issue:43525", + "accept": false, + "reason": "Similar symptom name, but different model families and code paths; not the same underlying fix." + }, + { + "left": "issue:43450", + "right": "issue:43575", + "accept": false, + "reason": "Video processor batching bug is unrelated to Qwen2-57B TP OOM." + }, + { + "left": "issue:42915", + "right": "issue:43824", + "accept": false, + "reason": "FP8 config failure for Qwen3Moe is unrelated to the Qwen2.5-VL import error." + }, + { + "left": "issue:42490", + "right": "issue:43493", + "accept": false, + "reason": "Local-path warning behavior is unrelated to the SigLIP2 discrepancy." + }, + { + "left": "issue:41720", + "right": "issue:46209", + "accept": false, + "reason": "CUDA assert under auto device mapping and FP8 inference failure on B200 are different hardware/runtime issues." + }, + { + "left": "issue:43824", + "right": "issue:43975", + "accept": false, + "reason": "Importing a class and tokenizer detokenization are unrelated." + }, + { + "left": "issue:44792", + "right": "issue:45563", + "accept": false, + "reason": "Failed image generation test and stale warning text are not the same bug." + }, + { + "left": "issue:43388", + "right": "issue:43526", + "accept": false, + "reason": "Different label-handling layers: metrics gathering vs. image processor reduction." + }, + { + "left": "issue:28218", + "right": "issue:35141", + "accept": false, + "reason": "Tokenizer spacing after added tokens is unrelated to embedding reinitialization after resize." + }, + { + "left": "issue:38175", + "right": "issue:43493", + "accept": false, + "reason": "Same model family, but one is zero probabilities and the other is a broader implementation discrepancy; not enough to treat as the same bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 26, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:29942", + "issue:32090", + "issue:37428", + "issue:38175", + "issue:43262", + "issue:43278", + "issue:43335", + "issue:43425", + "issue:43450", + "issue:43526", + "issue:43550", + "issue:43606", + "issue:43611", + "issue:43653", + "issue:43819", + "issue:43827", + "issue:43856", + "issue:43901", + "issue:44112", + "issue:44206", + "issue:44442", + "issue:44448", + "issue:44493", + "issue:44514", + "issue:44521", + "issue:44561", + "issue:44610", + "issue:44743", + "issue:44855", + "issue:44857", + "issue:45083", + "issue:45127", + "issue:45198", + "issue:45200", + "issue:45245", + "issue:45290", + "issue:45412", + "issue:45520", + "issue:45584", + "issue:45588", + "issue:45593", + "issue:45759", + "issue:46170", + "issue:46192", + "issue:46234", + "issue:46285", + "issue:46326", + "issue:46402" + ], + "soft_pairs": [ + "issue:43278|issue:43856", + "issue:43526|issue:46192", + "issue:43653|issue:44112", + "issue:44442|issue:45198", + "issue:45290|issue:46326", + "issue:38175|issue:45412", + "issue:43262|issue:43450", + "issue:45520|issue:46170", + "issue:45759|issue:46285", + "issue:44514|issue:44521", + "issue:44610|issue:44857", + "issue:44561|issue:46234", + "issue:44206|issue:45584", + "issue:43819|issue:44112", + "issue:43827|issue:43901", + "issue:43819|issue:45083", + "issue:44857|issue:45593", + "issue:43335|issue:43550", + "issue:43611|issue:46402", + "issue:37428|issue:43425", + "issue:43606|issue:45083", + "issue:29942|issue:32090", + "issue:44448|issue:44743", + "issue:45200|issue:45588", + "issue:45127|issue:45245", + "issue:43335|issue:43653", + "issue:43819|issue:44493", + "issue:44855|issue:45083" + ], + "prepared_review_unit_hash": "ec29193c84294c7e88b4afd86360a9942a412df55eed712685de61f8e747aded", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 36344, + "estimated_input_tokens": 9086, + "estimated_eval_tokens": 18428 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a grab-bag of mostly unrelated issue reports. There are a few loose thematic overlaps (chat-template bugs, tokenizer/model-loading regressions, import-utils symbols), but none are the same underlying defect, so no soft-edge pair should be merged.", + "confidence": 0.91, + "canonical_issue_reason": "issue:44521 is the strongest issue anchor here: it is open, high-activity, and has the broadest concrete reproduction around multimodal apply_chat_template behavior.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44521 is the best overall issue candidate because it is active, well-scoped, and central enough to serve as the representative issue for this mixed set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43278", + "right": "issue:43856", + "accept": false, + "reason": "Different bugs: eval dtype drift vs MoE training memory usage." + }, + { + "left": "issue:43526", + "right": "issue:46192", + "accept": false, + "reason": "Different processor bugs: label reduction vs return_tensors handling." + }, + { + "left": "issue:43653", + "right": "issue:44112", + "accept": false, + "reason": "Tokenizer special-token registration issue vs GraniteSpeech CI device-override test failure." + }, + { + "left": "issue:44442", + "right": "issue:45198", + "accept": false, + "reason": "Both audio-related, but different failure modes and different model/tokenizer stacks." + }, + { + "left": "issue:45290", + "right": "issue:46326", + "accept": false, + "reason": "Both touch apply_chat_template, but one is tool-call empty-content crashing and the other is mistral_common continue_final_message behavior." + }, + { + "left": "issue:38175", + "right": "issue:45412", + "accept": false, + "reason": "Different subsystems: unexpected probabilities vs RT-DETR memory release." + }, + { + "left": "issue:43262", + "right": "issue:43450", + "accept": false, + "reason": "Audio sampling-rate default bug vs video batch-shape bug." + }, + { + "left": "issue:45520", + "right": "issue:46170", + "accept": false, + "reason": "Unrelated platform/import problems: flash_attn lookup on Python 3.13 vs Windows ROCm torch.distributed dependency." + }, + { + "left": "issue:45759", + "right": "issue:46285", + "accept": false, + "reason": "Both concern composite model structure, but the concrete failures differ: text_config unwrapping vs CLIP .text_model inconsistency." + }, + { + "left": "issue:44514", + "right": "issue:44521", + "accept": false, + "reason": "Both are VL/chat-template issues, but one is a padding=False batch crash and the other is all-zero assistant masks." + }, + { + "left": "issue:44610", + "right": "issue:44857", + "accept": false, + "reason": "Different model and bug class: processor image size mismatch vs AMP loss crash." + }, + { + "left": "issue:44561", + "right": "issue:46234", + "accept": false, + "reason": "Both mention import_utils availability helpers, but they are missing different symbols with different breakage paths." + }, + { + "left": "issue:44206", + "right": "issue:45584", + "accept": false, + "reason": "Feature-extractor arg regression vs Whisper empty-transcription generation failure." + }, + { + "left": "issue:43819", + "right": "issue:44112", + "accept": false, + "reason": "Different defects: DAC latent parity vs GraniteSpeech CI device override." + }, + { + "left": "issue:43827", + "right": "issue:43901", + "accept": false, + "reason": "Both docs issues, but they target different pipeline APIs and different deprecations." + }, + { + "left": "issue:43819", + "right": "issue:45083", + "accept": false, + "reason": "Unrelated audio/codebook latent mismatch vs helper length-calculation behavior." + }, + { + "left": "issue:44857", + "right": "issue:45593", + "accept": false, + "reason": "Different detection-model bugs: float16 AMP crash vs auxiliary-loss omission." + }, + { + "left": "issue:43335", + "right": "issue:43550", + "accept": false, + "reason": "Config construction bug vs torch.compile/SDPA runtime incompatibility." + }, + { + "left": "issue:43611", + "right": "issue:46402", + "accept": false, + "reason": "Both affect model loading, but one is base_model_prefix handling and the other is missing Sam3 text-encoder weights." + }, + { + "left": "issue:37428", + "right": "issue:43425", + "accept": false, + "reason": "Specific missing flash_attention_utils symbol vs broad Torch 2.10 incompatibility." + }, + { + "left": "issue:43606", + "right": "issue:45083", + "accept": false, + "reason": "CPU-offload device mismatch vs audio feature-length helper behavior; not the same bug." + }, + { + "left": "issue:29942", + "right": "issue:32090", + "accept": false, + "reason": "Flash Attention 2 test failures are unrelated to NoneType GPU broadcast errors in Trainer." + }, + { + "left": "issue:44448", + "right": "issue:44743", + "accept": false, + "reason": "Different generation regressions: Pegasus v4/v5 output drift vs qwen3_5 recurrent state reset with cache." + }, + { + "left": "issue:45200", + "right": "issue:45588", + "accept": false, + "reason": "Different multimodal processor bugs: missing mm_token_type_ids default vs flash_attention.py s_aux=None crash." + }, + { + "left": "issue:45127", + "right": "issue:45245", + "accept": false, + "reason": "LoRA/tied-embedding collapse is unrelated to category-count overflow." + }, + { + "left": "issue:43335", + "right": "issue:43653", + "accept": false, + "reason": "Sparse-layer config bug vs tokenizer special-token registration bug." + }, + { + "left": "issue:43819", + "right": "issue:44493", + "accept": false, + "reason": "Different symptoms and code paths: DAC latent mismatch vs unexpected position-id keys across models." + }, + { + "left": "issue:44855", + "right": "issue:45083", + "accept": false, + "reason": "Python 3.13 decorator parsing/import error vs qwen3_omni_moe feature-length helper 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": 19, + "review_unit_index": 27, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:28218", + "issue:29127", + "issue:29942", + "issue:32090", + "issue:34689", + "issue:35141", + "issue:37428", + "issue:38175", + "issue:39692", + "issue:40444", + "issue:42490", + "issue:43278", + "issue:43335", + "issue:43504", + "issue:43526", + "issue:43550", + "issue:43606", + "issue:43701", + "issue:43720", + "issue:43746", + "issue:43825", + "issue:43854", + "issue:44112", + "issue:44206", + "issue:44351", + "issue:44360", + "issue:44448", + "issue:44493", + "issue:44561", + "issue:44779", + "issue:44857", + "issue:44977", + "issue:45072", + "issue:45083", + "issue:45229", + "issue:45290", + "issue:45310", + "issue:45381", + "issue:45459", + "issue:45478", + "issue:45588", + "issue:45593", + "issue:45823", + "issue:45859", + "issue:45920", + "issue:46192", + "issue:46209", + "issue:46234" + ], + "soft_pairs": [ + "issue:37428|issue:44351", + "issue:43550|issue:44112", + "issue:44206|issue:45823", + "issue:45290|issue:45381", + "issue:43335|issue:43504", + "issue:29127|issue:32090", + "issue:45588|issue:45859", + "issue:43278|issue:43825", + "issue:39692|issue:43526", + "issue:45310|issue:45478", + "issue:39692|issue:44493", + "issue:45459|issue:46234", + "issue:43335|issue:45083", + "issue:43720|issue:44112", + "issue:44448|issue:44779", + "issue:43335|issue:44112", + "issue:43335|issue:45593", + "issue:43606|issue:44857", + "issue:42490|issue:43854", + "issue:34689|issue:35141", + "issue:44360|issue:44561", + "issue:38175|issue:44493", + "issue:44857|issue:45072", + "issue:45920|issue:46192", + "issue:44448|issue:44977", + "issue:44977|issue:45229", + "issue:28218|issue:32090", + "issue:40444|issue:43526", + "issue:43746|issue:46209", + "issue:43701|issue:45083", + "issue:29942|issue:35141" + ], + "prepared_review_unit_hash": "ace45230a82d5a3e03ce3f2620a34902461580ac751fcf51d15de5f351044844", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 37048, + "estimated_input_tokens": 9262, + "estimated_eval_tokens": 18780 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a set of unrelated issue reports. The only clear duplicate pair is the repeated Qwen3.5 MoE `from_pretrained` error report; the rest differ by model, subsystem, or failure mode and should stay separate.", + "confidence": 0.93, + "canonical_issue_reason": "issue:45310 is the earlier, broader report of the same Qwen3.5 MoE `from_pretrained` regression as issue:45478, so it is the best issue to keep as canonical among the candidates.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45310 best represents the only true duplicate theme in this set: the Qwen3.5 MoE `from_pretrained` regression. It is the earlier report and captures the concrete failure more cleanly than the later repost.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:37428", + "right": "issue:44351", + "accept": false, + "reason": "Both are import errors, but they concern different missing symbols (`_flash_supports_window_size` vs `HybridCache`) and likely different code paths." + }, + { + "left": "issue:43550", + "right": "issue:44112", + "accept": false, + "reason": "Different models and failures: torch.compile/SDPA behavior in Bamba versus a GraniteSpeech CI device-override issue." + }, + { + "left": "issue:44206", + "right": "issue:45823", + "accept": false, + "reason": "One is a feature-extractor argument regression; the other is a Gemma4 device-mapping mismatch during forward. Not the same bug." + }, + { + "left": "issue:45290", + "right": "issue:45381", + "accept": false, + "reason": "Chat template tool-call crash and Qwen2.5-VL video position-id bug are separate failure modes." + }, + { + "left": "issue:43335", + "right": "issue:43504", + "accept": false, + "reason": "Different architectures and symptoms: SwitchTransformers config construction versus BEiT pretrained preset loading." + }, + { + "left": "issue:29127", + "right": "issue:32090", + "accept": false, + "reason": "An error-message clarity issue in LayoutLMv3 is unrelated to the Trainer `_gpu_broadcast_one` TypeError." + }, + { + "left": "issue:45588", + "right": "issue:45859", + "accept": false, + "reason": "Both touch model internals, but one is a flash-attention crash on `s_aux=None` and the other is a CPU-offload incompatibility in rotary embeddings." + }, + { + "left": "issue:43278", + "right": "issue:43825", + "accept": false, + "reason": "A dtype change between train/eval is unrelated to a pipeline warning about translation tasks in v5." + }, + { + "left": "issue:39692", + "right": "issue:43526", + "accept": false, + "reason": "Documentation/example errors for SigLIP2 are unrelated to BEiT Fast label reduction returning only one label." + }, + { + "left": "issue:45310", + "right": "issue:45478", + "accept": true, + "reason": "These are the same Qwen3.5 MoE `from_pretrained` error report; the titles and described failure match closely." + }, + { + "left": "issue:39692", + "right": "issue:44493", + "accept": false, + "reason": "A SigLIP2 doc example problem is not the same as the broad positional-id warning regression." + }, + { + "left": "issue:45459", + "right": "issue:46234", + "accept": false, + "reason": "Tokenizer protobuf error handling and a missing import-utils symbol are unrelated issues." + }, + { + "left": "issue:43335", + "right": "issue:45083", + "accept": false, + "reason": "SwitchTransformers sparse-layer construction and qwen3_omni_moe feature-length computation are different code paths." + }, + { + "left": "issue:43720", + "right": "issue:44112", + "accept": false, + "reason": "Packed-weight loading in BitNet is unrelated to the GraniteSpeech stale device-override CI failure." + }, + { + "left": "issue:44448", + "right": "issue:44779", + "accept": false, + "reason": "Both are output regressions, but one is Pegasus generation drift and the other is DeepSeek tokenizer tokenization; not the same bug." + }, + { + "left": "issue:43335", + "right": "issue:45593", + "accept": false, + "reason": "A SwitchTransformers config bug is not the same as D-FINE missing auxiliary losses when denoising is off." + }, + { + "left": "issue:43606", + "right": "issue:44857", + "accept": false, + "reason": "Both are runtime failures, but one is CPU-offload device mismatch and the other is AMP/CUDA loss behavior in LwDetrImageLoss." + }, + { + "left": "issue:42490", + "right": "issue:43854", + "accept": false, + "reason": "A local-path trailing-slash warning is unrelated to a GLM-4.7-Flash load failure in tests." + }, + { + "left": "issue:34689", + "right": "issue:35141", + "accept": false, + "reason": "Llama 3.2 Vision loading breakage and embedding reinitialization on resize are different issues." + }, + { + "left": "issue:44360", + "right": "issue:44561", + "accept": false, + "reason": "A DSA indexer activation omission is unrelated to the removal of `is_torch_fx_available` breaking `trust_remote_code` models." + }, + { + "left": "issue:38175", + "right": "issue:44493", + "accept": false, + "reason": "Unexpected zero probabilities in SigLIP2 and unexpected key/position-id warnings are not the same underlying defect." + }, + { + "left": "issue:44857", + "right": "issue:45072", + "accept": false, + "reason": "Both mention precision/device behavior, but they concern different models and different failure modes." + }, + { + "left": "issue:45920", + "right": "issue:46192", + "accept": false, + "reason": "Wrong token IDs across multiple models is a tokenizer regression; silently ignoring `return_tensors` is a separate API bug." + }, + { + "left": "issue:44448", + "right": "issue:44977", + "accept": false, + "reason": "Pegasus v4/v5 output drift and Qwen3.5 flash-attention generation failure are distinct model-specific problems." + }, + { + "left": "issue:44977", + "right": "issue:45229", + "accept": false, + "reason": "Flash-attention generation failure in Qwen3.5 is unrelated to Gemma4 multi-GPU OOM." + }, + { + "left": "issue:28218", + "right": "issue:32090", + "accept": false, + "reason": "Tokenizer spacing after added tokens has no connection to the Trainer broadcast TypeError." + }, + { + "left": "issue:40444", + "right": "issue:43526", + "accept": false, + "reason": "Multi-image IterableDataset fine-tuning failure is unrelated to BEiT Fast label reduction behavior." + }, + { + "left": "issue:43746", + "right": "issue:46209", + "accept": false, + "reason": "PEFT adapter loading from local checkpoints and FP8 inference on B200 are separate issues." + }, + { + "left": "issue:43701", + "right": "issue:45083", + "accept": false, + "reason": "Checkpoint key mismatch and qwen3_omni_moe output-length helper behavior are different bugs." + }, + { + "left": "issue:29942", + "right": "issue:35141", + "accept": false, + "reason": "Flash Attention 2 test failures and embedding reinitialization after resize are unrelated." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative: it only treats the exact Qwen3.5 MoE `from_pretrained` pair as a duplicate and keeps the other soft pairs separate based on different titles/failure modes. The canonical-issue choice is also supported by the earlier identical report. " + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 28, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:10000", + "issue:15354", + "issue:25251", + "issue:27957", + "issue:32090", + "issue:33290", + "issue:33357", + "issue:34689", + "issue:37428", + "issue:41720", + "issue:42490", + "issue:43278", + "issue:43329", + "issue:43381", + "issue:43412", + "issue:43550", + "issue:43575", + "issue:43653", + "issue:43873", + "issue:43927", + "issue:43976", + "issue:44206", + "issue:44448", + "issue:44466", + "issue:44479", + "issue:44561", + "issue:44568", + "issue:44743", + "issue:44779", + "issue:44855", + "issue:44857", + "issue:44871", + "issue:44977", + "issue:44987", + "issue:45083", + "issue:45198", + "issue:45229", + "issue:45245", + "issue:45372", + "issue:45406", + "issue:45440", + "issue:45491", + "issue:45563", + "issue:45588", + "issue:45698", + "issue:45803", + "issue:45823", + "issue:46399" + ], + "soft_pairs": [ + "issue:32090|issue:33357", + "issue:44855|issue:46399", + "issue:43550|issue:45083", + "issue:42490|issue:43575", + "issue:10000|issue:15354", + "issue:43927|issue:45698", + "issue:44561|issue:44779", + "issue:37428|issue:44987", + "issue:43873|issue:45229", + "issue:43278|issue:43329", + "issue:44857|issue:45588", + "issue:41720|issue:45229", + "issue:27957|issue:34689", + "issue:44977|issue:45563", + "issue:44568|issue:45491", + "issue:45372|issue:45588", + "issue:43278|issue:43381", + "issue:43412|issue:45588", + "issue:45588|issue:45823", + "issue:45372|issue:45406", + "issue:44448|issue:44466", + "issue:43976|issue:45440", + "issue:45198|issue:45803", + "issue:25251|issue:32090", + "issue:45245|issue:45563", + "issue:43653|issue:44743", + "issue:44206|issue:44479", + "issue:32090|issue:33290", + "issue:44871|issue:44977" + ], + "prepared_review_unit_hash": "abdbe7037d7b45237b9c7a8dc6b4f66123782354f9dd893356811d33e4da4fc8", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36700, + "estimated_input_tokens": 9175, + "estimated_eval_tokens": 18606 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is a mixed bag of unrelated Transformers issues rather than a duplicate set. No soft-edge pair looks like the same underlying bug or change. The most broadly representative issue is the save/load custom-module regression (#45698).", + "confidence": 0.82, + "canonical_issue_reason": "#45698 is the clearest broad regression with concrete repro and multiple references: `from_pretrained` loading the wrong custom module after `save_pretrained`.", + "canonical_pr_reason": null, + "best_issue_reason": "#45698 is the strongest issue candidate overall because it is specific, actionable, and affects a common serialization/loading path.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:32090", + "right": "issue:33357", + "accept": false, + "reason": "Different failures in different paths: Trainer GPU broadcast NoneType vs CLIP model bus error on macOS." + }, + { + "left": "issue:44855", + "right": "issue:46399", + "accept": false, + "reason": "Both mention parsing/registration problems, but they affect different models and unrelated code paths." + }, + { + "left": "issue:43550", + "right": "issue:45083", + "accept": false, + "reason": "Unrelated bugs: torch.compile/SDPA failure in Bamba vs helper output-length logic in qwen3_omni_moe." + }, + { + "left": "issue:42490", + "right": "issue:43575", + "accept": false, + "reason": "Import warning for local paths ending with slash is unrelated to Qwen2 TP OOM." + }, + { + "left": "issue:10000", + "right": "issue:15354", + "accept": false, + "reason": "Completely different models and issues: DistilBert error vs torch.jit GeneratorExp export failure." + }, + { + "left": "issue:43927", + "right": "issue:45698", + "accept": false, + "reason": "Both involve save/load, but one loses custom token IDs while the other loads the wrong custom module; different bugs." + }, + { + "left": "issue:44561", + "right": "issue:44779", + "accept": false, + "reason": "Removal of `is_torch_fx_available` breaks remote-code models; Deepseek tokenizer regression is a separate tokenizer issue." + }, + { + "left": "issue:37428", + "right": "issue:44987", + "accept": false, + "reason": "One is a missing import symbol in flash attention utils; the other is a model-loading failure for a specific repo." + }, + { + "left": "issue:43873", + "right": "issue:45229", + "accept": false, + "reason": "Quantized offloading behavior and multi-GPU CUDA OOM are different problems." + }, + { + "left": "issue:43278", + "right": "issue:43329", + "accept": false, + "reason": "BF16-to-FP32 dtype drift in eval is unrelated to the undefined-variable bug in multimodal token counting." + }, + { + "left": "issue:44857", + "right": "issue:45588", + "accept": false, + "reason": "Different crash sites: AMP image loss on CUDA vs flash_attention integration on sink-less models." + }, + { + "left": "issue:41720", + "right": "issue:45229", + "accept": false, + "reason": "Auto device-map assert on Qwen3 is not the same as Gemma4 multi-GPU OOM." + }, + { + "left": "issue:27957", + "right": "issue:34689", + "accept": false, + "reason": "Flash Attention 2 support issue and Llama 3.2 model loading regression are unrelated." + }, + { + "left": "issue:44977", + "right": "issue:45563", + "accept": false, + "reason": "Generation bug with flash-attention is unrelated to a stale warning in paged generate()." + }, + { + "left": "issue:44568", + "right": "issue:45491", + "accept": false, + "reason": "Tokenizer special-token handling and Gemma3 NaN embeddings are different failures." + }, + { + "left": "issue:45372", + "right": "issue:45588", + "accept": false, + "reason": "Gemma4 processor import dependency error is not the same as flash_attention.py crashing on `s_aux=None`." + }, + { + "left": "issue:43278", + "right": "issue:43381", + "accept": false, + "reason": "Dtype mismatch during eval is unrelated to the gradient-checkpointing eval-mode restriction." + }, + { + "left": "issue:43412", + "right": "issue:45588", + "accept": false, + "reason": "Executorch export failure and flash_attention `s_aux=None` crash are different code paths." + }, + { + "left": "issue:45588", + "right": "issue:45823", + "accept": false, + "reason": "Both are Gemma4-related, but one is a flash-attention attribute crash and the other is a device-map forward mismatch." + }, + { + "left": "issue:45372", + "right": "issue:45406", + "accept": false, + "reason": "ImportError from `mistral_common` and missing `_tokenizer` on Gemma4Processor are different regressions." + }, + { + "left": "issue:44448", + "right": "issue:44466", + "accept": false, + "reason": "Sequence-output differences and inconsistent `lm_head.weight` serialization are not the same bug." + }, + { + "left": "issue:43976", + "right": "issue:45440", + "accept": false, + "reason": "Python-version compatibility failure and DeepseekV3MoE divergence are unrelated." + }, + { + "left": "issue:45198", + "right": "issue:45803", + "accept": false, + "reason": "Wav2Vec2 save/tokenization failure and Fuyu bare-except cleanup bug do not share the same underlying defect." + }, + { + "left": "issue:25251", + "right": "issue:32090", + "accept": false, + "reason": "Pipeline `top_k` shape change is unrelated to the Trainer broadcast error." + }, + { + "left": "issue:45245", + "right": "issue:45563", + "accept": false, + "reason": "Category-count runtime limit and stale warning in paged generate are unrelated." + }, + { + "left": "issue:43653", + "right": "issue:44743", + "accept": false, + "reason": "Tokenizer special-token registration bug and recurrent-state reset in Qwen3.5 are unrelated." + }, + { + "left": "issue:44206", + "right": "issue:44479", + "accept": false, + "reason": "Feature-extractor center-arg regression and video-input regression are different issues." + }, + { + "left": "issue:32090", + "right": "issue:33290", + "accept": false, + "reason": "Unsupported NoneType broadcast and Adafactor/deepspeed OOM are unrelated training failures." + }, + { + "left": "issue:44871", + "right": "issue:44977", + "accept": false, + "reason": "Gemma eos_token mismatch is unrelated to Qwen3.5 flash-attention generation failure." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 29, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:27957", + "issue:29127", + "issue:29942", + "issue:30064", + "issue:33357", + "issue:34567", + "issue:35141", + "issue:39692", + "issue:42490", + "issue:42915", + "issue:42994", + "issue:43065", + "issue:43066", + "issue:43278", + "issue:43381", + "issue:43421", + "issue:43526", + "issue:43575", + "issue:43611", + "issue:43653", + "issue:43701", + "issue:43756", + "issue:43761", + "issue:43824", + "issue:43873", + "issue:43994", + "issue:44186", + "issue:44265", + "issue:44279", + "issue:44361", + "issue:44442", + "issue:44462", + "issue:44464", + "issue:44625", + "issue:45198", + "issue:45372", + "issue:45412", + "issue:45440", + "issue:45459", + "issue:45491", + "issue:45584", + "issue:45593", + "issue:45712", + "issue:45834", + "issue:45859", + "issue:46121", + "issue:46369", + "issue:46399" + ], + "soft_pairs": [ + "issue:43065|issue:45412", + "issue:44186|issue:45198", + "issue:43611|issue:46121", + "issue:42490|issue:43873", + "issue:44361|issue:45198", + "issue:33357|issue:39692", + "issue:29127|issue:34567", + "issue:43065|issue:43994", + "issue:45459|issue:45712", + "issue:42915|issue:42994", + "issue:43066|issue:44462", + "issue:45372|issue:45459", + "issue:29942|issue:30064", + "issue:45412|issue:46369", + "issue:43278|issue:43575", + "issue:43994|issue:45412", + "issue:44265|issue:45198", + "issue:43381|issue:43701", + "issue:43756|issue:45440", + "issue:43653|issue:44442", + "issue:43421|issue:45491", + "issue:27957|issue:29942", + "issue:29127|issue:33357", + "issue:43824|issue:44279", + "issue:43994|issue:46369", + "issue:45440|issue:45593", + "issue:43526|issue:45834", + "issue:44464|issue:45584", + "issue:46121|issue:46399", + "issue:43526|issue:44186", + "issue:29127|issue:35141", + "issue:44625|issue:45584", + "issue:43761|issue:45859" + ], + "prepared_review_unit_hash": "18588924c3926efe7337de00b0e0cc79a046b9feab2e680f40e235031901e1d1", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 33, + "serialized_chars": 37493, + "estimated_input_tokens": 9374, + "estimated_eval_tokens": 19004 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "False-positive cluster: the issues span many unrelated models, tokenizers, and failure modes. None of the soft-edge pairs look like the same underlying bug or a mergeable duplicate PR case.", + "confidence": 0.97, + "canonical_issue_reason": "No safe canonical issue stands out; the set is heterogeneous and does not center on one underlying defect.", + "canonical_pr_reason": null, + "best_issue_reason": "No single issue is a good global representative because the cluster mixes unrelated bugs across different subsystems and model families.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43065", + "right": "issue:45412", + "accept": false, + "reason": "Both mention model internals and memory, but one is a dummy Conv2d placeholder and the other is a garbage-collection memory leak; different bugs." + }, + { + "left": "issue:44186", + "right": "issue:45198", + "accept": false, + "reason": "Different tokenizer/model families and different failure paths; no shared code-path bug." + }, + { + "left": "issue:43611", + "right": "issue:46121", + "accept": false, + "reason": "Loading with base_model_prefix and rope-param dict conversion are unrelated issues." + }, + { + "left": "issue:42490", + "right": "issue:43873", + "accept": false, + "reason": "Local-path import warning and quantization/offloading behavior are unrelated." + }, + { + "left": "issue:44361", + "right": "issue:45198", + "accept": false, + "reason": "Both are tokenizer-related, but they affect different models with different symptoms and causes." + }, + { + "left": "issue:33357", + "right": "issue:39692", + "accept": false, + "reason": "MacOS bus error with CLIP is unrelated to SigLIP2 documentation/example errors." + }, + { + "left": "issue:29127", + "right": "issue:34567", + "accept": false, + "reason": "LayoutLMv3 error handling and TrainerState token counting are unrelated." + }, + { + "left": "issue:43065", + "right": "issue:43994", + "accept": false, + "reason": "Dummy Conv2d placeholder issue is not the same as quantized model saving failure." + }, + { + "left": "issue:45459", + "right": "issue:45712", + "accept": false, + "reason": "Tokenizer exception masking is unrelated to leftover dummy classes." + }, + { + "left": "issue:42915", + "right": "issue:42994", + "accept": false, + "reason": "Both involve quantization-adjacent topics, but one is FP8 loading and the other is quantized saving; different code paths." + }, + { + "left": "issue:43066", + "right": "issue:44462", + "accept": false, + "reason": "Wrong decoder type in v5 and ignoring tokenizer.json are separate tokenizer-loading bugs." + }, + { + "left": "issue:45372", + "right": "issue:45459", + "accept": false, + "reason": "Gemma 4 processor import failure and protobuf-related tokenizer error masking are unrelated." + }, + { + "left": "issue:29942", + "right": "issue:30064", + "accept": false, + "reason": "Flash Attention 2 test failures and void segmentation-map processing are different subsystems." + }, + { + "left": "issue:45412", + "right": "issue:46369", + "accept": false, + "reason": "Memory leak on deletion is not the same as memory overflow when calling sam3-litetext." + }, + { + "left": "issue:43278", + "right": "issue:43575", + "accept": false, + "reason": "Embedding dtype drift and tensor-parallel OOM are different problems." + }, + { + "left": "issue:43994", + "right": "issue:45412", + "accept": false, + "reason": "Quantized model saving failure is unrelated to model memory release on deletion." + }, + { + "left": "issue:44265", + "right": "issue:45198", + "accept": false, + "reason": "torch.export failure from torch_compilable_check is unrelated to tokenizer/save_pretrained issues." + }, + { + "left": "issue:43381", + "right": "issue:43701", + "accept": false, + "reason": "Eval-mode gradient checkpointing and checkpoint resume key mismatch are different issues." + }, + { + "left": "issue:43756", + "right": "issue:45440", + "accept": false, + "reason": "Both concern model implementation discrepancies, but they are for different model families and different missing behavior." + }, + { + "left": "issue:43653", + "right": "issue:44442", + "accept": false, + "reason": "BigBird special-token decode bug and AutoTokenizer loading failure are distinct tokenizer bugs." + }, + { + "left": "issue:43421", + "right": "issue:45491", + "accept": false, + "reason": "Runtime post-processor updates and Gemma3 NaN embeddings are unrelated." + }, + { + "left": "issue:27957", + "right": "issue:29942", + "accept": false, + "reason": "XLMRoberta Flash Attention 2 compatibility and a broader FA2 test failure are not clearly the same bug." + }, + { + "left": "issue:29127", + "right": "issue:33357", + "accept": false, + "reason": "LayoutLMv3 error messaging and CLIP MacOS bus error are unrelated." + }, + { + "left": "issue:43824", + "right": "issue:44279", + "accept": false, + "reason": "Missing Qwen2.5-VL import and a generic dependency issue do not match the same defect." + }, + { + "left": "issue:43994", + "right": "issue:46369", + "accept": false, + "reason": "Quantized saving failure and SAM3 memory overflow are unrelated." + }, + { + "left": "issue:45440", + "right": "issue:45593", + "accept": false, + "reason": "DeepSeekV3 implementation divergence and D-FINE auxiliary-loss behavior are different model bugs." + }, + { + "left": "issue:43526", + "right": "issue:45834", + "accept": false, + "reason": "BeitImageProcessorFast label reduction and Kosmos2.5 OCR index error are different image-processing failures." + }, + { + "left": "issue:44464", + "right": "issue:45584", + "accept": false, + "reason": "Chunked generation with compiled forward and Whisper empty-transcription failure are different generation paths." + }, + { + "left": "issue:46121", + "right": "issue:46399", + "accept": false, + "reason": "Rope-param dict conversion TypeError and missing decorator for rotary-pos-emb registration are unrelated." + }, + { + "left": "issue:43526", + "right": "issue:44186", + "accept": false, + "reason": "BeitImageProcessorFast label reduction and LayoutLMv2Tokenizer NER/padding crashes are unrelated." + }, + { + "left": "issue:29127", + "right": "issue:35141", + "accept": false, + "reason": "LayoutLMv3 error messaging and embedding reinitialization after resize are different bugs." + }, + { + "left": "issue:44625", + "right": "issue:45584", + "accept": false, + "reason": "Qwen3.5 num_labels propagation and Whisper empty transcription are unrelated." + }, + { + "left": "issue:43761", + "right": "issue:45859", + "accept": false, + "reason": "CLIPVision hidden_states regression and Qwen3_5Moe CPU-offload incompatibility are different code paths." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 30, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:27957", + "issue:29127", + "issue:33290", + "issue:34567", + "issue:34689", + "issue:35141", + "issue:42175", + "issue:42490", + "issue:42915", + "issue:43262", + "issue:43316", + "issue:43334", + "issue:43381", + "issue:43452", + "issue:43525", + "issue:43526", + "issue:43531", + "issue:43575", + "issue:43611", + "issue:43646", + "issue:43701", + "issue:43720", + "issue:43723", + "issue:43761", + "issue:43873", + "issue:43931", + "issue:44062", + "issue:44077", + "issue:44079", + "issue:44206", + "issue:44387", + "issue:44442", + "issue:44464", + "issue:44743", + "issue:44811", + "issue:44857", + "issue:44898", + "issue:44977", + "issue:45003", + "issue:45072", + "issue:45245", + "issue:45440", + "issue:45479", + "issue:45491", + "issue:45588", + "issue:45907", + "issue:46234" + ], + "soft_pairs": [ + "issue:44442|issue:45072", + "issue:43575|issue:43701", + "issue:42915|issue:43931", + "issue:42490|issue:43381", + "issue:43575|issue:44387", + "issue:43526|issue:44857", + "issue:43334|issue:43531", + "issue:44977|issue:45245", + "issue:43262|issue:44206", + "issue:44464|issue:45491", + "issue:27957|issue:29127", + "issue:43452|issue:45588", + "issue:44743|issue:44811", + "issue:27957|issue:33290", + "issue:43646|issue:44077", + "issue:33290|issue:34689", + "issue:44062|issue:45907", + "issue:33290|issue:34567", + "issue:33290|issue:35141", + "issue:43381|issue:43873", + "issue:44442|issue:44898", + "issue:45003|issue:45440", + "issue:43723|issue:45479", + "issue:43761|issue:44079", + "issue:42490|issue:44743", + "issue:43316|issue:43525", + "issue:42175|issue:46234", + "issue:34689|issue:43611", + "issue:43720|issue:44442" + ], + "prepared_review_unit_hash": "e0fd499cf4db813c151933d2712c5647c57a2a623de7505ae0864b2b3e23e002", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 29, + "serialized_chars": 35762, + "estimated_input_tokens": 8941, + "estimated_eval_tokens": 18138 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a mixed-topic similarity cluster, not a true duplicate set: it spans tokenizer/config API regressions, OOM/quantization issues, processor bugs, and unrelated model-loading failures. None of the soft pairs look like the same underlying bug/change closely enough to merge as duplicates. If one issue must serve as the anchor, the broad, high-activity quantization/offloading regression is the best representative.", + "confidence": 0.89, + "canonical_issue_reason": "Issue 43873 is the broadest and most active open thread in this set, with multiple inbound references and a large discussion around quantization/offloading memory behavior, so it is the best anchor if a single issue must represent the cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "43873 is the most suitable representative issue because it is current, high-signal, and covers the widest user-facing regression surface in the cluster; however, it is not a duplicate of the other items.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44442", + "right": "issue:45072", + "accept": false, + "reason": "Tokenizer loading failure vs dtype-mismatch inference bug; different code paths and symptoms." + }, + { + "left": "issue:43575", + "right": "issue:43701", + "accept": false, + "reason": "OOM during tensor-parallel model load vs resume_from_checkpoint key mismatch; unrelated bugs." + }, + { + "left": "issue:42915", + "right": "issue:43931", + "accept": false, + "reason": "FP8 config failure in Qwen3Moe vs Qwen3-VL weight-shape mismatch; different model families and failure modes." + }, + { + "left": "issue:42490", + "right": "issue:43381", + "accept": false, + "reason": "Local-path import warning vs gradient-checkpointing eval-mode restriction; no shared code-path." + }, + { + "left": "issue:43575", + "right": "issue:44387", + "accept": false, + "reason": "Both mention OOM, but one is TP load OOM and the other is int4 quantization memory growth; not the same bug." + }, + { + "left": "issue:43526", + "right": "issue:44857", + "accept": false, + "reason": "BeitImageProcessorFast label reduction bug vs LwDetrImageLoss AMP crash; unrelated subsystems." + }, + { + "left": "issue:43334", + "right": "issue:43531", + "accept": false, + "reason": "Missing pad_token_id on Qwen3-VL config vs sliding_window recurrent-state reset in Qwen3-MoE; different defects." + }, + { + "left": "issue:44977", + "right": "issue:45245", + "accept": false, + "reason": "Flash-attention generation issue vs category-cardinality limit; not the same concrete bug." + }, + { + "left": "issue:43262", + "right": "issue:44206", + "accept": false, + "reason": "Audio chat-template sampling-rate default vs unsupported 'center' arg in feature extractor; different processor paths." + }, + { + "left": "issue:44464", + "right": "issue:45491", + "accept": false, + "reason": "Compiled-forward chunked generation inconsistency vs NaNs from mixed-length sliding-window padding; different failure mechanisms." + }, + { + "left": "issue:27957", + "right": "issue:29127", + "accept": false, + "reason": "FlashAttention 2 support question vs LayoutLMv3 error-message clarity; unrelated issues." + }, + { + "left": "issue:43452", + "right": "issue:45588", + "accept": false, + "reason": "gguf_file from_pretrained breakage vs flash_attention s_aux=None crash; different loaders and code paths." + }, + { + "left": "issue:44743", + "right": "issue:44811", + "accept": false, + "reason": "Cache/recurrent-state reset bug vs Whisper batch_decode skip_special_tokens bug; unrelated functionality." + }, + { + "left": "issue:27957", + "right": "issue:33290", + "accept": false, + "reason": "Model/attention support request vs deepspeed Adafactor OOM; no shared underlying bug." + }, + { + "left": "issue:43646", + "right": "issue:44077", + "accept": false, + "reason": "Broad v5 custom-init regression vs patchtsmixer-specific post_init policy change; too different and too model-specific." + }, + { + "left": "issue:33290", + "right": "issue:34689", + "accept": false, + "reason": "Optimizer OOM in deepspeed vs Llama 3.2 Vision loading breakage; unrelated failures." + }, + { + "left": "issue:44062", + "right": "issue:45907", + "accept": false, + "reason": "AddedToken special-arg duplication vs list_repo_files revision forwarding; unrelated APIs." + }, + { + "left": "issue:33290", + "right": "issue:34567", + "accept": false, + "reason": "OOM in training/optimizer stack vs trainer token-count tracking not updating; different code paths." + }, + { + "left": "issue:33290", + "right": "issue:35141", + "accept": false, + "reason": "OOM vs embedding reinitialization after resize; unrelated bugs." + }, + { + "left": "issue:43381", + "right": "issue:43873", + "accept": false, + "reason": "Gradient checkpointing eval-mode restriction vs quantization/offloading memory regression; different concrete problems." + }, + { + "left": "issue:44442", + "right": "issue:44898", + "accept": false, + "reason": "Tokenizer load failure vs Perceiver interpolation/non-default resolution failure; different model classes and symptoms." + }, + { + "left": "issue:45003", + "right": "issue:45440", + "accept": false, + "reason": "Unsafe sys.modules access vs DeepSeekV3 implementation divergence; unrelated implementation issues." + }, + { + "left": "issue:43723", + "right": "issue:45479", + "accept": false, + "reason": "AutoTokenizer loading issue in v5 vs zero-loss sequence-classification regression; different failures." + }, + { + "left": "issue:43761", + "right": "issue:44079", + "accept": false, + "reason": "CLIP hidden_states regression vs ModelOutput key assignment bug; not the same code-path." + }, + { + "left": "issue:42175", + "right": "issue:46234", + "accept": false, + "reason": "Packaging/backends selection issue vs missing import_utils symbol; unrelated." + }, + { + "left": "issue:34689", + "right": "issue:43611", + "accept": false, + "reason": "Specific model-loading failure vs base_model_prefix loading regression; same broad area, but not the same bug." + }, + { + "left": "issue:43720", + "right": "issue:44442", + "accept": false, + "reason": "Accelerate loading/unpacking bug for BitNet vs tokenizer loading failure; unrelated." + }, + { + "left": "issue:44743", + "right": "issue:44811", + "accept": false, + "reason": "Recurrent-state reset bug vs Whisper decoding option bug; unrelated." + }, + { + "left": "issue:43316", + "right": "issue:43525", + "accept": false, + "reason": "Both touch config API shape, but one is a broad Gemma3TextConfig discrepancy and the other is a specific Llama4 pad_token_id AttributeError; not clearly the same underlying bug." + }, + { + "left": "issue:44077", + "right": "issue:44079", + "accept": false, + "reason": "PatchTSMixer post_init policy issue vs ModelOutput key handling bug; unrelated." + }, + { + "left": "issue:42175", + "right": "issue:46234", + "accept": false, + "reason": "Package install backend omission vs missing utility export; 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": 19, + "review_unit_index": 31, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:25251", + "issue:28282", + "issue:42617", + "issue:42907", + "issue:43065", + "issue:43232", + "issue:43295", + "issue:43316", + "issue:43388", + "issue:43493", + "issue:43550", + "issue:43575", + "issue:43606", + "issue:43643", + "issue:43650", + "issue:43701", + "issue:43742", + "issue:43749", + "issue:43761", + "issue:43825", + "issue:43827", + "issue:43873", + "issue:43931", + "issue:44077", + "issue:44188", + "issue:44190", + "issue:44442", + "issue:44464", + "issue:44479", + "issue:44496", + "issue:44617", + "issue:44855", + "issue:44871", + "issue:44913", + "issue:44987", + "issue:45137", + "issue:45216", + "issue:45290", + "issue:45632", + "issue:45698", + "issue:45759", + "issue:45803", + "issue:46178", + "issue:46202", + "issue:46365", + "issue:46421", + "issue:46424" + ], + "soft_pairs": [ + "issue:43232|issue:43388", + "issue:43761|issue:44188", + "issue:43931|issue:45216", + "issue:44077|issue:44479", + "issue:44987|issue:46178", + "issue:44464|issue:46421", + "issue:43493|issue:43827", + "issue:44442|issue:45803", + "issue:44479|issue:45290", + "issue:43873|issue:44871", + "issue:42907|issue:45698", + "issue:43749|issue:43931", + "issue:43643|issue:44913", + "issue:43316|issue:45759", + "issue:44855|issue:45137", + "issue:44190|issue:44496", + "issue:43650|issue:46202", + "issue:45632|issue:46424", + "issue:45803|issue:46365", + "issue:25251|issue:28282", + "issue:43295|issue:43825", + "issue:43550|issue:43701", + "issue:42617|issue:43575", + "issue:43606|issue:43701", + "issue:43065|issue:44617", + "issue:43575|issue:43742" + ], + "prepared_review_unit_hash": "28cac491a22c01e0e042c293f48473138f2105430c60a11e9db35ace58732b1a", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 34877, + "estimated_input_tokens": 8720, + "estimated_eval_tokens": 17696 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is heterogeneous: the soft pairs mostly connect unrelated issues that only share broad Transformers vocabulary (generation, loading, config, caching, offload, etc.). I found no true duplicate pair, so the safest interpretation is that this is not a real duplicate cluster.", + "confidence": 0.67, + "canonical_issue_reason": "Issue 45698 is the strongest standalone issue in this set: it is recent, active, open, and describes a concrete custom-module save/load regression with broad user impact. That said, the cluster itself is too mixed to indicate a true canonical duplicate.