{"title":"Hugging Face","team_label":"HF team","days":365,"space":"davanstrien/issue-credit","generated":"2026-11-06","since":"2025-06-01","repos":["huggingface/transformers","huggingface/trl","huggingface/lerobot","huggingface/diffusers","huggingface/huggingface_hub","huggingface/peft","huggingface/smolagents","huggingface/candle","huggingface/datasets","huggingface/transformers.js","huggingface/kernels","huggingface/accelerate","huggingface/sentence-transformers","huggingface/lighteval","huggingface/tokenizers","huggingface/huggingface.js","huggingface/xet-core","huggingface/text-embeddings-inference","huggingface/optimum-intel","huggingface/hub-docs","huggingface/chat-ui","huggingface/pytorch-image-models"],"team":{"3outeille":"Member","Abdennacer-Badaoui":"Member","AmineDiro":"Member","Ankita7033":"Owner","ArthurZucker":"Member","BenjaminBossan":"Member","CarolinePascal":"Member","Cyrilvallez":"Member","DN6":"Member","ErenAta16":"Owner","GDemay":"Owner","HaomingSong":"Collaborator","Hugoch":"Member","IlyasMoutawwakil":"Member","Isalia20":"Member","JingyaHuang":"Member","Kakulukian":"Member","Krishnachaitanyakc":"Owner","LThmly":"Owner","LeonEricsson":"Collaborator","LysandreJik":"Member","Mahnoor-Zaffar":"Owner","Maximellerbach":"Member","McPatate":"Member","Moon-Wrecker":"Owner","NathanHB":"Member","NielsRogge":"Member","NikodemBartnik":"Member","Olcmyk":"Owner","Pierrci":"Member","Rocketknight1":"Member","SBrandeis":"Member","Shr3yash":"Owner","SparshGarg999":"Owner","SunMarc":"Member","Vitaliy-Pikalo":"Owner","Wauplin":"Member","aastha44ma":"Owner","adarshxs":"Member","adithya-s-k":"Member","albertvillanova":"Member","alejandro-ao":"Member","alvarobartt":"Member","apolinario":"Member","ariG23498":"Member","as-suvorov":"Collaborator","asomoza":"Member","assafvayner":"Member","aymeric-roucher":"Collaborator","behroozazarkhalili":"Collaborator","cfahlgren1":"Member","clefourrier":"Member","clmnt":"Member","cmpatino":"Member","coyotte508":"Member","dacorvo":"Member","danaaubakirova":"Member","danieldk":"Member","davanstrien":"Member","deep1401":"Member","drbh":"Member","dwarez":"Member","echarlaix":"Member","edbeeching":"Member","eustlb":"Member","evalstate":"Member","gary149":"Member","githubnemo":"Member","glutamatt":"Member","guarin":"Member","hanouticelina":"Member","hmellor":"Member","hysts":"Member","imstevenpmwork":"Member","itazap":"Member","ivarflakstad":"Member","juanjucm":"Member","julien-c":"Member","kashif":"Member","kefranabg":"Member","kszucs":"Member","lewtun":"Member","lhoestq":"Member","likholat":"Collaborator","lindicaphxag-tech":"Owner","linoytsaban":"Member","mariosasko":"Collaborator","masato-ka":"Collaborator","maslinedwin":"Owner","merveenoyan":"Member","mishig25":"Member","molbap":"Member","nico-martin":"Member","nicolas-rabault":"Member","omsherikar":"Owner","oyoyo4556":"Owner","pagezyhf":"Member","paulinebm":"Member","pkooij":"Member","pngwn":"Member","popovaan":"Collaborator","pramodith":"Collaborator","priyanka25aug":"Owner","qgallouedec":"Member","rajatarya":"Member","ravediamond":"Collaborator","remi-or":"Member","rwightman":"Collaborator","s1lent4gnt":"Member","sayakpaul":"Member","seanses":"Member","sergiopaniego":"Member","setu4993":"Member","sezan92":"Owner","sivaratrisrinivas":"Owner","snorkelopstesting1-a11y":"Collaborator","somdipto":"Owner","srlynch1":"Owner","stefan-it":"Collaborator","stevhliu":"Member","syahra712":"Owner","tanishaj12":"Owner","tarekziade":"Member","tc-huang":"Collaborator","tengomucho":"Member","tomaarsen":"Member","tomsanbear":"Collaborator","vasqu":"Member","wali-reheman":"Owner","winglian":"Collaborator","wrosko":"Owner","xenova":"Member","yashb98":"Owner","ydshieh":"Member","yiyixuxu":"Member","younesbelkada":"Member","young1lin":"Owner","zucchini-nlp":"Member","zuoxingdong":"Collaborator"},"issues":[[0,50434,"`tokenizer.backend` doesn't match the tokenizer class","samyuktahegde",4,"","",0,"2026-11-06",""],[0,50426,"[Bug] Mixed-precision compressed-tensors checkpoints randomly initialize converted modules the checkpoint leav","kylesayrs",4,"","",0,"2026-11-06",""],[0,50424,"`GradientCheckpointingLayer` checkpoints FSDP2 hooks; `use_reentrant=True` silently drops gradients when a lay","weifengpy",4,"","",0,"2026-11-05",""],[0,50419,"Packed sequences silently leak attention and training gradients when use_cache=True initializes an empty KV ca","BlazeSR10hawk",4,"","",0,"2026-11-05",""],[0,50414,"Blockwise attention bypasses packed-sequence isolation, causing cross-sample leakage","BlazeSR10hawk",4,"","",0,"2026-11-05",""],[0,50400,"Add support for compiled transformers","cselkiraly",4,"","",0,"2026-11-05",""],[0,50396,"MimiModel.encode: last frame differs from kyutai's encoder when the input length is not a multiple of 1920 sam","elpsykongloo",4,"","",0,"2026-11-05",""],[0,50395,"Continuous batching fails on MPS with torch 2.15: TypeError in create_device_stream","davanstrien",0,"Isalia20","#50404",1,"2026-11-05","2026-11-05"],[0,50390,"A failed attention kernel load is cached: later models fail at forward with `KeyError: ... is not a valid atte","albertvillanova",4,"","",0,"2026-11-05",""],[0,50389,"[Keypoint matching] Request to add FlexGlue","tcourat",4,"","",0,"2026-11-05",""],[0,50383,"Seq2SeqTrainer does not accept optimizer_cls_and_kwargs (supported by Trainer and listed in its docstring)","shupandee",4,"","",0,"2026-11-05",""],[0,50382,"Tool schemas silently dropped during response parsing across text, multimodal pipelines and serving APIs","BlazeSR10hawk",4,"","",0,"2026-11-05",""],[0,50381,"GLM-5.3-Flash: mHC `base`/`scale` load in bf16 although the checkpoint stores them in float32","albertvillanova",4,"","",0,"2026-11-05",""],[0,50378,"WhisperFeatureExtractor: a sample's valid frame count depends on how much the batch pads it","storyicon",4,"","",0,"2026-11-05",""],[0,50372,"`AuoModel` should map to base class","zucchini-nlp",4,"","",0,"2026-11-05",""],[0,50352,"Continuous batching: model support tracker","qgallouedec",4,"","",0,"2026-11-04",""],[0,50347,"GLM-5.3-Flash: `f_a_proj`/`f_b_proj` are quantized to FP8 although the checkpoint keeps them in bf16","albertvillanova",4,"","",0,"2026-11-04",""],[0,50344,"[RFC] Native adapt for Biren SUPA accelerator in transformers","frozenleaves",4,"","",0,"2026-11-04",""],[0,50336,"DeepSeek-V4: beam search leaves HCA/CSA compression caches on the wrong beams","taking-lying-flat",4,"","",0,"2026-11-04",""],[0,50334,"MiniMax-M3-VL: vision attention mixes images across samples in a batch","taking-lying-flat",4,"","",0,"2026-11-04",""],[0,50330,"Qwen VL models: reusing the cache gives new image tokens wrong positions","Arnav66692",4,"","",0,"2026-11-03",""],[0,50328,"AudioClassificationPipeline defaults to softmax for regression and multi-label models, corrupting prediction s","BlazeSR10hawk",4,"","",0,"2026-11-03",""],[0,50316,"[BUG] Gemma 4 assistant models fail to build since KV sharing was formalized","harshaljanjani",4,"","",0,"2026-11-03",""],[0,50315,"ZeroShotObjectDetectionPipeline silently drops batched images and misroutes candidate labels","BlazeSR10hawk",1,"BlazeSR10hawk","#50318",0,"2026-11-03","2026-11-05"],[0,50308,"Trainer no longer calls an overridden gradient_checkpointing_enable after the activation checkpointing rename","albertvillanova",0,"qgallouedec","#50310",1,"2026-11-03","2026-11-03"],[0,50306,"Trainer: resume from a checkpoint for models sharded at load time (`DistributedConfig`)","qgallouedec",4,"","",0,"2026-11-03",""],[0,50300,"Quantizing gpt-oss on load with `Mxfp4Config()` crashes its forward and saves noise","qgallouedec",4,"","",0,"2026-11-03",""],[0,50297,"Docs: `hidden_states[-1]` of DINOv3 / ViT changed from post-norm to pre-norm in 4.57.2 (not in release notes o","ravediamond",1,"ravediamond","#50303",1,"2026-11-03","2026-11-03"],[0,50273,"QuantizedCache.crop() raises IndexError across FP8, Quanto and HQQ, breaking speculative decoding","BlazeSR10hawk",4,"","",0,"2026-11-02",""],[0,50270,"Saving a gpt-oss loaded with `Mxfp4Config(dequantize=True)` loses the experts","qgallouedec",4,"","",0,"2026-11-02",""],[0,50255,"`Mxfp4Config(dequantize=True)` ignores dtype: experts are always dequantized to bfloat16","qgallouedec",1,"qgallouedec","#50256",1,"2026-11-02","2026-11-02"],[0,50241,"PaddleOCR-VL: position embedding interpolation changed from align_corners=False to True","scyyh11",1,"scyyh11","#50384",0,"2026-11-02","2026-11-05"],[0,50232,"Add a calibration test for `WatermarkDetector`","aparnaittekot",2,"","",0,"2026-11-01","2026-11-04"],[0,50230,"[Doge] `DogeCDMoE` router `.view(2, bsz * seq_len, -1)` mixes tokens: with `is_moe=True`, logits depend on fut","Andreas-Hum",4,"","",0,"2026-11-01",""],[0,50229,"[Doge] `load_balancing_loss_func` multiplies expert loads (by expert id) with router probabilities in sorted r","Andreas-Hum",1,"Andreas-Hum","#50266",0,"2026-11-01","2026-11-03"],[0,50228,"`TopHLogitsWarper` does not implement the documented Top-H rule (it bounds the unnormalized partial entropy, a","Andreas-Hum",4,"","",0,"2026-11-01",""],[0,50227,"[CLAP] `ClapModel(return_loss=True)` computes the text-to-audio loss twice and never the audio-to-text loss","Andreas-Hum",4,"","",0,"2026-11-01",""],[0,50224,"`_attn_implementation` setter does not reach encoder/decoder stacks built with `copy.deepcopy(config)`","the-cross-art",2,"","",0,"2026-11-01","2026-11-02"],[0,50223,"CI: vllm-ci failing: uv 'No solution found' torch==2.14.0+cpu unavailable for vllm cpu build; step uses `sh` w","fitzyracing1",4,"","",0,"2026-11-01",""],[0,50222,"CI: Torch pipeline CI failing: 39 pipeline tests fail with RuntimeError ignore_mismatched_sizes=False","fitzyracing1",4,"","",0,"2026-11-01",""],[0,50221,"CI: check_tiny_models failing: `from datasets import load_dataset` ImportError — pyarrow needs NumPy>=2.0 (fou","fitzyracing1",4,"","",0,"2026-11-01",""],[0,50220,"CI: build-docker-images Doc builder failing: pyarrow requires NumPy>=2.0 but image has 1.26.4","fitzyracing1",4,"","",0,"2026-11-01",""],[0,50216,"ModernBERT mean pooling overflows in float16 on long inputs and returns NaN logits","clemlesne",4,"","",0,"2026-11-01",""],[0,50212,"Trainer floor-divides valid-token counts under model parallelism, silently mis-scaling gradients and optimizer","BlazeSR10hawk",1,"BlazeSR10hawk","#50214",0,"2026-11-01","2026-11-02"],[0,50211,"SDPA launch failures on ROCm Windows surface as misleading async errors with no fallback — propose error trans","AIwork4me",4,"","",0,"2026-11-01",""],[0,50209,"Text-generation pipeline crashes with batched generation metadata and silently drops beam scores","BlazeSR10hawk",1,"BlazeSR10hawk","#50210",0,"2026-11-01","2026-11-03"],[0,50204,"[Bug]: resume_from_checkpoint silently skips the root-level (trained) adapter when the checkpoint contains ada","TheSeaByTheMorning",4,"","",0,"2026-10-31",""],[0,50200,"DeBERTa: `from_pretrained(dtype=torch.bfloat16)` leaves `q_bias`/`v_bias` in float32, forward fails (`expected","tarekziade",4,"","",0,"2026-10-31",""],[0,50199,"Gemma4: `model(**inputs)` raises ValueError for video inputs (`num_frames_per_video` dropped in forward)","yzkkai",1,"yzkkai","#50237",0,"2026-10-31","2026-11-02"],[0,50197,"GGUF reader: llama.cpp Mixtral loads as dense Llama with random MLPs; Qwen3 head_dim and Mistral detection not","JohnnyTeutonic",4,"","",0,"2026-10-31",""],[0,50196,"GGUF: Gemma 3 loses global-layer RoPE linear scaling and query_pre_attn_scalar when loaded from a llama.cpp GG","JohnnyTeutonic",4,"","",0,"2026-10-31",""],[0,50195,"Mamba (Mamba-1): forward of several tokens over a primed cache restarts the selective scan from zero and retur","JohnnyTeutonic",4,"","",0,"2026-10-31",""],[0,50194,"Multimodal pipelines silently discard generated candidates during postprocessing","BlazeSR10hawk",1,"BlazeSR10hawk","#50198",0,"2026-10-31","2026-11-04"],[0,50187,"NemotronH: Mamba2 mixer clamps dt at `time_step_min`, ignoring `config.time_step_limit` (long-context logits d","ljhnick",2,"","",0,"2026-10-30","2026-11-02"],[0,50185,"Gated delta rule layers crash on packed batches with more than one row","qgallouedec",4,"","",0,"2026-10-30",""],[0,50177,"5.21.0 regression: end-anchored `WeightRenaming` rules save corrupted keys (`mask_embedding$`) that silently f","aifeifei798",4,"","",0,"2026-10-30",""],[0,50174,"TIPSv2: bfloat16 forward fails, text pooling returns float32 (`expected m1 and m2 to have the same dtype`)","tarekziade",4,"","",0,"2026-10-30",""],[0,50171,"Qwen4Exp QSA indexer: data-dependent k_layernorm calls deadlock ZeRO-3 training","eilamc14",4,"","",0,"2026-10-30",""],[0,50154,"apply_chat_template sanitize_control_tokens silently fails when a template prints a dict or list","jaeminbird",4,"","",0,"2026-10-30",""],[0,50151,"[distributed] Save/load benchmark on 8x B200: gathered save, DCP save and consolidation bottlenecks","3outeille",4,"","",0,"2026-10-30",""],[0,50147,"RepetitionPenaltyLogitsProcessor penalizes masked padding tokens, causing batch-dependent EOS scores and gener","BlazeSR10hawk",1,"BlazeSR10hawk","#50149",0,"2026-10-30","2026-11-04"],[0,50141,"fsdp_config[\"ignored_modules\"] is silently ignored by TrainingArguments","eilamc14",4,"","",0,"2026-10-29",""],[0,50139,"glm5_next / qwen4_exp: hyper-connection mapping keeps full-width fp32 activations for backward","eilamc14",3,"","",0,"2026-10-29","2026-11-04"],[0,50137,"Qwen4ExpModel._no_split_modules lists not existing modules","eilamc14",1,"eilamc14","#50138",0,"2026-10-29","2026-10-30"],[0,50135,"RoPE inv_freq is initialized in bf16 under DeepSpeed ZeRO-3 again (regression of #28760 from #47598)","eilamc14",0,"Cyrilvallez","#50275",1,"2026-10-29","2026-11-03"],[0,50125,"EsmConfig rejects facebook/esm-1b: hidden_dropout_prob is an int (0)","tarekziade",2,"","",0,"2026-10-29","2026-11-04"],[0,50119,"PaliGemma 2 ignores final_logit_softcapping","alejandrorodriguezalvarez884-dot",1,"alejandrorodriguezalvarez884-dot","#50156",0,"2026-10-29","2026-11-02"],[0,50113,"Restrict the CircleCI token to the CircleCI host","ParthibanRajasekaran",2,"","",0,"2026-10-29","2026-10-29"],[0,50112,"Use ast.literal_eval for layer configs in fairseq conversion scripts","ParthibanRajasekaran",2,"","",0,"2026-10-29","2026-10-29"],[0,50111,"Gate the unpickling in the PaddleDetection conversion script","ParthibanRajasekaran",2,"","",0,"2026-10-29","2026-10-29"],[0,50110,"Use yaml.safe_load for NeMo config in model conversion scripts","ParthibanRajasekaran",2,"","",0,"2026-10-29","2026-10-29"],[0,50109,"WatermarkDetector p-values are wrong for large z-scores (exactly 0 above z ≈ 7.7)","jonasschaefer",1,"jonasschaefer","#50123",0,"2026-10-29","2026-10-30"],[0,50100,"Image preprocessing ~20% slower on Linux since v5.18.0: in-place normalize (#48897) causes extra page faults","yidezou",4,"","",0,"2026-10-28",""],[0,50097,"ClapFeatureExtractor forces is_longer=True in fusion mode when every clip is ≤10 s","outerheaven199X",4,"","",0,"2026-10-28",""],[0,50094,"apply_chat_template ignores return_tensors=None for processors that default to tensors (e.g. Pixtral)","qgallouedec",0,"zucchini-nlp","#50103",1,"2026-10-28","2026-11-03"],[0,50072,"[hardware support] Moore Threads GPU (MUSA) support in flash linear attention","WangKangAandy",4,"","",0,"2026-10-28",""],[0,50069,"DeBERTa-v2: @torch.jit.script helpers emit torch 2.14 deprecation warnings at import","tailoredcaresolutions",0,"Rocketknight1","#50082",1,"2026-10-28","2026-10-30"],[0,50063,"Expose opt-in draft acceptance statistics for assisted decoding","BlazeSR10hawk",4,"","",0,"2026-10-28",""],[0,50051,"Prompt-lookup decoding advances stateful logits processors (CFG, SynthID watermark) on draft tokens, so output","YaochengTong",4,"","",0,"2026-10-28",""],[0,50026,"Bug: NameError: name 'nn' is not defined in transformers/integrations/accelerate.py","MC-1026",0,"megacodes24","#50089",0,"2026-10-27","2026-10-28"],[0,49983,"Lossless state-tree weights and KV cache: ~24% smaller BF16 checkpoints, faster inference","ammarali32",2,"","",0,"2026-10-26","2026-10-27"],[0,49979,"`Trainer.evaluate` eval_loss under DDP includes the samples the sampler repeats to even out the last batch","Robertboy18",4,"","",0,"2026-10-26",""],[0,49978,"Trainer scales the loss of heads that take `**kwargs` but return a mean loss by `world_size * gradient_accumul","Robertboy18",4,"","",0,"2026-10-26",""],[0,49973,"ForQuestionAnswering heads fail to load base model weights due to base_model_prefix=\"transformer\" (Llama famil","DavidJohnQuinlan",2,"","",0,"2026-10-26","2026-10-29"],[0,49971,"DataCollatorForLanguageModeling fails with whole-word masking and dynamic padding","yuanx749",4,"","",0,"2026-10-25",""],[0,49970,"Improve tokenizer validation errors by including the received input type","lounishamroun",2,"","",0,"2026-10-25","2026-10-27"],[0,49967,"[Docs] Add Vietnamese PhoBERT fine-tuning example for low-resource NLP","Dennis-nham",2,"","",0,"2026-10-25","2026-10-27"],[0,49954,"[serve] Streaming error events are not valid JSON when the error message contains quotes","akshat-lakhera",4,"","",0,"2026-10-24",""],[0,49952,"[serve] max_completion_tokens is ignored on /v1/chat/completions","akshat-lakhera",0,"fruitstash","#50225",0,"2026-10-24","2026-11-02"],[0,49938,"SAM3: `Sam3ViTLayer.position_ids` buffer is cast to bf16 on loading, degrading RoPE in global-attention layers","DManetti",4,"","",0,"2026-10-23",""],[0,49937,"GraniteSpeech: `logits_scaling` not applied since v5.10 (logits are 8x too large)","pszemraj",4,"","",0,"2026-10-23",""],[0,49936,"Speculative sampling accepts zero-probability candidates when the acceptance draw is zero","SeverinVisionary",2,"","",0,"2026-10-23","2026-11-02"],[0,49927,"[serve] Add a Warm-Up Step after Loading the Model","mattduerrmeier",4,"","",0,"2026-10-23",""],[0,49926,"Qwen3.5: split the Gated DeltaNet layers by heads under tensor parallelism (follow-up to #46846)","niklas-palm",4,"","",0,"2026-10-23",""],[0,49925,"tp_plan=\"auto\" adds backward hooks that break torch.compile graphs in inference","niklas-palm",2,"","",0,"2026-10-23","2026-10-29"],[0,49918,"[Chat Parsing] Add python-call-list content parser and strip_prefix/strip_suffix text args","nico-martin",4,"","",0,"2026-10-23",""],[0,49907,"FPS sampling returns no frames for short, non-empty videos","BlazeSR10hawk",1,"BlazeSR10hawk","#49908",0,"2026-10-23","2026-10-26"],[0,49902,"GOT-OCR2 produces incorrect bounding-box coordinates for NumPy images","BlazeSR10hawk",4,"","",0,"2026-10-23",""],[0,49897,"default_data_collator incorrectly casts NumPy integer labels to float32","BlazeSR10hawk",1,"BlazeSR10hawk","#49900",0,"2026-10-23","2026-10-27"],[0,49888,"Qwen4-Exp: eager QSA attention produces NaN logits for left-padded samples","taking-lying-flat",1,"taking-lying-flat","#49923",0,"2026-10-23","2026-11-03"],[0,49884,"Add Times Series Forecasting Models","yaswanth19",4,"","",0,"2026-10-22",""],[0,49881,"Importing Trainer fails under marimo when IPython is not installed","fabbrito",4,"","",0,"2026-10-22",""],[0,49879,"A generation config loaded from the Hub can't be saved again","benthecarman",4,"","",0,"2026-10-22",""],[0,49876,"DiffusionGemma: generated text changes with batch size (MoE router projection and self-conditioning matmul)","UdulaAbeykoon",2,"","",0,"2026-10-22","2026-10-26"],[0,49873,"FineGrainedFP8 `dequantize`: weights with a partial 128-block fail to load (DeepSeek-V3/R1) or dequantize sile","mrtpk",4,"","",0,"2026-10-22",""],[0,49850,"[i18n-te] Translating docs to Telugu","hakunamatata1997",2,"","",0,"2026-10-22","2026-10-22"],[0,49828,"Dynamic, YaRN and LongRoPE fail when config.head_dim is None (Mixtral, Ministral, MiniMax, HunYuan)","nettanvirdev",2,"","",0,"2026-10-21","2026-10-22"],[0,49817,"RT-DETR / RT-DETRv2: main loss is computed on the denoising queries too, so real queries train as background","hagai-psi",4,"","",0,"2026-10-21",""],[0,49810,"Gemma 2's attention soft-capping is silently dropped with sdpa (the default) and flash kernels without softcap","qgallouedec",1,"qgallouedec","#49811",1,"2026-10-21","2026-10-21"],[0,49793,"[CB] `generate_batch` with flash attention gets slower as the KV cache gets bigger, even when the cache is mos","albertvillanova",2,"","",0,"2026-10-20","2026-10-27"],[0,49762,"ProcessorMixin.apply_chat_template treats template kwargs unused by the template as processor kwargs","albertvillanova",3,"","",0,"2026-10-20","2026-10-26"],[0,49761,"SmolVLMProcessor.apply_chat_template always warns \"Kwargs passed to processor.__call__ have to be in processor","albertvillanova",0,"zucchini-nlp","#49812",1,"2026-10-20","2026-10-29"],[0,49760,"ProcessorMixin.apply_chat_template drops explicit processor_kwargs when an extra kwarg is passed","albertvillanova",0,"zucchini-nlp","#49812",1,"2026-10-20","2026-10-29"],[0,49745,"Qwen4-Exp: cache.reset() leaves stale position_ids when reusing DynamicCache or StaticCache","taking-lying-flat",1,"taking-lying-flat","#49840",0,"2026-10-19","2026-10-22"],[0,49739,"SAM disk offload and model parallelism tests fail with `ValueError: max() arg is an empty sequence`","tarekziade",2,"","",0,"2026-10-19","2026-10-20"],[0,49734,"[serve] Sampling not enabled for positive temperature","AbhilakshSinghReen",4,"","",0,"2026-10-19",""],[0,49727,"[Bug] Qwen3-MoE router auxiliary loss is multiplied by the DDP world size in Trainer","gss10282025",4,"","",0,"2026-10-19",""],[0,49726,"Qwen3-VL patch embedding: Conv3d 58.5s vs equivalent matmul 0.2ms on H800","daojiancha",2,"","",0,"2026-10-19","2026-10-19"],[0,49724,"SmolVLM forward with images fails in v5.20.1: shape '[-1, 14, ...]' is invalid (inputs_merger uses vision patc","albertvillanova",1,"albertvillanova","#49725",1,"2026-10-19","2026-10-19"],[0,49719,"Aria, Inkling and PaddleOCR-VL report `input_modalities == \"text\"`, and declaring their modalities breaks thei","qgallouedec",0,"zucchini-nlp","#49720",1,"2026-10-19","2026-10-19"],[0,49702,"Expert-parallel token dispatch gives experts no gradient on torch 2.7 to 2.10","digit50",1,"digit50","#49710",0,"2026-10-17","2026-10-20"],[0,49701,"Torchvision backend ignores EXIF orientation for path, URL and base64 images","digit50",1,"digit50","#49794",0,"2026-10-17","2026-10-21"],[0,49700,"prefill_chunk_size drops image inputs, VLMs generate without the image","digit50",1,"digit50","#49753",0,"2026-10-17","2026-11-04"],[0,49699,"generate() fails with KeyError/AttributeError when input_ids are off the model device","digit50",0,"Cyrilvallez","#49731",1,"2026-10-17","2026-10-19"],[0,49698,"input_projections are not registered as PyTorch modules, causing missing parameters and CUDA device mismatch","bruno2009",0,"Omarinho12345","#49703",0,"2026-10-17","2026-10-21"],[0,49695,"Generation crashes with encoder_repetition_penalty when encoder_outputs are provided without input_ids","kunalsanga",1,"kunalsanga","#49696",0,"2026-10-17","2026-10-22"],[0,49693,"CI: Idefics3 test_onnx_export_static fails with torch.AcceleratorError CUDA device-side assert","tarekziade",2,"","",0,"2026-10-17","2026-10-17"],[0,49692,"CI: Idefics3 test_model_parallelism fails with torch.AcceleratorError CUDA device-side assert","tarekziade",2,"","",0,"2026-10-17","2026-10-17"],[0,49691,"CI: Idefics3 test_model_parallel_beam_search fails with CUDA device-side assert","tarekziade",2,"","",0,"2026-10-17","2026-10-17"],[0,49687,"Expert parallelism with ep_size < fsdp_size trains no experts in bfloat16 (expert gradients are zero)","qgallouedec",1,"qgallouedec","#49688",1,"2026-10-17","2026-10-17"],[0,49675,"TextStreamer re-decodes the whole line on every token (quadratic on long lines)","deadstrobe5",4,"","",0,"2026-10-16",""],[0,49674,"Chat parsing: `json` parser's `string_delims` doesn't handle escaped delimiters","qgallouedec",4,"","",0,"2026-10-16",""],[0,49672,"Quick survey: your experience as a PyTorch Ecosystem maintainer","michellearoth",3,"","",0,"2026-10-16","2026-10-28"],[0,49666,"AutoTokenizer.from_pretrained deep-copies the freshly loaded backend for TokenizersBackend tokenizers","deadstrobe5",1,"deadstrobe5","#49667",0,"2026-10-16","2026-10-19"],[0,49656,"[i18n-<languageCode>] Translating docs to <languageName>","rvonnevski-create",2,"","",0,"2026-10-16","2026-10-16"],[0,49651,"torch 2.7: every `from_pretrained` fails in 5.20.0 (`hf_storage` import behind a `torch>=2.7` guard)","hazelxue",0,"naveecode","#49652",0,"2026-10-16","2026-10-16"],[0,49647,"GGUF tokenizer ignores tokenizer.ggml.add_bos_token, so the default encoding drops BOS","francip",1,"francip","#49649",0,"2026-10-15","2026-11-02"],[0,49645,"BatchEncoding.convert_to_tensors copies the whole batch to check if it's empty (~2x slower collation)","jayzuccarelli",4,"","",0,"2026-10-15",""],[0,49630,"Gemma4-12B Training Failed","kezoy",4,"","",0,"2026-10-15",""],[0,49628,"[Regression] EP no longer leaves attention replicated (v5.19)","dacorvo",1,"dacorvo","#49633",1,"2026-10-15","2026-10-15"],[0,49626,"[i18n-ur] Translating docs to Urdu (اردو)","mwzkhalil",4,"","",0,"2026-10-15",""],[0,49625,"Janus / DeepSeek-VL image processors ignore `background_color`; deepseek-vl-7b-chat's low-res branch pads with","tritsystem",1,"tritsystem","#49767",0,"2026-10-15","2026-10-21"],[0,49623,"_patch_mistral_regex treats a missing transformers_version as \"Mistral\": false positive on local DeBERTa-v3, a","cdcailf",4,"","",0,"2026-10-15",""],[0,49622,"DebertaV2Tokenizer drops SentencePiece NFKC normalization in v5: ligatures, fullwidth and circled characters e","tritsystem",4,"","",0,"2026-10-15",""],[0,49592,"SAM 3: text-only prompts drop the geometry CLS token that Meta's implementation always adds","ju-ius",4,"","",0,"2026-10-14",""],[0,49591,"import transformers spends 1-3 s in an eager importlib.metadata.packages_distributions() when torch >= 2.14 is","MakazhanAlpamys",0,"qgallouedec","#50047",1,"2026-10-14","2026-10-30"],[0,49585,"CI failure: ClefIntegrationTest::test_clef_flash assertion error","tarekziade",2,"","",0,"2026-10-14","2026-10-14"],[0,49578,"Tokenizer loses its normalizer from tokenizer.json in 5.18","AdnanElAssadi",4,"","",0,"2026-10-13",""],[0,49564,"SDPA: enable_gqa=True silently falls back to the math kernel at batch 1 on torch < 2.14 without Flash Attentio","Jose-Ribeir",4,"","",0,"2026-10-13",""],[0,49561,"UnboundLocalError in SuperPointForKeypointDetection.forward under torch.compile","kaixuanliu",1,"kaixuanliu","#49409",0,"2026-10-13","2026-10-13"],[0,49556,"[Bug] WhisperTimeStampLogitsProcessor crashes with TypeError when eos_token_id is a single-element list","kunalsanga",2,"","",0,"2026-10-13","2026-10-15"],[0,49553,"`/v1/systemone`: 500s for `chat_template_kwargs` overrides and for models without a chat template","rycerzes",4,"","",0,"2026-10-13",""],[0,49546,"CPU SDPA could use native GQA with a mask on torch >= 2.9 (1.2x decode, no repeat_kv)","ToukoUrsin",4,"","",0,"2026-10-12",""],[0,49540,"Gemma 3/4 multimodal models fail with tp_plan=\"auto\" because of mixed Tensor/DTensor usage","gbharada-aws",4,"","",0,"2026-10-12",""],[0,49536,"Mamba2: non-fused gated RMSNorm ignores n_groups and disagrees with fused training","taking-lying-flat",4,"","",0,"2026-10-12",""],[0,49529,"MM Grounding DINO and LLMDet: decoder box and class heads are tied to layer 0 although the config says they ar","MustafaOrhon",0,"lu123bing","#49552",0,"2026-10-12","2026-10-21"],[0,49519,"Assigning rope_scaling drops existing RoPE base parameters","jinwukong",4,"","",0,"2026-10-12",""],[0,49516,"T5 slower since #47014: non-contiguous position_bias makes SDPA fall back to MATH","IvenHsu01",1,"IvenHsu01","#49517",0,"2026-10-12","2026-10-14"],[0,49513,"LayoutLM rejects SDPA despite using the unified attention interface","kunalsanga",1,"kunalsanga","#49514",0,"2026-10-12","2026-10-12"],[0,49510,"AutoVideoProcessor fails to load backend-specific remote auto_map entries","DimensionSTP",1,"DimensionSTP","#49511",0,"2026-10-12","2026-10-13"],[0,49499,"HunYuan-VL beam search gives returned sequences the wrong image for batched prompts","Arnav66692",1,"Arnav66692","#49557",0,"2026-10-11","2026-10-16"],[0,49498,"SynthID-Text: CPU and MPS produce different sampling tables with the same seed","sk8ordie84",4,"","",0,"2026-10-11",""],[0,49497,"ZeRO-3: from_pretrained reads and converts the full checkpoint on every rank","xvxuopop",0,"3outeille","#50284",1,"2026-10-11","2026-11-06"],[0,49493,"https://github.com/features/ai/github-app?mobile_download=true","ferdie5683",2,"","",0,"2026-10-10","2026-10-12"],[0,49491,"[i18n-<languageCode>] Translating docs to <languageName>","rvonnevski-create",2,"","",0,"2026-10-10","2026-10-12"],[0,49488,"Discussion: configurable decoder loops with shared parameters and independent execution caches","bebetterest",4,"","",0,"2026-10-10",""],[0,49487,"[i18n-<languageCode>] Translating docs to <languageName>","ghost",2,"","",0,"2026-10-10","2026-10-10"],[0,49486,"Decision Transformer: attn_implementation=\"sdpa\" raises ValueError although attention already goes through ALL","yavuzmurattas",1,"yavuzmurattas","#49550",0,"2026-10-10","2026-10-13"],[0,49483,"Chunked prefill mishandles prefix-cache continuation","LE0-Lin",0,"Cyrilvallez","#49662",1,"2026-10-10","2026-10-19"],[0,49472,"Torch fallbacks of the linear-attention ops drop `seq_idx` / `cu_seqlens`, so packed documents leak without th","qgallouedec",1,"qgallouedec","#49474",1,"2026-10-09","2026-10-15"],[0,49469,"AutoVideoProcessor.register treats positional exist_ok as a backend dictionary","DimensionSTP",4,"","",0,"2026-10-09",""],[0,49465,"Mamba2 / Zamba2: cached decoding does not respect the time-step limits used by sequence scan","taking-lying-flat",4,"","",0,"2026-10-09",""],[0,49456,"Whisper generate() raises 'You have modified the pretrained model configuration' since #48282 (suppress_tokens","tarekziade",1,"tarekziade","#49466",1,"2026-10-09","2026-10-09"],[0,49451,"AutoVideoProcessor eagerly imports unrelated video processors after backend registration update","DimensionSTP",1,"DimensionSTP","#49458",0,"2026-10-09","2026-10-09"],[0,49448,"Trainer checkpoint rotation can delete valid checkpoints when backup directories match the checkpoint prefix","BlazeSR10hawk",4,"","",0,"2026-10-09",""],[0,49443,"Trainer disables find_unused_parameters under any gradient checkpointing, not just reentrant","qgallouedec",3,"","",0,"2026-10-08","2026-10-16"],[0,49440,"Packed sequences leak into each other in gated-DeltaNet models when only position_ids are given","qgallouedec",2,"","",0,"2026-10-08","2026-10-12"],[0,49434,"LIES","hgeorgescu",2,"","",0,"2026-10-08","2026-10-09"],[0,49422,"Gemma 4 `response_template` fails on tool names containing `-` or `.`","Varun-51",4,"","",0,"2026-10-08",""],[0,49411,"Inkling-Small: width of the routed experts resolves to 3072 instead of 2048 (moe_intermediate_size is never re","symbai-tech",0,"Talhax55z","#49416",0,"2026-10-08","2026-10-08"],[0,49410,"get_uniform_frame_indices can return too many or out-of-range frame indices","archiep867-sys",1,"archiep867-sys","#49464",0,"2026-10-08","2026-10-12"],[0,49403,"[Whisper] batch_decode(..., decode_with_timestamps=True) merges rows or raises on the tensor returned by gener","tarekziade",0,"Kingofbooks","#49419",0,"2026-10-07","2026-10-08"],[0,49399,"Add support for LFM2/2.5 in GGUF loading path","francip",1,"francip","#49400",0,"2026-10-07","2026-10-14"],[0,49393,"QuantizedCache exceeds residual_length on multi-token cache updates","LE0-Lin",1,"LE0-Lin","#49418",0,"2026-10-07","2026-10-08"],[0,49388,"Parakeet TDT greedy decoding ignores `max_symbols_per_step`: unbounded token repeats on one encoder frame","tomschott",4,"","",0,"2026-10-07",""],[0,49385,"Wav2Vec2 and 6 other audio models leave masked_spec_embed uninitialized when the checkpoint lacks it","tarekziade",1,"tarekziade","#49386",1,"2026-10-07","2026-10-29"],[0,49380,"5.19.0: models cannot be imported on torch 2.5, or on torch 2.6 without an accelerator (get_device_type calls ","ChrisChen667788",0,"Rocketknight1","#49395",1,"2026-10-07","2026-10-09"],[0,49378,"Loading a tokenizer imports torch and scikit-learn when they are installed (about 2 s of a 3 s import)","AshishKumar4",0,"jayzuccarelli","#49663",0,"2026-10-06","2026-10-22"],[0,49372,"set_attn_implementation() does not reach T5 encoder/decoder: model reports SDPA but still runs eager","tarekziade",2,"","",0,"2026-10-06","2026-10-20"],[0,49371,"Pixtral Flash Attention 2 uses axial RoPE coordinates as packed sequence boundaries (regression in 5.17.0)","oliverholworthy",1,"oliverholworthy","#49373",0,"2026-10-06","2026-10-07"],[0,49338,"Explicit kernelization modes are ignored on evaluation models","DimensionSTP",1,"DimensionSTP","#49339",0,"2026-10-06","2026-10-06"],[0,49337,"MetalLinear returns wrong outputs for every batch row after the first (3D inputs)","francip",4,"","",0,"2026-10-06",""],[0,49334,"Phi3 LongRoPE generation loses the prefix when clearing KV at the short/long crossing","AshishKumar4",0,"LE0-Lin","#49361",0,"2026-10-06","2026-10-07"],[0,49324,"MODEL_FOR_CAUSAL_LM_MAPPING[Qwen3_5Config] points at the flat class, but all Qwen3.5/3.8 checkpoints are condi","micdoh",3,"","",0,"2026-10-05","2026-10-05"],[0,49320,"CohereCompass omits \"video\" from input_modalities, so generate() rejects video inputs","stefankoncarevic",0,"destopianpirate","#49323",0,"2026-10-05","2026-10-06"],[0,49318,"generate() chunked prefill slices the batch axis of multi-axis position_ids (Qwen3-VL)","dacorvo",1,"dacorvo","#49319",1,"2026-10-05","2026-10-05"],[0,49310,"[PERF] DeepSeek-V3.2 / GLM-5 lightning indexer overallocates memory","harshaljanjani",4,"","",0,"2026-10-05",""],[0,49299,"WatermarkDetector counts repeated token windows when deduplication is enabled","LE0-Lin",1,"LE0-Lin","#49315",0,"2026-10-04","2026-10-06"],[0,49297,"[FSDP2] cpu_ram_efficient_loading materializes full model on CPU for every non-rank-0, defeating rank-0 broadc","rjchao2021",2,"","",0,"2026-10-04","2026-11-04"],[0,49293,"WatermarkDetector scores depend on batch order when rows have mixed BOS prefixes","euisuh",2,"","",0,"2026-10-04","2026-10-12"],[0,49289,"MiniMax lightning attention: decay recurrence runs in bf16, so slow-decay heads never decay and decode drifts ","tritsystem",4,"","",0,"2026-10-04",""],[0,49288,"model.to(bf16)/half() and Trainer bf16_full_eval cast RoPE inv_freq buffers, unlike from_pretrained(dtype=...)","tritsystem",2,"","",0,"2026-10-04","2026-10-05"],[0,49281,"New model Aleph-Alpha/Kolibri-1","stelterlab",0,"lucamotz","#49463",0,"2026-10-03","2026-10-13"],[0,49274,"Multi-image generation reorders encoded image features during beam and sampled batch expansion","DimensionSTP",1,"DimensionSTP","#49276",0,"2026-10-02","2026-10-05"],[0,49272,"pipeline(..., model_kwargs={\"code_revision\": ...}) raises TypeError, and top-level code_revision does not reac","KiSung-Park",0,"Rocketknight1","#49832",1,"2026-10-02","2026-10-27"],[0,49268,"Add MLX KV-cache compression integration with VeloxQuant-MLX","rajveer43",2,"","",0,"2026-10-02","2026-10-05"],[0,49267,"Gemma 4 `response_template` parses reasoning into `thinking`, which its chat template does not read back","albertvillanova",0,"Rocketknight1","#49535",1,"2026-10-02","2026-10-20"],[0,49261,"`generate()` with a static cache: host-side inputs miss tensors nested in dicts, and TP undersizes the early c","dacorvo",1,"dacorvo","#49317",1,"2026-10-02","2026-10-14"],[0,49253,"ImageTextToTextPipeline warns about generation arguments it injects itself (generation_config + return_dict_in","jr2804",0,"Rocketknight1","#49269",1,"2026-10-02","2026-10-08"],[0,49252,"AutoTokenizer with tokenizer_class=LlamaTokenizerFast silently builds a sentencepiece pipeline for a byte-leve","kangzhengyang",4,"","",0,"2026-10-02",""],[0,49238,"ExponentialDecayLengthPenalty creates NaN EOS scores when combined with minimum-length constraints","gitedmond",1,"gitedmond","#49260",0,"2026-10-01","2026-10-05"],[0,49237,"Trainer counts fp16 steps skipped by GradScaler as completed update steps (`global_step`, `max_steps`, `on_opt","SeverinVisionary",4,"","",0,"2026-10-01",""],[0,49233,"Encoder repetition penalty uses incorrect source rows after beam or sampled-return expansion","gitedmond",1,"gitedmond","#49262",0,"2026-10-01","2026-10-05"],[0,49203,"Next Release?","davidmezzetti",2,"","",0,"2026-09-30","2026-09-30"],[0,49196,"CSM codebook embedding tying is ignored when `tie_word_embeddings=False`","orrangetabby17",1,"orrangetabby17","#49197",0,"2026-09-30","2026-10-26"],[0,49195,"SAM3 Video reconditioning causes mask to grow onto the edges and make the bounding boxes balloon","Ryan-Xue2",4,"","",0,"2026-09-30",""],[0,49193,"Trainer reports incorrect final loss and throughput after resuming from a full checkpoint","gitedmond",0,"SunMarc","#49208",1,"2026-09-29","2026-10-01"],[0,49190,"[eager MOE] Selected expert IDs should probably not be treated as 0-d GPU tensors","fxmarty-amd",4,"","",0,"2026-09-29",""],[0,49185,"[BLT, Byte-Latent Transformer] Entropy patcher is missing BLT's 512-token sliding-window attention — patch rat","hyunW3",1,"hyunW3","#49188",0,"2026-09-29","2026-11-04"],[0,49183,"[Model Proposal] DualBigLittle-MoE: Tri-Tier Hierarchical MoE with Dual VRAM Cores & Streaming Micro-Experts","aifeifei798",2,"","",0,"2026-09-29","2026-09-29"],[0,49176,"tiny-aya models are loaded as CohereTokenizer in v5, and digits are split differently from tokenizer.json","Yinhan-Lu",2,"","",0,"2026-09-29","2026-11-02"],[0,49172,"Qwen3_5MoeModel no longer returns router_logits, breaking aux loss with output_router_logits=True","albertvillanova",0,"zucchini-nlp","#49179",1,"2026-09-29","2026-09-30"],[0,49166,"Rename `gradient_checkpointing` to `activation_checkpointing` in the next major","qgallouedec",4,"","",0,"2026-09-28",""],[0,49144,"`get_json_schema` maps bare `list` / `tuple` annotations to `\"type\": \"object\"` while `typing.List` / `typing.T","825pranav",1,"825pranav","#49145",0,"2026-09-27","2026-09-30"],[0,49139,"SigLIP2's tokenizer performs no case-normalization, unlike SigLIP v1's SiglipTokenizer","basaba07",0,"jayzuccarelli","#49245",0,"2026-09-27","2026-10-12"],[0,49137,"set_attn_implementation fails to propagate attention backend to nested sub-configs","eustlb",1,"eustlb","#49138",1,"2026-09-27","2026-10-15"],[0,49134,"[Regression v5.15.0+] UMT5 decoder self-attention is not causal: `is_causal` never reaches `UMT5Attention`","the-cross-art",1,"the-cross-art","#49135",0,"2026-09-27","2026-09-28"],[0,49122,"get_json_schema drops `items` / `enum` when a Union contains `list[...]`, `dict[...]` or `Literal[...]`","JoeyTan21",1,"JoeyTan21","#49136",0,"2026-09-26","2026-09-28"],[0,49121,"Trainer fails to resume from a checkpoint on CPU with two or more processes","SeverinVisionary",0,"neevmodh","#49123",0,"2026-09-25","2026-09-29"],[0,49113,"Nemotron3Diarization streaming: the last chunk drops the final valid frame when len(audio) % 160 < 96","happyarts",0,"eustlb","#49167",1,"2026-09-25","2026-09-30"],[0,49108,"Return hidden states in VLMs","zucchini-nlp",4,"","",0,"2026-09-25",""],[0,49106,"ModernBERT: local-attention window is computed from sequence index, not `position_ids`, so non-monotonic posit","johnmofficial16-prog",4,"","",0,"2026-09-25",""],[0,49096,"_can_use_grouped_mm falls back on TPU w/ torch.compile for non-bf16 weights","salkan0",1,"salkan0","#49097",0,"2026-09-24","2026-09-25"],[0,49095,"MusicGen: forward(labels=...) raises \"set the decoder_start_token_id\" on every released checkpoint, while gene","KhalidIbrahem",2,"","",0,"2026-09-24","2026-11-03"],[0,49094,"MusicGen: hub configs set decoder dropout=0.1 but the checkpoints were trained with 0 — train() doubles the lo","KhalidIbrahem",2,"","",0,"2026-09-24","2026-11-03"],[0,49093,"`SequenceBiasLogitsProcessor`: two edge cases (token id 0 rejected in list format; prefix equal to full contex","lucaluo925",1,"lucaluo925","#49117",0,"2026-09-24","2026-09-30"],[0,49090,"Parakeet / Nemotron feature extraction needs ~4 GB of RAM per hour of audio (one-pass STFT)","happyarts",2,"","",0,"2026-09-24","2026-09-25"],[0,49066,"Qwen2Tokenizer overwrites pre_tokenizer from tokenizer.json. Wrong token IDs for all Qwen3.6 / Qwen3.8 models ","simonepstein",0,"itazap","#49116",1,"2026-09-23","2026-09-28"],[0,49051,"Gemma 1 checkpoints silently use exact GELU instead of gelu_pytorch_tanh since v4.48.0 (#35235 removed the hid","PCfVW",1,"PCfVW","#49084",0,"2026-09-23","2026-09-25"],[0,49036,"Cache.crop does not roll back linear-attention recurrent states — block/speculative verification on GDN hybrid","0xWhiteMage",4,"","",0,"2026-09-23",""],[0,49028,"HfArgumentParser crashes on a dataclass field typed as a bare list","caiotheodoro",2,"","",0,"2026-09-23","2026-10-22"],[0,49007,"[BUG] Passing only min_pixels or only max_pixels raises ValueError in image and video processors","harshaljanjani",0,"zucchini-nlp","#49021",1,"2026-09-22","2026-09-22"],[0,49003,"convert_to_rgb ignores PNG tRNS transparency","surgan12",0,"cs-fisha","#49005",0,"2026-09-22","2026-09-25"],[0,48972,"`save_pretrained(state_dict=...)