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45698 is the best overall issue candidate for triage because it is the most actionable and currently active among the listed issues; however, it should not be treated as representative of the whole cluster beyond that.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43232", + "right": "issue:43388", + "accept": false, + "reason": "Generation kwargs handling after sync_gpus and gather_for_metrics label truncation are different bugs in different code paths." + }, + { + "left": "issue:43761", + "right": "issue:44188", + "accept": false, + "reason": "CLIP hidden_states regression and compile-time attention-kernel divergence are unrelated model/runtime problems." + }, + { + "left": "issue:43931", + "right": "issue:45216", + "accept": false, + "reason": "A Qwen3-VL shape-mismatch load failure is not the same as a Qwen3.5 save_pretrained checkpoint regression." + }, + { + "left": "issue:44077", + "right": "issue:44479", + "accept": false, + "reason": "PatchTSMixer post_init validation and Qwen video-input regression are different model-level issues." + }, + { + "left": "issue:44987", + "right": "issue:46178", + "accept": false, + "reason": "Loading a specific model failing under transformers>=5.1 and a backend mapping error for tensorflow_text are unrelated." + }, + { + "left": "issue:44464", + "right": "issue:46421", + "accept": false, + "reason": "Chunked generation inconsistency with compiled forward and cache lazy-init recompilation are only loosely related by compilation." + }, + { + "left": "issue:43493", + "right": "issue:43827", + "accept": false, + "reason": "SigLIP2 implementation drift and v5 docs still referencing pipeline() are different product areas." + }, + { + "left": "issue:44442", + "right": "issue:45803", + "accept": false, + "reason": "AutoTokenizer failing on a custom tokenizer and a bare-except bug in Fuyu tensor conversion are distinct issues." + }, + { + "left": "issue:44479", + "right": "issue:45290", + "accept": false, + "reason": "A video-input regression and a chat-template crash on tool-call messages do not share the same root cause." + }, + { + "left": "issue:43873", + "right": "issue:44871", + "accept": false, + "reason": "Quantization/offloading behavior and Gemma eos_token_id config mismatch are unrelated." + }, + { + "left": "issue:42907", + "right": "issue:45698", + "accept": false, + "reason": "Saving dequantized model artifacts and loading the wrong custom module after save_pretrained are separate save/load failures." + }, + { + "left": "issue:43749", + "right": "issue:43931", + "accept": false, + "reason": "Broken FSDP CPU RAM efficient loading and a Qwen3-VL weight-shape mismatch are different loading paths and failures." + }, + { + "left": "issue:43643", + "right": "issue:44913", + "accept": false, + "reason": "Missing fields from trust_remote_code AutoConfig and rotary_pct not persisting on reload are different config serialization bugs." + }, + { + "left": "issue:43316", + "right": "issue:45759", + "accept": false, + "reason": "Gemma3TextConfig API discrepancy and AutoModelForCausalLM.from_config not unwrapping text_config are related only at a broad API level, not the same bug." + }, + { + "left": "issue:44855", + "right": "issue:45137", + "accept": false, + "reason": "A Python 3.13 import parsing error and a DeepSpeed ZeRO3 deque IndexError are completely unrelated." + }, + { + "left": "issue:44190", + "right": "issue:44496", + "accept": false, + "reason": "Local dataset loading in an example script and an unrecognized model/config error are different failures." + }, + { + "left": "issue:43650", + "right": "issue:46202", + "accept": false, + "reason": "A placeholder-data issue and a Gemma pipeline roundtrip failure are not duplicates." + }, + { + "left": "issue:45632", + "right": "issue:46424", + "accept": false, + "reason": "trust_remote_code cache-path collision and max_cache_length only applying to Quantized Cache are different cache-related behaviors." + }, + { + "left": "issue:45803", + "right": "issue:46365", + "accept": false, + "reason": "A bare except swallowing errors and an import conflict involving Trainer/datasets are unrelated." + }, + { + "left": "issue:25251", + "right": "issue:28282", + "accept": false, + "reason": "Pipeline top_k output nesting and an AutoModel ImportError due to missing PyTorch are not the same underlying problem." + }, + { + "left": "issue:43295", + "right": "issue:43825", + "accept": false, + "reason": "A processor/tokenizer regression and a pipeline v5 error-message issue are different regression symptoms." + }, + { + "left": "issue:43550", + "right": "issue:43701", + "accept": false, + "reason": "Bamba torch.compile SDPA failure and resume_from_checkpoint key mismatch share no concrete code path." + }, + { + "left": "issue:42617", + "right": "issue:43575", + "accept": false, + "reason": "3d_parallel.py runtime failure and Qwen2-57B tensor-parallel OOM are different large-model issues." + }, + { + "left": "issue:43606", + "right": "issue:43701", + "accept": false, + "reason": "CPU-offload device mismatch and checkpoint key mismatch are unrelated." + }, + { + "left": "issue:43065", + "right": "issue:44617", + "accept": false, + "reason": "Dummy Conv2d in Sam3PixelDecoder and Sam3Video CUDA OOM are not the same bug." + }, + { + "left": "issue:43575", + "right": "issue:43742", + "accept": false, + "reason": "Tensor-parallel OOM for Qwen2-57B and a MobileLLM load KeyError are different model-loading failures." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 32, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:16998", + "issue:22355", + "issue:27957", + "issue:32090", + "issue:41950", + "issue:43295", + "issue:43316", + "issue:43388", + "issue:43404", + "issue:43441", + "issue:43452", + "issue:43475", + "issue:43504", + "issue:43540", + "issue:43572", + "issue:43646", + "issue:43673", + "issue:43688", + "issue:43716", + "issue:43761", + "issue:44077", + "issue:44162", + "issue:44479", + "issue:44496", + "issue:44514", + "issue:44534", + "issue:44610", + "issue:44617", + "issue:44734", + "issue:44743", + "issue:44792", + "issue:44871", + "issue:44877", + "issue:44913", + "issue:45127", + "issue:45200", + "issue:45406", + "issue:45542", + "issue:45584", + "issue:45698", + "issue:45803", + "issue:46050", + "issue:46170", + "issue:46178", + "issue:46234", + "issue:46369", + "issue:46421" + ], + "soft_pairs": [ + "issue:16998|issue:22355", + "issue:43540|issue:44514", + "issue:27957|issue:32090", + "issue:43316|issue:43716", + "issue:44610|issue:46369", + "issue:43295|issue:44871", + "issue:44871|issue:45584", + "issue:43316|issue:44877", + "issue:44162|issue:44792", + "issue:43404|issue:43441", + "issue:44077|issue:44534", + "issue:45803|issue:46170", + "issue:44871|issue:46050", + "issue:41950|issue:43475", + "issue:43452|issue:44496", + "issue:43761|issue:44479", + "issue:43646|issue:44913", + "issue:43388|issue:43688", + "issue:45127|issue:45200", + "issue:43673|issue:44743", + "issue:43572|issue:46178", + "issue:44617|issue:45584", + "issue:43504|issue:45698", + "issue:43673|issue:46421", + "issue:44734|issue:45406", + "issue:45542|issue:46234" + ], + "prepared_review_unit_hash": "5df06c03705811b50b6e6ece28636d1d163983b827357881ebd8113345f94ea7", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 26, + "serialized_chars": 34984, + "estimated_input_tokens": 8746, + "estimated_eval_tokens": 17748 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a loose semantic grouping of Transformers regressions and model-specific bugs rather than true duplicates. The only safe canonical choice is a broad v5 regression issue; the soft pairs are generally too different in model, code path, or symptom to merge.", + "confidence": 0.47, + "canonical_issue_reason": "issue:43646 is the broadest representative: a general Transformers v5 breakage around custom model initialization, which matches the cluster\u2019s overall regression theme better than the narrower model-specific reports.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43646 is the best umbrella issue for this set because it is broad, version-level, and likely to attract related reports; however, the cluster is heterogeneous and not a strong duplicate set.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:16998", + "right": "issue:22355", + "accept": false, + "reason": "Unrelated topics: DeBERTa model_max_length behavior vs an import error for transformers.onnx." + }, + { + "left": "issue:43540", + "right": "issue:44514", + "accept": false, + "reason": "Both involve multimodal Qwen processing, but one is video-input processing in Qwen3OmniMoe and the other is batched chat template padding in Qwen2.5-VL; different code paths and bugs." + }, + { + "left": "issue:27957", + "right": "issue:32090", + "accept": false, + "reason": "No shared underlying bug: FlashAttention support for XLM-Roberta vs a Trainer broadcast TypeError." + }, + { + "left": "issue:43316", + "right": "issue:43716", + "accept": false, + "reason": "Different model families and failures: Gemma3TextConfig API mismatch vs Mistral-3 image preprocessor dtype mismatch." + }, + { + "left": "issue:44610", + "right": "issue:46369", + "accept": false, + "reason": "Both are model-specific vision issues, but one is an input-size mismatch in OmDet-Turbo and the other is OOM in sam3-litetext; not the same fix." + }, + { + "left": "issue:43295", + "right": "issue:44871", + "accept": false, + "reason": "Processor/tokenizer regression in one issue vs eos_token_id config inconsistency in another; only loosely related to processors/configs." + }, + { + "left": "issue:44871", + "right": "issue:45584", + "accept": false, + "reason": "Different bugs: Gemma-3 eos token configuration vs Whisper empty-transcription generation after special-token alignment." + }, + { + "left": "issue:43316", + "right": "issue:44877", + "accept": false, + "reason": "Both concern config loading, but Gemma3TextConfig API discrepancy and granite_speech strict-config loading failure are distinct issues." + }, + { + "left": "issue:44162", + "right": "issue:44792", + "accept": false, + "reason": "Completely different models and symptoms: ESM2 breakage vs a failed janus image-generation test." + }, + { + "left": "issue:43404", + "right": "issue:43441", + "accept": false, + "reason": "Same broad model family, but different concrete failures: untied lm_head weights vs FlashAttention failure in Ministral-3." + }, + { + "left": "issue:44077", + "right": "issue:44534", + "accept": false, + "reason": "Optional post_init misuse in patchtsmixer is unrelated to non-persistent buffer junk values." + }, + { + "left": "issue:45803", + "right": "issue:46170", + "accept": false, + "reason": "Bare except swallowing interrupts in Fuyu conversion is unrelated to a Windows/ROCm torch.distributed import/type error." + }, + { + "left": "issue:44871", + "right": "issue:46050", + "accept": false, + "reason": "Different save/load regressions: Gemma-3 eos token mismatch vs sinq quantized model breaking after save_pretrained/from_pretrained." + }, + { + "left": "issue:41950", + "right": "issue:43475", + "accept": false, + "reason": "Both are video-adjacent, but one is pipeline processor selection and the other is a missing attribute in Sam3 video encoder output." + }, + { + "left": "issue:43452", + "right": "issue:44496", + "accept": false, + "reason": "Both are from_pretrained/config loading problems, but one is gguf/custom class loading and the other is an unrecognized model_type in config.json." + }, + { + "left": "issue:43761", + "right": "issue:44479", + "accept": false, + "reason": "Different regressions: CLIPVisionModel hidden_states handling vs Qwen video-input regression across multiple model variants." + }, + { + "left": "issue:43646", + "right": "issue:44913", + "accept": false, + "reason": "General custom model initialization breakage vs a specific GPTNeoXConfig rotary_pct persistence bug; not the same underlying defect." + }, + { + "left": "issue:43388", + "right": "issue:43688", + "accept": false, + "reason": "Training metric gathering dropping tuple labels is unrelated to auxiliary-loss normalization." + }, + { + "left": "issue:45127", + "right": "issue:45200", + "accept": false, + "reason": "LoRA merge with extended vocab collapsing models is unrelated to Gemma 4 mm_token_type_ids defaults." + }, + { + "left": "issue:43673", + "right": "issue:44743", + "accept": false, + "reason": "Both mention cache/recurrent state, but one is missing cache in chunked_prefill and the other is recurrent states resetting in a specific modular Qwen3.5 file; different bugs." + }, + { + "left": "issue:43572", + "right": "issue:46178", + "accept": false, + "reason": "StableLmConfig pad_token_idx regression is unrelated to tensorflow_text backend mapping errors." + }, + { + "left": "issue:44617", + "right": "issue:45584", + "accept": false, + "reason": "CUDA OOM in Sam3Video is unrelated to Whisper empty-transcription generation behavior." + }, + { + "left": "issue:43504", + "right": "issue:45698", + "accept": false, + "reason": "Legacy field blocks Beit preset loading vs wrong custom module loaded after save_pretrained; both are loading issues but not the same code path." + }, + { + "left": "issue:43673", + "right": "issue:46421", + "accept": false, + "reason": "Both involve chunked prefill caching, but one is a missing cache in v5.0.0 and the other is lazy-init causing spurious recompilation; related area, different bug." + }, + { + "left": "issue:44734", + "right": "issue:45406", + "accept": false, + "reason": "KV-cache continuation tensor indexing crash in serve is unrelated to Gemma4Processor missing _tokenizer." + }, + { + "left": "issue:45542", + "right": "issue:46234", + "accept": false, + "reason": "TensorBoard/TensorFlow backend installation issue is unrelated to a missing import utility helper for falcon mamba ssm." + } + ] + }, + "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": 19, + "review_unit_index": 33, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:25251", + "issue:27957", + "issue:35141", + "issue:41720", + "issue:42490", + "issue:43065", + "issue:43525", + "issue:43531", + "issue:43540", + "issue:43550", + "issue:43644", + "issue:43761", + "issue:43827", + "issue:43931", + "issue:43976", + "issue:43994", + "issue:44060", + "issue:44077", + "issue:44190", + "issue:44242", + "issue:44462", + "issue:44464", + "issue:44479", + "issue:44521", + "issue:44534", + "issue:44610", + "issue:44617", + "issue:44734", + "issue:44792", + "issue:44855", + "issue:44877", + "issue:44913", + "issue:44933", + "issue:44964", + "issue:44977", + "issue:45127", + "issue:45245", + "issue:45357", + "issue:45362", + "issue:45381", + "issue:45563", + "issue:45584", + "issue:45588", + "issue:45600", + "issue:45632", + "issue:45907", + "issue:46234", + "issue:46369" + ], + "soft_pairs": [ + "issue:43525|issue:43531", + "issue:43761|issue:44242", + "issue:45381|issue:46369", + "issue:44792|issue:44977", + "issue:45563|issue:45584", + "issue:43931|issue:44060", + "issue:44792|issue:45600", + "issue:44534|issue:44913", + "issue:44190|issue:44964", + "issue:44479|issue:44734", + "issue:35141|issue:42490", + "issue:44855|issue:45588", + "issue:44617|issue:46369", + "issue:43827|issue:43994", + "issue:41720|issue:45245", + "issue:43065|issue:44610", + "issue:44521|issue:45381", + "issue:44464|issue:45563", + "issue:44933|issue:45362", + "issue:43644|issue:44913", + "issue:43976|issue:44855", + "issue:43550|issue:44190", + "issue:44077|issue:44877", + "issue:43540|issue:44734", + "issue:45127|issue:45357", + "issue:44462|issue:45907", + "issue:25251|issue:27957", + "issue:45632|issue:46234" + ], + "prepared_review_unit_hash": "805253767e368899cb6e8d0a4ce2db46a9761e97a7d4eef18d4aba13ee2aee88", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 28, + "serialized_chars": 35893, + "estimated_input_tokens": 8974, + "estimated_eval_tokens": 18204 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The items are mostly unrelated Transformers bug reports rather than one duplicate cluster. None of the soft pairs describe the same underlying bug or change closely enough to merge; they usually differ by model, code path, or bug type.", + "confidence": 0.93, + "canonical_issue_reason": "Issue 44521 is the best representative issue in this set: it is open, has the strongest cross-references, and describes a concrete multimodal regression with clear downstream impact.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 44521 stands out as the strongest standalone bug report in the group, with multiple inbound references and active discussion, making it the most useful anchor issue here.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43525", + "right": "issue:43531", + "accept": false, + "reason": "Different models and different failure modes: missing pad_token_id on Llama4Config vs sliding_window behavior in Qwen3-MoE." + }, + { + "left": "issue:43761", + "right": "issue:44242", + "accept": false, + "reason": "Unrelated regressions: CLIPVision hidden_states handling vs MoE load-balancing loss when router logits are disabled." + }, + { + "left": "issue:45381", + "right": "issue:46369", + "accept": false, + "reason": "Both are model-specific multimodal bugs, but one is a qwen2.5-vl video position-id regression and the other is Sam3-litetext memory overflow." + }, + { + "left": "issue:44792", + "right": "issue:44977", + "accept": false, + "reason": "Different code paths and symptoms: Janus image-generation test failure vs Qwen3.5 flash-attention generation issue." + }, + { + "left": "issue:45563", + "right": "issue:45584", + "accept": false, + "reason": "A stale warning about num_return_sequences is not the same as Whisper failing on empty transcription after token alignment." + }, + { + "left": "issue:43931", + "right": "issue:44060", + "accept": false, + "reason": "One is a weight-shape mismatch on load, the other is an incorrect tied-weights warning involving a different Qwen3-Next parameter." + }, + { + "left": "issue:44792", + "right": "issue:45600", + "accept": false, + "reason": "A failing Janus generation test and a broken auto_mappings reference to removed Sam3LiteText configs are distinct issues." + }, + { + "left": "issue:44534", + "right": "issue:44913", + "accept": false, + "reason": "Generic non-persistent buffer initialization bug vs a GPTNeoX config field not persisting on reload." + }, + { + "left": "issue:44190", + "right": "issue:44964", + "accept": false, + "reason": "Local dataset loading in a training script is unrelated to loading a specific multimodal model with latest transformers." + }, + { + "left": "issue:44479", + "right": "issue:44734", + "accept": false, + "reason": "Video-input regression in Qwen multimodal models vs KV-cache continuation crash in transformers serve are different failures." + }, + { + "left": "issue:35141", + "right": "issue:42490", + "accept": false, + "reason": "Embedding reinitialization during post_init is unrelated to the import warning caused by a trailing slash in a local model path." + }, + { + "left": "issue:44855", + "right": "issue:45588", + "accept": false, + "reason": "Python 3.13 import parsing failure from torch.jit.script formatting is unrelated to flash_attention.py crashing on s_aux=None." + }, + { + "left": "issue:44617", + "right": "issue:46369", + "accept": false, + "reason": "Both mention memory pressure, but they affect different models and likely different root causes: Sam3Video OOM vs Sam3-litetext memory overflow." + }, + { + "left": "issue:43827", + "right": "issue:43994", + "accept": false, + "reason": "Docs still referencing pipeline() after removals is not the same as a SigLIP2 pipeline producing nonsensical outputs." + }, + { + "left": "issue:41720", + "right": "issue:45245", + "accept": false, + "reason": "CUDA assert during Qwen3 auto device mapping is unrelated to the 2^24 category limit error." + }, + { + "left": "issue:43065", + "right": "issue:44610", + "accept": false, + "reason": "A dummy Conv2d in Sam3PixelDecoder is a model-internal placeholder issue, not the OmDet-Turbo processor/model input-size mismatch." + }, + { + "left": "issue:44521", + "right": "issue:45381", + "accept": false, + "reason": "Both are multimodal, but one is all-zero assistant masks in apply_chat_template and the other is incorrect vision_position_ids in qwen2.5-vl video input processing." + }, + { + "left": "issue:44464", + "right": "issue:45563", + "accept": false, + "reason": "Compiled-forward chunked generation inconsistency is a different bug from a stale generation warning." + }, + { + "left": "issue:44933", + "right": "issue:45362", + "accept": false, + "reason": "A missing import in image_utils is unrelated to Qwen3.5-35B crashing in chat generation." + }, + { + "left": "issue:43644", + "right": "issue:44913", + "accept": false, + "reason": "A broad v5 buffer initialization regression is not the same as GPTNeoX rotary_pct being lost on config reload." + }, + { + "left": "issue:43976", + "right": "issue:44855", + "accept": false, + "reason": "General Python version incompatibility does not match a specific Python 3.13 IndentationError caused by a decorator/comment parsing edge case." + }, + { + "left": "issue:43550", + "right": "issue:44190", + "accept": false, + "reason": "torch.compile/SDPA failure in Bamba-9B-v2 is unrelated to local dataset loading in an image-classification training script." + }, + { + "left": "issue:44077", + "right": "issue:44877", + "accept": false, + "reason": "Optional post_init in patchtsmixer and strict config blocking granite_speech are different model/config validation issues." + }, + { + "left": "issue:43540", + "right": "issue:44734", + "accept": false, + "reason": "Processing video inputs in Qwen3OmniMoe is a different problem from server-side KV-cache continuation indexing." + }, + { + "left": "issue:45127", + "right": "issue:45357", + "accept": false, + "reason": "LoRA merge collapse with extended vocab and incorrect visual encoder keys in save_pretrained are separate serialization/training bugs." + }, + { + "left": "issue:44462", + "right": "issue:45907", + "accept": false, + "reason": "AutoTokenizer ignoring tokenizer.json is unrelated to list_repo_files failing to forward revision in CI." + }, + { + "left": "issue:25251", + "right": "issue:27957", + "accept": false, + "reason": "Pipeline top_k nesting behavior and XLMRoberta Flash Attention 2 support are unrelated." + }, + { + "left": "issue:45632", + "right": "issue:46234", + "accept": false, + "reason": "A cache-path collision for trust_remote_code is unrelated to a missing import-utils helper function." + } + ] + }, + "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": 19, + "review_unit_index": 34, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:27957", + "issue:35141", + "issue:36246", + "issue:38617", + "issue:41628", + "issue:41950", + "issue:42175", + "issue:43116", + "issue:43295", + "issue:43335", + "issue:43404", + "issue:43479", + "issue:43643", + "issue:43650", + "issue:43673", + "issue:43854", + "issue:43976", + "issue:44016", + "issue:44112", + "issue:44186", + "issue:44190", + "issue:44220", + "issue:44246", + "issue:44279", + "issue:44496", + "issue:44683", + "issue:44734", + "issue:44737", + "issue:44938", + "issue:44964", + "issue:44987", + "issue:45003", + "issue:45216", + "issue:45245", + "issue:45276", + "issue:45375", + "issue:45446", + "issue:45542", + "issue:45584", + "issue:45600", + "issue:45632", + "issue:45803", + "issue:46082", + "issue:46121", + "issue:46123", + "issue:46369", + "issue:46421", + "issue:46424" + ], + "soft_pairs": [ + "issue:44112|issue:45600", + "issue:43335|issue:44190", + "issue:44737|issue:45803", + "issue:43116|issue:44190", + "issue:45276|issue:45584", + "issue:43295|issue:45276", + "issue:44938|issue:45542", + "issue:46082|issue:46121", + "issue:45632|issue:46421", + "issue:45245|issue:45584", + "issue:36246|issue:41628", + "issue:42175|issue:44279", + "issue:43404|issue:45216", + "issue:43650|issue:44220", + "issue:44246|issue:45003", + "issue:36246|issue:38617", + "issue:41950|issue:44016", + "issue:44186|issue:44734", + "issue:43673|issue:46424", + "issue:43479|issue:45375", + "issue:44683|issue:45446", + "issue:43643|issue:44496", + "issue:44964|issue:45600", + "issue:43854|issue:44964", + "issue:44016|issue:44220", + "issue:44112|issue:44964", + "issue:46123|issue:46369", + "issue:27957|issue:35141", + "issue:43976|issue:44987" + ], + "prepared_review_unit_hash": "03a358bc17d297e717c933020a0bac8d2adf2eacf61b079c5f96688abbc1ccd0", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36512, + "estimated_input_tokens": 9128, + "estimated_eval_tokens": 18512 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is mostly heterogeneous: a few weakly similar import/cache/config issues, but no pair looks like the same underlying bug or change. I\u2019d use the resize-token-embeddings issue as the best representative of the only clearly coherent bug family here.", + "confidence": 0.9, + "canonical_issue_reason": "issue:35141 is the clearest, most general statement of a concrete bug in the resize_token_embeddings/post_init path; it best represents the only strong bug family in this set.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:35141 is the most broadly applicable and well-scoped issue among the candidates, with a concrete reproducible failure and a model-agnostic fix path.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44112", + "right": "issue:45600", + "accept": false, + "reason": "Different failures: GraniteSpeech CI device-override test vs removed Sam3LiteText auto-mapping references." + }, + { + "left": "issue:43335", + "right": "issue:44190", + "accept": false, + "reason": "Unrelated code paths: SwitchTransformers sparse-layer config vs image-classification script local dataset loading." + }, + { + "left": "issue:44737", + "right": "issue:45803", + "accept": false, + "reason": "Different bugs: XLNet CPU tensor allocation vs FuyuBatchFeature bare except swallowing errors." + }, + { + "left": "issue:43116", + "right": "issue:44190", + "accept": false, + "reason": "Both touch example/pipeline scripts, but one is multilabel classification output handling and the other is local dataset loading for image classification." + }, + { + "left": "issue:45276", + "right": "issue:45584", + "accept": false, + "reason": "Unrelated model behaviors: Gemma4 embedding resize vs Whisper empty-transcription generation." + }, + { + "left": "issue:43295", + "right": "issue:45276", + "accept": false, + "reason": "Different regressions in different paths: processor/tokenizer image handling vs resize_token_embeddings output-embedding updates." + }, + { + "left": "issue:44938", + "right": "issue:45542", + "accept": false, + "reason": "Both are environment/import issues, but one is Python 3.14 load failure and the other is missing TensorFlow backend wiring." + }, + { + "left": "issue:46082", + "right": "issue:46121", + "accept": false, + "reason": "Distinct config-validation bugs: Llama head_dim constraints vs rope-validation list typing." + }, + { + "left": "issue:45632", + "right": "issue:46421", + "accept": false, + "reason": "Both mention cache, but one is trust_remote_code cache-path collision and the other is chunked-prefill recompilation behavior." + }, + { + "left": "issue:45245", + "right": "issue:45584", + "accept": false, + "reason": "Completely different symptoms and subsystems: category-limit runtime error vs Whisper empty transcription handling." + }, + { + "left": "issue:36246", + "right": "issue:41628", + "accept": false, + "reason": "Both are import errors, but for different missing symbols and different modules." + }, + { + "left": "issue:42175", + "right": "issue:44279", + "accept": false, + "reason": "Dependency/backend packaging issue vs generic dependency issue; too broad and not the same defect." + }, + { + "left": "issue:43404", + "right": "issue:45216", + "accept": false, + "reason": "Different model-saving/training bugs: tied lm_head in Mistral3 vs Qwen3.5 checkpoint serialization." + }, + { + "left": "issue:43650", + "right": "issue:44220", + "accept": false, + "reason": "Malformed issue vs audio feature extraction bug; no shared underlying defect." + }, + { + "left": "issue:44246", + "right": "issue:45003", + "accept": false, + "reason": "Import-time slowness vs unsafe sys.modules access are different problems in different code paths." + }, + { + "left": "issue:36246", + "right": "issue:38617", + "accept": false, + "reason": "Both are missing-export import errors, but they concern different symbols and modules." + }, + { + "left": "issue:41950", + "right": "issue:44016", + "accept": false, + "reason": "Video pipeline processor-selection bug vs notebook syntax error; unrelated." + }, + { + "left": "issue:44186", + "right": "issue:44734", + "accept": false, + "reason": "Tokenizer padding/NER crash vs serve KV-cache continuation indexing bug; different subsystems." + }, + { + "left": "issue:43673", + "right": "issue:46424", + "accept": false, + "reason": "Both cache-related, but one is missing GenerationMixin cache in chunked prefill and the other is quantized-cache max length handling." + }, + { + "left": "issue:43479", + "right": "issue:45375", + "accept": false, + "reason": "Both are config-field issues, but they affect different models and different fields/serialization behavior." + }, + { + "left": "issue:44683", + "right": "issue:45446", + "accept": false, + "reason": "Related to flex_attention, but one is a compile failure on torch>=2.9 and the other is a version-check import guard bug." + }, + { + "left": "issue:43643", + "right": "issue:44496", + "accept": false, + "reason": "Different auto-config problems: missing fields with trust_remote_code vs unrecognized model missing model_type." + }, + { + "left": "issue:44964", + "right": "issue:45600", + "accept": false, + "reason": "Distinct model-loading failures: Phi-4-multimodal regression vs stale Sam3LiteText auto-mapping references." + }, + { + "left": "issue:43854", + "right": "issue:44964", + "accept": false, + "reason": "Different models and failure modes, both load issues but not the same bug." + }, + { + "left": "issue:44016", + "right": "issue:44220", + "accept": false, + "reason": "Notebook syntax error vs audio feature extraction bug; unrelated." + }, + { + "left": "issue:44112", + "right": "issue:44964", + "accept": false, + "reason": "CI test failure vs Phi-4-multimodal model loading regression; unrelated." + }, + { + "left": "issue:46123", + "right": "issue:46369", + "accept": false, + "reason": "MaskGenerationPipeline partial-batch bug vs sam3-litetext memory overflow; different symptoms and code paths." + }, + { + "left": "issue:27957", + "right": "issue:35141", + "accept": false, + "reason": "Different areas: Flash Attention support for XLM-Roberta vs resize-token-embeddings reinitialization." + }, + { + "left": "issue:43976", + "right": "issue:44987", + "accept": false, + "reason": "Python-version compatibility bug vs specific model load failure on 5.1.0; not the same underlying issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 35, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:28282", + "issue:35141", + "issue:38175", + "issue:42907", + "issue:43295", + "issue:43299", + "issue:43381", + "issue:43441", + "issue:43526", + "issue:43531", + "issue:43577", + "issue:43650", + "issue:43673", + "issue:43824", + "issue:43901", + "issue:43976", + "issue:44016", + "issue:44062", + "issue:44075", + "issue:44242", + "issue:44279", + "issue:44336", + "issue:44368", + "issue:44496", + "issue:44521", + "issue:44617", + "issue:44792", + "issue:44933", + "issue:45042", + "issue:45103", + "issue:45200", + "issue:45216", + "issue:45325", + "issue:45335", + "issue:45357", + "issue:45372", + "issue:45405", + "issue:45584", + "issue:45588", + "issue:45593", + "issue:45600", + "issue:45759", + "issue:46082", + "issue:46209", + "issue:46234", + "issue:46399", + "issue:46424" + ], + "soft_pairs": [ + "issue:43531|issue:45600", + "issue:42907|issue:43441", + "issue:43526|issue:43901", + "issue:43295|issue:45335", + "issue:44792|issue:46424", + "issue:28282|issue:35141", + "issue:44496|issue:45042", + "issue:46082|issue:46399", + "issue:43441|issue:45588", + "issue:45372|issue:46234", + "issue:44016|issue:44062", + "issue:43299|issue:45216", + "issue:43901|issue:44496", + "issue:43824|issue:44933", + "issue:43673|issue:44336", + "issue:43381|issue:45593", + "issue:43976|issue:44279", + "issue:44075|issue:44368", + "issue:44496|issue:45759", + "issue:45335|issue:45584", + "issue:43650|issue:46209", + "issue:43577|issue:45216", + "issue:45103|issue:45372", + "issue:44242|issue:45200", + "issue:38175|issue:44617", + "issue:44521|issue:45325", + "issue:44279|issue:45405", + "issue:43299|issue:45357" + ], + "prepared_review_unit_hash": "c41aab92e72b94ff3a60708eb49be9d7bc1c8bb43e2835b9fa4c031fc5e1d436", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 35662, + "estimated_input_tokens": 8916, + "estimated_eval_tokens": 18088 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "These items are mostly unrelated bug reports across different models, loaders, and utilities. None of the soft pairs look like the same underlying bug/change, so I rejected them all. No PRs are present in this cluster.", + "confidence": 0.93, + "canonical_issue_reason": "issue:44521 is the most active/open item in the set and has the strongest external signal (multiple inbound references and discussion), so it is the best representative issue for the cluster even though the cluster is not a true duplicate set.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:44521 is the best standalone issue to keep as the cluster representative because it is current, well-scoped, and has the most linkage/discussion among the listed issues.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43531", + "right": "issue:45600", + "accept": false, + "reason": "Different bugs: Qwen3-MoE sliding window behavior vs stale auto_mappings references. Same area of model support, but not the same code path." + }, + { + "left": "issue:42907", + "right": "issue:43441", + "accept": false, + "reason": "Both mention Ministral, but one is about saving dequantized weights and the other is a FlashAttention runtime failure. Not the same defect." + }, + { + "left": "issue:43526", + "right": "issue:43901", + "accept": false, + "reason": "BEIT processor label reduction bug vs docs mismatch for TextClassificationPipeline. Different subsystem and failure mode." + }, + { + "left": "issue:43295", + "right": "issue:45335", + "accept": false, + "reason": "Both involve embeddings/tokenizer-related behavior, but one is a processor/tokenizer regression and the other is t5gemma decoder embedding resize. Not one underlying bug." + }, + { + "left": "issue:44792", + "right": "issue:46424", + "accept": false, + "reason": "Janus image-generation test failure vs cache max length handling for quantized cache. Unrelated behaviors." + }, + { + "left": "issue:28282", + "right": "issue:35141", + "accept": false, + "reason": "AutoModel PyTorch import error vs post_init reinitializing output embeddings after resize. Different issue class and code path." + }, + { + "left": "issue:44496", + "right": "issue:45042", + "accept": false, + "reason": "Olmo model config import failure vs PIL backend processors requiring torchvision. Separate problems." + }, + { + "left": "issue:46082", + "right": "issue:46399", + "accept": false, + "reason": "LlamaConfig head_dim validation vs Qwen3.5 rotary kernel registration error. Different models and logic." + }, + { + "left": "issue:43441", + "right": "issue:45588", + "accept": false, + "reason": "Both touch FlashAttention, but one is a Ministral-specific failure and the other is an AttributeError for sink-less models in integration code. Not the same bug." + }, + { + "left": "issue:45372", + "right": "issue:46234", + "accept": false, + "reason": "Gemma 4 processor import breakage vs missing falcon_mamba availability helper. Unrelated imports." + }, + { + "left": "issue:44016", + "right": "issue:44062", + "accept": false, + "reason": "Notebook syntax error vs AddedToken keyword collision. No shared underlying code path." + }, + { + "left": "issue:43299", + "right": "issue:45216", + "accept": false, + "reason": "Qwen3.5 loading failure vs incorrect save_pretrained checkpoint contents. Same family, but different operations and bugs." + }, + { + "left": "issue:43901", + "right": "issue:44496", + "accept": false, + "reason": "Pipeline docs/behavior mismatch vs model config import error. Different issue type and scope." + }, + { + "left": "issue:43824", + "right": "issue:44933", + "accept": false, + "reason": "Two import errors, but for different symbols/modules and different root causes." + }, + { + "left": "issue:43673", + "right": "issue:44336", + "accept": false, + "reason": "Generation cache regression vs ANSI color codes in loading_report. Completely unrelated." + }, + { + "left": "issue:43381", + "right": "issue:45593", + "accept": false, + "reason": "Gradient checkpointing eval-mode behavior vs D-FINE auxiliary loss omission. Different training semantics." + }, + { + "left": "issue:43976", + "right": "issue:44279", + "accept": false, + "reason": "Python version compatibility issue vs generic dependency problem. Too broad and not the same bug." + }, + { + "left": "issue:44075", + "right": "issue:44368", + "accept": false, + "reason": "SGD optimizer args ignored vs tie_word_embeddings warning in Qwen fine-tuning. Different feature areas." + }, + { + "left": "issue:44496", + "right": "issue:45759", + "accept": false, + "reason": "Unrecognized model config vs AutoModelForCausalLM not unwrapping text_config for composite Qwen multimodal configs. Related to loading, but not the same defect." + }, + { + "left": "issue:45335", + "right": "issue:45584", + "accept": false, + "reason": "t5gemma embedding resize issue vs Whisper empty-transcription generation failure. Unrelated models and code paths." + }, + { + "left": "issue:43650", + "right": "issue:46209", + "accept": false, + "reason": "Placeholder issue title vs FP8 inference failure on B200. No overlap in bug report content." + }, + { + "left": "issue:43577", + "right": "issue:45216", + "accept": false, + "reason": "BLIP2 dtype propagation bug vs Qwen3.5 save_pretrained regression. Different models and operations." + }, + { + "left": "issue:45103", + "right": "issue:45372", + "accept": false, + "reason": "auto_docstring crash with future annotations vs Gemma 4 processor import failure. Different tooling and root cause." + }, + { + "left": "issue:44242", + "right": "issue:45200", + "accept": false, + "reason": "MoE load-balancing loss omission vs Gemma 4 mm_token_type_ids defaulting. Different model behavior." + }, + { + "left": "issue:38175", + "right": "issue:44617", + "accept": false, + "reason": "SigLIP2 zero probabilities vs Sam3Video CUDA OOM. Not the same bug or subsystem." + }, + { + "left": "issue:44521", + "right": "issue:45325", + "accept": false, + "reason": "Both multimodal, but assistant-mask generation and Qwen2.5-VL rope-position scaling are distinct code paths and symptoms." + }, + { + "left": "issue:44279", + "right": "issue:45405", + "accept": false, + "reason": "Generic transformer dependency issue vs unreleased MIN_PEFT_VERSION bump. Different release/dependency concerns." + }, + { + "left": "issue:43299", + "right": "issue:45357", + "accept": false, + "reason": "Both Qwen3.5 regressions, but loading breaks vs save_pretrained stores wrong visual encoder keys. Not the same concrete bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 36, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:33290", + "issue:39692", + "issue:41950", + "issue:42994", + "issue:43295", + "issue:43299", + "issue:43317", + "issue:43441", + "issue:43450", + "issue:43479", + "issue:43525", + "issue:43650", + "issue:43653", + "issue:43854", + "issue:43901", + "issue:43976", + "issue:44016", + "issue:44261", + "issue:44263", + "issue:44279", + "issue:44291", + "issue:44297", + "issue:44351", + "issue:44361", + "issue:44462", + "issue:44466", + "issue:44623", + "issue:44671", + "issue:44683", + "issue:44749", + "issue:44779", + "issue:45020", + "issue:45081", + "issue:45276", + "issue:45278", + "issue:45397", + "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:41950|issue:45020", + "issue:44351|issue:45278", + "issue:44779|issue:45812", + "issue:43295|issue:44779", + "issue:39692|issue:43901", + "issue:41950|issue:45823", + "issue:43479|issue:43525", + "issue:44683|issue:45542", + "issue:43976|issue:45405", + "issue:44462|issue:45081", + "issue:43299|issue:43854", + "issue:44297|issue:44462", + "issue:43450|issue:44279", + "issue:43650|issue:46132", + "issue:45081|issue:45907", + "issue:44466|issue:45276", + "issue:44016|issue:44291", + "issue:43441|issue:45834", + "issue:42994|issue:43317", + "issue:43653|issue:45593", + "issue:44623|issue:45397", + "issue:44623|issue:44749", + "issue:45600|issue:45712", + "issue:33290|issue:45656", + "issue:44261|issue:44263", + "issue:44361|issue:45593", + "issue:44671|issue:44779", + "issue:44016|issue:45823" + ], + "prepared_review_unit_hash": "d3b7d7781452c321599ef6d5c57ebe4a5b6e1b237a0f182defe7bf1e2a607c62", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 35380, + "estimated_input_tokens": 8845, + "estimated_eval_tokens": 17946 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is heterogeneous: it mixes unrelated Transformers regressions across tokenizers, processors, configs, CI, docs, quantization, and deepspeed. None of the soft pairs look like true duplicates under a strict bug-level definition.", + "confidence": 0.91, + "canonical_issue_reason": "No clear canonical duplicate target; the items are mostly distinct regressions with only broad topical overlap.", + "canonical_pr_reason": null, + "best_issue_reason": "If one issue must anchor the cluster, issue 45278 is the broadest umbrella-style regression report about imports breaking after an upgrade, but it still does not actually subsume the rest.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:41950", + "right": "issue:45020", + "accept": false, + "reason": "Different failures: a video-classification processor lookup bug vs a broad remote_code regression." + }, + { + "left": "issue:44351", + "right": "issue:45278", + "accept": false, + "reason": "One is a specific missing HybridCache import; the other is a generic report of many import errors after upgrade." + }, + { + "left": "issue:44779", + "right": "issue:45812", + "accept": false, + "reason": "Both are tokenizer regressions, but on different models and different concrete symptoms." + }, + { + "left": "issue:43295", + "right": "issue:44779", + "accept": false, + "reason": "Processor.tokenizer/custom image handling regression is not the same bug as Deepseek tokenizer misbehavior." + }, + { + "left": "issue:39692", + "right": "issue:43901", + "accept": false, + "reason": "Both are documentation-related, but one is a SigLIP2 example error and the other is stale pipeline docs for return_all_scores." + }, + { + "left": "issue:41950", + "right": "issue:45823", + "accept": false, + "reason": "Video-classification image-processor lookup and Gemma4 device-mismatch are unrelated bugs." + }, + { + "left": "issue:43479", + "right": "issue:43525", + "accept": false, + "reason": "Different config bugs: default multimodal sub-config initialization vs missing pad_token_id on Llama4Config." + }, + { + "left": "issue:44683", + "right": "issue:45542", + "accept": false, + "reason": "Unrelated code paths: flex_attention compilation vs tensorboard/TensorFlow backend setup." + }, + { + "left": "issue:43976", + "right": "issue:45405", + "accept": false, + "reason": "Python version compatibility failure is not the same as the PEFT minimum-version bump issue." + }, + { + "left": "issue:44462", + "right": "issue:45081", + "accept": false, + "reason": "Both involve tokenizer loading, but one ignores tokenizer.json while the other crashes in a Mistral regex patch." + }, + { + "left": "issue:43299", + "right": "issue:43854", + "accept": false, + "reason": "Different model-loading failures for different models and different underlying causes." + }, + { + "left": "issue:44297", + "right": "issue:44462", + "accept": false, + "reason": "Related serialization/loading area, but tokenizer_class mismatch is not the same concrete bug as ignoring tokenizer.json." + }, + { + "left": "issue:43450", + "right": "issue:44279", + "accept": false, + "reason": "Video processor shape bug and a dependency issue with transformers are not the same defect." + }, + { + "left": "issue:43650", + "right": "issue:46132", + "accept": false, + "reason": "Completely unrelated issues: placeholder text vs AttentionInterface.register behavior change." + }, + { + "left": "issue:45081", + "right": "issue:45907", + "accept": false, + "reason": "Tokenizer regex crash and list_repo_files revision forwarding are unrelated." + }, + { + "left": "issue:44466", + "right": "issue:45276", + "accept": false, + "reason": "lm_head.weight serialization inconsistency is different from Gemma4 resize_token_embeddings behavior." + }, + { + "left": "issue:44016", + "right": "issue:44291", + "accept": false, + "reason": "Notebook syntax error and init_empty_weights argument handling are unrelated." + }, + { + "left": "issue:43441", + "right": "issue:45834", + "accept": false, + "reason": "FlashAttention failure in Ministral-3 and long-ocr-input index error in Kosmos2.5 are different bugs." + }, + { + "left": "issue:42994", + "right": "issue:43317", + "accept": false, + "reason": "Both touch quantization/offload, but one is model saving and the other is loading dequantized models with device_map=auto." + }, + { + "left": "issue:43653", + "right": "issue:45593", + "accept": false, + "reason": "BigBirdTokenizer special-token registration and D-FINE auxiliary-loss behavior are unrelated." + }, + { + "left": "issue:44623", + "right": "issue:45397", + "accept": false, + "reason": "processor.save_pretrained missing files is not the same bug as gemma-4 zero3 from_pretrained failure." + }, + { + "left": "issue:44623", + "right": "issue:44749", + "accept": false, + "reason": "A save_pretrained file issue is unrelated to a post-upgrade data filtering slowdown report." + }, + { + "left": "issue:45600", + "right": "issue:45712", + "accept": false, + "reason": "Both are cleanup/CI issues, but they affect different files and different failure modes." + }, + { + "left": "issue:33290", + "right": "issue:45656", + "accept": false, + "reason": "Deepspeed OOM with Adafactor and optimizer.step being called twice are related only at a high level, not the same concrete bug." + }, + { + "left": "issue:44261", + "right": "issue:44263", + "accept": false, + "reason": "A precision/config issue in MLA is unrelated to torch.split return values in GlmMoeDsaIndexer." + }, + { + "left": "issue:44361", + "right": "issue:45593", + "accept": false, + "reason": "MLukeTokenizer AttributeError and D-FINE auxiliary-loss handling are unrelated." + }, + { + "left": "issue:44671", + "right": "issue:44779", + "accept": false, + "reason": "Different model-specific tokenizer/masked-LM regressions with no shared concrete code path." + }, + { + "left": "issue:44016", + "right": "issue:45823", + "accept": false, + "reason": "Notebook syntax error and Gemma4 device mismatch are unrelated." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 37, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:36246", + "issue:41628", + "issue:41950", + "issue:42175", + "issue:42491", + "issue:42757", + "issue:42886", + "issue:43262", + "issue:43299", + "issue:43441", + "issue:43525", + "issue:43526", + "issue:43531", + "issue:43643", + "issue:43673", + "issue:43723", + "issue:43761", + "issue:43784", + "issue:43824", + "issue:43881", + "issue:43901", + "issue:43976", + "issue:43994", + "issue:44162", + "issue:44188", + "issue:44220", + "issue:44242", + "issue:44246", + "issue:44295", + "issue:44297", + "issue:44336", + "issue:44373", + "issue:44496", + "issue:44521", + "issue:44589", + "issue:44683", + "issue:44871", + "issue:45003", + "issue:45103", + "issue:45362", + "issue:45397", + "issue:45542", + "issue:45561", + "issue:45563", + "issue:45593", + "issue:46132", + "issue:46399" + ], + "soft_pairs": [ + "issue:43723|issue:44162", + "issue:42175|issue:43976", + "issue:43441|issue:44683", + "issue:43901|issue:43994", + "issue:43262|issue:44521", + "issue:43673|issue:43761", + "issue:44297|issue:45397", + "issue:43824|issue:43976", + "issue:43525|issue:43881", + "issue:43901|issue:44246", + "issue:43525|issue:43643", + "issue:44589|issue:45362", + "issue:41628|issue:43976", + "issue:43784|issue:45542", + "issue:44188|issue:44336", + "issue:45103|issue:46399", + "issue:45561|issue:45563", + "issue:44246|issue:44496", + "issue:44336|issue:44373", + "issue:44295|issue:45397", + "issue:44242|issue:45593", + "issue:42491|issue:43299", + "issue:43531|issue:45362", + "issue:43526|issue:45003", + "issue:41950|issue:44220", + "issue:44246|issue:44871", + "issue:42886|issue:46132", + "issue:36246|issue:42757" + ], + "prepared_review_unit_hash": "55689a5cf6bb13384d540b27b661672356698774438dce0d070242fd509f7c94", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 28, + "serialized_chars": 35454, + "estimated_input_tokens": 8864, + "estimated_eval_tokens": 17984 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a broad cluster of mostly unrelated Transformers v5-era regression reports, import failures, and model-specific loading bugs. None of the soft pairs look like the same concrete underlying bug/change under a strict duplicate standard, so all are rejected. The strongest representative issue is the Qwen3 MoE/v5 loading regression report.", + "confidence": 0.33, + "canonical_issue_reason": "issue:43299 is the clearest, most concrete representative of the cluster\u2019s recurring v5 regression theme: a version-bounded model-loading failure with a detailed Qwen3 MoE repro and explicit hf4.x vs hf5.x breakage.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43299 best captures the dominant signal in this cluster: a specific, actionable Transformers v5 loading regression affecting Qwen3 MoE models, and it is closer to the cluster\u2019s main recurring pattern than the other, more isolated import/pipeline bugs.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43723", + "right": "issue:44162", + "accept": false, + "reason": "Both are load-time failures, but one is a generic AutoTokenizer regression and the other is an ESM2 model-specific breakage; different code paths." + }, + { + "left": "issue:42175", + "right": "issue:43976", + "accept": false, + "reason": "Packaging extras vs Python-version compatibility; unrelated issues." + }, + { + "left": "issue:43441", + "right": "issue:44683", + "accept": false, + "reason": "Both touch attention kernels, but one is Ministral FlashAttention in Transformers RC and the other is compiled flex_attention on torch>=2.9; not the same bug." + }, + { + "left": "issue:43901", + "right": "issue:43994", + "accept": false, + "reason": "Docs mismatch vs incorrect model outputs; different underlying problems." + }, + { + "left": "issue:43262", + "right": "issue:44521", + "accept": false, + "reason": "Different apply_chat_template bugs: audio sampling-rate default vs multimodal assistant-mask generation." + }, + { + "left": "issue:43673", + "right": "issue:43761", + "accept": false, + "reason": "Cache/chunked-prefill regression vs CLIP hidden_states regression; unrelated." + }, + { + "left": "issue:44297", + "right": "issue:45397", + "accept": false, + "reason": "Tokenizer save metadata mismatch vs DeepSpeed/from_pretrained loading failure; different paths." + }, + { + "left": "issue:43824", + "right": "issue:43976", + "accept": false, + "reason": "Specific Qwen2.5-VL import error vs Python 3.9+ support issue; unrelated." + }, + { + "left": "issue:43525", + "right": "issue:43881", + "accept": false, + "reason": "Both are model-load failures, but they affect different model families and have different missing-field symptoms." + }, + { + "left": "issue:43901", + "right": "issue:44246", + "accept": false, + "reason": "Documentation issue vs intermittent import slowness; unrelated." + }, + { + "left": "issue:43525", + "right": "issue:43643", + "accept": false, + "reason": "Different config-loading problems: missing pad_token_id for Llama4 vs missing fields with trust_remote_code." + }, + { + "left": "issue:44589", + "right": "issue:45362", + "accept": false, + "reason": "Different error modes and model paths; not the same underlying bug." + }, + { + "left": "issue:41628", + "right": "issue:43976", + "accept": false, + "reason": "ImportError for AutoImageProcessor vs Python-version compatibility regression; unrelated." + }, + { + "left": "issue:43784", + "right": "issue:45542", + "accept": false, + "reason": "Both are import/backend issues, but they involve different packages and code paths." + }, + { + "left": "issue:44188", + "right": "issue:44336", + "accept": false, + "reason": "Attention-kernel divergence vs ANSI formatting in loading_report; unrelated." + }, + { + "left": "issue:45103", + "right": "issue:46399", + "accept": false, + "reason": "auto_docstring crash with future annotations vs Qwen3.5 kernel-registration ValueError; unrelated." + }, + { + "left": "issue:45561", + "right": "issue:45563", + "accept": false, + "reason": "pytest-xdist file race vs stale warning in paged generate; different bugs." + }, + { + "left": "issue:44246", + "right": "issue:44871", + "accept": false, + "reason": "Import-performance issue vs Gemma eos_token_id config mismatch; unrelated." + }, + { + "left": "issue:44336", + "right": "issue:44373", + "accept": false, + "reason": "ANSI code emission in loading_report vs a docstring typo; not the same issue." + }, + { + "left": "issue:44295", + "right": "issue:45397", + "accept": false, + "reason": "Buffer read error vs model loading with zero3; different code paths." + }, + { + "left": "issue:44242", + "right": "issue:45593", + "accept": false, + "reason": "Both concern optional-loss branches, but they are different models and different mechanisms; not the same concrete bug." + }, + { + "left": "issue:42491", + "right": "issue:43299", + "accept": false, + "reason": "Both mention Qwen3 MoE/v5 compatibility, but one is a LoRA portability problem and the other is a Qwen3VL MoE loading regression." + }, + { + "left": "issue:43531", + "right": "issue:45362", + "accept": false, + "reason": "Both are Qwen-related, but sliding_window behavior and chat crashes are distinct failures." + }, + { + "left": "issue:43526", + "right": "issue:45003", + "accept": false, + "reason": "BeitImageProcessorFast label reduction vs unsafe sys.modules access; unrelated." + }, + { + "left": "issue:41950", + "right": "issue:44220", + "accept": false, + "reason": "Video-classification pipeline misrouting to image processors vs audio fbank extraction issue; different subsystems." + }, + { + "left": "issue:42886", + "right": "issue:46132", + "accept": false, + "reason": "Offline tokenizer cache loading vs AttentionInterface.register semantics; unrelated." + }, + { + "left": "issue:36246", + "right": "issue:42757", + "accept": false, + "reason": "Two different import errors involving unrelated APIs/packages; 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": 19, + "review_unit_index": 38, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:36246", + "issue:43257", + "issue:43262", + "issue:43352", + "issue:43475", + "issue:43596", + "issue:43673", + "issue:43901", + "issue:44162", + "issue:44242", + "issue:44246", + "issue:44295", + "issue:44336", + "issue:44351", + "issue:44466", + "issue:44479", + "issue:44485", + "issue:44496", + "issue:44561", + "issue:44734", + "issue:44749", + "issue:44779", + "issue:44857", + "issue:44877", + "issue:44987", + "issue:44995", + "issue:45003", + "issue:45137", + "issue:45200", + "issue:45250", + "issue:45292", + "issue:45335", + "issue:45375", + "issue:45397", + "issue:45412", + "issue:45459", + "issue:45561", + "issue:45584", + "issue:45656", + "issue:45701", + "issue:45759", + "issue:45859", + "issue:46123", + "issue:46170", + "issue:46291", + "issue:46369", + "issue:46399", + "issue:46424" + ], + "soft_pairs": [ + "issue:44734|issue:46424", + "issue:36246|issue:44351", + "issue:44336|issue:44485", + "issue:45003|issue:45561", + "issue:44779|issue:45292", + "issue:44336|issue:46424", + "issue:44242|issue:44857", + "issue:43475|issue:46369", + "issue:44295|issue:44749", + "issue:45859|issue:46399", + "issue:46291|issue:46399", + "issue:36246|issue:43596", + "issue:44162|issue:45200", + "issue:43901|issue:45003", + "issue:43352|issue:45250", + "issue:43673|issue:44561", + "issue:44496|issue:45003", + "issue:43257|issue:45137", + "issue:43262|issue:44246", + "issue:45292|issue:45584", + "issue:45412|issue:46123", + "issue:45375|issue:45759", + "issue:45292|issue:45701", + "issue:44561|issue:46399", + "issue:44877|issue:45375", + "issue:44987|issue:45459", + "issue:44995|issue:45200", + "issue:44466|issue:45335", + "issue:44479|issue:46170", + "issue:45397|issue:45656" + ], + "prepared_review_unit_hash": "5be94fde938ffc8d2b68019fb6dc065aa6efebcc6a498e2c29624f86199411bc", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36727, + "estimated_input_tokens": 9182, + "estimated_eval_tokens": 18620 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This set is highly heterogeneous and does not look like a true duplicate cluster. The only plausible representative issue is the broad Qwen video regression report, but even that is just a weak canonical choice.", + "confidence": 0.74, + "canonical_issue_reason": "Issue 44479 is the broadest and most concrete report here: a v5.3.0 regression affecting several Qwen VL/video variants with a clear symptom and scope.", + "canonical_pr_reason": null, + "best_issue_reason": "44479 is the best single issue to represent the set because it is specific, user-facing, and spans multiple related model families, unlike the other narrower or vaguer reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44734", + "right": "issue:46424", + "accept": false, + "reason": "Different bugs: KV-cache continuation crash in serve responses vs cache-length config handling." + }, + { + "left": "issue:36246", + "right": "issue:44351", + "accept": false, + "reason": "Both are import errors, but for unrelated symbols and model paths." + }, + { + "left": "issue:44336", + "right": "issue:44485", + "accept": false, + "reason": "ANSI-color output bug in loading_report vs GLM-5 RoPE implementation discussion." + }, + { + "left": "issue:45003", + "right": "issue:45561", + "accept": false, + "reason": "sys.modules access bug is unrelated to the xdist captured_info.txt race." + }, + { + "left": "issue:44779", + "right": "issue:45292", + "accept": false, + "reason": "Tokenization regression and resize_token_embeddings propagation are different code paths." + }, + { + "left": "issue:44336", + "right": "issue:46424", + "accept": false, + "reason": "Console coloring in loading_report is unrelated to cache-length behavior." + }, + { + "left": "issue:44242", + "right": "issue:44857", + "accept": false, + "reason": "MoE load-balancing loss omission vs an AMP/CUDA crash in LwDetrImageLoss." + }, + { + "left": "issue:43475", + "right": "issue:46369", + "accept": false, + "reason": "SAM3 vision encoder attribute error vs memory overflow in sam3-litetext are different failures." + }, + { + "left": "issue:44295", + "right": "issue:44749", + "accept": false, + "reason": "Position_ids buffer access error vs a slow-data-processing regression after upgrade." + }, + { + "left": "issue:45859", + "right": "issue:46399", + "accept": false, + "reason": "CPU-offload incompatibility in Qwen3_5MoeTextRotaryEmbedding is distinct from a kernels decorator registration crash." + }, + { + "left": "issue:46291", + "right": "issue:46399", + "accept": false, + "reason": "Kernels import/version requirement failure is not the same as the Qwen3.5 rotary-embedding registration error." + }, + { + "left": "issue:36246", + "right": "issue:43596", + "accept": false, + "reason": "Qwen2.5-VL import error and DeepSpeed ZeRO3/BertModel index error are unrelated." + }, + { + "left": "issue:44162", + "right": "issue:45200", + "accept": false, + "reason": "ESM2 model breakage and Gemma 4 token-type defaulting are different model-specific bugs." + }, + { + "left": "issue:43901", + "right": "issue:45003", + "accept": false, + "reason": "Docs mismatch for return_all_scores is unrelated to unsafe sys.modules access." + }, + { + "left": "issue:43352", + "right": "issue:45250", + "accept": false, + "reason": "Nemotron Flash Attention 2 unsupported is not the same as the vague generic Flash Attention 2 issue." + }, + { + "left": "issue:43673", + "right": "issue:44561", + "accept": false, + "reason": "Chunked-prefill cache bug and removal of is_torch_fx_available affect different paths." + }, + { + "left": "issue:44496", + "right": "issue:45003", + "accept": false, + "reason": "Unknown-model config loading failure is unrelated to sys.modules indexing." + }, + { + "left": "issue:43257", + "right": "issue:45137", + "accept": false, + "reason": "Qwen3 MOE weight conversion issue and ZeRO3 deque underflow are different DeepSpeed failures." + }, + { + "left": "issue:43262", + "right": "issue:44246", + "accept": false, + "reason": "Audio processor sampling-rate default bug is unrelated to import slowdown." + }, + { + "left": "issue:45292", + "right": "issue:45584", + "accept": false, + "reason": "Embedding resize propagation and Whisper empty-transcription failure are different behaviors." + }, + { + "left": "issue:45412", + "right": "issue:46123", + "accept": false, + "reason": "RT-DETR memory leak and MaskGenerationPipeline batch-finalization bug are unrelated." + }, + { + "left": "issue:45375", + "right": "issue:45759", + "accept": false, + "reason": "Missing deepstack_visual_indexes field and text_config unwrapping are separate Qwen config issues." + }, + { + "left": "issue:45292", + "right": "issue:45701", + "accept": false, + "reason": "Embedding resize bug and version-dependent tokenization change are not the same defect." + }, + { + "left": "issue:44561", + "right": "issue:46399", + "accept": false, + "reason": "is_torch_fx_available removal breakage is unrelated to the Qwen3.5 apply_rotary_pos_emb registration crash." + }, + { + "left": "issue:44877", + "right": "issue:45375", + "accept": false, + "reason": "granite_speech strict-config loading and Qwen3_5MoeVisionConfig field loss are different config problems." + }, + { + "left": "issue:44987", + "right": "issue:45459", + "accept": false, + "reason": "Loading a specific model package failing is unrelated to tokenizer errors hidden by protobuf exception handling." + }, + { + "left": "issue:44995", + "right": "issue:45200", + "accept": false, + "reason": "GlmMoeDsa stale cache on second forward and Gemma 4 mm_token_type_ids defaults are unrelated." + }, + { + "left": "issue:44466", + "right": "issue:45335", + "accept": false, + "reason": "lm_head.weight serialization inconsistency is distinct from resize_token_embeddings not updating decoder.embed_tokens." + }, + { + "left": "issue:44479", + "right": "issue:46170", + "accept": false, + "reason": "Qwen video-input regression and Windows ROCm distributed import failure are unrelated." + }, + { + "left": "issue:45397", + "right": "issue:45656", + "accept": false, + "reason": "Gemma-4 ZeRO3 loading and duplicate optimizer-step behavior with DeepSpeed are different 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": 19, + "review_unit_index": 39, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:42886", + "issue:42994", + "issue:43065", + "issue:43232", + "issue:43408", + "issue:43475", + "issue:43482", + "issue:43540", + "issue:43643", + "issue:43646", + "issue:43653", + "issue:43673", + "issue:43701", + "issue:43931", + "issue:44038", + "issue:44060", + "issue:44077", + "issue:44162", + "issue:44188", + "issue:44297", + "issue:44361", + "issue:44368", + "issue:44483", + "issue:44623", + "issue:44749", + "issue:44869", + "issue:44871", + "issue:44877", + "issue:44977", + "issue:44995", + "issue:45137", + "issue:45229", + "issue:45230", + "issue:45356", + "issue:45397", + "issue:45563", + "issue:45584", + "issue:45588", + "issue:45600", + "issue:45656", + "issue:45987", + "issue:46082", + "issue:46123", + "issue:46132", + "issue:46170", + "issue:46202", + "issue:46421" + ], + "soft_pairs": [ + "issue:45137|issue:45600", + "issue:45987|issue:46082", + "issue:43475|issue:43540", + "issue:44483|issue:46202", + "issue:43653|issue:44162", + "issue:42886|issue:45229", + "issue:44060|issue:44077", + "issue:44995|issue:45588", + "issue:44368|issue:44977", + "issue:43065|issue:43408", + "issue:44368|issue:45563", + "issue:44188|issue:46202", + "issue:44877|issue:46170", + "issue:43931|issue:44368", + "issue:45584|issue:46170", + "issue:44188|issue:46421", + "issue:43065|issue:46123", + "issue:44297|issue:44623", + "issue:44361|issue:44749", + "issue:44869|issue:45356", + "issue:44623|issue:45656", + "issue:43646|issue:44162", + "issue:44749|issue:45656", + "issue:43701|issue:44749", + "issue:43232|issue:43673", + "issue:44871|issue:46132", + "issue:45230|issue:45397", + "issue:42994|issue:43482", + "issue:44038|issue:44623", + "issue:43475|issue:43643" + ], + "prepared_review_unit_hash": "89f0df0ec15f372b06c5b64fb02374d5c62db277ad340f276f5df4416c451507", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 35928, + "estimated_input_tokens": 8982, + "estimated_eval_tokens": 18220 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No soft pairs look like true duplicates; they mostly share a broad subsystem or model family but describe different concrete failures, so all soft edges are rejected. For the issue set, issue 46421 is the strongest canonical tracking candidate because it is specific, open, and directly describes a cache-lazy-init recompilation regression in chunked prefill.", + "confidence": 0.86, + "canonical_issue_reason": "Issue 46421 is the clearest representative of the cache/chunked-prefill regression family: it states a concrete symptom, a narrow code path, and an active bug surface.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 46421 is the most actionable issue in the set: it is open, technically specific, and tied to a reproducible performance regression rather than a broad or ambiguous report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45137", + "right": "issue:45600", + "accept": false, + "reason": "DeepSpeed ZeRO3 deque exhaustion and removed auto_mappings config refs are unrelated failures in different code paths." + }, + { + "left": "issue:45987", + "right": "issue:46082", + "accept": false, + "reason": "StaticCache return-type mismatch and LlamaConfig head_dim validation are different bugs." + }, + { + "left": "issue:43475", + "right": "issue:43540", + "accept": false, + "reason": "SAM3 VisionEncoder attribute loss and Qwen3OmniMoe video-input ValueError are different model paths and symptoms." + }, + { + "left": "issue:44483", + "right": "issue:46202", + "accept": false, + "reason": "API request rejection and Gemma tokenizer roundtrip failure are unrelated." + }, + { + "left": "issue:43653", + "right": "issue:44162", + "accept": false, + "reason": "BigBirdTokenizer special-token registration and ESM2 breakage do not describe the same failure." + }, + { + "left": "issue:42886", + "right": "issue:45229", + "accept": false, + "reason": "Offline tokenizer cache loading and Gemma4 multi-GPU CUDA OOM are unrelated bugs." + }, + { + "left": "issue:44060", + "right": "issue:44077", + "accept": false, + "reason": "Tied-weights warning behavior and patchtsmixer post_init policy are different issues." + }, + { + "left": "issue:44995", + "right": "issue:45588", + "accept": false, + "reason": "Stale indexer cache on second forward and flash_attention s_aux=None AttributeError are different code paths." + }, + { + "left": "issue:44368", + "right": "issue:44977", + "accept": false, + "reason": "Tied embeddings warning and flash-attention generation failure are not the same concrete bug." + }, + { + "left": "issue:43065", + "right": "issue:43408", + "accept": false, + "reason": "Both are SAM3-related, but one concerns a dummy Conv2d in the pixel decoder while the other is a model-type mismatch warning." + }, + { + "left": "issue:44368", + "right": "issue:45563", + "accept": false, + "reason": "A tied-embeddings warning is unrelated to a stale num_return_sequences warning in paged generate()." + }, + { + "left": "issue:44188", + "right": "issue:46202", + "accept": false, + "reason": "Torch.compile kernel divergence and Gemma tokenizer roundtrip failure are unrelated." + }, + { + "left": "issue:44877", + "right": "issue:46170", + "accept": false, + "reason": "Strict granite_speech config loading and Windows ROCm import/type errors are different problems." + }, + { + "left": "issue:43931", + "right": "issue:44368", + "accept": false, + "reason": "Qwen3-VL weight-shape mismatch and tied-embeddings warnings are unrelated." + }, + { + "left": "issue:45584", + "right": "issue:46170", + "accept": false, + "reason": "Whisper empty-transcription generation failure and Windows ROCm import issues do not share the same code path." + }, + { + "left": "issue:43065", + "right": "issue:46123", + "accept": false, + "reason": "SAM3 pixel-decoder setup and MaskGenerationPipeline final-batch handling are unrelated." + }, + { + "left": "issue:44297", + "right": "issue:44623", + "accept": false, + "reason": "Tokenizer save_pretrained class mismatch and processor save_pretrained missing files are both save-pretrained issues, but not the same concrete failure." + }, + { + "left": "issue:44361", + "right": "issue:44749", + "accept": false, + "reason": "MLukeTokenizer AttributeError and a performance regression after upgrade are different bugs." + }, + { + "left": "issue:44869", + "right": "issue:45356", + "accept": false, + "reason": "Whisper decode crash on trailing replacement characters and Kimi-K2.5 codec regression are different tokenizer issues." + }, + { + "left": "issue:44623", + "right": "issue:45656", + "accept": false, + "reason": "Missing processor files and double optimizer stepping under deepspeed are unrelated." + }, + { + "left": "issue:43646", + "right": "issue:44162", + "accept": false, + "reason": "Both mention broad v5 breakage, but custom model initialization and ESM2 are different code paths and symptoms." + }, + { + "left": "issue:44749", + "right": "issue:45656", + "accept": false, + "reason": "A slowdown regression and optimizer.step being called twice are not the same bug." + }, + { + "left": "issue:43701", + "right": "issue:44749", + "accept": false, + "reason": "Checkpoint key mismatch and data-processing slowdown are unrelated." + }, + { + "left": "issue:43232", + "right": "issue:43673", + "accept": false, + "reason": "Generation kwargs update after sync_gpus and missing cache during chunked_prefill are both generation/cache topics, but they are distinct failure modes." + }, + { + "left": "issue:44871", + "right": "issue:46132", + "accept": false, + "reason": "Gemma eos_token_id inconsistency and AttentionInterface.register behavior change are unrelated." + }, + { + "left": "issue:45230", + "right": "issue:45397", + "accept": false, + "reason": "A generic bug report is not evidence of the same issue as gemma-4 zero3 from_pretrained." + }, + { + "left": "issue:42994", + "right": "issue:43482", + "accept": false, + "reason": "Quantized model saving failure and Qwen2.5-GGUF loading failure are separate model I/O problems." + }, + { + "left": "issue:44038", + "right": "issue:44623", + "accept": false, + "reason": "Qwen3-VL-Moe/v5 breakage and processor save_pretrained missing files are different bugs." + }, + { + "left": "issue:43475", + "right": "issue:43643", + "accept": false, + "reason": "SAM3 video encoder missing fpn_position_embeddings and trust_remote_code config-field loss are unrelated." + }, + { + "left": "issue:44188", + "right": "issue:46421", + "accept": false, + "reason": "Torch.compile branching causing divergent attention kernels and cache lazy init causing recompilation are related broadly, but not the same concrete code-path problem." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 40, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:28282", + "issue:30990", + "issue:36010", + "issue:41628", + "issue:42175", + "issue:42371", + "issue:42673", + "issue:42831", + "issue:42994", + "issue:43232", + "issue:43377", + "issue:43475", + "issue:43504", + "issue:43525", + "issue:43526", + "issue:43704", + "issue:43723", + "issue:43784", + "issue:43824", + "issue:43873", + "issue:44038", + "issue:44188", + "issue:44222", + "issue:44242", + "issue:44351", + "issue:44496", + "issue:44792", + "issue:44933", + "issue:44945", + "issue:44964", + "issue:45081", + "issue:45092", + "issue:45103", + "issue:45137", + "issue:45310", + "issue:45375", + "issue:45381", + "issue:45406", + "issue:45478", + "issue:45561", + "issue:45712", + "issue:45753", + "issue:46077", + "issue:46123", + "issue:46202", + "issue:46326", + "issue:46399" + ], + "soft_pairs": [ + "issue:28282|issue:30990", + "issue:43475|issue:45381", + "issue:45406|issue:46202", + "issue:41628|issue:44933", + "issue:42371|issue:45092", + "issue:45375|issue:45478", + "issue:43723|issue:46326", + "issue:42994|issue:43525", + "issue:43784|issue:46399", + "issue:45310|issue:45375", + "issue:45081|issue:45103", + "issue:43873|issue:44945", + "issue:46077|issue:46123", + "issue:42673|issue:43704", + "issue:41628|issue:44351", + "issue:43232|issue:44188", + "issue:44964|issue:45137", + "issue:44222|issue:45310", + "issue:44945|issue:45753", + "issue:42994|issue:43824", + "issue:44038|issue:45310", + "issue:36010|issue:41628", + "issue:43232|issue:44792", + "issue:42831|issue:44222", + "issue:42175|issue:45712", + "issue:44496|issue:45561", + "issue:43526|issue:45561", + "issue:43504|issue:45561", + "issue:42175|issue:44933", + "issue:43377|issue:44242" + ], + "prepared_review_unit_hash": "ced8dfcfe10754e79adb075424f0f1d34271a47c3d6205814059aa003cdb56bf", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 30, + "serialized_chars": 36060, + "estimated_input_tokens": 9015, + "estimated_eval_tokens": 18286 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly heterogeneous; only the Qwen3ForCausalLM VRAM-leak reports (42673/43704) are a clear duplicate pair, while the rest are distinct import/config/runtime bugs.", + "confidence": 0.94, + "canonical_issue_reason": "issue:42673 is the best duplicate anchor because issue:43704 repeats the exact same VRAM-leak title, and 42673 has the fuller report with more discussion/references.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:42673 is the strongest representative issue overall: it describes a concrete reproducible bug, has the clearest duplicate sibling, and is better documented than the other candidates.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:28282", + "right": "issue:30990", + "accept": false, + "reason": "Both involve loading/runtime failures, but one is a missing PyTorch dependency error and the other is Sentence Transformers hanging; different bugs." + }, + { + "left": "issue:43475", + "right": "issue:45381", + "accept": false, + "reason": "Both are multimodal/VL issues, but they fail on different model classes and different missing fields/position-id behavior." + }, + { + "left": "issue:45406", + "right": "issue:46202", + "accept": false, + "reason": "Gemma Processor attribute error and Gemma pipeline roundtrip tokenization failure are different code paths and symptoms." + }, + { + "left": "issue:41628", + "right": "issue:44933", + "accept": false, + "reason": "Both are import problems, but they concern different symbols and different modules; not the same underlying defect." + }, + { + "left": "issue:42371", + "right": "issue:45092", + "accept": false, + "reason": "TF32 API guidance and InternVL2 meta-init incompatibility are unrelated issues." + }, + { + "left": "issue:45375", + "right": "issue:45478", + "accept": false, + "reason": "These are both Qwen3.5/MoE-related, but one is a missing vision-config field and the other is a generic from_pretrained error; not enough to merge." + }, + { + "left": "issue:43723", + "right": "issue:46326", + "accept": false, + "reason": "Tokenizer loading in v5 and mistral_common continue_final_message are different tokenizer bugs." + }, + { + "left": "issue:42994", + "right": "issue:43525", + "accept": false, + "reason": "Quantized-model save failure and Llama4Config missing pad_token_id are unrelated failures." + }, + { + "left": "issue:43784", + "right": "issue:46399", + "accept": false, + "reason": "Import-time NameError in sentence-transformers and Qwen3.5 kernel-registration crash are different problems." + }, + { + "left": "issue:45310", + "right": "issue:45375", + "accept": false, + "reason": "One is a Qwen3.5-MoE from_pretrained error; the other is a missing deepstack_visual_indexes field in Qwen3_5MoeVisionConfig." + }, + { + "left": "issue:45081", + "right": "issue:45103", + "accept": false, + "reason": "Tokenizer regex patch crash and auto-docstring annotation crash are unrelated attribute errors." + }, + { + "left": "issue:43873", + "right": "issue:44945", + "accept": false, + "reason": "Quantization offloading behavior and pipeline-parallel output corruption are different bugs." + }, + { + "left": "issue:46077", + "right": "issue:46123", + "accept": false, + "reason": "Gemma4 attention-mask construction and MaskGenerationPipeline partial-batch handling are separate issues." + }, + { + "left": "issue:42673", + "right": "issue:43704", + "accept": true, + "reason": "Exact same Qwen3ForCausalLM VRAM-leak bug title and same failure mode; clearly duplicate reports." + }, + { + "left": "issue:41628", + "right": "issue:44351", + "accept": false, + "reason": "Both are import errors, but they are for different missing exports and different underlying causes." + }, + { + "left": "issue:43232", + "right": "issue:44188", + "accept": false, + "reason": "Generation kwarg sync behavior and torch.compile attention-kernel branching are different generation bugs." + }, + { + "left": "issue:44964", + "right": "issue:45137", + "accept": false, + "reason": "Phi-4 multimodal loading and DeepSpeed ZeRO3 deque underflow are unrelated." + }, + { + "left": "issue:44222", + "right": "issue:45310", + "accept": false, + "reason": "FP8 save_pretrained on MoE and Qwen3.5 MoE from_pretrained failure are distinct operations and symptoms." + }, + { + "left": "issue:44945", + "right": "issue:45753", + "accept": false, + "reason": "Pipeline-parallel output errors and Qwen3_5 image infinite-loop behavior are not the same bug." + }, + { + "left": "issue:42994", + "right": "issue:43824", + "accept": false, + "reason": "Quantized model saving and missing Qwen2_5_VLForConditionalGeneration import are unrelated." + }, + { + "left": "issue:44038", + "right": "issue:45310", + "accept": false, + "reason": "Qwen3-VL-Moe compatibility and Qwen3.5-MoE from_pretrained failure are different model/version problems." + }, + { + "left": "issue:36010", + "right": "issue:41628", + "accept": false, + "reason": "Missing GenerationMixin and missing AutoImageProcessor are separate import regressions." + }, + { + "left": "issue:43232", + "right": "issue:44792", + "accept": false, + "reason": "Generation kwargs after sync_gpus and janus image-generation test failure do not look like the same bug." + }, + { + "left": "issue:42831", + "right": "issue:44222", + "accept": false, + "reason": "FineGrainedFP8 accuracy drift and FP8 save_pretrained failure are different FP8-related issues." + }, + { + "left": "issue:42175", + "right": "issue:45712", + "accept": false, + "reason": "TensorFlow backend packaging and leftover dummy PT classes are unrelated packaging problems." + }, + { + "left": "issue:44496", + "right": "issue:45561", + "accept": false, + "reason": "Unrecognized model/config handling and pytest-xdist file-race in testing utils are unrelated." + }, + { + "left": "issue:43526", + "right": "issue:45561", + "accept": false, + "reason": "BeitImageProcessorFast label reduction bug and xdist race on captured_info.txt are unrelated." + }, + { + "left": "issue:43504", + "right": "issue:45561", + "accept": false, + "reason": "Legacy-field pretrained loading for Beit and xdist testing race are unrelated." + }, + { + "left": "issue:42175", + "right": "issue:44933", + "accept": false, + "reason": "TensorFlow backend installation and image_utils import issue are not the same bug." + }, + { + "left": "issue:43377", + "right": "issue:44242", + "accept": false, + "reason": "MIMI padding-mask batching bug and MoE load-balancing loss omission are different model behaviors." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. The only accepted soft edge, issue:42673 vs issue:43704, is well supported by the exact matching title and similar report content. The other pairwise rejections read appropriately cautious rather than overstated. No mergeability concerns apply because there are no PR pairs." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 41, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:22355", + "issue:27957", + "issue:28282", + "issue:30333", + "issue:30990", + "issue:33290", + "issue:34689", + "issue:35141", + "issue:36331", + "issue:42222", + "issue:42831", + "issue:42898", + "issue:43278", + "issue:43479", + "issue:43644", + "issue:43723", + "issue:43784", + "issue:43824", + "issue:43874", + "issue:43976", + "issue:44016", + "issue:44060", + "issue:44075", + "issue:44079", + "issue:44222", + "issue:44295", + "issue:44297", + "issue:44448", + "issue:44568", + "issue:44623", + "issue:44734", + "issue:44779", + "issue:44869", + "issue:44933", + "issue:45092", + "issue:45103", + "issue:45125", + "issue:45137", + "issue:45200", + "issue:45310", + "issue:45356", + "issue:45405", + "issue:45656", + "issue:45712", + "issue:45812", + "issue:46132", + "issue:46231" + ], + "soft_pairs": [ + "issue:42898|issue:44568", + "issue:43976|issue:44933", + "issue:43278|issue:45137", + "issue:30990|issue:35141", + "issue:44623|issue:45310", + "issue:42222|issue:45200", + "issue:30333|issue:33290", + "issue:43479|issue:44869", + "issue:43479|issue:44079", + "issue:42898|issue:45812", + "issue:43784|issue:43824", + "issue:43723|issue:44568", + "issue:42898|issue:44779", + "issue:42898|issue:45356", + "issue:44016|issue:46231", + "issue:44734|issue:45103", + "issue:43874|issue:45092", + "issue:44295|issue:45656", + "issue:44075|issue:46231", + "issue:42898|issue:44448", + "issue:44060|issue:45125", + "issue:45405|issue:45712", + "issue:44222|issue:44297", + "issue:44933|issue:45712", + "issue:44297|issue:45656", + "issue:44448|issue:46132", + "issue:44222|issue:44623", + "issue:27957|issue:30333", + "issue:30333|issue:34689", + "issue:43479|issue:43644", + "issue:42831|issue:44623", + "issue:30333|issue:36331", + "issue:30333|issue:35141", + "issue:22355|issue:28282" + ], + "prepared_review_unit_hash": "63045dc476d79f5ef3d39f042f4afdbcd9115f3d56351fd0f6a0a4ff27da2289", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 34, + "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": "The set is highly heterogeneous: most soft-similarity pairs share broad vocabulary or the same subsystem, but not the same concrete bug. I would not deduplicate any of the candidate pairs.", + "confidence": 0.64, + "canonical_issue_reason": "issue:42898 is the best cluster anchor because it is the clearest, most general v5 tokenizer-regression report and sits near several other tokenizer-related candidates, even though the set as a whole is not a true duplicate cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:42898 is the most representative issue among these entries: it is specific, well-scoped, and broadly phrased compared with the more model-specific or unrelated reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:42898", + "right": "issue:44568", + "accept": false, + "reason": "Both are tokenizer regressions in v5, but they describe different behaviors and different models; not the same bug." + }, + { + "left": "issue:43976", + "right": "issue:44933", + "accept": false, + "reason": "Unrelated failures: Python compatibility vs a missing image_utils import." + }, + { + "left": "issue:43278", + "right": "issue:45137", + "accept": false, + "reason": "Different code paths: dtype mismatch in evaluation vs a DeepSpeed ZeRO3 deque error." + }, + { + "left": "issue:30990", + "right": "issue:35141", + "accept": false, + "reason": "Stuck loading Sentence Transformers is unrelated to embedding reinitialization after resize." + }, + { + "left": "issue:44623", + "right": "issue:45310", + "accept": false, + "reason": "Save-pretrained missing files and Qwen3.5 MoE from_pretrained errors are different problems." + }, + { + "left": "issue:42222", + "right": "issue:45200", + "accept": false, + "reason": "VitPose breakage and Gemma 4 mm_token_type_ids behavior are unrelated multimodal issues." + }, + { + "left": "issue:30333", + "right": "issue:33290", + "accept": false, + "reason": "MLFlow reporting on failed jobs is unrelated to an Adafactor/DeepSpeed OOM." + }, + { + "left": "issue:43479", + "right": "issue:44869", + "accept": false, + "reason": "Different model components: multimodal config defaults vs Whisper timestamp decoding." + }, + { + "left": "issue:43479", + "right": "issue:44079", + "accept": false, + "reason": "Config initialization bug is unrelated to ModelOutput key assignment." + }, + { + "left": "issue:42898", + "right": "issue:45812", + "accept": false, + "reason": "Both are tokenizer regressions, but they affect different tokenization behaviors and different models." + }, + { + "left": "issue:43784", + "right": "issue:43824", + "accept": false, + "reason": "Sentence-transformers import failure and missing Qwen2.5-VL export are different import errors." + }, + { + "left": "issue:43723", + "right": "issue:44568", + "accept": false, + "reason": "General tokenizer loading issue and a specific BOS/EOS regression are not the same underlying bug." + }, + { + "left": "issue:42898", + "right": "issue:44779", + "accept": false, + "reason": "Both mention v5 tokenizer regressions, but the affected tokenizer and failure mode differ." + }, + { + "left": "issue:42898", + "right": "issue:45356", + "accept": false, + "reason": "Different tokenizer regressions for different models; not one concrete duplicate." + }, + { + "left": "issue:44016", + "right": "issue:46231", + "accept": false, + "reason": "Notebook syntax error and documentation parameter mismatch are unrelated." + }, + { + "left": "issue:44734", + "right": "issue:45103", + "accept": false, + "reason": "KV-cache continuation crash and auto-docstring annotation handling are unrelated." + }, + { + "left": "issue:43874", + "right": "issue:45092", + "accept": false, + "reason": "Different multimodal bugs: missing image patch method vs remote-code checkpoint meta-init incompatibility." + }, + { + "left": "issue:44295", + "right": "issue:45656", + "accept": false, + "reason": "Position_ids buffer read error is unrelated to duplicate optimizer stepping under DeepSpeed." + }, + { + "left": "issue:44075", + "right": "issue:46231", + "accept": false, + "reason": "SGD argument handling and a docs-only parameter name mismatch are unrelated." + }, + { + "left": "issue:42898", + "right": "issue:44448", + "accept": false, + "reason": "Both are v4-to-v5 regression reports, but for different models and different outputs." + }, + { + "left": "issue:44060", + "right": "issue:45125", + "accept": false, + "reason": "Incorrect tied-weights warning and missing tensor-parallel plan are separate Qwen3 issues." + }, + { + "left": "issue:45405", + "right": "issue:45712", + "accept": false, + "reason": "PEFT version pinning and leftover dummy classes are unrelated repo-maintenance issues." + }, + { + "left": "issue:44222", + "right": "issue:44297", + "accept": false, + "reason": "Both involve save_pretrained, but one is FP8 MoE serialization and the other is tokenizer config metadata." + }, + { + "left": "issue:27957", + "right": "issue:30333", + "accept": false, + "reason": "Flash Attention 2 on XLM-Roberta is unrelated to MLFlow reporting on failed training jobs." + }, + { + "left": "issue:30333", + "right": "issue:34689", + "accept": false, + "reason": "Training-job reporting and Llama 3.2 Vision model loading are unrelated." + }, + { + "left": "issue:43479", + "right": "issue:43644", + "accept": false, + "reason": "Default sub-config initialization and non-persistent buffer filling are different issues." + }, + { + "left": "issue:42831", + "right": "issue:44623", + "accept": false, + "reason": "FP8 accuracy concerns and processor save_pretrained missing files are unrelated." + }, + { + "left": "issue:30333", + "right": "issue:36331", + "accept": false, + "reason": "MLFlow failure reporting is unrelated to the CustomTrainer signature breakage." + }, + { + "left": "issue:30333", + "right": "issue:35141", + "accept": false, + "reason": "Failed-job MLFlow reporting is unrelated to token-embedding resizing behavior." + }, + { + "left": "issue:22355", + "right": "issue:28282", + "accept": false, + "reason": "Missing transformers.onnx module and missing PyTorch dependency are different import-time failures." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 42, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:30990", + "issue:38175", + "issue:41628", + "issue:42757", + "issue:42907", + "issue:43122", + "issue:43232", + "issue:43475", + "issue:43644", + "issue:43723", + "issue:43761", + "issue:44190", + "issue:44206", + "issue:44222", + "issue:44261", + "issue:44263", + "issue:44279", + "issue:44295", + "issue:44297", + "issue:44360", + "issue:44373", + "issue:44483", + "issue:44484", + "issue:44485", + "issue:44623", + "issue:44734", + "issue:44829", + "issue:44871", + "issue:44977", + "issue:44991", + "issue:45071", + "issue:45092", + "issue:45103", + "issue:45245", + "issue:45310", + "issue:45357", + "issue:45381", + "issue:45406", + "issue:45834", + "issue:46002", + "issue:46050", + "issue:46082", + "issue:46132", + "issue:46209", + "issue:46305", + "issue:46328", + "issue:46421", + "issue:46424" + ], + "soft_pairs": [ + "issue:45071|issue:46082", + "issue:45245|issue:46132", + "issue:43122|issue:46209", + "issue:44297|issue:45310", + "issue:41628|issue:42757", + "issue:43475|issue:46305", + "issue:44190|issue:46082", + "issue:43723|issue:44977", + "issue:44206|issue:45103", + "issue:45381|issue:46305", + "issue:43232|issue:46424", + "issue:44734|issue:46421", + "issue:44222|issue:44484", + "issue:43232|issue:45834", + "issue:44295|issue:44623", + "issue:44263|issue:45092", + "issue:44373|issue:44485", + "issue:43723|issue:44871", + "issue:44483|issue:45406", + "issue:44279|issue:46328", + "issue:43723|issue:44991", + "issue:42907|issue:46050", + "issue:30990|issue:38175", + "issue:44263|issue:44360", + "issue:45092|issue:45357", + "issue:43761|issue:46082", + "issue:43644|issue:46305", + "issue:44829|issue:46002", + "issue:44261|issue:45092" + ], + "prepared_review_unit_hash": "cc33674823a871156e0cb2d8800fb182ade56d541d5f6029a1583574eb53bf7a", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 29, + "serialized_chars": 36136, + "estimated_input_tokens": 9034, + "estimated_eval_tokens": 18324 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is highly heterogeneous: most items are unrelated Transformers v5 regressions or model-specific bugs that only share broad themes like loading, saving, generation, or config handling. None of the soft pairs look like the same underlying issue, so I would not merge any of them.", + "confidence": 0.37, + "canonical_issue_reason": "If one issue must serve as the representative, issue 43723 is the broadest and most central v5 loading regression in this set, but the cluster does not really have a true duplicate canonical.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43723 is the best overall representative because it is a concrete, user-facing tokenizer-loading regression on the common pretrained loading path; the other issues are mostly narrower, model-specific, or unrelated.