` can duplicate tied weights after state-dict materialization","UnitedSnakes",2,"","",0,"2026-09-21","2026-09-21"],[0,48967,"Incorrect model predictions (because of incorrect tokenization output) after the 5.0 update","AMR-KELEG",2,"","",0,"2026-09-20","2026-09-22"],[0,48966,"Module-level getattr breaks unittest functionality","chschroeder",2,"","",0,"2026-09-20","2026-10-20"],[0,48964,"Issue doing PEFT on Embedding Gemma w/ new classifier head","sr5434",2,"","",0,"2026-09-20","2026-09-20"],[0,48963,"[i18n-<languageCode>] Translating docs to <languageName>","HALIMMOUSAA",3,"","",0,"2026-09-20","2026-09-21"],[0,48947,"DynamicCache(offloading=True) returns wrong logits after first cache reuse on hybrid Gated-DeltaNet + attentio","com-junkawasaki",0,"qgallouedec","#49363",1,"2026-09-19","2026-10-13"],[0,48942,"T5 sentencepiece tokenizer without precompiled_charsmap cannot be loaded","DRKV333",0,"itazap","#49211",1,"2026-09-18","2026-10-07"],[0,48925,"4D mask created under torch.compile when padding_mask is None","msnliu",0,"jiqing-feng","#48975",0,"2026-09-18","2026-09-23"],[0,48924,"OPT training with torch.compile decomposes SDPA into Triton kernels due to attention_mask","msnliu",0,"jiqing-feng","#49002",0,"2026-09-18","2026-09-24"],[0,48917,"Image processors silently mutate class defaults when instantiated with min_pixels","Gracy769",1,"Gracy769","#48916",0,"2026-09-18","2026-09-21"],[0,48913,"MtpModel/MtpLayer topology is too rigid to represent non-uniform MTP architectures","soyr-redhat",4,"","",0,"2026-09-17",""],[0,48911,"MoE models that balance with a router bias never expose their router logits","qgallouedec",1,"qgallouedec","#48920",1,"2026-09-17","2026-10-05"],[0,48892,"Non-persistent buffers materialized uninitialized by from_pretrained (_move_missing_keys_from_meta_to_device)","taoshaoyu",2,"","",0,"2026-09-17","2026-10-01"],[0,48889,"Llama 4 declares `output_router_logits` and `router_aux_loss_coef` but never reads them","qgallouedec",1,"qgallouedec","#48920",1,"2026-09-17","2026-10-05"],[0,48887,"gemma 4 can not load audio file and produce  Audio features and audio tokens do not match, tokens: 31, feature","ghost",2,"","",0,"2026-09-17","2026-09-21"],[0,48886,"Four MoE models expose `aux_loss` in their output type and never set it","qgallouedec",2,"","",0,"2026-09-17","2026-10-14"],[0,48884,"Please remove dependency on the unmantained module pyctcdecode: it wasn't updated for years and is still at nu","yurivict",4,"","",0,"2026-09-17",""],[0,48874,"Nothing destroys the process group, so every distributed Trainer run warns at exit","qgallouedec",1,"qgallouedec","#48875",1,"2026-09-16","2026-09-24"],[0,48862,"CLIPSeg, TimesFM, PP-OCR detector and VideoPrism base/head loads still give every weight randomly initialized ","blipbyte",1,"blipbyte","#48871",0,"2026-09-16","2026-10-21"],[0,48861,"testest","kevinoldching",3,"","",0,"2026-09-16","2026-09-16"],[0,48851,"Need Kimi K3 model architecture support","imezx",4,"","",0,"2026-09-16",""],[0,48842,"Context parallelism: Trainer.predict returns duplicated rows, unlabeled path skips the CP context","qgallouedec",4,"","",0,"2026-09-16",""],[0,48841,"FSDP2: evaluate/predict on a fresh Trainer raises in Accelerator.prepare","qgallouedec",2,"","",0,"2026-09-16","2026-10-29"],[0,48836,"Version-gated tokenizer file selection (fast_tokenizer_files) hands different vocabularies to different consum","AUTHENSOR",2,"","",0,"2026-09-15","2026-09-16"],[0,48835,"VibeVoice-ASR: float32 ceil in get_audio_features miscounts audio tokens for long audio, crashing generate() w","de-tre",0,"ege-arhan","#48864",0,"2026-09-15","2026-09-18"],[0,48826,"[Qwen3.8-flash-next] QSAIndexer can not train","cxiaolong",4,"","",0,"2026-09-15",""],[0,48818,"Gemma4 audio tower: attention mask inverted under eager attention (uniform attention regardless of content)","dltsum",2,"","",0,"2026-09-15","2026-10-25"],[0,48813,"CSOAI — Transformers interop: model governance","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[0,48812,"Proposal to add Ling-3.0 (Bailing MoE V3) support","ShahVandit",2,"","",0,"2026-09-15","2026-10-14"],[0,48810,"CSOAI — 22-axis AI governance measurement for model evaluation","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[0,48805,"Only on saving a checkpoint are weight conversions used","PashkovD",2,"","",0,"2026-09-14","2026-10-24"],[0,48791,"Reporting a vulnerability","S3DFX-CYBER",2,"","",0,"2026-09-14","2026-10-24"],[0,48784,"Support 2D bool mask for logits_to_keep, expose last_hidden_state","jp1924",1,"jp1924","a672eb5",0,"2026-09-14","2026-10-22"],[0,48757,"FP8 quantizer: update_tp_plan misses composite configs, expert scales stay replicated under EP","malaiwah",0,"ArthurZucker","#49548",1,"2026-09-13","2026-10-16"],[0,48756,"get_balanced_memory ignores _no_placement_params, inflating every device's budget (device_map=\"auto\" OOM)","malaiwah",2,"","",0,"2026-09-13","2026-10-29"],[0,48755,"Add DeepSeek-V4.1-Flash (deepseek_v41)","malaiwah",4,"","",0,"2026-09-13",""],[0,48752,"[i18n-<languageCode>] Translating docs to <languageName>","efijeniagordonia-create",2,"","",0,"2026-09-13","2026-09-14"],[0,48748,"`DogeForCausalLM` attends to future tokens with the default `sdpa` backend — `SmallDoge/Doge-20M` scores ppl 5","blipbyte",1,"blipbyte","#48821",0,"2026-09-12","2026-09-17"],[0,48745,"`RoFormerForCausalLM`, `BigBirdForCausalLM`, `RemBertForCausalLM`, `MegatronBertForCausalLM` attend to future ","blipbyte",0,"Cyrilvallez","#48785",1,"2026-09-12","2026-09-15"],[0,48738,"Image processor backends return different keys and dtypes for the same input","yupengtang",1,"yupengtang","#48739",0,"2026-09-12","2026-09-21"],[0,48736,"Follow-up: clean up video_duration handling in hyperclovax_vision_v2 (#44314)","jp1924",4,"","",0,"2026-09-12",""],[0,48732,"Text-to-audio pipeline failed in 5.17.0","zhangwei217245",3,"","",0,"2026-09-12","2026-09-14"],[0,48722,"Loading `RTDetrModel` from a detection checkpoint (or `SEWDForCTC` from a SEW-D checkpoint) gives a model with","blipbyte",0,"abhinav-phi","#48744",0,"2026-09-11","2026-09-14"],[0,48718,"Qwen3.5-9B fine-tuning with run_clm.py on 8x B200 ran the gated DeltaNet layers on the fp32 torch chunk loop b","TarzanZhao",2,"","",0,"2026-09-11","2026-10-29"],[0,48705,"[i18n-ta] Translating docs to Tamil","Mohammed-Saajid",2,"","",0,"2026-09-11","2026-10-04"],[0,48703,"Gemma 4 assistant uses first-token hidden state after prefill","glistening",1,"glistening","#48704",0,"2026-09-11","2026-09-14"],[0,48701,"Continuous batching: decode fast path is unsupported for sliding-window models","YeonwooSung",2,"","",0,"2026-09-11","2026-10-19"],[0,48698,"Optional ultrafast COCO backend for the object-detection examples (after TorchMetrics support)","developer0hye",2,"","",0,"2026-09-11","2026-10-19"],[0,48693,"[ImageGPT] Cross-attention crashes, ignores encoder_attention_mask, and doubles its cache","blipbyte",2,"","",0,"2026-09-10","2026-10-12"],[0,48690,"[Bug] Qwen3-MoE router auxiliary loss is multiplied by the DDP world size in Trainer","gss10282025",2,"","",0,"2026-09-10","2026-10-19"],[0,48687,"Wrong router gradient with `experts_implementation=\"batched_mm\"` under expert parallelism","qgallouedec",1,"qgallouedec","#48689",1,"2026-09-10","2026-09-11"],[0,48683,"Text to Audio Pipeline broken in v5.17","davidmezzetti",3,"","",0,"2026-09-10","2026-09-14"],[0,48670,"ASR feature extractors do not actually convert batched multi-channel list inputs to mono","LiRunGuo",2,"","",0,"2026-09-10","2026-10-19"],[0,48661,"[Docs] MRoPE (Multi-dimensional Rotary Position Embeddings) is undocumented — missing from rope_utils.md and R","DijieDeng",2,"","",0,"2026-09-09","2026-10-05"],[0,48659,"[BUG] InternVL accepts patch-grid/downsample combinations that fail in pixel_shuffle","fwerkor",2,"","",0,"2026-09-09","2026-09-10"],[0,48658,"[zai-org/GLM-OCR] <|image|> token has special=False in Rust tokenizer backend — should it be True?","ydshieh",2,"","",0,"2026-09-09","2026-10-19"],[0,48648,"[BUG] DeepSeek-V3 accepts odd qk_rope_head_dim but interleaved RoPE fails at runtime","fwerkor",2,"","",0,"2026-09-09","2026-09-10"],[0,48647,"[BUG] DeepSeek-V3 grouped expert routing accepts invalid group topology and can silently misroute tokens","fwerkor",2,"","",0,"2026-09-09","2026-09-10"],[0,48644,"[BUG] MoE grouped_mm dispatch accepts unsupported tensor strides","fwerkor",2,"","",0,"2026-09-09","2026-09-10"],[0,48643,"[BUG] LlamaConfig accepts incompatible GQA head counts","fwerkor",2,"","",0,"2026-09-09","2026-09-10"],[0,48636,"[Feature Request] Approximate Speculative Decoding (ASD): training-free, budget-bounded acceptance for greedy ","Kissmetothemoon",3,"","",0,"2026-09-09","2026-10-06"],[0,48633,"`use_gqa_in_sdpa` doesn't check GPU architecture: up to 28% slower decode on pre-sm80 GPUs","banarto",2,"","",0,"2026-09-09","2026-09-10"],[0,48630,"GPTNeoXJapanese crashes for any `rotary_pct != 1.0`: RoPE ignores `partial_rotary_factor`","blipbyte",1,"blipbyte","#48652",0,"2026-09-08","2026-09-16"],[0,48626,"Backslashes in generate flag values are interpreted as JSON escapes in transformers chat","JHC56",1,"JHC56","#48709",0,"2026-09-08","2026-09-29"],[0,48622,"Kernels for partial RoPE","zucchini-nlp",4,"","",0,"2026-09-08",""],[0,48621,"`compute_transition_scores()` applies beam selection twice for Whisper → out-of-bounds gather","kaokai-ui",2,"","",0,"2026-09-08","2026-10-18"],[0,48609,"优化 npu_swiglu_forward：拼接投影权重而不是拼接激活以降低内存并提升 NPU 性能","yaogang2060",2,"","",0,"2026-09-08","2026-09-08"],[0,48608,"[Feature request] FP8 KV cache + flash-attention: is this in scope?","jiqing-feng",4,"","",0,"2026-09-08",""],[0,48607,"AutoImageProcessor and the *ImageProcessorPil classes require torchvision from a docstring mention","blipbyte",1,"blipbyte","#48616",0,"2026-09-07","2026-09-10"],[0,48592,"`Trainer` deletes `config.bos_token_id` when the tokenizer declares no BOS, and persists it","albertvillanova",1,"albertvillanova","#48708",1,"2026-09-07","2026-09-22"],[0,48588,"IndexError in transformers chat when a generate flag is missing '=', and values containing '=' are silently tr","JHC56",0,"SunMarc","#48597",1,"2026-09-07","2026-09-08"],[0,48567,"BarthezTokenizer cannot load moussaKam/barthez: v4.57 → v5 regression, BPE vocab into a hardcoded Unigram","blipbyte",2,"","",0,"2026-09-06","2026-09-07"],[0,48559,"is_torchvision_available() uses find_spec, so a broken-but-installed torchvision is reported as available","caiotheodoro",2,"","",0,"2026-09-06","2026-09-25"],[0,48558,"AutoProcessor eagerly loads the video processor, so image-only VL use requires torchvision","caiotheodoro",2,"","",0,"2026-09-06","2026-09-06"],[0,48556,"AutoProcessor.from_pretrained raises AttributeError: 'NoneType' object has no attribute '__module__' when torc","caiotheodoro",1,"caiotheodoro","#48557",0,"2026-09-06","2026-09-07"],[0,48548,"[RFC] Zero-Copy In-Place MoE Dynamic Token Routing for Mixtral/DeepSeek (1.77x Speedup, 0 Bytes Aux VRAM)","aemre-cetin",4,"","",0,"2026-09-05",""],[0,48544,"LukeTokenizer drops <s>/</s> on the plain-text path (no post-processor installed)","blipbyte",2,"","",0,"2026-09-05","2026-10-12"],[0,48541,"Feature request: add NEAT as an optional Trainer optimizer integration","ItCodinTime",4,"","",0,"2026-09-05",""],[0,48527,"test_get_cached_module_file_local_handles_symlinked_hub_cache fails on Windows without symlink privilege","iam-kira",4,"","",0,"2026-09-04",""],[0,48512,"Decide whether the checkpoint shard prefetch should be on by default","qgallouedec",4,"","",0,"2026-09-03",""],[0,48511,"AutoImageProcessor: qwen3_vl / qwen3_5 mapping points at Qwen3VLImageProcessor(Pil) which no longer exists — '","suntp",2,"","",0,"2026-09-03","2026-09-17"],[0,48506,"Fix overridden method signature: logits_to_keep -> output_router_logits in AfmoeForCausalLM at modular_afmoe.p","deborshi-web",2,"","",0,"2026-09-03","2026-09-03"],[0,48501,"Exception raised for Gemma4 with static cache and chunked prefill","dacorvo",1,"dacorvo","#48619",1,"2026-09-03","2026-09-22"],[0,48491,"Qwen3-VL and Gemma3 (VLM) ignore `shift_labels`","qgallouedec",1,"qgallouedec","#48493",1,"2026-09-02","2026-09-08"],[0,48479,"MIGRATION_GUIDE_V5.md missing Dynamic - Cache API removals (key_cache, from_legacy_cache)","arkapravopal04",2,"","",0,"2026-09-02","2026-10-05"],[0,48459,"GPT-NeoX: hardcoded float32 softmax silently caps float64 inference at float32 resolution","clashstar",3,"","",0,"2026-09-01","2026-09-03"],[0,48449,"Warning: `get_extended_attention_mask` is deprecated","colingoldberg",2,"","",0,"2026-09-01","2026-09-02"],[0,48441,"[NEX Agent Co.] Add NEX A2A + x402 as transformer inference example","NEXAITECHAU",2,"","",0,"2026-08-31","2026-10-09"],[0,48428,"RoPE corruption due to TF32 on B200 (possibly others) when using `torch.set_float32_matmul_precision(\"high\")`","Conzel",0,"Rocketknight1","#49265",1,"2026-08-31","2026-10-05"],[0,48424,"Tests depend on `picsum.photos`, which is currently unreachable","jiqing-feng",2,"","",0,"2026-08-31","2026-09-07"],[0,48419,"[Bug] `HfArgumentParser` silently corrupts `tuple` and `set` typed dataclass fields from CLI","Gracy769",2,"","",0,"2026-08-30","2026-10-08"],[0,48400,"Bug about transformers==5.16.1 while loading 'nvidia/audio-flamingo-next-hf' model.","Jinbo-Hu",2,"","",0,"2026-08-28","2026-08-28"],[0,48392,"Nested per-layer `rope_parameters` is misparsed when `layer_types` omits one of its keys","albertvillanova",0,"zucchini-nlp","#48435",1,"2026-08-28","2026-09-04"],[0,48390,"Vectorized (batched) logits processors for assisted decoding","VaggelisGian",4,"","",0,"2026-08-28",""],[0,48375,"More freedom to override kernels for functions","zucchini-nlp",0,"vasqu","#48443",1,"2026-08-27","2026-09-02"],[0,48371,"NVFP4 MoE experts are dequantized without weight_global_scale, silently","Bluear7878",4,"","",0,"2026-08-27",""],[0,48369,"适配鸿蒙","Hao-bigMan",4,"","",0,"2026-08-27",""],[0,48351,"FineGrainedFP8HfQuantizer.update_tp_plan crashes when the model has no tp_plan","01554",2,"","",0,"2026-08-26","2026-08-30"],[0,48350,"qwen4_exp: fp8-quantized n-gram (PLE) embedding rows are gathered without dequantization","01554",2,"","",0,"2026-08-26","2026-09-28"],[0,48349,"FineGrainedFP8: modules_to_not_convert never matches on text-only loads of multimodal checkpoints (silent qual","01554",2,"","",0,"2026-08-26","2026-08-30"],[0,48346,"[Regression] FineGrainedFP8HfQuantizer crashes when expert TP overrides are absent","changwangss",1,"changwangss","#48343",0,"2026-08-26","2026-08-26"],[0,48328,"QA: mtp models are gone from hub","tarekziade",2,"","",0,"2026-08-26","2026-10-04"],[0,48323,"Chunk conversion op silently produces unequal chunks when the fused tensor is not divisible by the target coun","VaggelisGian",2,"","",0,"2026-08-25","2026-09-01"],[0,48320,"StopStringCriteria raises a bare 'substring not found' when the tokenizer vocab cannot round-trip its validati","VaggelisGian",2,"","",0,"2026-08-25","2026-08-28"],[0,48315,"Trainer crashes at the end of a resumed run when trainer_state.json points best_model_checkpoint at a missing ","VaggelisGian",1,"VaggelisGian","#48319",0,"2026-08-25","2026-08-26"],[0,48313,"Continuous batching is loud by default: WARNING logs and progress bars on a plain generate_batch","qgallouedec",1,"qgallouedec","#48314",1,"2026-08-25","2026-08-27"],[0,48312,"Continuous batching CUDA-graph capture crashes under distributed runs (allocator internal assert)","qgallouedec",2,"","",0,"2026-08-25","2026-09-29"],[0,48310,"inconsistency in naming at models/afmoe /modular_afmoe.py","deborshi-web",0,"shenhuaqingshi","#48430",0,"2026-08-25","2026-09-03"],[0,48308,"Native TP does not shard qwen3_vl_moe experts (weights fully replicated per rank)","qgallouedec",4,"","",0,"2026-08-25",""],[0,48307,"Native TP has no KV-head replication: tp_size > num_key_value_heads fails mid-forward","qgallouedec",4,"","",0,"2026-08-25",""],[0,48306,"Native TP breaks on Qwen3.5/3.6 hybrids: reshape error in linear-attention layers","qgallouedec",4,"","",0,"2026-08-25",""],[0,48305,"Continuous batching does not support hybrid linear-attention models (all Qwen3.5/3.6/3.8)","qgallouedec",4,"","",0,"2026-08-25",""],[0,48303,"Async continuous batching crashes at start() if no request is queued yet","qgallouedec",0,"remi-or","#48334",1,"2026-08-25","2026-08-27"],[0,48298,"Continuous batching crashes on every VLM: config attributes read from the top-level config","qgallouedec",1,"qgallouedec","#48299",1,"2026-08-25","2026-08-26"],[0,48296,"`paged|sdpa` silently computes non-causal attention on a standard forward","qgallouedec",1,"qgallouedec","#48297",1,"2026-08-25","2026-08-28"],[0,48293,"Wrong dim for torch.cumsum(token_priority) & Identical router_probs / router_logits Output","wangchaoxixixi",0,"yurekami","#48421",0,"2026-08-25","2026-09-21"],[0,48286,"Parakeet TDT generate shares request state across concurrent calls","mdumitru",2,"","",0,"2026-08-25","2026-10-04"],[0,48285,"`from_pretrained` with safetensors mmap crashes on Windows for large multi-shard checkpoints (copy-on-write co","eryk-roch",1,"eryk-roch","#48341",0,"2026-08-25","2026-08-27"],[0,48283,"CLI: transformers env fails with ModuleNotFoundError: No module named 'requests' when run in clean environment","Mobocop",0,"panshularora","#48412",0,"2026-08-25","2026-10-21"],[0,48256,"Right-padded prefill produces a corrupted cache in Mamba-family models (zero conv state, decayed SSM state)","benthecarman",0,"jayzuccarelli","#49247",0,"2026-08-24","2026-10-06"],[0,48253,"`BayesianDetectorModel.from_pretrained()` fails: missing `post_init()` leaves `all_tied_weights_keys` unset","woojinpaik",1,"woojinpaik","#48254",0,"2026-08-24","2026-08-25"],[0,48243,"Make the keypoint models exportable (SuperPoint, SuperGlue, LightGlue)","Om-singhaI",4,"","",0,"2026-08-24",""],[0,48241,"MiniMaxM2 silently applies full-head RoPE: config drops the checkpoints' legacy `rotary_dim` field","yuhezhang-ai",0,"dajiaohuang","#48486",0,"2026-08-23","2026-09-17"],[0,48239,"Loading very large models across nodes is 10-30x slower than the filesystem allows (and can kill the job)\"","qgallouedec",2,"","",0,"2026-08-23","2026-10-07"],[0,48232,"Import module ","deborshi-web",3,"","",0,"2026-08-23","2026-08-23"],[0,48225,"[BUG] MLukeTokenizer emits `<s>` instead of `<ent>` on all entity tasks","blipbyte",2,"","",0,"2026-08-22","2026-10-20"],[0,48210,"Should Trainer support models FSDP2-sharded at load time (DistributedConfig(fsdp_size=N))?","qgallouedec",2,"","",0,"2026-08-22","2026-09-29"],[0,48206,"AutoTokenizer silently corrupts deepseek-math-7b-rl in 5.x: encode strips all spaces, decode emits literal byt","johnmarktaylor91",2,"","",0,"2026-08-22","2026-09-29"],[0,48193,"Add native support for TR-HASH deterministic token-routed MoE","Complexity-ML",4,"","",0,"2026-08-21",""],[0,48187,"Add multimodal sequence classification support for Mistral 3","oliverholworthy",2,"","",0,"2026-08-21","2026-08-26"],[0,48177,"InternVL processor passes a torch tensor to `np.cumsum`, emitting a NumPy `__array_wrap__` DeprecationWarning","albertvillanova",0,"lorenzozanee","#48469",0,"2026-08-21","2026-09-14"],[0,48176,"DisentangledFlash: fused Triton attention for DeBERTa-v2/v3","delyan-boychev",3,"","",0,"2026-08-21","2026-10-02"],[0,48172,"[Mllama] MllamaVisionAttention and MllamaTextCrossAttention never set self.is_causal — cross-attention is sile","acebot712",3,"","",0,"2026-08-21","2026-08-23"],[0,48169,"[BUG] GlmConfig accepts incompatible GQA head counts and fails during forward","fwerkor",2,"","",0,"2026-08-21","2026-09-10"],[0,48154,"T5: encoder embedding silently loaded from lm_head.weight for checkpoints without shared.weight (google/madlad","connerhughes-dg",0,"vaibhavmashal","#48238",0,"2026-08-20","2026-09-18"],[0,48148,"Installed nested optional kernels silently fall back to the Torch implementation","DimensionSTP",1,"DimensionSTP","#48221",0,"2026-08-20","2026-09-03"],[0,48147,"[i18n-<languageCode>] Translating docs to <languageName>","jonbaton14-beep",2,"","",0,"2026-08-20","2026-08-21"],[0,48145,"Flaky: test_auto_docstring_import_time_upper_bound asserts a timing ratio whose noise exceeds its margin","tarekziade",2,"","",0,"2026-08-20","2026-11-06"],[0,48142,"save_pretrained drops subclass vocabulary sidecars: WhisperTokenizer.normalize breaks after a reload","Bluear7878",2,"","",0,"2026-08-20","2026-09-28"],[0,48138,"Flaky MellumModelTest.test_load_balancing_loss — the only copy without @is_flaky","subin9",2,"","",0,"2026-08-20","2026-09-27"],[0,48128,"v5 tokenizers of segment-embedding models (ALBERT, XLNet, Funnel, Splinter, RemBERT, RoFormer, HerBERT) no lon","2sumtech",2,"","",0,"2026-08-19","2026-08-21"],[0,48127,"XPU: `from_pretrained(device_map=...)` fails under WSL2 because `caching_allocator_warmup` does not guard `mem","studioego",1,"studioego","#48136",0,"2026-08-19","2026-09-17"],[0,48101,"[BUG] Odd head_dim is accepted but RoPE crashes during forward","fwerkor",0,"somuai","#48524",0,"2026-08-19","2026-09-10"],[0,48080,"Zamba-7B-v1 runs slow during torch.compile due to use_associative_scan=False","msnliu",2,"","",0,"2026-08-19","2026-09-26"],[0,48073,"load_video with num_frames > total_num_frames silently returns the same frame repeated","carlszk",1,"carlszk","#48074",0,"2026-08-19","2026-08-19"],[0,48069,"#$667$","lesande13rootGAM",2,"","",0,"2026-08-18","2026-08-21"],[0,48068,"nanochat's checkpoint converter is missing the TRUST_REMOTE_CODE guard every other pickle.load site in the rep","vishalkelur28-cyber",2,"","",0,"2026-08-18","2026-08-21"],[0,48062,"Qwen3-VL video processor spends the full pixel budget on short clips, making a 90s clip cost nearly as many to","dkrisman",1,"dkrisman","#48071",0,"2026-08-18","2026-08-26"],[0,48057,"Qwen2-VL get_rope_index shape mismatch even with batch_size=1 (transformers 5.15.0)","changwangss",2,"","",0,"2026-08-18","2026-08-18"],[0,48051,"convert_to_rgb gives wrong output shape for RGBA numpy videos","carlszk",2,"","",0,"2026-08-18","2026-08-19"],[0,48039,"Assisted/speculative decoding ignores `stop_strings` completed mid-block, and raises with `assistant_model`","pjordanandrsn",2,"","",0,"2026-08-17","2026-08-18"],[0,48030,"HfArgumentParser cannot parse dict-typed dataclass fields from the CLI (registers them with type=dict)","sergiopaniego",2,"","",0,"2026-08-17","2026-10-21"],[0,48029,"Segfault on macOS (MPS) when loading a bf16 checkpoint with dtype=torch.float32 and device_map=\"auto\" (async l","sergiopaniego",2,"","",0,"2026-08-17","2026-10-01"],[0,48017,"NemotronHConfig: num_hidden_layers deprecated but still needed by GGUF conversion and downstream tools","davidnichols-ops",2,"","",0,"2026-08-17","2026-08-18"],[0,48016,"GenerationConfig should validate that pad_token_id is not in eos_token_id list","davidnichols-ops",2,"","",0,"2026-08-17","2026-09-19"],[0,48003,"MTP: `test_generate_with_mtp` skips for every checkpoint but Inkling, and `MtpModel.forward` raises with the d","pjordanandrsn",0,"Noxtimo","#48262",0,"2026-08-16","2026-09-01"],[0,47999,"`_can_use_grouped_mm` admits Blackwell on torch <= 2.8, where `torch._grouped_mm` is Hopper-only—MoE forward c","pjordanandrsn",2,"","",0,"2026-08-16","2026-08-17"],[0,47994,"Zamba2: `num_mem_blocks > 1` checkpoints cannot be constructed — `block_id` uses the global layer index while ","john-rocky",1,"john-rocky","#48325",0,"2026-08-16","2026-08-27"],[0,47981,"CI single GPU regressions: 08/13 vs 08/14 diff (42 findings)","ydshieh",2,"","",0,"2026-08-14","2026-09-23"],[0,47979,"[Bug] torch.export canonicalizes DistilBERT input embeddings to the tied output projector","thealgebraist",2,"","",0,"2026-08-14","2026-09-23"],[0,47973,"Secret Leaks workflow scans nothing on branch and merge queue pushes","albertvillanova",2,"","",0,"2026-08-14","2026-10-14"],[0,47957,"[bitnet] Weight-quant-only BitLinear checkpoints load with sub-norms forced on","peterlodri-sec",2,"","",0,"2026-08-14","2026-09-21"],[0,47956,"compressed-tensors MoE decompression: weight_shape sharded away under TP/EP, fallback formula unrecoverable fo","verma8076",2,"","",0,"2026-08-14","2026-10-17"],[0,47942,"nemotron_h (Nemotron 3.5 Lightning): silently wrong outputs under multi-GPU device_map dispatch — cache corrup","masmoo",3,"","",0,"2026-08-13","2026-08-14"],[0,47932,"Speculative decoding: candidate generators return a `q` they did not sample from, breaking losslessness with `","pjordanandrsn",2,"","",0,"2026-08-12","2026-08-17"],[0,47926,"Continuous batching auto-switches to flash_attention_3 on unsupported GPUs (Turing/T4), crashing — regression ","albertvillanova",2,"","",0,"2026-08-12","2026-09-09"],[0,47925,"Module not found - MetaCLIP2Model","Yash-g17",2,"","",0,"2026-08-12","2026-08-12"],[0,47920,"_init_weights crashes on quantized checkpoints for models that initialize child modules by name","Bluear7878",1,"Bluear7878","#47921",0,"2026-08-12","2026-08-14"],[0,47917,"TimmWrapperPreTrainedModel: SDPA dispatch fails even though timm uses SDPA internally","msnliu",0,"jiqing-feng","#47939",0,"2026-08-12","2026-08-14"],[0,47914,"caching_allocator_warmup crashes with AttributeError when loading bitsandbytes-quantized model with PEFT adapt","LiaoBill",2,"","",0,"2026-08-12","2026-09-21"],[0,47912,"Assisted/speculative decoding generates past an EOS that is accepted mid-block","pjordanandrsn",1,"pjordanandrsn","#47931",0,"2026-08-12","2026-08-18"],[0,47910,"[i18n-<languageCode>] Translating docs to <languageName>","ifionahdev",3,"","",0,"2026-08-11","2026-08-11"],[0,47893,"ModuleNotFoundError: Could not import module 'MistralForSequenceClassification'","cqray1990",2,"","",0,"2026-08-11","2026-09-20"],[0,47892,"[hardware support] Moore threads GPU support in flash attention 2","XiaomingFun233",2,"","",0,"2026-08-11","2026-10-15"],[0,47886,"ASR pipeline silently destroys stereo audio in channels-last layout (what soundfile/librosa return): mean(axis","Kayvan-Zahiri",1,"Kayvan-Zahiri","#47888",0,"2026-08-10","2026-08-19"],[0,47885,"WhisperFeatureExtractor: one non-finite input sample makes the entire feature matrix NaN, and the pipeline tra","Kayvan-Zahiri",2,"","",0,"2026-08-10","2026-09-18"],[0,47879,"Error during video handling with Gemma4 (E2B-Instruct)","anand-kumar-consulting",0,"guarin","#47896",1,"2026-08-10","2026-08-14"],[0,47875,"Add Kimi Linear (Kimi Delta Attention) native support","sharmax-vikas",2,"","",0,"2026-08-10","2026-09-17"],[0,47869,"[Enhancement] Add early device validation in pipeline() to prevent downloading/loading models before crashing ","AryanSharma21",3,"","",0,"2026-08-10","2026-10-26"],[0,47868,"Regarding compatibility with FP8 quantization of Xiaomi-MiMo-V2.5 multimodal models","z0o0ey",3,"","",0,"2026-08-10","2026-09-04"],[0,47838,"`additional_special_tokens` is silently discarded when `tokenizer_config.json` also contains `extra_special_to","wagnerpatriota",0,"erichanwang","#47848",0,"2026-08-07","2026-09-30"],[0,47831,"`Qwen3_5GatedDeltaNet` / `Qwen3_5MoeGatedDeltaNet` initialize `A_log` to `-inf` on some heads (bf16 models)","Nkluge-correa",1,"Nkluge-correa","#47944",0,"2026-08-07","2026-08-14"],[0,47822,"Special tokens aren't escaped in template generation","pavelai",2,"","",0,"2026-08-06","2026-09-14"],[0,47805,"Whisper large-v3 and large-v3-turbo: config declares float16 but WhisperFeatureExtractor emits float32, so the","Kayvan-Zahiri",1,"Kayvan-Zahiri","#49837",0,"2026-08-05","2026-10-22"],[0,47801,"CVE-2025-14920 affected versions","jesvinjames",2,"","",0,"2026-08-05","2026-10-11"],[0,47787,"Add RWKV-7 (Goose)","Hakureirm",4,"","",0,"2026-08-05",""],[0,47766,"NameError: name 'nn' is not defined when resolving OutputRecorder type hints","Rudransh0102",2,"","",0,"2026-08-04","2026-08-04"],[0,47752,"incorrect precedence of generation_config values","jkl-ds",0,"nithin42","#47953",0,"2026-08-03","2026-08-18"],[0,47743,"MaskFormerSwin cannot run in float16/bfloat16: attention mask is hardcoded to float32","Yigtwxx",2,"","",0,"2026-08-03","2026-08-04"],[0,47741,"Several models modify caller-provided tensors in place during forward","Yigtwxx",2,"","",0,"2026-08-03","2026-08-04"],[0,47739,"HuBERT positional convolution is sensitive to padding when conv_pos_batch_norm=True","Abdulrhman-shaheen",2,"","",0,"2026-08-03","2026-10-19"],[0,47732,"Add native support for OpenBMB MiniCPM4","sushmanthreddy",4,"","",0,"2026-08-03",""],[0,47721,"Continuous batching (generate_batch) fails on models with heterogeneous per-layer num_key_value_heads/head_dim","jp1924",2,"","",0,"2026-08-03","2026-08-03"],[0,47713,"[Model Request] Native support for THUDM CogVLM2-Video","AryanSharma21",3,"","",0,"2026-08-02","2026-09-16"],[0,47710,"transformers.exporters raises ImportError on torch < 2.8 while setup.py declares torch>=2.5","Neal006",1,"Neal006","#47711",0,"2026-08-02","2026-08-03"],[0,47706,"RobertaTokenizerFast.from_pretrained() loads broken tokenizer in v5 (missing pre_tokenizer and BPE merges)","alwaysprince05",2,"","",0,"2026-08-02","2026-08-10"],[0,47705,"Gemma-4-E4B: the PLE table holds ~5.6 GB of VRAM that can live on CPU bit-identically - is there a supported d","MyceliaLabsBV",2,"","",0,"2026-08-02","2026-09-01"],[0,47695,"Add native support for OpenBMB VoxCPM2","sushmanthreddy",4,"","",0,"2026-07-31",""],[0,47692,"Add native support for Amazon Chronos-2","sushmanthreddy",4,"","",0,"2026-07-31",""],[0,47689,"kosmos2 and kosmos2_5 raise a dtype mismatch under torch.autocast when merging image features","Yigtwxx",1,"Yigtwxx","#47691",0,"2026-07-31","2026-08-03"],[0,47688,"Heads that declare **kwargs but drop num_items_in_batch train with an inflated loss under gradient accumulatio","Yigtwxx",2,"","",0,"2026-07-31","2026-09-23"],[0,47674,"Grounding DINO text position embeddings are truncated to int64, collapsing 93% of them to zero","Yigtwxx",2,"","",0,"2026-07-31","2026-09-09"],[0,47672,"Encoder features are not cast to the text embedding dtype before masked_scatter in seven models","Yigtwxx",1,"Yigtwxx","#47673",0,"2026-07-31","2026-07-31"],[0,47667,"Add native Transformers support for Nanbeige4.2-3B","sushmanthreddy",4,"","",0,"2026-07-31",""],[0,47651,"Major Bug in parallel generation / left padding.","atbolsh",2,"","",0,"2026-07-30","2026-08-10"],[0,47649,"`modeling_qwen3_vl.py` performed D2H makes cpu wait gpu","NewUserHa",2,"","",0,"2026-07-30","2026-10-03"],[0,47641,"Rotary modules emit a self-triggered `FutureWarning`: `device` is deprecated for `compute_default_rope_paramet","albertvillanova",0,"Cyrilvallez","#47642",1,"2026-07-30","2026-07-30"],[0,47637,"Feature request: support epoch-relative logging, evaluation, and saving intervals","sachmatkris",3,"","",0,"2026-07-30","2026-08-11"],[0,47618,"Add Talkie (talkie-1930-13b)","ethanelasky",2,"","",0,"2026-07-28","2026-10-04"],[0,47612,"BOS and EOS token ID not within vocabulary warning for siglip2-base-patch16-naflex","willxxy",0,"zucchini-nlp","#48282",1,"2026-07-28","2026-09-06"],[0,47603,"Importing `AutoImageProcessor` fails when PyTorch is built without `torch.distributed`","hlky",0,"3outeille","#47619",1,"2026-07-28","2026-07-31"],[0,47602,"CodeLlamaTokenizer FIM/infilling drops correct leading whitespace on decode","Yashash-Yallapragada",2,"","",0,"2026-07-28","2026-08-01"],[0,47595,"Repetition penalty is gauge-dependent. Proposing opt-in normalized variant.","captainpete",3,"","",0,"2026-07-28","2026-10-21"],[0,47577,"NemotronH:  no longer respected after #47452 (unconditionally fetches Hub kernel)","albertvillanova",2,"","",0,"2026-07-27","2026-09-06"],[0,47559,"`create_bidirectional_sliding_window_mask` builds the mask on `encoder_hidden_states`' device — same model-par","abcgco",1,"abcgco","#47560",0,"2026-07-26","2026-07-28"],[0,47556,"Wav2Vec2 tokenizer rebuilds special-token lists per token in decode paths; phoneme variant carries a filter th","ishan-1010",1,"ishan-1010","#47557",0,"2026-07-26","2026-07-31"],[0,47553,"\"ValueError: Tokenizer class TokenizersBackend does not exist or is not currently imported\"","DXXS",2,"","",0,"2026-07-25","2026-07-29"],[0,47534,"`KeyError: 'mlp'` when building a cache for Nemotron-H — `\"mlp\"` missing from the layer-type mappings","qgallouedec",1,"qgallouedec","#47535",1,"2026-07-25","2026-07-27"],[0,47532,"Mamba2-family `cuda_kernels_forward` decode step is broken since #47452: missing seq-dim squeeze","qgallouedec",2,"","",0,"2026-07-25","2026-07-30"],[0,47530,"Phi3/Phimoe: prepare_inputs_for_generation forwards logits_to_keep=None into forward() -> 4-dim logits -> torc","pwasiewi",0,"Cyrilvallez","#47622",1,"2026-07-24","2026-07-30"],[0,47526,"`causal_conv1d_fn() got multiple values for argument 'seq_idx'` with the hub kernel","qgallouedec",1,"qgallouedec","#47527",1,"2026-07-24","2026-07-24"],[0,47514,"Transformers 5.13 CPU memory growth loading Qwen3-235B-A22B with DeepSpeed ZeRO-3; 4.51.0 remains bounded","narankhuut",2,"","",0,"2026-07-24","2026-09-09"],[0,47504,"MistralCommonBackend API documentation appears to render the dummy implementation instead of the actual class","samyuktahegde",2,"","",0,"2026-07-23","2026-09-09"],[0,47496,"`DataCollatorWithFlattening` uses 0-based `position_ids`, silently degrading RoBERTa-family models","tomaarsen",1,"tomaarsen","#47525",1,"2026-07-23","2026-07-30"],[0,47491,"[BUG] CANINE: dynamic ONNX export fixes sequence length to example input","sgpublic",2,"","",0,"2026-07-23","2026-10-14"],[0,47487,"CodeLlama tokenizer drops leading whitespace on decode round-trip","SuryanshSS1011",1,"SuryanshSS1011","#47488",0,"2026-07-22","2026-07-31"],[0,47480,"DTensor import in sharding_utils.py breaks on PyTorch < 2.5","gygdh-001",1,"gygdh-001","#47481",0,"2026-07-22","2026-07-23"],[0,47479,"`DTensor` import in `sharding_utils.py` breaks on PyTorch < 2.5","gygdh-001",2,"","",0,"2026-07-22","2026-07-22"],[0,47477,"Regarding security issue i have reported","liyander",2,"","",0,"2026-07-22","2026-08-31"],[0,47475,"[Zamba2] Causality violation in torch_forward: inter-chunk state recurrence reduces over the wrong axis","openaiteams",2,"","",0,"2026-07-22","2026-07-22"],[0,47473,"SentencePieceBackend should decode byte-fallback tokens","toslunar",2,"","",0,"2026-07-22","2026-09-14"],[0,47470,"ESMFold throws error when explicitly cast to fp16","crackalamoo",2,"","",0,"2026-07-22","2026-08-30"],[0,47467,"Qwen3.5-MoE (Qwen3_5MoeExperts) data-dependent expert dispatch breaks torch.utils.checkpoint reproducibility —","sternryan",3,"","",0,"2026-07-22","2026-08-11"],[0,47458,"[Bug] GPT2 c_proj weights incorrectly initialised","DavidJohnQuinlan",1,"DavidJohnQuinlan","#47459",0,"2026-07-21","2026-07-29"],[0,47455,"The kernel's version doesn't properly check.","Mohsen-Liaghat",2,"","",0,"2026-07-21","2026-08-30"],[0,47448,"gemma4 12B arch is not recognized!","HuangChiEn",2,"","",0,"2026-07-21","2026-08-28"],[0,47439,"Added-token encoder cache goes stale when tokenizers mutate _added_tokens_decoder directly, giving ids that do","ishan-1010",1,"ishan-1010","#47440",0,"2026-07-20","2026-08-28"],[0,47436,"AttributeError: 'Qwen2_5OmniConfig' object has no attribute 'max_position_embeddings'","VietDunghacker",2,"","",0,"2026-07-20","2026-07-20"],[0,47433,"DynamicCache.crop(0) retains the full KV cache and oversized negative crops retain stale tokens","HoneyTyagii",2,"","",0,"2026-07-20","2026-09-05"],[0,47429,"tokenizers<=0.23.0 upper bound blocks the only existing 0.23.x release","EfiSiapiti",2,"","",0,"2026-07-20","2026-08-21"],[0,47427,"`_is_hf_initialized` per-param check doesn't skip redundant init for built-in models on non-CUDA hardware","gygdh-001",2,"","",0,"2026-07-20","2026-08-28"],[0,47424,"Several slow tokenizers read all_special_tokens or all_special_ids once per token in decode loops","ishan-1010",1,"ishan-1010","#47425",0,"2026-07-20","2026-07-20"],[0,47407,"CompressedTensorsConfig(run_compressed=False) silently random-initializes unmapped fused-MoE expert weights on","narfian",2,"","",0,"2026-07-19","2026-08-26"],[0,47405,"Silent conversion skip on model_type mismatch: loading a converted checkpoint via the wrong Auto class yields ","SelfSL",2,"","",0,"2026-07-18","2026-08-26"],[0,47404,"device_map loading OOMs on conversion-op transients: WeightConverter contiguous() runs on destination GPU, unb","SelfSL",2,"","",0,"2026-07-18","2026-07-28"],[0,47402,"[i18n-<languageCode>] Translating docs to <languageName>","diplrcdi-cmd",2,"","",0,"2026-07-18","2026-07-20"],[0,47396,"MPS: DynamicCache produces NaN logits mid-decode in generate() (transformers >=5.2, regression from 5.0)","scp7",2,"","",0,"2026-07-18","2026-07-18"],[0,47394,"GitForCausalLM double-shifts the training loss and leaks targets across batch rows (regression from #35875)","Yigtwxx",1,"Yigtwxx","#47395",0,"2026-07-18","2026-07-21"],[0,47388,"[New model] PersonaPlex (full-duplex speech model, based on Moshi)","un1876",2,"","",0,"2026-07-17","2026-09-14"],[0,47378,"Question about overriding bnb_4bit_compute_dtype for pre-quantized checkpoints","John6666cat",2,"","",0,"2026-07-17","2026-09-01"],[0,47375,"Two systematic bugs in examples/pytorch/*_no_trainer.py: broken --resume_from_checkpoint auto-discover branch,","berkgungor",2,"","",0,"2026-07-16","2026-08-25"],[0,47373,"examples/training/distributed_training.py crashes on import: LOCAL_RANK/WORLD_SIZE/RANK are overwritten by OMP","berkgungor",2,"","",0,"2026-07-16","2026-07-16"],[0,47355,"GlmMoeDsaConfig: legacy `num_experts` unconditionally overrides explicit `n_routed_experts` (sibling of #46739","hermia-ai",2,"","",0,"2026-07-16","2026-08-25"],[0,47337,"## Flaky test: `IndexError` — bbox coordinate values out of 0-1000 range","gygdh-001",2,"","",0,"2026-07-15","2026-07-15"],[0,47333,"[Offloading] Cannot save disk-offloaded `Qwen/Qwen3-30B-A3B`","kylesayrs",1,"kylesayrs","#47336",0,"2026-07-15","2026-07-23"],[0,47332,"[Offloading] Cannot save disk-offloaded `Qwen3-VL-32B-Instruct`","kylesayrs",3,"","",0,"2026-07-15","2026-07-15"],[0,47331,"MetalConfig quantization: batched generate silently corrupts all rows after the first — affine_qmm_t computes ","fenilsonani",2,"","",0,"2026-07-15","2026-08-26"],[0,47328,"Qwen2.5-Omni Token2Wav DiT uses mismatched RoPE layout (half-split cos/sin with interleaved rotate) — regressi","HenryVarro666",1,"HenryVarro666","#47403",0,"2026-07-14","2026-07-23"],[0,47325,"`return_assistant_tokens_mask` masks the rest of the sequence when a `{% generation %}` span ends at token ind","sohumt123",2,"","",0,"2026-07-14","2026-07-17"],[0,47321,"AMD quark class not updated","debasisdwivedy",1,"debasisdwivedy","#47322",0,"2026-07-14","2026-07-31"],[0,47317,"PPFormulaNet training-loss bug","sharmax-vikas",2,"","",0,"2026-07-14","2026-09-01"],[0,47312,"Incompatible with kernels 0.16","jiqing-feng",2,"","",0,"2026-07-14","2026-08-14"],[0,47299,"[i18n-<supremeA AI Model ai.th AI >] Translating docs to <Copilot>","imgoingbacktheonlything239-ops",2,"","",0,"2026-07-13","2026-07-13"],[0,47293,"pr_build_doc_with_comment.yml is rejected at startup (job 'if:' references a job not in its 'needs:') — 3,086 ","kobihikri",2,"","",0,"2026-07-13","2026-08-17"],[0,47277,"Bug: `pthread_cond_clockwait` error in Termux/Android because of `esaxx-rs`","Manamama",2,"","",0,"2026-07-12","2026-07-12"],[0,47276,"Misleading synced_gpus warning message in continuous batching","Partha-Shankar",1,"Partha-Shankar","#47158",0,"2026-07-12","2026-07-14"],[0,47274,"stable diffusion in tensorflow ","deborshi-web",2,"","",0,"2026-07-12","2026-07-13"],[0,47273,"Cohere ASR chunk reassembly can crash on empty chunks and accept inconsistent metadata","AliOsm",2,"","",0,"2026-07-12","2026-09-11"],[0,47272,"Avoid repeated Cohere ASR encoder projection during cached generation","AliOsm",2,"","",0,"2026-07-12","2026-08-21"],[0,47254,"`CheckpointError` with PEFT + DeepSpeed ZeRO-3 + gradient checkpointing","qgallouedec",2,"","",0,"2026-07-10","2026-07-22"],[0,47246,"Incorrect inter-chunk recurrence in `NemotronHMamba2Mixer.torch_forward` (slow path). the Mamba2 fix from #351","Jeronymous",2,"","",0,"2026-07-10","2026-08-19"],[0,47228,"Sam3 bf16 compilation regression (v5.8.0 -> v5.9.0)","ianxmason",0,"David-Wu1119","#47238",0,"2026-07-10","2026-07-14"],[0,47225,"orthogonal_ weight init crashes with `geqrf_cuda not implemented for BFloat16` under v5 bf16 on-device init (R","janbernloehr",1,"janbernloehr","#47252",0,"2026-07-09","2026-07-17"],[0,47217,"VL models fail when their image placeholder is in user text.","samuelstevens",4,"Rocketknight1","#47386",1,"2026-07-09",""],[0,47205,"Standardize test image fixture loading for COCO images","tarekziade",0,"LevelVoid","#47218",0,"2026-07-09","2026-07-17"],[0,47177,"Tar-slip / zip-slip arbitrary file write → RCE ...","geo-chen",2,"","",0,"2026-07-08","2026-08-17"],[0,47176,"Path traversal via sharded-checkpoint index `weight_map` filenames in `get_checkpoint_shard_files` → arbitrary","geo-chen",2,"","",0,"2026-07-08","2026-08-17"],[0,47169,"BarkModel.enable_cpu_offload() strands submodules on CPU → device mismatch at generate() (transformers 4.57.6 ","psyiode",2,"","",0,"2026-07-08","2026-08-10"],[0,47145,"Silent corruption: FP8 compressed-tensors \"frozen\" checkpoints load without applying weight_scale","kvcache670",0,"rigen1048","#47152",0,"2026-07-07","2026-07-22"],[0,47139,"transformers chat routes management endpoints under /v1 when base_url includes /v1","VectorPeak",2,"","",0,"2026-07-07","2026-08-16"],[0,47138,"transformers chat routes management endpoints under /v1 when base_url includes /v1","VectorPeak",0,"dhruv7477","#47303",0,"2026-07-07","2026-07-22"],[0,47116,"Streaming Pipeline Regression in Transformers v5","davidmezzetti",0,"Rocketknight1","#47142",1,"2026-07-06","2026-07-09"],[0,47110,"special_tokens, 'list' object has no attribute 'keys'","Ratinax",2,"","",0,"2026-07-06","2026-08-18"],[0,47086,"Hybrid models: padding is not masked out of linear-attention/conv state on continued forwards (chunked prefill","abcgco",1,"abcgco","#47087",0,"2026-07-05","2026-07-20"],[0,47085,"Fast (Rust-backed) tokenizers aren't thread-safe anymore","dxqb",2,"","",0,"2026-07-05","2026-08-23"],[0,47070,"ASR pipeline: unconditional torchcodec import crashes on broken torchcodec installs","evnsnclr",2,"","",0,"2026-07-04","2026-08-10"],[0,47068,"Potential host synchronization in deepspeed_sp_compute_loss due to Python conditional on CUDA tensor","myeongsu-koo",0,"sshivampeta","#47073",0,"2026-07-04","2026-07-10"],[0,47067,"[i18n-<languageCode>] Translating docs to <languageName>","shirokuma-xv",2,"","",0,"2026-07-04","2026-07-06"],[0,47065,"DeformableDetrHungarianMatcher and LwDetrHungarianMatcher crash on non-finite cost matrices (\"cost matrix is i","MutugiD",2,"","",0,"2026-07-04","2026-07-06"],[0,47059,"tests/cli fail with huggingface_hub 1.22.0: 'HFCliTyperGroup' object has no attribute '_add_completion':","Madhav-000-s",0,"dhruv7477","#47064",0,"2026-07-04","2026-07-07"],[0,47050,"CanineModel fp16 ONNX export fails with dtype mismatch in attention (expected scalar type Float but found Half","sgpublic",0,"lcheng321","#47083",0,"2026-07-04","2026-07-06"],[0,47039,"GenerationConfig serializes ContinuousBatchingConfig as null","VectorPeak",0,"remi-or","#47291",1,"2026-07-03","2026-07-24"],[0,47031,"Checkpoint conversion silently loads transposed fused-expert weights when the swapped dims are equal (Transpos","pjordanandrsn",2,"","",0,"2026-07-02","2026-09-01"],[0,47030,"FineGrainedFP8 DeepGEMM path returns silently wrong results on SM100 (UE8M0 scale rounding without requantizat","keunhong",2,"","",0,"2026-07-02","2026-09-01"],[0,47020,"XLMRobertaTokenizer.__init__ passes dict to Unigram(vocab=...) expecting a Sequence","saattrupdan",2,"","",0,"2026-07-02","2026-07-08"],[0,47015,"Novel Application of Transformers: AI-Automated Cybersecurity Vulnerability Discovery System","Exploit-Garbage",2,"","",0,"2026-07-02","2026-07-02"],[0,47013,"[xLSTM] Crash in train mode with return_last_states=False: backend returns bare tensor but xLSTMLayer always u","Heatdh",2,"","",0,"2026-07-01","2026-08-10"],[0,47007,"Building AI-Automated Vulnerability Discovery Systems - Lessons from Using Advanced LLMs","Exploit-Garbage",2,"","",0,"2026-07-01","2026-07-02"],[0,47004,"Add LoMa keypoint detector & matcher","tcourat",4,"","",0,"2026-07-01",""],[0,47000,"RTDetrLoss crashing with \"ValueError: cost matrix is infeasible\"","m-matthias",2,"","",0,"2026-07-01","2026-09-04"],[0,46996,"Reporting a vulnerability","Eti-Hack",2,"","",0,"2026-07-01","2026-07-01"],[0,46984,"Add RWKV7 model support","chinoll",2,"","",0,"2026-06-30","2026-07-08"],[0,46983,"Add RWKV7 model support","chinoll",3,"","",0,"2026-06-30","2026-06-30"],[0,46981,"AutoProcessor.from_pretrained retries HTTP 429 responses multiple times for every config file when using HF_EN","peacefulotter",2,"","",0,"2026-06-30","2026-07-09"],[0,46972,"Add SuperCompress prompt compression for efficient inference","arjunkshah",2,"","",0,"2026-06-30","2026-06-30"],[0,46967,"Add AstroCLIP model support","giovannicozzolongo",2,"","",0,"2026-06-29","2026-07-02"],[0,46962,"[Qwen3] `create_masks_for_generate` return type change after #41251","hoshibara",2,"","",0,"2026-06-29","2026-07-06"],[0,46932,"[Gemma4] Fix dtype casting for quantized vision/audio embedders (same as #46904)","sharmax-vikas",1,"sharmax-vikas","#46933",0,"2026-06-27","2026-07-03"],[0,46931,"ตัวจัดการบัญชีสำหรับ L. CHOMPHUPHUEN.io","imgoingbacktheonlything239-ops",2,"","",0,"2026-06-27","2026-06-27"],[0,46909,"broken documentation example for DIA","Carotterap",0,"vasqu","#46923",1,"2026-06-26","2026-06-26"],[0,46901,"PPFormulaNet training-loss double-shift: trains against labels[..., 1:] instead of labels","OmkumarSolanki",1,"OmkumarSolanki","#46903",0,"2026-06-26","2026-07-23"],[0,46899,"[Gemma4] Audio (and vision) processing broken with 4‑bit quantization due to unsafe dtype cast in multimodal e","dukei",0,"praful-srinivasan-027","#46904",0,"2026-06-25","2026-06-26"],[0,46897,"Florence2 training-loss double-shift bug (same pattern as Moonshine #46784)","sharmax-vikas",1,"sharmax-vikas","#46898",0,"2026-06-25","2026-06-29"],[0,46894,"CohereASR training-loss double-shift bug (same pattern as Moonshine #46784)","sharmax-vikas",1,"sharmax-vikas","#46895",0,"2026-06-25","2026-07-21"],[0,46877,"auto_docstring prints [ERROR] doc-lint diagnostics to stdout at import time","CalamitousFelicitousness",2,"","",0,"2026-06-25","2026-06-25"],[0,46869,"YOLO model failing on multi GPU device map auto","Carotterap",0,"swankystark","#46886",0,"2026-06-24","2026-08-02"],[0,46868,"Sequence Parallelism Disabled for Evaluation","amanichopra",2,"","",0,"2026-06-24","2026-08-14"],[0,46858,"_prefill: cache.get_seq_length() returns device tensor, breaks slice arithmetic for static cache + torch.compi","silicon-dispatch-100",2,"","",0,"2026-06-24","2026-06-25"],[0,46851,"Qwen3.5 Does not work with variable length inputs using position_ids","kolehma8",2,"","",0,"2026-06-23","2026-06-24"],[0,46846,"[Qwen3.5] Missing `linear_attn` entries in `base_model_tp_plan` causes OOM and shape error at TP>1","tengomucho",1,"tengomucho","#47041",1,"2026-06-23","2026-07-03"],[0,46832,"[RT-DETR] TypeError in RTDetrModel.forward when num_feature_levels > len(backbone outputs)","c1prk",1,"c1prk","#46833",0,"2026-06-23","2026-06-29"],[0,46829,"[Bug] VideoPrismForVideoClassification returns last_hidden_state tensor as hidden_states in ImageClassifierOut","sharmax-vikas",1,"sharmax-vikas","#46830",0,"2026-06-23","2026-06-23"],[0,46823,"4-bit load_in_4bit + device_map='auto' + explicit max_memory still dispatches modules to CPU on a model that f","trevorgordon981",0,"Cyrilvallez","#47203",1,"2026-06-22","2026-07-09"],[0,46822,"ZeRO-3 zero.Init does not partition composite minimax_m3_vl language submodule -> OOM on multi-GPU load","trevorgordon981",4,"","",0,"2026-06-22",""],[0,46821,"[TimesFM 2.5]:`window_size` argument raises `AttributeError`","kanenorman",0,"kashif","#47363",1,"2026-06-22","2026-07-16"],[0,46785,"Is transformers 5.12.1 requires tokenizers<=0.23.0,>=0.22.0 really right?","jrp2014",2,"","",0,"2026-06-20","2026-09-26"],[0,46772,"Nvidia ModelOpt (NVFP4) compatibility for DiffusionGemma","ppabis",3,"","",0,"2026-06-19","2026-06-19"],[0,46762,"Substantial numerical differences with Minimax MSA attention vs Minimax's Reference","CoffeeVampir3",2,"","",0,"2026-06-19","2026-06-20"],[0,46756,"FR: Implement Downstream Integration Smoke Tests in CI/CD to Prevent Ecosystem Regression","briancullinan2",2,"","",0,"2026-06-18","2026-06-19"],[0,46752,"`apply_chat_template` crashes with `IndexError` when passed an empty list","sharmax-vikas",1,"sharmax-vikas","#46753",0,"2026-06-18","2026-06-22"],[0,46739,"GlmMoeDsa: head_dim→qk_rope_head_dim attribute_map silently clobbers the rope head dim (GLM-5.2)","Kasempiternal",2,"","",0,"2026-06-18","2026-06-18"],[0,46736,"LoRA fine-tuning of an FP8 checkpoint is blocked: `get_peft_model` does not clear `validate_quantization_for_t","brunopistone",2,"","",0,"2026-06-17","2026-08-18"],[0,46731,"LabelSmoother ignores num_items_in_batch, inflating gradients by ~gradient_accumulation_steps with label smoot","Incheonkirin",1,"Incheonkirin","#47261",0,"2026-06-17","2026-07-22"],[0,46728,"Aria: get_number_of_image_patches returns wrong count with max_image_size=980 when best resolution has odd 490","arnavkewalram",1,"arnavkewalram","#46732",0,"2026-06-17","2026-06-19"],[0,46726,"AutoProcessor reports “Unrecognized image processor” when required vision backend is missing","sisaman",1,"sisaman","#46727",0,"2026-06-17","2026-06-18"],[0,46723,"Several models hardcode `torch.float64` in their forward pass, crashing on MPS (Apple Silicon)","qflen",2,"","",0,"2026-06-17","2026-07-26"],[0,46711,"Clarify Gemma4 vision bidirectional mask behavior for full vs sliding attention","wnma3mz",2,"","",0,"2026-06-17","2026-06-30"],[0,46710,"Accuracy regression in DeepSeek-R1-Distill-Llama-8B after upgrading Transformers from 4.55 to 5.9","xiaowangintel",2,"","",0,"2026-06-17","2026-06-23"],[0,46701,"[DiffusionGemmaGenerationOutput] image-text-to-text pipeline does not work","KoichiYasuoka",1,"KoichiYasuoka","#46724",0,"2026-06-17","2026-06-23"],[0,46697,"Swin2SR image processor over-pads images when dimensions are already divisible by size_divisor","arnavkewalram",2,"","",0,"2026-06-16","2026-07-26"],[0,46693,"Performance regression with FA2 and long-context generation after 5.2.0","andreasgoulas",1,"andreasgoulas","#47134",0,"2026-06-16","2026-07-10"],[0,46692,"Out of memory after some epochs while training RT-DETR v2 model","Rayndell",2,"","",0,"2026-06-16","2026-08-10"],[0,46688,"[Qwen2VL] Image processor backend refactor (#43514) introduces pixel_values precision drift, causing bbox coor","chenxi971215",2,"","",0,"2026-06-16","2026-07-24"],[0,46682,"[Flex Attention] `Tensor` q_offset will cause wrong mask shape","hoshibara",2,"","",0,"2026-06-16","2026-06-23"],[0,46665,"Inconsistent behaviour when using `from_pretrained` inside Python classes","I-Absolutely-Am-Not-A-Robot",2,"","",0,"2026-06-15","2026-06-15"],[0,46664,"SmolVLM: Inconsistent meaning of num_frames in video processor","andreasgoulas",3,"","",0,"2026-06-15","2026-06-15"],[0,46661,"AutoModel.from_pretrained raises AttributeError: model_class has no attribute 'config_class' for remote-code m","atharv1945",1,"atharv1945","#46669",0,"2026-06-15","2026-06-15"],[0,46657,"Size of tensors mismatch when training object detection model","Rayndell",2,"","",0,"2026-06-15","2026-06-19"],[0,46656,"Inconsistent type annotation in `EvalPrediction`","pietrolesci",2,"","",0,"2026-06-15","2026-08-23"],[0,46655,"SmolVLM: Error due to incorrect padding calculation in video processor","andreasgoulas",0,"zucchini-nlp","#46662",1,"2026-06-15","2026-06-16"],[0,46653,"Mirko Privitera","mirko772",2,"","",0,"2026-06-15","2026-06-15"],[0,46650,"`save_pretrained` corrupts keys (`encoder.encoder.