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45071", + "right": "issue:46082", + "accept": false, + "reason": "Different bugs: PreTrainedConfig type-checking regression vs LlamaConfig head_dim validation error." + }, + { + "left": "issue:45245", + "right": "issue:46132", + "accept": false, + "reason": "Unrelated failures: category-count limit in tabular encoding vs AttentionInterface.register behavior change." + }, + { + "left": "issue:43122", + "right": "issue:46209", + "accept": false, + "reason": "Different code paths and symptoms: tokenizer output changes vs FP8 inference failure on B200." + }, + { + "left": "issue:44297", + "right": "issue:45310", + "accept": false, + "reason": "Both involve loading/saving, but the concrete bugs differ: tokenizer_class mismatch vs Qwen3.5 MoE from_pretrained failure." + }, + { + "left": "issue:41628", + "right": "issue:42757", + "accept": false, + "reason": "Both are import errors, but for different symbols and different dependencies; not the same underlying breakage." + }, + { + "left": "issue:43475", + "right": "issue:46305", + "accept": false, + "reason": "Different model/components: SAM3 video encoder attribute error vs MoonshineStreaming sliding-window off-by-one." + }, + { + "left": "issue:44190", + "right": "issue:46082", + "accept": false, + "reason": "Unrelated: local dataset loading script issue vs LlamaConfig configuration validation." + }, + { + "left": "issue:43723", + "right": "issue:44977", + "accept": false, + "reason": "Different problems: tokenizer loading regression vs Qwen3.5 flash-attention generation behavior." + }, + { + "left": "issue:44206", + "right": "issue:45103", + "accept": false, + "reason": "Different failure modes: unsupported center arg in feature extractor vs docstring processing crash with future annotations." + }, + { + "left": "issue:45381", + "right": "issue:46305", + "accept": false, + "reason": "Both are index/position-related, but they affect different pipelines and tensors; not the same bug." + }, + { + "left": "issue:43232", + "right": "issue:46424", + "accept": false, + "reason": "Generation kwargs update after sync_gpus is unrelated to cache max_length honoring." + }, + { + "left": "issue:44734", + "right": "issue:46421", + "accept": false, + "reason": "Both mention cache, but one is wrong tensor indexing during continuation and the other is lazy cache init causing recompilation." + }, + { + "left": "issue:44222", + "right": "issue:44484", + "accept": false, + "reason": "Different save_pretrained concerns: FP8 moe serialization vs shard size default question." + }, + { + "left": "issue:43232", + "right": "issue:45834", + "accept": false, + "reason": "Generation kwarg update bug is unrelated to Kosmos2.5 long OCR input indexing." + }, + { + "left": "issue:44295", + "right": "issue:44623", + "accept": false, + "reason": "Position_ids buffer read error vs processor.save_pretrained missing files are different save/load issues." + }, + { + "left": "issue:44263", + "right": "issue:45092", + "accept": false, + "reason": "torch.split return handling in a MoE indexer is unrelated to meta initialization compatibility for old InternVL2 checkpoints." + }, + { + "left": "issue:44373", + "right": "issue:44485", + "accept": false, + "reason": "Wrong docstring is documentation-only; GLM-5 RoPE is a runtime implementation issue." + }, + { + "left": "issue:43723", + "right": "issue:44871", + "accept": false, + "reason": "Tokenizer loading regression vs eos_token_id config mismatch; related area, but not the same underlying bug." + }, + { + "left": "issue:44483", + "right": "issue:45406", + "accept": false, + "reason": "Different server-side failures: request rejection in /v1/chat/completions vs missing _tokenizer on Gemma4Processor." + }, + { + "left": "issue:44279", + "right": "issue:46328", + "accept": false, + "reason": "Dependency issue with transformers is unrelated to the numpy-to-list padding inefficiency." + }, + { + "left": "issue:43723", + "right": "issue:44991", + "accept": false, + "reason": "Both are tokenizer-loading reports, but they do not identify the same concrete failure mode or fix." + }, + { + "left": "issue:42907", + "right": "issue:46050", + "accept": false, + "reason": "Different quantization paths and symptoms: dequantized Ministral/Devstral save failure vs sinq save/load breakage." + }, + { + "left": "issue:30990", + "right": "issue:38175", + "accept": false, + "reason": "Completely different models and bugs: Sentence Transformers load hang vs SigLIP2 zero probabilities." + }, + { + "left": "issue:44263", + "right": "issue:44360", + "accept": false, + "reason": "Same broad subsystem name, but different defects in the DSA indexer; not enough to merge." + }, + { + "left": "issue:45092", + "right": "issue:45357", + "accept": false, + "reason": "Old InternVL2 meta-init incompatibility is separate from Qwen3.5 save_pretrained visual key serialization." + }, + { + "left": "issue:43761", + "right": "issue:46082", + "accept": false, + "reason": "CLIPVisionModel hidden_states regression is unrelated to LlamaConfig head_dim validation." + }, + { + "left": "issue:43644", + "right": "issue:46305", + "accept": false, + "reason": "Non-persistent buffer junk fill is unrelated to streaming mask off-by-one behavior." + }, + { + "left": "issue:44829", + "right": "issue:46002", + "accept": false, + "reason": "Different training stability problems with different triggers and model paths." + }, + { + "left": "issue:44261", + "right": "issue:45092", + "accept": false, + "reason": "MLA layernorm epsilon precision issue is unrelated to InternVL2 meta-init checkpoint compatibility." + } + ] + }, + "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": 19, + "review_unit_index": 43, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:22355", + "issue:27957", + "issue:28218", + "issue:30990", + "issue:33357", + "issue:39401", + "issue:41628", + "issue:42371", + "issue:42757", + "issue:42831", + "issue:42886", + "issue:42907", + "issue:42994", + "issue:43066", + "issue:43122", + "issue:43232", + "issue:43408", + "issue:43479", + "issue:43550", + "issue:43576", + "issue:43646", + "issue:43756", + "issue:43784", + "issue:43824", + "issue:44038", + "issue:44060", + "issue:44117", + "issue:44188", + "issue:44263", + "issue:44295", + "issue:44297", + "issue:44368", + "issue:44373", + "issue:44521", + "issue:44568", + "issue:44704", + "issue:44913", + "issue:44945", + "issue:45278", + "issue:45310", + "issue:45593", + "issue:45753", + "issue:46002", + "issue:46132", + "issue:46305", + "issue:46326", + "issue:46369", + "issue:46402" + ], + "soft_pairs": [ + "issue:42831|issue:45310", + "issue:43479|issue:44913", + "issue:41628|issue:43824", + "issue:43408|issue:46402", + "issue:42831|issue:44038", + "issue:44263|issue:44373", + "issue:39401|issue:44568", + "issue:22355|issue:27957", + "issue:42757|issue:45278", + "issue:27957|issue:30990", + "issue:42886|issue:43066", + "issue:44368|issue:46305", + "issue:22355|issue:28218", + "issue:42886|issue:42994", + "issue:42831|issue:45753", + "issue:43408|issue:46369", + "issue:43232|issue:45593", + "issue:43576|issue:43784", + "issue:43646|issue:46002", + "issue:43756|issue:44060", + "issue:43550|issue:46132", + "issue:44295|issue:44297", + "issue:44117|issue:44704", + "issue:30990|issue:33357", + "issue:42371|issue:44263", + "issue:43122|issue:44521", + "issue:42907|issue:44945", + "issue:44188|issue:44373", + "issue:43122|issue:46326", + "issue:41628|issue:45278" + ], + "prepared_review_unit_hash": "505cb83f8d686abd8587ff79dddcba2f005c10348a54c751585cacad16978f3e", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36196, + "estimated_input_tokens": 9049, + "estimated_eval_tokens": 18354 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab bag of distinct Transformers regressions and API issues (imports, tokenizers, config reloads, model-specific bugs). The apparent similarities are broad subsystem overlaps, not the same concrete defect, so no soft-edge duplicate merges are justified.", + "confidence": 0.9, + "canonical_issue_reason": "issue:45278 is the broadest umbrella in the set for version-upgrade import regressions, but the cluster is too heterogeneous for a true duplicate canonical.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45278 is the closest cluster anchor because it is the most general and representative of the import/API-regression subset, even though it does not actually subsume the other reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:42831", + "right": "issue:45310", + "accept": false, + "reason": "Different bugs: FP8 accuracy drift vs a Qwen3.5-Moe from_pretrained loading error." + }, + { + "left": "issue:43479", + "right": "issue:44913", + "accept": false, + "reason": "Different config problems: multimodal default sub-config initialization vs GPTNeoX rotary_pct not persisting on reload." + }, + { + "left": "issue:41628", + "right": "issue:43824", + "accept": false, + "reason": "Both are import failures, but for different symbols/models and different code paths." + }, + { + "left": "issue:43408", + "right": "issue:46402", + "accept": false, + "reason": "Different SAM3 issues: a model-type warning vs missing text-encoder weight loading." + }, + { + "left": "issue:42831", + "right": "issue:44038", + "accept": false, + "reason": "Unrelated: FP8 accuracy issue vs a Qwen3-VL-Moe transformer's from_pretrained bug." + }, + { + "left": "issue:44263", + "right": "issue:44373", + "accept": false, + "reason": "Different scope: torch.split behavior in an indexer vs a docstring mistake for position_ids." + }, + { + "left": "issue:39401", + "right": "issue:44568", + "accept": false, + "reason": "Both are tokenization-related, but one is wrong offset mappings and the other is missing BOS/EOS insertion." + }, + { + "left": "issue:22355", + "right": "issue:27957", + "accept": false, + "reason": "Different areas: missing transformers.onnx module vs XLMRoberta Flash Attention 2 support." + }, + { + "left": "issue:42757", + "right": "issue:45278", + "accept": false, + "reason": "45278 is a broad import-regression umbrella, but this is a specific huggingface_hub symbol import breakage." + }, + { + "left": "issue:27957", + "right": "issue:30990", + "accept": false, + "reason": "XLMRoberta Flash Attention 2 support is unrelated to Sentence Transformers hanging on load." + }, + { + "left": "issue:42886", + "right": "issue:43066", + "accept": false, + "reason": "Different tokenizer bugs: offline cache loading vs wrong decoder type in v5." + }, + { + "left": "issue:44368", + "right": "issue:46305", + "accept": false, + "reason": "A warning about tie_word_embeddings is unrelated to the MoonshineStreaming off-by-one mask bug." + }, + { + "left": "issue:22355", + "right": "issue:28218", + "accept": false, + "reason": "onnx import failure and tokenizer added-space behavior are unrelated bugs." + }, + { + "left": "issue:42886", + "right": "issue:42994", + "accept": false, + "reason": "Offline tokenizer cache loading and quantized model saving failure are different code paths." + }, + { + "left": "issue:42831", + "right": "issue:45753", + "accept": false, + "reason": "Different failures: accuracy regression in FineGrainedFP8 vs Qwen3_5 infinite loop on a specific image." + }, + { + "left": "issue:43408", + "right": "issue:46369", + "accept": false, + "reason": "Both mention SAM3, but one is a warning about model type and the other is a memory overflow in a different model variant." + }, + { + "left": "issue:43232", + "right": "issue:45593", + "accept": false, + "reason": "Generation kwargs after sync_gpus is unrelated to D-FINE auxiliary-loss behavior." + }, + { + "left": "issue:43576", + "right": "issue:43784", + "accept": false, + "reason": "A broken transformers env command is unrelated to a sentence-transformers import NameError." + }, + { + "left": "issue:43646", + "right": "issue:46002", + "accept": false, + "reason": "Custom model initialization breakage and NaNs in classification heads are different symptoms and likely different fixes." + }, + { + "left": "issue:43756", + "right": "issue:44060", + "accept": false, + "reason": "Different model-specific regressions: Smollm3 RoPE layer handling vs Qwen3-Next tied-weights warning." + }, + { + "left": "issue:43550", + "right": "issue:46132", + "accept": false, + "reason": "torch.compile SDPA failure is not the same as AttentionInterface.register changing behavior." + }, + { + "left": "issue:44295", + "right": "issue:44297", + "accept": false, + "reason": "These are separate tokenizer/position-id issues: buffer access error vs tokenizer_class mismatch on save." + }, + { + "left": "issue:44117", + "right": "issue:44704", + "accept": false, + "reason": "TOKENIZER_MAPPING_NAMES returning None is a different failure from AutoProcessor not forwarding kwargs." + }, + { + "left": "issue:30990", + "right": "issue:33357", + "accept": false, + "reason": "Sentence Transformers loading hangs and a MacOS bus error with CLIP are distinct runtime failures." + }, + { + "left": "issue:42371", + "right": "issue:44263", + "accept": false, + "reason": "TF32 settings guidance and torch.split return values in GlmMoeDsaIndexer are unrelated." + }, + { + "left": "issue:43122", + "right": "issue:44521", + "accept": false, + "reason": "Different tokenizer behavior: version-dependent tokenization drift vs multimodal assistant_masks all-zero." + }, + { + "left": "issue:42907", + "right": "issue:44945", + "accept": false, + "reason": "Quantized model saving failure and incorrect pipeline-parallel LLM output are separate bugs." + }, + { + "left": "issue:44188", + "right": "issue:44373", + "accept": false, + "reason": "torch.compile attention-kernel divergence is unrelated to a docstring error." + }, + { + "left": "issue:43122", + "right": "issue:46326", + "accept": false, + "reason": "Tokenization drift across versions and continue_final_message for mistral_common tokenizers are different defects." + }, + { + "left": "issue:41628", + "right": "issue:45278", + "accept": false, + "reason": "Both are import problems, but 45278 is a broad report of many import errors while 41628 is a specific missing AutoImageProcessor symbol." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 44, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:22355", + "issue:29127", + "issue:29159", + "issue:36296", + "issue:39401", + "issue:42175", + "issue:42673", + "issue:42757", + "issue:42907", + "issue:43066", + "issue:43116", + "issue:43122", + "issue:43278", + "issue:43408", + "issue:43452", + "issue:43475", + "issue:43493", + "issue:43784", + "issue:43824", + "issue:43937", + "issue:43976", + "issue:43992", + "issue:44038", + "issue:44222", + "issue:44261", + "issue:44295", + "issue:44297", + "issue:44351", + "issue:44360", + "issue:44373", + "issue:44485", + "issue:44488", + "issue:44514", + "issue:44521", + "issue:44625", + "issue:44704", + "issue:44779", + "issue:44829", + "issue:44938", + "issue:44945", + "issue:44964", + "issue:45072", + "issue:45200", + "issue:45216", + "issue:45290", + "issue:45478", + "issue:45600", + "issue:45753" + ], + "soft_pairs": [ + "issue:42907|issue:43937", + "issue:39401|issue:44488", + "issue:44261|issue:44373", + "issue:43976|issue:44945", + "issue:43784|issue:45600", + "issue:44704|issue:45290", + "issue:42907|issue:44964", + "issue:39401|issue:44521", + "issue:43116|issue:45200", + "issue:44038|issue:44297", + "issue:43122|issue:44514", + "issue:44625|issue:44829", + "issue:44360|issue:44485", + "issue:44038|issue:44295", + "issue:43408|issue:43475", + "issue:42907|issue:43452", + "issue:43122|issue:44779", + "issue:44038|issue:44222", + "issue:45216|issue:45478", + "issue:43824|issue:44351", + "issue:36296|issue:43116", + "issue:42757|issue:44351", + "issue:22355|issue:29159", + "issue:43278|issue:44945", + "issue:42175|issue:44938", + "issue:43493|issue:45753", + "issue:43066|issue:44779", + "issue:42673|issue:45072", + "issue:22355|issue:29127", + "issue:43992|issue:44704" + ], + "prepared_review_unit_hash": "61f86d457a6a738d8d5600be048587a42e89c5ef077b34a937c47a9d1741eb7f", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36310, + "estimated_input_tokens": 9078, + "estimated_eval_tokens": 18412 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Most candidate edges are not true duplicates; they connect issues from different subsystems, models, or failure modes. I reject all soft edges. If one issue has to represent the cluster, issue 43122 is the broadest tokenizer-regression report, but it is not a duplicate of the others.", + "confidence": 0.66, + "canonical_issue_reason": "Issue 43122 is the broadest, most representative regression report in this set (tokenization changes across versions), while most other issues are narrower, model-specific, or unrelated.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43122 best serves as the cluster\u2019s representative because it captures a generic tokenizer-regression pattern and has more cross-references than the other candidates.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:42907", + "right": "issue:43937", + "accept": false, + "reason": "Different failure modes: dequantized save/load behavior vs GenerationConfig validation." + }, + { + "left": "issue:39401", + "right": "issue:44488", + "accept": false, + "reason": "Qwen3 tokenizer offset mapping bug and a separate model-loading failure are unrelated." + }, + { + "left": "issue:44261", + "right": "issue:44373", + "accept": false, + "reason": "LayerNorm epsilon precision issue and a docstring problem are not the same bug." + }, + { + "left": "issue:43976", + "right": "issue:44945", + "accept": false, + "reason": "Python-version compatibility and pipeline-parallelism output corruption are distinct issues." + }, + { + "left": "issue:43784", + "right": "issue:45600", + "accept": false, + "reason": "Import-time NameError in sentence-transformers is unrelated to stale auto_mappings references." + }, + { + "left": "issue:44704", + "right": "issue:45290", + "accept": false, + "reason": "cached_file kwargs forwarding and chat-template tool-call crashes are different code paths." + }, + { + "left": "issue:42907", + "right": "issue:44964", + "accept": false, + "reason": "Dequantized save_pretrained failure and Phi-4 multimodal loading failure are unrelated." + }, + { + "left": "issue:39401", + "right": "issue:44521", + "accept": false, + "reason": "Offset mapping correctness and multimodal assistant_masks generation are different tokenizer behaviors." + }, + { + "left": "issue:43116", + "right": "issue:45200", + "accept": false, + "reason": "Example-script empty outputs and mm_token_type_ids defaults are not the same underlying bug." + }, + { + "left": "issue:44038", + "right": "issue:44297", + "accept": false, + "reason": "Qwen3-VL-Moe issue and tokenizer_class serialization mismatch are separate problems." + }, + { + "left": "issue:43122", + "right": "issue:44514", + "accept": false, + "reason": "Both are tokenizer-related, but one is a version-wide regression and the other is a batched chat-template crash." + }, + { + "left": "issue:44625", + "right": "issue:44829", + "accept": false, + "reason": "num_labels propagation and flash_attention_3 training degeneration are unrelated." + }, + { + "left": "issue:44360", + "right": "issue:44485", + "accept": false, + "reason": "DSA indexer activation bug and GLM-5 RoPE implementation issue are not duplicates." + }, + { + "left": "issue:44038", + "right": "issue:44295", + "accept": false, + "reason": "Qwen3-VL-Moe loading behavior and position_ids buffer access are unrelated." + }, + { + "left": "issue:43408", + "right": "issue:43475", + "accept": false, + "reason": "Model-type warning and missing fpn_position_embeddings attribute are separate SAM3 issues." + }, + { + "left": "issue:42907", + "right": "issue:43452", + "accept": false, + "reason": "Dequantized save/load behavior and gguf_file loading failures do not share the same bug." + }, + { + "left": "issue:43122", + "right": "issue:44779", + "accept": false, + "reason": "Both mention tokenizer regressions, but they affect different models and likely different code paths." + }, + { + "left": "issue:44038", + "right": "issue:44222", + "accept": false, + "reason": "Qwen3-VL-Moe load issue and FP8 save_pretrained moe issue are different concrete failures." + }, + { + "left": "issue:45216", + "right": "issue:45478", + "accept": false, + "reason": "Related model/version, but save_pretrained corruption and from_pretrained load error are not clearly the same bug." + }, + { + "left": "issue:43824", + "right": "issue:44351", + "accept": false, + "reason": "Missing Qwen2_5_VLForConditionalGeneration import and missing HybridCache import are unrelated import errors." + }, + { + "left": "issue:36296", + "right": "issue:43116", + "accept": false, + "reason": "Tensor parallel training bug and multi-label classification example behavior are unrelated." + }, + { + "left": "issue:42757", + "right": "issue:44351", + "accept": false, + "reason": "huggingface_hub import failure and transformers import failure are different missing-symbol problems." + }, + { + "left": "issue:22355", + "right": "issue:29159", + "accept": false, + "reason": "Missing transformers.onnx module and slow/fast tokenizer inconsistency are unrelated." + }, + { + "left": "issue:43278", + "right": "issue:44945", + "accept": false, + "reason": "Embedding dtype drift in evaluate and pipeline-parallelism output errors are not the same issue." + }, + { + "left": "issue:42175", + "right": "issue:44938", + "accept": false, + "reason": "Package-extras backend omission and Python 3.14 load failure are unrelated." + }, + { + "left": "issue:43493", + "right": "issue:45753", + "accept": false, + "reason": "SigLIP2 implementation discrepancy and Qwen3_5 infinite-loop behavior are different problems." + }, + { + "left": "issue:43066", + "right": "issue:44779", + "accept": false, + "reason": "Both are tokenizer regressions in v5, but the reported symptoms and models differ too much to treat as one bug." + }, + { + "left": "issue:42673", + "right": "issue:45072", + "accept": false, + "reason": "VRAM leak in dataloader-thread usage and bfloat16 dtype mismatches are unrelated." + }, + { + "left": "issue:22355", + "right": "issue:29127", + "accept": false, + "reason": "Missing module import and LayoutLMv3 error-message clarity are not duplicates." + }, + { + "left": "issue:43992", + "right": "issue:44704", + "accept": false, + "reason": "Missing embed_tokens.weight on load and cached_file kwarg forwarding 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": 19, + "review_unit_index": 45, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:33453", + "issue:38617", + "issue:39401", + "issue:42222", + "issue:42371", + "issue:42673", + "issue:43064", + "issue:43116", + "issue:43122", + "issue:43441", + "issue:43450", + "issue:43493", + "issue:43611", + "issue:43673", + "issue:43825", + "issue:43827", + "issue:43874", + "issue:43906", + "issue:43937", + "issue:43976", + "issue:43986", + "issue:44038", + "issue:44075", + "issue:44117", + "issue:44188", + "issue:44291", + "issue:44355", + "issue:44380", + "issue:44466", + "issue:44485", + "issue:44779", + "issue:44936", + "issue:44977", + "issue:45071", + "issue:45081", + "issue:45083", + "issue:45278", + "issue:45357", + "issue:45405", + "issue:45446", + "issue:45478", + "issue:45563", + "issue:45588", + "issue:46002", + "issue:46077", + "issue:46231", + "issue:46298", + "issue:46328" + ], + "soft_pairs": [ + "issue:38617|issue:45278", + "issue:43937|issue:44485", + "issue:43611|issue:46002", + "issue:44188|issue:46328", + "issue:44355|issue:46231", + "issue:43116|issue:43493", + "issue:42222|issue:43441", + "issue:42371|issue:43874", + "issue:44075|issue:44936", + "issue:45357|issue:45478", + "issue:44038|issue:45478", + "issue:43976|issue:45446", + "issue:43116|issue:43825", + "issue:45083|issue:45446", + "issue:43064|issue:43673", + "issue:33453|issue:39401", + "issue:43937|issue:45563", + "issue:39401|issue:44779", + "issue:44117|issue:45588", + "issue:44380|issue:46077", + "issue:42673|issue:43906", + "issue:43827|issue:45446", + "issue:43450|issue:43986", + "issue:43673|issue:46328", + "issue:44291|issue:45071", + "issue:44466|issue:46002", + "issue:43937|issue:44977", + "issue:45405|issue:45446", + "issue:44117|issue:45081", + "issue:43122|issue:46298" + ], + "prepared_review_unit_hash": "c3334f1916fd2b48bc42a1bc07591c99501a68ccad829ca96ec1978705874c1d", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 30, + "serialized_chars": 36566, + "estimated_input_tokens": 9142, + "estimated_eval_tokens": 18540 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is a heterogeneous collection of Transformers v5 regression issues across unrelated subsystems (tokenizers, model loading, training, attention, docs, etc.). The soft links are mostly superficial title similarity rather than true duplicates. No PRs were provided.", + "confidence": 0.87, + "canonical_issue_reason": "issue:45278 is the broadest, most representative regression report in the set: it captures a general v5 upgrade breakage theme and is more suitable as an umbrella than the narrower model-specific bugs.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45278 is the best single issue representative because it is high-level, version-focused, and likely the easiest to route or cluster against compared with the many narrow, model-specific reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:38617", + "right": "issue:45278", + "accept": false, + "reason": "Both are import-related, but 45278 is a broad upgrade regression while 38617 is a specific missing symbol ImportError; not the same concrete bug." + }, + { + "left": "issue:43937", + "right": "issue:44485", + "accept": false, + "reason": "Same model family (GLM-5) but different failures: invalid GenerationConfig vs RoPE implementation discussion." + }, + { + "left": "issue:43611", + "right": "issue:46002", + "accept": false, + "reason": "Unrelated bugs: model loading with base_model_prefix vs NaNs in classification heads." + }, + { + "left": "issue:44188", + "right": "issue:46328", + "accept": false, + "reason": "Different subsystems entirely: attention kernel divergence vs SequenceFeatureExtractor.pad performance." + }, + { + "left": "issue:44355", + "right": "issue:46231", + "accept": false, + "reason": "Compiled Python file errors are unrelated to the set_training() docs parameter mismatch." + }, + { + "left": "issue:43116", + "right": "issue:43493", + "accept": false, + "reason": "Example script multi-label empty results and SigLIP2 implementation discrepancy are unrelated." + }, + { + "left": "issue:42222", + "right": "issue:43441", + "accept": false, + "reason": "Both are model-specific issues, but vitpose broken imports and Ministral-3 flash-attention failures are different bugs." + }, + { + "left": "issue:42371", + "right": "issue:43874", + "accept": false, + "reason": "TF32 API settings and a missing image-patch helper in GLM46V are unrelated changes." + }, + { + "left": "issue:44075", + "right": "issue:44936", + "accept": false, + "reason": "Optimizer argument handling is unrelated to trainer.evaluate() failing after trainer.train()." + }, + { + "left": "issue:45357", + "right": "issue:45478", + "accept": false, + "reason": "Same model family, but save_pretrained key serialization and from_pretrained loading failure are not the same concrete bug." + }, + { + "left": "issue:44038", + "right": "issue:45478", + "accept": false, + "reason": "Both mention Qwen3/Qwen3.5 MoE, but they concern different model variants and different failure modes." + }, + { + "left": "issue:43976", + "right": "issue:45446", + "accept": false, + "reason": "Python version compatibility and a PyTorch version check bug are unrelated." + }, + { + "left": "issue:43116", + "right": "issue:43825", + "accept": false, + "reason": "Classification example output bug and pipeline removal/error-message issue are different problems." + }, + { + "left": "issue:45083", + "right": "issue:45446", + "accept": false, + "reason": "Feature-extraction length behavior in qwen3_omni_moe is unrelated to the flex_attention import/version check." + }, + { + "left": "issue:43064", + "right": "issue:43673", + "accept": false, + "reason": "Training/FSDP optimizer-state corruption and GenerationMixin cache missing in chunked_prefill are unrelated." + }, + { + "left": "issue:33453", + "right": "issue:39401", + "accept": false, + "reason": "Both are tokenizer regressions, but loading regression and wrong offset_mapping are distinct bugs." + }, + { + "left": "issue:43937", + "right": "issue:45563", + "accept": false, + "reason": "GLM-5 invalid GenerationConfig and a stale warning for num_return_sequences are unrelated." + }, + { + "left": "issue:39401", + "right": "issue:44779", + "accept": false, + "reason": "Both are tokenizer correctness issues, but they affect different tokenizers and different failure modes." + }, + { + "left": "issue:44117", + "right": "issue:45588", + "accept": false, + "reason": "TOKENIZER_MAPPING_NAMES returning None is unrelated to the flash_attention crash on s_aux=None." + }, + { + "left": "issue:44380", + "right": "issue:46077", + "accept": false, + "reason": "Both involve attention behavior, but GPT2 backend scaling and Gemma4 bidirectional-mask construction are different bugs." + }, + { + "left": "issue:42673", + "right": "issue:43906", + "accept": false, + "reason": "VRAM leak in Qwen3ForCausalLM and a reproduction of another issue are not the same underlying bug." + }, + { + "left": "issue:43827", + "right": "issue:45446", + "accept": false, + "reason": "Docs still referencing pipeline() and a flex_attention PyTorch version check are unrelated." + }, + { + "left": "issue:43450", + "right": "issue:43986", + "accept": false, + "reason": "Video processor batched-shape bug and AutoProcessor crash without torchvision are different issues." + }, + { + "left": "issue:43673", + "right": "issue:46328", + "accept": false, + "reason": "Generation cache handling and SequenceFeatureExtractor.pad performance are unrelated." + }, + { + "left": "issue:44291", + "right": "issue:45071", + "accept": false, + "reason": "Both are model-loading/config-related regressions, but unexpected _is_hf_initialized argument and PretrainedConfig type checking are distinct problems." + }, + { + "left": "issue:44466", + "right": "issue:46002", + "accept": false, + "reason": "lm_head serialization inconsistency and NaNs in classification heads are unrelated." + }, + { + "left": "issue:43937", + "right": "issue:44977", + "accept": false, + "reason": "GLM-5 invalid GenerationConfig and Qwen3.5 flash-attention generation failure are different bugs." + }, + { + "left": "issue:45405", + "right": "issue:45446", + "accept": false, + "reason": "Package version bumping for PEFT and a PyTorch version check in flex_attention are unrelated." + }, + { + "left": "issue:44117", + "right": "issue:45081", + "accept": false, + "reason": "Tokenizer mapping fallback bug and Mistral regex patch crash are unrelated." + }, + { + "left": "issue:43122", + "right": "issue:46298", + "accept": false, + "reason": "Tokenizer behavior change across versions and a noisy logging message are not the same issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 46, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "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:43097", + "issue:43122", + "issue:43408", + "issue:43421", + "issue:43577", + "issue:43582", + "issue:43673", + "issue:43742", + "issue:43937", + "issue:44075", + "issue:44220", + "issue:44261", + "issue:44263", + "issue:44410", + "issue:44483", + "issue:44484", + "issue:44485", + "issue:44493", + "issue:44496", + "issue:44829", + "issue:44877", + "issue:44933", + "issue:45071", + "issue:45216", + "issue:45278", + "issue:45310", + "issue:45341", + "issue:45362", + "issue:45464", + "issue:45478", + "issue:45542", + "issue:45561", + "issue:45656", + "issue:46178", + "issue:46231", + "issue:46328" + ], + "soft_pairs": [ + "issue:44410|issue:45310", + "issue:45341|issue:45561", + "issue:38617|issue:41628", + "issue:45341|issue:46328", + "issue:42222|issue:46178", + "issue:44829|issue:45216", + "issue:42617|issue:45341", + "issue:42673|issue:43582", + "issue:45362|issue:46328", + "issue:43408|issue:43742", + "issue:39401|issue:43421", + "issue:42490|issue:42673", + "issue:44877|issue:46231", + "issue:42371|issue:44261", + "issue:44484|issue:45310", + "issue:38617|issue:44220", + "issue:43937|issue:44496", + "issue:44933|issue:45278", + "issue:43066|issue:43122", + "issue:42222|issue:45542", + "issue:44410|issue:45478", + "issue:44483|issue:45464", + "issue:42371|issue:42831", + "issue:43012|issue:43577", + "issue:44075|issue:45656", + "issue:42175|issue:42222", + "issue:43097|issue:44493", + "issue:43673|issue:45341", + "issue:41628|issue:45071", + "issue:44263|issue:44485", + "issue:45071|issue:45542" + ], + "prepared_review_unit_hash": "5206a7682bbcdcc2d031ebac25621e2a3f0dcc6acf586bb8bf9e69d42b191901", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 31, + "serialized_chars": 36531, + "estimated_input_tokens": 9133, + "estimated_eval_tokens": 18522 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is a loose thematic cluster of Transformers issues, not a true duplicate set: it mixes import/API regressions, tokenizer behavior changes, model-loading failures, backend/install problems, precision warnings, and testing utility bugs. I would not merge any of the soft edges.", + "confidence": 0.93, + "canonical_issue_reason": "No single canonical issue is safe to choose; the cluster is heterogeneous and the items are not the same underlying bug.", + "canonical_pr_reason": null, + "best_issue_reason": "If forced, issue 45278 is the broadest representative of the import-regression subtheme, but it still does not represent most of the cluster.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44410", + "right": "issue:45310", + "accept": false, + "reason": "Different Qwen bugs: missing projections in one model vs a from_pretrained loading error in another." + }, + { + "left": "issue:45341", + "right": "issue:45561", + "accept": false, + "reason": "Different test-utils problems: a generic bug in testing_utils.py vs a pytest-xdist file race." + }, + { + "left": "issue:38617", + "right": "issue:41628", + "accept": false, + "reason": "Both are import errors, but they concern different missing symbols and different code paths." + }, + { + "left": "issue:45341", + "right": "issue:46328", + "accept": false, + "reason": "One is a correctness/race issue in test helpers; the other is a performance inefficiency in feature padding." + }, + { + "left": "issue:42222", + "right": "issue:46178", + "accept": false, + "reason": "Unrelated: vitpose model breakage vs a TensorFlow backend mapping error." + }, + { + "left": "issue:44829", + "right": "issue:45216", + "accept": false, + "reason": "Different regressions: FA3 training degeneracy vs Qwen3.5 save_pretrained checkpoint corruption." + }, + { + "left": "issue:42617", + "right": "issue:45341", + "accept": false, + "reason": "A script execution problem is not the same bug as a testing_utils.py issue." + }, + { + "left": "issue:42673", + "right": "issue:43582", + "accept": false, + "reason": "Different failures: VRAM leak under threaded loading vs a TypeError on Apple Silicon warmup." + }, + { + "left": "issue:45362", + "right": "issue:46328", + "accept": false, + "reason": "Qwen chat crash and feature-extractor padding inefficiency are unrelated." + }, + { + "left": "issue:43408", + "right": "issue:43742", + "accept": false, + "reason": "A model-type warning is not the same as a key error when loading a specific checkpoint." + }, + { + "left": "issue:39401", + "right": "issue:43421", + "accept": false, + "reason": "Both touch tokenization, but offset_mapping is a different bug from runtime post-processor updates." + }, + { + "left": "issue:42490", + "right": "issue:42673", + "accept": false, + "reason": "A local-path import warning is unrelated to a multi-thread VRAM leak." + }, + { + "left": "issue:44877", + "right": "issue:46231", + "accept": false, + "reason": "Config loading strictness and a documentation parameter typo are different issues." + }, + { + "left": "issue:42371", + "right": "issue:44261", + "accept": false, + "reason": "TF32 behavior guidance is unrelated to the MLA rms_norm_eps precision bug." + }, + { + "left": "issue:44484", + "right": "issue:45310", + "accept": false, + "reason": "A question about save_pretrained shard size is not the same as a Qwen3.5 from_pretrained failure." + }, + { + "left": "issue:38617", + "right": "issue:44220", + "accept": false, + "reason": "ImportError for a missing symbol and an _torch_extract_fbank_features bug are different code paths." + }, + { + "left": "issue:43937", + "right": "issue:44496", + "accept": false, + "reason": "GenerationConfig validation failure is distinct from an unrecognized-model config.json error." + }, + { + "left": "issue:44933", + "right": "issue:45278", + "accept": false, + "reason": "Related theme of imports, but not the same concrete missing-symbol bug." + }, + { + "left": "issue:43066", + "right": "issue:43122", + "accept": false, + "reason": "Both are tokenizer regressions, but one is decoder type and the other is output tokenization differences." + }, + { + "left": "issue:42222", + "right": "issue:45542", + "accept": false, + "reason": "Different backend problems: model breakage vs TensorFlow being missing despite tensorboard being installed." + }, + { + "left": "issue:44410", + "right": "issue:45478", + "accept": false, + "reason": "Different Qwen3.5 failures with different symptoms and likely different code paths." + }, + { + "left": "issue:44483", + "right": "issue:45464", + "accept": false, + "reason": "Both involve chat/completions, but one is request validation and the other is streaming inference failure." + }, + { + "left": "issue:42371", + "right": "issue:42831", + "accept": false, + "reason": "TF32 API guidance is unrelated to FineGrainedFP8 accuracy degradation." + }, + { + "left": "issue:43012", + "right": "issue:43577", + "accept": false, + "reason": "A compile-time float32 warning is not the same as model dtype actually remaining float32 on load." + }, + { + "left": "issue:44075", + "right": "issue:45656", + "accept": false, + "reason": "SGD argument handling and double optimizer stepping under deepspeed are different bugs." + }, + { + "left": "issue:42175", + "right": "issue:42222", + "accept": false, + "reason": "Backend dependency installation and vitpose model breakage are unrelated." + }, + { + "left": "issue:43097", + "right": "issue:44493", + "accept": false, + "reason": "Removed API warning and unexpected position-id keys are different regressions." + }, + { + "left": "issue:43673", + "right": "issue:45341", + "accept": false, + "reason": "Chunked-prefill cache handling is unrelated to the testing_utils bug." + }, + { + "left": "issue:41628", + "right": "issue:45071", + "accept": false, + "reason": "An import error for AutoImageProcessor is not the same as a PreTrainedConfig type-checking regression." + }, + { + "left": "issue:44263", + "right": "issue:44485", + "accept": false, + "reason": "These are separate GLM-5 issues: torch.split return handling vs RoPE implementation details." + }, + { + "left": "issue:45071", + "right": "issue:45542", + "accept": false, + "reason": "Type-checking behavior and TensorFlow backend installation 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": 19, + "review_unit_index": 47, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:27957", + "issue:38617", + "issue:42757", + "issue:42890", + "issue:43064", + "issue:43097", + "issue:43335", + "issue:43352", + "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:44117", + "issue:44261", + "issue:44263", + "issue:44315", + "issue:44351", + "issue:44360", + "issue:44410", + "issue:44484", + "issue:44485", + "issue:44829", + "issue:44933", + "issue:44987", + "issue:45003", + "issue:45071", + "issue:45341", + "issue:45372", + "issue:45375", + "issue:45440", + "issue:45478", + "issue:45588", + "issue:45753", + "issue:46050", + "issue:46123", + "issue:46172", + "issue:46231", + "issue:46298", + "issue:46369" + ], + "soft_pairs": [ + "issue:27957|issue:43352", + "issue:43874|issue:44261", + "issue:44484|issue:45478", + "issue:44263|issue:46172", + "issue:43931|issue:44410", + "issue:44360|issue:46172", + "issue:43441|issue:44829", + "issue:43931|issue:44038", + "issue:43097|issue:43452", + "issue:43335|issue:45071", + "issue:45003|issue:45341", + "issue:43673|issue:46123", + "issue:43986|issue:45341", + "issue:44933|issue:46231", + "issue:45753|issue:46369", + "issue:44117|issue:44987", + "issue:43986|issue:44484", + "issue:43874|issue:44263", + "issue:43986|issue:45372", + "issue:43531|issue:45341", + "issue:45440|issue:46172", + "issue:43097|issue:46050", + "issue:43097|issue:45588", + "issue:44261|issue:44485", + "issue:38617|issue:42757", + "issue:43064|issue:44315", + "issue:44062|issue:44117", + "issue:44410|issue:45375", + "issue:43502|issue:43992", + "issue:44038|issue:44410", + "issue:42890|issue:46298", + "issue:38617|issue:44351", + "issue:43097|issue:43873" + ], + "prepared_review_unit_hash": "a3064feab6b6b2bb8d58fba6a334efbcd8f87897c6297c714f750f37fd467f34", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 47, + "item_count": 47, + "soft_pair_count": 33, + "serialized_chars": 36772, + "estimated_input_tokens": 9193, + "estimated_eval_tokens": 18642 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This cluster is mostly a grab-bag of unrelated Transformers regressions: import/API breakages, model-specific loading mismatches, FlashAttention issues, and a few training/pipeline bugs. I don\u2019t see any soft pair that looks like the same concrete underlying bug, so all candidate duplicates are rejected.", + "confidence": 0.79, + "canonical_issue_reason": "No single issue is a strong canonical duplicate for the cluster; if one must be chosen, issue 43873 is the least-bad representative because it is the broadest active bug report with the most discussion/inbound references.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43873 is the best representative by scope and activity, but only weakly: it covers a broad quantization/offloading regression and has the strongest signal of being a hub among these otherwise unrelated reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:27957", + "right": "issue:43352", + "accept": false, + "reason": "Both are FlashAttention-related, but they involve different models and different failure conditions; not the same code-path bug." + }, + { + "left": "issue:43874", + "right": "issue:44261", + "accept": false, + "reason": "Different bugs in different components: image patch counting vs MLA q_a_layernorm config precision." + }, + { + "left": "issue:44484", + "right": "issue:45478", + "accept": false, + "reason": "Unrelated symptoms: shard-size question vs Qwen3.5 MoE from_pretrained load error." + }, + { + "left": "issue:44263", + "right": "issue:46172", + "accept": false, + "reason": "Different model internals and different failure modes; no shared underlying defect." + }, + { + "left": "issue:43931", + "right": "issue:44410", + "accept": false, + "reason": "Both concern Qwen-family models, but one is a weight-shape mismatch and the other is missing projections in a different model; not the same bug." + }, + { + "left": "issue:44360", + "right": "issue:46172", + "accept": false, + "reason": "DSA ReLU concern is unrelated to the DeepSeek comb transpose issue." + }, + { + "left": "issue:43441", + "right": "issue:44829", + "accept": false, + "reason": "Both mention FlashAttention, but one is model support/loading and the other is degenerate training with flash_attention_3; different problems." + }, + { + "left": "issue:43931", + "right": "issue:44038", + "accept": false, + "reason": "Same general model family, but different concrete breakages: Qwen3-VL weight mismatch vs Qwen3-VL-MoE import/load failure." + }, + { + "left": "issue:43097", + "right": "issue:43452", + "accept": false, + "reason": "Different regressions: a removed config field vs gguf_file handling in tokenizer/model loading." + }, + { + "left": "issue:43335", + "right": "issue:45071", + "accept": false, + "reason": "SwitchTransformers layer creation bug is unrelated to PreTrainedConfig type checking." + }, + { + "left": "issue:45003", + "right": "issue:45341", + "accept": false, + "reason": "Unsafe sys.modules access is unrelated to the testing_utils bug." + }, + { + "left": "issue:43673", + "right": "issue:46123", + "accept": false, + "reason": "Generation cache handling and MaskGenerationPipeline final-batch handling are different code paths." + }, + { + "left": "issue:43986", + "right": "issue:45341", + "accept": false, + "reason": "Video AutoProcessor/torchvision crash is unrelated to the testing_utils issue." + }, + { + "left": "issue:44933", + "right": "issue:46231", + "accept": false, + "reason": "Nonexistent import from image_utils is a code bug; set_training docs parameter mismatch is documentation-only." + }, + { + "left": "issue:45753", + "right": "issue:46369", + "accept": false, + "reason": "Different multimodal failures with different models and symptoms; no shared underlying bug." + }, + { + "left": "issue:44117", + "right": "issue:44987", + "accept": false, + "reason": "Tokenizer mapping None-handling is a different failure than loading physical-intelligence/fast." + }, + { + "left": "issue:43986", + "right": "issue:44484", + "accept": false, + "reason": "Completely different topics: missing torchvision vs shard-size behavior." + }, + { + "left": "issue:43874", + "right": "issue:44263", + "accept": false, + "reason": "Unrelated GLM46V image patch API bug vs GlmMoeDsaIndexer split semantics." + }, + { + "left": "issue:43986", + "right": "issue:45372", + "accept": false, + "reason": "Missing torchvision crash is unrelated to the mistral_common ReasoningEffort import error." + }, + { + "left": "issue:43531", + "right": "issue:45341", + "accept": false, + "reason": "Qwen3-MoE sliding_window behavior has no connection to the testing_utils bug." + }, + { + "left": "issue:45440", + "right": "issue:46172", + "accept": false, + "reason": "DeepSeekV3MoE divergence from remote implementation is a broader model parity issue, not the same as the comb transpose bug." + }, + { + "left": "issue:43097", + "right": "issue:46050", + "accept": false, + "reason": "Both are save/load related, but one is an API removal and the other is a quantized model persistence regression." + }, + { + "left": "issue:43097", + "right": "issue:45588", + "accept": false, + "reason": "Removed tie_embeddings API and flash_attention s_aux=None crash are unrelated bugs." + }, + { + "left": "issue:44261", + "right": "issue:44485", + "accept": false, + "reason": "RMS norm epsilon precision discussion and GLM-5 RoPE implementation are different model math issues with no clear shared defect." + }, + { + "left": "issue:38617", + "right": "issue:42757", + "accept": false, + "reason": "Both are import errors, but they involve different missing symbols from different modules and do not look like the same bug." + }, + { + "left": "issue:43064", + "right": "issue:44315", + "accept": false, + "reason": "Optimizer-state corruption under FSDP2/PEFT is unrelated to Liger Kernel not being applied with model_init." + }, + { + "left": "issue:44062", + "right": "issue:44117", + "accept": false, + "reason": "AddedToken argument duplication and tokenizer mapping None are distinct tokenizer-layer bugs." + }, + { + "left": "issue:44410", + "right": "issue:45375", + "accept": false, + "reason": "Missing attn projections in qwen3next and dropped deepstack_visual_indexes in Qwen3_5MoeVisionConfig are different config/model issues." + }, + { + "left": "issue:43502", + "right": "issue:43992", + "accept": false, + "reason": "local_files_only network leakage and missing embed_tokens.weight are unrelated loading bugs." + }, + { + "left": "issue:44038", + "right": "issue:44410", + "accept": false, + "reason": "Both touch Qwen3 variants, but one is a Qwen3-VL-Moe from_pretrained issue and the other is missing projections in qwen3next; not the same code path." + }, + { + "left": "issue:42890", + "right": "issue:46298", + "accept": false, + "reason": "Test seeding failures and noisy logging are unrelated." + }, + { + "left": "issue:38617", + "right": "issue:44351", + "accept": false, + "reason": "Different missing imports from different APIs; same symptom type, not the same underlying bug." + }, + { + "left": "issue:43064", + "right": "issue:43873", + "accept": false, + "reason": "FSDP2/PEFT optimizer-state corruption and quantization offloading are different training-system 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": 19, + "review_unit_index": 48, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:36683", + "issue:38617", + "issue:42490", + "issue:42757", + "issue:42907", + "issue:43012", + "issue:43064", + "issue:43066", + "issue:43097", + "issue:43208", + "issue:43352", + "issue:43502", + "issue:43526", + "issue:43550", + "issue:43582", + "issue:43824", + "issue:43825", + "issue:43906", + "issue:43976", + "issue:43986", + "issue:43992", + "issue:44038", + "issue:44117", + "issue:44162", + "issue:44222", + "issue:44230", + "issue:44368", + "issue:44380", + "issue:44483", + "issue:44484", + "issue:44514", + "issue:44655", + "issue:44683", + "issue:44704", + "issue:44936", + "issue:44938", + "issue:45103", + "issue:45478", + "issue:45479", + "issue:45600", + "issue:45712", + "issue:45753", + "issue:46132", + "issue:46172", + "issue:46178", + "issue:46298", + "issue:46365", + "issue:46421" + ], + "soft_pairs": [ + "issue:44704|issue:45103", + "issue:44380|issue:46421", + "issue:43986|issue:44117", + "issue:43824|issue:43986", + "issue:44938|issue:46178", + "issue:43066|issue:43906", + "issue:43208|issue:43550", + "issue:43992|issue:44484", + "issue:43012|issue:43550", + "issue:43012|issue:44368", + "issue:43825|issue:44655", + "issue:45712|issue:46298", + "issue:42907|issue:44655", + "issue:42907|issue:45753", + "issue:43097|issue:46298", + "issue:43352|issue:44380", + "issue:43352|issue:43976", + "issue:44162|issue:46172", + "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:44936|issue:46365", + "issue:42490|issue:43012", + "issue:43012|issue:44683", + "issue:42757|issue:43824", + "issue:38617|issue:43502", + "issue:45600|issue:46298", + "issue:44038|issue:44230" + ], + "prepared_review_unit_hash": "f478899ba0bb9a519c3823b9164d8a5cf1857ac2711d799affcdf3e77ebd17b3", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 37089, + "estimated_input_tokens": 9273, + "estimated_eval_tokens": 18802 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "No soft pair looks like the same underlying bug/change, so I would not merge any of these issues as duplicates. The cluster is mostly a mix of unrelated import errors, model-specific regressions, and a few attention/compile/config bugs.", + "confidence": 0.84, + "canonical_issue_reason": "Issue 45103 is the cleanest standalone bug report: it has a precise crash condition, a clear subsystem, and a concrete reproduction path, making it the best canonical issue among this set.", + "canonical_pr_reason": null, + "best_issue_reason": "45103 is the most suitable representative issue because it is narrowly scoped, actionable, and currently open, while the rest are largely unrelated single-bug reports.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44704", + "right": "issue:45103", + "accept": false, + "reason": "Different subsystems and failures: AutoProcessor kwargs propagation vs auto_docstring crash with future annotations." + }, + { + "left": "issue:44380", + "right": "issue:46421", + "accept": false, + "reason": "One is GPT2 attention scaling ignored in SDPA/FlashAttention; the other is lazy cache init causing recompilation with chunked prefill." + }, + { + "left": "issue:43986", + "right": "issue:44117", + "accept": false, + "reason": "Video AutoProcessor/torchvision import crash is unrelated to TOKENIZER_MAPPING_NAMES returning None." + }, + { + "left": "issue:43824", + "right": "issue:43986", + "accept": false, + "reason": "Qwen2_5_VL class import failure is distinct from AutoProcessor failing without torchvision." + }, + { + "left": "issue:44938", + "right": "issue:46178", + "accept": false, + "reason": "Python 3.14 import/load failure is unrelated to a backend mapping error for tensorflow_text." + }, + { + "left": "issue:43066", + "right": "issue:43906", + "accept": false, + "reason": "Wrong tokenizer decoder type is not the same as an isolated reproduction of another issue." + }, + { + "left": "issue:43208", + "right": "issue:43550", + "accept": false, + "reason": "xLSTM training blockers and Bamba SDPA compile failure are different models and code paths." + }, + { + "left": "issue:43992", + "right": "issue:44484", + "accept": false, + "reason": "Missing embed_tokens weight on UMT5Encoder is unrelated to max_shard_size behavior in save_pretrained()." + }, + { + "left": "issue:43012", + "right": "issue:43550", + "accept": false, + "reason": "A PyTorch precision warning during compile is not the same bug as Bamba SDPA compile failure." + }, + { + "left": "issue:43012", + "right": "issue:44368", + "accept": false, + "reason": "Float32 warning on compile and tie_word_embeddings warning are unrelated configuration/runtime behaviors." + }, + { + "left": "issue:43825", + "right": "issue:44655", + "accept": false, + "reason": "Translation-task pipeline error is unrelated to saving Pipeline objects with save_pretrained()." + }, + { + "left": "issue:45712", + "right": "issue:46298", + "accept": false, + "reason": "Leftover dummy classes failing repo checks is not the same as a noisy logging message." + }, + { + "left": "issue:42907", + "right": "issue:44655", + "accept": false, + "reason": "Dequantized model save failure is different from generic pipeline serialization failure." + }, + { + "left": "issue:42907", + "right": "issue:45753", + "accept": false, + "reason": "Model save failure and Qwen3_5 image infinite loop are unrelated." + }, + { + "left": "issue:43097", + "right": "issue:46298", + "accept": false, + "reason": "Removed tie_embeddings_and_encoder_decoder is a behavior/regression issue, not a logging-noise issue." + }, + { + "left": "issue:43352", + "right": "issue:44380", + "accept": false, + "reason": "Unsupported Flash Attention 2.0 in NemotronH is unrelated to GPT2 attention scaling being ignored." + }, + { + "left": "issue:43352", + "right": "issue:43976", + "accept": false, + "reason": "Flash Attention support in a model is unrelated to a Python version compatibility problem." + }, + { + "left": "issue:44162", + "right": "issue:46172", + "accept": false, + "reason": "ESM2 breakage and a DeepSeek tensor transpose issue do not share the same failure mode." + }, + { + "left": "issue:44380", + "right": "issue:45479", + "accept": false, + "reason": "Attention scaling config loss is unrelated to a degenerate zero-loss classification bug." + }, + { + "left": "issue:44380", + "right": "issue:46132", + "accept": false, + "reason": "GPT2 backend scaling bug is distinct from AttentionInterface.register changing function behavior." + }, + { + "left": "issue:43064", + "right": "issue:44380", + "accept": false, + "reason": "FSDP2/PEFT optimizer-state corruption is unrelated to GPT2 attention scaling being ignored." + }, + { + "left": "issue:44483", + "right": "issue:44514", + "accept": false, + "reason": "Chat completions API rejection and Qwen2_5_VLProcessor batch-padding crash are different endpoints/code paths." + }, + { + "left": "issue:44222", + "right": "issue:45478", + "accept": false, + "reason": "FP8 save_pretrained behavior is different from a Qwen3.5 MoE from_pretrained error." + }, + { + "left": "issue:36683", + "right": "issue:43582", + "accept": false, + "reason": "Gemma3Config missing vocab_size and Apple Silicon caching_allocator_warmup TypeError are unrelated." + }, + { + "left": "issue:43208", + "right": "issue:43526", + "accept": false, + "reason": "xLSTM training bugs do not match BeitImageProcessorFast reduce_labels behavior." + }, + { + "left": "issue:44936", + "right": "issue:46365", + "accept": false, + "reason": "trainer.evaluate() failing after train is not the same as failing to import Trainer until datasets is uninstalled." + }, + { + "left": "issue:42490", + "right": "issue:43012", + "accept": false, + "reason": "Trailing-slash local model import warning is unrelated to a bfloat16 compile warning." + }, + { + "left": "issue:43012", + "right": "issue:44683", + "accept": false, + "reason": "General compile warning and compiled flex_attention failure on newer torch versions are different bugs." + }, + { + "left": "issue:42757", + "right": "issue:43824", + "accept": false, + "reason": "Missing is_offline_mode import from huggingface_hub is unrelated to missing Qwen2_5_VLForConditionalGeneration export." + }, + { + "left": "issue:38617", + "right": "issue:43502", + "accept": false, + "reason": "layer_type_validation import breakage is unrelated to local_files_only still allowing API requests." + }, + { + "left": "issue:45600", + "right": "issue:46298", + "accept": false, + "reason": "Broken auto_mappings references and a noisy logging message are not the same issue." + }, + { + "left": "issue:44038", + "right": "issue:44230", + "accept": false, + "reason": "Generic Qwen3-VL-Moe breakage and fp8 inference support are related in model family only, not the same concrete bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 49, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:36010", + "issue:36683", + "issue:42175", + "issue:42757", + "issue:42994", + "issue:43010", + "issue:43064", + "issue:43066", + "issue:43299", + "issue:43352", + "issue:43475", + "issue:43525", + "issue:43611", + "issue:43618", + "issue:43756", + "issue:43761", + "issue:43867", + "issue:43901", + "issue:43937", + "issue:43976", + "issue:43992", + "issue:44075", + "issue:44077", + "issue:44242", + "issue:44355", + "issue:44380", + "issue:44483", + "issue:44485", + "issue:44492", + "issue:44625", + "issue:44655", + "issue:44855", + "issue:44938", + "issue:45216", + "issue:45276", + "issue:45278", + "issue:45290", + "issue:45310", + "issue:45375", + "issue:45584", + "issue:45593", + "issue:45698", + "issue:46077", + "issue:46082", + "issue:46172", + "issue:46202", + "issue:46209", + "issue:46365" + ], + "soft_pairs": [ + "issue:45698|issue:46172", + "issue:43618|issue:44380", + "issue:43618|issue:43761", + "issue:44483|issue:45290", + "issue:43299|issue:43867", + "issue:36683|issue:43475", + "issue:42994|issue:44655", + "issue:43066|issue:46209", + "issue:43756|issue:44485", + "issue:45276|issue:46077", + "issue:43901|issue:44655", + "issue:36010|issue:42757", + "issue:43611|issue:44655", + "issue:43352|issue:43937", + "issue:36683|issue:44077", + "issue:43867|issue:45216", + "issue:44625|issue:46077", + "issue:43976|issue:44938", + "issue:45584|issue:46365", + "issue:42175|issue:46365", + "issue:43352|issue:45593", + "issue:46077|issue:46082", + "issue:36683|issue:43525", + "issue:43618|issue:45216", + "issue:44077|issue:46365", + "issue:36683|issue:44242", + "issue:44075|issue:46365", + "issue:43992|issue:45310", + "issue:43064|issue:44492", + "issue:43010|issue:45278", + "issue:44355|issue:44855", + "issue:43867|issue:46202", + "issue:43867|issue:45375" + ], + "prepared_review_unit_hash": "79ce4c8f80d2e8929e55232626f12177563a029b38a6b0a1861fe26298e59010", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 33, + "serialized_chars": 37164, + "estimated_input_tokens": 9291, + "estimated_eval_tokens": 18838 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is a loose mix of Transformers v5 regressions and unrelated import/load/save bugs. None of the soft pairs are the same concrete defect, so they should stay separate rather than be collapsed into one duplicate set.", + "confidence": 0.92, + "canonical_issue_reason": "No true canonical issue: the items are heterogeneous. If forced to pick one umbrella, issue 45278 is the broadest import-regression bucket and is the most reusable triage anchor among the set.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45278 is the best representative only as a broad umbrella for import failures after upgrading; the rest of the cluster is too diverse to treat as one bug.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45698", + "right": "issue:46172", + "accept": false, + "reason": "Save/load custom-module corruption and a deepseek-v4 tensor transpose bug are different code paths and symptoms." + }, + { + "left": "issue:43618", + "right": "issue:44380", + "accept": false, + "reason": "Missing CLIP attentions and ignored GPT2 attention scaling under SDPA/FlashAttention are unrelated regressions." + }, + { + "left": "issue:43618", + "right": "issue:43761", + "accept": false, + "reason": "Both involve CLIP/v5 output behavior, but one drops attentions while the other drops hidden_states; different concrete defects." + }, + { + "left": "issue:44483", + "right": "issue:45290", + "accept": false, + "reason": "An API request rejection and a chat-template crash on tool-call messages are different layers and bugs." + }, + { + "left": "issue:43299", + "right": "issue:43867", + "accept": false, + "reason": "Qwen3VL MoE loading breakage and a state_dict sorting load error are not the same failure mode." + }, + { + "left": "issue:36683", + "right": "issue:43475", + "accept": false, + "reason": "Missing Gemma3Config vocab_size and missing Sam3VisionEncoderOutput fpn_position_embeddings are unrelated attribute errors." + }, + { + "left": "issue:42994", + "right": "issue:44655", + "accept": false, + "reason": "Quantized model saving failure and Pipeline save_pretrained failure are both save-related but not the same bug." + }, + { + "left": "issue:43066", + "right": "issue:46209", + "accept": false, + "reason": "Wrong tokenizer decoder type and FP8 inference failure on B200 are unrelated issues." + }, + { + "left": "issue:43756", + "right": "issue:44485", + "accept": false, + "reason": "Different RoPE-related model bugs in different models; not the same concrete implementation defect." + }, + { + "left": "issue:45276", + "right": "issue:46077", + "accept": false, + "reason": "Gemma4 embedding resize not propagating and Gemma4 attention-mask construction are distinct code paths." + }, + { + "left": "issue:43901", + "right": "issue:44655", + "accept": false, + "reason": "Docs wording for return_all_scores and Pipeline serialization are unrelated." + }, + { + "left": "issue:36010", + "right": "issue:42757", + "accept": false, + "reason": "Both are import errors, but they concern different symbols and different dependency breakages." + }, + { + "left": "issue:43611", + "right": "issue:44655", + "accept": false, + "reason": "Base_model_prefix loading regression and Pipeline save_pretrained failure are not the same bug." + }, + { + "left": "issue:43352", + "right": "issue:43937", + "accept": false, + "reason": "FlashAttention support limitation and invalid GenerationConfig are unrelated model/config failures." + }, + { + "left": "issue:36683", + "right": "issue:44077", + "accept": false, + "reason": "A missing config attribute and a deprecation/allowlist issue in patchtsmixer are different bugs." + }, + { + "left": "issue:43867", + "right": "issue:45216", + "accept": false, + "reason": "State_dict ordering causing load failure is not the same as an incorrect checkpoint produced by save_pretrained." + }, + { + "left": "issue:44625", + "right": "issue:46077", + "accept": false, + "reason": "Qwen3.5 num_labels propagation and Gemma4 bidirectional-attention masking are unrelated." + }, + { + "left": "issue:43976", + "right": "issue:44938", + "accept": false, + "reason": "These are both version-compatibility problems, but for different Python versions and different failures." + }, + { + "left": "issue:45584", + "right": "issue:46365", + "accept": false, + "reason": "Whisper empty-transcription generation and a Trainer import conflict with datasets are unrelated." + }, + { + "left": "issue:42175", + "right": "issue:46365", + "accept": false, + "reason": "Missing TensorFlow backend in an install extra is unrelated to the Trainer import conflict." + }, + { + "left": "issue:43352", + "right": "issue:45593", + "accept": false, + "reason": "No FlashAttention 2 support and missing auxiliary losses are unrelated model bugs." + }, + { + "left": "issue:46077", + "right": "issue:46082", + "accept": false, + "reason": "Gemma4 causal-mask behavior and LlamaConfig head_dim validation are different configuration/architecture issues." + }, + { + "left": "issue:36683", + "right": "issue:43525", + "accept": false, + "reason": "Two missing config attributes on different model families are similar in shape but not the same underlying bug." + }, + { + "left": "issue:43618", + "right": "issue:45216", + "accept": false, + "reason": "Missing CLIP attentions and incorrect Qwen3.5 saved checkpoints are unrelated regressions." + }, + { + "left": "issue:44077", + "right": "issue:46365", + "accept": false, + "reason": "patchtsmixer post_init policy and the datasets-induced Trainer import error are unrelated." + }, + { + "left": "issue:36683", + "right": "issue:44242", + "accept": false, + "reason": "A missing Gemma3 config attribute and a missing load-balancing loss are different defects." + }, + { + "left": "issue:44075", + "right": "issue:46365", + "accept": false, + "reason": "SGD arguments being ignored and a Trainer import conflict are unrelated." + }, + { + "left": "issue:43992", + "right": "issue:45310", + "accept": false, + "reason": "UMT5Encoder missing embed_tokens.weight and Qwen3.5 MoE from_pretrained failure are different loading bugs." + }, + { + "left": "issue:43064", + "right": "issue:44492", + "accept": false, + "reason": "A distributed optimizer-state bug and a typo in cache strategies are not the same issue." + }, + { + "left": "issue:43010", + "right": "issue:45278", + "accept": false, + "reason": "Cache update no_grad behavior and broad import errors after upgrade do not share the same code path." + }, + { + "left": "issue:44355", + "right": "issue:44855", + "accept": false, + "reason": "Compiled-file execution errors and a Python 3.13 JIT parsing/indentation problem are distinct." + }, + { + "left": "issue:43867", + "right": "issue:46202", + "accept": false, + "reason": "State_dict-sorting load failure and Gemma pipeline roundtrip failure on \u2581 are different serialization bugs." + }, + { + "left": "issue:43867", + "right": "issue:45375", + "accept": false, + "reason": "Load failure from sorted state_dict and a silently dropped config field 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": 19, + "review_unit_index": 50, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:28282", + "issue:30333", + "issue:33453", + "issue:36296", + "issue:36683", + "issue:41669", + "issue:42548", + "issue:42890", + "issue:43010", + "issue:43064", + "issue:43065", + "issue:43232", + "issue:43502", + "issue:43575", + "issue:43576", + "issue:43618", + "issue:43723", + "issue:43867", + "issue:43976", + "issue:43992", + "issue:44062", + "issue:44077", + "issue:44112", + "issue:44222", + "issue:44246", + "issue:44351", + "issue:44355", + "issue:44360", + "issue:44387", + "issue:44485", + "issue:44556", + "issue:44625", + "issue:44749", + "issue:44857", + "issue:44938", + "issue:44964", + "issue:45071", + "issue:45125", + "issue:45357", + "issue:45375", + "issue:45440", + "issue:45478", + "issue:45561", + "issue:45600", + "issue:45850", + "issue:46132", + "issue:46285", + "issue:46365" + ], + "soft_pairs": [ + "issue:41669|issue:44246", + "issue:36683|issue:43064", + "issue:44938|issue:45071", + "issue:42890|issue:45561", + "issue:43010|issue:45850", + "issue:44625|issue:46365", + "issue:43992|issue:45478", + "issue:43867|issue:45357", + "issue:43010|issue:43976", + "issue:36296|issue:43575", + "issue:43618|issue:44077", + "issue:28282|issue:33453", + "issue:43618|issue:46285", + "issue:43618|issue:44857", + "issue:44360|issue:45440", + "issue:43502|issue:44062", + "issue:42890|issue:44964", + "issue:42890|issue:44112", + "issue:42890|issue:45600", + "issue:43723|issue:44556", + "issue:44387|issue:44556", + "issue:44222|issue:45125", + "issue:43010|issue:43065", + "issue:43576|issue:46132", + "issue:43502|issue:44485", + "issue:42548|issue:44351", + "issue:45125|issue:45478", + "issue:44355|issue:44749", + "issue:30333|issue:33453", + "issue:45125|issue:45375", + "issue:43010|issue:43232", + "issue:43502|issue:44222" + ], + "prepared_review_unit_hash": "dc5256ed377d879792e3997c42dece091188d2f610b88d6750e03b21b525b25f", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 36840, + "estimated_input_tokens": 9210, + "estimated_eval_tokens": 18676 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "Mostly a heterogeneous mix of unrelated Transformers issues. The only plausible duplicate thread is import-time slowdown/import-surface regression (41669 and 44246); the rest do not look like the same underlying bug.", + "confidence": 0.68, + "canonical_issue_reason": "44246 is the best available anchor because it captures the broad user-facing symptom in the only plausible shared theme (slow imports), though the cluster is not truly coherent.", + "canonical_pr_reason": null, + "best_issue_reason": "44246 is the most representative issue overall: it is broad, user-facing, and closest to the shared import-performance theme; the other issues are mostly unrelated point bugs.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:41669", + "right": "issue:44246", + "accept": true, + "reason": "Same import-performance regression theme: 41669 gives a likely root cause (`import *` in models), while 44246 reports the user-visible slowdown." + }, + { + "left": "issue:36683", + "right": "issue:43064", + "accept": false, + "reason": "Different failures: missing `vocab_size` in Gemma3 config vs distributed optimizer-state corruption in FSDP2/PEFT." + }, + { + "left": "issue:44938", + "right": "issue:45071", + "accept": false, + "reason": "Both are version-regression reports, but one is Python 3.14 import failure and the other is `PretrainedConfig` type-checking breakage; different bugs." + }, + { + "left": "issue:42890", + "right": "issue:45561", + "accept": false, + "reason": "Both are flaky test issues, but one is missing `set_seed()` in SamHQ tests and the other is an xdist race on `captured_info.txt`." + }, + { + "left": "issue:43010", + "right": "issue:45850", + "accept": false, + "reason": "`@torch.no_grad` on cache/layer updates is a specific API bug; 45850 is a broad complaint about many models breaking in v5." + }, + { + "left": "issue:44625", + "right": "issue:46365", + "accept": false, + "reason": "Unrelated: Qwen3.5 `num_labels` propagation vs inability to import `Trainer` until uninstalling `datasets`." + }, + { + "left": "issue:43992", + "right": "issue:45478", + "accept": false, + "reason": "Both involve `from_pretrained`, but the concrete problems differ: missing `embed_tokens.weight` vs Qwen3.5 MoE load error." + }, + { + "left": "issue:43867", + "right": "issue:45357", + "accept": false, + "reason": "Different save/load paths and model families: sorted state_dict loading error vs incorrect visual-encoder keys on save." + }, + { + "left": "issue:43010", + "right": "issue:43976", + "accept": false, + "reason": "No relation beyond being v5-related regressions; one is cache/layer update semantics, the other is a Python-version compatibility failure." + }, + { + "left": "issue:36296", + "right": "issue:43575", + "accept": false, + "reason": "Tensor-parallel training bug vs OOM while loading a specific Qwen2 model with TP; related area, but not the same concrete bug." + }, + { + "left": "issue:43618", + "right": "issue:44077", + "accept": false, + "reason": "Unrelated model/API issues: CLIP output attentions assignment vs patchtsmixer `post_init` allowance." + }, + { + "left": "issue:28282", + "right": "issue:33453", + "accept": false, + "reason": "Different subsystems and symptoms: missing PyTorch import for AutoModel vs tokenizer-loading regression." + }, + { + "left": "issue:43618", + "right": "issue:46285", + "accept": false, + "reason": "Both touch CLIP, but one is missing attentions in output and the other is an inconsistent internal structure change; not the same bug." + }, + { + "left": "issue:43618", + "right": "issue:44857", + "accept": false, + "reason": "CLIP attentions regression and LwDetr float16 AMP crash are unrelated." + }, + { + "left": "issue:44360", + "right": "issue:45440", + "accept": false, + "reason": "No shared concrete code-path: DSA indexer activation issue vs DeepseekV3MoE divergence from remote implementation." + }, + { + "left": "issue:43502", + "right": "issue:44062", + "accept": false, + "reason": "Different bug classes: local_files_only still making network requests vs AddedToken constructor argument collision." + }, + { + "left": "issue:42890", + "right": "issue:44964", + "accept": false, + "reason": "Test flakiness in SamHQ vs failure loading a Phi multimodal model; not the same underlying issue." + }, + { + "left": "issue:42890", + "right": "issue:44112", + "accept": false, + "reason": "Both are CI/test-related, but one is missing seeding and the other is a stale device override in GraniteSpeech." + }, + { + "left": "issue:42890", + "right": "issue:45600", + "accept": false, + "reason": "Unrelated test flake vs auto_mappings.py referencing removed Sam3LiteText configs." + }, + { + "left": "issue:43723", + "right": "issue:44556", + "accept": false, + "reason": "Both are loading regressions, but one is tokenizer loading in v5 and the other is checkpoint reload incompatibility after upgrading from v4.57." + }, + { + "left": "issue:44387", + "right": "issue:44556", + "accept": false, + "reason": "OOM under int4 quantization vs reload failure after version upgrade; different mechanisms and fixes." + }, + { + "left": "issue:44222", + "right": "issue:45125", + "accept": false, + "reason": "Different Qwen3.5 MoE issues: FP8 save_pretrained bug vs missing tensor-parallel plan." + }, + { + "left": "issue:43010", + "right": "issue:43065", + "accept": false, + "reason": "No relation: cache/layer no_grad decoration vs dummy Conv2d in Sam3PixelDecoder." + }, + { + "left": "issue:43576", + "right": "issue:46132", + "accept": false, + "reason": "`transformers env` command breakage is unrelated to AttentionInterface.register behavior changes." + }, + { + "left": "issue:43502", + "right": "issue:44485", + "accept": false, + "reason": "Local-files-only network bug vs GLM-5 RoPE implementation discussion; different problems." + }, + { + "left": "issue:42548", + "right": "issue:44351", + "accept": false, + "reason": "Both are import-name errors from `transformers`, but they concern different missing symbols and do not look like the same concrete defect." + }, + { + "left": "issue:45125", + "right": "issue:45478", + "accept": false, + "reason": "Both concern Qwen3.5 MoE, but one is missing `_tp_plan` and the other is a `from_pretrained` error in newer transformers." + }, + { + "left": "issue:44355", + "right": "issue:44749", + "accept": false, + "reason": "Compiled-Python-file errors vs performance slowdown after upgrading from 4.57.3 to 5.3.0 are unrelated." + }, + { + "left": "issue:30333", + "right": "issue:33453", + "accept": false, + "reason": "MLFlow reporting on failed training jobs has nothing to do with tokenizer loading regression." + }, + { + "left": "issue:45125", + "right": "issue:45375", + "accept": false, + "reason": "Both are Qwen3.5-related, but one is tensor parallelism setup and the other is a missing vision-config field silently dropped by strict validation." + }, + { + "left": "issue:43010", + "right": "issue:43232", + "accept": false, + "reason": "No shared bug: cache/layer update decoration vs generation state sync after `sync_gpus`." + }, + { + "left": "issue:43502", + "right": "issue:44222", + "accept": false, + "reason": "Different issues: unwanted API requests during offline loading vs FP8 MoE save_pretrained behavior." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative. It correctly treats 44246 as the best anchor for the only clearly related thread (import-time slowdown, with 41669), and the pair verdicts avoid overstating weaker similarities. No mergeability issue since there are no PR pairs." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 51, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:27957", + "issue:30333", + "issue:30990", + "issue:31795", + "issue:33453", + "issue:34689", + "issue:36296", + "issue:38617", + "issue:41628", + "issue:42548", + "issue:42757", + "issue:42831", + "issue:43299", + "issue:43502", + "issue:43526", + "issue:43576", + "issue:43673", + "issue:43824", + "issue:43827", + "issue:43874", + "issue:43901", + "issue:44038", + "issue:44222", + "issue:44230", + "issue:44273", + "issue:44315", + "issue:44355", + "issue:44360", + "issue:44393", + "issue:44485", + "issue:44492", + "issue:44556", + "issue:44623", + "issue:44655", + "issue:44908", + "issue:44938", + "issue:44961", + "issue:45071", + "issue:45125", + "issue:45230", + "issue:45362", + "issue:45468", + "issue:45542", + "issue:45850", + "issue:46095", + "issue:46143", + "issue:46144", + "issue:46209" + ], + "soft_pairs": [ + "issue:30990|issue:31795", + "issue:43299|issue:44556", + "issue:43502|issue:44961", + "issue:42831|issue:43874", + "issue:30333|issue:36296", + "issue:44355|issue:45071", + "issue:43827|issue:44961", + "issue:44355|issue:44655", + "issue:44315|issue:44360", + "issue:33453|issue:34689", + "issue:27957|issue:33453", + "issue:42757|issue:43576", + "issue:43576|issue:45071", + "issue:43576|issue:43824", + "issue:31795|issue:38617", + "issue:44961|issue:45468", + "issue:44222|issue:44230", + "issue:43576|issue:44938", + "issue:41628|issue:42548", + "issue:44908|issue:44961", + "issue:44485|issue:44961", + "issue:46143|issue:46144", + "issue:36296|issue:43901", + "issue:44315|issue:46209", + "issue:36296|issue:43526", + "issue:43673|issue:44492", + "issue:45362|issue:45850", + "issue:44623|issue:45230", + "issue:44273|issue:45542", + "issue:44038|issue:45125", + "issue:44230|issue:46095", + "issue:44393|issue:45071" + ], + "prepared_review_unit_hash": "ebd052d6e9a63d460f3df0529544a4047502c3fcc16261c22f9d2c3e942e584e", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 32, + "serialized_chars": 36415, + "estimated_input_tokens": 9104, + "estimated_eval_tokens": 18464 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The set is heterogeneous: a few pairs share broad themes like v5 regressions, loading failures, or import errors, but they describe different concrete bugs and should not be deduplicated. No pull requests are present.", + "confidence": 0.86, + "canonical_issue_reason": "No single issue is a strong canonical duplicate target here; the items span unrelated problems across loading, docs, FP8, import/export, CLI, and tensor-parallel behavior.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45850 is the broadest umbrella-style report, but it is too vague to be a reliable dedup target.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:30990", + "right": "issue:31795", + "accept": false, + "reason": "One is a Sentence Transformers load hang; the other is a docs wording complaint for model.forward." + }, + { + "left": "issue:43299", + "right": "issue:44556", + "accept": false, + "reason": "Both involve version-related loading regressions, but for different models and likely different failure modes." + }, + { + "left": "issue:43502", + "right": "issue:44961", + "accept": false, + "reason": "local_files_only API leakage is unrelated to the unrelated 'racoon' report." + }, + { + "left": "issue:42831", + "right": "issue:43874", + "accept": false, + "reason": "FP8 accuracy regression and a missing image-processor method are different bugs." + }, + { + "left": "issue:30333", + "right": "issue:36296", + "accept": false, + "reason": "MLFlow reporting on failed jobs is unrelated to tensor-parallel training." + }, + { + "left": "issue:44355", + "right": "issue:45071", + "accept": false, + "reason": "Compiled-file loading errors are not the same as PretrainedConfig type-checking breakage." + }, + { + "left": "issue:43827", + "right": "issue:44961", + "accept": false, + "reason": "Documentation references to removed pipeline() APIs are unrelated to 'racoon'." + }, + { + "left": "issue:44355", + "right": "issue:44655", + "accept": false, + "reason": "Compiled Python file errors and save_pretrained pipeline issues affect different code paths." + }, + { + "left": "issue:44315", + "right": "issue:44360", + "accept": false, + "reason": "Liger Kernel not applied with model_init is unrelated to the DSA/ReLU discussion." + }, + { + "left": "issue:33453", + "right": "issue:34689", + "accept": false, + "reason": "Tokenizer loading regression and Llama 3.2 vision model loading failure are different concrete load bugs." + }, + { + "left": "issue:27957", + "right": "issue:33453", + "accept": false, + "reason": "Flash Attention 2 support for XLM-Roberta is unrelated to tokenizer loading regression." + }, + { + "left": "issue:42757", + "right": "issue:43576", + "accept": false, + "reason": "Missing huggingface_hub import and the v5 env CLI breakage are not the same issue." + }, + { + "left": "issue:43576", + "right": "issue:45071", + "accept": false, + "reason": "Broken env command and PretrainedConfig type checking are separate regressions." + }, + { + "left": "issue:43576", + "right": "issue:43824", + "accept": false, + "reason": "CLI env failure is unrelated to the missing Qwen2_5_VL export." + }, + { + "left": "issue:31795", + "right": "issue:38617", + "accept": false, + "reason": "A docs confusion issue does not match an ImportError for layer_type_validation." + }, + { + "left": "issue:44961", + "right": "issue:45468", + "accept": false, + "reason": "The placeholder 'racoon' report does not match the Gemma-4 audio positional encoding bug." + }, + { + "left": "issue:44222", + "right": "issue:44230", + "accept": false, + "reason": "FP8 save_pretrained and FP8 inference support are different failures, not one code-path bug." + }, + { + "left": "issue:43576", + "right": "issue:44938", + "accept": false, + "reason": "Broken transformers env CLI and Python 3.14 import/load failure are distinct problems." + }, + { + "left": "issue:41628", + "right": "issue:42548", + "accept": false, + "reason": "Different missing exports: AutoImageProcessor versus PreTrainedModel." + }, + { + "left": "issue:44908", + "right": "issue:44961", + "accept": false, + "reason": "Scheduler kwargs handling is unrelated to the 'racoon' issue." + }, + { + "left": "issue:44485", + "right": "issue:44961", + "accept": false, + "reason": "GLM-5 RoPE implementation discussion is unrelated to the 'racoon' report." + }, + { + "left": "issue:46143", + "right": "issue:46144", + "accept": false, + "reason": "Both concern RoFormer, but one is about kwargs propagation and the other about attention-interface usage; not the same concrete bug." + }, + { + "left": "issue:36296", + "right": "issue:43901", + "accept": false, + "reason": "Tensor-parallel training and pipeline docs behavior are unrelated." + }, + { + "left": "issue:44315", + "right": "issue:46209", + "accept": false, + "reason": "Liger Kernel application in model_init is unrelated to FP8 inference on B200." + }, + { + "left": "issue:36296", + "right": "issue:43526", + "accept": false, + "reason": "Tensor-parallel training bug and BeitImageProcessorFast label reduction are different code paths." + }, + { + "left": "issue:43673", + "right": "issue:44492", + "accept": false, + "reason": "Generation cache missing during chunked_prefill is not a typo in cache strategy docs." + }, + { + "left": "issue:45362", + "right": "issue:45850", + "accept": false, + "reason": "A specific Qwen3.5-35B chat crash is not the same bug as a broad 'many models break in v5' complaint." + }, + { + "left": "issue:44623", + "right": "issue:45230", + "accept": false, + "reason": "Missing files in processor.save_pretrained is unrelated to the placeholder bug report." + }, + { + "left": "issue:44273", + "right": "issue:45542", + "accept": false, + "reason": "Lazy loading failures and the TensorFlow/tensorboard backend error are different issues." + }, + { + "left": "issue:44038", + "right": "issue:45125", + "accept": false, + "reason": "Qwen3-VL MoE breakage and missing _tp_plan for tensor parallelism are separate model-support gaps." + }, + { + "left": "issue:44230", + "right": "issue:46095", + "accept": false, + "reason": "FP8 inference support and a weight-conversion-script question are not the same underlying bug." + }, + { + "left": "issue:44393", + "right": "issue:45071", + "accept": false, + "reason": "Qwen3-VL 2D bounding-box output problems are unrelated to PretrainedConfig type-checking regressions." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 1, + "component_count": 19, + "review_unit_index": 52, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:31795", + "issue:33453", + "issue:34689", + "issue:41669", + "issue:42371", + "issue:42890", + "issue:43352", + "issue:43408", + "issue:43519", + "issue:43525", + "issue:43572", + "issue:43576", + "issue:43673", + "issue:43704", + "issue:43827", + "issue:43976", + "issue:44077", + "issue:44246", + "issue:44273", + "issue:44295", + "issue:44297", + "issue:44315", + "issue:44351", + "issue:44360", + "issue:44393", + "issue:44485", + "issue:44492", + "issue:44556", + "issue:44655", + "issue:44704", + "issue:44829", + "issue:44861", + "issue:44908", + "issue:45003", + "issue:45071", + "issue:45092", + "issue:45230", + "issue:45278", + "issue:45310", + "issue:45341", + "issue:45468", + "issue:45478", + "issue:45753", + "issue:45850", + "issue:46018", + "issue:46172", + "issue:46218", + "issue:46328" + ], + "soft_pairs": [ + "issue:44273|issue:45003", + "issue:43576|issue:44556", + "issue:44393|issue:45753", + "issue:43525|issue:45850", + "issue:41669|issue:43673", + "issue:44556|issue:45092", + "issue:43572|issue:45850", + "issue:44492|issue:45850", + "issue:45278|issue:45850", + "issue:42371|issue:44492", + "issue:41669|issue:46218", + "issue:42890|issue:44908", + "issue:43408|issue:44315", + "issue:44297|issue:45230", + "issue:44246|issue:44273", + "issue:43352|issue:44393", + "issue:44295|issue:45230", + "issue:43576|issue:44908", + "issue:44704|issue:44908", + "issue:43976|issue:44393", + "issue:44908|issue:45071", + "issue:41669|issue:44351", + "issue:44492|issue:46328", + "issue:41669|issue:45341", + "issue:43976|issue:45850", + "issue:44393|issue:45478", + "issue:44077|issue:44861", + "issue:43976|issue:46218", + "issue:43827|issue:44908", + "issue:31795|issue:33453", + "issue:44315|issue:44829", + "issue:43576|issue:44273", + "issue:44655|issue:44861", + "issue:45230|issue:45310", + "issue:44360|issue:45468", + "issue:44273|issue:44908", + "issue:43576|issue:46218", + "issue:43519|issue:44485", + "issue:46018|issue:46172", + "issue:43519|issue:44908", + "issue:41669|issue:43704", + "issue:31795|issue:34689" + ], + "prepared_review_unit_hash": "42adaef6177f8e02448c17a48b8f60049262999e74cb4d1b61b20971c3fa16ca", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 48, + "item_count": 48, + "soft_pair_count": 42, + "serialized_chars": 39548, + "estimated_input_tokens": 9887, + "estimated_eval_tokens": 20030 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The soft pairs are mostly false positives: they share broad themes like v5 regressions, import/loading problems, or model-family-specific bugs, but not the same concrete underlying defect. The closest umbrella issue is the general v5 import/loading breakage report, but there is no strong duplicate cluster here.", + "confidence": 0.89, + "canonical_issue_reason": "issue:45278 is the broadest concrete regression report in the set (many import errors after upgrading 4.57.0 -> 5.5.0), so it is the closest umbrella candidate for the import/load-related subset, though it is not a true duplicate of most others.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:45278 is the most representative and actionable issue for this cluster because it clearly describes a widespread upgrade regression affecting imports/loading, while the other issues are mostly narrower model-specific bugs or unrelated regressions.