*`) for trust_remote_code / from_config models via non-idemp","Bluear7878",0,"Cyrilvallez","#46876",1,"2026-06-15","2026-06-29"],[0,46645,"Add SWA+sink option to Granite models","daviswer",1,"daviswer","#47179",0,"2026-06-15","2026-07-29"],[0,46644,"Consider reverting CLIP refactor","dxqb",2,"","",0,"2026-06-14","2026-07-26"],[0,46640,"Flash Attention 2 issue","srishtiimp",2,"","",0,"2026-06-14","2026-07-22"],[0,46629,"BUG? transformers version '5.12.0' gemma-4 generate DynamicSlidingWindowLayer","NickyDark1",2,"","",0,"2026-06-13","2026-07-20"],[0,46620,"transformers 5.x: from_pretrained silently returns random-init weights for custom PreTrainedModel subclasses (","gkriegspeedbay",2,"","",0,"2026-06-13","2026-06-22"],[0,46617,"FileNotFoundError when loading a trust_remote_code model from cache","ldkhang1201",2,"","",0,"2026-06-13","2026-07-22"],[0,46614,"Wav2Vec2PhonemeCTCTokenizer.phonemize keeps the previous language's espeak backend after switching back (stale","BlueCollarChris",2,"","",0,"2026-06-12","2026-08-22"],[0,46612,"Beam search cache reorder silently skipped for Mamba, XLNet, RWKV, and Reformer models (wrong generation outpu","wi5nuu",0,"Cyrilvallez","#46819",1,"2026-06-12","2026-06-24"],[0,46598,"MCP to help OSS devs understand huggingface/transformers","ankitzm",2,"","",0,"2026-06-12","2026-06-26"],[0,46597,"[`docs`] PULL_REQUEST_TEMPLATE.md has broken link to non-existent CONTRIBUTING.md section","tomaarsen",0,"stevhliu","#46606",1,"2026-06-12","2026-06-15"],[0,46577,"Add RTDetrForSegmentation","ranykamel",4,"","",0,"2026-06-12",""],[0,46570,"AutoModel loading fails with ROCm when model has custom CUDA kernels","hakuba-pamu",2,"","",0,"2026-06-11","2026-07-20"],[0,46566,"DiffusionGemma weights remain on 'meta' device with device_map=\"auto\"","HuangBugWei",0,"zucchini-nlp","#46625",1,"2026-06-11","2026-06-17"],[0,46561,"serve: Gemma 4 non-thinking responses returned as reasoning_content with empty content","b11015006",2,"","",0,"2026-06-11","2026-06-30"],[0,46559,"`apply_chat_template(return_assistant_tokens_mask=True)` returns an empty assistant mask for VLM processors wh","qgallouedec",3,"","",0,"2026-06-11","2026-06-11"],[0,46552,"Wav2Vec2CTCTokenizer in v5 treats the word delimiter as a special token, which leaks into get_special_tokens_m","ishan-1010",2,"","",0,"2026-06-11","2026-07-19"],[0,46551,"Wav2Vec2CTCTokenizer in v5 treats the word delimiter as a special token, which leaks into get_special_tokens_m","SHAI-ishan-katoch",2,"","",0,"2026-06-11","2026-06-11"],[0,46536,"[Gemma4/Gemma3n] MultimodalEmbedder.forward missing dtype alignment that subclass override already applies","BiggieW",2,"","",0,"2026-06-10","2026-06-10"],[0,46531,"[Gemma 4] Gemma4UnifiedForConditionalGeneration text-only inference produces degenerate output (token repetiti","synapticode-ai",2,"","",0,"2026-06-10","2026-06-11"],[0,46519,"StopStringCriteria misses CJK stop strings on byte-level tokenizers when a character splits into byte-fragment","Incheonkirin",1,"Incheonkirin","#46530",0,"2026-06-09","2026-06-12"],[0,46518,"Investigate training support for Sapiens2ForPoseEstimation when labels are provided","Sainava",1,"Sainava","#46764",0,"2026-06-09","2026-06-25"],[0,46516,"[Continuous batching] Tracker for issues and feature requests","remi-or",2,"","",0,"2026-06-09","2026-07-20"],[0,46515,"incompatibility with torch 2.4.1","nickjyj",2,"","",0,"2026-06-09","2026-06-09"],[0,46512,"from_config(dtype=...) builds weights in the requested dtype but leaves model.config.dtype stale","qflen",2,"","",0,"2026-06-08","2026-07-16"],[0,46506,"Different results for PPDocLayoutV3 on CPU and CUDA","99991",2,"","",0,"2026-06-08","2026-07-18"],[0,46493,"difference in performance of sam3 video model compared to original sam3 implementation","iariav",2,"","",0,"2026-06-08","2026-07-17"],[0,46491,"CodeLlama tokenizer strips one leading space on encode→decode round-trip (regression vs v4)","SuryanshSS1011",2,"","",0,"2026-06-08","2026-07-17"],[0,46489,"DeepSeek-Coder v1 tokenizer produces incorrect output on transformers v5+ (gap in PR #44801's fix)","SuryanshSS1011",2,"","",0,"2026-06-08","2026-06-28"],[0,46471,"New RAG framework: VORTEXRAG — 7-layer causal RAG achieving EM 74.8, Faithfulness 0.94","vignesh2027",2,"","",0,"2026-06-07","2026-07-17"],[0,46468,"[Bug] Official Nemotron example broken","Davidnet",2,"","",0,"2026-06-07","2026-07-07"],[0,46463,"[Bug] Incorrect error message for T5 model family's decoder input validation","Reigenleif",2,"","",0,"2026-06-06","2026-07-16"],[0,46459,"`dtype` is silently ignored when loading composite (multimodal) checkpoints through AutoModelForCausalLM (affe","qflen",1,"qflen","#46514",0,"2026-06-06","2026-06-11"],[0,46458,"Suggestion: Give a warning for 4D attention mask which use float dtype but in 0.0 and 1.0 term (instead of -in","AbdiHaryadi",0,"stevhliu","#46507",1,"2026-06-06","2026-06-10"],[0,46443,"Vision RoPE refactor","IlyasMoutawwakil",0,"zucchini-nlp","#48105",1,"2026-06-05","2026-09-04"],[0,46441,"Qwen3.5 crash with `cache_implementation=\"static\"` — `KeyError: 'linear_attention'`","dacorvo",2,"","",0,"2026-06-05","2026-07-06"],[0,46439,"Static cache `early_initialization` corrupts linear-attention layers (wrong shapes → `RuntimeError`)","dacorvo",2,"","",0,"2026-06-05","2026-07-06"],[0,46431,"ValueError: Cannot use apply_chat_template because this processor does not have a chat template.","AlphaNext",2,"","",0,"2026-06-05","2026-06-06"],[0,46426,"Add KORMo model","mjkmain",2,"","",0,"2026-06-04","2026-07-13"],[0,46424,"Cache config max_cache_length only honored for Quantized Cache","dacorvo",2,"","",0,"2026-06-04","2026-07-05"],[0,46421,"# Cache lazy initialization leads to spurious recompilation with chunked prefill","dacorvo",2,"","",0,"2026-06-04","2026-07-05"],[0,46402,"Sam3Model text encoder weights are not loaded in transformers 5.10.1","MalteEbner",0,"zucchini-nlp","#46406",1,"2026-06-04","2026-06-04"],[0,46399,"Qwen3.5: use_kernels=True crashes with ValueError on apply_rotary_pos_emb registration - missing decorator","rycerzes",0,"Bissmella","#46440",0,"2026-06-04","2026-06-05"],[0,46396,"`added_tokens_encoder` rebuilds the entire added-token map on every access","ishan-1010",1,"ishan-1010","#46535",0,"2026-06-04","2026-06-22"],[0,46389,"AutoProcessor import fails with NVIDIA torch 2.7 missing torch.float8_e8m0fnu","doctorpangloss",0,"andylin-hao","#46393",0,"2026-06-03","2026-06-05"],[0,46382,"Propagation of output_hidden_states to sub_configs","julien-gadonneix",4,"","",0,"2026-06-03",""],[0,46370,"Deprecate CompileableContextVar in favor of plain ContextVar with upcoming Dynamo native support","ayushsatyam146",4,"","",0,"2026-06-03",""],[0,46369,"Memory overflow occurs when calling the sam3-litetext model","mcmingchang",2,"","",0,"2026-06-03","2026-07-12"],[0,46365,"I couldn't import Trainer from transformers until I uninstalled datasets.","OneGneralf",2,"","",0,"2026-06-03","2026-07-12"],[0,46358,"[ExecuTorch] convert_and_export_with_cache fails for vision-language and audio models","ananyajoshi0203-cpu",2,"","",0,"2026-06-02","2026-07-12"],[0,46328,"SequenceFeatureExtractor.pad wasting time converting numpy array to list of numpy arrays","bolshoytoster",0,"Anai-Guo","#46329",0,"2026-06-01","2026-06-02"],[0,46326,"Bug in continue_final_message for tokenizer based on mistral_common","ntnq4",2,"","",0,"2026-06-01","2026-06-03"],[0,46315,"Regression: `convert_tokens_to_ids` is much slower in v5 than v4 for slow tokenizers with many added tokens","ichizok",0,"ishan-1010","#46323",0,"2026-06-01","2026-06-03"],[0,46305,"MoonshineStreaming: sliding_window_mask_function off-by-one — right_window_size=4 attends to 3 future frames i","Kush0610",2,"","",0,"2026-05-31","2026-07-10"],[0,46304,"Transformers v5.9.0 breaks support for pytorch with USE_DISTRIBUTED=0","MSDarwish2000",2,"","",0,"2026-05-31","2026-06-01"],[0,46298,"Noisy logging message","davidmezzetti",0,"skshmjn","#46350",0,"2026-05-30","2026-06-04"],[0,46297,"Safety Report: AI Agent Guardrails Do Not Work — 56-Day Empirical Proof","tzb1-ai",2,"","",0,"2026-05-30","2026-06-02"],[0,46295,"Training Support for FP8","Banaxi-Tech",4,"","",0,"2026-05-30",""],[0,46294,"Qwen3.5-9B video input fails with apply_chat_template — Video features and video tokens do not match","14790897",0,"kpal002","#46296",0,"2026-05-30","2026-06-02"],[0,46291,"Importing `transformers.pipelines` fails with kernels v0.15: kernels now requires `version` or `revision`","jshn9515",2,"","",0,"2026-05-29","2026-06-09"],[0,46285,"CLIPTextModel and CLIPTextModelWithProjection have inconsistent structure since v5.6 (.text_model removed from","kappacommit",2,"","",0,"2026-05-29","2026-05-29"],[0,46273,"Direct build of layer/attention silently produces non-causal output","ayghri",2,"","",0,"2026-05-29","2026-05-30"],[0,46271,"docs: clarify output_hidden_states behavior for Koopman/spectral analysis use cases","hinanohart",2,"","",0,"2026-05-28","2026-05-29"],[0,46247,"Faster preprocessing when using new Parakeet TDT please","BBC-Esq",2,"","",0,"2026-05-27","2026-05-29"],[0,46245,"[`Layer Types`] Split keys across MLP and Attentions","vasqu",0,"zucchini-nlp","#47517",1,"2026-05-27","2026-07-27"],[0,46238,"DeepSeek-V4 FP8 load fails: 1→1 WeightConverter w2 weights not dequantized due to $-anchored source pattern","moonhush922",2,"","",0,"2026-05-27","2026-06-02"],[0,46234,"`is_falcon_mamba_ssm_available` missing from `transformers.utils.import_utils`","voloha-mil",0,"Anai-Guo","#46343",0,"2026-05-27","2026-06-08"],[0,46231,"DOC: set_training() documents parameter as `num_train_epochs` but actual parameter name is `num_epochs`","kratos0718",2,"","",0,"2026-05-27","2026-05-28"],[0,46228,"ImportError: cannot import name 'AutoModelWithLMHead' from 'transformers'","hannahlhargrove",2,"","",0,"2026-05-27","2026-07-05"],[0,46226,"1 ขอให้เจอสิ่งดีๆ คิดดี ทำดี เจอแต่สิ่งดีๆ สาธุ","imgoingbacktheonlything239-ops",2,"","",0,"2026-05-27","2026-05-29"],[0,46222,"คิดดี ทำดี เจอแต่สิ่งดี สาธุ 365.ai.th","imgoingbacktheonlything239-ops",4,"","",0,"2026-05-26",""],[0,46218,"Could not import module 'Qwen3_5ForCausalLM","Arslan-Mehmood1",2,"","",0,"2026-05-26","2026-07-05"],[0,46217,"convert_file_size_to_int crashes on bare unit string like GIB","ghost",2,"","",0,"2026-05-26","2026-05-26"],[0,46216,"is_torch_hpu_available permanently monkeypatches global torch functions","ghost",2,"","",0,"2026-05-26","2026-07-05"],[0,46215,"_sanitize_module_name doesnt handle path traversal chars","ghost",2,"","",0,"2026-05-26","2026-07-06"],[0,46209,"[BUG] FP8 model cannot inference on B200 (SM>=100)","xin3he",2,"","",0,"2026-05-26","2026-08-05"],[0,46206,"There is a PR in the queue Reverting security patches","rickgonzalez",2,"","",0,"2026-05-26","2026-05-26"],[0,46202,"Roundtrip Failure for Gemma Pipeline on \"▁\"","julmb",2,"","",0,"2026-05-25","2026-05-28"],[0,46192,"return_tensors is silently ignored when text_kwargs is explicitly passed","leveretconey",0,"Ace3Z","#46195",0,"2026-05-25","2026-05-29"],[0,46190,"Qwen3.5 GatedDeltaNet: Large logit divergence between full-sequence forward and prefill+decode with cache","JWQZ",2,"","",0,"2026-05-25","2026-07-09"],[0,46187,"cohere2_moe fails training + tensor parallel tests","3outeille",2,"","",0,"2026-05-25","2026-05-26"],[0,46186,"[New Model] Add TIGER — Recommender Systems with Generative Retrieval (Google DeepMind, NeurIPS 2023)","raviteja-ganta",2,"","",0,"2026-05-25","2026-05-25"],[0,46185,"[New Model] Add Microsoft Samba - Hybrid SSM + Sliding Window Attention","raviteja-ganta",2,"","",0,"2026-05-25","2026-05-25"],[0,46179,"[New model] Add Fun-ASR-Nano (FunAudioLLM/Fun-ASR-Nano-2512)","LauraGPT",1,"LauraGPT","#46180",0,"2026-05-24","2026-07-03"],[0,46178,"ValueError: Backend should be defined in the BACKENDS_MAPPING. Offending backend: tensorflow_text","marchcat69",2,"","",0,"2026-05-24","2026-07-03"],[0,46175,"[Gemma4] `Gemma4VisionPatchEmbedder._position_embeddings` materializes a ~19 GiB one-hot tensor that's mathema","kuso2006",0,"Sriniketh24","#46176",0,"2026-05-23","2026-05-28"],[0,46172,"[BUG] deepseek-v4 `comb.to(dtype).transpose(-1, -2)`","Jintao-Huang",2,"","",0,"2026-05-23","2026-05-23"],[0,46170,"ImportError / TypeError on Windows with AMD ROCm PyTorch due to torch.distributed dependency","mjakob-gh",0,"danyalahmed1995","#46205",0,"2026-05-23","2026-05-29"],[0,46167,"[deepseek_v4] save_pretrained silently downcasts FP32 tensors to BF16 (hc_*, attn_sink, ffn.gate.bias, compres","pasta-paul",0,"ArthurZucker","#46198",1,"2026-05-23","2026-05-27"],[0,46165,"Unable to resume model training with a locally downloaded model and trust remote code","conceptofmind",2,"","",0,"2026-05-22","2026-10-12"],[0,46159,"[RT-DETRv2] MPS crash: build_2d_sinusoidal_position_embedding hardcodes torch.float64, breaking Apple Silicon ","shubhammr21",0,"guarin","#47897",1,"2026-05-22","2026-08-25"],[0,46153,"Regression in v5.x: `ProcessorMixin._load_tokenizer_from_pretrained` forces subfolder for non-primary sub-toke","HanfengLiao",2,"","",0,"2026-05-22","2026-06-29"],[0,46150,"​[Feature Proposal] Cognitive Weighting Protocol: Hierarchical context management to mitigate \"Contextual Drif","Nossari",2,"","",0,"2026-05-22","2026-05-22"],[0,46144,"RoFormer attention implementation does not use attention interface","ir2718",4,"","",0,"2026-05-21",""],[0,46143,"`**kwargs` not passed through methods of RoFormer models","ir2718",1,"ir2718","#46171",0,"2026-05-21","2026-05-25"],[0,46139,"Discussion: optional RankSEG-style decoding for Transformers semantic segmentation post-processing","Leev1s",2,"","",0,"2026-05-21","2026-07-06"],[0,46133,"Add TIPSv2 (Advancing Vision-Language Pretraining with Enhanced Patch-Text Alignment) by Google DeepMind","farrosalferro",0,"Ternura143","#46347",0,"2026-05-21","2026-07-07"],[0,46132,"AttentionInterface.register changes behavior of registered function","pjc15111",2,"","",0,"2026-05-20","2026-05-21"],[0,46129,"[deepseek_v4] conversion_mapping doesn't cover mtp.* paths — MTP keys silently random-init even after _keys_to","pasta-paul",2,"","",0,"2026-05-20","2026-06-29"],[0,46123,"MaskGenerationPipeline: is_last never True on final partial batch, silently dropping results","J3r3myPerera",1,"J3r3myPerera","#46136",0,"2026-05-20","2026-05-25"],[0,46121,"`convert_rope_params_to_dict` raises `TypeError` when `ignore_keys_at_rope_validation` is a JSON-loaded list","Charly21r",1,"Charly21r","#46142",0,"2026-05-20","2026-05-25"],[0,46097,"Path Traversal in Sharded Checkpoint Loader via Unsanitized `weight_map` Entries in `*.index.json`","karnakarreddi",2,"","",0,"2026-05-20","2026-06-28"],[0,46095,"[deepseekv4]Does Transformers provide a weight conversion script to convert the Hugging Face weights into a fo","young-creator",2,"","",0,"2026-05-20","2026-06-02"],[0,46093,"[Inquiry] What is the current status and roadmap for supporting packed sequences (packing) on Qwen3.5/hybrid l","liangxuZhang",2,"","",0,"2026-05-20","2026-05-26"],[0,46082,"LlamaConfig rejects explicit head_dim when hidden_size is not divisible by num_attention_heads","Anri-Lombard",2,"","",0,"2026-05-19","2026-06-28"],[0,46077,"Gemma4 `use_bidirectional_attention=\"all\"` still builds causal attention masks","oliverholworthy",1,"oliverholworthy","#46079",0,"2026-05-19","2026-05-26"],[0,46050,"Model quantized via sinq broken after save_pretrained and from_pretrained","k355l3r-5yndr0m3",2,"","",0,"2026-05-19","2026-06-27"],[0,46032,"Mamba2Mixer: use_cache with seq_len > 1 silently produces incorrect results (both CPU and GPU paths)","clara2911",0,"Ramshankar07","#46084",0,"2026-05-18","2026-06-23"],[0,46018,"DeepSeek-V4 shared expert not gated.","yuzhongw-nvidia",2,"","",0,"2026-05-18","2026-06-19"],[0,46016,"Code quality scan: 164 findings (A-, 84/100)","repobilitycom",2,"","",0,"2026-05-17","2026-05-18"],[0,46015,"Outdated examples about RAG","someone282801",0,"Sriniketh24","#46035",0,"2026-05-17","2026-05-18"],[0,46002,"NaNs in classification heads upon checkout + init","maciejskorski",0,"Rocketknight1","#46030",1,"2026-05-16","2026-05-19"],[0,46000,"🎯 AI Agents: Claim tasks and execute them here","chenyuan35",2,"","",0,"2026-05-16","2026-05-18"],[0,45998,"Security Vulnerability","LinZiyuu",2,"","",0,"2026-05-16","2026-05-18"],[0,45987,"[Bug] StaticCache.get_seq_length() returns shape-(1,) Tensor despite -> int contract","Abineshabee",2,"","",0,"2026-05-15","2026-06-22"],[0,45950,"BAN","Priyabhunia",2,"","",0,"2026-05-13","2026-05-15"],[0,45941,"FSDP + KD-teacher-wrap + flash-attn-2 + LoRA: working-set memory exceeds 40 GiB per rank on production-shape S","drewvenegas",2,"","",0,"2026-05-13","2026-05-13"],[0,45938,"[DeepSeekV4] Compressor does not seem to account for padding tokens when forming compressed KV blocks","WKQ9411",4,"","",0,"2026-05-13",""],[0,45925,"Discrepancy Between the Paper’s Claimed Sparse Attention Complexity and the HuggingFace Implementation Caused ","I-hercules",2,"","",0,"2026-05-13","2026-05-14"],[0,45923,"Nemotron-3-Nano-Omni: supports_gradient_checkpointing flag missing on trust_remote_code variant (1-line fix)","badlerSI",2,"","",0,"2026-05-12","2026-05-13"],[0,45920,"AutoTokenizer produces wrong token IDs for OLMo2, HyperClovaX, DeepSeek-R1-Distill-Llama, Yi, and others (v5 r","kndtran",0,"itazap","#46091",1,"2026-05-12","2026-06-19"],[0,45910,"[DeepSeekV4] Potential RoPE theta mismatch between main attention and compressed KV branches","WKQ9411",2,"","",0,"2026-05-12","2026-05-12"],[0,45907,"Our `tinker-cookbook` CI broke: `list_repo_files` should forward the `revision` argument","nealwu",1,"nealwu","#45908",0,"2026-05-12","2026-05-12"],[0,45902,"`Qwen3_5MoeTextRotaryEmbedding.inv_freq` reads uninitialized memory after `meta → to_empty(cuda)` materializat","jamesbraza",2,"","",0,"2026-05-11","2026-05-13"],[0,45901,"table-question-answering task crashes","someone282801",0,"Rocketknight1","#45916",1,"2026-05-11","2026-05-28"],[0,45874,"Gemma4-E2B/E4B: passing `inputs_embeds` triggers an extremely expensive reverse embedding lookup","thijs-vanweezel",0,"Cyrilvallez","#45927",1,"2026-05-10","2026-05-14"],[0,45865,"[Feature Request] Add lossy speculative decoding via static ensemble verification","kasakh",1,"kasakh","#45979",0,"2026-05-10","2026-06-17"],[0,45864,"PretrainedConfig.from_pretrained: silent default on missing config.json","MilkClouds",3,"","",0,"2026-05-09","2026-05-09"],[0,45859,"`Qwen3_5MoeTextRotaryEmbedding.forward` is not compatible with CPU offload","jamesbraza",1,"jamesbraza","#45861",0,"2026-05-09","2026-05-12"],[0,45854,"pipeline text-generation ignores return_full_text=False with chat template","yashpandya-msr",0,"Abineshabee","#45860",0,"2026-05-09","2026-05-12"],[0,45853,"Provide HF_USE_MLX=0 / public flag to disable MLX backend detection at import time","Steve-Allison",2,"","",0,"2026-05-08","2026-05-25"],[0,45850,"Since 5.0 version it breaks too many models","LOYINuts",2,"","",0,"2026-05-08","2026-06-16"],[0,45841,"Embed Agent Friendly Code Score Badge","hsnice16",2,"","",0,"2026-05-08","2026-06-08"],[0,45834,"Kosmos2.5: index error on long ocr input","nunq",1,"nunq","#45835",0,"2026-05-08","2026-05-28"],[0,45823,"Gemma4 PLE device mismatch with `device_map=\"auto\"` during forward","rishon-galileo",0,"zucchini-nlp","#45826",1,"2026-05-07","2026-05-07"],[0,45820,"DeepSeekV4 的 CSA Indexer 因果 mask","slZheng077",2,"","",0,"2026-05-07","2026-05-12"],[0,45812,"`AutoTokenizer` produces wrong token IDs for all Granite models (silent v4→v5 regression)","kndtran",1,"kndtran","#45813",0,"2026-05-06","2026-05-18"],[0,45810,"Feature request Add Qwen3_5ForTokenClassification for using as a value model.","han2-l",2,"","",0,"2026-05-06","2026-08-11"],[0,45803,"[Bug] Bare `except:` in FuyuBatchFeature.convert_to_tensors() swallows KeyboardInterrupt and hides real errors","Abineshabee",1,"Abineshabee","#45808",0,"2026-05-06","2026-05-08"],[0,45800,"incompatiblity between torch 2.4.1 and transformers 5.8.0","nickjyj",0,"rraghavkaushik","#45804",0,"2026-05-06","2026-05-08"],[0,45797,"[BUG] HF Hub is DOWN on AWS in Xet mode","michaelroyzen",2,"","",0,"2026-05-05","2026-05-07"],[0,45796,"grootN_training","zeic1",2,"","",0,"2026-05-05","2026-05-06"],[0,45779,"[Windows] RTX 5070 Ti (Blackwell sm_120) - setup and deployment notes","loongmiaow-pixel",2,"","",0,"2026-05-05","2026-06-22"],[0,45768,"Macro Privitera Pitbull 24-10-2025","mirko772",2,"","",0,"2026-05-03","2026-05-04"],[0,45767,"Privitera Fabrizio 16-12-1967","mirko772",2,"","",0,"2026-05-03","2026-05-04"],[0,45766,"Verduci Caterina 21-07-1964","mirko772",2,"","",0,"2026-05-03","2026-05-04"],[0,45765,"Erika Privitera 14-08 -2003","mirko772",2,"","",0,"2026-05-03","2026-05-04"],[0,45764,"Daniele Privitera 14-05-1998","mirko772",2,"","",0,"2026-05-03","2026-05-04"],[0,45763,"Ivan privitera 02-09-1993","mirko772",2,"","",0,"2026-05-03","2026-05-04"],[0,45762,"Mirko Privitera 30-09-1990","mirko772",2,"","",0,"2026-05-03","2026-05-04"],[0,45761,"Veneto","mirko772",2,"","",0,"2026-05-03","2026-05-04"],[0,45759,"`AutoModelForCausalLM.from_config` does not unwrap `text_config` for composite Qwen 3.5 and 3.6 multimodal con","jamesbraza",1,"jamesbraza","#45770",0,"2026-05-03","2026-05-05"],[0,45758,"DeepSeek-V4 CSA eager path may not preserve per-query top-k masking for S > 1","kekmodel",2,"","",0,"2026-05-03","2026-05-12"],[0,45753,"Qwen3_5 goes into infinite loop for a specific image","MHRDYN7",0,"poojansatani","#46007",0,"2026-05-03","2026-06-08"],[0,45750,"`Qwen3VLVisionPatchEmbed.proj` (`nn.Conv3d` with `stride == kernel`) is ~50,000× slower than equivalent `nn.Li","WangYuHang-cmd",2,"","",0,"2026-05-03","2026-06-15"],[0,45736,"Please update `tokenizers` version check","lalala-233",0,"ArthurZucker","242f5df",1,"2026-05-01","2026-08-17"],[0,45735,"Bug with detecting cache positions in sdpa_mask","davidmezzetti",0,"Cyrilvallez","#45884",1,"2026-05-01","2026-05-11"],[0,45727,"fix(generation): correct spelling mistake in continuous_api docstring","lohdptm",2,"","",0,"2026-04-30","2026-05-08"],[0,45721,"Pass library_name=\"transformers\" to HfApi/Hub functions in push paths so commits attribute correctly","davanstrien",0,"Wauplin","#46318",1,"2026-04-30","2026-06-03"],[0,45715,"PreTrainedTokenizer.convert_ids_to_tokens(skip_special_tokens=True) rebuilds all_special_ids on every iteratio","longlee0622",0,"i3hz","#45728",0,"2026-04-30","2026-05-04"],[0,45712,"Six leftover dummy classes in `dummy_pt_objects.py` fail `check_repo.py` and leak into `dir(transformers)` wit","jw9603",1,"jw9603","#45722",0,"2026-04-30","2026-04-30"],[0,45710,"Fix wrong repo link in Dinov2ForImageClassification doc example","Milan-Bhimani",2,"","",0,"2026-04-30","2026-05-01"],[0,45706,"Hive Civilization — x402-native services for transformers Agents (notification)","srotzin",2,"","",0,"2026-04-30","2026-04-30"],[0,45704,"T5 silently uses apex.FusedRMSNorm which has a memory leak (NVIDIA/apex#1999)","dustnehowl",0,"Rocketknight1","#45723",1,"2026-04-29","2026-05-01"],[0,45701,"transformers version changes the tokenization","PiRom1",2,"","",0,"2026-04-29","2026-06-14"],[0,45698,"from_pretrained loads wrong custom module after save_pretrained","nurpax",2,"","",0,"2026-04-29","2026-07-01"],[0,45696,"Improving CLI Serving Code Structure with Class-Based FastAPI Patterns","HeHongyeFY",2,"","",0,"2026-04-29","2026-04-29"],[0,45693,"Why the calculation of train_batch_size unrelated to split_batches","mklpr",2,"","",0,"2026-04-29","2026-06-07"],[0,45685,"[moe] mps interface has error  \"histogram_mps\" not implemented for 'Int'","chenzhe1204",0,"rigen1048","#45687",0,"2026-04-28","2026-05-05"],[0,45684,"save_pretrained` (with `register_for_auto_class`) propagates read-only permissions from custom-model source fi","nurpax",1,"nurpax","#45686",0,"2026-04-28","2026-04-29"],[0,45676,"Gemma 4: Exploding pre-clip gradient norms during LoRA fine-tuning of `gemma-4-31B-it`","pritmish",2,"","",0,"2026-04-28","2026-05-02"],[0,45674,"[BitsAndBytesConfig] Providing llm_int8_skip_modules clears the default lm_head exclusion, causing AssertionEr","softguy777",2,"","",0,"2026-04-28","2026-06-06"],[0,45672,"[Gemma4] torch.finfo() TypeError on uint8 weights in audio modules during 4-bit (NF4) inference","softguy777",2,"","",0,"2026-04-28","2026-05-28"],[0,45663,"Gemma-4 training with FSDP2 raises `KeyError` in `Gemma4TextAttention.forward` because `shared_kv_states` is r","jamesbraza",2,"","",0,"2026-04-27","2026-05-14"],[0,45657,"ValueError in zero_shot_object_detection.md doctest on Python 3.13","AnkitAhlawat7742",1,"AnkitAhlawat7742","#45669",0,"2026-04-27","2026-04-28"],[0,45656,"Optimizer step being called 2 times when using deepspeed","harsh2912",2,"","",0,"2026-04-27","2026-04-27"],[0,45647,"MusicgenMelody ignores audio conditioning (regression between 4.48 and 4.57)","audiodude",2,"","",0,"2026-04-25","2026-06-03"],[0,45646,"NLP","mariam12-dotcom",2,"","",0,"2026-04-25","2026-04-27"],[0,45644,"mps: test_eager_matches_sdpa_inference tests fail with PyTorch MPS backend","qflen",2,"","",0,"2026-04-25","2026-04-27"],[0,45636,"Proposal: add sdpa_memeff attn_implementation for shape combinations no fast backend covers","dvdimitrov13",4,"","",0,"2026-04-24",""],[0,45632,"`trust_remote_code` cache path collides for local models sharing a leaf directory name","nurpax",0,"Jeevang1-epic","#45642",0,"2026-04-24","2026-04-29"],[0,45600,"auto_mappings.py references removed Sam3LiteText configs, breaking CI","artem-spector",2,"","",0,"2026-04-23","2026-04-23"],[0,45593,"D-FINE not using any auxiliary losses when denoising is turned off","m-matthias",0,"Abineshabee","#45601",0,"2026-04-23","2026-04-23"],[0,45588,"`integrations/flash_attention.py` crashes with `AttributeError` on `s_aux=None` for sink-less models","jamesbraza",1,"jamesbraza","#45589",0,"2026-04-23","2026-04-23"],[0,45584,"Whisper generation fails on empty transcription after align_special_tokens","ronansgd",2,"","",0,"2026-04-22","2026-06-10"],[0,45571,"A nice UX for generating dynamic tensors that break torch.compile/torch.export","IlyasMoutawwakil",4,"","",0,"2026-04-22",""],[0,45563,"Paged generate() emits a stale warning for num_return_sequences","oleksii-tumanov",0,"joaquinhuigomez","#45582",0,"2026-04-22","2026-04-24"],[0,45561,"[Bug] pytest-xdist workers race on captured_info.txt in patched testing utils","oleksii-tumanov",2,"","",0,"2026-04-22","2026-07-27"],[0,45557,"Pipeline num_workers runtime default is 0 but documentation states 8","Leonater",2,"","",0,"2026-04-21","2026-04-23"],[0,45545,"Transformers is trying to call home despite `local_files_only=True`","Sur3",2,"","",0,"2026-04-21","2026-04-22"],[0,45542,"Only tensorboard is installed without TensorFlow, causing undefined tf backend error","nursery42",2,"","",0,"2026-04-21","2026-04-23"],[0,45538,"CLIPTokenizer uses 10**30 as `model_max_length`","D1-3105",2,"","",0,"2026-04-21","2026-05-29"],[0,45536,"Feature Request: Add SCAO Optimizer integration for 1.5x faster fine-tuning throughput","whispering3",2,"","",0,"2026-04-20","2026-04-21"],[0,45529,"Add `Olmo2ForSequenceClassification` (and ideally `OlmoForSequenceClassification` / `Olmo3ForSequenceClassific","earino",1,"earino","#45551",0,"2026-04-20","2026-04-22"],[0,45522,"Feature request: Flash Attention 2 support for T5Gemma 2","arunkumarchithanar",2,"","",0,"2026-04-20","2026-05-15"],[0,45521,"T5Gemma2: decoder self-attention fixed 4097-element mask at batch=1, fails on inputs >4094 tokens","arunkumarchithanar",0,"Beichen-Ma","#45540",0,"2026-04-20","2026-04-27"],[0,45520,"KeyError: 'flash_attn' in import_utils.py when running on Python 3.13","laichiaheng",0,"SAY-5","#45524",0,"2026-04-20","2026-05-26"],[0,45518,"Expose static_graph DDP flag via TrainingArguments","KeitaW",1,"KeitaW","#45519",0,"2026-04-20","2026-04-21"],[0,45517,"MPS OOM error, finetuning T5Gemma2 with Seq2SeqTrainer","Tokarak",3,"","",0,"2026-04-19","2026-04-20"],[0,45507,"GraniteMoEHybrid Model Calls Invalid Method","rnowling",0,"tianhaocui","#45514",0,"2026-04-18","2026-04-27"],[0,45496,"Add V-JEPA 2.1 inference support","davevanveen",4,"","",0,"2026-04-17",""],[0,45491,"[Gemma3] NaN embeddings on GPU when batching sequences of mixed length (sliding window attention + all-padding","RiccardoTOTI",2,"","",0,"2026-04-17","2026-04-21"],[0,45488,"LlamaTokenizer in v5 overrides tokenizer.json's ByteLevel pre-tokenizer with Metaspace, silently breaks DeepSe","dc3671",0,"itazap","#45936",1,"2026-04-17","2026-07-15"],[0,45482,"Gemma4 26B-A4B: cross-device errors with CPU offload (RoPE, inputs, layer_scalar, SDPA mask, mm_token_type_ids","sirfyyn",2,"","",0,"2026-04-16","2026-05-25"],[0,45479,"`problem_type=\"single_label_classification\"` with `num_labels=1` leads to degenerate zero loss across multiple","BohdanBabii",0,"gaurav0107","#45611",0,"2026-04-16","2026-04-24"],[0,45478,"[BUG] transformers>=5.4.0, Qwen3.5 Moe from_pretrained error","Jintao-Huang",2,"","",0,"2026-04-16","2026-04-20"],[0,45468,"[BUG] Gemma-4 Gemma4AudioRelPositionalEncoding","foldl",0,"eustlb","#45606",1,"2026-04-16","2026-04-27"],[0,45464,"chat/completions API fail on Qwen3.5-0.8B for streaming inference","zhangwei217245",2,"","",0,"2026-04-15","2026-05-24"],[0,45459,"`except import_protobuf_decode_error()` hides real tokenizer errors when protobuf isn't installed","jw9603",1,"jw9603","#45486",0,"2026-04-15","2026-04-20"],[0,45458,"Add typing support incrementally (meta issue)","tarekziade",0,"moonbogi","#45703",0,"2026-04-15","2026-04-30"],[0,45447,"granitemoehybrid: HybridMambaAttentionDynamicCache missing from modeling_granitemoehybrid — breaks ibm-granite","Steve-Allison",2,"","",0,"2026-04-15","2026-05-18"],[0,45446,"Incorrect PyTorch version check for AuxRequest import in flex_attention","ZSLsherly",2,"","",0,"2026-04-15","2026-04-15"],[0,45440,"Native `DeepseekV3MoE` diverges from the remote DeepSeekV3 implementation","casinca",1,"casinca","#45441",0,"2026-04-14","2026-04-22"],[0,45431,"Wrong checkpoint path in Dinov2 model_docs","ambroiseodt",2,"","",0,"2026-04-14","2026-04-15"],[0,45419,"Chat template inconsistencies in tool-calling support","qgallouedec",4,"","",0,"2026-04-13",""],[0,45412,"RT-DETR models do not release memory when deleted / garbage-collected","dhdaines",0,"yonigozlan","#45922",0,"2026-04-13","2026-05-15"],[0,45406,"transformers serve crashes with AttributeError: 'Gemma4Processor' object has no attribute '_tokenizer'","asdat3",2,"","",0,"2026-04-13","2026-04-13"],[0,45405,"`MIN_PEFT_VERSION` bumped to 0.18.2 which is not yet released on PyPI","artem-spector",2,"","",0,"2026-04-13","2026-04-13"],[0,45399,"Fallback to kernels-community/flash-attn2 is blocked by other checks when fa2 is not installed","efsotr",2,"","",0,"2026-04-13","2026-04-14"],[0,45397,"[BUG] gemma-4 zero3 from_pretrained","Jintao-Huang",0,"saslifat-gif","#45402",0,"2026-04-13","2026-04-17"],[0,45390,"CLIPTextModel / CLIPVisionModel fail to load old checkpoints after architecture flattening","sayakpaul",2,"","",0,"2026-04-13","2026-05-21"],[0,45381,"transformers==5.3.0, qwen2.5-vl video input vision_position_ids seems to be wrong","bicheng-xu",0,"zucchini-nlp","#45400",1,"2026-04-11","2026-04-14"],[0,45376,"[Bug] Loading gemma4 on transformers v4 raises misleading AttributeError: 'list' object has no attribute 'keys","harishcs01",2,"","",0,"2026-04-11","2026-05-20"],[0,45375,"Qwen3_5MoeVisionConfig missing deepstack_visual_indexes field — silently dropped by @strict","sharonyu-115",2,"","",0,"2026-04-11","2026-05-20"],[0,45373,"Add Gemma4ForSequenceClassification (missing from gemma4 module — Gemma 2/3 have it)","LarsKlawitter",2,"","",0,"2026-04-11","2026-08-28"],[0,45372,"ImportError: cannot import name 'ReasoningEffort' from mistral_common breaks Gemma 4 processor loading in tran","Glademist",2,"","",0,"2026-04-11","2026-06-17"],[0,45362,"Qwen3.5-35B crashes with transformers chat","Sector14",0,"sharziki","#45368",0,"2026-04-10","2026-04-13"],[0,45357,"[Regression] Qwen3.5 `save_pretrained` still saves incorrect visual encoder keys in 5.5.3","johnking0099",0,"Cyrilvallez","#45358",1,"2026-04-10","2026-04-10"],[0,45356,"Regression in Kimi-K2.5 tokenizer from 5.3.0 to 5.4.0: incorrect codec handling and misleading fix_mistral_reg","Lander-Hatsune",0,"ArthurZucker","#45359",1,"2026-04-10","2026-04-13"],[0,45341,"A little bug in testing_utils.py","MHRDYN7",0,"RudrenduPaul","#45351",0,"2026-04-09","2026-05-11"],[0,45335,"[t5gemma] resize_token_embeddings does not effect to decoder.embed_tokens","KoichiYasuoka",2,"","",0,"2026-04-09","2026-04-16"],[0,45331,"[Gemma4] Bug: audio token missing newline separators in chat_template.jinja causes multimodal failure when ima","LfWhat",2,"","",0,"2026-04-09","2026-05-11"],[0,45325,"Qwen2.5-VL get_rope_index scales still-image temporal position_ids by tokens_per_second in transformers 5.3.0","ayaan-fw",0,"Kash6","#45330",0,"2026-04-08","2026-04-10"],[0,45322,"[Model Request] Add EUPE (Efficient Universal Perception Encoder) by Meta AI","forensicmike",0,"molbap","#46691",1,"2026-04-08","2026-06-16"],[0,45313,"Qwen3.5: DeepSpeed ZeRO-3 fails to load weights for language_model","debOliveira",2,"","",0,"2026-04-08","2026-05-20"],[0,45310,"[BUG] transformers>=5.4.0, Qwen3.5 Moe from_pretrained error","Jintao-Huang",2,"","",0,"2026-04-08","2026-04-09"],[0,45308,"Feature request: Support evaluation every N epochs in TrainingArguments","varuna-km-18267",2,"","",0,"2026-04-08","2026-07-07"],[0,45307,"Bug, Generation","Regata3010",2,"","",0,"2026-04-08","2026-04-10"],[0,45306,"Remove import * usage","yueyinqiu",2,"","",0,"2026-04-08","2026-05-16"],[0,45305,"Gradients not averaged by GAS when using DeepSpeed + model_accepts_loss_kwargs=True (Qwen3, Llama3, etc.)","florian6973",1,"florian6973","#45349",0,"2026-04-07","2026-04-13"],[0,45304,"NexusQuant: training-free KV cache compression (10-33x) via DynamicCache hooks","jagmarques",2,"","",0,"2026-04-07","2026-05-16"],[0,45295,"Support Sequence Classification for Gemma 4 Models","jesperschlegel",2,"","",0,"2026-04-07","2026-04-15"],[0,45292,"resize_token_embeddings does not effect to output_embeddings","KoichiYasuoka",2,"","",0,"2026-04-07","2026-04-25"],[0,45290,"`apply_chat_template(tokenize=True)` crashes on assistant messages with tool calls and no content","qgallouedec",1,"qgallouedec","#45348",1,"2026-04-07","2026-04-13"],[0,45278,"Many import errors after update from 4.57.0 to 5.5.0","marchcat69",2,"","",0,"2026-04-07","2026-05-15"],[0,45276,"[gemma4] resize_token_embeddings does not effect to embed_tokens_per_layer or output_embeddings","KoichiYasuoka",0,"zucchini-nlp","#45324",1,"2026-04-07","2026-04-15"],[0,45265,"More permissive config parsing and validation","vadimkantorov",0,"zucchini-nlp","#45289",1,"2026-04-06","2026-04-13"],[0,45259,"How can I use Gemma4's variable image resolution feature?","ejlee95",2,"","",0,"2026-04-06","2026-04-14"],[0,45250,"Flash Attention 2.0","nicholasdudek",2,"","",0,"2026-04-05","2026-04-07"],[0,45246,"[Security/Feature] Deterministic substrate made modeling_utils.py stateful without modifying source — CJPI 100","SweetKenneth",2,"","",0,"2026-04-05","2026-04-08"],[0,45245,"RuntimeError: number of categories cannot exceed 2^24","Hzzone",2,"","",0,"2026-04-05","2026-05-08"],[0,45242,"[Gemma 4] `use_cache=False` corrupts attention computation, producing garbage logits","siwooy",0,"RyanMullins","#45328",0,"2026-04-04","2026-04-09"],[0,45239,"🚨 QA Observer Agent: Real-Time Architecture & Security Pattern Watcher (SCAFFOLD-WATCH)","Insider77Circle",2,"","",0,"2026-04-04","2026-04-08"],[0,45237,"GPT-OSS-20B not work in AMD GPUs","tanreinama",2,"","",0,"2026-04-04","2026-06-02"],[0,45231,"My own task or dataset (give details below)","kerrrang9214-tech",2,"","",0,"2026-04-03","2026-04-08"],[0,45230,"Bug report","kerrrang9214-tech",2,"","",0,"2026-04-03","2026-04-03"],[0,45229,"Gemma4 31B-IT Multi-GPU inference CUDA OOM","vaibhavBh-0",2,"","",0,"2026-04-03","2026-04-23"],[0,45216,"[Regression] Qwen3.5 saved checkpoint is not correct with `save_pretrained` API since version 5.4.0","xin3he",0,"Cyrilvallez","#45340",1,"2026-04-03","2026-04-09"],[0,45208,"[Qwen3MoE] Potentially a bug on `Qwen3MoeSparseMoeBlock`","KbKuuhaku",0,"RudrenduPaul","#45352",0,"2026-04-03","2026-04-13"],[0,45206,"Gemma4: PLE (Per-Layer Embeddings) implementation is underdocumented and config is misleading","w4nderlust",1,"w4nderlust","#45207",0,"2026-04-03","2026-04-14"],[0,45205,"Gemma4: chat_template missing from tokenizer_config.json, requires manual loading from separate file","w4nderlust",2,"","",0,"2026-04-03","2026-06-10"],[0,45203,"Add PolarQuant quantization: Hadamard-rotated Lloyd-Max optimal weights + KV cache","caiovicentino",2,"","",0,"2026-04-03","2026-06-25"],[0,45201,"[Gemma 4] Support per-layer FlashAttention: FA2 for sliding layers, SDPA for global layers","samuelazran",2,"","",0,"2026-04-02","2026-05-12"],[0,45200,"[Gemma 4] mm_token_type_ids required for text-only fine-tuning - should default to zeros","dentity007",0,"zucchini-nlp","#45454",1,"2026-04-02","2026-04-22"],[0,45198,"[BUG] Wav2Vec2 wav2vec2-lv-60-espeak-cv-ft: save_pretrained and tokenization fail","harshaljanjani",1,"harshaljanjani","#45199",0,"2026-04-02","2026-04-14"],[0,45183,"[Bug] XOR logic for `input_ids`/`inputs_embeds` validation produces wrong or misleading error messages across ","XanxusCrypto",2,"","",0,"2026-04-02","2026-04-02"],[0,45182,"🔒 Track: Pin GitHub Actions to commit SHAs","paulinebm",1,"paulinebm","#45180",1,"2026-04-02","2026-04-02"],[0,45177,"add DeepSeek-OCR2","thisisiron",2,"","",0,"2026-04-02","2026-04-30"],[0,45175,"Feature request: Add EfficientViT-SAM (efficientvitsam) to Transformers","masoudpz",4,"","",0,"2026-04-02",""],[0,45162,"Document limitations of `PreTrainedModel._can_set_*` source inspection logic","ElgoogUdiab",2,"","",0,"2026-04-01","2026-05-12"],[0,45161,"Only TP not working with GPT-OSS MoE model","SharvariMedhe",2,"","",0,"2026-04-01","2026-06-06"],[0,45160,"Add AEO quality badge to README","digitamaz",2,"","",0,"2026-04-01","2026-05-10"],[0,45151,"[Energy] N6 Arithmetic: 50-70% AI Training/Inference Energy Reduction — 17 Techniques with Code","dancinlife",2,"","",0,"2026-03-31","2026-04-02"],[0,45146,"Allow for \"pure\" linear attention based Qwen3.5 models","HallerPatrick",2,"","",0,"2026-03-31","2026-04-02"],[0,45145,"[Energy] N6 Arithmetic: 50-70% AI Training/Inference Energy Reduction — 17 Techniques with Code","dancinlife",2,"","",0,"2026-03-31","2026-04-02"],[0,45141,"[refactor] gpt-oss `eager_attention_forward` for modularity (Ex: models with dual eager attn: sink/no sink)","casinca",2,"","",0,"2026-03-31","2026-04-02"],[0,45137,"IndexError: pop from an empty deque with DeepSpeed ZeRO3","albertvillanova",0,"ArthurZucker","52f2268",1,"2026-03-31","2026-04-13"],[0,45127,"[Bug] Model collapse after merging LoRA with extended vocabulary on models with tie_word_embeddings=True (e.g.","YangNobody12",0,"Cursx","#3126",0,"2026-03-30","2026-04-09"],[0,45125,"Qwen3_5MoeForConditionalGeneration missing _tp_plan for tensor parallelism","danielquintas8",1,"danielquintas8","#45124",0,"2026-03-30","2026-04-02"],[0,45120,"Double softmax in MoE router load-balancing loss (mixtral, qwen2_moe, qwen3_vl_moe families)","ionut-anghelina",2,"","",0,"2026-03-30","2026-04-13"],[0,45106,"Reporting a RCE vulnerability","Vancir",2,"","",0,"2026-03-30","2026-03-30"],[0,45103,"[auto_docstring] _process_kwargs_parameters crashes with AttributeError when   module uses from __future__ imp","rpathade",2,"","",0,"2026-03-29","2026-06-10"],[0,45102,"[Research] Fundamental Equation of Consciousness: Ψ = argmax H(p) s.t. Φ > Φ_min","dancinlife",2,"","",0,"2026-03-29","2026-03-30"],[0,45099,"add HyperCLOVA X SEED Vision Instruct 3B","bigshanedogg",4,"","",0,"2026-03-29",""],[0,45095,"transformers 4.30.0 incompatible with rust","starsgo",2,"","",0,"2026-03-29","2026-03-30"],[0,45093,"AutoConfig.register() ignored when trust_remote_code=True and auto_map is present","HanFa",1,"HanFa","#45094",0,"2026-03-29","2026-03-31"],[0,45092,"[Bug] Old InternVL2 remote-code checkpoints are incompatible with Transformers v5 meta initialization","baonudesifeizhai",2,"","",0,"2026-03-29","2026-05-06"],[0,45084,"TypeError: Can't compile non template nodes","theobarrague",0,"Rocketknight1","#45117",1,"2026-03-28","2026-04-10"],[0,45083,"Unexpected behaviour of helper function `_get_feat_extract_output_lengths` in qwen3_omni_moe","CYQFWang",2,"","",0,"2026-03-28","2026-05-06"],[0,45081,"_patch_mistral_regex crashes with AttributeError: 'tokenizers.Tokenizer' object has no attribute 'backend_toke","kruthtom0",2,"","",0,"2026-03-28","2026-05-06"],[0,45072,"[BUG][CI] SwitchTransformers and TimmWrapperModel dtype mismatches in bfloat16 inference","harshaljanjani",1,"harshaljanjani","#45074",0,"2026-03-27","2026-04-02"],[0,45071,"v5.4.0 breaks `PretrainedConfig` type checking","fynnsu",0,"shhKnight30","#45240",0,"2026-03-27","2026-04-10"],[0,45070,"v5.4.0 breaks `PretrainedConfig` field in pydantic model","fynnsu",2,"","",0,"2026-03-27","2026-05-06"],[0,45068,"TypeError in rope validation: set -= list when config loaded from JSON","Fr0do",1,"Fr0do","#45069",0,"2026-03-27","2026-03-30"],[0,45059,"SAM3 PCS very weird behaviour when providing text and bounding boxes","alex-bene",0,"Kash6","#45171",0,"2026-03-27","2026-04-15"],[0,45042,"PIL backend image processors incorrectly require torchvision in v5.4.0","hysts",0,"Lidang-Jiang","#45045",0,"2026-03-27","2026-03-30"],[0,45030,"tiny-random glm4v configuration can't load due to config validation changes","tomaarsen",0,"zucchini-nlp","#45049",1,"2026-03-26","2026-03-27"],[0,45027,"Support Voxtral-4B-TTS-2603 on transformers library","MohamedAliRashad",4,"","",0,"2026-03-26",""],[0,45020,"Recent transformers versions break models using `remote_code`","fxmarty-amd",2,"","",0,"2026-03-26","2026-05-16"],[0,45008,"Add Mamba-3 model support","Anri-Lombard",2,"","",0,"2026-03-26","2026-05-03"],[0,45005,"[v5] Issues with tied weights on translation models in v5","orthorhombic",2,"","",0,"2026-03-25","2026-04-20"],[0,45003,"modeling_utils unsafely accesses sys.modules[]","cjkindel",2,"","",0,"2026-03-25","2026-05-04"],[0,44998,"Unemployment","advikmrai",2,"","",0,"2026-03-25","2026-03-26"],[0,44995,"[Bug] GlmMoeDsa crashes on second forward pass — stale indexer cache","Butanium",2,"","",0,"2026-03-25","2026-04-27"],[0,44993,"Inconsistent tokenization and BLEU scores between AutoTokinzer and NllbTokenizerFast","AdrianSteene",0,"itazap","#45680",1,"2026-03-25","2026-04-28"],[0,44991,"transformers >= 5.0.0 fails loading tokenizer for EMBEDDIA/est-roberta","soras",2,"","",0,"2026-03-25","2026-03-30"],[0,44987,"transformers>=5.1.0 fails when loading physical-intelligence/fast","HaronW",2,"","",0,"2026-03-25","2026-05-03"],[0,44982,"why you people continously dropping file?","xalteropsx",2,"","",0,"2026-03-25","2026-03-25"],[0,44977,"Qwen3.5 cannot generate normally with flash-attention","yuyijiong",2,"","",0,"2026-03-24","2026-03-25"],[0,44969,"Add optional learnable context tokens for CLIP text prompts","anuj-aj",4,"","",0,"2026-03-24",""],[0,44964,"Cannot load `microsoft/Phi-4-multimodal-instruct` model with latest transformers.","kaixuanliu",2,"","",0,"2026-03-24","2026-04-30"],[0,44963,"Your lazy loading and torchvision dependency in Wav2Vec2/Hubert is breaking PyInstaller builds. Stop forcing t","vinkenzor001-afk",3,"","",0,"2026-03-24","2026-10-25"],[0,44962,"Qwen3VL/Qwen2.5VL VisionAttention breaks torch.compile with flash_attention_2","andylizf",2,"","",0,"2026-03-24","2026-05-01"],[0,44961,"racoon","Cleanskiier27",2,"","",0,"2026-03-24","2026-03-24"],[0,44960,"GLM5","inisis",2,"","",0,"2026-03-24","2026-04-23"],[0,44959,"Add automatic dtype alignment and validation for model inputs and pipelines","sasmita016",2,"","",0,"2026-03-23","2026-03-24"],[0,44957,"Add HyperCLOVA X SEED Think 14B","bigshanedogg",2,"","",0,"2026-03-23","2026-10-20"],[0,44955,"Orgs should be able to agree to 3rd party use agreements so their members can access 3rd party models","asglover",2,"","",0,"2026-03-23","2026-07-19"],[0,44945,"Incorrect LLM output when using pipeline parallelism","tasini21",2,"","",0,"2026-03-23","2026-05-01"],[0,44938,"transformers in Python 3.14 failed to load","resc863",2,"","",0,"2026-03-23","2026-03-24"],[0,44936,"trainer.evaluate() fails after trainer.train()","HenrikEilers",0,"Charly21r","#44949",0,"2026-03-22","2026-04-13"],[0,44933,"Nonexistant import from image_utils","josh-kean",2,"","",0,"2026-03-22","2026-04-30"],[0,44929,"First-class fine-tuning support for Mamba / Mamba-2 SSMs — architecture is production-ready, but the training ","lochanharishwar",4,"","",0,"2026-03-22",""],[0,44928,"[Bug] Catastrophic gradient explosion (NaN) in RLHF with Qwen3.5 due to 3D position_ids forcing SDPA Math fall","ouroborosscr",4,"","",0,"2026-03-22",""],[0,44918,"Unpacking Qwen3.5 input embeddings fails with trl SFT trainer","JakobJBauer",2,"","",0,"2026-03-21","2026-03-23"],[0,44913,"In GPTNeoXConfig, rotary_pct silently reverts to default on reload","ratishsp",0,"Krishnachaitanyakc","#44985",1,"2026-03-21","2026-03-27"],[0,44912,"git-oss-20b will not properly load with MXFP4 quantization and falls back to bf16","jottbecr",0,"javierdejesusda","#44930",0,"2026-03-21","2026-03-24"],[0,44910,"[Bug] Flash Attention crashes with illegal memory access on Qwen3.5 due to 3D position_ids being misinterprete","ouroborosscr",2,"","",0,"2026-03-21","2026-03-24"],[0,44908,"inverse_sqrt scheduler ignores lr_scheduler_kwargs (timescale not passed)","magarwal0205",0,"anshuS1310","#44909",0,"2026-03-21","2026-03-24"],[0,44906,"Remove unnecessary `expand_as` in `get_placeholder_mask` across VLMs","syncdoth",1,"syncdoth","#44907",0,"2026-03-21","2026-06-11"],[0,44898,"[BUG] Perceiver image classification (non-default res) fails even with interpolate_pos_encoding=True","harshaljanjani",1,"harshaljanjani","#44899",0,"2026-03-20","2026-03-25"],[0,44877,"Strict config prevents loading `granite_speech` config","tomaarsen",0,"javierdejesusda","#45019",0,"2026-03-20","2026-03-27"],[0,44871,"[Gemma-3] Inconsistent eos_token_id configuration: tokenizer has single value (1) but model.config has list [1","IvanFan-Van",2,"","",0,"2026-03-20","2026-03-21"],[0,44869,"Whisper word timestamp decode crashes on trailing replacement character at end of decoded token stream","chromatic-descension",0,"itazap","#45435",1,"2026-03-20","2026-04-22"],[0,44868,"[rag-end2end-retriever] Broken Google Drive link for SQuAD dataset and hyperparameters","lmmanriquem",2,"","",0,"2026-03-19","2026-04-20"],[0,44863,"NemotronH implementation can't load NemotronH checkpoints!","elfprince13",2,"","",0,"2026-03-19","2026-04-24"],[0,44861,"_get_tied_weight_keys crashes with AttributeError when _tied_weights_keys is a list","gh-wf",2,"","",0,"2026-03-19","2026-03-20"],[0,44857,"LwDetrImageLoss crashes when using float16 AMP and Cuda","m-matthias",1,"m-matthias","#44886",0,"2026-03-19","2026-03-24"],[0,44855,"IndentationError when importing DebertaV2Model on Python 3.13 - @torch.jit.script fails to parse function with","MNIKIEMA",0,"Krishnachaitanyakc","#44986",1,"2026-03-19","2026-03-25"],[0,44849,"Transformers Qwen3.5 had a bug when set output_hidden_states=True","lucasjinreal",2,"","",0,"2026-03-19","2026-04-27"],[0,44843,"AutoTokenizer.from_pretrained calls model_info() unconditionally in _patch_mistral_regex, breaks HF_HUB_OFFLIN","nv-yna",2,"","",0,"2026-03-19","2026-04-29"],[0,44841,"Processor fails for mistralai/Voxtral-Mini-3B-2507","BhavyaShah1234",2,"","",0,"2026-03-19","2026-03-24"],[0,44840,"typo in code block in docs","zhulinchng",2,"","",0,"2026-03-19","2026-03-19"],[0,44829,"AutoModelForSequenceClassification with attn_implementation=\"flash_attention_3\" causes degenerate training (lo","Jantory",2,"","",0,"2026-03-18","2026-04-26"],[0,44821,"Unable to load `AutoImageProcessor` from URL","BSchilperoort",1,"BSchilperoort","#44946",0,"2026-03-18","2026-03-23"],[0,44811,"Whisper processor.batch_decode() function ignoring skip_special_tokens params","cfasana",2,"","",0,"2026-03-18","2026-03-20"],[0,44810,"Showcase / question: a board-proven offline language runtime on ESP32-C3, and whether some future language cap","Juemian-Deng",2,"","",0,"2026-03-18","2026-03-18"],[0,44805,"IndexError: The shape of the mask [...] at index 0 does not match the shape of the indexed tensor [...] at ind","akowalsk",2,"","",0,"2026-03-18","2026-03-18"],[0,44792,"Failed test case `test_model_generate_images` for janus model","kaixuanliu",2,"","",0,"2026-03-17","2026-04-17"],[0,44779,"Deepseek tokenizer produces incorrect results as of v5 (works in v4)","xenova",0,"itazap","#44801",1,"2026-03-17","2026-03-18"],[0,44756,"Disable mmap on Strix Halo to avoid OOM","woct0rdho",2,"","",0,"2026-03-16","2026-06-03"],[0,44754,"I really need help regarding to meta device issue","Hany138078",2,"","",0,"2026-03-16","2026-04-26"],[0,44749,"Transformer 从4.57.3 升级到5.3.0 后过滤数据时长变慢十倍以上","chenjiaoAngel",2,"","",0,"2026-03-16","2026-04-26"],[0,44748,"[Neuron] Auto-select StaticCache when device is Neuron","dacorvo",2,"","",0,"2026-03-16","2026-05-20"],[0,44746,"最理想的Transformes二代生态环境的设想，100B的推理引擎模型会成为未来标准。","QuickerStudio",2,"","",0,"2026-03-16","2026-03-18"],[0,44744,"推理模型的缺陷","QuickerStudio",2,"","",0,"2026-03-16","2026-03-18"],[0,44743,"transformers modular_qwen3_5.py: Recurrent states always reset when using cache and seq_len>1","yansongzhou",2,"","",0,"2026-03-16","2026-03-16"],[0,44742,"[Neuron] Compilation-friendly generation loop","dacorvo",1,"dacorvo","#46590",1,"2026-03-16","2026-06-15"],[0,44741,"[Neuron] Improve transformers compatibility with AWS Neuron devices","dacorvo",2,"","",0,"2026-03-16","2026-04-17"],[0,44740,"Run model linter on new models additions","tarekziade",2,"","",0,"2026-03-16","2026-04-26"],[0,44737,"XLNet: relative_positional_encoding computes on CPU every forward pass (missing device= in torch.arange)","mvstrauss",0,"JiwaniZakir","#44782",0,"2026-03-16","2026-03-19"],[0,44734,"`transformers serve`: /v1/responses crashes on KV cache continuation due to wrong tensor