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44273", + "right": "issue:45003", + "accept": false, + "reason": "Both involve import/lazy-loading behavior, but one is a generic lazy-loading complaint and the other is a specific sys.modules access concern; not clearly the same bug." + }, + { + "left": "issue:43576", + "right": "issue:44556", + "accept": false, + "reason": "Both are v5 breakage reports, but one is about the `transformers env` command and the other about checkpoint reload compatibility; different code paths." + }, + { + "left": "issue:44393", + "right": "issue:45753", + "accept": false, + "reason": "Same model family, but one is a 2D bounding-box output issue and the other is an infinite-loop on a specific image; different failures." + }, + { + "left": "issue:43525", + "right": "issue:45850", + "accept": false, + "reason": "One is a specific `Llama4Config` attribute error; the other is a vague general complaint about many models breaking in 5.0." + }, + { + "left": "issue:41669", + "right": "issue:43673", + "accept": false, + "reason": "One concerns expensive `import *` usage and import-time overhead; the other is a GenerationMixin cache problem during chunked prefill." + }, + { + "left": "issue:44556", + "right": "issue:45092", + "accept": false, + "reason": "Both mention v5 reload/incompatibility after upgrades, but they concern different model families and likely different underlying breakages." + }, + { + "left": "issue:43572", + "right": "issue:45850", + "accept": false, + "reason": "Specific missing `pad_token_idx` in StableLmConfig is not the same as a broad claim that 5.0 breaks many models." + }, + { + "left": "issue:44492", + "right": "issue:45850", + "accept": false, + "reason": "A cache-strategy typo is unrelated to the generic v5 model-breakage complaint." + }, + { + "left": "issue:45278", + "right": "issue:45850", + "accept": false, + "reason": "Both are broad v5 regression reports, but one is specifically about import errors after upgrade and the other is an unspecific blanket complaint; not the same defect." + }, + { + "left": "issue:42371", + "right": "issue:44492", + "accept": false, + "reason": "TF32 API guidance and a cache-strategy typo are unrelated." + }, + { + "left": "issue:41669", + "right": "issue:46218", + "accept": false, + "reason": "One is about import-time cost from `import *`; the other is a missing `Qwen3_5ForCausalLM` import, which is a separate symbol/export issue." + }, + { + "left": "issue:42890", + "right": "issue:44908", + "accept": false, + "reason": "A flaky test due to missing `set_seed()` is unrelated to scheduler kwargs handling." + }, + { + "left": "issue:43408", + "right": "issue:44315", + "accept": false, + "reason": "Model-type warning for sam3_tracker/sam3_video and Liger Kernel not applied with `model_init` are unrelated." + }, + { + "left": "issue:44297", + "right": "issue:45230", + "accept": false, + "reason": "Tokenizer save metadata mismatch and a generic bug report do not describe the same bug." + }, + { + "left": "issue:44246", + "right": "issue:44273", + "accept": false, + "reason": "Slow imports and broken lazy loading may be related at a high level, but the reports do not identify the same concrete defect." + }, + { + "left": "issue:43352", + "right": "issue:44393", + "accept": false, + "reason": "Flash Attention 2 support for Nemotron and Qwen3-VL bounding-box errors are unrelated." + }, + { + "left": "issue:44295", + "right": "issue:45230", + "accept": false, + "reason": "A buffer/`position_ids` access error is not the same as a generic bug report." + }, + { + "left": "issue:43576", + "right": "issue:44908", + "accept": false, + "reason": "The `transformers env` command and inverse_sqrt scheduler kwargs are different areas of the codebase." + }, + { + "left": "issue:44704", + "right": "issue:44908", + "accept": false, + "reason": "AutoProcessor kwarg forwarding and scheduler kwargs are unrelated parameter-passing bugs." + }, + { + "left": "issue:43976", + "right": "issue:44393", + "accept": false, + "reason": "Python-version compatibility and Qwen3-VL bounding-box output are unrelated issues." + }, + { + "left": "issue:44908", + "right": "issue:45071", + "accept": false, + "reason": "Scheduler kwargs handling and `PretrainedConfig` type-checking are different bugs." + }, + { + "left": "issue:41669", + "right": "issue:44351", + "accept": false, + "reason": "Import performance from `import *` is unrelated to failing to import `HybridCache`." + }, + { + "left": "issue:44492", + "right": "issue:46328", + "accept": false, + "reason": "A typo in cache strategies is unrelated to `SequenceFeatureExtractor.pad` converting arrays inefficiently." + }, + { + "left": "issue:41669", + "right": "issue:45341", + "accept": false, + "reason": "Import-star cleanup/performance and a small testing_utils bug are unrelated." + }, + { + "left": "issue:43976", + "right": "issue:45850", + "accept": false, + "reason": "Python version support and generic v5 model breakage are not the same underlying bug." + }, + { + "left": "issue:44393", + "right": "issue:45478", + "accept": false, + "reason": "Both concern Qwen-related models, but one is a VL bounding-box issue and the other is a Qwen3.5 MoE `from_pretrained` failure." + }, + { + "left": "issue:44077", + "right": "issue:44861", + "accept": false, + "reason": "A `patchtsmixer` initialization policy issue is unrelated to `_get_tied_weight_keys` crashing on a list." + }, + { + "left": "issue:43976", + "right": "issue:46218", + "accept": false, + "reason": "Python-version compatibility and a missing `Qwen3_5ForCausalLM` import are unrelated." + }, + { + "left": "issue:43827", + "right": "issue:44908", + "accept": false, + "reason": "Docs referencing removed pipelines and scheduler kwargs are unrelated." + }, + { + "left": "issue:31795", + "right": "issue:33453", + "accept": false, + "reason": "Documentation confusion in `model.forward` is unrelated to tokenizer loading regression." + }, + { + "left": "issue:44315", + "right": "issue:44829", + "accept": false, + "reason": "Liger Kernel not applied with `model_init` and degenerate training under flash_attention_3 are different training/model-init problems." + }, + { + "left": "issue:43576", + "right": "issue:44273", + "accept": false, + "reason": "`transformers env` being broken and lazy loading not working are both broad runtime complaints, but not the same specific bug." + }, + { + "left": "issue:44655", + "right": "issue:44861", + "accept": false, + "reason": "Saving Pipeline objects and `_get_tied_weight_keys` crashing are unrelated failures." + }, + { + "left": "issue:45230", + "right": "issue:45310", + "accept": false, + "reason": "A generic bug report and a specific Qwen3.5 MoE `from_pretrained` error are not duplicates." + }, + { + "left": "issue:44360", + "right": "issue:45468", + "accept": false, + "reason": "A DSA indexer activation issue is unrelated to Gemma4AudioRelPositionalEncoding." + }, + { + "left": "issue:44273", + "right": "issue:44908", + "accept": false, + "reason": "Lazy loading problems and inverse_sqrt scheduler kwargs are unrelated." + }, + { + "left": "issue:43576", + "right": "issue:46218", + "accept": false, + "reason": "Broken env command and missing Qwen3_5 import are different issues." + }, + { + "left": "issue:43519", + "right": "issue:44485", + "accept": false, + "reason": "Qwen3VL timestamp calculation and GLM-5 RoPE implementation are different model-specific bugs." + }, + { + "left": "issue:46018", + "right": "issue:46172", + "accept": false, + "reason": "DeepSeek-V4 shared expert gating and a DeepSeek-V4 transpose/dtype bug are different concrete defects." + }, + { + "left": "issue:43519", + "right": "issue:44908", + "accept": false, + "reason": "Timestamp calculation in a processor and scheduler kwargs handling are unrelated." + }, + { + "left": "issue:41669", + "right": "issue:43704", + "accept": false, + "reason": "Import-star cleanup/performance and a VRAM leak in multi-threaded dataloaders are unrelated." + }, + { + "left": "issue:31795", + "right": "issue:34689", + "accept": false, + "reason": "A documentation issue and a model-loading regression for Llama 3.2 Vision 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": 19, + "review_unit_index": 53, + "review_unit_count": 53, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-10000-450", + "nodes": [ + "issue:30333", + "issue:31515", + "issue:31795", + "issue:33453", + "issue:36296", + "issue:37934", + "issue:38617", + "issue:42371", + "issue:42913", + "issue:42936", + "issue:43066", + "issue:43240", + "issue:43298", + "issue:43352", + "issue:43502", + "issue:43519", + "issue:43525", + "issue:43572", + "issue:43606", + "issue:43618", + "issue:43668", + "issue:43704", + "issue:43717", + "issue:43813", + "issue:44031", + "issue:44230", + "issue:44261", + "issue:44263", + "issue:44351", + "issue:44485", + "issue:44492", + "issue:44623", + "issue:44861", + "issue:44908", + "issue:44998", + "issue:45290", + "issue:45310", + "issue:45341", + "issue:45468", + "issue:45850", + "issue:46018", + "issue:46095", + "issue:46144", + "issue:46218" + ], + "soft_pairs": [ + "issue:31515|issue:31795", + "issue:43704|issue:44485", + "issue:44485|issue:45468", + "issue:43298|issue:46095", + "issue:45850|issue:46218", + "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:44351|issue:46218", + "issue:43352|issue:46218", + "issue:42913|issue:43066", + "issue:44908|issue:44998", + "issue:44492|issue:44998", + "issue:44998|issue:45850", + "issue:43618|issue:44998", + "issue:44861|issue:45290", + "issue:44998|issue:45341", + "issue:43525|issue:44861", + "issue:43572|issue:44861", + "issue:31515|issue:38617", + "issue:44230|issue:44623", + "issue:30333|issue:31515", + "issue:31515|issue:36296", + "issue:44230|issue:45310", + "issue:31515|issue:33453", + "issue:43298|issue:44230", + "issue:43717|issue:46144", + "issue:42936|issue:46018", + "issue:43668|issue:44031", + "issue:42936|issue:43240", + "issue:43240|issue:46144", + "issue:42936|issue:44031", + "issue:42936|issue:46144", + "issue:42936|issue:43813", + "issue:43240|issue:43668", + "issue:43668|issue:43813", + "issue:37934|issue:43668", + "issue:42936|issue:43668" + ], + "prepared_review_unit_hash": "41043ad1267420e234cb6cfd9c1b4bf3355c2f12583e8f29ffad08d8e038b73b", + "component_budget": { + "node_count": 450, + "item_count": 450, + "soft_pair_count": 1664, + "serialized_chars": 861331, + "estimated_input_tokens": 215333, + "estimated_eval_tokens": 430922 + }, + "budget": { + "node_count": 44, + "item_count": 44, + "soft_pair_count": 42, + "serialized_chars": 37124, + "estimated_input_tokens": 9281, + "estimated_eval_tokens": 18818 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "The cluster is highly heterogeneous: most items are model-specific regressions, tokenizer/config issues, or unrelated API/docs bugs. The soft-similarity pairs mostly share only a broad subsystem, not the same concrete defect, so I reject all soft edges.", + "confidence": 0.69, + "canonical_issue_reason": "Issue 45850 is the broadest umbrella in the set and best matches the apparent recurring theme of Transformers v5 regressions affecting multiple models, but the cluster is not truly duplicate-heavy.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 45850 is the best overall representative because it captures the widest common theme across the set (version-5 breakages), even though it is too broad to be a precise duplicate target.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:31515", + "right": "issue:31795", + "accept": false, + "reason": "Different problems: checkpoint loading speed versus documentation wording for forward inputs." + }, + { + "left": "issue:43704", + "right": "issue:44485", + "accept": false, + "reason": "Different model/code paths: VRAM leak in dataloader threads versus GLM-5 RoPE implementation." + }, + { + "left": "issue:44485", + "right": "issue:45468", + "accept": false, + "reason": "Both are model-specific bugs, but they affect different models and distinct logic." + }, + { + "left": "issue:43298", + "right": "issue:46095", + "accept": false, + "reason": "Continuous batching audio support and DeepSeek-V4 weight conversion are unrelated." + }, + { + "left": "issue:45850", + "right": "issue:46218", + "accept": false, + "reason": "One is a broad version-5 complaint; the other is an import error for a specific class." + }, + { + "left": "issue:43502", + "right": "issue:43519", + "accept": false, + "reason": "Local-files-only network requests and Qwen3VL timestamp math are different bugs." + }, + { + "left": "issue:42371", + "right": "issue:43704", + "accept": false, + "reason": "TF32 settings guidance and a VRAM leak are unrelated." + }, + { + "left": "issue:43519", + "right": "issue:44623", + "accept": false, + "reason": "Timestamp calculation and missing save_pretrained files are different code paths." + }, + { + "left": "issue:43519", + "right": "issue:43704", + "accept": false, + "reason": "No concrete overlap beyond both being runtime bugs." + }, + { + "left": "issue:43606", + "right": "issue:45468", + "accept": false, + "reason": "CPU offload device mismatch and Gemma-4 positional encoding are unrelated." + }, + { + "left": "issue:44263", + "right": "issue:45468", + "accept": false, + "reason": "torch.split return handling in GLM-MoE is not the same as Gemma-4 audio positional encoding." + }, + { + "left": "issue:44261", + "right": "issue:45468", + "accept": false, + "reason": "Different model internals and different failure modes." + }, + { + "left": "issue:44351", + "right": "issue:46218", + "accept": false, + "reason": "Both are import errors, but for different symbols and likely different fixes." + }, + { + "left": "issue:43352", + "right": "issue:46218", + "accept": false, + "reason": "Flash Attention support gating and a module import failure are unrelated." + }, + { + "left": "issue:42913", + "right": "issue:43066", + "accept": false, + "reason": "Both concern tokenizer v5 behavior, but one is a broad behavior change and the other is a decoder-type mapping bug." + }, + { + "left": "issue:44908", + "right": "issue:44998", + "accept": false, + "reason": "Scheduler kwargs bug versus an unrelated placeholder issue." + }, + { + "left": "issue:44492", + "right": "issue:44998", + "accept": false, + "reason": "A typo in cache strategies is unrelated to the other issue." + }, + { + "left": "issue:44998", + "right": "issue:45850", + "accept": false, + "reason": "One is not a technical bug; the other is a broad Transformers regression report." + }, + { + "left": "issue:43618", + "right": "issue:44998", + "accept": false, + "reason": "CLIPOutput attentions regression and an unrelated placeholder issue." + }, + { + "left": "issue:44861", + "right": "issue:45290", + "accept": false, + "reason": "Tied-weights crash and chat-template tool-call crash are different code paths." + }, + { + "left": "issue:44998", + "right": "issue:45341", + "accept": false, + "reason": "Unrelated issues; no shared defect." + }, + { + "left": "issue:43525", + "right": "issue:44861", + "accept": false, + "reason": "Missing config attribute in Llama4 versus tied-weights list handling are unrelated." + }, + { + "left": "issue:43572", + "right": "issue:44861", + "accept": false, + "reason": "StableLm config padding field regression is not the same bug as tied-weights handling." + }, + { + "left": "issue:31515", + "right": "issue:38617", + "accept": false, + "reason": "Slow checkpoint loading and an ImportError for layer_type_validation are unrelated." + }, + { + "left": "issue:44230", + "right": "issue:44623", + "accept": false, + "reason": "FP8 support request and processor.save_pretrained missing files are different issues." + }, + { + "left": "issue:30333", + "right": "issue:31515", + "accept": false, + "reason": "MLFlow reporting on failed training jobs is unrelated to checkpoint loading speed." + }, + { + "left": "issue:31515", + "right": "issue:36296", + "accept": false, + "reason": "Different symptoms and different subsystems." + }, + { + "left": "issue:44230", + "right": "issue:45310", + "accept": false, + "reason": "FP8 feature support request versus from_pretrained error in Qwen3.5 MoE are not the same defect." + }, + { + "left": "issue:31515", + "right": "issue:33453", + "accept": false, + "reason": "Checkpoint loading performance and tokenizer loading regression are different problems." + }, + { + "left": "issue:43298", + "right": "issue:44230", + "accept": false, + "reason": "Audio continuous batching support and FP8 inference support are unrelated." + }, + { + "left": "issue:43717", + "right": "issue:46144", + "accept": false, + "reason": "Mamba init_weights and RoFormer attention interface are separate model implementations." + }, + { + "left": "issue:42936", + "right": "issue:46018", + "accept": false, + "reason": "Different model-specific bugs with no shared concrete code path." + }, + { + "left": "issue:43668", + "right": "issue:44031", + "accept": false, + "reason": "Incorrect type hint and tokenizer regex warning are unrelated." + }, + { + "left": "issue:42936", + "right": "issue:43240", + "accept": false, + "reason": "One is a missing use of ignore_value; the other is kwargs not passed into loss calculation." + }, + { + "left": "issue:43240", + "right": "issue:46144", + "accept": false, + "reason": "Loss kwargs handling and RoFormer attention interface are unrelated." + }, + { + "left": "issue:42936", + "right": "issue:44031", + "accept": false, + "reason": "Different bugs: ignore_value usage versus regex warning." + }, + { + "left": "issue:42936", + "right": "issue:46144", + "accept": false, + "reason": "Different model internals and failure modes." + }, + { + "left": "issue:42936", + "right": "issue:43813", + "accept": false, + "reason": "A model-loss bug and a typo in quantization operation naming are unrelated." + }, + { + "left": "issue:43240", + "right": "issue:43668", + "accept": false, + "reason": "Loss kwargs propagation and a config type-hint issue are unrelated." + }, + { + "left": "issue:43668", + "right": "issue:43813", + "accept": false, + "reason": "Type-hint correction and typo in cache conversion naming are not the same bug." + }, + { + "left": "issue:37934", + "right": "issue:43668", + "accept": false, + "reason": "RMSNorm question for Llama4 and a ModernBERT type-hint bug are unrelated." + }, + { + "left": "issue:42936", + "right": "issue:43668", + "accept": false, + "reason": "Different model-specific issues with no concrete overlap." + } + ] + }, + "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": 19, + "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": "e3740ca505d98751732cb5dfb22c3cef79ead20a31a24fb521d5b38b2628895a", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 2878, + "estimated_input_tokens": 720, + "estimated_eval_tokens": 1696 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 2878, + "estimated_input_tokens": 720, + "estimated_eval_tokens": 1696 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These three issues are not duplicates of one another; they request unrelated features in different parts of the library.", + "confidence": 0.98, + "canonical_issue_reason": "No single canonical issue: the items cover unrelated requests (token truncation behavior, ReduceLROnPlateau-like scheduling, and Wav2Vec2 word offsets).", + "canonical_pr_reason": null, + "best_issue_reason": "issue:10519 is the most concrete and narrowly scoped request, but it should not be treated as a canonical duplicate for the others.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:10519", + "right": "issue:11307", + "accept": false, + "reason": "Different topics: truncation from the beginning vs. Wav2Vec2 word start/end offsets." + }, + { + "left": "issue:10519", + "right": "issue:11005", + "accept": false, + "reason": "Different topics: truncation behavior vs. ReduceLROnPlateau-like learning-rate scheduling." + }, + { + "left": "issue:11005", + "right": "issue:11307", + "accept": false, + "reason": "Different topics: scheduler functionality vs. extracting Wav2Vec2 word timing 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": 19, + "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": "ff92f07787566a7c34c16ba09c17d7b146ba179e536a63c41cc087e29de0c599", + "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 issues do not look like true duplicates: one is about Tapas tokenization mismatch, one about an incorrect auth-related error when downloading private models, and one about a generic failure to load bert-base-uncased. The only mild overlap is around model-loading UX, not the same underlying bug.", + "confidence": 0.31, + "canonical_issue_reason": "No strong canonical issue exists because the topics differ. If one must be chosen, #16618 is the broadest model-loading report and the closest to the other two by title, but it is still not the same bug as the auth error or Tapas tokenization issue.", + "canonical_pr_reason": null, + "best_issue_reason": "#16618 is the best representative only in the sense that it is the most general and closest to the shared 'model loading' wording; however, it should not be treated as a true duplicate target for the other issues.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:13244", + "right": "issue:16618", + "accept": false, + "reason": "Tapas tokenization mismatch and bert-base-uncased loading failure are unrelated code paths; one concerns tokenizer parity, the other a model load error." + }, + { + "left": "issue:15980", + "right": "issue:16618", + "accept": false, + "reason": "#15980 is specifically about the wrong error when a private model download is attempted without login, while #16618 is a generic load failure for a public model; same broad area, different 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": 4, + "component_count": 19, + "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": "8c45da95a32f4cf0d9ad84a91386f4dfd723ffda7cd05e0da5c978614dd51fce", + "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 feature requests: 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 or code path.", + "confidence": 0.98, + "canonical_issue_reason": "Issue 26350 is the stronger anchor only by activity/engagement, but it is not a duplicate of 33260 because the requested features are different.", + "canonical_pr_reason": null, + "best_issue_reason": "26350 has much higher discussion activity and is the more established tracker, so it is the best representative issue here even though the requests are unrelated.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:26350", + "right": "issue:33260", + "accept": false, + "reason": "Different feature requests (Flash Attention 2 vs GGUF support); shared wording about 'more architectures' is too broad to indicate the same underlying bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 5, + "component_count": 19, + "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-26413-2", + "nodes": [ + "issue:26413", + "issue:31501" + ], + "soft_pairs": [ + "issue:26413|issue:31501" + ], + "prepared_review_unit_hash": "37670e0b969fa8bf6c1a2ee26fa2a530c83bd1951e908505673cc72dc9a44a50", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1593, + "estimated_input_tokens": 399, + "estimated_eval_tokens": 1054 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1593, + "estimated_input_tokens": 399, + "estimated_eval_tokens": 1054 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues both concern training startup failures, but they do not clearly describe the same underlying bug. The checkpoint-resume issue has a concrete error path; the FineWeb training issue is broader and could have many causes.", + "confidence": 0.76, + "canonical_issue_reason": "issue #26413 is the more specific, technically grounded report with a clear failure signature (`resume_from_checkpoint` / empty `epoch_iterator`), making it the better anchor for this cluster.", + "canonical_pr_reason": null, + "best_issue_reason": "issue #26413 is the best representative because it states an exact error condition and code path, while #31501 is a high-level symptom report without enough evidence of the same root cause.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:26413", + "right": "issue:31501", + "accept": false, + "reason": "Both are training-related, but one is a specific checkpoint-resume exception and the other is a generic \u201ctraining doesn\u2019t start\u201d report. There is not enough evidence they describe 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": 6, + "component_count": 19, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-27088-8", + "nodes": [ + "issue:27088", + "issue:28936", + "issue:38435", + "issue:43834", + "issue:44365", + "issue:44370", + "issue:46222", + "issue:46226" + ], + "soft_pairs": [ + "issue:27088|issue:44370", + "issue:27088|issue:44365", + "issue:27088|issue:43834", + "issue:27088|issue:38435", + "issue:27088|issue:28936", + "issue:28936|issue:43834", + "issue:28936|issue:44365", + "issue:28936|issue:38435", + "issue:28936|issue:44370", + "issue:38435|issue:44370", + "issue:38435|issue:44365", + "issue:38435|issue:43834", + "issue:38435|issue:46222", + "issue:38435|issue:46226", + "issue:43834|issue:44370", + "issue:43834|issue:44365", + "issue:43834|issue:46222", + "issue:43834|issue:46226", + "issue:44365|issue:44370", + "issue:44365|issue:46222", + "issue:44365|issue:46226", + "issue:44370|issue:46222", + "issue:44370|issue:46226", + "issue:46222|issue:46226" + ], + "prepared_review_unit_hash": "17c198444253ef129a9b2afc134de84cbdaac3f35c23cd662abb3b2fe589547e", + "component_budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 24, + "serialized_chars": 13530, + "estimated_input_tokens": 3383, + "estimated_eval_tokens": 7022 + }, + "budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 24, + "serialized_chars": 13530, + "estimated_input_tokens": 3383, + "estimated_eval_tokens": 7022 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is mostly a set of similarly worded i18n doc-translation requests, but they are not the same underlying issue across languages. The only clear duplicate-like pair is the exact Thai duplicate 46222/46226; the rest look like separate translation requests despite high textual similarity from the shared template.", + "confidence": 0.67, + "canonical_issue_reason": "issue:46222 is the strongest duplicate anchor because it is text-identical to issue:46226, making it the clearest representative of the only confirmed duplicate subcluster.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:46222 is the best overall issue representative because it cleanly matches an exact duplicate (46226) and has the strongest evidence of being the canonical copy among the issues here.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:27088", + "right": "issue:44370", + "accept": false, + "reason": "Different language translation requests; same template wording but not the same underlying work." + }, + { + "left": "issue:27088", + "right": "issue:44365", + "accept": false, + "reason": "Different language translation requests; similarity is from the shared i18n template, not a duplicate." + }, + { + "left": "issue:27088", + "right": "issue:43834", + "accept": false, + "reason": "Turkish vs a separate placeholder translation issue; not enough evidence of the same concrete request." + }, + { + "left": "issue:27088", + "right": "issue:38435", + "accept": false, + "reason": "Different languages (Turkish vs Romanian); these are separate translation tasks." + }, + { + "left": "issue:27088", + "right": "issue:28936", + "accept": false, + "reason": "Different languages (Turkish vs Spanish); not a duplicate." + }, + { + "left": "issue:28936", + "right": "issue:43834", + "accept": false, + "reason": "Spanish translation request versus a separate placeholder translation issue; high similarity is just template overlap." + }, + { + "left": "issue:28936", + "right": "issue:44365", + "accept": false, + "reason": "Different translation requests; not the same underlying issue." + }, + { + "left": "issue:28936", + "right": "issue:38435", + "accept": false, + "reason": "Spanish vs Romanian translation requests are distinct." + }, + { + "left": "issue:28936", + "right": "issue:44370", + "accept": false, + "reason": "Shared template only; no evidence they refer to the same language/change." + }, + { + "left": "issue:38435", + "right": "issue:44370", + "accept": false, + "reason": "Romanian translation request versus a separate placeholder issue; not clearly the same work." + }, + { + "left": "issue:38435", + "right": "issue:44365", + "accept": false, + "reason": "Same broad i18n area, but different translation requests." + }, + { + "left": "issue:38435", + "right": "issue:43834", + "accept": false, + "reason": "Romanian vs another placeholder translation issue; similarity is not sufficient to call a duplicate." + }, + { + "left": "issue:38435", + "right": "issue:46222", + "accept": false, + "reason": "Different language requests; not the same underlying change." + }, + { + "left": "issue:38435", + "right": "issue:46226", + "accept": false, + "reason": "Different language requests; not a duplicate." + }, + { + "left": "issue:43834", + "right": "issue:44370", + "accept": false, + "reason": "Near-identical template text, but no concrete evidence they are the same translation request." + }, + { + "left": "issue:43834", + "right": "issue:44365", + "accept": false, + "reason": "Looks like separate placeholder issues rather than one underlying task." + }, + { + "left": "issue:43834", + "right": "issue:46222", + "accept": false, + "reason": "Different language/issue context; not enough to merge." + }, + { + "left": "issue:43834", + "right": "issue:46226", + "accept": false, + "reason": "Different language/issue context; not a confirmed duplicate." + }, + { + "left": "issue:44365", + "right": "issue:44370", + "accept": false, + "reason": "Very similar template wording, but no explicit evidence of the same underlying issue." + }, + { + "left": "issue:44365", + "right": "issue:46222", + "accept": false, + "reason": "Different translation requests; do not merge on template similarity alone." + }, + { + "left": "issue:44365", + "right": "issue:46226", + "accept": false, + "reason": "Different translation requests; not the same bug/change." + }, + { + "left": "issue:44370", + "right": "issue:46222", + "accept": false, + "reason": "Different translation requests; shared template text is not enough." + }, + { + "left": "issue:44370", + "right": "issue:46226", + "accept": false, + "reason": "Different translation requests; not a duplicate." + }, + { + "left": "issue:46222", + "right": "issue:46226", + "accept": true, + "reason": "These are text-identical duplicate Thai translation issues with no distinguishing evidence." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet and stays conservative: the only fully supported duplicate pair is issue:46222/issue:46226, which have identical titles and maximal similarity, while the other high-similarity pairs are reasonably treated as separate translation requests because the titles indicate different languages or placeholder variants. No overreach on mergeability or PR pairing." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 7, + "component_count": 19, + "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": "3868bfb8e2cf6ceb881c130fe300e5589df9acbf7e016de416c1d232464df89c", + "component_budget": { + "node_count": 16, + "item_count": 16, + "soft_pair_count": 36, + "serialized_chars": 21660, + "estimated_input_tokens": 5415, + "estimated_eval_tokens": 11086 + }, + "budget": { + "node_count": 16, + "item_count": 16, + "soft_pair_count": 36, + "serialized_chars": 21660, + "estimated_input_tokens": 5415, + "estimated_eval_tokens": 11086 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is a set of independent model-addition/support requests across different model families and tasks (BERT variants, detection, audio/TTS/ASR, OCR, vision-language). They share the broad pattern of \"add model support,\" but none appear to be the same underlying change.", + "confidence": 0.92, + "canonical_issue_reason": "issue:43671 is the most central and broadly connected request in the set, with the strongest soft links to many of the other model-support issues. It best represents the cluster theme, but it is still a specific Qwen3-TTS request rather than a duplicate of the others.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43671 is the best representative issue for this cluster because it sits at the center of the similarity graph and reflects the common model-support request pattern most clearly. It should not be treated as a duplicate target for the other items.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:36817", + "right": "issue:41211", + "accept": false, + "reason": "Different model families: EuroBert config vs DEIMv2 support." + }, + { + "left": "issue:36817", + "right": "issue:43671", + "accept": false, + "reason": "BERT config request vs Qwen3-TTS audio support; different task and code path." + }, + { + "left": "issue:36817", + "right": "issue:36879", + "accept": false, + "reason": "EuroBert vs RF-DETR are unrelated models." + }, + { + "left": "issue:36879", + "right": "issue:41211", + "accept": false, + "reason": "RF-DETR detection support is unrelated to DEIMv2." + }, + { + "left": "issue:36879", + "right": "issue:43909", + "accept": false, + "reason": "RF-DETR detection vs LFM2.5 audio support are different changes." + }, + { + "left": "issue:36879", + "right": "issue:44121", + "accept": false, + "reason": "Different model families and implementation paths." + }, + { + "left": "issue:36879", + "right": "issue:43837", + "accept": false, + "reason": "RF-DETR vs Qwen3-ASR are different modalities and models." + }, + { + "left": "issue:36879", + "right": "issue:43671", + "accept": false, + "reason": "RF-DETR and Qwen3-TTS are unrelated requests." + }, + { + "left": "issue:37015", + "right": "issue:43837", + "accept": false, + "reason": "NeoBERT support is unrelated to Qwen3-ASR." + }, + { + "left": "issue:37015", + "right": "issue:44121", + "accept": false, + "reason": "NeoBERT and weight-sparse transformer are different model additions." + }, + { + "left": "issue:37015", + "right": "issue:43909", + "accept": false, + "reason": "NeoBERT vs audio model support are different changes." + }, + { + "left": "issue:36879", + "right": "issue:37015", + "accept": false, + "reason": "RF-DETR and NeoBERT are distinct model requests." + }, + { + "left": "issue:37015", + "right": "issue:42503", + "accept": false, + "reason": "NeoBERT and ModernVBERT are different model families." + }, + { + "left": "issue:41115", + "right": "issue:43837", + "accept": false, + "reason": "MiniCPM3 support is separate from Qwen3-ASR support." + }, + { + "left": "issue:41115", + "right": "issue:44121", + "accept": false, + "reason": "MiniCPM3 and weight-sparse transformer are not the same code change." + }, + { + "left": "issue:41115", + "right": "issue:42503", + "accept": false, + "reason": "MiniCPM3 vs ModernVBERT are different model requests." + }, + { + "left": "issue:41115", + "right": "issue:45027", + "accept": false, + "reason": "MiniCPM3 and Voxtral TTS are unrelated models." + }, + { + "left": "issue:41211", + "right": "issue:43671", + "accept": false, + "reason": "DEIMv2 and Qwen3-TTS are different models and tasks." + }, + { + "left": "issue:42503", + "right": "issue:43837", + "accept": false, + "reason": "ModernVBERT models and Qwen3-ASR are unrelated." + }, + { + "left": "issue:42503", + "right": "issue:43671", + "accept": false, + "reason": "ModernVBERT and Qwen3-TTS are different support requests." + }, + { + "left": "issue:42503", + "right": "issue:42738", + "accept": false, + "reason": "Related BERT wording, but not the same underlying change: model addition vs RoPE variant support." + }, + { + "left": "issue:42738", + "right": "issue:45027", + "accept": false, + "reason": "BERT-like RoPE support is unrelated to Voxtral TTS." + }, + { + "left": "issue:43671", + "right": "issue:43837", + "accept": false, + "reason": "Qwen3-TTS and Qwen3-ASR are different tasks and would not be one merged PR." + }, + { + "left": "issue:43671", + "right": "issue:45027", + "accept": false, + "reason": "Both are TTS-related in title only; they target different model families and code paths." + }, + { + "left": "issue:43671", + "right": "issue:45177", + "accept": false, + "reason": "TTS support and OCR2 support are unrelated." + }, + { + "left": "issue:43671", + "right": "issue:44121", + "accept": false, + "reason": "Qwen3-TTS support is unrelated to the sparse transformer request." + }, + { + "left": "issue:43671", + "right": "issue:43909", + "accept": false, + "reason": "Qwen3-TTS and LFM2.5 Audio are distinct model additions." + }, + { + "left": "issue:43837", + "right": "issue:43909", + "accept": false, + "reason": "ASR support vs audio model support are different requests." + }, + { + "left": "issue:43837", + "right": "issue:45322", + "accept": false, + "reason": "Qwen3-ASR and EUPE are different model families and tasks." + }, + { + "left": "issue:43837", + "right": "issue:44121", + "accept": false, + "reason": "ASR support is unrelated to the weight-sparse transformer request." + }, + { + "left": "issue:43909", + "right": "issue:44205", + "accept": false, + "reason": "Audio model support vs SAM3-LiteText are unrelated." + }, + { + "left": "issue:44121", + "right": "issue:45322", + "accept": false, + "reason": "Weight-sparse transformer and EUPE are different underlying changes." + }, + { + "left": "issue:43837", + "right": "issue:45027", + "accept": false, + "reason": "Qwen3-ASR and Voxtral TTS are different model support changes." + }, + { + "left": "issue:37015", + "right": "issue:45027", + "accept": false, + "reason": "NeoBERT and Voxtral TTS are unrelated model requests." + }, + { + "left": "issue:43837", + "right": "issue:45177", + "accept": false, + "reason": "ASR support and DeepSeek-OCR2 support are different tasks." + }, + { + "left": "issue:45322", + "right": "issue:46133", + "accept": false, + "reason": "EUPE and TIPSv2 are distinct vision-language model requests." + } + ] + }, + "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": 19, + "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": "afcf19f5cfa2617fa95ec40963375d01fc2748d28373b1e14f13e5d9b068fa8f", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1480, + "estimated_input_tokens": 370, + "estimated_eval_tokens": 996 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1480, + "estimated_input_tokens": 370, + "estimated_eval_tokens": 996 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are not duplicates: one asks for Set Block Decoding, the other requests DeepSeek v3.2 support. They describe different user-facing problems and have no shared targets or code-path evidence.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": "Issue #43037 is the more concrete and actionable request, but it is still unrelated to #41084 rather than a duplicate of it.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:41084", + "right": "issue:43037", + "accept": false, + "reason": "Different topics entirely: block decoding vs model support for DeepSeek v3.2. No shared filenames, targets, or code path evidence." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 9, + "component_count": 19, + "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": "ee7f3caea1d265221a4c99a923b80b6e92f37127fcfe2ab261200f352e17fbe9", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1584, + "estimated_input_tokens": 396, + "estimated_eval_tokens": 1048 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1584, + "estimated_input_tokens": 396, + "estimated_eval_tokens": 1048 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related only at a very high level (Trainer customization), but they describe different problems: one is about `remove_unused_column` behavior, the other about FLOPS estimation for custom trainers. I would not treat them as duplicates.", + "confidence": 0.94, + "canonical_issue_reason": "Issue 43598 is the more general Trainer customizability request and is the better representative if one issue must be kept.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43598 is the better global anchor for this cluster because it centers on Trainer behavior customization rather than a narrower utility concern.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43598", + "right": "issue:43600", + "accept": false, + "reason": "They concern different code paths and outcomes: `remove_unused_column` customizability versus FLOPS estimation for custom trainers. Shared Trainer context is too broad to imply 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": 10, + "component_count": 19, + "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": "888e9bf61323e8c0a8cdd3ba357fe6d5dc49ce63bf5a15a6f5b244489aa8219b", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1611, + "estimated_input_tokens": 403, + "estimated_eval_tokens": 1062 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1611, + "estimated_input_tokens": 403, + "estimated_eval_tokens": 1062 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related only at a very broad level (Qwen model support), but they describe different model families and different auto-class registration gaps, so they should not be deduplicated.", + "confidence": 0.89, + "canonical_issue_reason": "Issue 43684 is the broader, more central request: it asks for model registration in AutoModel and AutoModelForConditionalGeneration, which is the more general missing-support problem.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue 43684 is the best representative because it covers the core registration gap in the auto-model machinery, whereas 44405 is a narrower sequence-classification-specific request for a different config family.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43684", + "right": "issue:44405", + "accept": false, + "reason": "Different underlying changes: one requests Qwen3-Omni registration for generic/conditional-generation auto classes, the other requests AutoModelForSequenceClassification support for Qwen3.5. Same broad area, but not the same bug or mergeable 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": 19, + "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": "0092da70d7afde1e86b61af67a0bfb8ef34dbc43c9f41841072d29b6b41776b0", + "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": "All four items are low-information closed issues. Three share the exact title \"huggingface \" but there is no body/context to confirm they describe the same underlying bug; the other issue titled \"Fett\" looks unrelated. I would keep one of the repeated \"huggingface \" reports as the representative issue, but I would not merge any pair based on the available evidence.", + "confidence": 0.89, + "canonical_issue_reason": "issue:43845 is the best representative of the repeated \"huggingface \" reports because it is the earliest of the near-identical titles and has no less evidence than the later duplicates.", + "canonical_pr_reason": null, + "best_issue_reason": "issue:43845 is the strongest global issue candidate: it matches the repeated title cluster, is earliest among the duplicates, and there is no evidence that another issue is a better source report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:43835", + "right": "issue:43845", + "accept": false, + "reason": "Different titles and no body details. \"Fett\" has no visible relation to the repeated \"huggingface \" reports." + }, + { + "left": "issue:43835", + "right": "issue:43846", + "accept": false, + "reason": "Different titles and no substantive description to indicate the same underlying bug." + }, + { + "left": "issue:43835", + "right": "issue:43859", + "accept": false, + "reason": "Different titles and no evidence tying this issue to the \"huggingface \" reports." + }, + { + "left": "issue:43845", + "right": "issue:43846", + "accept": false, + "reason": "Titles match, but the issues are too empty to confirm they report the same concrete problem; identical phrasing alone is not enough." + }, + { + "left": "issue:43845", + "right": "issue:43859", + "accept": false, + "reason": "Same generic title, but there is no body or technical context to prove these are duplicate reports of one bug." + }, + { + "left": "issue:43846", + "right": "issue:43859", + "accept": false, + "reason": "Same generic title, but the available evidence is insufficient to establish they describe 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": 12, + "component_count": 19, + "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": "7ade19427739b93dd009c2ff6f807100171492bf62a5d0bd034c9ce677180429", + "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": "This cluster looks like unrelated low-information/spam issues rather than true duplicates. There is no evidence of a shared bug or request, and the only somewhat descriptive title is still too vague to anchor a canonical duplicate.", + "confidence": 0.93, + "canonical_issue_reason": "No strong canonical issue: the items have different titles, no bodies or linked references, and no shared technical details. They do not appear to describe the same underlying problem.", + "canonical_pr_reason": null, + "best_issue_reason": "If forced to pick a representative, issue 44418 is the least opaque because it has the only non-spam-like title, but it is still not a convincing canonical duplicate target.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:44371", + "right": "issue:44572", + "accept": false, + "reason": "Both are titled \"\" with no body or links, so there is no concrete evidence they report the same issue." + }, + { + "left": "issue:44371", + "right": "issue:44573", + "accept": false, + "reason": "Same generic spam title, but no technical content or references to show a shared underlying bug." + }, + { + "left": "issue:44371", + "right": "issue:44418", + "accept": false, + "reason": "One is spam-like and the other is a generic documentation suggestion; they do not look like the same report." + }, + { + "left": "issue:44371", + "right": "issue:44937", + "accept": false, + "reason": "The titles are unrelated and there is no content suggesting a common root cause." + }, + { + "left": "issue:44418", + "right": "issue:44572", + "accept": false, + "reason": "A vague documentation suggestion does not match a spam placeholder issue." + }, + { + "left": "issue:44418", + "right": "issue:44573", + "accept": false, + "reason": "No shared subject matter beyond being low-information issues; not the same underlying bug or change." + }, + { + "left": "issue:44572", + "right": "issue:44573", + "accept": false, + "reason": "Identical spam titles are not enough to merge them as duplicates without any substantive body or evidence." + }, + { + "left": "issue:44572", + "right": "issue:44937", + "accept": false, + "reason": "The titles are unrelated and the available metadata gives no indication of the same problem." + }, + { + "left": "issue:44573", + "right": "issue:44937", + "accept": false, + "reason": "No shared technical details, targets, or references; these are not the same issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 13, + "component_count": 19, + "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": "ec34912142b4604103177982616d4fa4e7c58c20934f235dbb2601b3e38fc1a5", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1645, + "estimated_input_tokens": 412, + "estimated_eval_tokens": 1080 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1645, + "estimated_input_tokens": 412, + "estimated_eval_tokens": 1080 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two closed issues appear likely to be duplicates: they share the same title and very similar body sizes, and #45145 has the only inbound reference, so it is the better canonical representative.", + "confidence": 0.87, + "canonical_issue_reason": "#45145 appears to be the better canonical issue because it has an inbound reference, making it slightly better connected in the discussion trail.", + "canonical_pr_reason": null, + "best_issue_reason": "#45145 is the strongest issue representative because it appears to be the more connected copy of the same likely duplicate report.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45145", + "right": "issue:45151", + "accept": true, + "reason": "Identical titles and very similar body lengths make these issues appear to describe the same likely report, though the evidence is limited." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The pairing is plausible, but the writeup is a bit too strong for the evidence available. The packet only shows identical titles, similar body lengths, and a soft_similarity signal; it does not establish that these are definitely duplicates. Please soften the language to \u2018likely\u2019/\u2018appear to be\u2019 throughout, especially in the soft-edge reason and confidence." + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 14, + "component_count": 19, + "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": 19, + "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": "d22306be406110f45c523b58ae1363b61120985338f421e14dc925759f2a4732", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1670, + "estimated_input_tokens": 418, + "estimated_eval_tokens": 1092 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1670, + "estimated_input_tokens": 418, + "estimated_eval_tokens": 1092 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related 4-bit inference issues, but they describe different bugs: one is a Gemma4/audio-module uint8 dtype failure in `torch.finfo()`, the other is a `BitsAndBytesConfig.llm_int8_skip_modules` regression that drops the default `lm_head` exclusion and triggers an assertion. They should not be deduplicated.", + "confidence": 0.91, + "canonical_issue_reason": "Issue #45674 is the more general and actionable report: it names the configuration regression, the affected setting, and the concrete 4-bit inference failure path.", + "canonical_pr_reason": null, + "best_issue_reason": "Issue #45674 is the best representative of the cluster because it describes a central library-side bug rather than a model-specific symptom.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45672", + "right": "issue:45674", + "accept": false, + "reason": "Different underlying bugs and code paths. #45672 is Gemma4/audio-module-specific `torch.finfo()` TypeError on uint8 weights; #45674 is a `BitsAndBytesConfig` regression where `llm_int8_skip_modules` clears the default `lm_head` exclusion and causes an `AssertionError` in 4-bit inference." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 16, + "component_count": 19, + "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": "bfbb7a22b10662deda6f06fa65fa922505f2f0dcd0cbe7881d8d2718c7b2406d", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1477, + "estimated_input_tokens": 370, + "estimated_eval_tokens": 996 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1477, + "estimated_input_tokens": 370, + "estimated_eval_tokens": 996 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two closed issues are not duplicates; they are unrelated placeholder/personal-name titles with no shared bug report context.", + "confidence": 0.98, + "canonical_issue_reason": "No canonical issue: the issues do not describe the same problem or request, so neither can serve as a representative duplicate target.", + "canonical_pr_reason": null, + "best_issue_reason": "Neither issue is a good global representative; both are standalone, unrelated entries with no actionable problem statement.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45762", + "right": "issue:45763", + "accept": false, + "reason": "Titles only share a surname-like token and both are bare name/date strings; there is no evidence they report 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": 19, + "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": "5c956266222604b78382b0c18326765328a9b9a0b956c2cd79323214457cc791", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1480, + "estimated_input_tokens": 370, + "estimated_eval_tokens": 996 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1480, + "estimated_input_tokens": 370, + "estimated_eval_tokens": 996 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The two issues are not convincing duplicates: they are just two different personal-name/date titles with no body, no links, and no shared technical context. The soft similarity signal is too weak to treat them as the same underlying issue.", + "confidence": 0.96, + "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": "Different names and dates, no description or evidence of a shared bug/report; not the same underlying issue." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 18, + "component_count": 19, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45810-2", + "nodes": [ + "issue:45810", + "issue:46295" + ], + "soft_pairs": [ + "issue:45810|issue:46295" + ], + "prepared_review_unit_hash": "da00cc043a3158f7fd01531d3ec003a10d2f5e0b4ac107f1930e188cf5e09d38", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1540, + "estimated_input_tokens": 385, + "estimated_eval_tokens": 1026 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 1540, + "estimated_input_tokens": 385, + "estimated_eval_tokens": 1026 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two issues are not duplicates: one requests a new Qwen token-classification model class, while the other asks for FP8 training support.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45810", + "right": "issue:46295", + "accept": false, + "reason": "Different feature requests in different areas: model class addition vs FP8 training support. They do not describe the same underlying bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "issue", + "component_index": 19, + "component_count": 19, + "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": "3fd48f9b1e16ab7fbc31db8c262ebaef503f8fe96e9b9af2d69cff3719906124", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 8, + "serialized_chars": 5727, + "estimated_input_tokens": 1432, + "estimated_eval_tokens": 3120 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 8, + "serialized_chars": 5727, + "estimated_input_tokens": 1432, + "estimated_eval_tokens": 3120 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The cluster most likely contains two strong duplicate pairs\u201445870/45871 and 45872/45873\u2014plus one unrelated dataset post (45869). The evidence supports likely duplicates, but not confirmed exact duplication from the packet alone.", + "confidence": 0.9, + "canonical_issue_reason": "No single canonical issue fits all five items because the cluster splits into two separate likely-duplicate pairs with different concrete topics, plus one unrelated issue.", + "canonical_pr_reason": null, + "best_issue_reason": "45870 is the earliest clean representative of the 'Example: Financial time series prediction with DataForge' pair, but it only represents that pair and not the separate scraping-service pair.", + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "issue:45869", + "right": "issue:45870", + "accept": false, + "reason": "Dataset bundle promo vs. financial prediction example: related finance wording, but different underlying request." + }, + { + "left": "issue:45869", + "right": "issue:45871", + "accept": false, + "reason": "Dataset bundle promo vs. financial prediction example; not the same issue." + }, + { + "left": "issue:45870", + "right": "issue:45871", + "accept": true, + "reason": "Same title and matching body length with very strong similarity; this is a likely duplicate pair, though not proven exact from metadata alone." + }, + { + "left": "issue:45870", + "right": "issue:45872", + "accept": false, + "reason": "Prediction example vs. scraping-service integration are different concrete asks." + }, + { + "left": "issue:45870", + "right": "issue:45873", + "accept": false, + "reason": "Prediction example vs. scraping-service integration are different concrete asks." + }, + { + "left": "issue:45871", + "right": "issue:45872", + "accept": false, + "reason": "Different underlying topics despite broad DataForge/finance overlap." + }, + { + "left": "issue:45871", + "right": "issue:45873", + "accept": false, + "reason": "Different underlying topics despite broad DataForge/finance overlap." + }, + { + "left": "issue:45872", + "right": "issue:45873", + "accept": true, + "reason": "Same title and matching body length with very strong similarity; this is a likely duplicate pair, though not proven exact from metadata alone." + } + ] + }, + "evaluator_result": { + "accept": false, + "feedback": "The analysis is too strong for the evidence in the packet. We only have titles, body lengths, and soft-similarity signals; there is no body text proving exact duplication. In particular, calling 45870/45871 and 45872/45873 \"exact duplicate\" pairs overstates what is grounded here. A more conservative summary would say they are likely duplicates or strong matches, not confirmed exact duplicates. " + }, + "evaluator_used": true, + "retried": true, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 1, + "component_count": 33, + "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-39895-16", + "nodes": [ + "pull_request:39895", + "pull_request:41855", + "pull_request:42668", + "pull_request:43251", + "pull_request:43254", + "pull_request:43424", + "pull_request:43823", + "pull_request:43995", + "pull_request:44827", + "pull_request:45170", + "pull_request:45213", + "pull_request:45655", + "pull_request:46048", + "pull_request:46267", + "pull_request:46344", + "pull_request:46367" + ], + "soft_pairs": [ + "pull_request:39895|pull_request:43424", + "pull_request:41855|pull_request:43823", + "pull_request:41855|pull_request:43424", + "pull_request:42668|pull_request:43823", + "pull_request:43251|pull_request:43823", + "pull_request:43251|pull_request:44827", + "pull_request:43424|pull_request:44827", + "pull_request:43424|pull_request:43823", + "pull_request:43823|pull_request:45655", + "pull_request:43823|pull_request:43995", + "pull_request:43995|pull_request:45213", + "pull_request:43995|pull_request:44827", + "pull_request:44827|pull_request:46344", + "pull_request:44827|pull_request:45170", + "pull_request:44827|pull_request:46267", + "pull_request:44827|pull_request:46367", + "pull_request:44827|pull_request:45213", + "pull_request:43823|pull_request:45213", + "pull_request:44827|pull_request:45655", + "pull_request:46048|pull_request:46367", + "pull_request:44827|pull_request:46048", + "pull_request:43251|pull_request:43254" + ], + "prepared_review_unit_hash": "66c7f538f489f1a16daabb1d18d018c2f853d2d72de246a5d90ffc4651784c72", + "component_budget": { + "node_count": 16, + "item_count": 16, + "soft_pair_count": 22, + "serialized_chars": 43086, + "estimated_input_tokens": 10772, + "estimated_eval_tokens": 21800 + }, + "budget": { + "node_count": 16, + "item_count": 16, + "soft_pair_count": 22, + "serialized_chars": 43086, + "estimated_input_tokens": 10772, + "estimated_eval_tokens": 21800 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Only one true duplicate pair stands out: two PRs that both extend `fixed_cross_entropy` to forward `weight` and `label_smoothing` into `nn.functional.cross_entropy`. The rest are unrelated changes that only overlap at a broad subsystem level or via generic soft similarity.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 43251 is the cleaner canonical representative of the shared loss-function fix: it has the explicit issue target (43240) and directly patches `fixed_cross_entropy` to pass the missing kwargs through.", + "best_issue_reason": null, + "best_pr_reason": "PR 43251 is the best overall representative of the cluster because it is the only clearly grounded bug-fix PR here, with a concrete issue link and a narrowly scoped implementation in the core loss helper.", + "soft_edge_verdicts": [ + { + "left": "pull_request:39895", + "right": "pull_request:43424", + "accept": false, + "reason": "Different changes: VideoPrism model/docs scaffolding vs an ExecuTorch dynamic-shape test." + }, + { + "left": "pull_request:41855", + "right": "pull_request:43823", + "accept": false, + "reason": "Mistral tokenizer missing methods vs adding MobileLLM; unrelated model areas and code paths." + }, + { + "left": "pull_request:41855", + "right": "pull_request:43424", + "accept": false, + "reason": "Tokenizer method additions vs ExecuTorch export test; no shared bug or change." + }, + { + "left": "pull_request:42668", + "right": "pull_request:43823", + "accept": false, + "reason": "Processor auto-loading robustness changes vs MobileLLM model addition; unrelated." + }, + { + "left": "pull_request:43251", + "right": "pull_request:43823", + "accept": false, + "reason": "Loss-function kwargs fix vs MobileLLM model addition; different concrete changes." + }, + { + "left": "pull_request:43251", + "right": "pull_request:44827", + "accept": false, + "reason": "Loss-function kwargs fix vs Mistral4 test fixes; unrelated code paths." + }, + { + "left": "pull_request:43424", + "right": "pull_request:44827", + "accept": false, + "reason": "ExecuTorch dynamic-shape test vs Mistral4 test fixes; not the same underlying change." + }, + { + "left": "pull_request:43424", + "right": "pull_request:43823", + "accept": false, + "reason": "ExecuTorch export test vs MobileLLM model addition; unrelated." + }, + { + "left": "pull_request:43823", + "right": "pull_request:45655", + "accept": false, + "reason": "MobileLLM model addition vs config-resolution ordering fix; different targets." + }, + { + "left": "pull_request:43823", + "right": "pull_request:43995", + "accept": false, + "reason": "MobileLLM model addition vs Falcon output-interface refactor; unrelated." + }, + { + "left": "pull_request:43995", + "right": "pull_request:45213", + "accept": false, + "reason": "Falcon refactor vs model-creation skill draft; not the same bug or feature." + }, + { + "left": "pull_request:43995", + "right": "pull_request:44827", + "accept": false, + "reason": "Falcon output-interface refactor vs Mistral4 test fix; unrelated." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46344", + "accept": false, + "reason": "Mistral4 test fixes vs VitPose flip_back graph-break fix; different models and issues." + }, + { + "left": "pull_request:44827", + "right": "pull_request:45170", + "accept": false, + "reason": "Mistral4 test fixes vs layernorm renaming/mapping fixes; not the same change." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46267", + "accept": false, + "reason": "Mistral4 test fixes vs chat-response parsing updates; unrelated." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46367", + "accept": false, + "reason": "Mistral4 test fixes vs finegrained FP8 kernel rename/update; unrelated." + }, + { + "left": "pull_request:44827", + "right": "pull_request:45213", + "accept": false, + "reason": "Mistral4 test fixes vs model-creation skill draft; unrelated." + }, + { + "left": "pull_request:43823", + "right": "pull_request:45213", + "accept": false, + "reason": "MobileLLM model addition vs model-creation skill draft; unrelated artifacts." + }, + { + "left": "pull_request:44827", + "right": "pull_request:45655", + "accept": false, + "reason": "Mistral4 test fixes vs config-resolution ordering fix; unrelated." + }, + { + "left": "pull_request:46048", + "right": "pull_request:46367", + "accept": false, + "reason": "TDT loss kernel integration vs finegrained FP8 kernel API rename; different kernels and purposes." + }, + { + "left": "pull_request:44827", + "right": "pull_request:46048", + "accept": false, + "reason": "Mistral4 test fixes vs TDT loss kernel integration; unrelated." + }, + { + "left": "pull_request:43251", + "right": "pull_request:43254", + "accept": true, + "reason": "Both PRs fix the same `fixed_cross_entropy` bug by forwarding supported kwargs (`weight`, `label_smoothing`) into `nn.functional.cross_entropy`, and both point at issue 43240." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is grounded in the packet, and the lone accepted pair is a conservative call: PRs 43251 and 43254 make the same narrow `fixed_cross_entropy` kwargs-forwarding change in the same file and share issue target 43240, so they look mergeable into one bugfix PR. The rejected soft pairs are also conservatively framed as unrelated or merely broadly similar." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 2, + "component_count": 33, + "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-40546-3", + "nodes": [ + "pull_request:40546", + "pull_request:42781", + "pull_request:46331" + ], + "soft_pairs": [ + "pull_request:40546|pull_request:42781", + "pull_request:42781|pull_request:46331" + ], + "prepared_review_unit_hash": "506819152e21cdc077232213ab320c8eba4f72998d0c73feb12eca92731b379a", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 10379, + "estimated_input_tokens": 2595, + "estimated_eval_tokens": 5446 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 10379, + "estimated_input_tokens": 2595, + "estimated_eval_tokens": 5446 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are three distinct feature PRs: an initial VibeVoice integration, a later VibeVoice Realtime extension, and an unrelated Parakeet-RNNT integration. They share some auto-registry and speech-model scaffolding, but they do not look like the same concrete change.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 40546 is the clearest canonical PR for the VibeVoice cluster because it is the original, broad VibeVoice implementation and docs addition; the others are separate follow-on or unrelated model integrations rather than the same change.", + "best_issue_reason": null, + "best_pr_reason": "PR 40546 best represents the underlying VibeVoice feature introduction. PR 42781 is a later realtime/acoustic-tokenizer extension, and PR 46331 is for a different model family (Parakeet-RNNT).", + "soft_edge_verdicts": [ + { + "left": "pull_request:40546", + "right": "pull_request:42781", + "accept": false, + "reason": "Both are about VibeVoice, but 42781 is a distinct follow-up adding realtime and acoustic-tokenizer support plus additional generation/runtime changes. It is not the same concrete fix/feature as the original VibeVoice implementation and would not plausibly be merged as a single duplicate PR." + }, + { + "left": "pull_request:42781", + "right": "pull_request:46331", + "accept": false, + "reason": "These are different model integrations: VibeVoice Realtime versus Parakeet-RNNT. Shared auto-mapping edits are generic registry updates, not evidence of the same underlying bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 3, + "component_count": 33, + "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": "313538dd2696a58deaa39e47334fe6003ca3e7adab0fa25912113659a43c82b9", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4675, + "estimated_input_tokens": 1169, + "estimated_eval_tokens": 2594 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4675, + "estimated_input_tokens": 1169, + "estimated_eval_tokens": 2594 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two pull requests are unrelated: one adds covariate forecasting support for TimesFM, while the other adds DINOv3 image-classification support. They do not describe the same bug, change, or code path.", + "confidence": 0.99, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: the PRs target different model families and implement unrelated features.", + "best_issue_reason": null, + "best_pr_reason": "Neither PR is a good global representative for the cluster because there is no duplicate relationship to collapse.", + "soft_edge_verdicts": [ + { + "left": "pull_request:40755", + "right": "pull_request:41224", + "accept": false, + "reason": "Different subsystems and features: TimesFM covariate forecasting vs. DINOv3 image classification. They do not fix or implement 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": 4, + "component_count": 33, + "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": 33, + "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": "1758fc83c7883da502f043db78a9be497b6b6f78c94810db2edcb33ca13c32c0", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 10433, + "estimated_input_tokens": 2609, + "estimated_eval_tokens": 5474 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 10433, + "estimated_input_tokens": 2609, + "estimated_eval_tokens": 5474 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Three unrelated model-addition PRs were grouped by generic Hugging Face model-registration/doc changes, but they target different models and codepaths, so they are not duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR: each PR introduces a different model package and docs (DeepSeek OCR, Isaac, Kimi2-6), and the shared auto-registration/doc edits are boilerplate rather than the same underlying change.", + "best_issue_reason": null, + "best_pr_reason": "pull_request:45630 is the strongest standalone candidate because it has the most discussion/review activity and the most complete model integration surface, but it is still not merge-equivalent to the others.", + "soft_edge_verdicts": [ + { + "left": "pull_request:41797", + "right": "pull_request:45186", + "accept": false, + "reason": "Different model additions (DeepSeek OCR vs Isaac) with separate model directories and docs; shared registration updates are generic boilerplate, not the same fix/change." + }, + { + "left": "pull_request:45186", + "right": "pull_request:45630", + "accept": false, + "reason": "Different model additions (Isaac vs Kimi2-6); despite overlapping auto-mapping/doc files, they introduce distinct model implementations and should not be merged as one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 6, + "component_count": 33, + "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": "ed8da67cb21ef2260ba97a52f108240dea1b860e1414c41db26da9f797c6830f", + "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 bumps core workflow actions for Node 24 compatibility, the other updates Docker build actions and a few related workflow actions to newer versions. They overlap in files but address different action families and different upgrade goals.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 42975 is the more clearly scoped representative of the cluster because its title and diff directly target a specific compatibility issue across many workflows; PR 42976 is a separate generic action-version refresh focused on Docker-related actions.", + "best_issue_reason": null, + "best_pr_reason": "PR 42975 is the best representative PR for this cluster because it has the clearest, most concrete problem statement and broad workflow impact, while still being distinct from the Docker action upgrades in PR 42976.", + "soft_edge_verdicts": [ + { + "left": "pull_request:42975", + "right": "pull_request:42976", + "accept": false, + "reason": "Do not merge as duplicates: PR 42975 upgrades checkout/cache/upload-artifact for Node 24 compatibility, while PR 42976 upgrades docker/build-push-action and other workflow actions to latest versions. They overlap only in general CI/workflow maintenance, not the same underlying code-path 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": 7, + "component_count": 33, + "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:46411" + ], + "soft_pairs": [ + "pull_request:42978|pull_request:43448", + "pull_request:42978|pull_request:43451", + "pull_request:42978|pull_request:45490", + "pull_request:42978|pull_request:46411", + "pull_request:43448|pull_request:43451", + "pull_request:43448|pull_request:45490", + "pull_request:43448|pull_request:46411", + "pull_request:43451|pull_request:45490", + "pull_request:43451|pull_request:46411", + "pull_request:45490|pull_request:46411" + ], + "prepared_review_unit_hash": "e16c1d768cfbdd61b5dc731a854a114d5deec2e2fed90db57ab261baacba004e", + "component_budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 10, + "serialized_chars": 17815, + "estimated_input_tokens": 4454, + "estimated_eval_tokens": 9164 + }, + "budget": { + "node_count": 5, + "item_count": 5, + "soft_pair_count": 10, + "serialized_chars": 17815, + "estimated_input_tokens": 4454, + "estimated_eval_tokens": 9164 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "All items are separate model-addition PRs for different architectures (ViT NEPA, Molmo, Molmo2, CTSM, MiniMax M2.7). The overlap is only at the template level of adding model docs/registry/tests, not the same underlying bug or change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "PR 43451 is the most substantial and actively reviewed model-addition PR, but it still represents a distinct Molmo2 integration rather than a duplicate of any other PR here.", + "soft_edge_verdicts": [ + { + "left": "pull_request:42978", + "right": "pull_request:43448", + "accept": false, + "reason": "Different models and code paths: ViT NEPA vs Molmo. Shared auto/docs plumbing is generic model-integration work, not the same change." + }, + { + "left": "pull_request:42978", + "right": "pull_request:43451", + "accept": false, + "reason": "Different model families: ViT NEPA vs Molmo2. Similar file patterns, but no shared concrete bug or feature." + }, + { + "left": "pull_request:42978", + "right": "pull_request:45490", + "accept": false, + "reason": "ViT NEPA and CTSM are unrelated architectures (vision-language vs time-series); only the repo integration pattern overlaps." + }, + { + "left": "pull_request:42978", + "right": "pull_request:46411", + "accept": false, + "reason": "ViT NEPA model addition is unrelated to the MiniMax demo/code drop; they do not fix or add the same thing." + }, + { + "left": "pull_request:43448", + "right": "pull_request:43451", + "accept": false, + "reason": "Molmo and Molmo2 are related by name, but they are distinct model ports with different files and capabilities; not the same underlying change." + }, + { + "left": "pull_request:43448", + "right": "pull_request:45490", + "accept": false, + "reason": "Molmo vs CTSM are completely different domains and implementations; shared top-level registry/doc edits are incidental." + }, + { + "left": "pull_request:43448", + "right": "pull_request:46411", + "accept": false, + "reason": "Molmo model integration and MiniMax demo/code are unrelated; no common code-path problem or mergeable duplicate change." + }, + { + "left": "pull_request:43451", + "right": "pull_request:45490", + "accept": false, + "reason": "Molmo2 and CTSM address different model stacks; overlap is only in generic auto-registration/docs." + }, + { + "left": "pull_request:43451", + "right": "pull_request:46411", + "accept": false, + "reason": "Molmo2 integration and MiniMax demo code are unrelated changes, not the same feature or bug fix." + }, + { + "left": "pull_request:45490", + "right": "pull_request:46411", + "accept": false, + "reason": "CTSM model integration and a MiniMax demo file have no shared concrete code-path issue; they are separate additions." + } + ] + }, + "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": 33, + "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-43291-4", + "nodes": [ + "pull_request:43291", + "pull_request:43665", + "pull_request:44815", + "pull_request:46380" + ], + "soft_pairs": [ + "pull_request:43291|pull_request:43665", + "pull_request:43291|pull_request:44815", + "pull_request:43291|pull_request:46380", + "pull_request:43665|pull_request:44815", + "pull_request:43665|pull_request:46380", + "pull_request:44815|pull_request:46380" + ], + "prepared_review_unit_hash": "5430e326dbfd546fd09438bef8ab2986e6a2900789ee99abff1b7f9229008ebf", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 6, + "serialized_chars": 11142, + "estimated_input_tokens": 2786, + "estimated_eval_tokens": 5828 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 6, + "serialized_chars": 11142, + "estimated_input_tokens": 2786, + "estimated_eval_tokens": 5828 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are four mostly unrelated pull requests that happen to have similar \u201cfix\u201d titles and some superficial overlap in touched areas/tests. They do not look like duplicate implementations of the same bug or change.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #46380 is the strongest canonical representative: it fixes a concrete runtime bug in generation config validation, has a clear code-path change, and includes targeted follow-up tests.", + "best_issue_reason": null, + "best_pr_reason": "PR #46380 is the best overall PR in this set because it addresses a specific correctness issue in core generation behavior rather than test-only edits or an unrelated subsystem change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43291", + "right": "pull_request:43665", + "accept": false, + "reason": "Both are test-oriented and share a generic \u201cfix\u201d theme, but they target different models/code paths: Whisper tokenizer decoding vs. CLIP/ViT test edits. No same underlying bug." + }, + { + "left": "pull_request:43291", + "right": "pull_request:44815", + "accept": false, + "reason": "Whisper decoding tests/code and FP8 dequantization/model loading are unrelated problems in different subsystems." + }, + { + "left": "pull_request:43291", + "right": "pull_request:46380", + "accept": false, + "reason": "Whisper tokenizer decoding changes and generation watermarking validation are distinct code paths and bugs; only a broad \u201cfix\u201d label overlaps." + }, + { + "left": "pull_request:43665", + "right": "pull_request:44815", + "accept": false, + "reason": "One PR is test assertions in CLIP/ViT, the other is FP8 dequant/model-loading logic; they are not the same change or bug." + }, + { + "left": "pull_request:43665", + "right": "pull_request:46380", + "accept": false, + "reason": "CLIP/ViT test-only edits do not match the generation-config runtime fix in PR #46380." + }, + { + "left": "pull_request:44815", + "right": "pull_request:46380", + "accept": false, + "reason": "FP8 dequantization/padding and generation watermarking validation are different concrete bugs in different areas; they could not plausibly be merged as one fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 9, + "component_count": 33, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-43488-2", + "nodes": [ + "pull_request:43488", + "pull_request:43532" + ], + "soft_pairs": [ + "pull_request:43488|pull_request:43532" + ], + "prepared_review_unit_hash": "68d63d27bc514066a7aa832f6bba3ec69f9236e2e33d4abf4b07a6dd2e598090", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4563, + "estimated_input_tokens": 1141, + "estimated_eval_tokens": 2538 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4563, + "estimated_input_tokens": 1141, + "estimated_eval_tokens": 2538 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These two pull requests are not duplicates: 43488 is about setup/versioning and a test dependency change, while 43532 is about workflow/CI behavior and related utility changes. They only overlap on a test file and do not appear to fix the same underlying code-path problem.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "PR 43532 is the more substantial and self-contained change set, but it still addresses a different area than 43488 and should not be treated as the same fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43488", + "right": "pull_request:43532", + "accept": false, + "reason": "Shared vit test files are not enough here: 43488 changes packaging/version metadata and adds pytest-random-order, while 43532 rewrites CI/workflow behavior and notification utilities. Different code paths and different goals." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 10, + "component_count": 33, + "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": "0775009ddf58f9203a2a77d244d3e358ef8f4e51bc1d8f89cccbbc19d0c57c5c", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6080, + "estimated_input_tokens": 1520, + "estimated_eval_tokens": 3296 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6080, + "estimated_input_tokens": 1520, + "estimated_eval_tokens": 3296 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two open PRs add similarly named but distinct segmentation pipelines: one for promptable concept segmentation and one for promptable visual segmentation. They share some scaffolding, but they target different features/models and should not be treated as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 43613 is the broader and more complete addition, covering the visual segmentation pipeline with more model integrations and test coverage.", + "best_issue_reason": null, + "best_pr_reason": "PR 43613 is the best representative of the cluster because it is the more substantial, end-to-end pipeline implementation.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43612", + "right": "pull_request:43613", + "accept": false, + "reason": "Reject: these are different features, not the same underlying change. One adds Promptable Concept Segmentation, the other Promptable Visual Segmentation, with different docs, pipeline modules, and model/test coverage. Shared boilerplate does not make them mergeable into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 11, + "component_count": 33, + "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": "de73c04e8345a45e7c0285b0fb51338c270535040863c5047fc1792ce0ed0dd3", + "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": "These two PRs both touch `trainer.py` and add extensibility hooks, but they address different code paths: gradient-accumulation loss scaling vs. signature-based column filtering. They are related in theme, not duplicates.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No single canonical PR: 43651 and 43663 change different Trainer behaviors and would not be merged as one duplicate fix.", + "best_issue_reason": null, + "best_pr_reason": "PR 43651 is the better representative of the cluster because it makes a concrete training-step behavior change with a clear API hook; 43663 is a separate hook for input-column filtering.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43651", + "right": "pull_request:43663", + "accept": false, + "reason": "Reject: both are Trainer extensibility changes, but they fix different problems in different code paths. 43651 changes gradient-accumulation loss normalization; 43663 changes how signature columns are derived for filtering." + } + ] + }, + "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": 33, + "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": "b9723602bddeacfebe346049128f67c1783a4fe8a1c5c276403557d66b5e418e", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3663, + "estimated_input_tokens": 916, + "estimated_eval_tokens": 2088 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 3663, + "estimated_input_tokens": 916, + "estimated_eval_tokens": 2088 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs touch the same file but address different TypeAdapter/runtime typing problems in `src/transformers/cli/serve.py`. They should stay separate; PR 43842 is the more focused standalone fix for the missing-pydantic case.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": "43842 is the cleaner, more targeted fix: it explicitly guards the `TypeAdapter` import behind `TYPE_CHECKING` and quotes the annotation, matching the stated missing-pydantic NameError.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43656", + "right": "pull_request:43842", + "accept": false, + "reason": "Reject: both touch `serve.py`, but they fix different concrete issues. 43656 is a broader annotation/future-import cleanup, while 43842 specifically fixes `TypeAdapter` when pydantic is absent." + } + ] + }, + "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": 33, + "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": 33, + "review_unit_index": 1, + "review_unit_count": 2, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-43996-28", + "nodes": [ + "pull_request:43996", + "pull_request:43997", + "pull_request:44007", + "pull_request:44010", + "pull_request:44013", + "pull_request:44017", + "pull_request:44018", + "pull_request:44019", + "pull_request:44044", + "pull_request:44056", + "pull_request:44059", + "pull_request:44066", + "pull_request:44068", + "pull_request:44071", + "pull_request:44072", + "pull_request:44073", + "pull_request:44074", + "pull_request:44076", + "pull_request:44085", + "pull_request:44086", + "pull_request:44098", + "pull_request:44101", + "pull_request:44116", + "pull_request:44129", + "pull_request:44154", + "pull_request:44161", + "pull_request:44722" + ], + "soft_pairs": [ + "pull_request:43996|pull_request:44085", + "pull_request:43996|pull_request:44044", + "pull_request:44066|pull_request:44085", + "pull_request:44007|pull_request:44072", + "pull_request:44072|pull_request:44722", + "pull_request:44066|pull_request:44072", + "pull_request:44013|pull_request:44044", + "pull_request:44066|pull_request:44086", + "pull_request:44066|pull_request:44071", + "pull_request:44018|pull_request:44068", + "pull_request:44018|pull_request:44066", + "pull_request:44066|pull_request:44068", + "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:44116", + "pull_request:44018|pull_request:44116", + "pull_request:44068|pull_request:44722", + "pull_request:44071|pull_request:44722", + "pull_request:44018|pull_request:44085", + "pull_request:44018|pull_request:44086", + "pull_request:44068|pull_request:44086", + "pull_request:44071|pull_request:44085", + "pull_request:44059|pull_request:44086", + "pull_request:44085|pull_request:44086", + "pull_request:44098|pull_request:44101", + "pull_request:44059|pull_request:44722", + "pull_request:44018|pull_request:44071", + "pull_request:44017|pull_request:44722", + "pull_request:44017|pull_request:44161", + "pull_request:44017|pull_request:44073", + "pull_request:44068|pull_request:44071", + "pull_request:44071|pull_request:44086", + "pull_request:43997|pull_request:44073", + "pull_request:44017|pull_request:44071", + "pull_request:44073|pull_request:44101", + "pull_request:44059|pull_request:44071", + "pull_request:44018|pull_request:44059", + "pull_request:44073|pull_request:44161", + "pull_request:44056|pull_request:44101", + "pull_request:44059|pull_request:44068", + "pull_request:44101|pull_request:44161", + "pull_request:44017|pull_request:44018", + "pull_request:43997|pull_request:44056", + "pull_request:44071|pull_request:44076", + "pull_request:44074|pull_request:44101", + "pull_request:43997|pull_request:44018", + "pull_request:44056|pull_request:44161", + "pull_request:43997|pull_request:44161", + "pull_request:44074|pull_request:44161", + "pull_request:44018|pull_request:44161", + "pull_request:43997|pull_request:44071", + "pull_request:44010|pull_request:44071", + "pull_request:44018|pull_request:44129", + "pull_request:44010|pull_request:44073", + "pull_request:43997|pull_request:44074", + "pull_request:44071|pull_request:44129", + "pull_request:44010|pull_request:44018", + "pull_request:43997|pull_request:44129", + "pull_request:43996|pull_request:44161", + "pull_request:43996|pull_request:44017", + "pull_request:44010|pull_request:44044", + "pull_request:44044|pull_request:44074", + "pull_request:44066|pull_request:44722", + "pull_request:44085|pull_request:44722", + "pull_request:44085|pull_request:44129", + "pull_request:43996|pull_request:44007", + "pull_request:43996|pull_request:44129", + "pull_request:44044|pull_request:44129", + "pull_request:44044|pull_request:44154" + ], + "prepared_review_unit_hash": "351319817012d0d54a2c857aebe183700fdb17fa867f17c6cf597052a9dbc88b", + "component_budget": { + "node_count": 28, + "item_count": 28, + "soft_pair_count": 78, + "serialized_chars": 84896, + "estimated_input_tokens": 21224, + "estimated_eval_tokens": 42704 + }, + "budget": { + "node_count": 27, + "item_count": 27, + "soft_pair_count": 72, + "serialized_chars": 75092, + "estimated_input_tokens": 18773, + "estimated_eval_tokens": 37802 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "This is mostly a theme cluster of standardized output-tracing refactors across many different models. Only the GPT-J/CodeGen and GPT-Neo pairs look like true duplicates; the rest are separate model-specific changes that merely share the same refactor pattern.", + "confidence": 0.85, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 44722 is the best anchor for the only real duplicate subcluster: it covers the same GPT-J/CodeGen output-tracing refactor as the overlapping GPT-J PRs, with the clearest and broadest concrete code-path overlap.", + "best_issue_reason": null, + "best_pr_reason": "PR 44722 is the strongest single representative in this cluster because it reflects the shared GPT-J/CodeGen decorator migration most directly and appears to subsume the closely related GPT-J variants.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43996", + "right": "pull_request:44085", + "accept": false, + "reason": "Different model families; same refactor theme only." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44044", + "accept": false, + "reason": "CVT/FNet output tracing vs DeBERTa output tracing are unrelated code paths." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44085", + "accept": true, + "reason": "Both target the GPT-J output-tracing refactor in the same file family; they look like overlapping versions of the same change." + }, + { + "left": "pull_request:44007", + "right": "pull_request:44072", + "accept": false, + "reason": "RegNet/ResNet refactors and EfficientNet refactors are separate model changes." + }, + { + "left": "pull_request:44072", + "right": "pull_request:44722", + "accept": false, + "reason": "EfficientNet and GPT-J are unrelated model code paths." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44072", + "accept": false, + "reason": "GPT-J/CodeGen output tracing is unrelated to EfficientNet output tracing." + }, + { + "left": "pull_request:44013", + "right": "pull_request:44044", + "accept": false, + "reason": "MobileNetV2 and DeBERTa are unrelated models; only the refactor pattern matches." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-J/CodeGen and MGP-STR are different code paths." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44071", + "accept": false, + "reason": "GPT-J/CodeGen and MPT are separate model implementations." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44068", + "accept": true, + "reason": "Both are GPT-Neo output-tracing refactors against the same model/file path, so they appear to be the same underlying change." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44066", + "accept": false, + "reason": "GPT-Neo and GPT-J are different models; same umbrella refactor is not enough." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44068", + "accept": false, + "reason": "GPT-J/CodeGen vs GPT-Neo are distinct model code paths." + }, + { + "left": "pull_request:44019", + "right": "pull_request:44722", + "accept": false, + "reason": "ResNet output tracing and GPT-J/CodeGen output tracing are unrelated changes." + }, + { + "left": "pull_request:44086", + "right": "pull_request:44722", + "accept": false, + "reason": "MGP-STR and GPT-J touch different model implementations." + }, + { + "left": "pull_request:44019", + "right": "pull_request:44085", + "accept": false, + "reason": "ResNet and GPT-J are different concrete model paths." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44019", + "accept": false, + "reason": "GPT-Neo and ResNet are different models; not a duplicate change." + }, + { + "left": "pull_request:44019", + "right": "pull_request:44071", + "accept": false, + "reason": "ResNet and MPT are unrelated implementations." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44116", + "accept": false, + "reason": "GPT-Neo and Flaubert are separate model-specific refactors." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44116", + "accept": false, + "reason": "GPT-Neo and Flaubert do not share the same code path." + }, + { + "left": "pull_request:44068", + "right": "pull_request:44722", + "accept": false, + "reason": "GPT-Neo output tracing is unrelated to GPT-J/CodeGen output tracing." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44722", + "accept": false, + "reason": "MPT and GPT-J/CodeGen are different model implementations." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44085", + "accept": false, + "reason": "GPT-Neo and GPT-J are different models." + }, + { + "left": "pull_request:44018", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-Neo and MGP-STR are unrelated code paths." + }, + { + "left": "pull_request:44098", + "right": "pull_request:44101", + "accept": false, + "reason": "ViLT and Flaubert are separate model-specific changes." + }, + { + "left": "pull_request:44059", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-2 and MGP-STR are unrelated concrete implementations." + }, + { + "left": "pull_request:44085", + "right": "pull_request:44086", + "accept": false, + "reason": "GPT-J and MGP-STR are different models." + }, + { + "left": "pull_request:44098", + "right": "pull_request:44161", + "accept": false, + "reason": "ViLT and LongT5 are unrelated model paths." + }, + { + "left": "pull_request:44073", + "right": "pull_request:44101", + "accept": false, + "reason": "VisualBert and Flaubert are different models." + }, + { + "left": "pull_request:44059", + "right": "pull_request:44071", + "accept": false, + "reason": "GPT-2 and MPT are unrelated code paths." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44076", + "accept": false, + "reason": "MPT and ImageGPT are separate model refactors." + }, + { + "left": "pull_request:44074", + "right": "pull_request:44161", + "accept": false, + "reason": "TextNet and LongT5 are different model implementations." + }, + { + "left": "pull_request:44071", + "right": "pull_request:44129", + "accept": false, + "reason": "MPT and SpeechT5 are unrelated concrete changes." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44073", + "accept": false, + "reason": "SqueezeBert and VisualBert are different models." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44018", + "accept": false, + "reason": "SqueezeBert and GPT-Neo do not share the same code path." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44056", + "accept": false, + "reason": "RegNet and MPNet are separate model-specific refactors." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44071", + "accept": false, + "reason": "SqueezeBert and MPT are unrelated implementations." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44074", + "accept": false, + "reason": "RegNet and TextNet are different models." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44161", + "accept": false, + "reason": "RegNet and LongT5 are unrelated changes." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44161", + "accept": false, + "reason": "MPNet and LongT5 are different model paths." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44017", + "accept": false, + "reason": "CVT/FNet refactors and Segformer refactors are separate model-specific changes." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44044", + "accept": false, + "reason": "SqueezeBert and DeBERTa are unrelated models." + }, + { + "left": "pull_request:44044", + "right": "pull_request:44074", + "accept": false, + "reason": "DeBERTa and TextNet are different code paths." + }, + { + "left": "pull_request:44066", + "right": "pull_request:44722", + "accept": true, + "reason": "Same GPT-J/CodeGen output-tracing change; 44722 looks like the consolidated version of the overlapping refactor." + }, + { + "left": "pull_request:44085", + "right": "pull_request:44722", + "accept": true, + "reason": "Both touch the same GPT-J output-tracing refactor in the same file area, so they are effectively duplicates." + }, + { + "left": "pull_request:44085", + "right": "pull_request:44129", + "accept": false, + "reason": "GPT-J and SpeechT5 are unrelated model changes." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44007", + "accept": false, + "reason": "CVT/FNet and RegNet/ResNet are different model-specific refactors." + }, + { + "left": "pull_request:43996", + "right": "pull_request:44129", + "accept": false, + "reason": "CVT/FNet and SpeechT5 are unrelated code paths." + }, + { + "left": "pull_request:44044", + "right": "pull_request:44129", + "accept": false, + "reason": "DeBERTa and SpeechT5 are different models." + }, + { + "left": "pull_request:44044", + "right": "pull_request:44154", + "accept": false, + "reason": "DeBERTa and VITS are separate model implementations." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "Grounded and conservative overall. The accepted soft edges stay within the same model/file families with clear diff overlap (GPT-J/CodeGen and GPT-Neo), while cross-family pairs are appropriately rejected. The summary does not overstate broader duplication beyond what the packet supports." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 14, + "component_count": 33, + "review_unit_index": 2, + "review_unit_count": 2, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-43996-28", + "nodes": [ + "pull_request:43997", + "pull_request:44010", + "pull_request:44024", + "pull_request:44056", + "pull_request:44073", + "pull_request:44074", + "pull_request:44076", + "pull_request:44129" + ], + "soft_pairs": [ + "pull_request:44024|pull_request:44076", + "pull_request:44073|pull_request:44074", + "pull_request:44056|pull_request:44073", + "pull_request:44056|pull_request:44074", + "pull_request:43997|pull_request:44010", + "pull_request:44010|pull_request:44129" + ], + "prepared_review_unit_hash": "52d22dd3050bb234fe5a8981fa793614fa7b9e6244a6f815e5f50b8f63728e30", + "component_budget": { + "node_count": 28, + "item_count": 28, + "soft_pair_count": 78, + "serialized_chars": 84896, + "estimated_input_tokens": 21224, + "estimated_eval_tokens": 42704 + }, + "budget": { + "node_count": 8, + "item_count": 8, + "soft_pair_count": 6, + "serialized_chars": 17125, + "estimated_input_tokens": 4282, + "estimated_eval_tokens": 8820 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": true, + "analyst_result": { + "summary": "All items are model-specific PRs that apply the same output-tracing refactor pattern to different code paths, not duplicates of one another. The cluster is best represented by the SpeechT5 PR because it is the most concrete and has an explicit issue target.