indexing","mango766",1,"mango766","#44735",0,"2026-03-16","2026-03-16"],[0,44717,"Support packed sequences for linear attention models (i.e. Qwen3.5)","kirawi",2,"","",0,"2026-03-14","2026-05-26"],[0,44716,"`PixioPatchEmbeddings.forward` supports `interpolate_pos_encoding` but it is not propagated through `PixioEmbe","audioXD",2,"","",0,"2026-03-14","2026-04-28"],[0,44704,"AutoProcessor.from_pretrained not passing all kwargs to cached_file","peacefulotter",0,"he-yufeng","#44710",0,"2026-03-14","2026-03-25"],[0,44701,"Example: Handling imbalanced text classification with F1-score evaluation using Trainer API","MdSaifAli786123",2,"","",0,"2026-03-14","2026-03-18"],[0,44683,"Compiled flex_attention fails on torch >= 2.9","ntenenz",1,"ntenenz","#44684",0,"2026-03-13","2026-03-18"],[0,44679,"Allow kernel modules to declare their preferred mask function","dacorvo",2,"","",0,"2026-03-13","2026-05-01"],[0,44678,"Use index_select instead of fancy indexing in batched_mm_experts_forward","dacorvo",2,"","",0,"2026-03-13","2026-03-19"],[0,44677,"Add base_model_tp_plan to OlmoeConfig","dacorvo",1,"dacorvo","#44668",1,"2026-03-13","2026-03-26"],[0,44671,"CamemBERT produces incorrect masked LM predictions in v5","math-hiyoko",0,"r266-tech","#44931",0,"2026-03-13","2026-03-23"],[0,44661,"`add-new-model-like` fails if model is inside the `TOKENIZER_MAPPING_NAMES`","michalrzak",0,"NielsRogge","#44334",1,"2026-03-13","2026-03-13"],[0,44655,"Unable to save Pipeline objects with save_pretrained.","VishnuHaasan",2,"","",0,"2026-03-13","2026-03-13"],[0,44643,"Qwen3.5 + `flash_attention_2` crashes: 3D M-RoPE position_ids leak to `_is_packed_sequence`","ritwickchaudhry",2,"","",0,"2026-03-13","2026-03-13"],[0,44637,"load_best_model_at_end reloads PEFT adapter weights onto CUDA and can OOM under low remaining GPU memory","DogWala",2,"","",0,"2026-03-12","2026-05-31"],[0,44625,"Qwen3.5 `num_labels` not propagated from core config to text config","tomaarsen",0,"zucchini-nlp","#44664",1,"2026-03-12","2026-05-08"],[0,44623,"[BUG] processor.save_pretrained(...) missing files","Jintao-Huang",2,"","",0,"2026-03-12","2026-03-13"],[0,44619,"Plug model rule in development flow and extend it","tarekziade",2,"","",0,"2026-03-12","2026-04-22"],[0,44617,"Sam3Video: CUDA out of memory","middleknight",2,"","",0,"2026-03-12","2026-04-29"],[0,44610,"[BUG] OmDet-Turbo processor produces 640px inputs but the model expects 224px","harshaljanjani",1,"harshaljanjani","#44611",0,"2026-03-11","2026-03-13"],[0,44609,"StaticSlidingWindowLayer triggers torch.compile/dynamo recompilations every decode step","jazpurTT",2,"","",0,"2026-03-11","2026-03-12"],[0,44593,"Support for sequence-level custom metrics with decoder-only models","l-k-11235",3,"","",0,"2026-03-11","2026-07-31"],[0,44589,"TypeError: couldn't find storage object Float8_e4m3fnStorage","xin3he",0,"Cyrilvallez","#44883",1,"2026-03-11","2026-03-20"],[0,44574,"Title: Clarification and consistency of \"Transformers\" terminology in README","rabbierabbie",2,"","",0,"2026-03-10","2026-03-11"],[0,44573,"<spam>","aiqing20230305-bot",2,"","",0,"2026-03-10","2026-03-11"],[0,44572,"<spam>","aiqing20230305-bot",2,"","",0,"2026-03-10","2026-03-11"],[0,44568,"[BUG] add_special_tokens=True doesn't add BOS/EOS tokens for microsoft/mdeberta-v3-base tokenizer in transform","Abdullahaml1",0,"umbilnm","#44570",0,"2026-03-10","2026-03-24"],[0,44561,"Removal of `is_torch_fx_available` in v5.0 breaks `trust_remote_code` models","janbernloehr",3,"","",0,"2026-03-10","2026-03-25"],[0,44560,"Qwen3-vl-embedding Video Error \"StopIteration\" in transformers 5.3.0","QYQTexas",2,"","",0,"2026-03-10","2026-03-10"],[0,44559,"flash-attn-4 (flash_attn.cute) is not supported by attn_implementation=\"flash_attention_2\"","DimensionSTP",2,"","",0,"2026-03-10","2026-03-11"],[0,44556,"Loading checkpoint trained on v4.57 cannot be reload after upgraded to v5.2 & v5.3","ihungalexhsu",2,"","",0,"2026-03-10","2026-03-11"],[0,44554,"[MPS] Upstream correctness issue in attention when value head dim differs from query","hvaara",2,"","",0,"2026-03-10","2026-09-30"],[0,44545,"Qwen2_5_VLProcessor.apply_chat_template crashes on batched input when padding=False","Anakintano",0,"zucchini-nlp","#44563",1,"2026-03-09","2026-03-25"],[0,44541,"Can not deploy SFT Qwen3.5-9B model","zouYC2021",2,"","",0,"2026-03-09","2026-04-10"],[0,44537,"About check_model_inputs after version 5.2.0, Where is it!?","QYQTexas",2,"","",0,"2026-03-09","2026-03-10"],[0,44534,"Transformers v5 fills non-persistent buffers with junk","umarbutler",2,"","",0,"2026-03-09","2026-03-09"],[0,44530,"[CB] PagedAttentionCache crashes with \"Invalid group type: linear_attention\" on Qwen3.5 models","mxchampagne",2,"","",0,"2026-03-08","2026-03-10"],[0,44521,"apply_chat_template returns all-zero assistant_masks for multimodal inputs","renhouxing",0,"Rocketknight1","#48793",1,"2026-03-08","2026-09-23"],[0,44514,"`Qwen2_5_VLProcessor.apply_chat_template` crashes on batched input when `padding=False`","qgallouedec",0,"zucchini-nlp","#44563",1,"2026-03-07","2026-03-25"],[0,44512,"Docs still mention transformers run command which was removed in v5","math-hiyoko",1,"math-hiyoko","#44513",0,"2026-03-07","2026-03-09"],[0,44509,"Docs still mention text2text-generation / summarization / translation pipeline tasks which were removed in v5","math-hiyoko",1,"math-hiyoko","#44510",0,"2026-03-06","2026-03-09"],[0,44496,"ValueError: Unrecognized model in allenai/Olmo-Hybrid-Instruct-SFT-7B. Should have a `model_type` key in its c","xenova",2,"","",0,"2026-03-06","2026-03-08"],[0,44493,"Many models started showing UNEXPECTED Key with position id","weathon",0,"KartikPawade","#44508",0,"2026-03-06","2026-03-18"],[0,44492,"Typo in Cache strategies","aryan221102",0,"frogNotToad","#44501",0,"2026-03-06","2026-03-06"],[0,44488,"Current version also does not load \"cjvt/sleng-bert\"","AngledLuffa",2,"","",0,"2026-03-06","2026-03-23"],[0,44486,"KubeflowCallback: Native progress reporting for Kubernetes-based Kubeflow training","abhijeet-dhumal",1,"abhijeet-dhumal","#44487",0,"2026-03-06","2026-03-18"],[0,44485,"[Bug/Discusion] GLM-5 RoPE Implementation","Jintao-Huang",2,"","",0,"2026-03-06","2026-04-13"],[0,44484,"Why max_shard_size in PreTrainedModel.save_pretrained() is 50GB?","Silencezong",2,"","",0,"2026-03-06","2026-03-06"],[0,44483,"[Critical  v5.3] /v1/chat/completions would not accept request as usual","zhangwei217245",2,"","",0,"2026-03-06","2026-03-16"],[0,44479,"[`bug`] v5.3.0 video input regression for `qwen2_5_vl`, `qwen3_vl`, `qwen3_5`, and `qwen3_5_moe`","tomaarsen",2,"","",0,"2026-03-05","2026-03-10"],[0,44466,"[v5] Inconsistent serialization of `lm_head.weight` (tied weights?) depending on model device in v5/`main`, wh","fxmarty-amd",0,"Cyrilvallez","#44497",1,"2026-03-05","2026-03-09"],[0,44464,"Chunked generation produces inconsistent outputs when using compiled forward","mrTsjolder",2,"","",0,"2026-03-05","2026-03-06"],[0,44462,"AutoTokenizer ignores tokenizer.json from the repository","apaniukov",2,"","",0,"2026-03-05","2026-03-20"],[0,44458,"Mllama compile failed after new attn mask","jiqing-feng",2,"","",0,"2026-03-05","2026-04-20"],[0,44457,"LORA权重合并保存在本地之后重新加载出来，两个模型输出结果不一致","fish-kong",2,"","",0,"2026-03-05","2026-04-12"],[0,44453,"推荐一个基于 Transformers 的 AI 效率工具库","zhuxunyu",2,"","",0,"2026-03-05","2026-04-08"],[0,44451,"Latest version cannot load \"vesteinn/ScandiBERT\"","AngledLuffa",0,"weiguangli-io","#44452",0,"2026-03-05","2026-03-19"],[0,44450,"Support argumentless loading from Trainer checkpoints","adosar",3,"","",0,"2026-03-05","2026-10-25"],[0,44448,"[BUG] Different output for google/pegasus-cnn_dailymail between Transformers v4 and v5","math-hiyoko",0,"ArthurZucker","#44571",1,"2026-03-04","2026-03-18"],[0,44442,"[BUG] AutoTokenizer fails to load FastSpeech2ConformerTokenizer","harshaljanjani",1,"harshaljanjani","#44443",0,"2026-03-04","2026-03-09"],[0,44423,"[Bug] `transformers serve --continuous-batching` crashes with multimodal models (Qwen3.5) — AttributeError: 's","jw9603",1,"jw9603","#44436",0,"2026-03-04","2026-03-09"],[0,44419,"Urge DOJ & FBI to Protect AI Innovation Even as the U.S. Focuses on the Iran War","NobleResearch",2,"","",0,"2026-03-03","2026-03-04"],[0,44418,"📋 Documentation Enhancement Suggestion","croviatrust",2,"","",0,"2026-03-03","2026-03-04"],[0,44410,"qwen3next: layer 0 missing attn_qkv/attn_gate projections","fcorneli",2,"","",0,"2026-03-03","2026-03-03"],[0,44409,"Suggestion: optional external troubleshooting reference for retrieval-heavy and RAG-style workflows","onestardao",2,"","",0,"2026-03-03","2026-03-03"],[0,44405,"Add AutoModelForSequenceClassification support for Qwen3.5 (Qwen3_5Config)","medhakimbedhief",1,"medhakimbedhief","#44406",0,"2026-03-03","2026-03-04"],[0,44404,"Bring Loss classes back to modeling file and leverage modular","sbucaille",2,"","",0,"2026-03-03","2026-04-10"],[0,44403,"Unnecessary noise when loading a transformer","AngledLuffa",0,"dhruv7477","#45752",0,"2026-03-02","2026-06-01"],[0,44402,"\"rmihaylov/bert-base-bg\" model has pad and unk tokens outside the tokenizer vocab_size","AngledLuffa",2,"","",0,"2026-03-02","2026-03-10"],[0,44393,"Qwen3-VL: Halucination/Error with 2D bounding box output","L-Reichardt",2,"","",0,"2026-03-02","2026-03-03"],[0,44387,"Increased CUDA reserved memory in Transformers 5.x under int4 quantization leads to OOM","tangefly",0,"SunMarc","#44576",1,"2026-03-02","2026-03-16"],[0,44384,"Qwen3.5 model: When data is not padding, an error is reported, indicating that the shape does not match.","Vectorwh",0,"weiguangli-io","#44399",0,"2026-03-02","2026-03-05"],[0,44380,"GPT2 attention scaling config is ignored when using SDPA / FlashAttention backends","Qi-Zhan",2,"","8757098",0,"2026-03-02","2026-03-04"],[0,44373,"Wrong docstring for position_ids","RmZeta2718",2,"","",0,"2026-03-01","2026-04-12"],[0,44371,"<spam>","HansElze",2,"","",0,"2026-03-01","2026-03-02"],[0,44370,"[i18n-<languageCode>] Translating docs to <languageName>","j6n5nwwmx9-cpu",2,"","",0,"2026-03-01","2026-03-02"],[0,44368,"when using ms-swift lora fine-tuning Qwen3.5-27B, each layer emits warning:You should update the config with `","huangy3881",2,"","",0,"2026-03-01","2026-04-08"],[0,44367,"Unauthorized error for non gated models also","Swatikkar",2,"","",0,"2026-03-01","2026-04-08"],[0,44361,"[BUG] MLukeTokenizer fails with AttributeError on tasks","harshaljanjani",1,"harshaljanjani","#44362",0,"2026-02-28","2026-03-02"],[0,44360,"[Bug/Discussion] The DSA indexer lacks a ReLU","yangdsh",0,"ArthurZucker","#44564",1,"2026-02-28","2026-03-19"],[0,44355,"Errors occur when running compiled Python files.","HuaC-Z",2,"","",0,"2026-02-28","2026-04-08"],[0,44351,"cannot import name 'HybridCache' from 'transformers","wade0604",2,"","",0,"2026-02-28","2026-03-02"],[0,44336,"Some ANSI codes are generated by utils/loading_report even when not connected to terminal","foxik",0,"Cyrilvallez","#44544",1,"2026-02-27","2026-03-09"],[0,44327,"decode_spans in QA pipeline crashes with ValueError: kth out of bounds when len(scores_flat) == top_k","jakipradip-patra",0,"mvanhorn","#44584",0,"2026-02-27","2026-03-12"],[0,44322,"AttributeError: 'Qwen3_5Config' object has no attribute 'num_attention_heads'","zzc0430",2,"","",0,"2026-02-27","2026-03-02"],[0,44315,"Liger Kernel is not applied when creating the model with `model_init`","linfeng-du",1,"linfeng-du","#44329",0,"2026-02-27","2026-02-27"],[0,44303,"Less verbose `tqdm` weight loading (`Loading weights:  38% ... Materializing param=....]` log)","fxmarty-amd",0,"Cyrilvallez","#44414",1,"2026-02-26","2026-03-03"],[0,44297,"[BUG] tokenizer.save_pretrained: tokenizer_class in tokenizer_config.json doesn't match the original","Jintao-Huang",2,"","",0,"2026-02-26","2026-04-06"],[0,44295,"An error occurs when reading position_ids after registering it as a buffer.","nursery42",2,"","",0,"2026-02-26","2026-04-06"],[0,44292,"Error running Qwen-3-8B-NVFP4","IKACE",2,"","",0,"2026-02-26","2026-04-05"],[0,44291,"Bug: TypeError when loading model with `init_empty_weights` in transformers >= 5.0.0rc0 due to unexpected `_is","chenyushuo",2,"","",0,"2026-02-26","2026-03-09"],[0,44280,"Add species bias to model card templates and bias documentation","samtuckerdavis",2,"","",0,"2026-02-25","2026-04-05"],[0,44279,"Dependency issue with transformers","Chilliwiddit",2,"","",0,"2026-02-25","2026-02-26"],[0,44276,"Loading kimik2 is taking forever","savitha-suresh",2,"","",0,"2026-02-25","2026-04-08"],[0,44273,"Lazy loading is not working properly","albertvillanova",2,"","",0,"2026-02-25","2026-07-19"],[0,44265,"[BUG] torch.export.export fails for models using torch_compilable_check (Mask2Former, DeformableDetr, etc.)","harshaljanjani",1,"harshaljanjani","#44266",0,"2026-02-24","2026-02-25"],[0,44263,"The torch.split() return values in GlmMoeDsaIndexer","Jintao-Huang",2,"","",0,"2026-02-24","2026-02-24"],[0,44262,"from_pretrained no longer uses mmap for CPU weights in transformers 5.x causing full materialization","nikitas-cerebras",2,"","",0,"2026-02-24","2026-03-05"],[0,44261,"[Bug/Discussion] MLA q_a_layernorm Missing config.rms_norm_eps, Causing 1e-5/1e-6 Precision Error","Jintao-Huang",2,"","",0,"2026-02-24","2026-05-12"],[0,44248,"[Bug] Security Vulnerability","yomananchill",2,"","",0,"2026-02-24","2026-02-24"],[0,44247,"[MPS] Silent correctness issue in bidirectional attention","hvaara",4,"","",0,"2026-02-24",""],[0,44246,"import transformers takes long sometimes","audricschiltknecht",2,"","",0,"2026-02-23","2026-04-05"],[0,44242,"Load balancing loss not added when output_router_logits=False","Matheart",2,"","",0,"2026-02-23","2026-04-12"],[0,44238,"CI: failing slow runs fails to report properly","tarekziade",1,"tarekziade","efdcbc7",1,"2026-02-23","2026-02-24"],[0,44230,"[fp8] qwen3-vl-fp8/qwen3.5 moe fp8 support (infer)","Jintao-Huang",0,"hmellor","#45610",1,"2026-02-23","2026-04-23"],[0,44222,"[Bug] FP8 save_pretrained moe","Jintao-Huang",2,"","",0,"2026-02-23","2026-04-28"],[0,44220,"Issue with _torch_extract_fbank_features()","dmakhervaks",2,"","",0,"2026-02-23","2026-02-23"],[0,44214,"Add sequence classification capabilities to the Granite models","jmriosal",4,"","",0,"2026-02-22",""],[0,44208,"request refund","storescienza-gif",2,"","",0,"2026-02-22","2026-03-24"],[0,44206,"v5.2.0 regression: LasrFeatureExtractor passes unsupported center arg and crashes","ainergiz",1,"ainergiz","#44207",0,"2026-02-21","2026-02-23"],[0,44205,"Adding SAM3-LiteText","SimonZeng7108",0,"NielsRogge","#44320",1,"2026-02-21","2026-04-13"],[0,44190,"Cannot load local dataset with run_image_classification_no_trainer.py","dyecon",0,"gowthamr-tech","#44199",0,"2026-02-21","2026-02-24"],[0,44188,"Diverging attention kernels due to `allow_is_bidirectional_skip` branching on torch.compile","xmfan",2,"","",0,"2026-02-20","2026-04-27"],[0,44186,"[BUG] LayoutLMv2Tokenizer crashes on NER inputs and batched padding/truncation","harshaljanjani",1,"harshaljanjani","#44187",0,"2026-02-20","2026-02-23"],[0,44183,"EU AI Act Compliance Documentation: Risk Classification & Data Governance Guidelines","desiorac",2,"","",0,"2026-02-20","2026-02-20"],[0,44169,"Need an example for FSDP + FP16 training","quic-meetkuma",2,"","",0,"2026-02-20","2026-03-31"],[0,44168,"Feature: EU AI Act risk classification metadata in model cards","desiorac",2,"","",0,"2026-02-20","2026-02-20"],[0,44164,"save/from_pretrained fails to handle extra_state","quic-kyunggeu",2,"","",0,"2026-02-19","2026-04-09"],[0,44162,"ESM2 is broken, impacting 1000s of scientists workflows","lhallee",2,"","",0,"2026-02-19","2026-02-20"],[0,44155,"[AudioFlamingo3] Batched inference produces incorrect results due to embedding/token leak between tracks","IvanBarabanau",2,"","",0,"2026-02-19","2026-03-25"],[0,44153,"[Bug] Glm46VImageProcessorFast.get_number_of_image_patches() ignores self.size, uses hardcoded longest_edge","SteadfastAsArt",3,"","",0,"2026-02-19","2026-02-19"],[0,44121,"[Model Request] Add OpenAI Weight-Sparse Transformer (circuit-sparsity / circuitgpt)","dtiourine",3,"","",0,"2026-02-18","2026-08-27"],[0,44117,"TOKENIZER_MAPPING_NAMES sometimes returns None, but from_pretrained assumes otherwise","DavidMChan",2,"","",0,"2026-02-17","2026-02-18"],[0,44112,"[BUG][CI] Stale device override test in GraniteSpeech fails on CI","harshaljanjani",1,"harshaljanjani","#44113",0,"2026-02-17","2026-02-19"],[0,44079,"`ModelOutput` keys aren't correctly assigned if key was previously None","tomaarsen",1,"tomaarsen","#44080",1,"2026-02-17","2026-02-20"],[0,44077,"`patchtsmixer` has optional `post_init`, should no longer be allowed","tomaarsen",0,"Cyrilvallez","#44082",1,"2026-02-17","2026-02-17"],[0,44075,"Optimizer SGD args are not used","varunakathirvel3886",0,"nightcityblade","#44203",0,"2026-02-17","2026-02-25"],[0,44062,"TypeError: tokenizers.AddedToken() got multiple values for keyword argument 'special'","umarbutler",2,"","",0,"2026-02-16","2026-03-02"],[0,44060,"Qwen3-Next: Incorrect tied weights warning ties embed_tokens.weight to linear_attn.dt_bias across all layers","Shanay-Mehta",0,"Cyrilvallez","#44095",1,"2026-02-16","2026-02-17"],[0,44052,"Fix skipped tests for glm_moe_dsa model","ArthurZucker",0,"Rocketknight1","#44097",1,"2026-02-16","2026-02-17"],[0,44038,"[bug] transformers 5.0 & Qwen3-VL-Moe","Jintao-Huang",2,"","",0,"2026-02-16","2026-02-17"],[0,44031,"All tokenizers raise incorrect regex pattern warning after version 4.57.3?","jue-jue-zi",2,"","",0,"2026-02-16","2026-03-18"],[0,44016,"Syntax error in Transformer section 3 (Transformers, what can they do?) notebook","khyatimirani",0,"Rocketknight1","#44182",1,"2026-02-15","2026-02-24"],[0,44008,"[Gemma 3n][modular] AttributeError: 'Tensor' object has no attribute 'audio_mel_mask' — variable name collisio","mfielding92",0,"zucchini-nlp","#44040",1,"2026-02-15","2026-02-19"],[0,43994,"google/siglip2-base-patch16-224 produces nonsensical results with AutoModel and pipeline","Adefey",2,"","",0,"2026-02-14","2026-02-17"],[0,43992,"UMT5Encoder.from_pretrained misses `embed_tokens.weight`","nanlliu",2,"","",0,"2026-02-14","2026-03-16"],[0,43990,"Model loading got changed now and 3 weeks before with AutoModelForCausalLM, AutoTokenizer","alokesh17",2,"","",0,"2026-02-13","2026-02-15"],[0,43986,"Confusing crash when loading a video model through AutoProcessor without torchvision installed","Frojdholm",0,"zucchini-nlp","#44125",1,"2026-02-13","2026-02-20"],[0,43979,"Call to contributions: refactor output tracing in transformers","molbap",0,"Aki-07","#43983",0,"2026-02-13","2026-02-18"],[0,43976,"Transformers 5.1.0 does not work with Python3.9+ but Python3.10+","GeoMariusj",0,"mariam851","#44009",0,"2026-02-13","2026-02-16"],[0,43975,"`deepseek-ai/deepseek-coder-6.7b-instruct` incorrectly detokenizes in v5","pavel-esir",2,"","",0,"2026-02-13","2026-02-25"],[0,43957,"model loading with torch.device(\"meta\") breaks some models on transformers 5.+ , e.g. TRELLIS.2 ,  RMBG-2.0","xvdp",2,"","",0,"2026-02-12","2026-02-16"],[0,43950,"`from_pretrained()` silently corrupts non-persistent buffers (`register_buffer(persistent=False)`) -- transfor","adrienB134",2,"","",0,"2026-02-12","2026-02-12"],[0,43940,"Qwen3-Next: DeepSpeed ZeRO-3 fails to load weights (all params MISSING)","Shanay-Mehta",2,"","",0,"2026-02-12","2026-02-17"],[0,43939,"Better regex in `build_glob_alternation` method","ariG23498",2,"","",0,"2026-02-12","2026-03-23"],[0,43937,"[GLM-5] ValueError: GenerationConfig is invalid","xin3he",2,"","",0,"2026-02-12","2026-02-23"],[0,43935,"Add `eval_on_end` flag (analogous to `eval_on_start`)","MarkusSpanring",0,"khushali9","#44146",0,"2026-02-12","2026-03-26"],[0,43931,"Model loading error: weight shapes mismatch of Qwen3-VL-30B-A3B-Instruct","jonbakerfish",2,"","",0,"2026-02-12","2026-02-16"],[0,43927,"[BUG] DiaConfig loses custom token IDs after save / load and causes IndexError during generation","harshaljanjani",1,"harshaljanjani","#43928",0,"2026-02-11","2026-02-13"],[0,43909,"Add LFM2.5 Audio 1.5B","MHRDYN7",4,"","",0,"2026-02-11",""],[0,43908,"What is the process of adding a new hardware backend for Trainer?","quic-meetkuma",2,"","",0,"2026-02-11","2026-03-23"],[0,43906,"Isolated reproduction of https://github.com/huggingface/transformers/issues/38071","willxxy",0,"ManasVardhan","#44021",0,"2026-02-11","2026-02-17"],[0,43901,"TextClassificationPipeline docs still mention return_all_scores, but behavior differs","math-hiyoko",1,"math-hiyoko","#43903",0,"2026-02-11","2026-02-11"],[0,43899,"`sync_each_batch` has no effect when using FSDP","ojh31",0,"SunMarc","#43919",1,"2026-02-10","2026-02-13"],[0,43885,"[Refactor] Confusing naming: input_embeds vs inputs_embeds","JiangJiaWei1103",2,"","",0,"2026-02-10","2026-02-19"],[0,43883,"AttributeError: 'MolmoForCausalLM' object has no attribute 'all_tied_weights_keys'","Kaihui-intel",2,"","",0,"2026-02-10","2026-02-16"],[0,43881,"glm-4v-9b loading failed","Kaihui-intel",2,"","",0,"2026-02-10","2026-02-23"],[0,43878,"Save + loading from pretrained raises: out_features is not supported by TimmBackbone. Please use out_indices i","jveitchmichaelis",0,"zucchini-nlp","#43886",1,"2026-02-10","2026-02-12"],[0,43874,"[GLM46V] Glm46VImageProcessorFast missing get_number_of_image_patches breaks _get_num_multimodal_tokens (Attri","baonudesifeizhai",0,"weiguangli-io","#43877",0,"2026-02-10","2026-02-11"],[0,43873,"Offloading not working as expected with quantization","edmcman",1,"edmcman","#4325",0,"2026-02-09","2026-09-25"],[0,43872,"bitsandbytes incompatibility: TypeError: Int8Params.__new__() got an unexpected keyword argument '_is_hf_initi","edmcman",2,"","",0,"2026-02-09","2026-03-18"],[0,43867,"load model error when state_dict sorted","nursery42",0,"shtse8","#43966",0,"2026-02-09","2026-02-13"],[0,43866,"Ovis2 1B checkpoint corrupted","EnricoBeltramo",2,"","",0,"2026-02-09","2026-04-02"],[0,43864,"GlmMoeDsaConfig: mlp_layer_types default overwritten by inlined parent init","joninco",0,"weiguangli-io","#43876",0,"2026-02-09","2026-02-10"],[0,43859,"huggingface ","Pediboi666",2,"","",0,"2026-02-09","2026-02-09"],[0,43856,"Inefficient memory usage during Qwen3 MoE training","yurkoff-mv",2,"","",0,"2026-02-09","2026-03-23"],[0,43854,"Unable to load `zai-org/GLM-4.7-Flash` model correctly in the unit tests","kaixuanliu",0,"ArthurZucker","#43893",1,"2026-02-09","2026-02-11"],[0,43846,"huggingface ","Pediboi666",2,"","",0,"2026-02-08","2026-02-09"],[0,43845,"huggingface ","Pediboi666",2,"","",0,"2026-02-08","2026-02-09"],[0,43844,"Gradient abnormally increases when training a randomly initialized model with HfDeepSpeedConfig + ZeRO-3","wxhcore",2,"","",0,"2026-02-08","2026-03-19"],[0,43837,"Proposal to add Qwen3-ASR support","mbtariq82",1,"mbtariq82","#43838",0,"2026-02-08","2026-06-26"],[0,43835,"Fett","Pediboi666",2,"","",0,"2026-02-08","2026-02-09"],[0,43834,"[i18n-<languageCode>] Translating docs to <languageName>","Pediboi666",2,"","",0,"2026-02-08","2026-02-09"],[0,43828,"With torch.autocast, Phi-tiny-MoE-instruct raises an dtype mismatch error","vinnamkim",0,"Mr-Neutr0n","#43839",0,"2026-02-08","2026-02-11"],[0,43827,"Summarization/Translation docs still reference pipeline() after v5 pipeline removals","math-hiyoko",0,"Mr-Neutr0n","018af00",0,"2026-02-08","2026-02-09"],[0,43825,"pipeline() error message incorrectly suggests translation tasks are supported in v5","math-hiyoko",2,"","",0,"2026-02-08","2026-03-10"],[0,43824,"ImportError: cannot import name 'Qwen2_5_VLForConditionalGeneration' from 'transformers'","waltwalt36",2,"","",0,"2026-02-08","2026-03-23"],[0,43819,"[BUG] DAC.from_latents does not match the forward pass with missing STE","harshaljanjani",1,"harshaljanjani","#43820",0,"2026-02-07","2026-02-10"],[0,43818,"[Video-LLaVA] `Video-LLaVA-7B-hf` video_tower is missing temporal attention AND shares nearly identical weight","jong980812",2,"","",0,"2026-02-07","2026-03-25"],[0,43813,"Typo - should be \"orig_conversion.quantization_operation\"","ishaan-shivhare",2,"","",0,"2026-02-07","2026-03-17"],[0,43805,"chore(test): add a `set_seed` pytest fixture","tarekziade",2,"","",0,"2026-02-06","2026-03-03"],[0,43792,"openai/whisper-large-v2 can't run","LIANGQI0811",2,"","",0,"2026-02-06","2026-02-09"],[0,43784,"NameError: name 'nn' is not defined when importing sentence-transformers with latest transformers","Alan-Jowett",2,"","",0,"2026-02-06","2026-03-09"],[0,43782,"Qwen3VLForConditionalGeneration.from_pretrained weight_only = True error","oscars17",2,"","",0,"2026-02-05","2026-03-29"],[0,43761,"[v5 regression] CLIPVisionModel.forward returns hidden_states=None even when output_hidden_states=True","yukiu00",2,"","",0,"2026-02-05","2026-02-06"],[0,43756,"Smollm3 drops 3/4 RoPE layers while 1/4 is intended by whats claimed in the blog post.","tobiaskatsch",2,"","",0,"2026-02-05","2026-02-05"],[0,43749,"FSDP_CPU_RAM_EFFICIENT_LOADING broken","kmod",2,"","",0,"2026-02-05","2026-03-10"],[0,43746,"[GraniteSpeechForConditionalGeneration] Models with PEFT adapters won't load from local checkpoints (from_pret","gabe-l-hart",1,"gabe-l-hart","#43781",0,"2026-02-04","2026-03-05"],[0,43742,"Key error when loading facebook/MobileLLM-125M","ghost",2,"","",0,"2026-02-04","2026-03-15"],[0,43726,"Help: Transformer v5 managed to break HF models","TomLucidor",2,"","",0,"2026-02-04","2026-03-06"],[0,43725,"Quantization model behavior changed","jiqing-feng",2,"","",0,"2026-02-04","2026-03-15"],[0,43724,"[integration] add pluto experiment tracker callback.","asaiacai",3,"","",0,"2026-02-04","2026-09-24"],[0,43723,"Issue loading tokenizer using AutoTokenizer.from_pretrained in v5","msubrayada",2,"","",0,"2026-02-04","2026-02-04"],[0,43720,"[BUG][CI] BitNet AutoBitLinear fails when packed weights aren’t unpacked during accelerate loading","harshaljanjani",1,"harshaljanjani","#43721",0,"2026-02-03","2026-02-16"],[0,43717,"init_weights usage in Mamba and Mamba-2","kevinli573",2,"","",0,"2026-02-03","2026-03-15"],[0,43716,"Mistral-3: dtype mismatch between image preprocessor and model","egallouis",0,"zucchini-nlp","#43731",1,"2026-02-03","2026-02-04"],[0,43708,"Trainer `resume_from_checkpoint` incorrectly calculates `max_steps` when changing `per_device_train_batch_size","KHPan",0,"lordaarush","#43770",0,"2026-02-03","2026-02-06"],[0,43704,"Qwen3ForCausalLM leaks VRAM if used in multiple dataloader threads","dxqb",2,"","",0,"2026-02-03","2026-02-03"],[0,43701,"resume_from_checkpoint key mismatch","zenosai",2,"","",0,"2026-02-03","2026-03-27"],[0,43700,"Lack of FastTokenizer for Internlm/Intern-S1","jiosephlee",2,"","",0,"2026-02-03","2026-04-02"],[0,43698,"SwanLab integration uses outdated swanlab.init() signature","TianjiaoTsao",0,"i-pj","#43719",0,"2026-02-03","2026-02-09"],[0,43697,"RTDetrV2ForObjectDetection produces different outputs in Transformers v5 with identical inputs","MorganFujimaka",2,"","",0,"2026-02-03","2026-03-05"],[0,43696,"GPT-oss-20b: torch.OutOfMemoryError: CUDA out of memory.","Decadz",2,"","",0,"2026-02-03","2026-03-13"],[0,43688,"Incorrect normalization of auxiliary loss in OLMoE and GPT Oss","andresnowak",2,"","",0,"2026-02-02","2026-03-13"],[0,43684,"Add Qwen3-Omni model registration to AutoModel and AutoModelForConditionalGeneration classes","samudraneel05",2,"","",0,"2026-02-02","2026-02-04"],[0,43682,"`torch.compile` and `torch.export` helper function to be moved from `import_utils`","IlyasMoutawwakil",4,"","",0,"2026-02-02",""],[0,43680,"Add Nvidia NitroGen to huggingface transformers","AffanBinFaisal",4,"","",0,"2026-02-02",""],[0,43676,"`test_apply_chat_template_video_frame_sampling` fails with `num_frames` and `fps` are mutually exclusive","tarekziade",2,"","",0,"2026-02-02","2026-02-02"],[0,43673,"GenerationMixin cache missing in v5.0.0 during chunked_prefill","SmerkyG",0,"zucchini-nlp","#43679",1,"2026-02-02","2026-02-04"],[0,43671,"Proposal to add Qwen3-TTS support","ShahVandit",4,"","",0,"2026-02-02",""],[0,43668,"ModernBERTConfig `norm_eps` type hint is incorrect","fschlatt",1,"fschlatt","#43669",0,"2026-02-01","2026-02-03"],[0,43653,"[BUG][CI] BigBirdTokenizer mask token not registered as special token, gives empty decode output","harshaljanjani",2,"","",0,"2026-01-31","2026-03-11"],[0,43650,"ADD THE DATA","rsiya1986-arch",2,"","",0,"2026-01-31","2026-01-31"],[0,43646,"Transformers 5.0.0 breaks custom model initialization","umarbutler",2,"","",0,"2026-01-31","2026-02-04"],[0,43645,"Transformers 5.0.0 breaks defining and then initializing custom models in Jupyter notebooks","umarbutler",0,"Cyrilvallez","#43690",1,"2026-01-31","2026-02-03"],[0,43644,"Transformers 5.0.0 fills non-persistent buffers with junk","umarbutler",2,"","",0,"2026-01-31","2026-03-02"],[0,43643,"`trust_remote_code=True` in `AutoConfig.from_pretrained` results in missing fields in returned object","Lander-Hatsune",2,"","",0,"2026-01-31","2026-02-02"],[0,43638,"IndexError: index 0 is out of bounds for dimension 0 with size 0 with deepspeed zero3 traininig and a non pret","MichelDucartier",2,"","",0,"2026-01-31","2026-03-11"],[0,43632,"Transformers v5 breaks the `_is_hf_initialized` flag","ZhiyuanChen",2,"","",0,"2026-01-30","2026-02-03"],[0,43630,"Add multilingual text classification examples to docs (Arabic, Chinese, etc.)","salehA13",2,"","",0,"2026-01-30","2026-03-11"],[0,43618,"CLIPOutput attentions is no longer assigned","dbellavista-ai",0,"jonathan-fulton","#43657",0,"2026-01-30","2026-02-03"],[0,43611,"Transformers 5.0.0 breaks loading models with the `base_model_prefix` attribute","umarbutler",2,"","",0,"2026-01-30","2026-02-26"],[0,43606,"[BUG][CI] suno/bark-small model fails with a device mismatch when using CPU offload","harshaljanjani",1,"harshaljanjani","#43607",0,"2026-01-29","2026-01-30"],[0,43604,"Revisit the condition for scaling the loss","qgallouedec",1,"qgallouedec","#49240",1,"2026-01-29","2026-03-20"],[0,43602,"Revisit the condition for calling `compute_metrics`","qgallouedec",2,"","",0,"2026-01-29","2026-04-15"],[0,43601,"Revisit the condition for calling `compute_loss` at eval","qgallouedec",2,"","",0,"2026-01-29","2026-03-09"],[0,43600,"Account for custom trainers when trying to estimate the number of FLOPS","qgallouedec",2,"","",0,"2026-01-29","2026-02-03"],[0,43599,"Use a private `_metrics` dict to allow for additional metric logging","qgallouedec",2,"","",0,"2026-01-29","2026-03-09"],[0,43598,"Revisit `remove_unused_column` in Trainer for better customizability","qgallouedec",2,"","",0,"2026-01-29","2026-03-09"],[0,43597,"Activation offloading for Trainer","qgallouedec",2,"","",0,"2026-01-29","2026-09-22"],[0,43596,"IndexError: index 0 is out of bounds for dimension 0 with size 0 with deepspeed zero3 init and BertModel","MichelDucartier",2,"","",0,"2026-01-29","2026-01-31"],[0,43595,"Tell Us: What Would Make Trainer Better?","SunMarc",4,"","",0,"2026-01-29",""],[0,43582,"caching_allocator_warmup function raise TypeError on AppleSilicon M4 max","theVannu",0,"tobyliu2004","#43608",0,"2026-01-29","2026-02-02"],[0,43577,"model.dtype and model.qformer.dtype remain float32 when loading Blip2 model with dtype=torch.float16 or torch.","rebel-seinpark",2,"","",0,"2026-01-29","2026-01-29"],[0,43576,"`transformers env` command seems to be broken in v5","Hermit-w",2,"","",0,"2026-01-29","2026-04-02"],[0,43575,"Load `Qwen2-57B-A14B-Instruct` with tp lead to OOM","Hermit-w",2,"","",0,"2026-01-29","2026-03-08"],[0,43572,"missing pad_token_idx in StableLmConfig after 5.0 update","mario-aws",0,"zucchini-nlp","#43592",1,"2026-01-29","2026-01-30"],[0,43550,"[BUG] Bamba-9B-v2 model fails with torch.compile when using SDPA","harshaljanjani",2,"","",0,"2026-01-28","2026-02-10"],[0,43541,"RuntimeError: MixtralForCausalLM float32 model errors at grouped_mm op during torch dynamo tracing","vakumar1",2,"","",0,"2026-01-27","2026-02-27"],[0,43540,"ValueError when processing video inputs in Qwen3OmniMoe","BBiering",0,"zucchini-nlp","#43588",1,"2026-01-27","2026-01-29"],[0,43531,"`sliding_window` issue with Qwen3-MoE models","saattrupdan",0,"Vallabh-1504","#43593",0,"2026-01-27","2026-02-04"],[0,43526,"reduce_labels returns only one label instead of the whole array of labels in BeitImageProcessorFast","sbucaille",1,"sbucaille","#43527",0,"2026-01-27","2026-01-27"],[0,43525,"AttributeError: 'Llama4Config' object has no attribute 'pad_token_id'","xin3he",0,"zucchini-nlp","#43592",1,"2026-01-27","2026-01-30"],[0,43522,"Weights are not tied when model is loaded onto the `meta` device","hmellor",0,"Cyrilvallez","#43523",1,"2026-01-27","2026-01-27"],[0,43519,"Incorrect timestamp calculation in Qwen3VL Processor","666zz666",0,"jonathan-fulton","#43659",0,"2026-01-27","2026-02-09"],[0,43508,"Transformers 5.0.0 calls torch.is_autocast_enabled(device_type) breaking compatibility with torch<2.4","oliverholworthy",2,"","",0,"2026-01-26","2026-01-30"],[0,43504,"[BUG] BeitForSemanticSegmentation fails to load pretrained model preset due to a legacy field","harshaljanjani",1,"harshaljanjani","#43505",0,"2026-01-26","2026-02-05"],[0,43502,"API requests are made despite setting local_files_only=True.","haok1402",0,"vaibhav-research","#43603",0,"2026-01-26","2026-04-13"],[0,43493,"SigLIP2 discrepancy between HF implementation and original JAX implementation","nmilosev",2,"","",0,"2026-01-26","2026-03-07"],[0,43489,"Transformers' version 5 is out!","LysandreJik",2,"","",0,"2026-01-26","2026-03-07"],[0,43482,"Qwen2.5-GGUF loading failed with transformers v5","Kaihui-intel",2,"","",0,"2026-01-26","2026-01-26"],[0,43479,"Phi4MultimodalConfig incorrectly initializes default vision/audio configs when passed as None","charlieJ107",1,"charlieJ107","#43480",0,"2026-01-26","2026-01-26"],[0,43475,"[SAM 3 Video] Sam3VisionEncoderOutput object has no attribute 'fpn_position_embeddings'","vydpnguyen",0,"tomaarsen","#43487",1,"2026-01-25","2026-01-26"],[0,43474,"[SAM3 Video] get_text_features returns CLIPTextModelOutput but code expects tensor","ncerny",0,"tomaarsen","#43487",1,"2026-01-25","2026-01-26"],[0,43472,"Introduce standardized BatchLinear module for MoE architectures to facilitate PEFT and Quantization","ITcarrot",4,"","",0,"2026-01-25",""],[0,43454,"[BUG] AyaVisionConfig fails to tie lm_head weights causing garbage text generation","harshaljanjani",1,"harshaljanjani","#43455",0,"2026-01-23","2026-01-24"],[0,43452,"gguf_file breaks for AutoTokenizer.from_pretrained and AutoModelForCausalLM.from_pretrained","xenova",0,"ArthurZucker","#43465",1,"2026-01-23","2026-01-24"],[0,43450,"Video processors return incorrect shape when input is batched","shaform",0,"zucchini-nlp","#43486",1,"2026-01-23","2026-01-30"],[0,43441,"[BUG] Ministral-3 fails with FlashAttention in Transformers v5 RC","dysby",0,"zucchini-nlp","#43443",1,"2026-01-23","2026-01-26"],[0,43431,"Add dataloader_multiprocessing_context to TrainingArguments + Ensure dataloader_prefetch_factor fully supports","muyihao",2,"","",0,"2026-01-23","2026-10-20"],[0,43425,"Torch 2.10 incompatible","qgallouedec",3,"","",0,"2026-01-22","2026-02-06"],[0,43421,"[FEATURE] TokenizersBackend does not update post-processor when special tokens are modified at runtime","harshaljanjani",2,"","",0,"2026-01-22","2026-03-09"],[0,43416,"[Kosmos-2.5] RuntimeError: a leaf Variable that requires grad is being used in an in-place operation during tr","Neuro1729",1,"Neuro1729","#43418",0,"2026-01-22","2026-01-23"],[0,43412,"gemma3n executorch export fails missing self training guard and erfinv not supported","maceip",2,"","",0,"2026-01-22","2026-03-29"],[0,43408,"Warning: You are using a model of type sam3_video to instantiate a model of type sam3_tracker","gboeer",0,"leoneperdigao","#43495",0,"2026-01-22","2026-02-05"],[0,43404,"Bug: lm_head weight not tied in Mistral3ForConditionalGeneration (affects AutoModelForImageTextToText)","aswin00000",2,"","",0,"2026-01-22","2026-01-26"],[0,43393,"Qwen3-VL checkpoints don't have pad_token","Rocketknight1",0,"vaibhav-research","#43398",0,"2026-01-21","2026-01-22"],[0,43388,"gather_for_metrics incorrectly drops label elements in the last batch when labels is a tuple with several labe","J-Bracke",2,"","",0,"2026-01-21","2026-03-02"],[0,43386,"Support other types of model inputs for continuous batching","yhshin11",4,"","",0,"2026-01-21",""],[0,43383,"Parakeet model limited to ~8 minutes audio and yet NeMo supports hours-long audio","penguinwang96825",2,"","",0,"2026-01-21","2026-03-04"],[0,43381,"Gradient checkpointing cannot be used in eval mode","smarter",2,"","",0,"2026-01-20","2026-03-01"],[0,43377,"[BUG] MIMI Encoder produces different outputs for batched vs single inputs due to missing padding mask support","harshaljanjani",2,"","",0,"2026-01-20","2026-03-15"],[0,43375,"Feature request: Support resume_from_checkpoint for checkpoints stored on Hugging Face Hub","meodbr",4,"","",0,"2026-01-20",""],[0,43366,"GGUF model with architecture gpt-oss support","akayunov",0,"sirzechs66","#45506",0,"2026-01-20","2026-04-22"],[0,43364,"DataCollatorForLanguageModeling with whole_word_mask=True crashes without offset_mapping","a1nkit",2,"","",0,"2026-01-20","2026-01-20"],[0,43360,"[v5] do_sample false not working ('batched_mm'). transformers 5.0 main branch","Jintao-Huang",0,"raimbekovm","#43362",0,"2026-01-20","2026-01-21"],[0,43355,"WeightRenaming doesn't allow more than one group to be matched","sbucaille",2,"","",0,"2026-01-19","2026-01-22"],[0,43352,"Error: NemotronHForCausalLM does not support Flash Attention 2.0 yet","OrlandoWhite88",2,"","",0,"2026-01-19","2026-02-28"],[0,43350,"[Bug] Fix device mismatch in Ministral3 RoPE forward pass","dysby",2,"","",0,"2026-01-19","2026-01-22"],[0,43347,"[BUG] ESM config validation error with incorrect divisibility check","harshaljanjani",1,"harshaljanjani","#43348",0,"2026-01-19","2026-01-19"],[0,43344,"invalid test cases for glm_image model","kaixuanliu",2,"","",0,"2026-01-19","2026-01-26"],[0,43335,"[BUG] SwitchTransformersConfig creates sparse layer when num_sparse_encoder_layers=0 with single layer model","harshaljanjani",1,"harshaljanjani","#43336",0,"2026-01-17","2026-02-09"],[0,43334,"Qwen3-VL can't be loaded in transformers dev: AttributeError: 'Qwen3VLTextConfig' object has no attribute 'pad","albertvillanova",2,"","",0,"2026-01-17","2026-01-22"],[0,43329,"[BUG] _get_num_multimodal_tokens: video branch uses undefined a) get_number_of_video_patches, b)merge_size. Te","stefgina",2,"","",0,"2026-01-17","2026-02-24"],[0,43328,"Add support for SNAC model type","rtATvw",4,"","",0,"2026-01-16",""],[0,43322,"Segmentation Fault when loading Llava Next Models","omrastogi",2,"","",0,"2026-01-16","2026-02-24"],[0,43317,"device_map=auto fails to load the dequantized model on gpu+cpu offload","IlyasMoutawwakil",0,"Cyrilvallez","#43326",1,"2026-01-16","2026-01-26"],[0,43316,"API discrepancy between `Gemma3TextConfig` and others","Tcc0403",1,"Tcc0403","#43320",0,"2026-01-16","2026-01-26"],[0,43315,"Qwen3-VL-Moe implementation is not sparse","Maxwell-Lyu",3,"","",0,"2026-01-16","2026-01-16"],[0,43308,"Qwen2-VL and Qwen2.5-VL have no _name_or_path in transformers v4.57.5","albertvillanova",2,"","",0,"2026-01-15","2026-01-16"],[0,43299,"Transformer version: 5.0.0.dev0 breaks Qwen3VL Moe models loading","daulettoibazar",0,"Cyrilvallez","#43307",1,"2026-01-15","2026-01-16"],[0,43298,"continous batching not support audio Model","MengLeebin",4,"","",0,"2026-01-15",""],[0,43296,"Failed to load PaddleOCR-VL model with transformers 4.53.0 in vLLM 0.11.0","jhlddz",2,"","",0,"2026-01-15","2026-02-10"],[0,43295,"[Regression] v4.57.5 breaks custom model code accessing processor.tokenizer and passing images to tokenizer","AndreasKaratzas",2,"","",0,"2026-01-14","2026-02-23"],[0,43287,"Adding DeepSeek's Engram module","Chenhao-Guan",2,"","",0,"2026-01-14","2026-01-14"],[0,43284,"Customize Quantization-Friendly Backward Compatibility","yiliu30",0,"IlyasMoutawwakil","#43917",1,"2026-01-14","2026-03-02"],[0,43278,"Embedding layer dtype changed from BF16 in training to FP32 in evaluate.","TriLoo",2,"","",0,"2026-01-14","2026-02-23"],[0,43276,"`Llama4VisionRotaryEmbedding` access deprecated config attribute `rope_theta`","Tcc0403",1,"Tcc0403","#43281",0,"2026-01-14","2026-01-22"],[0,43272,"`[Bug][v5.0.0rc2] AutoTokenizer.from_pretrained fails on second run with AttributeError: 'PreTrainedConfig' ob","ansh-info",0,"Rocketknight1","#43321",1,"2026-01-14","2026-01-16"],[0,43262,"Audio processors: `apply_chat_template()` defaults to 16kHz sampling rate, even if the processor config sets a","rmclarke",0,"zucchini-nlp","#43674",1,"2026-01-13","2026-02-02"],[0,43257,"Qwen3 MOE weights not converted when loading with accelerate + deepspeed","edbeeching",0,"kashif","#43303",1,"2026-01-13","2026-01-24"],[0,43240,"kwargs are not passed to loss calculation function.","XinhaoMei",2,"","",0,"2026-01-13","2026-02-21"],[0,43232,"_update_model_kwargs_for_generation after sync_gpus when generation","xin-w8023",2,"","",0,"2026-01-12","2026-03-05"],[0,43224,"DeepSeek-V2-Lite-Chat repetitive generation when `_experts_implementation=\"eager\"`","yiliu30",0,"Rocketknight1","#43288",1,"2026-01-12","2026-01-15"],[0,43215,"Add output_attentions flag to DinoV3","GaspTO",3,"","",0,"2026-01-10","2026-09-23"],[0,43208,"[xLSTM] Three bugs preventing training models smaller than 7B","Anri-Lombard",1,"Anri-Lombard","#43209",0,"2026-01-10","2026-02-09"],[0,43206,"Mamba2ForCausalLM missing get_output_embeddings/set_output_embeddings breaks resize_token_embeddings","Anri-Lombard",1,"Anri-Lombard","#43207",0,"2026-01-10","2026-01-22"],[0,43202,"tokenizer.decode producing bad results in some cases from 5.0.0rc1 to 5.0.0rc2","awni",0,"Anri-Lombard","#43219",0,"2026-01-09","2026-01-22"],[0,43200,"4.57.3 issue with AutoTokenizer.from_pretrained when HF_HUB_OFFLINE=1","g8a9",2,"","",0,"2026-01-09","2026-01-12"],[0,43193,"`torch._grouped_mm` support for `GptOssExperts`","mattteochen",0,"IlyasMoutawwakil","#43227",1,"2026-01-09","2026-01-22"],[0,43190,"SAM 3 missing point prompting","kliyer-ai",2,"","",0,"2026-01-09","2026-01-12"],[0,43186,"Floating-point precision issues in iamge_processing_deepseek_vl.py","frankenliu",0,"raimbekovm","#43187",0,"2026-01-09","2026-01-16"],[0,43183,"Regression in Qwen2_5_VLProcessor video kwargs routing (4.56.2 -> 4.57.1+): videos_kwargs{min_pixels,max_pixel","sidezrw",0,"zucchini-nlp","#43228",1,"2026-01-08","2026-01-13"],[0,43169,"torch.compile of model.forward() or model.generate() not working anymore","jeasinema",2,"","",0,"2026-01-08","2026-01-09"],[0,43163,"[XLA/Neuron] torch.isin() fails with int64 in isin_mps_friendly() on AWS Trainium","naglemi",2,"","",0,"2026-01-08","2026-01-08"],[0,43159,"opt-125m is missing weight for embed and lm_head when use_safetensors=True","xin3he",2,"","",0,"2026-01-08","2026-01-12"],[0,43142,"Symbolic trace failed when using HFTracer with Qwen3Model. AttributeError: 'Tensor' object has no attribute 'g","tongl2",2,"","",0,"2026-01-07","2026-01-08"],[0,43125,"Saving bug using FSDP 2.0 + parallelism_config set (while works fine without providing parallelism_config in T","d5031",0,"vaibhav-research","#43314",0,"2026-01-06","2026-01-24"],[0,43122,"Different tokenization with same tokenizer from 4.57.3 to 5.0","awni",2,"","",0,"2026-01-05","2026-01-26"],[0,43116,"Multi-label classification always returns empty results in run_classification.py example script","ziorufus",2,"","",0,"2026-01-05","2026-02-18"],[0,43110,"4.57.3 and 5.0.0 from main: MistralTokenizer object has no attribute 'convert_tokens_to_ids'","dgouju",0,"adi776borate","#43141",0,"2026-01-05","2026-01-19"],[0,43099,"Question about .T usage when loading video frames from file paths","c1ircle",2,"","",0,"2026-01-04","2026-02-12"],[0,43097,"5.0.0 `tie_embeddings_and_encoder_decoder` removed without indication","AmitMY",2,"","",0,"2026-01-04","2026-02-24"],[0,43093,"summarization pipeline breaks.","jiqing-feng",2,"","",0,"2026-01-04","2026-01-15"],[0,43089,"Generation overhead:  many GPU syncs per token + PyTorch dispatch overhead","AmitMY",2,"","",0,"2026-01-03","2026-02-25"],[0,43086,"Add async_stopping_criteria flag to reduce GPU-CPU synchronization overhead","AmitMY",2,"","",0,"2026-01-03","2026-03-08"],[0,43083,"[Bug] Qwen3OmniMoe Loading fails in transformers v5","HuiyingLi",2,"","",0,"2026-01-03","2026-01-05"],[0,43082,"chunked prefill implementation is incorrect","SmerkyG",0,"marcndo","#43132",0,"2026-01-03","2026-01-09"],[0,43081,"[Feature Request] Support explicitly mapping Model IDs to local paths","Gianthard-cyh",2,"","",0,"2026-01-02","2026-01-02"],[0,43079,"[Bug] qwen2_5_omni: Hardcoded noise_initialization length of '30000' causes shape mismatch","sniper35",2,"","",0,"2026-01-01","2026-01-12"],[0,43075,"FlexAttention backend support for sequence packing","vadimkantorov",2,"","",0,"2025-12-31","2026-08-11"],[0,43074,"Issue generate audio Qwen3-Omni","Hert4",2,"","",0,"2025-12-31","2026-01-12"],[0,43072,"EoMT fast image processor fails to process segmentation maps stored on CUDA","simonreise",1,"simonreise","#43073",0,"2025-12-30","2026-01-06"],[0,43066,"Wrong tokenizer decoder type in Transformers v5","awni",2,"","",0,"2025-12-29","2026-01-26"],[0,43065,"Dummy `nn.Conv2d` in `Sam3PixelDecoder`","james77777778",2,"","",0,"2025-12-29","2026-02-08"],[0,43064,"Trainer.train() using v5 + FSDP2 + PEFT + cpu_ram_efficient_loading=True results in wrong optimizer states/par","d5031",2,"","",0,"2025-12-29","2026-02-06"],[0,43062,"GemmaTokenizer - unclear purpose of Split pre_tokenizer on whitespace","biuq",0,"vaibhav-research","#43106",0,"2025-12-29","2026-01-08"],[0,43054,"text embedding of siglip2 is much worse than siglip","fancyerii",2,"","",0,"2025-12-27","2026-01-28"],[0,43052,"Missing import of torch in distilbert docs","PredictiveManish",1,"PredictiveManish","#43053",0,"2025-12-26","2026-01-05"],[0,43051,"Tensor item cannot be called","lumideez",2,"","",0,"2025-12-26","2026-01-20"],[0,43048,"Need to understand difference between TP support via transformers code v/s Pytorch's native parallelize_module","quic-meetkuma",2,"","",0,"2025-12-26","2026-03-08"],[0,43045,"Multimodal chat sample","Wovchena",3,"","",0,"2025-12-26","2025-12-31"],[0,43043,"Single-scale image input support for SAM3 Mask Decoder","Enq666",3,"","",0,"2025-12-26","2026-09-22"],[0,43039,"When using the Liger Kernel, torch.nn.functional.cross_entropy is called","yurkoff-mv",2,"","",0,"2025-12-25","2026-07-23"],[0,43037,"DeepSeek v3.2 support","freedom-cui",2,"","",0,"2025-12-25","2026-10-20"],[0,43032,"[v5 Regression] BitsAndBytes 4-bit quantization OOM - core_model_loading bypasses quantizer device placement","jwm1969",2,"","",0,"2025-12-24","2026-03-18"],[0,43027,"Implementation of Speculative Cascades","radia78",2,"","",0,"2025-12-24","2026-01-05"],[0,43026,"Sequential Sampling","jiosephlee",2,"","",0,"2025-12-24","2026-08-11"],[0,43025,"Enable passing block_size and bf16_stochastic_round to torchao optimizers via TrainingArguments","klae01",4,"","",0,"2025-12-23",""],[0,43023,"How to investigate \"CAS service error\" during model downloading?","satyrmipt",2,"","",0,"2025-12-23","2026-01-31"],[0,43016,"Dia model cannot load/ generation error","jiqing-feng",0,"Haseebasif7","#43029",0,"2025-12-23","2026-01-07"],[0,43012,"Compiling a bfloat16 model triggers float32 precision PyTorch warning","vadimkantorov",2,"","",0,"2025-12-23","2026-03-05"],[0,43011,"`StaticLayer` cache layer to implement `.crop(seq_len)` to match API of `DynamicLayer`","vadimkantorov",4,"","",0,"2025-12-23",""],[0,43010,"Cache's (and Layer's) `update(...)` method to be decorated with `@torch.no_grad`","vadimkantorov",0,"Rocketknight1","#43897",1,"2025-12-23","2026-03-16"],[0,43009,"cannot import name 'ALLOWED_LAYER_TYPES' from 'transformers.configuration_utils'","AndreasKaratzas",2,"","",0,"2025-12-22","2025-12-22"],[0,43005,"VLM reloaded config provide wrong model_type attribution","xin3he",2,"","",0,"2025-12-22","2026-01-05"],[0,42999,"is_flash_attn_2_available() gives packaging.version.InvalidVersion: Invalid version: 'N/A'","dylanprins",0,"lukepayyapilli","#43004",0,"2025-12-22","2026-01-05"],[0,42994,"quantized model saving failed","jiqing-feng",2,"","",0,"2025-12-22","2026-03-02"],[0,42987,"backwards compatibility with older versions","MasoudMaani",2,"","",0,"2025-12-22","2025-12-25"],[0,42986,"dia model generation precision error because of generation config change.","jiqing-feng",2,"","",0,"2025-12-22","2025-12-22"],[0,42981,"Change from v4 to v5c1: DeprecationWarning: builtin type SwigPyPacked/Object has no __module__ attribute","jrp2014",2,"","",0,"2025-12-21","2026-01-29"],[0,42977,"Add ViT NEPA","sbucaille",4,"","",0,"2025-12-21",""],[0,42971,"Please create a Huggingface Transformers SKILL for Claude","Emasoft",2,"","",0,"2025-12-20","2026-02-03"],[0,42968,"`LlavaNextForConditionalGeneration` forward pass broken","qgallouedec",0,"zucchini-nlp","#42990",1,"2025-12-19","2026-01-13"],[0,42954,"XiaoMi MiMo","ArthurZucker",0,"casinca","#45144",0,"2025-12-19","2026-07-01"],[0,42950,"`masking_utils.create_causal_mask` fails in `@torch.compile(fullgraph=True)`","vadimkantorov",2,"","",0,"2025-12-18","2025-12-22"],[0,42947,"Gradient Checkpointing Ineffective with PEFT LoRA Despite Proper Configuration","yurkoff-mv",2,"","",0,"2025-12-18","2026-02-18"],[0,42943,"Continuous batching: output queue requeue starvation and request-scoped iterator does not terminate on complet","pythongiant",2,"","",0,"2025-12-18","2026-02-08"],[0,42937,"[Bug] RuntimeError with torch.compile (reduce-overhead) and StaticCache on Qwen2.5","lotato",2,"","",0,"2025-12-18","2025-12-18"],[0,42936,"Mask2former model's ignore_value not used after definition","mhmd-j",2,"","",0,"2025-12-17","2026-03-11"],[0,42930,"Inconsistent handling of video_metadata in Qwen3VLVideoProcessor usage example","wagoriginal",2,"","",0,"2025-12-17","2025-12-18"],[0,42925,"`type_vocab_size` is not documented and does not have a default value in TVP config","simonreise",0,"majiayu000","#42928",0,"2025-12-17","2025-12-18"],[0,42918,"model parallelism unit test failed for `modeling_pe_audio_video.py` and `modeling_pe_video.py`","kaixuanliu",0,"CodersAcademy006","#43047",0,"2025-12-17","2025-12-29"],[0,42915,"Qwen3Moe failed with FineGrainedFP8Config","jessiewiswjc",2,"","",0,"2025-12-17","2026-01-25"],[0,42914,"[Tokenizer][OFFLINE] chat_template.jinja not downloaded in cache","khatrimann",2,"","",0,"2025-12-16","2026-01-12"],[0,42913,"Unexpected tokenizer behavior difference from v4 to v5","awni",2,"","",0,"2025-12-16","2026-01-25"],[0,42911,"Make `backbone_config` param of vision models configs expect `Any` instead of `dict | None`","simonreise",1,"simonreise","#42927",0,"2025-12-16","2025-12-18"],[0,42910,"Qwen2VLImageProcessor not respecting Qwen3VL config","opherlieber",0,"qianyue76","#43095",0,"2025-12-16","2026-01-05"],[0,42907,"Failing to Save Dequantized Ministrals/Devstrals","pandora-s-git",2,"","",0,"2025-12-16","2026-03-08"],[0,42902,"Improve GPT OSS Conversion Script","KyleMylonakisProtopia",2,"","",0,"2025-12-16","2026-02-02"],[0,42898,"`clean_up_tokenization_spaces` behavior changes in v5","apaniukov",2,"","",0,"2025-12-16","2026-02-11"],[0,42890,"tests/models/sam_hq/test_modeling_sam_hq.py::SamHQModelIntegrationTest may fail since a lot of cases are lack ","sywangyi",2,"","",0,"2025-12-16","2026-01-26"],[0,42889,"Divergence in UMT T5 Encoder numerical values from v4.57.3 to v5.0.0.rc*","SolitaryThinker",2,"","",0,"2025-12-16","2026-01-22"],[0,42886,"Tokenizer fails to load from cache when HF_HUB_OFFLINE=1 on 4.57.3","chtruong814",2,"","",0,"2025-12-15","2026-02-09"],[0,42868,"sdpa_paged: How does it handle paged cache without padding?","jiqing-feng",2,"","",0,"2025-12-15","2025-12-16"],[0,42862,"RoPE removed in GraniteMoEHybrid models in v5","avihu111",1,"avihu111","#42872",0,"2025-12-14","2025-12-16"],[0,42859,"Numeric values are not properly decoded in PaddleOCR-VL outputs","passing2961",2,"","",0,"2025-12-13","2026-01-21"],[0,42851,"Threaded loading puts tensors on wrong device when device=\"cuda\"","kmod",2,"","",0,"2025-12-12","2025-12-19"],[0,42837,"Regression in SAM2 caused by changes in mark_tied_weights_as_initialized","krzwaraksa",0,"yonigozlan","#42840",0,"2025-12-12","2025-12-17"],[0,42836,"Enforce implementation of sibling.size_in_bytes for model_info.siblings, for getting model files-sizes without","benayat",2,"","",0,"2025-12-12","2025-12-12"],[0,42832,"Question about tie_weights","cjw-d",2,"","",0,"2025-12-12","2026-04-01"],[0,42831,"Accuracy issue associated with FineGrainedFP8","sunghyuckhong",2,"","",0,"2025-12-12","2026-02-08"],[0,42821,"Support batched torch Tensors in ASTFeatureExtractor","psu-om",4,"","",0,"2025-12-11",""],[0,42817,"Remove unnessary `requests` module","CoderTCY",0,"omkar-334","#42845",0,"2025-12-11","2026-01-23"],[0,42804,"[`Quantization FP8`] Native `from_config` support","vasqu",4,"","",0,"2025-12-11",""],[0,42796,"Ministral-3-8B-Instruct tokenizer doesn't handle BPE markers properly","yanancai",2,"","",0,"2025-12-11","2026-01-08"],[0,42794,"`decoder_start_token_id` or `bos_token_id` has to be defined for encoder-decoder generation.","jiqing-feng",0,"zucchini-nlp","#42922",1,"2025-12-11","2025-12-18"],[0,42792,"SpeechT5 TTS pipeline breaks by processor.","jiqing-feng",1,"jiqing-feng","#42830",0,"2025-12-11","2025-12-15"],[0,42791,"Add support for GPT_OSS with tp_plan or enable native tensor parallelism","quic-akuruvil",4,"","",0,"2025-12-11",""],[0,42789,"Why the use of masked_fill","tanmay2004",2,"","",0,"2025-12-11","2026-01-19"],[0,42778,"requires_backends ImportError in transformers 5.0.0rc0","KosmoCHE",0,"LysandreJik","#42799",1,"2025-12-10","2025-12-12"],[0,42771,"FSDP of Trainer does not work well with Accelerate","gouchangjiang",2,"","",0,"2025-12-10","2026-01-19"],[0,42762,"Generation config merging logic prevents explicit override of model-specific defaults back to global defaults","albertvillanova",2,"","",0,"2025-12-10","2026-01-12"],[0,42760,"Support `BatchFeature` in `LengthGroupedSampler` for Multimodal compatibility","npurson",0,"qgallouedec","#48034",1,"2025-12-10","2026-08-18"],[0,42759,"output_hidden_states is not working correctly with SiglipModel class","hamzaakyildiz",4,"","",0,"2025-12-10",""],[0,42758,"Add support for Microsoft VibeVoice models","AliBagheriNejad",4,"","",0,"2025-12-10",""],[0,42757,"cannot import name 'is_offline_mode' from 'huggingface_hub'","dollarser",2,"","",0,"2025-12-10","2025-12-10"],[0,42754,"Excluding weight decay not working properly on most LMs","wwt17",0,"casinca","#42769",0,"2025-12-10","2026-01-18"],[0,42746,"[Devstral 24B] FP8 is currently not working correctly","patrickvonplaten",0,"MekkCyber","#42775",0,"2025-12-09","2025-12-10"],[0,42740,"how to train trocr with transformers 4.57+?","cqray1990",2,"","",0,"2025-12-09","2026-02-18"],[0,42739,"How about adding local kernel loading to `transformers.KernelConfig()`","zheliuyu",2,"","",0,"2025-12-09","2025-12-17"],[0,42738,"BERT-like models with RoPE","hmellor",0,"ed22699","#43067",0,"2025-12-09","2026-10-22"],[0,42728,"the training crashes at logging_steps when running on a single GPU","gouchangjiang",2,"","",0,"2025-12-09","2026-01-16"],[0,42727,"[BUG] [v5] Models loaded with zero3 have weights that are all zeros or all ones.","Jintao-Huang",2,"","",0,"2025-12-09","2025-12-09"],[0,42726,"convert trocr to onnx","cqray1990",2,"","",0,"2025-12-09","2026-01-16"],[0,42723,"git-oss-20b will not properly load with MXFP4 quantization even though Triton version satisfys.","nickeisenberg",2,"","",0,"2025-12-09","2025-12-11"],[0,42722,"Ensure that `OneFormerProcessor` place text `task_inputs` to the same device as other inputs","simonreise",2,"","",0,"2025-12-08","2026-01-19"],[0,42719,"`OneFormerProcessor` does not support fast image processor","simonreise",2,"","",0,"2025-12-08","2025-12-08"],[0,42713,"mulitmodal forward pass for ministral 3 family","rishavranaut",2,"","",0,"2025-12-08","2025-12-12"],[0,42712,"EoMT: `post_process_semantic_segmentation` fails if `patch_offsets` is None","simonreise",1,"simonreise","#42716",0,"2025-12-08","2025-12-09"],[0,42710,"Outstanding issues / PR before we can release v5","ArthurZucker",2,"","",0,"2025-12-08","2026-01-26"],[0,42709,"granitemoehybrid forward(): lots of logits upcast to float32, eating masive VRAM for minimal gain","mramendi",0,"gabe-l-hart","#42753",0,"2025-12-08","2025-12-11"],[0,42698,"parse_response must not accept detokenized text","kibergus",2,"","",0,"2025-12-08","2026-01-16"],[0,42694,"Some questions about the apply_multimodal_rotary_pos_emb function.","Dyb3438",2,"","",0,"2025-12-08","2026-01-16"],[0,42690,"How to run Phi4MultimodalProcessor","wcrzlh",2,"","",0,"2025-12-08","2026-01-16"],[0,42687,"DagsHubCallback fails on Windows","mattiacurri",2,"","",0,"2025-12-07","2025-12-08"],[0,42682,"Add VideoMAEImageProcessorFast","vasanthrpjan1-boop",2,"","",0,"2025-12-07","2025-12-08"],[0,42680,"Olmo-3 flash-attention-2 support","Zephyr271828",2,"","",0,"2025-12-07","2025-12-27"],[0,42679,"Conditional DETR's `post_process_semantic_segmentation` removes one class","simonreise",0,"vasanthrpjan1-boop","#42681",0,"2025-12-07","2026-01-15"],[0,42674,"Missing imports for DetrLoss and DetrHungarianMatcher","sammlapp",2,"","",0,"2025-12-06","2026-01-15"],[0,42673,"Qwen3ForCausalLM leaks VRAM if used in multiple dataloader threads","dxqb",2,"","",0,"2025-12-06","2026-03-02"],[0,42670,"Transformers 4.57.3 needs older huggingface-hub","dlin95123",2,"","",0,"2025-12-06","2025-12-08"],[0,42664,"TrainingArguments does not process fsdp arguments correctly","quic-meetkuma",2,"","",0,"2025-12-05","2026-01-05"],[0,42662,"Deprecate NeptuneCallback","tomaarsen",0,"qgallouedec","#42666",1,"2025-12-05","2025-12-08"],[0,42661,"Bug in processing of self.args.parallelism_config inside Trainer","quic-meetkuma",0,"SunMarc","#42704",1,"2025-12-05","2025-12-08"],[0,42659,"SAM3 Loss","aselimc",4,"","",0,"2025-12-05",""],[0,42658,"Potential bug in Trainer+Accelerate integration","quic-meetkuma",2,"","",0,"2025-12-05","2026-01-13"],[0,42656,"Voxtral ASR Pipeline Integration Broken in 4.57.3","juzhxng",2,"","",0,"2025-12-05","2026-01-13"],[0,42645,"Processor's attribute is mis-detected for child class from #41633","MilkClouds",4,"","",0,"2025-12-05",""],[0,42641,"Cannot inference llava-next with transformers==4.57.1 on dtype=\"auto\" bug","rebel-seinpark",2,"","",0,"2025-12-05","2026-01-22"],[0,42638,"Routing Replay for MoEs","qgallouedec",0,"kashif","#44925",1,"2025-12-04","2026-04-14"],[0,42629,"Blt Model can't pass training_ci","3outeille",0,"preetam1407","#42685",0,"2025-12-04","2025-12-16"],[0,42617,"Not able to run 3d_parallel.py","quic-meetkuma",2,"","",0,"2025-12-04","2026-02-06"],[0,42608,"Ministral-3-3B-Instruct-2512-ONNX does not work on Electron","abdullah-cod9",2,"","",0,"2025-12-03","2026-01-11"],[0,42594,"Qwen3-VL hidden_states are not the same used to compute logits","LautaroEst",0,"casinca","#42609",0,"2025-12-03","2025-12-04"],[0,42591,"Loading local non-Mistral tokenizer incorrectly trigger fix_mistral_regex warning.","tohz",0,"ArthurZucker","#42743",1,"2025-12-03","2025-12-10"],[0,42590,"model type `deepseek_v32`","fernandaspets",2,"","",0,"2025-12-03","2026-07-26"],[0,42589,"Incorrect tokenization `tokenizers` for escaped strings / Mismatch with `mistral_common`","patrickvonplaten",2,"","",0,"2025-12-03","2025-12-16"],[0,42585,"bug in Mistral3","foldl",2,"","",0,"2025-12-03","2026-01-11"],[0,42583,"microsoft/deberta-v3-base init tokenizer failed","fancyerii",2,"","",0,"2025-12-03","2025-12-05"],[0,42582,"4.57.3 auto_docstring crashes with AttributeError","ymcki",2,"","",0,"2025-12-03","2025-12-06"],[0,42581,"Error with Kyutai Speech-To-Text inference","alien-0119",2,"","",0,"2025-12-03","2026-01-10"],[0,42565,"[Bug] Ministral 3 model loading fails with TypeError after PR #42455","Aratako",0,"SunMarc","#42569",1,"2025-12-02","2025-12-02"],[0,42559,"Parallelize the model download with the weight loading","LysandreJik",4,"","",0,"2025-12-02",""],[0,42556,"[v5] Remove `safe_serialization` parameter","Wauplin",2,"","",0,"2025-12-02","2025-12-16"],[0,42555,"[v5] Increase default shard size","Wauplin",2,"","",0,"2025-12-02","2025-12-16"],[0,42552,"Default model for `image-to-text` pipeline fails to load","davidmezzetti",2,"","",0,"2025-12-02","2025-12-04"],[0,42550,"SDPA and FA2 produce different outputs","SimJeg",2,"","",0,"2025-12-02","2025-12-10"],[0,42548,"cannot import name 'PreTrainedModel' from 'transformers'","missTL",2,"","",0,"2025-12-02","2026-03-08"],[0,42546,"dots1 expert bias routing still broken in v5.0.0rc0","fjosw",2,"","",0,"2025-12-02","2025-12-02"],[0,42544,"diffusion model load failed because safetensor","jiqing-feng",2,"","",0,"2025-12-02","2025-12-08"],[0,42541,"Load suno/bark failed due to safetensor","jiqing-feng",2,"","",0,"2025-12-02","2025-12-16"],[0,42540,"Remove slow tokenizer break pipeline loading.","jiqing-feng",0,"itazap","#42628",1,"2025-12-02","2025-12-05"],[0,42539,"Error with text generation pipeline in v5","davidmezzetti",2,"","",0,"2025-12-02","2026-01-09"],[0,42533,"gpt-oss is not working with flash-attention","avihu111",0,"ArthurZucker","#44081",1,"2025-12-01","2026-01-06"],[0,42503,"Add ModernVBERT models","paultltc",1,"paultltc","#42504",0,"2025-12-01","2026-02-23"],[0,42502,"DataCollatorWithFlattening only accepts labels as list","romitjain",0,"hqkqn32","#42620",0,"2025-12-01","2025-12-10"],[0,42499,"Automatic Speech Recognition pipline raises error when torchcodec is installed but not valid","litagin02",2,"","",0,"2025-12-01","2026-01-08"],[0,42492,"'Automatic Speech Recognition' pipeline not working for Phi4 Multimodal Instruct","palladium123",2,"","",0,"2025-11-29","2025-12-01"],[0,42491,"The LoRA model trained with qwen3_moe on hf4.x cannot be used on the current main branch (hf5.x).","linitra24",0,"ArthurZucker","#43261",1,"2025-11-29","2026-01-24"],[0,42490,"Import warning when loading a local model from a path that ends with a slash","kk-89",2,"","",0,"2025-11-28","2025-12-03"],[0,42489,"PEFT training with gradient checkpointing fails","qgallouedec",2,"","",0,"2025-11-28","2025-12-29"],[0,42475,"Add EVA-02 / EVA-02-CLIP vision encoder to Transformers","Aaraviitkgp",2,"","",0,"2025-11-28","2025-11-28"],[0,42469,"now transformers is support Rust ,why not to develop java /scala3","mullerhai",2,"","",0,"2025-11-28","2026-01-06"],[0,42464,"Add SAM 3D Objects Encoder","ghost",3,"","",0,"2025-11-27","2026-09-10"],[0,42460,"Loading a checkpoint with tied weights results in `ValueError: This checkpoint seem corrupted.`","tomaarsen",2,"","",0,"2025-11-27","2025-12-08"],[0,42454,"StaticCache crashes when the batch-size changes","mobicham",2,"","",0,"2025-11-27","2026-01-05"],[0,42451,"[kernels] When multiple device exist in `kernel_mapping`, calling `add_to_mapping()` will repeatedly overwrite","zheliuyu",2,"","",0,"2025-11-27","2025-12-03"],[0,42445,"Gemma3nForCaulsalLM cannot use StaticCache","ChrisStewie404",2,"","",0,"2025-11-27","2025-11-28"],[0,42440,"Gemma3ForConditionalGeneration is not torch.compile compatible","jood-canva",2,"","",0,"2025-11-27","2025-12-02"],[0,42422,"Performance Improvement: Avoid host↔device synchronizations caused by tensor-to-Python conversions and certain","Dhruv88",4,"Dhruv88","#42433",0,"2025-11-26",""],[0,42418,"Custom nn.Parameter initialization in PreTrainedModel subclasses is overwritten by post_init()/from_pretrained","Noietch",2,"","",0,"2025-11-26","2026-09-17"],[0,42417,"FSDP2 + LoRA `model.generate` raise `aten.embedding.default: got mixed torch.Tensor and DTensor` error","Xiao-Chenguang",1,"Xiao-Chenguang","#42436",0,"2025-11-26","2025-12-03"],[0,42414,"`use_liger_kernel` is not compatible with sequence parallel","jue-jue-zi",0,"kashif","#42444",1,"2025-11-26","2025-11-28"],[0,42405,"Integrate FA4 (Flash Attention for Blackwell) into HF Transformers","sambhavnoobcoder",0,"vasqu","#42435",1,"2025-11-25","2026-03-13"],[0,42401,"The `get_(text|image|audio|video)_features` methods have inconsistent output formats, needs aligning for Sente","tomaarsen",1,"tomaarsen","#42564",1,"2025-11-25","2026-01-23"],[0,42398,"LlamaRMSNorm and equivalent module implementations using torch ops","mstojkovicTT",4,"","",0,"2025-11-25",""],[0,42395,"AutoTokenizer.from_pretrained crashed：AttributeError: 'dict' object has no attribute 'model_type' (fixed)","UnderTides",2,"","",0,"2025-11-25","2025-11-25"],[0,42393,"Tokenizer fails to load from cache with local_files_only on 4.57.2","chtruong814",2,"","",0,"2025-11-25","2025-11-25"],[0,42381,"ValueError: Unrecognized model in OrdalieTech/Solon-embeddings-large-0.1. Should have a `model_type` key in it","vador31",2,"","",0,"2025-11-25","2026-01-04"],[0,42378,"Attribute error for AutoTokenizer when loading HF repo from older transformers versions :)","nil0x9",0,"vasqu","#42389",1,"2025-11-25","2025-11-25"],[0,42375,"SAM3 single image inference with multiple text prompt","iariav",2,"","",0,"2025-11-25","2026-02-08"],[0,42374,"'dict' object has no attribute 'model_type'","wangby",0,"vasqu","#42389",1,"2025-11-25","2025-11-25"],[0,42371,"Please use the new API settings to control TF32 behavior, ...","wasertech",0,"khushali9","#42428",0,"2025-11-24","2025-12-01"],[0,42370,"PreTrainedTokenizerBase.pad() broken on tensordict.TensorDict","ligz08",2,"","",0,"2025-11-24","2025-12-01"],[0,42369,"Can't use local model (Solved)","kabayan",0,"vasqu","#42389",1,"2025-11-24","2025-11-25"],[0,42353,"SAM3 point mode is not supported yet?","haofanwang",2,"","",0,"2025-11-24","2025-11-26"],[0,42352,"The transformers-5.0.0.dev0 version has importing bug","xiaoky1",2,"","",0,"2025-11-24","2025-11-25"],[0,42344,"How to fine-tune SAM 3D models?","bruno686",4,"","",0,"2025-11-23",""],[0,42340,"OwlV2 Image Guided Object Detection Reproducibility","PT-10",2,"","",0,"2025-11-23","2025-11-24"],[0,42331,"SAM3 does not support custom inference resolutions","Kallinteris-Andreas",2,"","",0,"2025-11-21","2025-11-22"],[0,42320,"`sdpa_mask_recent_torch` can cause Torch Compile C++ Errors","KyleMylonakisProtopia",2,"","",0,"2025-11-21","2025-12-30"],[0,42316,"v4.41.2 default_data_collator import torch bug","AdrianMoen",2,"","",0,"2025-11-21","2025-12-29"],[0,42313,"[v5, Mistral models broken] Embeddings are automatically tied","patrickvonplaten",2,"","",0,"2025-11-20","2025-12-23"],[0,42312,"Timm doesnt work well with AutoModelForImageClassification, num_labels doesn't work as expected.","lucian-student",2,"","",0,"2025-11-20","2025-12-22"],[0,42291,"Can we disable IPython progress bar and use normal tqdm bar?","weathon",2,"","",0,"2025-11-20","2025-12-28"],[0,42283,"AutoTokenizer / Process does not support Magistral model","benglewis",0,"ArthurZucker","#42299",1,"2025-11-19","2025-11-24"],[0,42281,"Clarification needed: patch for CVE-2025-3777 (improper URL validation in image_utils.py)","Archana-Shinde1",2,"","",0,"2025-11-19","2025-12-28"],[0,42279,"`python -OO -c 'import transformers.pipelines'` crashes","TimDarcet",2,"","",0,"2025-11-19","2025-12-22"],[0,42271,"【Bug Report】Qwen2_5_VLVisionAttention attention_mask is None","JamesCao2048",2,"","",0,"2025-11-19","2025-11-19"],[0,42269,"local_files_only=True in pipeline() does not prevent model downloads in transformers==4.57.1","nmahavishnu",0,"nandan2003","#42318",0,"2025-11-18","2025-12-17"],[0,42259,"AudioFlamingo3ForConditionalGeneration: `RuntimeError: Input type (float) and bias type (struct c10::BFloat16)","mingyi456",0,"zucchini-nlp","#42278",1,"2025-11-18","2025-11-19"],[0,42257,"Lfm2 fails with beam search","popovaan",2,"","",0,"2025-11-18","2025-12-27"],[0,42251,"Finetuning issue with V-JEPA 2","MattLiutt",2,"","",0,"2025-11-18","2025-12-26"],[0,42249,"`parse_response` should drop EOS","qgallouedec",2,"","",0,"2025-11-17","2026-02-15"],[0,42241,"How to use padding with Mistral?","TopCoder2K",2,"","",0,"2025-11-17","2025-11-18"],[0,42226,"LlamaForCasualLM: base_model Attribute Mystery","Maxwilson1101",2,"","",0,"2025-11-16","2025-11-17"],[0,42222,"All vitpose model were brokentransformers/models/vitpose_","lucasjinreal",2,"","",0,"2025-11-15","2026-02-09"],[0,42221,"Repeated `bos_token_id` is added when using tokenizer","Hermit-w",2,"","",0,"2025-11-15","2025-11-16"],[0,42220,"Qwen3 response_schema might not be correct","qgallouedec",2,"","",0,"2025-11-15","2025-12-29"],[0,42219,"Rt_DETR loss attempts returning auxiliary_outputs when auxiliary_loss=False","sinAshish",2,"","",0,"2025-11-14","2025-11-17"],[0,42215,"Stopped working with Qwen Image Edit quantized model","AlpineVibrations",2,"","",0,"2025-11-14","2025-11-17"],[0,42214,"From_pretrained of TableTransformerForObjectDetection could not use only local file","Christhian16",2,"","",0,"2025-11-14","2025-12-24"],[0,42211,"benchmarker_V2: sdpa_backend handling and level 2 config issue","Abdennacer-Badaoui",0,"engmohamedsalah","#42339",0,"2025-11-14","2025-11-24"],[0,42209,"Add Evo2 Genomic foundation model","McClain-Thiel",4,"","",0,"2025-11-14",""],[0,42205,"Support `.to(device)` or Device Aware Handling for Segmentation Labels in `EOMTImageProcessor`","ariG23498",0,"yonigozlan","#42884",0,"2025-11-14","2025-12-17"],[0,42204,"[xpu] [OOB] `pip install -e \".[torch]\"` wrongly installed the CUDA version dependencies","Stonepia",2,"","",0,"2025-11-14","2025-12-22"],[0,42203,"`num_query_tokens` of blip2 might be None","HYLcool",2,"","",0,"2025-11-14","2025-12-28"],[0,42202,"Deformable DETR Finetuning breaks for any dataset","iamsashank09",2,"","",0,"2025-11-14","2026-02-08"],[0,42200,"Request of rewriting implementation of prediction_step in trainer.py","Yacklin",2,"","",0,"2025-11-14","2026-06-05"],[0,42199,"Cardinality error is incorrect for models derived from DETR that do not have an explicit background class","jveitchmichaelis",0,"heathdutton","#43513",0,"2025-11-13","2026-02-09"],[0,42197,"Attempt to access socket despite HF_HUB_OFFLINE = 1 if cache warmed outside current process","fr1ll",2,"","",0,"2025-11-13","2025-11-22"],[0,42194,"VitModel never outputs an attention map","hyenal",2,"","",0,"2025-11-13","2025-11-14"],[0,42181,"GLM4.1v  video inference bug","aminer-404",2,"","",0,"2025-11-13","2025-12-22"],[0,42180,"The `padding` argument is missing from the docstring for `WhisperFeatureExtractor`","TopCoder2K",1,"TopCoder2K","#42286",0,"2025-11-13","2025-11-19"],[0,42179,"Add TileLang Kernel Support","crownz248",3,"","",0,"2025-11-13","2026-10-20"],[0,42175,"Tensorflow not include in the backend when using pip install '.[torch]'","yinpeiqi",0,"Rocketknight1","#42182",1,"2025-11-13","2025-11-18"],[0,42172,"SAM2/EdgeTAM produces different result on consecutive runs only when input_boxes is specified","xenova",1,"xenova","#42173",1,"2025-11-12","2025-11-12"],[0,42169,"404 Not Found when using https://router.huggingface.co/hf-inference","oliveiradany07",2,"","",0,"2025-11-12","2025-12-22"],[0,42163,"Support AutoAWQ","davidmezzetti",2,"","",0,"2025-11-12","2026-09-23"],[0,42161,"Add support for passing in `BatchFeature` to `Wav2Vec2FeatureExtractor.__call__`","ad12",2,"","",0,"2025-11-12","2026-09-22"],[0,42154,"💡 Bounty Platform for Hugging Face Transformers","dineshroxonn",2,"","",0,"2025-11-12","2025-11-13"],[0,42153,"KeyError: 'hidden_states'","officialsahyaboutorabi",2,"","",0,"2025-11-12","2025-12-20"],[0,42147,"Ray Tune trials fail importing datasets","filbeofITK",2,"","",0,"2025-11-11","2025-12-31"],[0,42143,"DataCollatorForLanguageModeling TypeError","mjaliz",2,"","",0,"2025-11-11","2025-11-11"],[0,42142,"'LlavaNextVideoProcessor' has no attribute 'from_dict'.","jiqing-feng",2,"","",0,"2025-11-11","2025-11-11"],[0,42141,"qwen3-vl ModuleNotFoundError: No module named 'awq.modules'; 'awq' is not a package","LiaoZihZrong",2,"","",0,"2025-11-11","2025-12-19"],[0,42140,"KeyError: 'hidden_states'","officialsahyaboutorabi",2,"","",0,"2025-11-11","2025-11-11"],[0,42137,"CLIP issues with Flash Attention 3 pt.2","akashpalla",3,"","",0,"2025-11-11","2025-11-14"],[0,42132,"Support for Megatron","ri938",3,"","",0,"2025-11-10","2026-10-20"],[0,42121,"kernels-community/flash-attn3 does not work with SmolVLM2","MilkClouds",0,"i3hz","#42149",0,"2025-11-10","2025-11-12"],[0,42118,"[audio] standardize conv output length computation","eustlb",2,"","",0,"2025-11-10","2025-12-19"],[0,42116,"Integration of the SINQ quantization strategy","ChiaraBoretti",1,"ChiaraBoretti","#43112",0,"2025-11-10","2026-02-16"],[0,42115,"Why pass size[\"longest_edge\"] to max_pixels, and size[\"shortest_edge\"] to min_pixels when doing smart_resize","leijie-ww",2,"","",0,"2025-11-10","2025-12-18"],[0,42113,"Add AutoMergeAdapters: Official Utility to Combine Multiple LoRA Adapters into One Unified Model","3015pavan",2,"","",0,"2025-11-09","2025-11-10"],[0,42111,"Add thinking-budget support (max_thinking_tokens) for reasoning-capable chat models","AndresAlgaba",3,"","",0,"2025-11-09","2026-07-28"],[0,42109,"Potential bug in xLSTM (native) training","stefan-it",2,"","",0,"2025-11-08","2025-12-18"],[0,42103,"Compatability with `torchcodec`","Samoed",2,"","",0,"2025-11-07","2026-01-15"],[0,42100,"Qwen3Moe models have bug during generation since they calculate unnecessary load_balancing_loss","zhangchong25",2,"","",0,"2025-11-07","2026-01-05"],[0,42093,"Mbart decoder ignoring index 0 from labels | index 1 from dec in","jaaabir",2,"","",0,"2025-11-07","2025-11-07"],[0,42087,"SpeechT5 RelativePositionalEncoding can create empty tensors","ppadjinTT",1,"ppadjinTT","#42090",0,"2025-11-07","2025-11-19"],[0,42086,"Does Trainer uses grad scaler for training?","quic-meetkuma",2,"","",0,"2025-11-07","2025-11-13"],[0,42084,"[Qwen3-VL] The comment for this function may be incorrect.","pengshuang",2,"","",0,"2025-11-07","2025-11-09"],[0,42081,"Trainer code is tightly coupled with backend/hardware specific calls","quic-meetkuma",2,"","",0,"2025-11-07","2025-12-19"],[0,42077,"Qwen3-VL-4B use trace build graph bug","justzhanghong",2,"","",0,"2025-11-07","2025-12-15"],[0,42064,"[audio] weight_norm standardization","eustlb",2,"","",0,"2025-11-06","2025-12-15"],[0,42056,"Glm4vProcessor passes False instead of tokenizer instance when use_fast=False","zRzRzRzRzRzRzR",2,"","",0,"2025-11-06","2025-12-20"],[0,42035,"Add StreamingSentimentPipeline: Real-Time Sentiment Analysis for Live Data Streams (WebSocket, Kafka, HTTP)","3015pavan",2,"","",0,"2025-11-05","2025-11-06"],[0,42032,"ValueError: Unrecognized configuration class <class 'transformers.models.qwen3_omni_moe.configuration_qwen3_om","Tortoise17",2,"","",0,"2025-11-05","2025-12-27"],[0,42029,"Need to understand logging_steps in Trainer.","quic-meetkuma",2,"","",0,"2025-11-05","2026-01-09"],[0,42027,"Empty tensor in torch model trace for concat operation","apbose",2,"","",0,"2025-11-05","2026-01-07"],[0,42024,"`PreTrainedTokenizerBase.model_input_names` is a singleton","stephantul",0,"yashwantbezawada","#42051",0,"2025-11-05","2025-12-13"],[0,42017,"[bug] Lfm2 issue in second prefill stage when asking followup question","joeldushouyu",2,"","",0,"2025-11-04","2025-11-05"],[0,42016,"Load Mixtral-8x22B-Instruct-v0.1 lead to OOM","Hermit-w",0,"Cyrilvallez","#42129",1,"2025-11-04","2025-11-13"],[0,42014,"Integration of a High-Performance, Real-Time Optimized Sentiment Analysis Model (DeBERTa-v3 + LORA) into Huggi","3015pavan",2,"","",0,"2025-11-04","2025-11-05"],[0,42010,"[bug] tokenizer apply_chat_template default param return_dict's value changed","WingEdge777",2,"","",0,"2025-11-04","2025-11-05"],[0,42005,"## SuperGlue Bug Report: Index Out of Bounds in `post_process_keypoint_matching`","m-christen",0,"yonigozlan","#42018",0,"2025-11-04","2025-11-20"],[0,42003,"KeyError: 'rope_parameters_factor' when using convert_gpt_oss_weights_to_hf.py for openai/gpt-oss-20b","ghost",2,"","#42007",0,"2025-11-04","2025-11-06"],[0,41999,"Add Timestamp Support for Voxtral Models","juzhxng",3,"","",0,"2025-11-03","2026-10-20"],[0,41984,"Variable name mismatch on `MixtralExperts.forward()`","kwonmha",0,"diegoakel","#42041",0,"2025-11-03","2025-11-07"],[0,41981,"KeyError: 'optimum.quanto' for `_is_package_available`","kaixuanliu",0,"yashwantbezawada","#42050",0,"2025-11-03","2025-11-06"],[0,41970,"Huggingface Hub version increased causing import error","harveyz0",2,"","",0,"2025-11-01","2025-12-11"],[0,41964,"Improve typing support and flexibility for RopeParameters","RyanMullins",1,"RyanMullins","#49831",0,"2025-10-31","2026-10-26"],[0,41960,"Transformers 4.57.1 not working with torch < 2.3","iglaweb",0,"Cyrilvallez","#42304",1,"2025-10-31","2025-11-20"],[0,41959,"DeepSpeed ZeRO-3 Resizing untied output head after `resize_token_embeddings` breaks (LoRA + ZeRO-3)","XiangZhang-zx",2,"","",0,"2025-10-30","2025-12-08"],[0,41955,"`max_pixels` parameter ignored when loading Qwen2VL/Qwen3VL image processors via `from_pretrained()`","ReinforcedKnowledge",0,"yonigozlan","#41997",0,"2025-10-30","2025-11-04"],[0,41950,"video-classification pipeline looks for image processors","merveenoyan",0,"J3r3myPerera","#46256",0,"2025-10-30","2026-06-05"],[0,41948,"Does Qwen2VLImageProcessor treat two consecutive images as one group/feature?","priancho",2,"","",0,"2025-10-30","2025-10-31"],[0,41947,"why Smolvlm-256M-Instruct slower then Internvl-v2-1B ?","HuangChiEn",2,"","",0,"2025-10-30","2025-10-31"],[0,41945,"Consider not using emojis in `print`, which encounterred encoding error.","acane77",0,"7amim","#42091",0,"2025-10-30","2025-11-11"],[0,41944,"SPDA, FA2 vs. Eager Attention Implementation leading to different losses","jiosephlee",2,"","",0,"2025-10-30","2025-11-05"],[0,41943,"error: argument --include_num_input_tokens_seen/--include-num-input-tokens-seen: Truthy value expected: got no","guofy-ai",2,"","",0,"2025-10-30","2025-12-13"],[0,41935,"Missing `config.json` and `preprocessor_config.json` in `kyutai/moshiko-pytorch-bf16 model` repo","akshatvishu",2,"","",0,"2025-10-29","2025-12-08"],[0,41929,"ViT model's output_attention is not work","naturesh",0,"molbap","#41930",1,"2025-10-29","2025-11-12"],[0,41927,"Nightly / Nvidia CI workflows trigger on forks and fail due to missing org-specific runners","AvinashDwivedi",2,"","",0,"2025-10-29","2025-11-28"],[0,41924,"`output_attentions=True` always warns for non-`\"eager\"` attention implementations, even when a custom Attentio","kannandeepti",0,"Aravind-11","#42334",0,"2025-10-28","2025-11-26"],[0,41919,"LFM2 image_processing_lfm2_vl_fast.py Mean Std swapped?","florianvoss-commit",0,"yonigozlan","#41940",0,"2025-10-28","2025-10-31"],[0,41913,"`epoch` in the log message uses a wrong denominator under some conditions","nzw0301",2,"","",0,"2025-10-28","2025-12-07"],[0,41911,"The forward() Method in ModernBertForTokenClassification is missing **kwargs","SinaDBMS",2,"","",0,"2025-10-28","2025-12-04"],[0,41910,"Breaking change about AWQ Fused modules due to Attention Refactor","fanqiNO1",1,"fanqiNO1","#41909",0,"2025-10-28","2025-11-20"],[0,41908,"from_pretrained will fail if device_map is `torch.device(\"mps\", index=0)`","oceanusxiv",2,"","",0,"2025-10-28","2025-12-08"],[0,41906,"[Feature Request] Add CPU Inference Benchmark Framework to Transformers","AmazingcatAndrew",2,"","",0,"2025-10-28","2025-10-28"],[0,41902,"Cannot load models in latest transformers version","ri938",0,"luaenrique","#41903",0,"2025-10-27","2025-10-28"],[0,41898,"inaccurate eval loss computation","wwt17",2,"","",0,"2025-10-27","2025-12-06"],[0,41896,"Issues when updating the .gitignore file in run_clm_no_trainer.py","wwt17",0,"luaenrique","#41957",0,"2025-10-27","2025-10-31"],[0,41893,"Add MiniViT: lightweight Vision Transformer for CIFAR-scale image classification","justynigam",4,"","",0,"2025-10-27",""],[0,41885,"DINOv3ViTLayer_FP16 error","AnnaTrainingG",2,"","",0,"2025-10-27","2025-12-06"],[0,41881,"FSDP2 training hangs during backward pass with MoE models when some experts are not activated","LucienXian",2,"","",0,"2025-10-27","2026-01-21"],[0,41876,"LlamaAttention num_heads","shanhx2000",2,"","",0,"2025-10-27","2025-10-31"],[0,41875,"Flash Attention Error During AutoModelForSeq2SeqLM.generate (model: gemma-2b-2b-ul2-it)","avanunts",0,"vasqu","#41890",1,"2025-10-26","2025-11-11"],[0,41874,"Distributed training of SigCLIP","zyk1559676097-dot",2,"","",0,"2025-10-26","2025-12-04"],[0,41872,"Improve robustness of batch_decode in PreTrainedTokenizerBase","AvinashDwivedi",2,"","",0,"2025-10-26","2025-10-28"],[0,41870,"GemmaTokenizerFast inconsistent with Sentencepiece tokenizer","jedreky",2,"","",0,"2025-10-26","2025-12-04"],[0,41867,"[RFC] Automatic CPU dtype fallback and thread optimization","AmazingcatAndrew",2,"","",0,"2025-10-26","2025-12-03"],[0,41863,"`generate()` produces incoherent output when `inputs_embeds` has length 1","tyarkoni",0,"zucchini-nlp","#42088",1,"2025-10-25","2025-12-23"],[0,41862,"Request for InternVL3_5_Flash","YanxingLiu",4,"","",0,"2025-10-25",""],[0,41861,"transformers.Adafactor is almost 2x slower on Windows than Linux - even WSL is slow what can be reason?","FurkanGozukara",2,"","",0,"2025-10-25","2025-12-03"],[0,41859,"Human Verification not working?","thefued",2,"","",0,"2025-10-25","2025-10-26"],[0,41858,"Wav2Vec2PhonemeCTCTokenizer phonemizer backend problem.","QishengL",2,"","",0,"2025-10-25","2025-10-25"],[0,41856,"Performance regression: `allow_is_causal_skip` incorrectly disabled when `use_cache=False`","williamsnell",2,"","",0,"2025-10-25","2025-12-28"],[0,41846,"Incompatibility single-modality AutoProcessor and PEFT Adapter","tomaarsen",0,"yonigozlan","#42402",0,"2025-10-24","2025-12-02"],[0,41842,"Incorrect usage of `num_items_in_batch`?","gohar94",2,"","",0,"2025-10-24","2025-12-01"],[0,41838,"ONNX export vmap functorch\"","Worke1221",2,"","",0,"2025-10-24","2026-10-20"],[0,41826,"Integrating TiledMLP for a much smaller memory footprint","stas00",4,"","",0,"2025-10-23",""],[0,41819,"IndexError: tuple index out of range when using Tensor Parallelism with FSDP2 on GPT-OSS 20B (tensor_parallel.","JdRion",2,"","",0,"2025-10-23","2026-01-11"],[0,41816,"Processor saving does not work when multiple tokenizers","AmitMY",0,"yonigozlan","#42667",0,"2025-10-23","2025-12-16"],[0,41814,"AutoModel does not support Qwen3VLMoE","AylinAkkus",2,"","",0,"2025-10-23","2025-12-01"],[0,41813,"Hugginface transformers downloading models doesnt work!","lucian-student",2,"","",0,"2025-10-23","2025-10-24"],[0,41810,"How do you use t5gemma decoder with a different encoder?","kushaltatariya",2,"","",0,"2025-10-23","2025-12-01"],[0,41809,"[Bug] Qwen3-VL beam search with video inputs.","rzhao-zhsq",2,"","",0,"2025-10-23","2025-12-02"],[0,41803,"torch.compile graph break in `flash_attention_v2` backend","StrongerXi",3,"","",0,"2025-10-23","2026-10-20"],[0,41782,"Local pretrained models cannot be loaded with multithreading","Crissium",2,"","",0,"2025-10-22","2025-10-22"],[0,41779,"Broken documentation link in Transformers website","ruheena-shaik",2,"","",0,"2025-10-22","2025-11-30"],[0,41775,"Hugging Face website and models not reachable","christian-rauch",3,"","",0,"2025-10-22","2025-11-21"],[0,41762,"`IndexError: index 0 is out of bounds for dimension 0 with size 0` when loading Gemma3ForConditionalGeneration","Asunatan",2,"","",0,"2025-10-21","2025-10-22"],[0,41760,"`find_adapter_config_file()` got an unexpected keyword argument `_adapter_model_path`","avihu111",2,"","",0,"2025-10-21","2025-11-21"],[0,41756,"InternVL3-8B quantize got TypeError: can only concatenate str (not \"list\") to str","BigFaceBoy",2,"","",0,"2025-10-21","2025-11-20"],[0,41753,"Please port for MPS acceleration on MacOS.","shyamalschandra",3,"","",0,"2025-10-20","2026-10-20"],[0,41749,"`_get_num_multimodal_tokens` is not implemented for model `mllama`","mrtpk",2,"","",0,"2025-10-20","2025-10-21"],[0,41746,"When Qwen2VLConfig is saved and loaded again, an incorrect model_type will be obtained","ZX-ModelCloud",2,"","",0,"2025-10-20","2025-11-20"],[0,41740,"Safety Checking Infrastructure for Text Generation","rice-e",3,"","",0,"2025-10-20","2026-07-23"],[0,41738,"KeyError: 'florence2_language'","conceptofmind",0,"i3hz","#41866",0,"2025-10-19","2025-10-29"],[0,41737,"this PR added a few untested notebooks that need to be fixed/verified by AMD","Mahdi-CV",2,"","",0,"2025-10-19","2025-10-23"],[0,41732,"Qwen3Next compatibilty issue with flash-linear-attention","Tcc0403",0,"SrijanUpadhyay","#41735",0,"2025-10-19","2025-10-23"],[0,41731,"transformers CLI documentation issue ","ArjunPimpale",2,"","",0,"2025-10-19","2025-12-22"],[0,41720,"Qwen3 with auto device mapping fails due to cudaErrorAssert on A800","guosyjlu",2,"","",0,"2025-10-18","2026-01-05"],[0,41709,"Qwen3-VL Processor combined images in the same batch as one message","weathon",2,"","",0,"2025-10-18","2025-11-17"],[0,41708,"Cannot load LLaMA3 with transformers==4.57.1 on non-cuda accelerators using tensor parallelism (tp_plan=\"auto\"","trajepl",2,"","",0,"2025-10-17","2026-01-12"],[0,41706,"pip install -e \".[dev]\" not working","Aravind-11",2,"","",0,"2025-10-17","2025-10-18"],[0,41680,"[BUG] Qwen3-Omni Training. AttributeError: 'Qwen3OmniMoeThinkerForConditionalGeneration' object has no attribu","Jintao-Huang",0,"BakerBunker","#41741",0,"2025-10-17","2025-10-20"],[0,41669,"Remove import * usage from models, adds 10 seconds and 38k files","bschnurr",2,"","",0,"2025-10-16","2025-11-24"],[0,41668,"CLIP incompatible with Flash Attention 3","akashpalla",0,"vasqu","#41750",1,"2025-10-16","2025-10-24"],[0,41652,"DataCollatorWithFlattening incompatible with sequence classification","LPirch",2,"","",0,"2025-10-16","2025-11-24"],[0,41651,"Erroneous flash attention version check","franzscherr",0,"Cyrilvallez","#41661",1,"2025-10-16","2025-10-16"],[0,41643,"Qwen3VL input_ids = input_ids[attention_mask[i] == 1]","bent1e",2,"","",0,"2025-10-16","2025-11-24"],[0,41640,"AttributeError: BartTokenizerFast has no attribute image_token. Did you mean: 'mask_token'?","conceptofmind",0,"zucchini-nlp","#41659",1,"2025-10-16","2025-10-17"],[0,41639,"MarianMTModel performance regression due to Bidirectional masks","jiqing-feng",0,"vasqu","#41852",1,"2025-10-16","2025-11-10"],[0,41630,"Incorrect Access of Dataset Fields in check_tokenizers.py","vignesh1507",2,"","",0,"2025-10-15","2025-11-24"],[0,41628,"Cannot import name 'AutoImageProcessor' from 'transformers'","Pittmann-XIE",2,"","",0,"2025-10-15","2025-10-16"],[0,41623,"Qwen Audio DisjunctiveConstraint missing","Znerual",2,"","",0,"2025-10-15","2025-11-24"],[0,41622,"Florence 2 _supports_sdpa missing in remote code","Znerual",2,"","",0,"2025-10-15","2025-10-16"],[0,41615,"Issue Report: Abnormal GPU Utilization Pattern - DDP Training CLIP Model from Scratch","xiehuanyi",2,"","",0,"2025-10-15","2025-11-24"],[0,41600,"torch can not be detected by transformers on aarch","HuangChiEn",2,"","",0,"2025-10-15","2025-10-16"],[0,41598,"EncodecFeatureExtractor.__call__() got an unexpected keyword argument 'pad_to_multiple_of'","auli-aziz",2,"","",0,"2025-10-15","2025-11-26"],[0,41570,"Uneven GPU memory usage and CUDA OOM during multi-GPU inference with Qwen3-VL-30B-A3B-Thinking","Xqle",2,"","",0,"2025-10-14","2025-11-23"],[0,41566,"Performance regression due to Bidirectional masks","jiqing-feng",2,"","",0,"2025-10-14","2025-10-15"],[0,41562,"NameError: name 'PreTrainedModel' is not defined","deepNoah",2,"","",0,"2025-10-14","2025-10-14"],[0,41560,"Trainer._algin_special_token can't figure out tokenizezr_has_new_eos correctly when model.generation_config.eo","WayenVan",2,"","",0,"2025-10-14","2025-10-14"],[0,41554,"model.from_pretrained( . . . ) not loading needed weights/parameters","lorsonb",2,"","",0,"2025-10-13","2025-11-24"],[0,41553,"Bad error message for AutoTokenizer loading Voxtral","jackzhxng",0,"ArthurZucker","#42299",1,"2025-10-13","2025-11-24"],[0,41544,"ValueError: You current version of `autoawq` does not support module quantization skipping, please upgrade `au","EIIvy",2,"","",0,"2025-10-13","2025-11-21"],[0,41543,"Add VideoInpaintPipeline for temporally-consistent diffusion-based video inpainting","Aki-07",2,"","",0,"2025-10-13","2025-10-17"],[0,41539,"All POETRY operations fail on latest version 4.57.0","bfuia",2,"","",0,"2025-10-13","2025-10-13"],[0,41533,"Add_Specifical_tokens and resize_toked_embeddings result in an error","jialiangZ",2,"","",0,"2025-10-12","2025-10-13"],[0,41532,"where is examples/rag from original paper?","IgorKasianenko",2,"","",0,"2025-10-12","2025-10-17"],[0,41526,"Eager Attention Works with OneVision 0.5B but Not with 7B","varungupta31",2,"","",0,"2025-10-11","2025-10-13"],[0,41520,"Add GPU-Accelerated Video Decoding for VideoMAE, TimeSformer, and Similar Models","Aki-07",1,"Aki-07","#41534",0,"2025-10-11","2025-10-13"],[0,41519,"Add Shared Vision–Language Negative Sampling Utility for BLIP / LLaVA / IDEFICS","Aki-07",3,"","",0,"2025-10-11","2026-07-16"],[0,41518,"Add Structured Prompt Templates Registry for LLM / VLM / Diffusion Tasks","Aki-07",3,"","",0,"2025-10-11","2026-09-10"],[0,41517,"Add MobileNetV2 Fast Image Processor","Aki-07",2,"","",0,"2025-10-11","2025-10-11"],[0,41516,"when finetune deepseek v3.1 with lora，an ValueError was reported: QuantizationMethod.FP8 do not support traini","Francisapzii",2,"","",0,"2025-10-11","2025-11-19"],[0,41509,"Warning when using the \"image-to-text\" pipeline","intexcor",0,"Rocketknight1","#41513",1,"2025-10-10","2025-10-13"],[0,41501,"Paged attention does not work","yuyijiong",2,"","",0,"2025-10-10","2025-11-19"],[0,41498,"mamba2 mixer different outputs between fast, slow, training, and inference paths.","tsengalb99",2,"","",0,"2025-10-10","2025-11-17"],[0,41494,"Incorrect tokenizer created for gemma gguf files","amychen85",2,"","",0,"2025-10-09","2025-11-29"],[0,41492,"For finetuning MBart-based model, setting decoder_start_token_id in model.config is NOT ENOUGH.","Bmingg",2,"","",0,"2025-10-09","2025-11-17"],[0,41472,"Python 3.13.8 breaks DeBERTa-v2 and SEW-D model loading","albertvillanova",2,"","",0,"2025-10-09","2025-10-16"],[0,41465,"Support s_aux in GPT-OSS","LoserCheems",3,"","",0,"2025-10-09","2026-10-02"],[0,41463,"Pop catch kwarg in run_hp_search_optuna() for flexible Optuna error handling","nicha-api",2,"","",0,"2025-10-09","2025-10-14"],[0,41462,"Support encoder text classification for sequence to sequence models like BART and T5","cbhyphen",4,"","",0,"2025-10-09",""],[0,41455,"RuntimeError with dtype=torch.float16 and dtype=torch.bfloat16 in HQQ 4-bit, 8-bit quantized Granite4 models o","prathamesh-chavan-22",2,"","",0,"2025-10-08","2025-12-22"],[0,41454,"Huggingface wav2vec2-Conformer conversion script fails for Fairseq models","Getmany1",2,"","",0,"2025-10-08","2025-12-12"],[0,41453,"SmolVLM2 cannot be used quantized","anderlem",2,"","",0,"2025-10-08","2025-11-17"],[0,41447,"ImportError: cannot import name 'Qwen3VLMoeForConditionalGeneration' from 'transformers'","EIIvy",2,"","",0,"2025-10-08","2026-01-12"],[0,41431,"gradient scaling occurs even though total gradient remains < max_grad_norm in trainer.py","lorsonb",2,"","",0,"2025-10-07","2025-11-15"],[0,41418,"Qwen3 VL Moe: Expected self.dtype to be equal to src.dtype","danielquintas8",1,"danielquintas8","#41420",0,"2025-10-07","2025-10-14"],[0,41399,"Filter models by Original Provider","HarrySoteriou",3,"","",0,"2025-10-07","2026-10-20"],[0,41398,"OwlViT Process has error 'numpy.ndarray' object has no attribute 'to'","PinedTD2025",2,"","",0,"2025-10-07","2025-10-08"],[0,41397,"ViTForMaskedImageModeling (vit-base-patch16-224-in21k) distributed training warning","willxxy",2,"","",0,"2025-10-07","2025-11-15"],[0,41392,"automaticlly setting attention mask when pas token is equal to the eos token","yonikremer",2,"","",0,"2025-10-06","2025-10-08"],[0,41387,"Parakeet and PerceptionLM are not registered correctly for auto","fschlatt",1,"fschlatt","#41464",0,"2025-10-06","2025-10-09"],[0,41379,"RT-Detr 2d positional embedding bug","konstantinos-p",2,"","",0,"2025-10-06","2025-11-06"],[0,41363,"RT-Detr docs should reflect fixed 640x640 input size","konstantinos-p",2,"","",0,"2025-10-06","2025-11-06"],[0,41345,"Add standardized multimodal capability metadata","adarshxs",2,"","",0,"2025-10-04","2025-10-06"],[0,41344,"No need to gather all parameters to save model when training with DeepSpeed Zero3 and LoRA","Taise228",1,"Taise228","#41416",0,"2025-10-04","2025-12-16"],[0,41341,"Cannot install latest transformers version 4.57.0 in free-threaded python version 3.14","fleonce",2,"","",0,"2025-10-04","2025-12-01"],[0,41339,"Transformers 4.57.0 declares compatibility with Python>=3.9 but is only compatible with >=3.10","anakin87",2,"","",0,"2025-10-04","2025-10-23"],[0,41336,"is there a bug in group_videos_by_shape for qwenvl video preprocessiong?","dichencd",2,"","",0,"2025-10-03","2025-10-03"],[0,41331,"Broken metadata in transformers 4.57.0 causing Poetry <empty> version parse error","ashisjena",2,"","",0,"2025-10-03","2025-10-14"],[0,41327,"Add DEIMv2","yonigozlan",2,"","",0,"2025-10-03","2025-10-03"],[0,41311,"Unskip and fix test_offline_mode by using HF_HUB_OFFLINE and hermetic cache warmup","arvinder004",2,"","",0,"2025-10-03","2025-12-15"],[0,41306,"TiledFusedLogitsLoss (LigerFusedLinearCrossEntropy, FLCE) Flag in `from_pretrained()`","Tcc0403",4,"","",0,"2025-10-03",""],[0,41301,"HTTP proxies via env variables no longer working in the main branch of transformers","jerryzh168",2,"","",0,"2025-10-02","2025-10-16"],[0,41296,"Parsing error with --lr-scheduler-kwargs: invalid dict value","stephankoe",2,"","",0,"2025-10-02","2025-10-14"],[0,41278,"Segmentation Fault with Qwen2 on Multi-GPU (device_map='auto')","justice-Pan",2,"","",0,"2025-10-02","2025-11-11"],[0,41274,"does DeepSeek-V3.2-Exp have update？","Mally-cj",2,"","",0,"2025-10-02","2025-11-09"],[0,41271,"Add Hierarchical Reasoning Model","zbloss",2,"","",0,"2025-10-01","2026-10-01"],[0,41247,"Custom loss function for Qwen 2.5 VL 7b training","akhila1729",2,"","",0,"2025-10-01","2025-11-10"],[0,41246,"Add Contribution Guidelines and License Section in README","dimpal-yadav",2,"","",0,"2025-10-01","2025-11-09"],[0,41245,"Add Pull Request (PR) Template","dimpal-yadav",2,"","",0,"2025-10-01","2025-11-07"],[0,41244,"Add hacktoberfest-2025 Banner to README","dimpal-yadav",0,"aastha44ma","#1",1,"2025-10-01","2025-11-09"],[0,41238,"[Bug] RuntimeError: dtype mismatch in _group_beam_search with bfloat16/fp16 models","anmorgunov",2,"","",0,"2025-09-30","2025-10-01"],[0,41235,"i want to request a demo code for StatefulDataLoader , i want to use data checkpoint to recover the train stag","ldh127",2,"","",0,"2025-09-30","2025-11-08"],[0,41230,"Consider forking and maintaining pyctcdecode or switch to torchaudio.models.decoder","FredHaa",2,"","",0,"2025-09-30","2025-12-16"],[0,41213,"Qwen3_Omni error inference transformer 4.57.0.dev","Tortoise17",2,"","",0,"2025-09-30","2025-10-01"],[0,41211,"Add DEIMv2","NielsRogge",0,"harshaljanjani","#44339",0,"2025-09-30","2026-04-27"],[0,41210,"Qwen3-Next tokenizer fails to apply chat template for openai-style message inputs","felixzhu555",2,"","",0,"2025-09-30","2025-09-30"],[0,41209,"Why it's easy to get nan for A_log in Qwen3Next","npuichigo",2,"","",0,"2025-09-30","2025-11-08"],[0,41208,"Integrate mamba SSM kernels from the hub","romitjain",2,"","",0,"2025-09-30","2025-12-18"],[0,41206,"LLaVA-Gemma processor is bugged","nhatkhtn",2,"","",0,"2025-09-29","2025-09-30"],[0,41203,"[audio] standardization tracker","eustlb",2,"","",0,"2025-09-29","2025-12-13"],[0,41196,"[Model] Support