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 44129 is the strongest representative: it has an explicit issue target, a clear standardized output-capture refactor, and enough context to describe the cluster\u2019s common theme.", + "best_issue_reason": null, + "best_pr_reason": "44129 is the most suitable standalone representative because it is self-contained, clearly scoped to one model, and anchored to a linked issue, whereas the others are similar but equally model-specific refactors.", + "soft_edge_verdicts": [ + { + "left": "pull_request:44024", + "right": "pull_request:44076", + "accept": false, + "reason": "Both are output-tracing refactors, but they touch different model implementations (FocalNet vs ImageGPT) and different code paths; not the same underlying change." + }, + { + "left": "pull_request:44073", + "right": "pull_request:44074", + "accept": false, + "reason": "Same refactor pattern, but VisualBert and TextNet are separate model-specific implementations; these are parallel changes, not duplicate PRs." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44073", + "accept": false, + "reason": "MPNet and VisualBert are distinct model code paths; the similarity is architectural, not a shared concrete bug or change." + }, + { + "left": "pull_request:44056", + "right": "pull_request:44074", + "accept": false, + "reason": "Both are standardized-output refactors, but they are separate model migrations in different files with no shared concrete fix to merge into one PR." + }, + { + "left": "pull_request:43997", + "right": "pull_request:44010", + "accept": false, + "reason": "RegNet and SqueezeBert are unrelated model implementations; the PRs share a refactor theme only, not the same underlying issue." + }, + { + "left": "pull_request:44010", + "right": "pull_request:44129", + "accept": false, + "reason": "SqueezeBert and SpeechT5 modify different model-specific forward/output plumbing, so they are not duplicate fixes even though both standardize output capture." + } + ] + }, + "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": 33, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-43998-13", + "nodes": [ + "pull_request:43998", + "pull_request:43999", + "pull_request:44000", + "pull_request:44001", + "pull_request:44002", + "pull_request:44003", + "pull_request:44004", + "pull_request:44025", + "pull_request:44026", + "pull_request:44027", + "pull_request:44028", + "pull_request:44029", + "pull_request:44030" + ], + "soft_pairs": [ + "pull_request:43998|pull_request:44002", + "pull_request:43998|pull_request:44000", + "pull_request:43998|pull_request:44001", + "pull_request:43998|pull_request:44003", + "pull_request:43998|pull_request:44029", + "pull_request:43999|pull_request:44001", + "pull_request:43999|pull_request:44002", + "pull_request:43999|pull_request:44025", + "pull_request:43999|pull_request:44026", + "pull_request:43999|pull_request:44027", + "pull_request:44000|pull_request:44001", + "pull_request:44000|pull_request:44002", + "pull_request:44000|pull_request:44003", + "pull_request:44000|pull_request:44025", + "pull_request:44001|pull_request:44002", + "pull_request:44001|pull_request:44003", + "pull_request:44001|pull_request:44025", + "pull_request:44002|pull_request:44029", + "pull_request:44002|pull_request:44003", + "pull_request:44003|pull_request:44004", + "pull_request:43998|pull_request:44004", + "pull_request:44001|pull_request:44004", + "pull_request:44002|pull_request:44004", + "pull_request:44000|pull_request:44004", + "pull_request:44025|pull_request:44026", + "pull_request:44025|pull_request:44027", + "pull_request:44025|pull_request:44028", + "pull_request:44025|pull_request:44029", + "pull_request:44025|pull_request:44030", + "pull_request:44026|pull_request:44029", + "pull_request:44026|pull_request:44027", + "pull_request:44026|pull_request:44028", + "pull_request:44026|pull_request:44030", + "pull_request:44027|pull_request:44028", + "pull_request:44027|pull_request:44029", + "pull_request:44027|pull_request:44030", + "pull_request:44028|pull_request:44029", + "pull_request:44028|pull_request:44030", + "pull_request:44029|pull_request:44030" + ], + "prepared_review_unit_hash": "12ae77e0d182a15f237b4e8ee98ea2c820851f0c9d062c0473b053e6b1f68a91", + "component_budget": { + "node_count": 13, + "item_count": 13, + "soft_pair_count": 39, + "serialized_chars": 39788, + "estimated_input_tokens": 9947, + "estimated_eval_tokens": 20150 + }, + "budget": { + "node_count": 13, + "item_count": 13, + "soft_pair_count": 39, + "serialized_chars": 39788, + "estimated_input_tokens": 9947, + "estimated_eval_tokens": 20150 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is a set of similar-looking but model-specific output-tracing refactors across unrelated code paths. They share a theme and some naming, but not the same underlying bug/change, so none of the soft edges should be merged as duplicates.", + "confidence": 0.93, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 44025 is a good cluster representative: it clearly shows the core refactor pattern (adding `can_return_tuple` and simplifying forward output handling) and spans two related files, making it a strong exemplar of the shared theme.", + "best_issue_reason": null, + "best_pr_reason": "PR 44025 is the best overall representative for the cluster because its diff is straightforward, clearly documents the refactor intent, and captures the common output-tracing change pattern without being overly specialized to a single code path.", + "soft_edge_verdicts": [ + { + "left": "pull_request:43998", + "right": "pull_request:44002", + "accept": false, + "reason": "Different model families and different files; same refactor theme only, not the same concrete bug or change." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44000", + "accept": false, + "reason": "Both are output-tracing refactors, but they target different architectures and code paths." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44001", + "accept": false, + "reason": "Unrelated model implementations; similar decorator/import cleanup is not enough to treat as one change." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44003", + "accept": false, + "reason": "Timm backbone vs Mamba/Falcon Mamba: different model behavior and separate forward paths." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44004", + "accept": false, + "reason": "CodeGen\u2019s forward/cache handling is a different change from timm_backbone output tracing." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44029", + "accept": false, + "reason": "These are distinct model stacks; similarity is limited to the generic output-tracing refactor pattern." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44001", + "accept": false, + "reason": "MobileNetV1 and UnivNet are unrelated models; not the same underlying fix." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44002", + "accept": false, + "reason": "Different model code paths and different output types; only broad refactor overlap." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44025", + "accept": false, + "reason": "MobileNetV1 vs Depth Anything: same repo-wide refactor style, but not a shared bug/change." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44026", + "accept": false, + "reason": "Separate model wrappers with different forward signatures and paths." + }, + { + "left": "pull_request:43999", + "right": "pull_request:44027", + "accept": false, + "reason": "MobileNetV1 and Speech Encoder-Decoder are unrelated; no shared concrete defect." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44001", + "accept": false, + "reason": "Vision-Text Dual Encoder and UnivNet are different model families; similarity is only in the refactor pattern." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44002", + "accept": false, + "reason": "Different output-tracing changes in distinct models, not one mergeable fix." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44003", + "accept": false, + "reason": "Vision-Text Dual Encoder and Mamba/Falcon Mamba touch different code paths and APIs." + }, + { + "left": "pull_request:44000", + "right": "pull_request:44025", + "accept": false, + "reason": "These are separate model implementations; the common theme is too broad to duplicate." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44002", + "accept": false, + "reason": "UnivNet and UperNet are unrelated; same cleanup style does not imply same underlying issue." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44003", + "accept": false, + "reason": "Different model stacks and semantics; not plausibly one combined PR." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44025", + "accept": false, + "reason": "Depth Anything vs UnivNet: no shared code path or concrete bug." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44026", + "accept": false, + "reason": "Unrelated model wrappers; similar decorator changes only." + }, + { + "left": "pull_request:44002", + "right": "pull_request:44003", + "accept": false, + "reason": "UperNet and Mamba/Falcon Mamba are different subsystems with different output handling." + }, + { + "left": "pull_request:44002", + "right": "pull_request:44004", + "accept": false, + "reason": "UperNet segmentation output refactor is not the same as CodeGen cache/output typing changes." + }, + { + "left": "pull_request:44003", + "right": "pull_request:44004", + "accept": false, + "reason": "Mamba/Falcon Mamba and CodeGen have different forward behavior and problem scope." + }, + { + "left": "pull_request:43998", + "right": "pull_request:44004", + "accept": false, + "reason": "Timm backbone tracing cleanup and CodeGen output refactor are unrelated concrete changes." + }, + { + "left": "pull_request:44001", + "right": "pull_request:44004", + "accept": false, + "reason": "UnivNet and CodeGen do not share the same code path or defect." + }, + { + "left": "pull_request:44002", + "right": "pull_request:44004", + "accept": false, + "reason": "Different model families and different return semantics." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44026", + "accept": false, + "reason": "Depth Anything and Vision/Encoder-Decoder are distinct model wrappers; same refactor theme only." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44027", + "accept": false, + "reason": "Depth Anything and Speech Encoder-Decoder are not the same underlying bug/change." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44028", + "accept": false, + "reason": "Depth estimation vs SuperPoint keypoint model: unrelated implementation paths." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44029", + "accept": false, + "reason": "Depth Anything and Mamba/Falcon Mamba are different subsystems; no shared fix." + }, + { + "left": "pull_request:44025", + "right": "pull_request:44030", + "accept": false, + "reason": "Depth Anything and DPR touch different model behaviors and output contracts." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44027", + "accept": false, + "reason": "Vision Encoder-Decoder and Speech Encoder-Decoder are separate wrappers; similarity is broad architectural pattern only." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44028", + "accept": false, + "reason": "Different tasks and code paths; not the same concrete issue." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44029", + "accept": false, + "reason": "Distinct model families; no evidence they fix the same bug." + }, + { + "left": "pull_request:44026", + "right": "pull_request:44030", + "accept": false, + "reason": "Vision/Speech encoder-decoder wrappers are unrelated to DPR's output handling." + }, + { + "left": "pull_request:44027", + "right": "pull_request:44028", + "accept": false, + "reason": "Speech Encoder-Decoder and SuperPoint are unrelated models with different forward contracts." + }, + { + "left": "pull_request:44027", + "right": "pull_request:44029", + "accept": false, + "reason": "Different model families and output mechanisms." + }, + { + "left": "pull_request:44027", + "right": "pull_request:44030", + "accept": false, + "reason": "Speech Encoder-Decoder and DPR are separate code paths; not a duplicate." + }, + { + "left": "pull_request:44028", + "right": "pull_request:44029", + "accept": false, + "reason": "SuperPoint and Mamba/Falcon Mamba are unrelated; same refactor motif is insufficient." + }, + { + "left": "pull_request:44028", + "right": "pull_request:44030", + "accept": false, + "reason": "SuperPoint keypoint model vs DPR retrieval model: different concrete changes." + }, + { + "left": "pull_request:44029", + "right": "pull_request:44030", + "accept": false, + "reason": "Mamba/Falcon Mamba and DPR do not share the same underlying bug or change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 16, + "component_count": 33, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-44662-2", + "nodes": [ + "pull_request:44662", + "pull_request:45401" + ], + "soft_pairs": [ + "pull_request:44662|pull_request:45401" + ], + "prepared_review_unit_hash": "15e041515752d0ca8f0dc56774432caae3973f9bc02c45f63f795910be52f14b", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6551, + "estimated_input_tokens": 1638, + "estimated_eval_tokens": 3532 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6551, + "estimated_input_tokens": 1638, + "estimated_eval_tokens": 3532 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two unrelated model-addition PRs: one adds PenguinVL, the other adds Voxtral TTS support. They share only broad Transformers integration files and docs plumbing, not the same underlying feature or bug.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #45401 is the more self-contained representative: it targets a specific tracked issue, has a focused scope around Voxtral TTS support, and includes the full implementation/testing stack for that model.", + "best_issue_reason": null, + "best_pr_reason": "PR #45401 is the best cluster representative because it is the more explicit, issue-backed feature addition and is easier to interpret as the cluster's exemplar model-support PR.", + "soft_edge_verdicts": [ + { + "left": "pull_request:44662", + "right": "pull_request:45401", + "accept": false, + "reason": "Different models and code paths: PenguinVL vs Voxtral TTS. Shared files are only common Transformers registration/documentation surfaces, which is too broad to indicate the same underlying change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 17, + "component_count": 33, + "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-44979-4", + "nodes": [ + "pull_request:44979", + "pull_request:45363", + "pull_request:45666", + "pull_request:46339" + ], + "soft_pairs": [ + "pull_request:44979|pull_request:45363", + "pull_request:45363|pull_request:46339", + "pull_request:45666|pull_request:46339" + ], + "prepared_review_unit_hash": "f290302ae2b505321b825f2508a04d3f5f0eb1744efeafc904120166a57c02b4", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 9634, + "estimated_input_tokens": 2409, + "estimated_eval_tokens": 5074 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 3, + "serialized_chars": 9634, + "estimated_input_tokens": 2409, + "estimated_eval_tokens": 5074 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster is mostly a chain of related kernel-fusion PRs. The latest PR, 46339, looks like the best canonical PR for the concrete n-to-1 KernelConfig fusion feature, while 44979 is a separate lower-level module-fusion API and should not be merged with the later KernelConfig work.", + "confidence": 0.87, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46339 is the most complete and latest version of the same n-to-1 KernelConfig fusion feature, adding the broader integration/simplification work that builds on the earlier PRs.", + "best_issue_reason": null, + "best_pr_reason": "46339 is the best standalone representative because it appears to subsume the earlier KernelConfig fusion iterations and is the most polished implementation of the shared code-path change.", + "soft_edge_verdicts": [ + { + "left": "pull_request:44979", + "right": "pull_request:45363", + "accept": false, + "reason": "44979 introduces a foundational Module Fusion API, while 45363 adds a specific KernelConfig-based n-to-1 kernel fusion integration; related, but not the same concrete change." + }, + { + "left": "pull_request:45363", + "right": "pull_request:46339", + "accept": true, + "reason": "Both target the same concrete n-to-1 kernel fusion via KernelConfig path, touch overlapping integration files, and 46339 reads like a follow-up simplification/extension of 45363." + }, + { + "left": "pull_request:45666", + "right": "pull_request:46339", + "accept": true, + "reason": "These are the same KernelConfig-based n-to-1 fusion feature area, with 46339 appearing to extend and simplify the exact mechanism introduced in 45666." + } + ] + }, + "evaluator_result": { + "accept": true, + "feedback": "The summary is well grounded in the packet: 44979 is clearly the separate Module Fusion API PR, while 45363/45666/46339 all center on KernelConfig-based n-to-1 kernel fusion and share overlapping files/titles. The soft-edge verdicts are conservative and do not overstate the evidence." + }, + "evaluator_used": true, + "retried": false, + "accepted_nontrivial_soft_edge": true, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 18, + "component_count": 33, + "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": "61cbe46094d9d2184a0bec3bb0f07407b7bd7d745ab33d022c2c7c9d62954757", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4414, + "estimated_input_tokens": 1104, + "estimated_eval_tokens": 2464 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4414, + "estimated_input_tokens": 1104, + "estimated_eval_tokens": 2464 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both pull requests target the same Mistral regex patch bug in `tokenization_utils_tokenizers.py`; the later PR adds tests and is the stronger canonical candidate.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "No issue node is present. Between the PRs, #45317 is the better canonical PR because it fixes the same AttributeError but also adds test coverage, making it the more complete change.", + "best_issue_reason": null, + "best_pr_reason": "PR #45317 is the best overall representative: it addresses the same concrete `_patch_mistral_regex` AttributeError and includes validation tests, while #45086 is a narrower earlier version of the same fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45086", + "right": "pull_request:45317", + "accept": true, + "reason": "Same bug, same code path, same file, and nearly identical fix; #45317 is essentially an expanded version of #45086 with tests, so they are duplicate/mergeable." + } + ] + }, + "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": 33, + "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": "25ca93cd0c9d3a9f49c7d80a6c84dcda1e42b265617ee910f4bbbe9c8bfb6254", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6066, + "estimated_input_tokens": 1517, + "estimated_eval_tokens": 3290 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 6066, + "estimated_input_tokens": 1517, + "estimated_eval_tokens": 3290 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related NemotronH model work items, but they are not the same fix: one is a narrower refactor/conversion-mapping update, while the other splits NemotronH into dense/sparse variants with broader API, docs, and test changes.", + "confidence": 0.91, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 45569 is the better canonical PR because it is the broader, more complete model-structure change: it introduces the dense/sparse variants, updates auto/model registrations, adds docs, and includes tests. PR 45115 is narrower and mainly refactors inheritance/conversion mappings.", + "best_issue_reason": null, + "best_pr_reason": "PR 45569 is the best representative of the cluster since it addresses the main architectural change and touches the full user-facing surface. PR 45115 looks like a partial supporting refactor rather than the primary end-state implementation.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45115", + "right": "pull_request:45569", + "accept": false, + "reason": "They overlap on NemotronH internals, but 45115 is about inheritance/conversion mapping cleanup for NemotronH, while 45569 is a larger split into NemotronH dense/sparse variants with docs, auto-mapping, and tests. Not the same concrete change and not safely mergeable into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 20, + "component_count": 33, + "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": "fbef38615b0bdc887bdabbda27a14d8ef4bbd53418cdad7f8b46a2de1920bc78", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5466, + "estimated_input_tokens": 1367, + "estimated_eval_tokens": 2990 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5466, + "estimated_input_tokens": 1367, + "estimated_eval_tokens": 2990 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two related but distinct PRs for heterogeneous model support: one adds per-layer config infrastructure, the other adds heterogeneous modeling/cache changes. They overlap on shared config plumbing but are not the same concrete change.", + "confidence": 0.84, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #45332 is the broader umbrella change and includes the modeling-side integration; #45333 is a narrower config-only foundation. If picking one representative, #45332 is the better canonical PR, but they are not duplicates.", + "best_issue_reason": null, + "best_pr_reason": "#45332 is the best representative of the cluster because it spans the larger end-to-end heterogeneous model support surface, while #45333 is a prerequisite/config subset.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45332", + "right": "pull_request:45333", + "accept": false, + "reason": "They share heterogeneous-config plumbing, but one is a broad modeling/cache/masking PR and the other is a config-only PR. Different concrete code-paths and not safe to treat 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": 21, + "component_count": 33, + "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": "19e3bcce600d7e3f35500917b22fadb3a3354b17d5227fe61d35da67962e0bee", + "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 items are typing-related PRs, but they address different scopes: #45415 is a broad repo-wide type-checking sweep, while #45425 is a more targeted typing/refactor PR centered on modeling_utils and shared typing protocols. They are not duplicates.", + "confidence": 0.94, + "canonical_issue_reason": null, + "canonical_pr_reason": "No clear duplicate PR in this cluster. If one must be selected, #45425 is the better representative because it is a focused, self-contained typing change, whereas #45415 is a wide-ranging type-checking cleanup across many files.", + "best_issue_reason": null, + "best_pr_reason": "#45425 is the best representative of the cluster because it is narrowly scoped and technically coherent; #45415 is broader and mixes many unrelated typing edits, making it a poorer canonical stand-in.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45415", + "right": "pull_request:45425", + "accept": false, + "reason": "Shared typing-related files are not enough here: #45415 is a broad typing sweep, while #45425 introduces specific typing protocols and annotations. They do not look like the same underlying fix and would not plausibly merge into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 22, + "component_count": 33, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45513-2", + "nodes": [ + "pull_request:45513", + "pull_request:46084" + ], + "soft_pairs": [ + "pull_request:45513|pull_request:46084" + ], + "prepared_review_unit_hash": "8d6dd30b13f1bae327f83e28b182395920e3236527105fc96895cef4ab218543", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4858, + "estimated_input_tokens": 1215, + "estimated_eval_tokens": 2686 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4858, + "estimated_input_tokens": 1215, + "estimated_eval_tokens": 2686 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs address cache-enabled forward paths with seq_len > 1, but they target different model families and distinct code paths, so they do not look like the same underlying fix.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46084 is the cleaner representative: it is narrowly scoped to Mamba2 cache handling and explicitly tied to issue #46032.", + "best_issue_reason": null, + "best_pr_reason": "PR 46084 is the best single PR to represent this cluster because it is tightly scoped, issue-linked, and more clearly describes one concrete bug fix.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45513", + "right": "pull_request:46084", + "accept": false, + "reason": "Different architectures and different implementations: Qwen/OLMo linear-attention cached forward vs Mamba2 CUDA-kernel cache handling. Same broad theme, not the same concrete bug." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 23, + "component_count": 33, + "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": "6c3cb2761dce33a74873dcb9c6035c69d7d54407de6a5a7aa75e73798fdcb76f", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4507, + "estimated_input_tokens": 1127, + "estimated_eval_tokens": 2510 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4507, + "estimated_input_tokens": 1127, + "estimated_eval_tokens": 2510 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are related by expert-parallel loading infrastructure, but they fix different bugs: PR 45548 is about preserving EP/DeepSpeed ZeRO-3 loading metadata through accelerate launch, while PR 45662 targets expert parameters being overwritten during FSDP2 loading. Not a duplicate pair.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 45662 is the better cluster representative because it addresses a concrete expert-parameter corruption bug and includes the broader set of loading-path changes around that failure mode.", + "best_issue_reason": null, + "best_pr_reason": "PR 45662 is the strongest overall representative: its title and diff describe a specific, high-signal EP loading bug with direct code-path fixes, whereas 45548 is narrower and centered on config serialization/launch behavior.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45548", + "right": "pull_request:45662", + "accept": false, + "reason": "Shared expert-parallel context and one common file are not enough; 45548 fixes config serialization/launch handling for ZeRO-3, while 45662 fixes expert tensors being overwritten in FSDP2 loading. Different underlying bugs and not plausibly one merged PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 24, + "component_count": 33, + "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": "b401dbd0659916b5294fc7147db4101cd24cf9e6aa07295e15cf0f3377db2caf", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4589, + "estimated_input_tokens": 1148, + "estimated_eval_tokens": 2552 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4589, + "estimated_input_tokens": 1148, + "estimated_eval_tokens": 2552 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Both PRs address the same xdist collision problem around captured_info output in testing_utils; PR 45645 appears to be the broader, more complete version with CI log aggregation added.", + "confidence": 0.79, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 45645 is the better canonical PR because it covers the core xdist-safe output fix plus the downstream CI/workflow handling needed to read multiple captured_info files.", + "best_issue_reason": null, + "best_pr_reason": "PR 45645 is the strongest representative: it addresses the same underlying captured_info xdist collision and also preserves/debug-prints the resulting artifacts in CI.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45639", + "right": "pull_request:45645", + "accept": true, + "reason": "Same underlying bug: xdist workers colliding on captured_info output for patched testing logs. PR 45645 extends the same fix into CI artifact collection, so these are plausibly one combined change." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 25, + "component_count": 33, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45670-3", + "nodes": [ + "pull_request:45670", + "pull_request:46320", + "pull_request:46401" + ], + "soft_pairs": [ + "pull_request:45670|pull_request:46320", + "pull_request:46320|pull_request:46401" + ], + "prepared_review_unit_hash": "7307da740cca9642070b10d873f943f0e2e69e50b2858d5ec6f3993400c75195", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 7029, + "estimated_input_tokens": 1758, + "estimated_eval_tokens": 3772 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 7029, + "estimated_input_tokens": 1758, + "estimated_eval_tokens": 3772 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "The cluster is mostly about model-specific attn_implementation dispatch support, plus one unrelated AutoModel registration nit. The RecurrentGemma and BLIP-2 PRs are similar in theme but target different code paths and should not be treated as duplicates.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46320 is the best representative of the cluster\u2019s main theme: a focused model-specific attn_implementation dispatch implementation for RecurrentGemma, with supporting conversion and tests.", + "best_issue_reason": null, + "best_pr_reason": "PR 46320 is the strongest canonical PR because it is the clearest, most self-contained example of the attention-dispatch feature work in this cluster.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45670", + "right": "pull_request:46320", + "accept": false, + "reason": "Different changes: AutoModel mapping for glmasr versus RecurrentGemma attention-dispatch support. They do not fix the same code-path problem." + }, + { + "left": "pull_request:46320", + "right": "pull_request:46401", + "accept": false, + "reason": "Both add attn_implementation dispatch, but for different model families and different implementations. They are similar feature work, not the same concrete bug/change, and would not naturally merge into one PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 26, + "component_count": 33, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45720-2", + "nodes": [ + "pull_request:45720", + "pull_request:46259" + ], + "soft_pairs": [ + "pull_request:45720|pull_request:46259" + ], + "prepared_review_unit_hash": "b326c2c80e43af865f995d624df965cc2cce76de41b05bb9696d24425e0c1268", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 7188, + "estimated_input_tokens": 1797, + "estimated_eval_tokens": 3850 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 7188, + "estimated_input_tokens": 1797, + "estimated_eval_tokens": 3850 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two unrelated open PRs were clustered only by soft similarity. PR #45720 implements TRF003 tuple/return_dict handling across model forwards, while PR #46259 adds a different modeling rule about do_* flags in image/video processors. They do not look like the same underlying change.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": "No canonical PR as a duplicate target: the PRs address different lint rules and different code paths, so they should not be merged or treated as one change.", + "best_issue_reason": null, + "best_pr_reason": "PR #45720 is the cleaner representative of the cluster if one must be chosen, because it is a focused code-path fix across model forwards; PR #46259 is a broader, different rule addition spanning processors and docs.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45720", + "right": "pull_request:46259", + "accept": false, + "reason": "Different underlying tasks: tuple/return_dict forward cleanup vs do_* flag consumption in image/video processors. Shared filenames are incidental and not evidence of the same fix." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 27, + "component_count": 33, + "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": 28, + "component_count": 33, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-45932-2", + "nodes": [ + "pull_request:45932", + "pull_request:46100" + ], + "soft_pairs": [ + "pull_request:45932|pull_request:46100" + ], + "prepared_review_unit_hash": "701eba2c86a20efc1e8a909c4ba2bc1bf500df8228552da7ef88b42407d7c6ab", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4301, + "estimated_input_tokens": 1076, + "estimated_eval_tokens": 2408 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4301, + "estimated_input_tokens": 1076, + "estimated_eval_tokens": 2408 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are not duplicates. PR #45932 is a broad model/config/workflow change, while PR #46100 is a tiny test-only change that intentionally adds failing assertions and targets a different code path. They should not be clustered as the same underlying change.", + "confidence": 0.98, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR #45932 is the better canonical PR because it contains the substantive code changes across model implementations and workflow/secrets plumbing, whereas PR #46100 is just a test-migration stub with deliberate failures.", + "best_issue_reason": null, + "best_pr_reason": "PR #45932 is the most representative and materially meaningful change in the set; PR #46100 is a narrow test-only edit and is not a good canonical representative.", + "soft_edge_verdicts": [ + { + "left": "pull_request:45932", + "right": "pull_request:46100", + "accept": false, + "reason": "Different purpose and code paths: #45932 changes multiple model implementations and CI secrets, while #46100 only edits a ViT test file and adds failing assertions. They do not appear to fix the same bug or implement the same feature." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 29, + "component_count": 33, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46073-3", + "nodes": [ + "pull_request:46073", + "pull_request:46154", + "pull_request:46196" + ], + "soft_pairs": [ + "pull_request:46073|pull_request:46154", + "pull_request:46073|pull_request:46196", + "pull_request:46154|pull_request:46196" + ], + "prepared_review_unit_hash": "f605969488e985877dc33290309e829e6f53dd1992380ea53f75f1975535c6e4", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 8083, + "estimated_input_tokens": 2021, + "estimated_eval_tokens": 4298 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 3, + "serialized_chars": 8083, + "estimated_input_tokens": 2021, + "estimated_eval_tokens": 4298 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These are three separate PRs adding SDPA dispatch support to different model families (CTRL, T5, and wav2vec2/SeamlessM4T). They share a broad theme but not the same concrete bug or code path, so they should not be clustered as duplicates.", + "confidence": 0.96, + "canonical_issue_reason": null, + "canonical_pr_reason": "No PR is a true duplicate of the others. If a representative must be chosen, 46154 is the cleanest single-model PR and the most self-contained of the set.", + "best_issue_reason": null, + "best_pr_reason": "46154 is the most focused and least broad: it targets one model family (T5) with a clear SDPA dispatch change, making it the best representative of the cluster, though not a duplicate of the others.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46073", + "right": "pull_request:46154", + "accept": false, + "reason": "Both are SDPA-related, but they modify different model implementations (CTRL vs T5) and different code paths. Same theme, not the same underlying fix." + }, + { + "left": "pull_request:46073", + "right": "pull_request:46196", + "accept": false, + "reason": "CTRL SDPA support is unrelated to the wav2vec2_bert / wav2vec2_conformer / SeamlessM4T changes. The PRs touch different architectures and cannot plausibly be merged as one fix." + }, + { + "left": "pull_request:46154", + "right": "pull_request:46196", + "accept": false, + "reason": "T5 dispatch support and the multi-model wav2vec2/SeamlessM4T dispatch work address different model-specific implementations, so they are not duplicates." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 30, + "component_count": 33, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46126-2", + "nodes": [ + "pull_request:46126", + "pull_request:46269" + ], + "soft_pairs": [ + "pull_request:46126|pull_request:46269" + ], + "prepared_review_unit_hash": "41468e419c5c5cb67fcf6a981df82e5667fca5841c45d993f3c5e15f21d04f15", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5571, + "estimated_input_tokens": 1393, + "estimated_eval_tokens": 3042 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 5571, + "estimated_input_tokens": 1393, + "estimated_eval_tokens": 3042 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "Two pull requests overlap on distributed/tensor-parallel files, but they are not the same change: PR 46126 is a focused EP + Trainer integration patch, while PR 46269 is a broad distributed branch base touching many unrelated areas. This is not a duplicate pair.", + "confidence": 0.95, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46126 is the more focused, self-contained change set centered on EP and Trainer integration; PR 46269 is a large umbrella branch with many unrelated distributed updates, so it is a poorer canonical duplicate target.", + "best_issue_reason": null, + "best_pr_reason": "PR 46126 is the best representative of a concrete code-path change. PR 46269 is too broad and mixed to serve as a clean duplicate anchor.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46126", + "right": "pull_request:46269", + "accept": false, + "reason": "Shared filenames and soft similarity are not enough here: 46126 implements a specific EP/Trainer integration, while 46269 is a large distributed branch base spanning many subsystems. They do not look like the same underlying bug or mergeable single PR." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 31, + "component_count": 33, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46152-2", + "nodes": [ + "pull_request:46152", + "pull_request:46207" + ], + "soft_pairs": [ + "pull_request:46152|pull_request:46207" + ], + "prepared_review_unit_hash": "b06bec4a63189d5d5108cb1eec88b0f437d7b18ec7e8b55b375ab80503d5f269", + "component_budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4276, + "estimated_input_tokens": 1069, + "estimated_eval_tokens": 2394 + }, + "budget": { + "node_count": 2, + "item_count": 2, + "soft_pair_count": 1, + "serialized_chars": 4276, + "estimated_input_tokens": 1069, + "estimated_eval_tokens": 2394 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are related only at a high level: both stop using direct file-open source inspection, but they fix different code paths in different modules with different mechanisms. They should not be merged as duplicates.", + "confidence": 0.89, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46207 is the better representative of the cluster because it targets one concrete source-reading check in `modeling_utils.py` and includes tests, whereas PR 46152 is a broader import-utils refactor with a different implementation strategy.", + "best_issue_reason": null, + "best_pr_reason": "PR 46207 is the strongest standalone candidate: it addresses a specific bug-prone source inspection path and is narrowly scoped enough to evaluate on its own.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46152", + "right": "pull_request:46207", + "accept": false, + "reason": "Different underlying changes: 46152 refactors internal source-path resolution in `import_utils.py`, while 46207 changes `_can_set_attn_implementation` in `modeling_utils.py` from file-open to `inspect.getsource`. Same broad theme, but not the same concrete bug or patch." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 32, + "component_count": 33, + "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-46337-4", + "nodes": [ + "pull_request:46337", + "pull_request:46362", + "pull_request:46366", + "pull_request:46368" + ], + "soft_pairs": [ + "pull_request:46337|pull_request:46368", + "pull_request:46337|pull_request:46362", + "pull_request:46362|pull_request:46368", + "pull_request:46362|pull_request:46366" + ], + "prepared_review_unit_hash": "daf9670b7a981f8840e7eed5a9545e310174299c0652fcb79e09e0a428fa375b", + "component_budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 4, + "serialized_chars": 8227, + "estimated_input_tokens": 2057, + "estimated_eval_tokens": 4370 + }, + "budget": { + "node_count": 4, + "item_count": 4, + "soft_pair_count": 4, + "serialized_chars": 8227, + "estimated_input_tokens": 2057, + "estimated_eval_tokens": 4370 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "These PRs are mostly unrelated model-specific test adjustments. The only close soft link is between the two DeepSeek V4 PRs, but they fix different problems: one skips an incompatible quant-cache test, the other adds XPU/CUDA backend selection for distributed testing. I would not merge any of the soft pairs as duplicates.", + "confidence": 0.92, + "canonical_issue_reason": null, + "canonical_pr_reason": "PR 46366 is the most substantive standalone change in the set: it addresses distributed-test backend selection and is more representative of an actual code-path fix than the model-specific skip-only PRs.", + "best_issue_reason": null, + "best_pr_reason": "PR 46366 is the strongest standalone candidate to keep because it fixes environment/backend handling rather than merely skipping a failing test, and it is not tied to a single model-specific incompatibility.", + "soft_edge_verdicts": [ + { + "left": "pull_request:46337", + "right": "pull_request:46368", + "accept": false, + "reason": "Both are model-test maintenance PRs, but they target different models and different failure modes: Cohere2 MoE test setup vs NemotronH quant-cache skipping." + }, + { + "left": "pull_request:46337", + "right": "pull_request:46362", + "accept": false, + "reason": "Different models and different changes. One adjusts Cohere2 MoE test configuration; the other skips a DeepSeek V4 quant-cache test." + }, + { + "left": "pull_request:46362", + "right": "pull_request:46368", + "accept": false, + "reason": "Superficially similar skips, but they are for different model code paths and different cache implementations; not the same underlying bug/change." + }, + { + "left": "pull_request:46362", + "right": "pull_request:46366", + "accept": false, + "reason": "Same file, but the changes are unrelated: one skips a quant-cache test, the other adds XPU fallback logic for a distributed worker." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + }, + { + "label": "pull_request", + "component_index": 33, + "component_count": 33, + "review_unit_index": 1, + "review_unit_count": 1, + "status": "reviewed", + "reason": null, + "source": "llm", + "cache_hit": false, + "model": "gpt-5.4-mini?service_tier=flex", + "cluster_id": "cluster-46376-3", + "nodes": [ + "pull_request:46376", + "pull_request:46417", + "pull_request:46425" + ], + "soft_pairs": [ + "pull_request:46376|pull_request:46417", + "pull_request:46376|pull_request:46425" + ], + "prepared_review_unit_hash": "18e3dad1947182c71e5c4dd657bec9ca5de2262d5b8e1c24c9c5a379958ead4d", + "component_budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 9319, + "estimated_input_tokens": 2330, + "estimated_eval_tokens": 4916 + }, + "budget": { + "node_count": 3, + "item_count": 3, + "soft_pair_count": 2, + "serialized_chars": 9319, + "estimated_input_tokens": 2330, + "estimated_eval_tokens": 4916 + }, + "overflow_policy": "truncate_then_skip", + "trimmed": false, + "aggressively_trimmed": false, + "split": false, + "analyst_result": { + "summary": "This cluster groups three unrelated PRs: multimodal continuous batching infrastructure, a new HunYuan VL model integration, and RADIO model support. They touch different codepaths and represent separate feature additions, not duplicate fixes.", + "confidence": 0.97, + "canonical_issue_reason": null, + "canonical_pr_reason": null, + "best_issue_reason": null, + "best_pr_reason": null, + "soft_edge_verdicts": [ + { + "left": "pull_request:46376", + "right": "pull_request:46417", + "accept": false, + "reason": "Different changesets and goals: continuous batching multimodal support vs. adding a new HunYuan VL model package and docs. No shared concrete bug or mergeable duplicate implementation." + }, + { + "left": "pull_request:46376", + "right": "pull_request:46425", + "accept": false, + "reason": "These are unrelated feature PRs: one modifies continuous batching internals, the other adds RADIO model support across model files and auto mappings. They do not fix the same underlying problem." + } + ] + }, + "evaluator_result": null, + "evaluator_used": false, + "retried": false, + "accepted_nontrivial_soft_edge": false, + "error_kind": null, + "error_message": null + } + ] +} diff --git a/snapshots/20260605T000034Z/analysis-runs/hybrid-model-20260605t000034z/manifest.json b/snapshots/20260605T000034Z/analysis-runs/hybrid-model-20260605t000034z/manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..ed7d165d56041650561aaf6f94edb193b991f690 --- /dev/null +++ b/snapshots/20260605T000034Z/analysis-runs/hybrid-model-20260605t000034z/manifest.json @@ -0,0 +1,14 @@ +{ + "analysis_id": "hybrid-model-20260605t000034z", + "artifacts": { + "hybrid": "snapshots/20260605T000034Z/analysis-runs/hybrid-model-20260605t000034z/analysis-report-hybrid.json", + "hybrid_reviews": "snapshots/20260605T000034Z/analysis-runs/hybrid-model-20260605t000034z/analysis-report-hybrid.llm-reviews.json" + }, + "channel": "canonical", + "model": null, + "published_at": "2026-06-05T00:09:30Z", + "repo": "huggingface/transformers", + "schema_version": 1, + "snapshot_id": "20260605T000034Z", + "variant": "hybrid" +} diff --git a/snapshots/20260605T000034Z/manifest.json b/snapshots/20260605T000034Z/manifest.json index ed45dd6b580b9803886e99222798721fbbd902f2..7765f70b64799f4fffeb6434c9a2328ed50ea7db 100644 --- a/snapshots/20260605T000034Z/manifest.json +++ b/snapshots/20260605T000034Z/manifest.json @@ -32,6 +32,24 @@ "timeline_events": 100 }, "extracted_at": "2026-06-05T00:00:34Z", + "published_analysis": { + "canonical_analysis_id": "hybrid-model-20260605t000034z", + "runs": { + "hybrid-model-20260605t000034z": { + "analysis_id": "hybrid-model-20260605t000034z", + "artifacts": { + "hybrid": "snapshots/20260605T000034Z/analysis-runs/hybrid-model-20260605t000034z/analysis-report-hybrid.json", + "hybrid_reviews": "snapshots/20260605T000034Z/analysis-runs/hybrid-model-20260605t000034z/analysis-report-hybrid.llm-reviews.json" + }, + "channel": "canonical", + "manifest_path": "snapshots/20260605T000034Z/analysis-runs/hybrid-model-20260605t000034z/manifest.json", + "model": null, + "published_at": "2026-06-05T00:09:30Z", + "variant": "hybrid" + } + }, + "schema_version": 1 + }, "repo": "huggingface/transformers", "snapshot_id": "20260605T000034Z", "watermark": {