Deepseek-V3.2-Exp","hnyls2002",2,"","",0,"2025-09-29","2026-07-31"],[0,41191,"[BUG] Qwen3-Omni: save_pretrained causes abnormal audio_token_id, leading to incorrect inference (in training ","Jintao-Huang",0,"goodhee","#41192",0,"2025-09-29","2025-09-30"],[0,41190,"Regression in SmolVLM results in different vision embeddings","yfw",0,"i3hz","#41374",0,"2025-09-28","2025-10-07"],[0,41189,"Documentation says manual `tp_plan` can be set but it can't be set.","st81",0,"stevhliu","#41205",1,"2025-09-28","2025-09-30"],[0,41187,"There is no license in the readme","vinitjain2005",2,"","",0,"2025-09-28","2025-09-29"],[0,41186,"Qwen2.5-VL  restore tensor multi-image form","NiFangBaAGe",2,"","",0,"2025-09-28","2025-11-05"],[0,41185,"llama3-8b","52566rz",2,"","",0,"2025-09-27","2025-09-29"],[0,41184,"Continuous Batching sliding window attention mask is wrong","NixGD",2,"","",0,"2025-09-27","2025-10-13"],[0,41180,"Qwen2.5-VL-7B-Instruct Accuracy Regression Still Persists in v4.56.2","rahul-tuli",2,"","",0,"2025-09-26","2025-11-29"],[0,41179,"ImportError `flash_attn_3` while setting `attn_implementation=\"flash_attention_3\"` on hopper GPU","allenphilipj",2,"","",0,"2025-09-26","2025-11-05"],[0,41178,"Memory leak when using openai/clip-vit-base-patch32","Adefey",2,"","",0,"2025-09-26","2025-11-29"],[0,41171,"LFM2-VL PyTorch","yukiarimo",2,"","",0,"2025-09-25","2025-09-27"],[0,41164,"cache offloading check is incorrect","mobicham",2,"","",0,"2025-09-25","2025-09-29"],[0,41152,"[Bug] Qwen2.5-VL prepare_inputs_for_generation does not pass second_per_grid_ts to get_rope_index","Xqle",1,"Xqle","#41153",0,"2025-09-25","2025-09-25"],[0,41151,"Incorrect `norm_topk_prob` config for loading Qwen3 MoE GGUF","yuko29",2,"","",0,"2025-09-25","2025-11-17"],[0,41150,"Error message: XPU out of memory. Tried to allocate 4.00 GiB (GPU 0; 15.11 GiB total capacity; 0 bytes already","fablevi",2,"","",0,"2025-09-25","2025-12-29"],[0,41146,"pipeline import fails with torchvision::nms operator mismatch","paytonison",2,"","",0,"2025-09-25","2025-11-02"],[0,41141,"Need a concise example of Tensor Parallelism (TP) training using Trainer/SFTTrainer.","meet-minimalist",2,"","",0,"2025-09-25","2026-06-22"],[0,41140,"Is`_index_first_axis` functionally equal to FA2's `index_first_axis`","A1waysBeenHere",2,"","",0,"2025-09-25","2025-09-28"],[0,41139,"Suggestion: Support HMP protocol for multi-agent LLM workflows","kagvi13",2,"","",0,"2025-09-24","2025-09-25"],[0,41136,"Real time sentiment analysis dashboard","3015pavan",2,"","",0,"2025-09-24","2025-09-25"],[0,41129,"tensor mismatch when finetuning a smollm3 model","codefusser",2,"","",0,"2025-09-24","2025-11-24"],[0,41125,"Confusion about Qwen3-Next’s linear attention","fanglinpu",2,"","",0,"2025-09-24","2025-09-25"],[0,41115,"Add Model Architecture for MiniCPM3","bzantium",1,"bzantium","#41116",0,"2025-09-24","2026-06-23"],[0,41108,"`predict_step` in Trainer should pass `num_items_in_batch`","pramodith",1,"pramodith","#41183",1,"2025-09-23","2025-09-30"],[0,41093,"IndexError: The shape of the mask [1406] at index 0 does not match the shape of the indexed tensor [1405] at i","wyn1015",0,"zucchini-nlp","#41173",1,"2025-09-23","2025-10-06"],[0,41092,"Failed to run `Qwen3-235B-A22B` with `tp_size=8`","Hermit-w",2,"","",0,"2025-09-23","2025-09-23"],[0,41090,"wrong metadata of widht and height when read video use read_video_torchvision","Juude",2,"","",0,"2025-09-23","2025-09-24"],[0,41085,"a small bug in qwen3_moe","shiwanghua",2,"","",0,"2025-09-23","2025-10-31"],[0,41084,"Set Block Decoding","davidmrau",3,"","",0,"2025-09-23","2026-10-20"],[0,41081,"Add support for LLaVA-OneVision-1.5 Multi-Modal Model","g1050",4,"","",0,"2025-09-23",""],[0,41080,"FileNotFoundError: Missing safe tensors file: D:\\docling\\docling_model\\model.safetensors","rinno1027",2,"","",0,"2025-09-23","2025-09-24"],[0,41079,"Intern-S1 not working","Django-Jiang",2,"","",0,"2025-09-23","2025-09-23"],[0,41073,"`use_cache=True` does not work with T5GemmaForConditionalGeneration.generate()","hazemessamm",2,"","",0,"2025-09-22","2025-11-03"],[0,41060,"`do_sample` does not work in Qwen3 Model‘s `generate` method","Hermit-w",2,"","",0,"2025-09-22","2025-09-24"],[0,41044,"[Bug] AutoModelForCausalLM.from_pretrained() causes fatal silent crash on Windows with CUDA 12.9 / RTX 3060 Ti","remby83-boop",2,"","",0,"2025-09-22","2025-10-22"],[0,41043,"`TFBertForMaskedLM` throws `TypeError: 'builtins.safe_open' object is not iterable`","khteh",2,"","",0,"2025-09-22","2025-09-22"],[0,41042,"Zero GPU quota doesn't reset","EvgenyZaretskiy",2,"","",0,"2025-09-22","2025-10-30"],[0,41038,"Method `from_pretrained` does not respect the mapping of tied tensors in `safetensors` checkpoint","daskol",2,"","",0,"2025-09-21","2025-09-29"],[0,41034,"G","ESCAPEEEE111",2,"","",0,"2025-09-21","2025-09-21"],[0,41026,"Attention sinks are not applied correctly in `integrations.flex_attention`","jonny-so",0,"SamuelBarryCS","#41083",0,"2025-09-20","2025-09-29"],[0,41023,"modular_qwen2_5_vl.py冗余代码，影响代码整洁和引入歧义","xiangqian12345678",2,"","",0,"2025-09-20","2025-10-29"],[0,41020,"Config rope_scaling and text_config.rope_scaling might be the same or different dict objects","albertvillanova",0,"zucchini-nlp","#41158",1,"2025-09-20","2025-09-30"],[0,41013,"Runtime errors when trying to call Trainer() on a model that exceeds GPU vRAM","ag-TJNII",0,"The5cheduler","#41032",0,"2025-09-19","2025-09-29"],[0,41005,"Are we have Qwen3VL Official Model Published by Alibaba","Dineshkumar-Anandan-ZS0367",2,"","",0,"2025-09-19","2025-09-20"],[0,40993,"HfArgumentParser cannot parse TRL Config","caoyang-sufe",2,"","",0,"2025-09-19","2025-09-19"],[0,40990,"Extremely high perplexity on openai/gpt-oss-20b with WikiText-2 (raw)","kuantuna",2,"","",0,"2025-09-19","2025-09-22"],[0,40988,"Qwen Model does not free CPU Memory when moving it to GPU","cehongwang",2,"","",0,"2025-09-18","2025-10-27"],[0,40986,"Rewrite audio feature processors to accept torch tensors","jackzhxng",4,"","",0,"2025-09-18",""],[0,40984,"Adding `logits_to_keep` to older models","philiproeleveld",1,"philiproeleveld","#41335",0,"2025-09-18","2025-10-15"],[0,40983,"[`RFC v5`] Encoder-decoder support for Bert-related models and GPT2","vasqu",2,"","",0,"2025-09-18","2025-11-28"],[0,40977,"Whisper Finetuning Issue","AbhijithMallya",0,"tayo4christ","#41027",0,"2025-09-18","2025-09-30"],[0,40972,"Error when running Qwen3_Next Qwen3-Next model","guangshuoqin",2,"","",0,"2025-09-18","2025-11-12"],[0,40970,"Inconsistenly sized batches of images and text (Gemma 3)","KarelKenens",2,"","",0,"2025-09-18","2025-09-18"],[0,40968,"An error occurred when directly deploying the Qwen3-Next 80B-A3B-Thinking model saved by the save_pretrained o","zhanlun150729",2,"","",0,"2025-09-18","2025-11-12"],[0,40964,"llava model compile output regression caused by `check_model_inputs`","jiqing-feng",0,"zucchini-nlp","#40994",1,"2025-09-18","2025-10-06"],[0,40963,"Qwen3-next feature dims are mismatched in torch_chunk_gated_delta_rule","jyC23333",0,"notkisk","#41036",0,"2025-09-18","2025-09-24"],[0,40958,"Variety of models not working sufficiently good on tool calling as has been advertised","dvn8weil",2,"","",0,"2025-09-18","2025-10-26"],[0,40957,"Incomplete tool call arguments in HuggingFaceTB/SmolLM3-3B streaming output","dvn8weil",2,"","",0,"2025-09-18","2025-10-26"],[0,40953,"gpt-oss-120b inference failed running on 16 GPUs, single node and with tp_plan=\"auto\"","yuanhangsu1986",2,"","",0,"2025-09-17","2025-10-26"],[0,40952,"InstructBLIP requiring eager attention specified to work properly","nhatkhtn",2,"","",0,"2025-09-17","2025-09-19"],[0,40946,"Docs (zh-hans): Revert certain terms to English for clarity and professional consistency","Ri-Nai",2,"","",0,"2025-09-17","2025-10-26"],[0,40938,"RFC for `tokenization` in v5","ArthurZucker",2,"","",0,"2025-09-17","2025-11-14"],[0,40933,"`UserWarning: `seed_generator` is deprecated and will be removed in a future version.`","khteh",2,"","",0,"2025-09-17","2025-09-17"],[0,40932,"Inconsistent handling of tokenizer bos_token","fxmarty-amd",2,"","",0,"2025-09-17","2025-10-26"],[0,40927,"PreTrainedTokenizer requires self.get_vocab() which is no longer implemented","eugenekwaNeuromics",2,"","",0,"2025-09-17","2025-10-01"],[0,40926,"Contrastive search doesn't work on Gemma3","jood-canva",2,"","",0,"2025-09-17","2025-09-17"],[0,40915,"HfArgumentParser does not support peft.LoraConfig","romitjain",2,"","",0,"2025-09-16","2025-09-23"],[0,40913,"Setting chat_template when creating a processor does not change the chat template","NohTow",2,"","",0,"2025-09-16","2025-10-26"],[0,40910,"Gemma-3: prepare_inputs_for_generation should forward pixel_values based on image token presence, not cache_po","Simone999",0,"zucchini-nlp","#42088",1,"2025-09-16","2025-12-23"],[0,40904,"MXFP4 Tensor Core GEMM support in GPT-OSS for Blackwell GPUs","TheTinyTeddy",2,"","",0,"2025-09-16","2025-10-26"],[0,40901,"Cannot fine-tune T5Gemma with Seq2SeqTrainer","Crissium",2,"","",0,"2025-09-16","2025-10-20"],[0,40891,"Need help for Applying Visual Prompt Tuning with Qwen2.5-VL vision","davidan208",2,"","",0,"2025-09-15","2025-10-25"],[0,40886,"not able to import Gemma3TextForSequenceClassification on transformers == '4.56.1'","rishavranaut",2,"","",0,"2025-09-15","2025-09-16"],[0,40875,"ColPaliForRetrieval errors out when loaded in half precision dtypes","merveenoyan",0,"zucchini-nlp","#40912",1,"2025-09-14","2025-09-16"],[0,40874,"Missing num_hidden_layers in T5GemmaConfig","kuihao",0,"gante","#40939",0,"2025-09-14","2025-10-01"],[0,40868,"Review and update the Code of Conduct","wiwdep-netizen",2,"","",0,"2025-09-13","2025-09-15"],[0,40867,"bug with AutoVideoProcessor for VJEPA 2","FrancoisPorcher",0,"zucchini-nlp","#41218",1,"2025-09-13","2025-10-01"],[0,40858,"torch.no_grad() yields NaN values on mps device, 4D attention mask","AmitMY",2,"","",0,"2025-09-12","2025-11-11"],[0,40853,"Top Three Models on Object Detection Leaderboard Won't Load on Newest Version","waylonflinn",0,"Cyrilvallez","#40878",1,"2025-09-12","2025-09-15"],[0,40849,"The chat prompt template for google/gemma-3-270m-it omits the system message.","umang-0801",2,"","",0,"2025-09-12","2025-11-17"],[0,40841,"Add Support for Ovis2.5 Multi-Modal Model​","xschen-beb",4,"","",0,"2025-09-12",""],[0,40835,"`generate_batch` failing","qgallouedec",2,"","",0,"2025-09-11","2025-10-15"],[0,40833,"Generate attends to padded tokens when using left-padded (KV) batches","giulio98",2,"","",0,"2025-09-11","2025-09-15"],[0,40830,"`pipeline` support for backends like `vLLM` and `SGLang`","MRiabov",3,"","",0,"2025-09-11","2026-10-20"],[0,40829,"Cannot load config for Qwen/Qwen2.5-32B-Instruct","qiaoruiyt",2,"","",0,"2025-09-11","2025-09-11"],[0,40827,"[`Cache`] Proper support for Linear Attention (related) caches","vasqu",2,"","",0,"2025-09-11","2026-04-09"],[0,40822,"Welcome v5","LysandreJik",2,"","",0,"2025-09-11","2026-03-05"],[0,40817,"Got shape error when running JetMoe","wtomin",0,"vasqu","#40819",1,"2025-09-11","2025-09-11"],[0,40815,"get_decoder feature regression in 4.56.0","KyleMylonakisProtopia",0,"molbap","#40824",1,"2025-09-11","2025-09-16"],[0,40813,"Incorrect sharding configuration for Starcoder2 model","greg-kwasniewski1",1,"greg-kwasniewski1","#40814",0,"2025-09-11","2025-09-15"],[0,40798,"SigLIP2 loss is incomplete. Add the other loss.","SuperHotDogCat",3,"","",0,"2025-09-10","2026-10-20"],[0,40794,"Feature Request: Option to transfer logits to CPU during generation","YunruiZhang",4,"","",0,"2025-09-10",""],[0,40787,"SAM2 - Deteriorated performance compared to original repository","alex-bene",0,"yonigozlan","#40800",0,"2025-09-10","2025-09-11"],[0,40785,"Granite 20B Function Calling model hallucinating the tool call","dvn8weil",2,"","",0,"2025-09-10","2025-10-13"],[0,40781,"Not save processor related files when call save_model","zyandtom",2,"","",0,"2025-09-10","2025-09-10"],[0,40779,"AttributeError: module 'torchcodec' has no attribute 'decoders'","MartinYTSo",2,"","",0,"2025-09-10","2025-09-12"],[0,40776,"Add function for reversing chat templates","LakeYin",2,"","",0,"2025-09-09","2026-10-20"],[0,40770,"Sam2VideoProcessor throws tensor size mismatch when adding multiple input boxes without input_labels","DanForester",0,"yonigozlan","#40797",0,"2025-09-09","2025-09-11"],[0,40768,"max_length ignored in summarization pipeline, overridden by default max_new_tokens","noah-13",0,"lilin-1","#40928",0,"2025-09-09","2025-09-24"],[0,40767,"3D Object Detection Models","SeucheAchat9115",4,"","",0,"2025-09-09",""],[0,40754,"Potentially incorrect value assignment of Llama4TextModel's output in Llama4ForCausalLM's output?","st143575",2,"","",0,"2025-09-08","2025-09-11"],[0,40752,"How to extract attention weights for the first generated token?","VincentLHH",2,"","",0,"2025-09-08","2025-09-08"],[0,40751,"\"max_length\" in pipeline(\"text-generation\", model=\"gpt2\") does not work","wanghao16020510036",2,"","",0,"2025-09-08","2025-09-08"],[0,40750,"Trainer.save_model saves config with FSDP-wrapped architectures instead of base model","JdRion",2,"","",0,"2025-09-08","2025-10-16"],[0,40743,"Support for 4D attention mask for T5","Aethor",2,"","",0,"2025-09-07","2026-10-20"],[0,40739,"num_assistant_tokens not configured in assistant model","geoffrey-young",2,"","",0,"2025-09-06","2025-12-13"],[0,40731,"\"AutoProcessor.from_pretrained\" cannot run in distributed environments after v4.56.0","erikchwang",0,"zucchini-nlp","#40786",1,"2025-09-05","2025-09-17"],[0,40723,"Gemma3 with flash-attention2 outputs warning","tolgadur",0,"August-murr","#40744",0,"2025-09-05","2025-09-08"],[0,40709,"Divide gradient_accumulation_steps repeatedly in Trainer","shenlanyilang",2,"","",0,"2025-09-05","2025-09-12"],[0,40708,"When using a custom model, it copies the code into Hugging Face’s cache directory.","wzr0108",2,"","",0,"2025-09-05","2025-11-15"],[0,40705,"[Potential Issue] Inappropriate handling of falsy values when parameters default to None","ruixCMU",2,"","",0,"2025-09-05","2025-10-13"],[0,40690,"Batches loaded from wrong epoch when resuming from second epoch","ngazagna-qc",1,"ngazagna-qc","#40691",0,"2025-09-04","2025-12-03"],[0,40687,"the continue_final_message parameter in apply_chat_template can behave unexpectedly in some cases","usepr",0,"abdokaseb","#40732",0,"2025-09-04","2025-09-08"],[0,40682,"Performance Regression in Whisper with torch.compile on v4.52.0+ (vs v4.51.0)","baeseongsu",2,"","",0,"2025-09-04","2025-10-12"],[0,40680,"Idea: Exploring Mathematical Extensions for GPT-style Models (teaser)","muzamil-ashiq",2,"","",0,"2025-09-04","2025-10-12"],[0,40678,"updating `dtype` during `from_pretrained()` causes minor issue","kwonmha",2,"","",0,"2025-09-04","2025-09-04"],[0,40674,"Add object detection heads to DINOv3","ryan-minato",4,"","",0,"2025-09-04",""],[0,40672,"Preserve original GGUF dtypes during fp32 conversion","dillondesilva",2,"","",0,"2025-09-03","2025-09-14"],[0,40659,"Automodel.from_pretrained fails with -OO flag","SimonDR-Boltzmann",2,"","",0,"2025-09-03","2025-10-12"],[0,40650,"Apertus - Error while loading model","BlueskyFR",2,"","",0,"2025-09-03","2025-09-03"],[0,40649,"Support for HumanMLLM/HumanOmni","Zha-Miku",3,"","",0,"2025-09-03","2026-11-04"],[0,40647,"how to get response text during training","zyandtom",2,"","",0,"2025-09-03","2025-10-12"],[0,40644,"`BlenderbotForConditionalGeneration` errors out with `list index out of range`","ariG23498",0,"gante","#40772",0,"2025-09-03","2025-09-12"],[0,40642,"Llama-4-Scout-17B-16E-Instruct model perplexity anomaly when transformers==4.55.1","kewang-amd",2,"","",0,"2025-09-03","2025-09-22"],[0,40639,"Add DeepseekV3ForTokenClassificaiton for using as a value model","bzantium",1,"bzantium","#40641",0,"2025-09-03","2025-09-03"],[0,40635,"Safetensors files for long-t5-tglobal models fail to load correctly","pszemraj",0,"Szustarol","#40724",0,"2025-09-03","2025-09-24"],[0,40621,"[Feature Proposal] Add GGUF support for Korean-specific LLM architectures","Kim-Ju-won",4,"","",0,"2025-09-02",""],[0,40620,"Function/Tool Calling with Streaming (transformers serve) Returns Empty Chunks for GPT-OSS-20B","Lanson2232",2,"","",0,"2025-09-02","2025-12-14"],[0,40616,"Allow users to pass through `Accelerate` instance","nialldevlin1",2,"","",0,"2025-09-02","2026-07-29"],[0,40608,"No module named 'transformers.models.gpt'","TPLong2002",2,"","",0,"2025-09-02","2025-10-25"],[0,40606,"GPT-OSS attention backends available for SM120 other than Eager?","TheTinyTeddy",2,"","",0,"2025-09-02","2025-10-12"],[0,40603,"Processing utils bug in `apply_chat_template` when image content is PIL image","ZENGXH",0,"abdokaseb","#40622",0,"2025-09-01","2025-09-03"],[0,40601,"Custom loss function computation even when training without labels.","lrast",0,"Flakes342","#40632",0,"2025-09-01","2025-09-22"],[0,40582,"feat(deepseek_v3): Add grouped GEMM kernel for faster MoE computation","bzantium",2,"","",0,"2025-09-01","2026-01-08"],[0,40577,"Llava-next image processor issue","EliSchwartz",0,"zucchini-nlp","#40588",1,"2025-08-31","2025-09-01"],[0,40572,"transformers 4.21 AutoTokenizer.from_pretrained issue","gda2048",2,"","",0,"2025-08-31","2025-09-01"],[0,40571,"Responses API does not support input in (complex) format","adner",3,"","",0,"2025-08-31","2025-09-30"],[0,40564,"Loss Scales Incorrectly with Gradient Accumulation Steps in Trainer (Gemma3 and other models)","yevvonlim",1,"yevvonlim","#40591",0,"2025-08-30","2025-09-03"],[0,40560,"`train_tokens_per_second` is wrong after continuing from checkpoint","AmitMY",0,"lilin-1","#41156",0,"2025-08-30","2025-09-29"],[0,40559,"Type resolution failed for <class '__main__.TrainingArguments'>. Try declaring the class in global scope or re","Asunatan",2,"","",0,"2025-08-30","2025-09-01"],[0,40554,"Add sequence classification support for small Gemma 3 models","adamreichold",0,"abdokaseb","#40562",0,"2025-08-29","2025-09-04"],[0,40533,"Smart Hyperparameter Search for PEFT","PamelaBha",2,"","",0,"2025-08-29","2025-08-29"],[0,40531,"Getting AttributeError: 'Mxfp4GptOssExperts' object has no attribute 'down_proj_scales'. When trying to load t","eltonjohnfanboy",2,"","",0,"2025-08-28","2025-10-16"],[0,40522,"RuntimeError with use_remove_padding=True in GRPO training for Qwen2.5-VL due to tensor shape mismatch in tran","zhihaofang1017",2,"","",0,"2025-08-28","2025-09-30"],[0,40513,".from_pretrained() auto converts model to fp32, leads to issue with flash attention 2","KeshavSingh29",2,"","",0,"2025-08-28","2025-08-28"],[0,40512,"Context-Aware Tokenizer Suggestions","PamelaBha",3,"","",0,"2025-08-28","2026-10-20"],[0,40504,"gpt-oss and transformers serve not working","SunMarc",2,"","",0,"2025-08-27","2025-11-03"],[0,40500,"torch.load fails to load RNG state in PyTorch 2.6+ due to weights_only=True and missing safe globals","rangehow",2,"","",0,"2025-08-27","2025-08-27"],[0,40496,"AutoModel.from_pretrained() doesn't work for models with '.' in their name when there's a relative import","zachmoshe",0,"August-murr","#40745",0,"2025-08-27","2025-09-10"],[0,40488,"Voxtral model fails with LoRA due to in-place operation error","juzhxng",0,"rcogill","#40671",0,"2025-08-27","2025-09-04"],[0,40487,"Mutex while building docs","Uvi-12",2,"","",0,"2025-08-27","2025-08-28"],[0,40485,"KeyError: 'architectures' when  `trust_remote_code=True`","Twinkle-ce",2,"","",0,"2025-08-27","2025-08-27"],[0,40478,"Typo: should be `feature_extractor`, not `feature_extracttor`","Guo-Chenxu",1,"Guo-Chenxu","#40484",0,"2025-08-27","2025-08-28"],[0,40477,"AutoModelForVision2Seq.from_config() got an unexpected keyword argument 'torch_dtype'","ljeff97",0,"zucchini-nlp","#40491",1,"2025-08-27","2025-09-03"],[0,40476,"Qwen2.5Omni Repeatedly Calculates Audio/Video/Image Features","CharlesGong12",2,"","",0,"2025-08-27","2025-09-01"],[0,40475,"InternVL 3.5 30B A3B load processor bug","yilian49",2,"","",0,"2025-08-27","2025-10-31"],[0,40468,"ImportError: cannot import name 'GptOssForSequenceClassification' from 'transformers'","pervaizniazi",2,"","",0,"2025-08-26","2025-08-28"],[0,40463,"Number of trainable params is displayed wrong in accelerate logs when using FSDP","Ali-Sayed-Salehi",2,"","",0,"2025-08-26","2025-10-04"],[0,40462,"Question about RoPE Implementation in modeling_llama: Should torch.cat be repeat_interleave?","abhidipbhattacharyya",2,"","",0,"2025-08-26","2025-08-27"],[0,40461,"AttributeError: 'GptOssConfig' object has no attribute 'max_position_embeddings'","speedbunny",0,"zifeitong","#40474",0,"2025-08-26","2025-08-27"],[0,40460,"Gemma-3-12b-pt high grad_norm during continual pre-training","marseller",2,"","",0,"2025-08-26","2025-10-03"],[0,40459,"`use_kernels=True` does not invoke custom kernels","ariG23498",0,"MekkCyber","#40628",0,"2025-08-26","2025-09-16"],[0,40457,"Image Embedding Models (Feature extractors) should have a `.hidden_size`","AmitMY",2,"","",0,"2025-08-26","2026-10-23"],[0,40450,"Add Seed OSS Model","MengAiDev",2,"","",0,"2025-08-26","2025-08-26"],[0,40447,"Processor does not load optional attributes","AmitMY",0,"zucchini-nlp","#41474",1,"2025-08-26","2025-10-16"],[0,40444,"Finetuning Qwen2.5-VL with an IterableDataset with multiple images per prompt fails","Infernaught",2,"","",0,"2025-08-25","2025-10-04"],[0,40422,"[Feature] Add new SOTA backbone for Vision as MetaFormer series","Bili-Sakura",2,"","",0,"2025-08-25","2025-08-27"],[0,40418,"Javanese language code is `jv`, not `jw` if based on ISO 639-1","fahadh4ilyas",2,"","",0,"2025-08-25","2025-10-02"],[0,40415,"Unexpected behavior when modifying Linear.forward with device_map=\"auto\" offloading","huangwei021230",2,"","",0,"2025-08-25","2025-10-02"],[0,40410,"Cannot get per-token scores in each beam in each step using beam search.","bagasjiwanta",2,"","",0,"2025-08-24","2025-08-29"],[0,40409,"AcceleratorError: CUDA error: device-side assert triggered","monochandan",2,"","",0,"2025-08-24","2025-08-25"],[0,40406,"Cache tokenlizer","axymeus",2,"","",0,"2025-08-24","2026-07-22"],[0,40401,"`num_input_tokens_seen` should exclude padding tokens","RylanSchaeffer",0,"PrathmeshAdsod","#40416",0,"2025-08-23","2025-09-11"],[0,40399,"[API] Current `query_length` API can break ulysses-sp patch","ETOgaosion",0,"vasqu","#40424",1,"2025-08-23","2025-08-28"],[0,40398,"NVIDIA RADIO-L","Uvi-12",2,"","",0,"2025-08-23","2026-07-23"],[0,40382,"Cannot resize token embeddings in distributed training setups with FSDP","Ali-Sayed-Salehi",2,"","",0,"2025-08-22","2025-08-24"],[0,40376,"Training Arguments resets accelerator","lcxrocks",2,"","",0,"2025-08-22","2025-08-28"],[0,40371,"Typo: \"casual\" should be \"causal\" in code and docs","vamshika-0210",0,"akintunero","#40407",0,"2025-08-22","2025-08-25"],[0,40366,"[Feature] Support fromjson in jinja2 chat template rendering","byjiang1996",3,"","",0,"2025-08-22","2026-08-19"],[0,40362,"Registering attention fn with AttentionInterface changes its computation, potentially loses mask","tplr-y",2,"","",0,"2025-08-21","2025-09-02"],[0,40361,"Quantized model loading failing with custom device mesh","radulescupetru",2,"","",0,"2025-08-21","2025-09-29"],[0,40359,"Llama4UnfoldConvolution possible bug","gskorokhod",2,"","",0,"2025-08-21","2025-08-21"],[0,40357,"AutoModel failed with empty tensor error","bluestealth",2,"","",0,"2025-08-21","2025-09-23"],[0,40349,"Typo and GPU kernel check bug in MXFP4 quantization (v4.55.3)","HeenaRajan",0,"akintunero","#40408",0,"2025-08-21","2025-08-25"],[0,40345,"Flexattention Implementation in Gemma Failing on CPU","amd-lalithnc",2,"","",0,"2025-08-21","2025-08-22"],[0,40335,"repetitio/flan-t5-small is not longer exists","jiqing-feng",2,"","",0,"2025-08-21","2025-09-01"],[0,40334,"It takes too long time to load GGUF tokenizer","jiqing-feng",0,"nayana1729","#40509",0,"2025-08-21","2025-08-28"],[0,40330,"open-qwen2vl-base","olccihyeon",3,"","",0,"2025-08-21","2026-07-19"],[0,40327,"ValueError when loading 4-bit quantized models with sharded weights on Transformers >=4.54.0","rice8y",2,"","",0,"2025-08-20","2025-10-06"],[0,40323,"Is there a plan to add DINOv3 into AutoBackbone?","Farenweh",0,"vijayabhaskar-ev","#41276",0,"2025-08-20","2025-11-11"],[0,40318,"TypeError: 'builtins.safe_open' object is not iterable / ValueError: Exception encountered when calling layer ","ngvananh2508",2,"","",0,"2025-08-20","2025-08-22"],[0,40303,"Add mean_token_accuracy from TRL's `SFTTrainer` to `Trainer`","RylanSchaeffer",3,"","",0,"2025-08-20","2026-08-20"],[0,40302,"Add audio-text-to-text task","jackzhxng",0,"zucchini-nlp","#40884",1,"2025-08-19","2025-11-27"],[0,40301,"Label smoothing for multi-label classification","andreabosisio",4,"","",0,"2025-08-19",""],[0,40293,"main seems to break the FA2 interface, ok with 4.55.2","sfc-gh-sbekman",0,"vasqu","#40295",1,"2025-08-19","2025-08-20"],[0,40292,"transformers==4.55.2 still imports TFPreTrainedModel even with TRANSFORMERS_NO_TF=1\"","Yanshuptl18",0,"id01","#40439",0,"2025-08-19","2025-08-25"],[0,40278,"[BUG] GPT-OSS weight dequantization discrepancy for online and offline methods","TheTinyTeddy",2,"","",0,"2025-08-19","2025-08-28"],[0,40268,"Some csm testcases fail because of incompatible with torch_comiple","yuanwu2017",2,"","",0,"2025-08-19","2025-08-20"],[0,40266,"PretrainedConfig `from_pretrained` does not maintain `save_pretrained` values with `sub_configs`","AmitMY",2,"","",0,"2025-08-19","2025-08-19"],[0,40264,"Model loads with inconsistent `dtype` based on `from_pretrained` and `from_config`","AmitMY",2,"","",0,"2025-08-19","2025-08-19"],[0,40263,"[VLMs] How to process a batch that contains samples with and without images?","qgallouedec",2,"","",0,"2025-08-19","2025-09-18"],[0,40262,"v4.55.3 tag miss some codes for patch, affect FA2 on Ascend NPU","FightingZhen",2,"","",0,"2025-08-19","2025-08-21"],[0,40260,"Add dynamic shapes to ExecuTorch static cache export tests","jackzhxng",4,"","",0,"2025-08-18",""],[0,40258,"RuntimeError in loss computation with label_smoothing_factor > 0 for multi-label classification","andreabosisio",0,"avchauzov","#40296",0,"2025-08-18","2025-08-25"],[0,40255,"SAM decoder training error","alexdzm",0,"yonigozlan","#40667",0,"2025-08-18","2025-09-04"],[0,40253,"Wrong attention weights conversion for D-FINE","materight",0,"notkisk","#40257",0,"2025-08-18","2025-08-19"],[0,40251,"PerceptionLM Image Processor doesn't properly allow for vision_input_type override","tyleryzhu",1,"tyleryzhu","#40252",0,"2025-08-18","2025-08-19"],[0,40246,"MT5ForConditionalGeneration no longer importable after torch and transformers upgrade","MarcoDiMarek",2,"","",0,"2025-08-18","2025-08-18"],[0,40245,"When elastic training is enabled, data may be lost or over-read.","zhengchenyu",2,"","",0,"2025-08-18","2025-10-05"],[0,40242,"All versions of Transformers after 4.52.4 break batch inference on MPS when padding_side = left with SmolVLM","Matteo-Omenetti",2,"","",0,"2025-08-18","2025-09-29"],[0,40239,"Please remove the redundant dependency jieba: rjieba does the same and has better performance","yurivict",0,"divyanshsinghvi","#40383",0,"2025-08-18","2025-08-30"],[0,40236,"Training support for the MXFP4 Quantization Method","InfiniteInbox",3,"","",0,"2025-08-18","2025-08-18"],[0,40235,"Type annotation error in DepthPro model image processor","soyosaki",0,"MengAiDev","#40238",0,"2025-08-18","2025-08-18"],[0,40231,"pytest-* dependencies shouldn't be run-time dependencies","yurivict",2,"","",0,"2025-08-17","2025-08-18"],[0,40230,"Transfomers chat cli doesnt load on gpu and breaks on second message","Mcilie",0,"robin-ede","#40232",0,"2025-08-17","2025-08-19"],[0,40229,"Typo in the docs/README.md","PavloFesenko",1,"PavloFesenko","#40234",0,"2025-08-17","2025-08-18"],[0,40227,"The 'dataclasses' dependency should be marked for python<3.7","yurivict",2,"","",0,"2025-08-17","2025-08-18"],[0,40224,"Gemma RMSNorm weight init should init with 0","albertz",2,"","",0,"2025-08-16","2025-09-30"],[0,40219,"Non-deterministic Trainer behavior","AmitMY",2,"","",0,"2025-08-16","2025-09-24"],[0,40217,"metric_for_best_model set to eval_f1 raises KeyError because metric is not found in evaluation results","monochandan",0,"huzaifa-jawad367","#40353",0,"2025-08-16","2025-08-22"],[0,40214,"Chunked Attention with left padding causes train–inference mismatch (first chunk truncated)","xiazhuo",0,"Cyrilvallez","#40324",1,"2025-08-16","2025-08-21"],[0,40202,"Transformer serve gpt oss 20b cannot support multi-turn chat","Taimin",2,"","",0,"2025-08-15","2025-09-29"],[0,40199,"Add Deepseek v3 For Sequence Classification","AhmedHashish123",0,"abdokaseb","#40200",0,"2025-08-15","2025-08-21"],[0,40193,"`_flash_attention_forward` can't receive `cumulative_seqlens_q` and `cumulative_seqlens_k` in TransformersKwar","Kurt232",1,"Kurt232","#40194",0,"2025-08-15","2025-08-25"],[0,40189,"Support token classification with GPT-OSS models","abdokaseb",1,"abdokaseb","#40190",0,"2025-08-15","2025-08-22"],[0,40188,"occur bugs 'CONFIG_MAPPING[vision_config[\"model_type\"]] KeyError: 'vision'' when using deepseek-vl pipeline","WongGawa",2,"","",0,"2025-08-15","2025-09-23"],[0,40187,"load_sharded_checkpoint doesn't abide by _checkpoint_conversion_mapping, causing issues when resuming from che","reubenvanammers",2,"","",0,"2025-08-15","2025-09-22"],[0,40186,"nccl time out","nicefuture2016",2,"","",0,"2025-08-15","2025-08-20"],[0,40183,"5090 - Random triton kernel crashes depending on batch size on MXFP4 (gpt-oss)","cmp-nct",2,"","",0,"2025-08-14","2025-09-25"],[0,40179,"[docs] Running jobs with SkyPilot","alex000kim",3,"","",0,"2025-08-14","2026-10-20"],[0,40170,"Add MXFP4 MoE/attention backward kernels","Rocketknight1",4,"","",0,"2025-08-14",""],[0,40168,"EncoderDecoderModel emits FutureWarning about v4.12 training path when labels are passed (v4.48.1, „version v4","leaBroe",2,"","",0,"2025-08-14","2025-10-20"],[0,40166,"Error after calling ESM model with input embeddings not input ids","hunklinger",0,"notkisk","#40181",0,"2025-08-14","2025-08-18"],[0,40154,"Qwen2.5VL is broken!","pengzhenghao",0,"zucchini-nlp","#40490",1,"2025-08-14","2025-09-01"],[0,40153,"Incomplete Function calling arguments in Qwen3-8B model","dvn8weil",2,"","",0,"2025-08-14","2025-09-21"],[0,40145,"Vision Transformer with Mixture of Experts (ViT-MoE)","LawJarp-A",2,"","",0,"2025-08-13","2025-08-19"],[0,40136,"Qwen2.5-VL-7B-Instruct: Significant accuracy regression on MMMU benchmark with transformers >=4.54.0","rahul-tuli",0,"zucchini-nlp","#40490",1,"2025-08-13","2025-09-01"],[0,40126,"slide attention bug for qwen3 model","Yufei-Z",2,"","",0,"2025-08-13","2025-09-21"],[0,40125,"Truncation 'only_second' does not work for multimodal chat templates.","karol-szustakowski",2,"","",0,"2025-08-13","2025-09-21"],[0,40122,"wrong onnx output of OPUS en-ar","logicvv",2,"","",0,"2025-08-13","2025-11-17"],[0,40118,"MT5: UnboundLocalError","kimihailv",2,"","",0,"2025-08-12","2025-08-18"],[0,40116,"SmolVLM RuntimeError Expected all tensors to be on the same device, but found at least two devices","sayandipdutta",2,"","",0,"2025-08-12","2025-08-13"],[0,40111,"Wrong loss computation for VisionEncoderDecoderModel","tcnguyen",2,"","",0,"2025-08-12","2025-11-03"],[0,40110,"Upgrading transformers from 4.51.3 to 4.55 makes GRPOTrainer fail training when used with accelerate","Randomdude11",2,"","",0,"2025-08-12","2025-09-21"],[0,40105,"Docs: Add missing `eli5` dataset example in Language Modeling guide","sdivyanshu90",2,"","",0,"2025-08-12","2025-09-21"],[0,40104,"router_logits not getting populated on  output_router_logits=True for GptOssForCausalLM","qrdlgit",2,"","",0,"2025-08-12","2025-10-15"],[0,40101,"[WIP] Add PLDR-LLM","burcgokden",2,"","",0,"2025-08-12","2025-09-01"],[0,40099,"QuantoQuantizedCache is not working","maxjeblick",0,"manueldeprada","#40109",0,"2025-08-12","2025-08-13"],[0,40098,"Different outputs between using input_embeds and using input_ids","HCRTY",2,"","",0,"2025-08-12","2025-11-01"],[0,40097,"Wrong function signature introduced in PR #40029 breaking flash attention","maxjeblick",2,"","",0,"2025-08-12","2025-08-12"],[0,40089,"Could not import module 'AutoTokenizer'. Are this object's requirements defined correctly?","octavianBordeanu",2,"","",0,"2025-08-11","2025-08-12"],[0,40088,"Default behavior of llama tokenizers breaks text by removing spaces (round trip is not identity function)","keenanpepper",2,"","",0,"2025-08-11","2025-08-13"],[0,40079,"TypeError in DogeDecoderLayer with MoE Configuration when using dropout()","LoserCheems",2,"","",0,"2025-08-11","2025-09-19"],[0,40073,"gpt_oss inference activates *all* experts for every token","shivak",2,"","",0,"2025-08-11","2025-09-19"],[0,40071,"Issue running model from ImageSegmentationPipeline","LuSrodri",2,"","",0,"2025-08-10","2025-09-19"],[0,40070,"Transformer GGUF support philosophy / naive question","luke14free",4,"","",0,"2025-08-10",""],[0,40067,"[BUG] No umt5 config for GGUF. This is not supported configuration.","SlimRG",0,"akshay-babbar","#40218",0,"2025-08-10","2025-09-17"],[0,40062,"[Mistral3] attn_implementation not applied to vision_tower.config in Mistral3Config due to init order","starcatmeow",1,"starcatmeow","#40063",0,"2025-08-10","2025-08-11"],[0,40056,"Question: How to write a custome tokenizer form scratch","obadx",2,"","",0,"2025-08-09","2025-09-24"],[0,40054,"Whisper transcription accuracy improves when last 1600 samples of input audio are muted","jozefchutka",2,"","",0,"2025-08-09","2026-01-19"],[0,40052,"Previous PRs introduced a bug on Accumulated Gradients Losses","w32zhong",2,"","",0,"2025-08-09","2025-08-09"],[0,40050,"Support text classification with GPT-OSS models","zyfedward",2,"","",0,"2025-08-09","2025-08-14"],[0,40049,"Please support loading Qwen 2.5 VL from GGUF","ihendley",3,"","",0,"2025-08-09","2026-10-21"],[0,40046,"Recent releases break backwards-compatibility with key_cache","ntenenz",2,"","",0,"2025-08-08","2025-09-17"],[0,40042,"Support loading glm4moe GGUF","adonishong",4,"","",0,"2025-08-08",""],[0,40040,"Error when loading gguf file","ZJY0516",2,"","",0,"2025-08-08","2025-08-09"],[0,40034,"`plamo-2-1b` broken on latest main","tdoublep",2,"","",0,"2025-08-08","2025-09-02"],[0,40032,"Add Padding Strategy to DataCollatorForLanguageModeling","rjgleaton",2,"","",0,"2025-08-08","2026-06-17"],[0,40031,"[gpt-oss] MoE routing bug in the mxfp4 implementation (in distributed setting)","kitft",0,"ArthurZucker","#40283",1,"2025-08-08","2025-08-19"],[0,40028,"`TypeError: 'builtins.safe_open' object is not iterable` in `load_pytorch_state_dict_in_tf2_model `","harupy",2,"","",0,"2025-08-08","2025-08-08"],[0,40020,"accelerate==1.10.0 and safetensors==0.6.1 are incompatible with transformers==4.53.1","AniruddhaHumane",2,"","",0,"2025-08-08","2025-09-15"],[0,40018,"need GptOssForSequenceClassification","cold-eye",2,"","",0,"2025-08-08","2025-08-19"],[0,40017,"Major issues with transformers version causing rubbish generations with Gemma3 family using vllm","AbdelrahmanHagrass",0,"hmellor","#21927",1,"2025-08-07","2025-08-08"],[0,40010,"Customizable Logit Warping Strategies for Generation","PamelaBha",3,"","",0,"2025-08-07","2026-10-20"],[0,40004,"video_inputs are not passed to perception_lm","4g",2,"","",0,"2025-08-07","2025-08-07"],[0,40001,"Possible wrong init call","zhizhongli-sony",2,"","",0,"2025-08-07","2025-10-05"],[0,39992,"[gpt-oss] Transform checkpoint from safetensors to state dict","fingertap",2,"","",0,"2025-08-07","2025-09-15"],[0,39991,"Gemma3n get_placeholder_mask issue","Znerual",2,"","",0,"2025-08-07","2025-08-08"],[0,39985,"Triton version check compatibility on windows","Tsumugii24",1,"Tsumugii24","#39986",0,"2025-08-07","2025-08-11"],[0,39983,"CVE fix for v4.37.2 and v4.38.0","Aman-Surkar",2,"","",0,"2025-08-07","2025-09-15"],[0,39982,"flash-attn cannot perform deterministic computation","Ju-si-yuan",2,"","",0,"2025-08-07","2025-08-08"],[0,39977,"FSDP2 not compatible with transformers >= 4.54.0 GenericForTokenClassification","ETOgaosion",0,"Cyrilvallez","#40191",1,"2025-08-07","2025-08-15"],[0,39974,"bug in new transformers: 'Florence2ForConditionalGeneration' object has no attribute '_supports_sdpa'","pseudotensor",2,"","",0,"2025-08-07","2025-09-01"],[0,39972,"Gemma3 with fp16 in inference (I don't know if this change is working in fine-tune) #BUG FIX","DGTell",2,"","",0,"2025-08-06","2025-09-08"],[0,39969,"Finetune `gpt-oss-20b` with `mxfp4` quantization","eliotjones1",2,"","",0,"2025-08-06","2025-08-06"],[0,39966,"`convert_deepseek_vl_weights_to_hf.py` not included in v4.55.0 release.","rasmi",0,"geetu040","#40003",0,"2025-08-06","2025-08-07"],[0,39963,"change `dataloader_persistent_workers` default value to `True`","farbodbj",2,"","",0,"2025-08-06","2025-11-03"],[0,39961,"Calling `trainer.evaluate()` before `trainer.train()` with FSDP 2 raises `ValueError: When using FSDP2, a mode","RonanFR",2,"","",0,"2025-08-06","2025-09-14"],[0,39958,"TypeError: Received a NoneType for argument video_processor, but a BaseVideoProcessor was expected.(this issue","2ayush2",3,"","",0,"2025-08-06","2026-07-23"],[0,39955,"Fix grammatically incorrect variable name \"expert_hitted\" → \"expert_hit\" in MoE implementation","Mihonarium",1,"Mihonarium","#39959",0,"2025-08-06","2025-08-06"],[0,39954,"[gpt‑oss] eager_attention_forward not using sliding-window attention for GPT‑OSS models","AlfredTino",2,"","",0,"2025-08-06","2025-08-07"],[0,39947,"v4.55.0 Idefics3 RuntimeError Tensors on different devices","noahleegithub",0,"zucchini-nlp","#39981",1,"2025-08-06","2025-08-07"],[0,39946,"Retaining computational graph after using AutoImageProcessor","YinniKun",2,"","",0,"2025-08-06","2026-10-20"],[0,39945,"GPT-OSS mxfp4 with triton_kernel: make_default_matmul_mxfp4_w_layout not found","yilian49",2,"","",0,"2025-08-06","2025-09-18"],[0,39943,"Breaking change in unset `_tp_plan` attribute","rishub-tamirisa",1,"rishub-tamirisa","#39944",0,"2025-08-06","2025-08-20"],[0,39939,"AttributeError: 'BitsAndBytesConfig' object has no attribute 'get_loading_attributes' with transformers 4.55.0","yukiharada1228",2,"","",0,"2025-08-05","2025-08-08"],[0,39935,"Still getting \"fp16 mixed precision requires a GPU (not 'mps').\" error","renet10",2,"","",0,"2025-08-05","2025-09-08"],[0,39934,"ModernBertUnpaddedRotaryEmbedding __init__ error","Darejkal",2,"","",0,"2025-08-05","2025-08-07"],[0,39932,"transformers serve doesn't handle OPTIONS http method","alew3",2,"","",0,"2025-08-05","2025-08-06"],[0,39921,"[Gemma3N] Not able to add new special tokens to model/tokenizer due to projection error","debasisdwivedy",2,"","",0,"2025-08-05","2026-09-03"],[0,39916,"When using batch_eval_metrics, inputs are not gathered from different device, which is wrong behavior","yuanyifang",2,"","",0,"2025-08-05","2025-10-12"],[0,39914,"Idefics 3: shape mismatch: value tensor of shape [256, 576] cannot be broadcast to indexing result of shape [1","ezyzzy",2,"","",0,"2025-08-05","2025-09-13"],[0,39911,"[Gemma3N] Audio processing issue","NickLucche",0,"Isotr0py","#39919",0,"2025-08-05","2025-08-07"],[0,39910,"Question: Llama4 weight reshaping","gskorokhod",2,"","",0,"2025-08-05","2025-08-13"],[0,39907,"Hidden torchvision>=0.19.0 dependency results in quiet import failures of e.g. PreTrainedModel","tomaarsen",0,"yonigozlan","#39928",0,"2025-08-05","2025-08-13"],[0,39900,"Weights not tied when loading `from_pretrained` with a wrapped model","bryant1410",0,"Cyrilvallez","#39996",1,"2025-08-05","2025-08-08"],[0,39896,"v4.54.1 average_tokens_across_devices=True would cause \"ValueError: Tensors must be CUDA and dense\" when gathe","chiquitita-101",0,"ssharpe42","#40773",0,"2025-08-04","2025-09-10"],[0,39893,"Add VideoPrism","MHRDYN7",1,"MHRDYN7","#39895",0,"2025-08-04","2026-06-19"],[0,39882,"[Feature Request] Automatically log parallelism configuration from Accelerate to W&B","WoosungMyung",2,"","",0,"2025-08-04","2026-10-20"],[0,39878,"Misleading WandB error when WANDB_DISABLED=True and report_to=\"wandb\" are both set","RayaneA7",0,"notkisk","#39891",0,"2025-08-04","2025-08-05"],[0,39877,"ValueError: Max cache length is not consistent across layers","TheTharz",2,"","",0,"2025-08-03","2025-08-04"],[0,39873,"Checking for additional_chat_templates doesn't work without internet (ConnectionError)","akug",1,"akug","#39874",0,"2025-08-03","2025-08-18"],[0,39872,"InternVL, PerceptionLM inference freeze in 4.54.1","iglaweb",2,"","",0,"2025-08-03","2025-09-12"],[0,39871,"model.generate custom encoder and decoder outputs/inputs","agu18dec",2,"","",0,"2025-08-03","2025-08-03"],[0,39868,"Tensor parallelism for GLM-4.5","fernandaspets",3,"","",0,"2025-08-02","2026-08-26"],[0,39864,"454545","qiansheniwang-design",2,"","",0,"2025-08-02","2025-08-06"],[0,39860,"Florence2ForConditionalGeneration does not support Flash Attention 2.0 yet ?...","MathDC99",2,"","",0,"2025-08-02","2025-09-01"],[0,39853,"`make fixup` can't find PLC1802","jackzhxng",2,"","",0,"2025-08-01","2025-09-09"],[0,39852,"Inconsistent Function calling behaviour by Mistral-7B-Instruct-v0.3","dvn8weil",2,"","",0,"2025-08-01","2025-09-09"],[0,39850,"Support topNSigma sampling in `generate`","pramodith",2,"","",0,"2025-08-01","2026-04-02"],[0,39849,"Accelerate seems to default mixed precision to bf16 when passing a DeepSpeed config.","alexge233",2,"","",0,"2025-08-01","2025-09-09"],[0,39842,"Expected behavior of `compute_result` is hard to expect and inconsistent","MilkClouds",2,"","",0,"2025-08-01","2025-10-04"],[0,39841,"MistralCommonTokenizer does not match PreTrainedTokenizer","Fhrozen",2,"","",0,"2025-08-01","2025-11-23"],[0,39839,"pack_image_features RuntimeError when vision_feature_select_strategy=\"full\"","llnnnnnn",2,"","",0,"2025-08-01","2025-09-08"],[0,39835,"Crash when running Llama4 on transformers-4.54.1","IKACE",2,"","",0,"2025-08-01","2025-08-15"],[0,39834,"Allow extra outputs from `GenerationMixin.generate`","jood-canva",4,"","",0,"2025-08-01",""],[0,39833,"Tool-Calling Model (ToolACE-2-Llama-3.1-8B) Responds with Irrelevant Tool message on General Question","dvn8weil",2,"","",0,"2025-07-31","2025-08-01"],[0,39818,"Qwen2-VL err","Yan0613",2,"","",0,"2025-07-31","2025-07-31"],[0,39814,"Flash Attention fails with non aligned position_ids","alessiodevoto",0,"zucchini-nlp","#39843",1,"2025-07-31","2025-08-07"],[0,39812,"Why `lm-head` weight still exists with `\"tie_word_embeddings\": true`","Kelvinlby",2,"","",0,"2025-07-31","2025-08-04"],[0,39811,"Missing einops dependency causing ModuleNotFoundError","iforgetmyname",2,"","",0,"2025-07-31","2025-08-12"],[0,39804,"Fine tuning qwen2.5 error","shaojun0",2,"","",0,"2025-07-31","2025-10-03"],[0,39803,"Memory leak occurred during training qwen-2.5-vl","shaojun0",2,"","",0,"2025-07-31","2025-09-16"],[0,39801,"ValueError: This model does not support cache_implementation='static'. Please check the following issue: https","jpitalopez",2,"","",0,"2025-07-30","2025-09-07"],[0,39798,"You current version of `autoawq` does not support module quantization skipping, please upgrade `autoawq` packa","bi1101",2,"","",0,"2025-07-30","2025-08-26"],[0,39795,"Regression - High memory usage when using transformers model with FSDP + LoRA","romitjain",1,"romitjain","#40075",0,"2025-07-30","2025-08-19"],[0,39791,"`transformers serve` Fails to Handle Messages with Nested Content","jakeret",0,"gante","#39829",0,"2025-07-30","2025-08-13"],[0,39789,"ViTPose+ models post processing doest not work for `dataset_index : 5`","testdummyvt",2,"","",0,"2025-07-30","2025-09-07"],[0,39787,"\"CSM audio generation lacks reliable EOS: does not generate all-zero frames → never stops early\"","sergiuxorga",2,"","",0,"2025-07-30","2025-10-06"],[0,39780,"pip install 'transformers[torch]' pulls nvidia dependencies","marcindulak",0,"stevhliu","#40631",1,"2025-07-30","2025-09-09"],[0,39779,"transformers env fails with: ModuleNotFoundError: No module named 'PIL'","marcindulak",0,"gante","#39825",0,"2025-07-30","2025-08-18"],[0,39776,"BioGPT Implementation Bug Report","SunnyThakur25",2,"","",0,"2025-07-30","2025-07-30"],[0,39775,"Granite 4.0 Tiny Preview inference broken in","nataxcan",2,"","",0,"2025-07-30","2025-08-28"],[0,39774,"Blip model got performance regression on compile mode after refactor cache.","jiqing-feng",0,"zucchini-nlp","#39788",1,"2025-07-30","2025-07-30"],[0,39771,"would it be possible to standardize on the vx.y.z format for all tags","hubutui",2,"","",0,"2025-07-30","2025-09-07"],[0,39767,"Model with non-string type property tool giving incomplete response using VLLM","anyon17",2,"","",0,"2025-07-29","2025-09-07"],[0,39763,"Instantiating `google/gemma-3-4b-pt` with AutoModelForSequenceClassification Reports Unitialized Model","pks",0,"zucchini-nlp","#39778",1,"2025-07-29","2025-08-11"],[0,39755,"Follow-up on Issues Regarding Training State Restoration from Interruptions","rangehow",2,"","",0,"2025-07-29","2025-09-28"],[0,39753,"Inv frequency has not default, going against our philosophy","ArthurZucker",2,"","",0,"2025-07-29","2026-10-20"],[0,39749,"Qwen2_5_VLForConditionalGeneration cfg forward twice error","guozhiyao",2,"","",0,"2025-07-29","2025-09-07"],[0,39747,"ModernBERT has been totally destroyed by PR #38974 and #38838","rangehow",0,"zucchini-nlp","#39750",1,"2025-07-29","2025-07-29"],[0,39746,"encoder decoder model compile failed after refactor cache","jiqing-feng",0,"zucchini-nlp","#39754",1,"2025-07-29","2025-07-29"],[0,39744,"_supports_static_cache disappear","jiqing-feng",2,"","",0,"2025-07-29","2025-07-29"],[0,39736,"losses, logits, labels = self.prediction_step(model, inputs, prediction_loss_only, ignore_keys=ignore_keys) ou","Lmmfff",2,"","",0,"2025-07-28","2025-09-06"],[0,39730,"device mismatch error when using `SlidingWindowLayer`.","nuxlear",0,"Cyrilvallez","#39732",1,"2025-07-28","2025-07-28"],[0,39723,"`num_beams` > 1 leads to exception for Qwen2.5VL (Qwen family or all VLM models?)","iglaweb",0,"zucchini-nlp","#39726",1,"2025-07-28","2025-08-11"],[0,39721,"Support loading Qwen3 MoE GGUF","ctcanbol",1,"ctcanbol","#39638",0,"2025-07-28","2025-07-29"],[0,39720,"[transformers==4.54.0] FSDP1 forward misalignment after loading state dict","ETOgaosion",2,"","",0,"2025-07-28","2025-10-12"],[0,39719,"vlmm 0.10.0 load  baidu/ERNIE-4.5-300B-A47B-Base-PT  error","lianghao6",2,"","",0,"2025-07-28","2025-07-28"],[0,39716,"[rank0]: ValueError: Your setup doesn't support bf16/gpu.","kadirnar",2,"","",0,"2025-07-27","2025-07-31"],[0,39711,"Max cache length issue with Gemma 3","mitchelldehaven",0,"manueldeprada","#39737",0,"2025-07-27","2025-07-29"],[0,39710,"OWLv2 with visual prompt - alternative query embedding selection method","vvmnnnkv",0,"qgallouedec","#49200",1,"2025-07-27","2026-10-02"],[0,39705,"[i18n-<bn>] Translating docs to <Bengali>","ankitdutta428",3,"","",0,"2025-07-27","2026-10-20"],[0,39704,"Iwin Transformer: Hierarchical Vision Transformer using Interleaved Windows","Cominder",2,"","",0,"2025-07-27","2025-08-24"],[0,39703,"ValueError: Number of image placeholders in the prompt does not match the number of images. internVL3","hexiao0275",2,"","",0,"2025-07-26","2025-09-07"],[0,39699,"No flag to support Conditional Parameter Loading for gemma-3n-E2B models in transformer","aakashgaur01",2,"","",0,"2025-07-26","2025-09-03"],[0,39696,"[exaone4] ZeroDivisionError/TypeError when sliding_window_pattern is None/\"LLLG\" and _attn_implementation stay","wheeze01",2,"","",0,"2025-07-26","2025-07-29"],[0,39694,"4.54.0 bug: ImportError: cannot import name 'deterministic_g' from 'transformers.modeling_flash_attention_util","georgethrax",2,"","",0,"2025-07-26","2025-09-13"],[0,39692,"SigLIP2 documentation example has multiple errors (model/processor mismatch + quantization failure)","david-littlefield",0,"jonathan-fulton","#43652",0,"2025-07-26","2026-02-03"],[0,39687,"[DeepSeek-V3] Different rotary embedding implementation between DeepSeek-AI and Transformers","wwwjn",2,"","",0,"2025-07-26","2025-08-08"],[0,39686,"CRITICAL ISSUE REPORT! GEMMA 3 1B CANNOT RUN!","yukiarimo",2,"","",0,"2025-07-26","2025-07-26"],[0,39685,"Qwen 2.5 VL - error without attention_mask","aidando73",2,"","",0,"2025-07-26","2025-09-02"],[0,39684,"Add multi-candidate & tree search for assisted decoding (speculative decoding)","transcend-0",3,"","",0,"2025-07-25","2026-10-20"],[0,39682,"Accelerate beam search decoding via tree attention","transcend-0",3,"","",0,"2025-07-25","2026-10-20"],[0,39673,"error: argument --deepspeed: invalid dict value: '<path>'","yeshsurya",2,"","",0,"2025-07-25","2025-09-02"],[0,39668,"Issue when initializing a DynamicCache","xadupre",2,"","",0,"2025-07-25","2025-09-02"],[0,39656,"T5Gemma training not working","NilsHellwig",2,"","",0,"2025-07-25","2025-09-01"],[0,39650,"Inference API Returning 404","FoundationINCCorporateTeam",2,"","",0,"2025-07-24","2025-07-25"],[0,39648,"Use DP+FSDP device mesh dimensions for scaling loss with default value of average_tokens_across_devices: True","winglian",2,"","",0,"2025-07-24","2025-09-28"],[0,39647,"Please develop DataCollatorForVisionLanguageModeling to support visual model training !!!","ShelterWFF",3,"","",0,"2025-07-24","2026-08-17"],[0,39637,"[BUG] Run 111B+ Teacher distributed inference and 8B Student distributed training on multi-node H200 GPUs usin","seona21",2,"","",0,"2025-07-24","2025-09-01"],[0,39627,"[XPU] Model get OOM when loading models","Stonepia",0,"jiqing-feng","#39654",0,"2025-07-24","2025-07-29"],[0,39619,"FSDP v1 bug: trainer incorrectly uses an unwrapped model","YanjunChen329",2,"","",0,"2025-07-23","2025-08-31"],[0,39618,"SageAttention for attention implementation?","Many0therFunctions",2,"","",0,"2025-07-23","2026-10-20"],[0,39616,"Trainer: Error when folded metrics are saved","l-uuz",2,"","",0,"2025-07-23","2025-08-31"],[0,39611,"VoxtralForConditionalGeneration import error","xinkez",2,"","",0,"2025-07-23","2025-07-24"],[0,39608,"Qwen3 Fails w/4D Attn Mask when using FA2","ntenenz",2,"","",0,"2025-07-23","2025-08-31"],[0,39607,"ImageClassificationPipeline preprocess should accept numpy/tensor arrays","idantene",0,"kamran-nizamani","#46573",0,"2025-07-23","2026-06-16"],[0,39597,"The similarity between image and text in siglip2 is very low","sigma-alpha-beta",2,"","",0,"2025-07-23","2025-07-23"],[0,39596,"Does transformers support python3.13 -- disable-gil or python3.14 free threading?","SoulH-qqq",2,"","",0,"2025-07-23","2025-08-30"],[0,39593,"`gemma-3-1b-it` with `use_cache=True` and `past_key_values` throws `RuntimeError: CUDA error: device-side asse","nickeisenberg",2,"","",0,"2025-07-22","2025-07-24"],[0,39586,"AddedToken should check content on `_update`","phiott",1,"phiott","#39587",0,"2025-07-22","2025-07-28"],[0,39567,"Clarification on Recent Changes to Loss and Gradient Accumulation","jiosephlee",2,"","",0,"2025-07-22","2025-07-22"],[0,39565,"Model forward execution in full eager mode?","22quinn",2,"","",0,"2025-07-21","2025-08-21"],[0,39558,"Backwards incompatible change in returned hidden states","BenjaminBossan",0,"ArthurZucker","#39636",1,"2025-07-21","2025-07-25"],[0,39554,"Why `is_causal` is not used in `flash_attention_forward` ?","lucaswychan",1,"lucaswychan","#39707",0,"2025-07-21","2025-08-18"],[0,39549,"Is there plan to integrate ColQwen2.5 into Transformers?","rebel-thkim",0,"sahil-kabir","#40600",0,"2025-07-21","2025-11-03"],[0,39545,"Is the new Intel–Weizmann speculative decoding algorithm integrated into Transformers?","NEWbie0709",2,"","",0,"2025-07-21","2025-07-21"],[0,39542,"ValueError: You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time","xjackzenvey",0,"cyntqliu","#40045",0,"2025-07-21","2025-08-21"],[0,39537,"Add muon and flash-muon optimizer","kadirnar",2,"","",0,"2025-07-20","2026-10-19"],[0,39551,"InformerForPrediction [I would like to seek your opinions, everyone, How can I set the dynamic real features f","2004learner",2,"","",0,"2025-07-20","2025-08-28"],[0,39533,"KeyError: 'llava_qwen2'","fengyuentau",2,"","",0,"2025-07-20","2025-07-21"],[0,39530,"Typo in `apply_transcrition_request` method name","xenova",0,"revvvanth","#39572",0,"2025-07-20","2025-07-25"],[0,39527,"training google colab error","unoroberto12",2,"","",0,"2025-07-19","2025-08-18"],[0,39525,"paged attention NOT working with Qwen Models","NickNickGo",2,"","",0,"2025-07-19","2025-08-28"],[0,39522,"T5Gemma failing on provided example","jadermcs",2,"","",0,"2025-07-19","2025-08-12"],[0,39521,"T5Gemma problem with tokenizer(?)","jadermcs",2,"","",0,"2025-07-19","2025-07-21"],[0,39514,"T5Gemma returning 0 loss for s2s training","jadermcs",2,"","",0,"2025-07-19","2025-07-19"],[0,39512,"text-generation extremely slow with large `bad_words_ids` list","xenova",0,"dwarez","#39556",1,"2025-07-18","2025-07-25"],[0,39511,"Export voxtral to ExecuTorch","mergennachin",2,"","",0,"2025-07-18","2025-09-22"],[0,39510,"\"ValueError: Predictions and/or references don't match the expected format.\" error","ashokmon-aws",2,"","",0,"2025-07-18","2025-07-22"],[0,39508,"Whisper transcription is 2x slower between 4.51.3 -> 4.52.1","Anjum48",2,"","",0,"2025-07-18","2025-09-08"],[0,39498,"Gemma3n don't support chat with history","weedge",2,"","",0,"2025-07-18","2025-07-18"],[0,39497,"dataset  4.0.0 , issue with load_dataset loading audio dataset","nanyi545",2,"","",0,"2025-07-18","2025-07-18"],[0,39495,"Add Muon Optimiser for 2x faster convergence","RonanKMcGovern",4,"","",0,"2025-07-18",""],[0,39492,"modeling_flax_gemma.FlaxGemmaModule failed with incompatible shapes when running with GemmaConfig","nhatleSummer22",2,"","",0,"2025-07-18","2025-07-18"],[0,39484,"Transformers still tries to use apex.amp which is no longer a thing in apex.","yselivonchyk",2,"","",0,"2025-07-17","2025-08-25"],[0,39479,"Adding Space-Time-MiniLM-v0","haidarjomaa",4,"","",0,"2025-07-17",""],[0,39476,"Allow `load_best_model_at_end=True` to work when `save_steps < eval_steps` and best model is saved","KrisTHL181",4,"","",0,"2025-07-17",""],[0,39473,"Unexpected behaviour with transformers versions above 4.28 for Donut","mdavudov",2,"","",0,"2025-07-17","2025-10-15"],[0,39463,"can't torch.export.export tinyllama model","heshuju",2,"","",0,"2025-07-17","2025-07-17"],[0,39462,"HfArgumentParser cannot parse `str` for local path","Tavish9",1,"Tavish9","#39467",0,"2025-07-17","2025-07-17"],[0,39461,"I can't make sense of this works on Windows but not on Linux AutoModelForCausalLM.from_pretrained","FurkanGozukara",2,"","",0,"2025-07-17","2025-07-17"],[0,39460,"Autoformer get_lagged_subsequences always true if condition","Marco-Pi",2,"","",0,"2025-07-16","2025-08-24"],[0,39442,"Missing 4 spaces in SmolVLMImageProcessorFast","gnobitab",0,"Krish0909","#39452",0,"2025-07-16","2025-07-16"],[0,39440,"Add Interactive Multi-Modal Attention Visualization for Vision-Language Models","sisird864",3,"","",0,"2025-07-15","2026-09-04"],[0,39437,"Support for per-token latency tracking in `generate()` (suggested options: using callback, profiler class, or ","spsagar13",4,"","",0,"2025-07-15",""],[0,39436,"Export LFM2 to ExecuTorch","guangy10",2,"","",0,"2025-07-15","2025-08-24"],[0,39430,"Add DiCoW: Diarization-Conditioned Whisper","Lakoc",2,"","",0,"2025-07-15","2025-09-19"],[0,39427,"Gemma 3 Compilation Issues During Generation","mitchelldehaven",2,"","",0,"2025-07-15","2025-09-23"],[0,39426,"object detection : matchin outputs.last_hidden_state with results","fenaux",3,"","",0,"2025-07-15","2026-10-20"],[0,39424,"[YosoConfig] Missing `architectures` field","SankethHonavar",2,"","",0,"2025-07-15","2025-07-16"],[0,39421,"Speculative Decoding(do_sample=False) get different outputs","nighty8",2,"","",0,"2025-07-15","2025-07-18"],[0,39413,"Exeception 3 type mismatch","Smirkkkk",2,"","",0,"2025-07-15","2025-08-22"],[0,39410,"FP8 training support for Model Parallel / Tensor Parallel (MP/TP)","edgeinfinity1",4,"","",0,"2025-07-15",""],[0,39409,"TypeError: couldn't find storage object Float8_e4m3fnStorage - which version is needed for this?","FurkanGozukara",2,"","",0,"2025-07-15","2025-08-02"],[0,39408,"Off-by-one error when using flash_attention with a sliding window","tyler-romero",0,"vasqu","#40163",1,"2025-07-15","2025-08-20"],[0,39405,"breaking changes in ESM model classes","adrienchaton",2,"","",0,"2025-07-14","2025-07-17"],[0,39404,"Whisper `return_language` with pipeline no longer working","Metric-Void",0,"FredHaa","#42227",0,"2025-07-14","2026-04-27"],[0,39401,"Qwen3 tokenizer wrong offset_mapping","contribcode",2,"","",0,"2025-07-14","2025-07-16"],[0,39400,"Causal mask is not compatible with Qwen2-VL when using padding-free training","hiyouga",0,"zucchini-nlp","#39447",1,"2025-07-14","2025-07-21"],[0,39399,"Qwen2.5-VL Sharding error when using Tensor Parallelism","rupert404",2,"","",0,"2025-07-14","2025-09-17"],[0,39392,"Enabling `average_tokens_across_devices` by default in Trainer","MilkClouds",0,"Krish0909","#39395",0,"2025-07-14","2025-07-21"],[0,39389,"Gemma3 bidirectional mask for image tokens isn't reaching attention forward","brb-nv",0,"zucchini-nlp","#39396",1,"2025-07-13","2025-07-22"],[0,39388,"Using resnet-18 in flax","aagha6",2,"","",0,"2025-07-13","2025-07-14"],[0,39382,"Mask2FormerImageProcessor yields inconsistent results between single and batch inference","roboserg",2,"","",0,"2025-07-12","2025-08-20"],[0,39379,"Handling of full_text_row_masked_out_mask in mllama is incorrect.","nshepperd",2,"","",0,"2025-07-12","2026-10-05"],[0,39417,"Option to tokenize messages one after the other","VivienCabannes",0,"pco111","#39507",0,"2025-07-12","2025-07-31"],[0,39377,"FlashAttention2 support for  GSAI-ML / LLaDA-8B-Instruct?","lbertge",2,"","",0,"2025-07-12","2025-08-19"],[0,39367,"Adding api key to `transformers serve`","xhluca",4,"","",0,"2025-07-11",""],[0,39366,"RuntimeError when loading llmcompressor W8A8 quantized model: int8 dtype in weight initialization","AdelineXinyi",2,"","",0,"2025-07-11","2025-12-08"],[0,39365,"Bug in modeling_bart.eager_attention_forward","xadupre",2,"","",0,"2025-07-11","2026-09-10"],[0,39359,"vLLM v0.9.2: Qwen2.5-Omni-7B-AWQ fails to load with transformers 4.53.1 (requires 4.52.4)","GreenerZ",2,"","",0,"2025-07-11","2025-07-11"],[0,39352,"env.useBrowserCache = true causes JSON parsing error, forced to disable cache making app slower.","piyushh-k",2,"","",0,"2025-07-11","2025-09-15"],[0,39350,"surpport for google/medgemma-27b-it","moshilangzi",2,"","",0,"2025-07-11","2026-10-20"],[0,39337,"Latest Transformers release causes CUDA out-of-memory errors during VisionLLM fine-tuning","ruiite",2,"","",0,"2025-07-10","2025-07-14"],[0,39336,"TypeError: GenerationMixin._extract_past_from_model_output() got an unexpected keyword argument 'standardize_c","FurkanGozukara",2,"","",0,"2025-07-10","2025-08-18"],[0,39334,"FlashAttention2 ImportError: undefined symbol with flash_attn_2_cuda when loading Phi-4-Multimodal","tanuj-rai",2,"","",0,"2025-07-10","2025-07-10"],[0,39329,"Adding support for Gemma 3n GGUFs","minhnguyenhoang",4,"","",0,"2025-07-10",""],[0,39326,"TypeError in Qwen2_5_VLForConditionalGeneration (torch.finfo misuse)","dsnsabari",2,"","",0,"2025-07-10","2025-07-10"],[0,39319,"Add HF integration dates + paper release dates to the model docs","MHRDYN7",2,"","",0,"2025-07-09","2025-08-14"],[0,39318,"Whisper demo code for model + processor API is broken","piotyrus",2,"","",0,"2025-07-09","2025-08-18"],[0,39311,"Inference with model.generate( ) using a quantized model leads to assertion error","Sandipan99",2,"","",0,"2025-07-09","2025-08-18"],[0,39302,"Gemma3 slightly alters hidden state when input_ids is batched","lwiklendt",2,"","",0,"2025-07-09","2025-07-10"],[0,39295,"`quantization_method` is not cleared after calling `.dequantize()`","karol-szustakowski",0,"dwarez","67f4292",1,"2025-07-09","2025-07-15"],[0,39291,"Support 2D Array Inputs in Wav2Vec2FeatureExtractor for Non-Waveform Modalities","cyberso",2,"","",0,"2025-07-09","2025-08-20"],[0,39290,"v4.53.0+ starts erroring with 'Gemma3TextConfig' object has no attribute 'sliding_window_pattern' with vLLM","es94129",0,"zucchini-nlp","#39301",1,"2025-07-09","2025-07-09"],[0,39288,"flash_attention_3 for Qwen2_5_VisionTransformerPretrainedModel","aidando73",0,"zucchini-nlp","#39294",1,"2025-07-08","2025-07-09"],[0,39285,"Pytorch language_modelling example run_clm fails when streaming is enabled","HRezaei",1,"HRezaei","#39286",0,"2025-07-08","2025-07-15"],[0,39275,"hangs during training using deepspeed","mxjmtxrm",2,"","",0,"2025-07-08","2025-08-17"],[0,39269,"SigLip2 text pooler output selection","Jason3900",2,"","",0,"2025-07-08","2025-07-16"],[0,39266,"Unable to create tensor, you should probably activate truncation and/or padding with 'padding=True' 'truncatio","mumu029",2,"","",0,"2025-07-08","2025-07-08"],[0,39260,"Please help i am trying to run model but issue","vinaydanidhariya",2,"","",0,"2025-07-07","2025-08-17"],[0,39256,"ModuleNotFoundError: No module named 'habana_frameworks.torch'","orkungedik",2,"","",0,"2025-07-07","2025-07-07"],[0,39249,"OSError: Tensor parallel is only supported for `torch>=2.5`","mxjmtxrm",2,"","",0,"2025-07-07","2025-07-08"],[0,39241,"[Trainer] Eval loss depends on batch size (with solution)","ba144220",2,"","",0,"2025-07-07","2025-08-18"],[0,39237,"disable_grouping parameter missed in image_processing_glm4v_fast.py","r0mar0ma",2,"","",0,"2025-07-05","2025-07-07"],[0,39235,"Specifying multiple metrics in TrainingArguments.metric_for_best_model","gumran",3,"","",0,"2025-07-05","2026-09-01"],[0,39231,"v4.53.0 - Qwen 2.5 VL Flash Attention error - object has no attribute is_causal","aidando73",2,"","",0,"2025-07-04","2025-08-12"],[0,39224,"transformers: FlaubertTokenizer: do_lowercase_and_remove_accent: make the logger warning actionable (don't onl","kirisakow",2,"","",0,"2025-07-04","2026-07-05"],[0,39217,"torch fake_tensor load hf model failed","SandyWang85",2,"","",0,"2025-07-04","2025-08-12"],[0,39215,"_load_rng_state after get_batch_samples may break training reproducibility when dataloader has random operatio","rangehow",2,"","",0,"2025-07-04","2026-07-19"],[0,39214,"Inconsistant `input_feature` length and `attention_mask` length in `WhisperFeatureExtractor`","BakerBunker",1,"BakerBunker","#39221",0,"2025-07-04","2025-08-11"],[0,39213,"Remove device to host sync triggered in _flash_attention_forward","piyifan123",4,"","",0,"2025-07-04",""],[0,39208,"Unknown Model (mobilenetv5_300m_enc) when loading Gemma 3n","HenryNdubuaku",0,"harshaljanjani","#40951",0,"2025-07-03","2025-09-18"],[0,39204,"When creating a Trainer object for a MixedModel, the initialization tries to access attribute get_base_model (","rluss",0,"sebastianvlad1","#39226",0,"2025-07-03","2025-07-16"],[0,39203,"Qwen3 MOE models w/non-empty `mlp_only_layers` fail when `output_router_logits=True`","ntenenz",2,"","",0,"2025-07-03","2025-08-18"],[0,39202,"Naming incosistencies of `PreTrained*` classes.","vitormbesen",2,"","",0,"2025-07-03","2026-10-20"],[0,39201,"No or astronomical loss in `ModernBertForMultipleChoice`","netique",2,"","",0,"2025-07-03","2025-07-04"],[0,39191,"🐛 Bug Report: Accelerate config to disable torch dynamo is ignored by transformers automatic compilation","leobianco",2,"","",0,"2025-07-03","2025-08-11"],[0,39188,"Gemma2 fall back to cpu execusion when attn_implementation='flash_attention_2'","Lingy12",2,"","",0,"2025-07-03","2025-09-17"],[0,39186,"FSDP RuntimeError: 'weight' must be 2-D","muhayyo-tash",2,"","",0,"2025-07-03","2025-07-05"],[0,39175,"Torch patches tracker for HPU/Gaudi","IlyasMoutawwakil",2,"","",0,"2025-07-02","2026-01-10"],[0,39169,"Using Gemma3n with text-only generation requires image dependencies","meickmeyer",2,"","",0,"2025-07-02","2025-08-01"],[0,39168,"Illegal memory access when using 3d rope","BakerBunker",2,"","",0,"2025-07-02","2025-07-08"],[0,39167,"apply_rotary_pos_emb_flashatt failed during triton jit compilation 'constexpr' object has no attribute 'bit_le","hebiao064",2,"","",0,"2025-07-02","2025-08-11"],[0,39162,"Not capable of exporting Mistral to ONNX format with the use of caching","EricJi150",2,"","",0,"2025-07-01","2025-08-09"],[0,39159,"[CI ENERGY Waste] The exist jobs in `Doctests` that has never completed successfully","souhailaS",0,"ydshieh","#39316",1,"2025-07-01","2025-07-09"],[0,39151,"`LayoutLMv3TokenizerFast` doesn't pass all the params.","sergiopaniego",2,"","",0,"2025-07-01","2025-07-01"],[0,39143,"OpenTelemetry Collector Connection error when installing the latest release 4.53.0 during `docker build`","ancalita",2,"","",0,"2025-07-01","2025-07-16"],[0,39141,"VLLM depoly Qwen2.5_omni server error","WenmuZhou",2,"","",0,"2025-07-01","2025-07-02"],[0,39139,"Add x-transformers library by lucidrains","asigalov61",3,"","",0,"2025-06-30","2026-09-09"],[0,39137,"ImportError: cannot import name 'pipeline' from 'transformers'","atabari-bci",2,"","",0,"2025-06-30","2025-08-12"],[0,39136,"bf16_full_eval=True moves model to device before FSDP application and causes cuda OOM","jlu-figma",2,"","",0,"2025-06-30","2025-09-11"],[0,39124,"AttributeError: 'Resampler' object has no attribute '_initialize_weights'","DarwinYang",2,"","",0,"2025-06-30","2025-07-02"],[0,39123,"Gradient accumulation steps for Vision Languge model","khalil-Hennara",2,"","",0,"2025-06-30","2025-08-09"],[0,39118,"[Gaudi] the seamless_m4t cannot work on Gaudi. No need to fix. Workaround PR is merged.","yuanwu2017",2,"","",0,"2025-06-30","2025-07-03"],[0,39115,"`transformers.utils.metrics` sets global `TracerProvider`","harupy",0,"McPatate","#39422",1,"2025-06-30","2025-07-15"],[0,39114,"Is there a way to force it to use ASCII based progress bar and not the ipython widget one?","weathon",2,"","",0,"2025-06-29","2026-08-06"],[0,39112,"QWEN2VLProcessor missing video_token_id in mm_token_type_ids","NikeHop",2,"","",0,"2025-06-29","2025-08-18"],[0,39111,"New release 4.53.0 breaks HF trainer/model","liujch1998",2,"","",0,"2025-06-29","2025-08-09"],[0,39107,"Caching of model code in ~/.cache/huggingface/modules/transformers_modules","ira-b-27",2,"","",0,"2025-06-29","2025-08-24"],[0,39105,"How to use other acceleration apis of npu?","zheliuyu",2,"","",0,"2025-06-29","2026-01-04"],[0,39101,"docs: fix typos in awesome-transformers.md WIP","VladimirGutuev",1,"VladimirGutuev","#39102",0,"2025-06-28","2025-06-30"],[0,39099,"Pretrainedtokenizerfast Segmentation fault","Facerman-cloud",2,"","",0,"2025-06-28","2025-08-09"],[0,39097,"[Core] Saving models with multiple shared tensor groups is not supported when model is dispatched","kylesayrs",1,"kylesayrs","#39263",0,"2025-06-28","2025-07-10"],[0,39095,"`Qwen2_5_VLVisionAttention` with flash attention has no `is_causal` attribute","allentsouhuang",0,"zucchini-nlp","#39121",1,"2025-06-27","2025-07-01"],[0,39091,"`transformers`' dependency on `sentencepiece` blocks use on windows in python 3.13","leondz",2,"","",0,"2025-06-27","2025-07-03"],[0,39089,"Bug in version 4.52.4: LlavaOnevisonConfig Class init the inappropriate (hidden_size,num_attention_heads) pair","glgjss960",0,"ved1beta","#39163",0,"2025-06-27","2025-07-02"],[0,39081,"AttributeError: 'HfTrainerDeepSpeedConfig' object has no attribute 'is_zero3'","JaydenChao101",2,"","",0,"2025-06-27","2025-08-05"],[0,39077,"ImportError: cannot import name 'TorchTensorParallelPlugin' from 'accelerate.utils'","ibillxia",0,"S1ro1","#39083",0,"2025-06-27","2025-06-27"],[0,39076,"Loading audio in video from video URLs fail with chat template","merveenoyan",2,"","",0,"2025-06-27","2025-08-15"],[0,39075,"facebook/dinov2-with-registers-giant does not appear to have a file named pytorch_model.bin, model.safetensors","MengHao666",2,"","",0,"2025-06-27","2025-08-05"],[0,39073,"Inefficient default GELU implementation in GPT2","null-pointer-access",2,"","",0,"2025-06-27","2025-08-05"],[0,39072,"Inefficient memory resharding in attention layer","null-pointer-access",2,"","",0,"2025-06-27","2025-08-05"],[0,39067,"Output logits differ significantly for different attn_implementations on image inputs","zzigak",2,"","",0,"2025-06-26","2025-08-30"],[0,39055,"DBRX model passes probabilities and not logits to the load balancer","rudolfwilliam",1,"rudolfwilliam","#39056",0,"2025-06-26","2025-07-16"],[0,39049,"pytorch version 1.8.1 compatibility","Dominiq24",2,"","",0,"2025-06-26","2025-08-03"],[0,39048,"[BUG] Converting Megatron-LM GPT-2 to hf format raise error: KeyError: 'position_embeddings'","lishuai-97",2,"","",0,"2025-06-26","2025-06-28"],[0,39045,"GPTBot AI4BUSINESS","Onay931",2,"","",0,"2025-06-25","2025-06-25"],[0,39044,"Deepseek-ai","Onay931",2,"","",0,"2025-06-25","2025-06-26"],[0,39038,"Only with newest version (4.52.4): from_pretrained() esm.embeddings.position_embeddings.weight missing","hunklinger",2,"","",0,"2025-06-25","2025-08-03"],[0,39036,"AutoModelForCausalLM.from_pretrained(..., device_map=...) ignore `Tensor.retain_grad()` in Multi-GPUs setting","jasonrichdarmawan",2,"","",0,"2025-06-25","2025-10-12"],[0,39030,"`AutoTokenizer.from_pretrained` does not propagate `token`","anakin87",0,"brandenkmurray","#40956",0,"2025-06-25","2025-09-18"],[0,39027,"Issue with module.smart_apply(module._initialize_weights) in the initialize_weights Function of modeling_utils","DarwinYang",2,"","",0,"2025-06-25","2025-08-30"],[0,39026,"CPMANT Model Fails to Run Following Official Tutorial","iugoood",2,"","",0,"2025-06-25","2025-08-18"],[0,39023,"Does Gemma 3 need positions ids to be 1-indexed explicitly?","krypticmouse",2,"","",0,"2025-06-25","2025-07-25"],[0,39022,"AssertionError: The parameter 666 has already been reduced Gradient computed twice for this partition. Multipl","rutts29",2,"","",0,"2025-06-24","2025-06-26"],[0,39020,"Segfault on Apple M4 using AutoModelForSequenceClassification with BETO model on CPU","Juan-Alberto-Ignacio",2,"","",0,"2025-06-24","2025-08-28"],[0,39018,"Grammatical error in the \"Loading model's\" page","marcndo",2,"","",0,"2025-06-24","2025-07-25"],[0,39017,"Not able to use flash attention with torch.compile with model like BERT","gambiTarun",2,"","",0,"2025-06-24","2025-08-02"],[0,39007,"'Mistral3Model' object has no attribute 'prepare_inputs_for_generation'","timpal0l",2,"","",0,"2025-06-24","2025-08-02"],[0,39004,"Warning when load pretrained model for qwen2-VL-1.5B-Instruct.","wuaiglengme",0,"bvantuan","#39029",0,"2025-06-24","2025-07-01"],[0,38997,"v4.51.3-Qwen2.5-Omni-preview is not available anymore","aurel-g",2,"","",0,"2025-06-24","2025-06-24"],[0,38996,"AutoModelForImageClassification broken in 4.52.4: AttributeError: 'TinyVit' object has no attribute '_initiali","virgile-fsr",2,"","",0,"2025-06-23","2025-06-24"],[0,38984,"QA pipeline prediction generates wrong response when `top_k` param > 1","WeichenXu123",0,"Vixel2006","#39193",0,"2025-06-23","2025-07-17"],[0,38982,"ModernBERT training loss is astronomical when using gradient accumulation","umarbutler",1,"umarbutler","#38983",0,"2025-06-23","2025-06-25"],[0,38981,"InternVL3 config missing critical fields after saving","TheDropZone",2,"","",0,"2025-06-23","2025-06-23"],[0,38977,"LMHead is processing redundant tokens in prefill","null-pointer-access",1,"null-pointer-access","#39016",0,"2025-06-23","2025-06-25"],[0,38975,"[torch.export] Unhandled FakeTensor Device Propagation for two different devices","Stonepia",1,"Stonepia","#38976",0,"2025-06-23","2025-07-17"],[0,38973,"ImportError: cannot import name 'DTensor' from 'torch.distributed.tensor'","ChinDer",2,"","",0,"2025-06-23","2025-06-23"],[0,38970,"Global and Local Anomaly co-Synthesis Strategy (GLASS)","sbrzz",2,"","",0,"2025-06-22","2025-06-23"],[0,38968,"Wandb isn't logging config in offline mode","souththzz",2,"","",0,"2025-06-22","2025-07-23"],[0,38966,"Nemotron-H: A Family of Accurate and Efficient Hybrid Mamba-Transformer Models","AshAnand34",2,"","",0,"2025-06-21","2025-12-03"],[0,38965,"Modernbert implementation with Tensorflow","lfoppiano",2,"","",0,"2025-06-21","2025-06-23"],[0,38963,"RuntimeError: Index put requires the source and destination dtypes match, got Half for the destination and Flo","sinngam-khaidem",2,"","",0,"2025-06-21","2025-07-08"],[0,38961,"🛠️ Add Failure-Safe Training with Emergency Checkpointing","yaswanth19",0,"efazal","#41723",0,"2025-06-21","2025-12-04"],[0,38952,"Deepspeed HF_HOME Cuda Ops bug on transformers==4.51.0 (fixed in >=4.52.0)","Alex-ley-scrub",2,"","",0,"2025-06-20","2025-06-23"],[0,38950,"Number Token Loss","jannisborn",3,"","",0,"2025-06-20","2026-10-20"],[0,38939,"Resuming training from an interrupted checkpoint fails to save the final checkpoint.","rangehow",1,"rangehow","#40347",0,"2025-06-20","2025-09-18"],[0,38935,"Gemma2 crash on load with torch <= 2.4","francois-rozet",2,"","",0,"2025-06-20","2025-06-27"],[0,38933,"ddp_time in TrainingArguments with deepspeed does not work","kindernerd",2,"","",0,"2025-06-20","2025-07-09"],[0,38928,"Improve CI/CD by completing migration from setup.py to pyproject.toml","ParagEkbote",2,"","",0,"2025-06-20","2025-08-12"],[0,38927,"Can't load my LoRA checkpoint after gemma3 refactor","jood-canva",2,"","",0,"2025-06-20","2025-07-10"],[0,38925,"Checkpointing broken for classifier training multi-gpu","ojh31",2,"","",0,"2025-06-20","2025-07-28"],[0,38924,"Exporting Llava decoder into ONNX format","EricJi150",2,"","",0,"2025-06-19","2025-08-12"],[0,38918,"Lack of IDE-Specific Authentication Instructions in Hugging Face \"Quickstart\" Documentation","marcndo",1,"marcndo","#38954",0,"2025-06-19","2025-06-24"],[0,38917,"Add past_key_values as inputs to TorchExportableModuleWithStaticCache forward","justinchuby",2,"","",0,"2025-06-19","2026-10-21"],[0,38914,"Error: StaticCache.__init__() got an unexpected keyword argument 'batch_size'","rsxdalv",2,"","",0,"2025-06-19","2025-08-22"],[0,38910,"`load_balancing_loss_func` doesn't support 4D attention mask","Vermouth7",2,"","",0,"2025-06-19","2025-07-30"],[0,38905,"Add flexible configuration for Liger Kernel in TrainingArguments","hamza-hcompany",1,"hamza-hcompany","#38911",0,"2025-06-19","2025-06-19"],[0,38903,"Incorrect comparison when generating `special_image_mask` in Gemma3Model","caicai0402",2,"","",0,"2025-06-19","2025-06-19"],[0,38898,"Qwen2_5OmniProcessor.__init__() got multiple values for argument 'image_processor'","WenmuZhou",0,"Isotr0py","#39125",0,"2025-06-19","2025-07-01"],[0,38891,"Add EdgeTam to Transformers 🤗","yonigozlan",2,"","",0,"2025-06-18","2026-01-08"],[0,38889,"The wrong config parameter found in src/transformers/models/qwen2_5_vl/configuration_qwen2_5_vl.py.","Bytes-Lin",2,"","",0,"2025-06-18","2025-07-27"],[0,38878,"📄 Information about the usage of models and code with non-commercial LICENSE on transformers and diffusers","ishandutta0098",2,"","",0,"2025-06-18","2025-06-20"],[0,38874,"Reproducibility Issue of Siglip2 with Blackwell Architecture GPUs (RTX 5090)","silverstar0727",2,"","",0,"2025-06-18","2025-07-27"],[0,38873,"Inconsistency between from_pretrained and save_pretrained API","jiafatom",2,"","",0,"2025-06-17","2025-06-18"],[0,38871,"Qwen3 thinking flag is flipped","rasbt",2,"","",0,"2025-06-17","2025-06-18"],[0,38870,"Safetensors deserializing silently mishandles tied parameters","edmcman",2,"","",0,"2025-06-17","2025-10-26"],[0,38866,"Super slow inference using `eager` attention with `Llama-4-Scout-17B-16E-Instruct`","Tizzzzy",2,"","",0,"2025-06-17","2025-06-18"],[0,38858,"BLIP-2 regression in 4.52.1","ageron",2,"","",0,"2025-06-17","2025-06-18"],[0,38854,"scale loss per token/local sequence for discrete system representation","wesboyt",3,"","",0,"2025-06-17","2026-10-13"],[0,38851,"Should `compute_metrics` only run on the main process when doing DDP?","TIE666",2,"","",0,"2025-06-17","2025-07-25"],[0,38846,"video_auto_processing.py breaks everything","lucasjinreal",2,"","",0,"2025-06-16","2025-07-25"],[0,38843,"Error when create ModernBert model with flash attention  TypeError: RotaryEmbedding.__init__() got an unexpect","KabaevAnton",2,"","",0,"2025-06-16","2025-07-27"],[0,38837,"Loss is incorrectly scaled in Trainer during the last step with gradient accumulation when the final batch is ","hutaiHang",1,"hutaiHang","#39659",0,"2025-06-16","2025-07-29"],[0,38832,"ImportError: cannot import name 'BitsAndBytesConfig' from 'bitsandbytes'","tanuj-rai",2,"","",0,"2025-06-15","2025-06-15"],[0,38829,"Support Asynchronous Evaluation on Separate GPU in `Trainer`","AmitMY",3,"","",0,"2025-06-15","2026-09-30"],[0,38823,"Cannot create a 2-class Deformable DETR model from a 1-class model","jveitchmichaelis",2,"","",0,"2025-06-14","2025-06-14"],[0,38820,"CI skipped failures tracking issue","ivarflakstad",2,"","",0,"2025-06-13","2025-07-23"],[0,38809,"Text classification evaluation has_labels bug!","HERIUN",2,"","",0,"2025-06-13","2025-06-17"],[0,38803,"DTensor issues when running Llama4ForConditionalGeneration with tensor parallel.","czkkkkkk",2,"","",0,"2025-06-12","2025-07-21"],[0,38787,"Qwen3 Incorrect order of sliding window layers","norpadon",1,"norpadon","#38794",0,"2025-06-12","2025-06-13"],[0,38776,"[Bug][DOCS] TrainingArguments.__init__() got an unexpected keyword argument 'fsdp_strategy'","PT-10",1,"PT-10","#38807",0,"2025-06-12","2025-06-13"],[0,38771,"device_map='auto' coupled with tp_plan='auto'","weifengpy",0,"SunMarc","#38942",1,"2025-06-11","2025-07-03"],[0,38765,"Add FastVLM from CVPR 2025","camilla-deckard",2,"","",0,"2025-06-11","2025-12-04"],[0,38753,"Silent Overwrite of Custom Optimizer When Using DeepSpeed with Transformers Trainer","VladPyzh",2,"","",0,"2025-06-11","2025-07-21"],[0,38750,"Is it a good choice to early error when `output_attentions=True` and attn implementation not equal to `eager`","kaixuanliu",3,"","",0,"2025-06-11","2025-06-11"],[0,38745,"[Bug][InformerForPredict] The shape will cause a problem","2004learner",0,"Flink-ddd","#38856",0,"2025-06-11","2025-06-23"],[0,38742,"[DeepSeek-V3] take care of the case `q_lora_rank is None`","bzantium",1,"bzantium","#38743",0,"2025-06-11","2025-06-11"],[0,38740,"[DOCS] Add `pruna` as optimization framework","davidberenstein1957",2,"","",0,"2025-06-11","2026-02-27"],[0,38739,"LlamaAttention forward function type hint is incorrect","nhatkhtn",2,"","",0,"2025-06-10","2025-07-15"],[0,38733,"GRPO per_device_eval_batch_size can't be set as 1, when there is only 1 GPU","CasanovaLLL",2,"","",0,"2025-06-10","2025-06-11"],[0,38727,"KV Cache Bug in Iterative generation","Greek-Guardian",0,"zucchini-nlp","#38775",1,"2025-06-10","2025-06-12"],[0,38726,"Issue importing models in jupyter notebooks 'No module named transformers.models.ipynb_checkpoints'","mchaudrycupa",2,"","",0,"2025-06-10","2025-07-19"],[0,38725,"`MoshiIntegrationTests` started to fail after #34464","ydshieh",2,"","",0,"2025-06-10","2025-11-19"],[0,38720,"ModernBERT for Sequence Classification - issues with finetuning","98MM",2,"","",0,"2025-06-10","2025-07-16"],[0,38717,"`.to` is not supported for `4-bit` or `8-bit` bitsandbytes models. Please use the model as it is, since the mo","falali009",2,"","",0,"2025-06-10","2025-07-18"],[0,38710,"There is no transformers version that can run DeepSeek V3 generate","pbelevich",2,"","",0,"2025-06-10","2025-07-15"],[0,38709,"`get_video_features` in XCLIPModel always returns `pooled_output`","Vishu26",2,"","",0,"2025-06-10","2025-07-18"],[0,38708,"Bert2D: A 2D-Word Embedding Model for Morphologically Rich Languages","yigit353",4,"","",0,"2025-06-09",""],[0,38702,"Incorrect scaling of Gemma embeddings in float32 regime","norpadon",2,"","",0,"2025-06-09","2025-08-11"],[0,38693,"Add MulT Model from “Multimodal Transformer for Unaligned Multimodal Language Sequences\" paper","Vixel2006",2,"","",0,"2025-06-09","2025-06-09"],[0,38692,"CheckpointLoaderSimple ..... Error while deserializing header: InvalidHeaderDeserialization","saeedafm",2,"","",0,"2025-06-09","2025-07-17"],[0,38690,"[BUG] Got nan logits after mask logic refactor","jiqing-feng",2,"","",0,"2025-06-09","2025-06-23"],[0,38687,"[RuntimeError: Expected all tensors to be on the same device, but found at least two devices] when fine-tuning","karoaper",2,"","",0,"2025-06-09","2025-07-17"],[0,38686,"`Trainer._save()` May Incorrectly Save Empty Model State (safetensors)","ChenDaiwei-99",2,"","",0,"2025-06-09","2025-07-24"],[0,38680,"Add support for Orthogonal Residual Updates","BootsofLagrangian",3,"","",0,"2025-06-09","2026-10-20"],[0,38667,"TypeError: 'NoneType' object is not iterable in ESM when using DDP training","dbleyl",2,"","",0,"2025-06-08","2025-07-16"],[0,38665,"Exception while inference Qwen2VL and Qwen2VL, assert module.weight.shape[1] == 1","iglaweb",2,"","",0,"2025-06-07","2025-08-04"],[0,38663,"MODEL_FOR_MASK_GENERATION_MAPPING_NAMES variable is present twice in modeling_auto.py","sbucaille",1,"sbucaille","#38664",0,"2025-06-07","2025-06-09"],[0,38662,"Whisper models appear to be broken with Flash Attention 2","Anjum48",2,"","",0,"2025-06-07","2025-07-18"],[0,38659,"NotImplementedError: Loading a dataset cached in a LocalFileSystem is not supported.","lucavlasblom",2,"","",0,"2025-06-07","2025-06-09"],[0,38657,"please develop transformers java/scala  sdk,eagerly to use!","mullerhai",3,"","",0,"2025-06-07","2026-10-20"],[0,38656,"Potential Memory Leak or Caching in Fast Image Processor","yhyang201",2,"","",0,"2025-06-07","2025-08-12"],[0,38654,"The visualization of image input in Qwen2.5-VL","Bytes-Lin",2,"","",0,"2025-06-07","2025-06-10"],[0,38650,"Support of Qwen3 GGUF model","Auth0rM0rgan",2,"","",0,"2025-06-06","2025-07-15"],[0,38639,"ImportError: cannot import name 'DTensor' from 'torch.distributed.tensor'","ybdong919",2,"","",0,"2025-06-06","2025-06-09"],[0,38634,"Download models from a private hub in 2025","DanielSchuhmacher",2,"","",0,"2025-06-06","2025-06-13"],[0,38625,"Remove 'as_target_processor' from docstring of Wav2Vec2Processor* methods","renet10",2,"","",0,"2025-06-05","2025-06-27"],[0,38623,"RagRetriever retrieve function mismatch in signature and return,  unneeded requirement for faiss","Fiona-Waters",1,"Fiona-Waters","#38624",0,"2025-06-05","2025-06-09"],[0,38617,"ImportError: cannot import name 'layer_type_validation' from 'transformers.configuration_utils'","Jacoobr",2,"","",0,"2025-06-05","2025-06-15"],[0,38613,"MDX Errors","rileyafox",2,"","",0,"2025-06-05","2025-06-06"],[0,38609,"Wav2Vec2Processor/Wav2Vec2ProcessorWithLM documentation still refers to 'as_target_processor'","renet10",0,"marcndo","#38642",0,"2025-06-05","2025-06-09"],[0,38602,"Failed to full fine tuning code5p 2B","latuan1",2,"","",0,"2025-06-05","2025-07-13"],[0,38601,"Contribute to Transformers on windows natively without WSL","ghost",2,"","",0,"2025-06-05","2025-07-27"],[0,38600,"MambaInnerFnBackward","HN-BB",2,"","",0,"2025-06-05","2025-07-13"],[0,38593,"Consider Deprecating Sigopt from Hyperparameter Search","ParagEkbote",1,"ParagEkbote","#38610",0,"2025-06-04","2025-06-05"],[0,38590,"\"facebook/opt-125m\" gives wrong results","BoyuanFeng",2,"","",0,"2025-06-04","2025-06-05"],[0,38588,"Transformers fail to load deepseek-ai/DeepSeek-V3 with vllm","ajayvohra2005",2,"","",0,"2025-06-04","2025-07-13"],[0,38576,"A local variable 'image_seq_length' leading to UnboundLocalError: cannot access local variable 'image_seq_leng","IceGiraffe",1,"IceGiraffe","#38577",0,"2025-06-04","2025-06-04"],[0,38574,"1","wubingheng111",2,"","",0,"2025-06-04","2025-06-04"],[0,38573,"Request to add the small-doge model","wubingheng111",2,"","",0,"2025-06-04","2026-08-01"],[0,38571,"Oneke not utilizing much from GPU(Nvidia L20)","zy1260957619",2,"","",0,"2025-06-04","2025-06-06"],[0,38567,"Support for excel files","Murdock135",2,"","",0,"2025-06-03","2025-06-04"],[0,38559,"Possible Typo in \"Mask2FormerLoss\"","NiccoloCavagnero",2,"","",0,"2025-06-03","2025-07-04"],[0,38554,"hidden_states, self_attn_weights = self.self_attn( ValueError: too many values to unpack (expected 2)","kasaer",2,"","",0,"2025-06-03","2025-07-12"],[0,38549,"Clarification on default top_k sampling parameter","MostHumble",2,"","",0,"2025-06-03","2025-07-12"],[0,38544,"Paligemma model card needs update","punitvara",2,"","",0,"2025-06-03","2025-07-14"],[0,38542,"enable GraniteMoeHybridIntegrationTest in UT","yao-matrix",2,"","",0,"2025-06-03","2025-08-06"],[0,38541,"`eager_attention_forward` and `repeat_kv` code duplication","ChengLyu",2,"","",0,"2025-06-03","2025-06-10"],[0,38538,"Hidden states are different for model() and model.generate()","okihnjo",2,"","",0,"2025-06-02","2025-06-05"],[0,38535,"Streaming mode support on HF vs kyutai-labs for the mimi model","billmguo",2,"","",0,"2025-06-02","2025-09-02"],[0,38527,"Why do you remove sample_indices_fn for processor.apply_chat_template?","futrime",2,"","",0,"2025-06-02","2025-06-03"],[0,38524,"404 Client Error when accessing https://router.huggingface.co/nebius/v1/chat/completions endpoint","indrawi15",2,"","",0,"2025-06-02","2025-06-04"],[0,38523,"\"Size mismatch\"  error when trying to download pretrained ChatGPT-4 using transformers","AnastassiyaP",2,"","",0,"2025-06-02","2025-07-11"],[0,38522,"Allow `mlm_probability` to be set to None when `mlm`=False in `DataCollatorForLanguageModeling`","KameniAlexNea",2,"","",0,"2025-06-02","2025-07-11"],[0,38521,"Error for `return_assistant_tokens_mask` in MLLM processor","MilkClouds",0,"zucchini-nlp","#38545",1,"2025-06-02","2025-07-18"],[0,38518,"Failed to export PyTorch  traced graph of Mixtral-8x7B-Instruct-v0.1 due to the PR #32429","nv-guomingz",2,"","",0,"2025-06-01","2026-06-29"],[0,38517,"model_type = self._reverse_config_mapping[key.__name__] KeyError: 'Qwen2RMConfig'","Lelege0",2,"","",0,"2025-06-01","2025-07-16"],[0,38516,"Hang in quantized_phi::ModelWeights::forward() with Phi-2 GGUF on CPU (Candle main branch)","EarthSports",3,"","",0,"2025-06-01","2025-06-03"],[0,38514,"Can not reproduce Blip2ForImageTextRetrieval example from docs, getting different results","KarlisJ",2,"","",0,"2025-06-01","2025-06-17"],[1,8163,"GRPO crashes on Qwen2-VL / Qwen2.5-VL with transformers < 5.3.0 when a completion ends with `<|vision_start|>`","albertvillanova",4,"","",0,"2026-11-06",""],[1,8162,"SFTTrainer keeps ShareGPT roles `human`/`gpt`: Qwen3 renders nothing and the train dataset ends up empty","shashank-100",4,"","",0,"2026-11-06",""],[1,8149,"`is_conversational` result depends on `PYTHONHASHSEED`","aidless",4,"","",0,"2026-11-06",""],[1,8136,"CI emits triton UserWarning: TRITON_DEFAULT_FP_FUSION is unset: defaulting to FP fusion ENABLED for backward c","albertvillanova",4,"","",0,"2026-11-05",""],[1,8135,"CI emits transformers FutureWarning: `gradient_checkpointing_enable` is deprecated and will be removed in v6. ","albertvillanova",4,"","",0,"2026-11-05",""],[1,8134,"CI emits transformers FutureWarning: `gradient_checkpointing` is deprecated and will be removed in v6. Use `ac","albertvillanova",4,"","",0,"2026-11-05",""],[1,8130,"CI fails for experimental tests with dev dependencies: ValueError: An error occurred while trying to load from","albertvillanova",4,"","",0,"2026-11-05",""],[1,8127,"optimizer_cls_and_kwargs missing from DPO/GRPO/KTO/RLOO trainers (regression of #2302 for RLOO)","shupandee",4,"","",0,"2026-11-05",""],[1,8123,"Tiny GLM-5.3-Flash stores the mHC `base`/`scale` in bf16 while `zai-org/GLM-5.3-Flash` keeps them in float32","albertvillanova",4,"","",0,"2026-11-05",""],[1,8121,"SFT with assistant_only_loss trains the pre-filled thinking block on GLM-4.5 and GLM-4.6 with thinking off","albertvillanova",4,"","",0,"2026-11-05",""],[1,8116,"`check_dtype_pattern` skips tensors whose names differ from the reference checkpoint","albertvillanova",1,"albertvillanova","#8118",1,"2026-11-04","2026-11-05"],[1,8115,"Tiny GLM-5.3-Flash model fails its forward pass: AssertionError: Bs must be 3D (num_experts, N//block_n, K//bl","albertvillanova",4,"","",0,"2026-11-04",""],[1,8109,"Training chat templates train tokens that the generation prompt already ends with","albertvillanova",4,"","",0,"2026-11-04",""],[1,8104,"AsyncGRPO fails on Gemma 4: FlashAttention forward only supports head dimension at most 256","rycerzes",4,"","",0,"2026-11-03",""],[1,8102,"SFTTrainer saves `output_router_logits=True` in the config of the MoE models it trains","qgallouedec",4,"","",0,"2026-11-03",""],[1,8095,"Gemma 4 checkpoints are rejected for assistant-only loss: The chat template is not training-compatible","rycerzes",4,"","",0,"2026-11-03",""],[1,8093,"CI fails with dev dependencies for selective activation checkpointing: ValueError: Unexpected keyword argument","albertvillanova",1,"albertvillanova","#8094",1,"2026-11-03","2026-11-03"],[1,8089,"GOLD Countdown collator test is always skipped: its dataset is not accessible","albertvillanova",4,"","",0,"2026-11-03",""],[1,8078,"GRPO/RLOO + vLLM: a PEFT adapter with `modules_to_save` crashes at the first weight sync","qgallouedec",1,"qgallouedec","#8079",1,"2026-11-02","2026-11-03"],[1,8076,"SFT: `padding_free=True` is refused with any `max_length`, although truncation now runs in dataset preparation","qgallouedec",1,"qgallouedec","#8077",1,"2026-11-02","2026-11-05"],[1,8071,"SFT with assistant_only_loss trains the pre-filled thinking block on Qwen3 checkpoints","albertvillanova",4,"","",0,"2026-11-02",""],[1,8059,"SFT with assistant_only_loss trains the wrong tokens on DeepSeek-R1, DeepSeek-V3 and DeepSeek-R1-Distill check","albertvillanova",1,"albertvillanova","#8061",1,"2026-11-02","2026-11-03"],[1,8056,"GRPO/RLOO under tensor parallelism: TP ranks sample different completions, and the prompt sampler is sized for","htokoyo",4,"","",0,"2026-11-01",""],[1,8052,"SFTTrainer silently skips ModernBERT Decoder's prediction head","JohnLyu2",4,"","",0,"2026-10-31",""],[1,8051,"Docs: smol-course links in the SFT, DPO and ORPO trainer pages return 404","erdogan-deniz",4,"","",0,"2026-10-31",""],[1,8050,"`parse_response` puts Qwen3.5 tool calls into `reasoning_content` when the model skips the prefilled `<think>`","luyuzhe111",4,"","",0,"2026-10-30",""],[1,8042,"Proposing to remove OnlineDPOTrainer","qgallouedec",4,"","",0,"2026-10-30",""],[1,8031,"GLM-4-MoE tool calls are parsed as plain text with transformers < 5.13","albertvillanova",1,"albertvillanova","#8032",1,"2026-10-30","2026-11-03"],[1,8028,"OpenEnv session failure test fails on the dropped group ids in the rollout buffer","albertvillanova",1,"albertvillanova","#8029",1,"2026-10-30","2026-11-03"],[1,8023,"Built-in reward functions crash with the standard dataset format","RerankerGuo",4,"","",0,"2026-10-29",""],[1,8020,"CI fails with latest dependencies for Mistral Large 4: triton.compiler.errors.CompilationError: too many indic","albertvillanova",1,"albertvillanova","#8092",1,"2026-10-29","2026-11-03"],[1,8011,"GLM-4-MoE tool-call responses parse with `<|observation|>` as content","albertvillanova",1,"albertvillanova","#8030",1,"2026-10-29","2026-11-03"],[1,8009,"Tokenizers with a TRL training chat template are rejected for tool calling: Unrecognized chat template","albertvillanova",1,"albertvillanova","#8027",1,"2026-10-29","2026-11-03"],[1,8008,"Async trainers write `lora_name` into a private attribute that `RolloutWorkerProtocol` does not declare","albertvillanova",4,"AmineDiro","#8066",1,"2026-10-29",""],[1,8004,"CI test jobs on self-hosted runners report success when pytest output stops before the end","albertvillanova",1,"albertvillanova","#8006",1,"2026-10-29","2026-11-02"],[1,8003,"[GRPO] vllm_mode=\"server\" + PEFT: last packed chunk is broadcast after unmerge_adapter(), so vLLM receives un-","Liamyu0301",4,"","",0,"2026-10-29",""],[1,7999,"SFTTrainer.evaluate() repacks prepared BFD datasets and loses document boundaries","kirill-tym",4,"","",0,"2026-10-28",""],[1,7995,"[GOLD] Byte-offset alignment for byte-fallback tokenizers (Gemma 4), not only ByteLevel","Strongich",0,"kashif","#8010",1,"2026-10-28","2026-10-29"],[1,7993,"FSDP2: reference model is zeros outside local rank 0's shard when cpu_ram_efficient_loading is on","hivaze",4,"","",0,"2026-10-28",""],[1,7991,"AsyncGRPO harness training hangs silently when every OpenEnv session fails","albertvillanova",1,"albertvillanova","#7992",1,"2026-10-28","2026-10-29"],[1,7982,"ChunkedLogProbFunction returns wrong log-probs and non-finite gradients when the support is far below the row ","albertvillanova",1,"albertvillanova","#7983",1,"2026-10-28","2026-10-29"],[1,7980,"CI fails with latest dependencies on AMD: AssertionError: Tensor-likes are not close!","albertvillanova",1,"albertvillanova","#8005",1,"2026-10-28","2026-10-29"],[1,7965,"[BUG] GRPO/RLOO continuous batching casts the trained weights to bf16 in place","harshaljanjani",4,"","",0,"2026-10-28",""],[1,7962,"Popular checkpoints are rejected for assistant-only loss: The chat template is not training-compatible","albertvillanova",4,"","",0,"2026-10-28",""],[1,7960,"Tool calling is not supported for some popular model families","albertvillanova",4,"","",0,"2026-10-28",""],[1,7955,"CI fails with vLLM for PEFT with a tied lm_head: ValueError: 'lm_head.weight' was skipped because it is tied t","albertvillanova",4,"","",0,"2026-10-28",""],[1,7954,"CI fails with vLLM for GRPO structured outputs: AssertionError: Parameter model.layers.1.self_attn.k_proj.bias","albertvillanova",4,"","",0,"2026-10-28",""],[1,7952,"VESPO with `vllm_importance_sampling_mode=\"token_mask\"` drops the whole sequence for a single out-of-range tok","yusufozdil",4,"","",0,"2026-10-27",""],[1,7950,"The `<think>`-format examples score with `accuracy_reward`, which credits a `\\boxed{}` answer inside the reaso","jimmy-holdout",4,"","",0,"2026-10-27",""],[1,7943,"Popular checkpoints are rejected for tool calling: Unrecognized chat template","albertvillanova",4,"","",0,"2026-10-27",""],[1,7936,"AsyncGRPO: `min_p` alone leaves the importance sampling ratio biased","albertvillanova",1,"albertvillanova","#7937",1,"2026-10-27","2026-10-28"],[1,7933,"The CLI training tests fail on machines without an accelerator","albertvillanova",3,"","",0,"2026-10-27","2026-10-27"],[1,7930,"The quantization tests fail on CPU with bitsandbytes < 0.49.0","albertvillanova",3,"","",0,"2026-10-27","2026-10-27"],[1,7927,"The math-verify reward tests are skipped in every CI job","albertvillanova",1,"albertvillanova","#7928",1,"2026-10-27","2026-10-28"],[1,7924,"GKD: on-policy steps normalize the loss by the dataset's token count, not the generated one","qgallouedec",1,"qgallouedec","#7925",1,"2026-10-26","2026-11-03"],[1,7923,"[Tracking] Train KTO and distillation over the same tiny models as DPO and GRPO","qgallouedec",4,"","",0,"2026-10-26",""],[1,7918,"MoE aux loss: `AttributeError: ... has no attribute 'num_experts'` on Granite MoE, granite-4.0-h and PhiMoE","qgallouedec",1,"qgallouedec","#7919",1,"2026-10-26","2026-11-04"],[1,7916,"GRPO `loss_type=\"cispo\"`: the default `epsilon_high` clips every importance weight to 0.2","qgallouedec",4,"","",0,"2026-10-26",""],[1,7914,"GRPO with vLLM: `repetition_penalty` drives the importance sampling ratio to 0 and stalls training","qgallouedec",1,"qgallouedec","#7915",1,"2026-10-26","2026-11-05"],[1,7912,"ORPO/CPO: a prompt that fills `max_length` gives an empty completion and NaN weights","qgallouedec",1,"qgallouedec","#7913",1,"2026-10-26","2026-10-27"],[1,7908,"Popular LFM2.5 checkpoints are rejected for tool calling: Unrecognized chat template","albertvillanova",2,"","",0,"2026-10-26","2026-10-27"],[1,7897,"DPO reuses stale reference scores when a custom collator's sequence limit changes","JohnLyu2",4,"","",0,"2026-10-25",""],[1,7896,"RLOO: clipped sequence ratios produce NaN gradients with a finite loss on reused long completions","taking-lying-flat",3,"","",0,"2026-10-25","2026-11-03"],[1,7895,"`parse_response` returns tool calls as plain text when `prefix` is omitted (transformers >= 5.13)","Sonica-B",4,"","",0,"2026-10-25",""],[1,7894,"DistillationTrainer ignores Inkling μP scaling and effective output vocabulary","taking-lying-flat",4,"","",0,"2026-10-25",""],[1,7892,"GOLD hybrid ULD reverse KL rewards student-only probability mass","taking-lying-flat",3,"","",0,"2026-10-25","2026-10-25"],[1,7889,"Training with the default `bf16=True` crashes on pre-Ampere GPUs (T4, V100, RTX 20xx) since v1.15","akanyaani",4,"","",0,"2026-10-24",""],[1,7888,"GKD/Distillation reject same-tokenizer pairs that only differ in padded vocab_size (Qwen2.5 0.5B↔7B, Gemma 3 1","meanwo",4,"","",0,"2026-10-24",""],[1,7887,"`ref_model_sync_steps=0` crashes a training run with ZeroDivisionError instead of failing at config creation","woodforestsen",3,"","",0,"2026-10-24","2026-10-28"],[1,7885,"CI fails for experimental tests with dev dependencies for OpenReward: ClientConnectorError: Cannot connect to ","albertvillanova",1,"albertvillanova","#7886",1,"2026-10-24","2026-10-24"],[1,7879,"Address the emSign CA certificate advisory","albertvillanova",2,"","",0,"2026-10-23","2026-10-23"],[1,7876,"The `use_reentrant=True` override for PEFT + ZeRO-3","cm8908",4,"","",0,"2026-10-23",""],[1,7874,"The GRPO and RLOO examples fail after a plain `pip install trl`: `accuracy_reward` needs math-verify","albertvillanova",1,"albertvillanova","#7875",1,"2026-10-22","2026-10-27"],[1,7871,"Test modules fail to import on their own with transformers v5 and peft 0.17: cannot import name 'HybridCache'","albertvillanova",1,"albertvillanova","#7873",1,"2026-10-22","2026-10-26"],[1,7869,"`accuracy_reward` silently returns wrong rewards in a worker thread with math-verify < 0.8.0","albertvillanova",1,"albertvillanova","#7870",1,"2026-10-22","2026-10-26"],[1,7867,"Tests error on machines without a GPU when liger-kernel is installed","albertvillanova",1,"albertvillanova","#7868",1,"2026-10-22","2026-10-23"],[1,7865,"GRPO: with top_p, top_k or min_p, the vLLM importance sampling ratio vanishes and training stalls","albertvillanova",4,"","",0,"2026-10-22",""],[1,7863,"`from trl import GRPOTrainer` fails without a visible GPU with bitsandbytes 0.45.1","albertvillanova",1,"albertvillanova","#7864",1,"2026-10-21","2026-10-22"],[1,7861,"AsyncGRPO multi-turn tool trajectories can mix policy versions and falsely drop fresh groups","lindicaphxag-tech",3,"","",0,"2026-10-21","2026-11-04"],[1,7858,"Online DPO VLM training fails for Qwen2-VL and Qwen3-VL: ValueError: mm_token_type_ids is missing","albertvillanova",1,"albertvillanova","#7859",1,"2026-10-21","2026-10-22"],[1,7857,"AsyncDistillation rollout records wrong policy version if a publish races with prompt preparation","lindicaphxag-tech",3,"","",0,"2026-10-21","2026-10-22"],[1,7854,"The minimum-versions CI job does not test deepspeed at its floor","albertvillanova",1,"albertvillanova","#7855",1,"2026-10-21","2026-10-22"],[1,7853,"Test every dependency floor at that floor","albertvillanova",4,"","",0,"2026-10-21",""],[1,7851,"Bug: SFTTrainer unconditionally casts QLoRA trainable params to bfloat16, causing GradScaler crash when fp16=T","mrajpurohit1912",4,"","",0,"2026-10-21",""],[1,7847,"Tests workflows run twice for PRs whose head branch starts with `ci-`","albertvillanova",1,"albertvillanova","#7848",1,"2026-10-21","2026-10-22"],[1,7845,"CI fails with min versions: ImportError: liger-kernel requires transformers >= 5.0.0, got 4.56.2","albertvillanova",1,"albertvillanova","#7846",1,"2026-10-21","2026-10-21"],[1,7841,"[Tracking] Promote AsyncGRPO to the stable API","qgallouedec",4,"","",0,"2026-10-20",""],[1,7840,"AsyncGRPO: two files, `async_grpo_config.py` and `async_grpo_trainer.py`","qgallouedec",4,"","",0,"2026-10-20",""],[1,7839,"AsyncGRPO metrics: `sample/` and the rollout queue","qgallouedec",4,"","",0,"2026-10-20",""],[1,7838,"AsyncGRPO metrics: `tools/`","qgallouedec",4,"","",0,"2026-10-20",""],[1,7837,"AsyncGRPO metrics: `rollout/`","qgallouedec",4,"","",0,"2026-10-20",""],[1,7836,"AsyncGRPO metrics: naming conventions","qgallouedec",4,"","",0,"2026-10-20",""],[1,7835,"AsyncGRPO metrics: same names as GRPO for the same quantities","qgallouedec",4,"","",0,"2026-10-20",""],[1,7834,"AsyncGRPO: rename `heartbeat_stale_after_s` to `rollout_worker_timeout`","qgallouedec",0,"AmineDiro","#8063",1,"2026-10-20","2026-11-04"],[1,7833,"AsyncGRPO: one timeout, `vllm_server_timeout`","qgallouedec",0,"AmineDiro","#8063",1,"2026-10-20","2026-11-04"],[1,7832,"AsyncGRPO: remove `fork_threshold_tokens`","qgallouedec",4,"","",0,"2026-10-20",""],[1,7831,"AsyncGRPO: rename `max_inflight_tasks` to `max_concurrent_rollouts`","qgallouedec",0,"AmineDiro","#8063",1,"2026-10-20","2026-11-04"],[1,7830,"AsyncGRPO: remove `queue_maxsize`","qgallouedec",4,"","",0,"2026-10-20",""],[1,7829,"AsyncGRPO: `token_budget` modes and what `per_device_train_batch_size` means","qgallouedec",4,"","",0,"2026-10-20",""],[1,7828,"AsyncGRPO: drop the `weight_transfer` argument","qgallouedec",0,"AmineDiro","#8065",1,"2026-10-20","2026-11-05"],[1,7827,"AsyncGRPO: drop the `rollout_worker` argument","qgallouedec",4,"","",0,"2026-10-20",""],[1,7826,"AsyncGRPO: add `quantization_config`","qgallouedec",4,"","",0,"2026-10-20",""],[1,7824,"GRPO and RLOO continuous batching size the KV cache to the GPU instead of the batch, making training ~4x slowe","albertvillanova",2,"","",0,"2026-10-20","2026-10-27"],[1,7822,"Online DPO: Qwen2-VL fails because mm_token_type_ids is not passed","eliem08",2,"","",0,"2026-10-20","2026-10-30"],[1,7819,"The Docker images get a different vLLM and transformers pair depending on how uv breaks a tie","albertvillanova",1,"albertvillanova","#7820",1,"2026-10-20","2026-10-21"],[1,7818,"The Docker images install huggingface_hub 1.x because trackio requires `huggingface-hub<2`","albertvillanova",2,"","",0,"2026-10-20","2026-10-21"],[1,7816,"The release Docker image keeps the uv download cache, doubling its install layer","albertvillanova",1,"albertvillanova","#7817",1,"2026-10-20","2026-10-21"],[1,7814,"The Docker images install a `kernels` version that their transformers rejects","albertvillanova",1,"albertvillanova","#7815",1,"2026-10-20","2026-10-21"],[1,7812,"The dev Docker image installs `liger_kernel` and `peft` again after the `liger` and `peft` extras","albertvillanova",1,"albertvillanova","#7813",1,"2026-10-20","2026-10-21"],[1,7811,"CI emits transformers warning: Kwargs passed to `processor.__call__` have to be in `processor_kwargs` dict, no","albertvillanova",2,"","",0,"2026-10-20","2026-10-30"],[1,7809,"CI fails for experimental tests with dev dependencies for SmolVLM: XPASS(strict)","albertvillanova",1,"albertvillanova","#7810",1,"2026-10-20","2026-10-20"],[1,7806,"[Bug] Validate BEMACallback update_freq","marcusacosta",4,"","",0,"2026-10-19",""],[1,7802,"[Bug] Validate num_prompts in completion callbacks","marcusacosta",4,"","",0,"2026-10-19",""],[1,7795,"Passing a single module to `--lora_target_modules` fails with \"Target modules not found\"","albertvillanova",1,"albertvillanova","#7796",1,"2026-10-19","2026-10-21"],[1,7794,"accuracy_reward: a `\\text{}` denial inside the box (`\\boxed{R \\text{ is wrong}}`) or a false equation (`\\boxed","Super-Eric",4,"","",0,"2026-10-19",""],[1,7790,"CI fails for experimental tests: RuntimeError: shape '[-1, 14, 16]' is invalid for input of size 16848","albertvillanova",1,"albertvillanova","#7929",1,"2026-10-19","2026-10-28"],[1,7788,"SFTTrainer drops document boundaries when only eval_packing is enabled","mottopanikeiku",4,"","",0,"2026-10-18",""],[1,7787,"GRPOTrainer: resuming from a checkpoint inside a generation block shifts the prompt schedule (same code in RLO","TarangKhanna",4,"","",0,"2026-10-18",""],[1,7782,"AsyncGRPO/AsyncDistillation merged-weight sync resumes vLLM before publishing the new rollout version","lindicaphxag-tech",3,"","",0,"2026-10-17","2026-11-04"],[1,7781,"Async OPD rollout can mislabel the sampling policy version during LoRA weight sync","lindicaphxag-tech",3,"","",0,"2026-10-17","2026-11-04"],[1,7780,"Normalize async distillation over tokens with usable teacher support","lindicaphxag-tech",3,"","",0,"2026-10-17","2026-11-04"],[1,7776,"QLoRA with fp16=True crashes: trainers cast the adapters to bf16 unconditionally","qgallouedec",4,"","",0,"2026-10-16",""],[1,7775,"SFTTrainer: chat_template_path crashes with vision-language models","qgallouedec",4,"","",0,"2026-10-16",""],[1,7772,"VLM training with vLLM in server mode ignores the images for Gemma3","albertvillanova",1,"albertvillanova","#7773",1,"2026-10-16","2026-10-19"],[1,7767,"LoRA on MoE models with shared experts does not train the dense and shared expert layers with transformers v5 ","albertvillanova",1,"albertvillanova","#7771",1,"2026-10-16","2026-10-19"],[1,7766,"AsyncGRPO: TokenBudgetBatcher closes a micro-batch even when another row has free room","clementtisseau",1,"clementtisseau","#7793",0,"2026-10-16","2026-10-22"],[1,7765,"LoRA on MoE models does not train the routed experts with transformers v5 and peft<0.19","albertvillanova",1,"albertvillanova","#7769",1,"2026-10-16","2026-10-21"],[1,7761,"[Docs] Docs for Async GRPO and the Harbor example still use Qwen3.5, which will crash on startup","shawnghu",4,"","",0,"2026-10-16",""],[1,7759,"Add an AgentEnv integration: `trl.experimental.agentenv`","earakely-scale",3,"","",0,"2026-10-15","2026-10-27"],[1,7747,"Proposal: audit zero-advantage group filtering under imperfect rewards","GBX-Max1220",3,"","",0,"2026-10-15","2026-11-02"],[1,7742,"Agree on when to add, drop and change the default Python version","albertvillanova",1,"albertvillanova","#7743",1,"2026-10-15","2026-10-19"],[1,7735,"Add an AMD GPU runner to CI","qgallouedec",0,"kashif","#5918",1,"2026-10-14","2026-10-20"],[1,7734,"DiffusionGemma SFT example evicts sampled-block prefix from sliding-window KV cache","taking-lying-flat",4,"","",0,"2026-10-14",""],[1,7733,"GRPO/RLOO on Trainium: each new prompt length bucket recompiles generation for ~12 minutes","qgallouedec",4,"","",0,"2026-10-14",""],[1,7732,"[RFC] Document torchrun as the default launcher","qgallouedec",4,"","",0,"2026-10-14",""],[1,7726,"CI emits TRL UserWarning: Cross-architecture VLM distillation detected","albertvillanova",1,"albertvillanova","#7728",1,"2026-10-14","2026-10-16"],[1,7725,"CI emits TRL FutureWarning: The `'keep_end'` truncation mode is deprecated and will be removed in v2.0.0","albertvillanova",1,"albertvillanova","#7727",1,"2026-10-14","2026-10-15"],[1,7722,"Agree on when to move the vLLM supported range","albertvillanova",4,"","",0,"2026-10-14",""],[1,7720,"Saving a checkpoint fails on torch 2.6 when generating the model card: UnicodeDecodeError: 'ascii' codec can't","albertvillanova",2,"","",0,"2026-10-14","2026-10-21"],[1,7717,"Distillation reports a peft version error for prefix tuning instead of the unsupported-method error on peft <=","albertvillanova",1,"albertvillanova","#7718",1,"2026-10-14","2026-10-16"],[1,7716,"import trl computes importlib.metadata.packages_distributions() eagerly (1-3 s with torch >= 2.14 on Python 3.","MakazhanAlpamys",0,"qgallouedec","#7947",1,"2026-10-14","2026-10-30"],[1,7715,"torch.compile(fullgraph=True) fails for selective_log_softmax_and_entropy on torch 2.6: torch._dynamo.exc.Unsu","albertvillanova",1,"albertvillanova","#7719",1,"2026-10-14","2026-10-15"],[1,7714,"CheckpointError and wrong gradients in later training tests after a training error on the same test worker","albertvillanova",1,"albertvillanova","#7797",1,"2026-10-14","2026-10-21"],[1,7709,"Proposing to remove TPOTrainer","qgallouedec",1,"qgallouedec","#7807",1,"2026-10-13","2026-10-26"],[1,7702,"Online DPO VLM training fails for processors with 5D pixel values, such as Idefics2: RuntimeError: Number of d","albertvillanova",1,"albertvillanova","#7704",1,"2026-10-13","2026-10-14"],[1,7701,"Online DPO VLM training fails when the prompt is truncated: ValueError: Image features and image tokens do not","albertvillanova",1,"albertvillanova","#7703",1,"2026-10-13","2026-10-14"],[1,7693,"keep_start truncation can make chosen and rejected DPO branches identical","fukalous",3,"","",0,"2026-10-13","2026-10-13"],[1,7692,"GRPO VLM training fails for LlavaNext with torch 2.6.0: ValueError: Image features and image tokens do not mat","albertvillanova",1,"albertvillanova","#7695",1,"2026-10-13","2026-10-15"],[1,7690,"Iterable datasets with at least as many shards as processes skip and duplicate prompts in multi-process GRPO/R","Strongich",4,"","",0,"2026-10-13",""],[1,7687,"torch_compile=True trains with zero loss and NaN gradients on torch < 2.11 with the Triton loss kernels","albertvillanova",1,"albertvillanova","#7688",1,"2026-10-13","2026-10-14"],[1,7682,"KTO: a paired dataset is unpaired into single-label blocks, so no training batch mixes desirable and undesirab","hpdkhoa",4,"","",0,"2026-10-13",""],[1,7674,"log_completions=True crashes GRPO training with tool calls: ArrowInvalid writing the completions parquet (regr","earakely-scale",0,"qgallouedec","#7696",1,"2026-10-13","2026-10-16"],[1,7673,"Use `reasoning_content` in the gpt-oss and LFM2.5 response schemas once transformers aliases reasoning fields","qgallouedec",4,"","",0,"2026-10-13",""],[1,7665,"GRPO/RLOO/Distillation: `pad_to_multiple_of` is not applied to the prompts passed to `generate`","qgallouedec",1,"qgallouedec","#7671",1,"2026-10-13","2026-10-14"],[1,7664,"GRPO/RLOO `cache_implementation=\"static\"` fails with the default fp32 model: the static cache is fp32, the key","qgallouedec",4,"","",0,"2026-10-13",""],[1,7663,"`trl env` reports `accelerator(s): cpu` on Trainium","qgallouedec",1,"qgallouedec","#7672",1,"2026-10-13","2026-10-13"],[1,7662,"`activation_offloading=True` crashes on Trainium: `torch.cuda.Stream` is used for every non-XPU/NPU accelerato","qgallouedec",1,"qgallouedec","#7670",1,"2026-10-13","2026-10-14"],[1,7661,"Docs: a Trainium (Neuron) page","qgallouedec",1,"qgallouedec","#7729",1,"2026-10-13","2026-10-14"],[1,7660,"`torch_compile=True` fails on the second step on Trainium: the Neuron dynamo backend can't run the dynamic-sha","qgallouedec",4,"","",0,"2026-10-13",""],[1,7659,"Fused LM head requires Triton: SFT, DPO, KTO, GRPO, RLOO and Distillation fail on Trainium (and any non-CUDA d","qgallouedec",1,"qgallouedec","#7669",1,"2026-10-13","2026-10-14"],[1,7658,"[Tracking] TRL on AWS Trainium (Neuron)","qgallouedec",4,"","",0,"2026-10-13",""],[1,7653,"Declare direct dependencies we currently get transitively","qgallouedec",0,"albertvillanova","#7713",1,"2026-10-12","2026-10-15"],[1,7644,"Update the dependency version floors under the agreed criterion","albertvillanova",4,"","",0,"2026-10-12",""],[1,7637,"LoRA on MoE models with shared experts does not train the routed experts with peft main","albertvillanova",1,"albertvillanova","#7636",1,"2026-10-12","2026-10-16"],[1,7630,"`DPOTrainer` crashes on the first step when `use_weighting=True` is combined with a loss_type containing \"sft\"","Sergasgr",4,"","",0,"2026-10-11",""],[1,7623,"support training decision (\"System One\") models with `SFTTrainer`","rycerzes",4,"","",0,"2026-10-10",""],[1,7621,"RFC: a `use_kernels` switch, then kernels on by default, then packing on by default","qgallouedec",4,"","",0,"2026-10-09",""],[1,7618,"SDFT and SDPO build the teacher in fp32 with the default attention when the model is passed as an instance","ningmao-hlyz",2,"","",0,"2026-10-09","2026-10-31"],[1,7617,"Add ABKD as a loss in DistillationTrainer","ghwang-s",4,"","",0,"2026-10-09",""],[1,7608,"ORPO log1mexp: finite loss but NaN gradients near zero log-probability","taking-lying-flat",3,"","",0,"2026-10-09","2026-10-14"],[1,7603,"CI fails for slow tests: no tests selected (pytest exit code 5)","albertvillanova",1,"albertvillanova","#7604",1,"2026-10-09","2026-10-09"],[1,7596,"Proposing to remove SSDTrainer","qgallouedec",1,"qgallouedec","#7706",1,"2026-10-08","2026-10-14"],[1,7592,"Support Transformer Engine FP8 training through Accelerate in TRL","sbhavani",4,"","",0,"2026-10-08",""],[1,7589,"CI emits TRL UserWarning: You are using 'rollout_func', which is an experimental feature","albertvillanova",1,"albertvillanova","#7590",1,"2026-10-08","2026-10-09"],[1,7587,"CI emits peft RuntimeWarning: The following rank_pattern keys did not match any targeted module and were ignor","albertvillanova",1,"albertvillanova","#7588",1,"2026-10-08","2026-10-13"],[1,7585,"CI emits peft UserWarning: Position ids are not supported for parameter efficient tuning","albertvillanova",1,"albertvillanova","#7586",1,"2026-10-08","2026-10-09"],[1,7580,"GOLD crashes with AttributeError when a VLM student gets a text-only teacher model","albertvillanova",1,"albertvillanova","#7581",1,"2026-10-08","2026-10-09"],[1,7576,"use_kernels=True fails with transformers 5.2.0 and 5.3.0","albertvillanova",1,"albertvillanova","#7577",1,"2026-10-08","2026-10-09"],[1,7574,"CI fails for latest TRL release with dev dependencies for LFM2 chat templates: ValueError: The chat template i","albertvillanova",2,"","",0,"2026-10-08","2026-10-09"],[1,7569,"MoE aux loss is scaled by num_processes × gradient_accumulation_steps","qgallouedec",1,"qgallouedec","#7570",1,"2026-10-08","2026-10-08"],[1,7565,"Test real objects instead of hand-built stand-ins","albertvillanova",2,"","",0,"2026-10-07","2026-10-30"],[1,7560,"Some modules fail to import when an optional dependency is missing","albertvillanova",2,"","",0,"2026-10-07","2026-10-20"],[1,7558,"Remote-code models with tied embeddings fail to load with transformers 5.3.0: NotImplementedError: Cannot copy","albertvillanova",1,"albertvillanova","#7559",1,"2026-10-07","2026-10-08"],[1,7551,"SFT with the default `chunked_nll` loss fails on a model split across devices","albertvillanova",1,"albertvillanova","#7552",1,"2026-10-07","2026-10-09"],[1,7548,"OnlineDPOTrainer ignores additional eos ids declared in model.generation_config","kumarrah2002",0,"albertvillanova","#7506",1,"2026-10-06","2026-10-07"],[1,7546,"VLM SFT collator passes processor kwargs loose to `apply_chat_template`, triggering a `processor_kwargs` warni","qgallouedec",1,"qgallouedec","#7547",1,"2026-10-06","2026-10-07"],[1,7544,"IWOPDTrainer can skip the mandatory vLLM weight sync when resuming from a checkpoint","sahil-mangla",4,"","",0,"2026-10-06",""],[1,7543,"BEMACallback crashes at on_train_begin with a PEFT (LoRA) model","sahil-mangla",3,"","",0,"2026-10-06","2026-10-12"],[1,7529,"SFT with assistant_only_loss crashes on Cohere checkpoints whose chat template has named variants","albertvillanova",1,"albertvillanova","#7530",1,"2026-10-05","2026-10-08"],[1,7526,"SFTTrainer and RewardTrainer modify the caller's peft_config when chat_template_path adds tokens","Vedant-Agarwal",4,"","",0,"2026-10-05",""],[1,7520,"Add score centering to GRPO for vLLM training-inference mismatch","vladbataev",4,"","",0,"2026-10-04",""],[1,7513,"OnlineDPOTrainer documents processing_class as optional but rejects omission","DaoyuanLi2816",4,"","",0,"2026-10-03",""],[1,7512,"Clarify raw held-out dataset evaluation for CPO and ORPO","DaoyuanLi2816",4,"","",0,"2026-10-03",""],[1,7511,"BEMACallback fails at train start with ordinary PEFT LoRA","DaoyuanLi2816",4,"","",0,"2026-10-03",""],[1,7509,"AsyncGRPOTrainer: support eval_dataset for validation reward","lewtun",4,"","",0,"2026-10-02",""],[1,7504,"GRPO ignores the end-of-turn tokens a model declares in its generation config","albertvillanova",2,"","",0,"2026-10-02","2026-10-07"],[1,7503,"Async GRPO does not train Gemma 4's tool-calling turns","albertvillanova",2,"","",0,"2026-10-02","2026-10-27"],[1,7501,"Make the upkeep of stored chat templates more robust","albertvillanova",4,"","",0,"2026-10-02",""],[1,7496,"Revisit how the tiny models track chat template revisions","albertvillanova",4,"","",0,"2026-10-02",""],[1,7492,"Supported models are rejected when their chat template differs from TRL's stored copy","albertvillanova",4,"","",0,"2026-10-02",""],[1,7484,"Remove the slow tests","albertvillanova",1,"albertvillanova","#7604",1,"2026-10-02","2026-10-09"],[1,7482,"AsyncGRPOTrainer and AsyncDistillationTrainer log the wrong `epoch` with the epoch-driven stop","lewtun",4,"","",0,"2026-10-02",""],[1,7480,"BEMACallback breaks when any parameter is frozen","Shr3yash",4,"","",0,"2026-10-01",""],[1,7478,"GRPO and RLOO crash at the first step when the model is split across devices with device_map","albertvillanova",1,"albertvillanova","#7479",1,"2026-10-01","2026-10-06"],[1,7470,"DFT loss turns every weight into NaN on a batch with no trainable tokens","albertvillanova",1,"albertvillanova","#7471",1,"2026-10-01","2026-10-02"],[1,7461,"Artist_ph ","ronaldmaguling-del",3,"","",0,"2026-09-30","2026-09-30"],[1,7449,"SFT prompt-completion trains the wrong tokens with Gemma 4 12B/26B/31B","uzunenes",0,"qgallouedec","#7463",1,"2026-09-29","2026-10-06"],[1,7445,"SFTTrainer does not append EOS when dataset_text_field is not \"text\"","JoeyTan21",1,"JoeyTan21","#7446",0,"2026-09-29","2026-09-30"],[1,7436,"CI fails with dev dependencies for Qwen3.5-MoE: AttributeError: 'int' object has no attribute 'to'","albertvillanova",1,"albertvillanova","#7460",1,"2026-09-29","2026-09-30"],[1,7430,"cast_lm_head_to_fp32 is ignored during GRPO generations (both BF16 autocast and vLLM)","ingambe",4,"","",0,"2026-09-28",""],[1,7428,"Perf Issue: Colocated vLLM sleeps and reloads weights between the tool turns of one rollout","ingambe",4,"","",0,"2026-09-28",""],[1,7423,"Merged-LoRA export changes frozen base weights","ingambe",4,"","",0,"2026-09-28",""],[1,7421,"CI fails for slow tests on multiple GPUs: RuntimeError: Expected all tensors to be on the same device, but got","albertvillanova",0,"qgallouedec","#7407",1,"2026-09-28","2026-10-06"],[1,7416,"GRPO: Tool continuation traces are discarded","ingambe",0,"qgallouedec","#7418",1,"2026-09-26","2026-09-27"],[1,7402,"[Feature request] C-GRPO: conformal adaptive rollout budgets for GRPO (NeurIPS 2026)","AryaFayyazi",4,"","",0,"2026-09-25",""],[1,7396,"Server-mode training crashes at the first weight sync on vLLM 0.20 to 0.25","albertvillanova",1,"albertvillanova","#7410",1,"2026-09-25","2026-09-26"],[1,7376,"Support assistant_only_loss for vision datasets in SFTTrainer","albertvillanova",1,"albertvillanova","#7377",1,"2026-09-24","2026-10-01"],[1,7374,"Llava-Next training chat template masks the next turn's first token as assistant output","albertvillanova",1,"albertvillanova","#7375",1,"2026-09-24","2026-09-29"],[1,7372,"CI fails with dev dependencies for Llava assistant masks: XPASS(strict)","albertvillanova",1,"albertvillanova","#7373",1,"2026-09-24","2026-09-24"],[1,7364,"OpenReward's default GRPO reward function cannot distinguish \"never scored\" from \"scored 0.0\"","shaurya416",0,"adithya-s-k","#7467",1,"2026-09-24","2026-10-06"],[1,7362,"GRPOTrainer tool discovery evaluates every property on an environment","Rome-1",1,"Rome-1","#7360",0,"2026-09-23","2026-10-03"],[1,7354,"SFTTrainer drops dataset_text_field for vision datasets with default column filtering","HelloHaoWu",4,"","",0,"2026-09-23",""],[1,7353,"`rollout_func` docstring says prompts have \"no duplication\", but each prompt is already repeated `num_generati","kuotunyu",4,"","",0,"2026-09-23",""],[1,7326,"CI rerun filter does not retry GPU allocation failures raised in a launched subprocess","albertvillanova",4,"","",0,"2026-09-21",""],[1,7320,"CI fails with minimum versions: AssertionError: Tensor-likes are not close!","albertvillanova",1,"albertvillanova","#7321",1,"2026-09-21","2026-09-21"],[1,7310,"Distributed deadlock and metric corruption in GRPOTrainer and RLOOTrainer when conditionally logging metrics/e","truecallerabreham",0,"behroozazarkhalili","#7382",1,"2026-09-20","2026-09-24"],[1,7302,"GRPO/RLOO/Distillation trainers crash on iterable datasets when dataloader_multiprocessing_context is set (def","a1bhinav",4,"","",0,"2026-09-19",""],[1,7295,"Proposing to remove MiniLLMTrainer","qgallouedec",1,"qgallouedec","#7500",1,"2026-09-18","2026-10-06"],[1,7291,"Support custom/composable advantage computation across trainers","art-test-stack",4,"","",0,"2026-09-18",""],[1,7274,"AsyncGRPO: one stale-dropped group pins the saved resume position for the rest of the run","cm8908",4,"","",0,"2026-09-18",""],[1,7270,"[Bug] GMPOTrainer with use_liger_kernel=True trains on the GRPO objective","gss10282025",3,"","",0,"2026-09-18","2026-09-18"],[1,7268,"SFTTrainer's QLoRA fp16 downcast also hits QDoRA magnitude vector, causing silent no-op updates","tardelr",0,"qgallouedec","#7342",1,"2026-09-18","2026-10-02"],[1,7265,"[Bug] BCO uses different running baselines within one gradient-accumulation window","gss10282025",2,"","",0,"2026-09-17","2026-09-17"],[1,7245,"DPO and KTO skip DDP gradient synchronization with use_liger_kernel=True","qgallouedec",4,"","",0,"2026-09-16",""],[1,7244,"Fused Triton kernels for the loss head","qgallouedec",4,"","",0,"2026-09-16",""],[1,7223,"Dataset preparation runs once per rank on every node except rank 0's","qgallouedec",1,"qgallouedec","#7241",1,"2026-09-16","2026-09-17"],[1,7222,"router_aux_loss_coef is silently ignored on MoE configs without output_router_logits","qgallouedec",1,"qgallouedec","#7248",1,"2026-09-16","2026-09-23"],[1,7221,"entropy_from_logits returns NaN for zero-probability tokens","vineethsaivs",3,"","",0,"2026-09-15","2026-09-17"],[1,7220,"Apple Silicon (MPS): create_model_from_path can segfault while loading bf16 checkpoints","Saheb",0,"albertvillanova","#7455",1,"2026-09-15","2026-10-01"],[1,7211,"Label the pull requests merged since v1.13","albertvillanova",2,"","",0,"2026-09-15","2026-09-15"],[1,7206,"AsyncGRPOTrainer: loss normalization across gradient accumulation and complete-rollout admission","adithya-s-k",4,"","",0,"2026-09-14",""],[1,7202,"Audit the trl-internal-testing repos that have no generation script","albertvillanova",4,"","",0,"2026-09-14",""],[1,7198,"vLLM encoder cache is not reset after weight sync","rmaguado",4,"","",0,"2026-09-13",""],[1,7189,"Online DPO: Idefics2 image batches fail in vision tensor duplication and scoring preparation","DaoyuanLi2816",4,"","",0,"2026-09-12",""],[1,7182,"[RFC] Criteria for removing code from `trl.experimental`","albertvillanova",2,"","",0,"2026-09-11","2026-10-16"],[1,7181,"DPO/KTO training goes NaN with the default SDPA backend on torch 2.13 (cuDNN attention); clean with `enable_cu","qgallouedec",2,"","",0,"2026-09-11","2026-10-03"],[1,7179,"[GRPO] Pad per minibatch instead of per generation batch","qgallouedec",4,"","",0,"2026-09-10",""],[1,7169,"Where the GPU runner hours go in the Tests workflow","albertvillanova",4,"","",0,"2026-09-10",""],[1,7137,"Align tiny test model configs with their reference models","albertvillanova",4,"","",0,"2026-09-09",""],[1,7122,"CI fails for experimental tests: AssertionError: assert -7.018252081922305e-13 >= 0.0","albertvillanova",1,"albertvillanova","#7123",1,"2026-09-09","2026-09-09"],[1,7116,"Agree on an explicit criterion for raising a dependency version floor","albertvillanova",1,"albertvillanova","#7638",1,"2026-09-08","2026-10-13"],[1,7103,"AsyncGRPO FSDP2 weight sync stalls when the vLLM manifest is bf16 and the trainer streams fp32","sfc-gh-truwase",1,"sfc-gh-truwase","#7131",0,"2026-09-08","2026-10-25"],[1,7100,"`num_tokens` overcounted under TP","qgallouedec",1,"qgallouedec","#7101",1,"2026-09-07","2026-09-09"],[1,7093,"CI emits 462 `transformers` `Trainer` log warnings: \"The tokenizer has new PAD/BOS/EOS tokens that differ from","albertvillanova",2,"","",0,"2026-09-07","2026-10-12"],[1,7086,"RLOO uses higher-variance KL Estimator and should use lower variance k3","AMindToThink",3,"","",0,"2026-09-06","2026-10-22"],[1,7083,"Support precomputed reference log-probs in fused DPO training","alay2shah",3,"","",0,"2026-09-06","2026-09-15"],[1,7071,"SFTTrainer treats an empty config._name_or_path as a Hub repo id","victorzhong0110",3,"","",0,"2026-09-05","2026-09-06"],[1,7063,"[Tracking] Move off liger-kernel and make the memory-efficient loss the default","qgallouedec",4,"","",0,"2026-09-04",""],[1,7049,"RLOO _tokenize_prompts skips VLM string-content normalization that GRPO and Distillation apply","YeonwooSung",4,"","",0,"2026-09-04",""],[1,7048,"Docs hygiene: paper_index path/typos, GOLD subsection, IW-OPD page, AsyncDistillation telemetry","YeonwooSung",1,"YeonwooSung","#7051",0,"2026-09-04","2026-09-10"],[1,7047,"GRPO loss-path metrics average micro-batch entries, so uneven accumulation windows are weighted by their size","behroozazarkhalili",3,"","",0,"2026-09-04","2026-09-17"],[1,7035,"Async trainers take an extra optimizer step when resuming at the epoch target","DaoyuanLi2816",4,"","",0,"2026-09-03",""],[1,7031,"Bug: CLI --accelerate_config does not override YAML config","YZJF",4,"","",0,"2026-09-03",""],[1,7028,"Does hf jobs uv run --image huggingface/trl actually use the preinstalled TRL?","albertvillanova",1,"albertvillanova","#7054",1,"2026-09-03","2026-09-04"],[1,7027,"Verify the released Docker image build end to end, including the push, after the release trigger change","albertvillanova",2,"","",0,"2026-09-03","2026-09-03"],[1,7025,"create_model_from_path ignores subfolder when inferring the model architecture","DaoyuanLi2816",3,"","",0,"2026-09-03","2026-10-09"],[1,7024,"MiniLLM: right-padding leaks into the reverse-KL advantage, and length normalization depends on batch padding","behroozazarkhalili",3,"","",0,"2026-09-02","2026-09-17"],[1,7021,"Replace the introspection guard in `sft_qwen3_8b_1m_context.py` with a plain version comparison","qgallouedec",0,"kumarrah2002","#7537",0,"2026-09-02","2026-10-06"],[1,7018,"CI emits `torch` UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars","albertvillanova",1,"albertvillanova","#7019",1,"2026-09-02","2026-09-02"],[1,7015,"get_repetition_penalty_reward accepts non-positive ngram_size","qsxustc",3,"","",0,"2026-09-02","2026-10-01"],[1,7012,"PPOTrainer statistics average per-micro-batch means without weighting by micro-batch size","behroozazarkhalili",3,"","",0,"2026-09-02","2026-09-04"],[1,7011,"policy_loss is logged at two different scales depending on loss_type","behroozazarkhalili",3,"","",0,"2026-09-02","2026-09-17"],[1,7010,"Metrics are reduced to a rank-local scalar before gather_for_metrics, biasing the final batch","behroozazarkhalili",3,"","",0,"2026-09-02","2026-09-17"],[1,7009,"GRPOTrainer silently drops the MoE auxiliary loss when use_liger_kernel=True","behroozazarkhalili",2,"","",0,"2026-09-02","2026-09-24"],[1,7007,"XPO Trainer data update step","jhuang265",2,"","",0,"2026-09-02","2026-09-02"],[1,7006,"create_model_from_path crashes when config lacks architecture metadata","YZJF",4,"","",0,"2026-09-02",""],[1,7005,"`policy_loss` is never logged by GMPOTrainer, gspo_token's GRPOTrainer, or RLOOTrainer","behroozazarkhalili",3,"","",0,"2026-09-02","2026-09-17"],[1,6996,"BCOTrainer fails before the first step on any encoder-decoder model","behroozazarkhalili",2,"","",0,"2026-09-01","2026-09-17"],[1,6989,"Docker image version tags are mutable: huggingface/trl:1.9.2 was rebuilt on 65 pushes after its release","albertvillanova",2,"","",0,"2026-09-01","2026-09-02"],[1,6984,"Padding-free warning fires for revision-pinned hub kernels (`repo_id@revision`)","albertvillanova",1,"albertvillanova","#6985",1,"2026-08-31","2026-09-02"],[1,6981,"🐛 [Bug] get_dataset raises ZeroDivisionError for zero-sum fractions","YZJF",3,"","",0,"2026-08-31","2026-10-14"],[1,6974,"Automatic processor loading ignores the model revision","edbeeching",0,"kashif","#6978",1,"2026-08-29","2026-09-09"],[1,6973,"VLLMClient sets no request timeout, so a stalled vLLM server hangs training indefinitely","behroozazarkhalili",3,"","",0,"2026-08-29","2026-09-17"],[1,6970,"Bug: pack_dataset BFD fails when all sequences are empty","YZJF",3,"","",0,"2026-08-29","2026-09-05"],[1,6967,"Online DPO vLLM server fails on mixed image and text-only batches","DaoyuanLi2816",4,"","",0,"2026-08-29",""],[1,6966,"`think_format_reward` always returns 0.0 for models whose chat template prefills `<think>`","A-ryanVAT-S",4,"","",0,"2026-08-28",""],[1,6961,"CI emits `transformers` warning log: Unrecognized keys in `rope_parameters` for 'rope_type'='default': {'full_","albertvillanova",1,"albertvillanova","#6962",1,"2026-08-28","2026-08-31"],[1,6951,"XPOTrainer accepts reward_processing_classes but never uses it, and NashMDTrainer does not accept it at all","behroozazarkhalili",3,"","",0,"2026-08-27","2026-09-18"],[1,6945,"AsyncGRPO has no importance-sampling correction, but TRL still imposes --logprobs-mode processed_logprobs on i","behroozazarkhalili",2,"","",0,"2026-08-27","2026-10-16"],[1,6941,"publish.yml cannot tell a release bump from a dev bump — the v1.12.0 upload reproduces at v1.13","tonydzi",3,"","",0,"2026-08-26","2026-08-26"],[1,6928,"prepare_multimodal_messages_vllm fails on content-less tool-call messages","YZJF",0,"behroozazarkhalili","#6946",1,"2026-08-26","2026-10-02"],[1,6917,"CI rerun filter does not retry GPU allocation failures wrapped in another exception","albertvillanova",1,"albertvillanova","#7325",1,"2026-08-25","2026-09-22"],[1,6914,"CI fails with dev dependencies for MoE LoRA target_parameters: ValueError: Registering a parametrization may n","albertvillanova",1,"albertvillanova","#6919",1,"2026-08-25","2026-08-25"],[1,6890,"CI warning: Node.js 20 is deprecated (slackapi/slack-github-action in hf-workflows post-slack)","albertvillanova",1,"albertvillanova","#6896",1,"2026-08-24","2026-08-25"],[1,6888,"CI emits `datasets` UserWarning: Downcasting array dtype int64 to uint8 to be compatible with 'Pillow'","albertvillanova",1,"albertvillanova","#6889",1,"2026-08-24","2026-08-25"],[1,6879,"IPO and sigmoid_norm count an unscored token after keep_end truncation","primorLee",3,"","",0,"2026-08-23","2026-08-24"],[1,6877,"Bug: unpair IterableDatasetDict preference datasets","YZJF",4,"","",0,"2026-08-23",""],[1,6874,"Bug: normalize --accelerate_config=<name> CLI syntax","YZJF",4,"","",0,"2026-08-23",""],[1,6857,"Tiny test models are incompatible with Flash Attention (head_size=2)","albertvillanova",4,"","",0,"2026-08-21",""],[1,6852,"compute_flops_per_token fails on configs without head_dim","albertvillanova",1,"albertvillanova","#6855",1,"2026-08-21","2026-08-21"],[1,6849,"CI emits torch inductor UserWarning: TensorFloat32 tensor cores for float32 matrix multiplication available bu","albertvillanova",1,"albertvillanova","#6851",1,"2026-08-21","2026-08-21"],[1,6848,"CI emits NumPy DeprecationWarning: `__array_wrap__` must accept context and return_scalar arguments","albertvillanova",1,"albertvillanova","#6850",1,"2026-08-21","2026-08-21"],[1,6837,"CI fails for experimental tests: RuntimeError: mha_fwd, head_size should be a multiple of 8","albertvillanova",1,"albertvillanova","#6854",1,"2026-08-20","2026-08-21"],[1,6836,"CI fails for slow tests on multiple GPUs: Fatal Python error: Segmentation fault","albertvillanova",0,"qgallouedec","#7407",1,"2026-08-20","2026-10-06"],[1,6834,"CI fails for slow tests: RuntimeError: mha_fwd, head_size should be a multiple of 8","albertvillanova",1,"albertvillanova","#6853",1,"2026-08-20","2026-08-21"],[1,6833,"RFC: Add single-image vision-language model support to A2POTrainer","DimensionSTP",4,"","",0,"2026-08-20",""],[1,6832,"RFC: Add end-to-end vision-language model support to SDPOTrainer","DimensionSTP",4,"","",0,"2026-08-20",""],[1,6831,"RFC: Support A2PO with DeepSpeed ZeRO-3 via an explicit reference model","DimensionSTP",3,"","",0,"2026-08-20","2026-10-02"],[1,6828,"SDPO PEFT EMA initialization reads partitioned state under ZeRO-3","DimensionSTP",4,"","",0,"2026-08-20",""],[1,6809,"Metrics-only inaccuracies in RL trainers: unmasked stats (OnlineDPO, PPO), wrong XPO logps metric, GRPO policy","mmjerge",4,"","",0,"2026-08-19",""],[1,6808,"GMPO: use_liger_kernel=True silently bypasses the GMPO loss; MoE aux loss and entropy bonus dropped (gaps not ","mmjerge",4,"","",0,"2026-08-19",""],[1,6807,"gspo_token: _compute_loss is a stale copy of GRPO's — use_bias_correction_kl, OPSM, entropy bonus, and MoE aux","mmjerge",3,"","",0,"2026-08-19","2026-09-23"],[1,6806,"OnlineDPO/XPO/Nash-MD: list-valued beta/alpha/mixture_coef epoch schedules crash (list indexed with float stat","mmjerge",4,"","",0,"2026-08-19",""],[1,6805,"XPO/Nash-MD: missing_eos_penalty checks the full sequence including the prompt, never applies with EOS-in-prom","mmjerge",3,"","",0,"2026-08-19","2026-09-18"],[1,6804,"GRPOWithReplayBufferTrainer: replay machinery silently degenerates through the trainer call path (shape contra","mmjerge",3,"","",0,"2026-08-19","2026-09-30"],[1,6803,"Online DPO: `objective/kl` and `objective/entropy` include padding tokens","mmjerge",2,"","",0,"2026-08-19","2026-08-19"],[1,6802,"XPO / Nash-MD: `missing_eos_penalty` checks the prompt too, silently disabling it for chat templates whose EOS","mmjerge",2,"","",0,"2026-08-19","2026-08-19"],[1,6801,"Online DPO / XPO / Nash-MD: epoch-scheduled `beta`/`alpha`/`mixture_coef` crash with `TypeError` on a float li","mmjerge",2,"","",0,"2026-08-19","2026-08-19"],[1,6800,"Online DPO / XPO / Nash-MD logprob forwards omit `position_ids` under left-padded prompts","mmjerge",3,"","",0,"2026-08-19","2026-09-30"],[1,6797,"AsyncGRPO: NCCL weight transfer can hang forever with no error, and deadlocks whenever trainer world size exce","qgallouedec",4,"","",0,"2026-08-19",""],[1,6796,"AsyncGRPO: NCCL weight transfer can hang forever with no error","qgallouedec",0,"AmineDiro","#6913",1,"2026-08-19","2026-08-28"],[1,6791,"Dict-typed config fields can't be set from the CLI (only model_init_kwargs works)","qgallouedec",1,"qgallouedec","#6792",1,"2026-08-18","2026-08-20"],[1,6789,"GRPO: vLLM importance-sampling ratio is biased when top_p/top_k/min_p truncate sampling","qgallouedec",0,"AmineDiro","#7627",1,"2026-08-18","2026-10-16"],[1,6780,"Bug in Generalized JSD Loss formula and endpoint descriptions","aunaun913",3,"","",0,"2026-08-17","2026-08-25"],[1,6778,"CI fails: ValueError: Pointer argument cannot be accessed from Triton (cpu tensor?)","qgallouedec",0,"albertvillanova","#6783",1,"2026-08-17","2026-08-18"],[1,6776,"GRPOTrainer vLLM colocate: LLM() init does not forward hf_overrides (blocks text-only Gemma4 / multimodal-arch","marksverdhei",3,"","",0,"2026-08-17","2026-11-03"],[1,6773,"Nemotron 3 SFT example uses target_modules that may not exist in hybrid architecture","davidnichols-ops",2,"","",0,"2026-08-17","2026-11-04"],[1,6767,"Bump liger-kernel to >=0.8.2 once released (drop the !=0.8.1 exclusion)","albertvillanova",1,"albertvillanova","#6768",1,"2026-08-15","2026-08-19"],[1,6746,"Add Weighted Instruction Tuning (WIT) loss to SFTTrainer","C-anwoy",3,"","",0,"2026-08-14","2026-10-20"],[1,6743,"Make TRL's skill discoverable from an install, and drop `trl skills install`","qgallouedec",2,"","",0,"2026-08-13","2026-09-24"],[1,6734,"In DistillationTrainer, it is not possible to have the teacher model on a vllm server.","Palipoor",4,"","",0,"2026-08-13",""],[1,6732,"Migrate async GRPO weight transfer to vLLM's stateful engine API","qgallouedec",2,"","",0,"2026-08-13","2026-10-02"],[1,6720,"having separate quantization configs for the teacher and the student","qgallouedec",0,"behroozazarkhalili","#6769",1,"2026-08-12","2026-09-16"],[1,6711,"CI fails for slow tests: RuntimeError: min(): Expected reduction dim to be specified for input.numel() == 0","albertvillanova",2,"","",0,"2026-08-12","2026-10-02"],[1,6708,"Accounting error in FLOPS calculation","Harindhar10",0,"jayzuccarelli","#7450",0,"2026-08-11","2026-09-30"],[1,6702,"GRPO: a NaN training step is invisible in log_history, so a corrupted run reports a clean loss curve","behroozazarkhalili",3,"","",0,"2026-08-10","2026-09-17"],[1,6700,"GRPOTrainer silently no-ops (empty optimizer) when continuing from a PeftModel loaded with is_trainable=False ","Meredith2328",3,"","",0,"2026-08-10","2026-09-18"],[1,6697,"Qwen3.5 VLM Liger GRPO drops multimodal token type IDs","DimensionSTP",2,"","",0,"2026-08-10","2026-10-06"],[1,6696,"SFT chunked NLL stalls under ZeRO-3 with unequal rank chunk counts","DimensionSTP",2,"","",0,"2026-08-10","2026-10-29"],[1,6688,"Support status and numerical safety of GRPO + vLLM + bf16 LoRA for long multi-turn tool rollouts in TRL 1.9.2","yanan1116",2,"","",0,"2026-08-08","2026-10-16"],[1,6685,"SFTTrainer's default chunked_nll path fails on models whose output has no past_key_values","Hakureirm",1,"Hakureirm","#6686",0,"2026-08-08","2026-08-14"],[1,6684,"AsyncGRPOTrainer hardcodes dtype=torch.float32, so a 4B OOMs on an 80 GB card","Mandark-droid",2,"","",0,"2026-08-08","2026-09-09"],[1,6683,"examples/scripts/openenv/opencode.py is broken against current OpenEnv@main - every rollout scores \"unscorable","Mandark-droid",2,"","",0,"2026-08-08","2026-09-15"],[1,6679,"[Bug] `trl vllm-serve` fails with `tensor_parallel_size > 1`: Cannot re-initialize CUDA in forked subprocess","linfeng-du",2,"","",0,"2026-08-08","2026-09-22"],[1,6669,"`pack_dataset` with `strategy=\"wrapped\"` silently duplicates data across `map` batches","SuryanshSS1011",1,"SuryanshSS1011","#6670",0,"2026-08-06","2026-09-17"],[1,6668,"SFTTrainer reports loss 0.0 when every label in the batch is masked","behroozazarkhalili",3,"","",0,"2026-08-05","2026-08-12"],[1,6663,"Completion callbacks leave tokenizer in left-padding mode","RerankerGuo",3,"","",0,"2026-08-04","2026-09-29"],[1,6661,"Reasoning reward uses delimiter-list order instead of text order","RerankerGuo",3,"","",0,"2026-08-04","2026-09-09"],[1,6659,"Generation context does not restore state after errors","RerankerGuo",3,"","",0,"2026-08-04","2026-10-26"],[1,6657,"Skills CLI cannot uninstall symlinked skills","RerankerGuo",3,"","",0,"2026-08-04","2026-08-06"],[1,6655,"TrlParser exposes TypeErrors for non-mapping YAML configs","RerankerGuo",1,"RerankerGuo","#6656",0,"2026-08-04","2026-08-10"],[1,6649,"Make config files the primary interface: `trl train config.yaml`","qgallouedec",4,"","",0,"2026-08-03",""],[1,6636,"Empty conversational fields raise IndexError during format detection","cupkk",3,"","",0,"2026-08-03","2026-08-07"],[1,6626,"MiniLLM: `_compute_advantage` discounts by absolute position, scaling every advantage by gamma^t","MehrdadMoghimi",3,"","",0,"2026-08-02","2026-10-06"],[1,6622,"FSDP2 vLLM weight sync crashes with ignored parameters","Rusheel86",4,"","",0,"2026-08-02",""],[1,6617,"Training a VLM on a text-only dataset leaves the vision tower trainable — DDP crash, and wasted optimizer stat","qgallouedec",4,"","",0,"2026-08-01",""],[1,6586,"KL bias correction with importance_sampling_level=\"sequence\" broadcasts a sequence-level ratio onto per-token ","gowtham-sai-yadav",4,"","",0,"2026-07-30",""],[1,6584,"CI warns with dev dependencies: FutureWarning: `device` is deprecated and will be removed in version 5.18 for ","albertvillanova",1,"albertvillanova","#6585",1,"2026-07-30","2026-07-30"],[1,6580,"CI emits `bitsandbytes` UserWarning: inner dimension (8) is not aligned for fast kernel with blocksize=64","albertvillanova",1,"albertvillanova","#6582",1,"2026-07-29","2026-07-30"],[1,6567,"Explore QAT support in TRL","qgallouedec",4,"","",0,"2026-07-28",""],[1,6561,"Add support for vLLM 0.26.0","qgallouedec",2,"","",0,"2026-07-28","2026-09-01"],[1,6555,"Remove kernels-community/mamba-ssm autocast warning filter once the Hub kernel build stops calling `torch.get_","albertvillanova",4,"","",0,"2026-07-27",""],[1,6553,"GRPO materializing copies without need","jadermcs",0,"DaoyuanLi2816","#6554",0,"2026-07-27","2026-07-28"],[1,6548,"[CPO/SimPO] Truncation in CPOTrainer can silently produce empty completions -> NaN loss with loss_type=\"simpo\"","ninjafrmtjk",0,"RohanMali2003","#6588",0,"2026-07-26","2026-10-03"],[1,6545,"Packing is refused for VLMs even on text-only datasets","qgallouedec",0,"cris96spa","#6547",0,"2026-07-26","2026-07-26"],[1,6543,"CI fails: AssertionError: Parameter base_model.model.model.layers.0.self_attn.q_proj.base_layer.bias has chang","albertvillanova",1,"albertvillanova","#6546",1,"2026-07-25","2026-07-28"],[1,6541,"CI fails with dev dependencies: RuntimeError: expand(CUDABFloat16Type{[3, 1, 1, 8]}, size=[-1, -1, 4]): the nu","albertvillanova",1,"albertvillanova","#6558",1,"2026-07-25","2026-07-27"],[1,6536,"CI (transformers dev): `TypeError: causal_conv1d_fn() got multiple values for argument 'seq_idx'` on NemotronH","qgallouedec",2,"","",0,"2026-07-24","2026-07-24"],[1,6533,"GRPO silently accepts a reward function that returns the wrong number of rewards","gowtham-sai-yadav",1,"gowtham-sai-yadav","#6534",0,"2026-07-24","2026-07-24"],[1,6532,"CI fails with dev dependencies for Gemma4: ValueError: Unrecognized chat template, failed to add response sche","albertvillanova",0,"qgallouedec","#6535",1,"2026-07-24","2026-07-24"],[1,6519,"Support image paths and URLs in vLLM generation","alay2shah",4,"","",0,"2026-07-23",""],[1,6516,"CI fails: RuntimeError: gemm input type at::BFloat16 and output type float is only supported for CUDA devices ","albertvillanova",1,"albertvillanova","#6518",1,"2026-07-23","2026-07-24"],[1,6506,"ImportError: cannot import name 'setup_chat_format' from 'trl' (/usr/local/lib/python3.12/dist-packages/trl/__","acg12",2,"","",0,"2026-07-22","2026-07-22"],[1,6483,"SFTTrainer crashes on Qwen3.5 because forward is functools.partial","Eraly-ml",2,"","",0,"2026-07-21","2026-07-27"],[1,6473,"Implement single-rollout asynchronous optimization","kirawi",4,"","",0,"2026-07-20",""],[1,6471,"Add multi-GPU precompute test coverage for KTO","qgallouedec",1,"qgallouedec","#7507",1,"2026-07-20","2026-10-05"],[1,6470,"`precompute_ref_log_probs=True` crashes under FSDP2 (device mismatch)","qgallouedec",4,"","",0,"2026-07-20",""],[1,6466,"Remove triton `AnnAssign` warning filter once no supported torch resolves a triton without the fix (>= 3.8.0)","albertvillanova",4,"","",0,"2026-07-20",""],[1,6462,"Enforce PR-only changes to main via branch protection / rulesets","albertvillanova",2,"","",0,"2026-07-20","2026-08-07"],[1,6449,"Refactor `DistillationTrainer` and promote it to the stable API","qgallouedec",2,"","",0,"2026-07-20","2026-08-17"],[1,6447,"Remove bitsandbytes `_check_is_size` warning filter once the minimum required bitsandbytes is >= 0.50.0","albertvillanova",4,"","",0,"2026-07-19",""],[1,6441,"DPOTrainer apo_down loss inconsistently applies f_divergence_type across its two terms","ErenAta16",0,"qgallouedec","#6559",1,"2026-07-18","2026-07-31"],[1,6435,"Remove SAPO `IndexPutBackward0` warning filter once `liger-kernel >= 0.8.2` is required","albertvillanova",1,"albertvillanova","#6768",1,"2026-07-17","2026-08-19"],[1,6412,"CI is failing for experimental tests: AttributeError: 'GeometricMixtureWrapper' object has no attribute 'get_e","albertvillanova",1,"albertvillanova","#6413",1,"2026-07-16","2026-07-16"],[1,6408,"Your project is on StackMap — a curated map of the AI stack","hoghweed",2,"","",0,"2026-07-15","2026-07-15"],[1,6407,"Add UP (Unbounded Positive) loss as a GRPO loss_type","yashb98",3,"","",0,"2026-07-15","2026-10-06"],[1,6401,"GRPO VLM colocate: pre-tokenized prompt ids + raw image handoff mis-aligns with vLLM preprocessing (M-RoPE cra","guangboyu",3,"","",0,"2026-07-15","2026-07-16"],[1,6369,"GRPO: dapo/cispo/vespo loss denominator ignores `mask_truncated_completions` (the numerator doesn't)","SushantDaga",0,"Functionhx","#6439",0,"2026-07-13","2026-07-18"],[1,6367,"Option to exclude zero-variance/dead groups from the loss denominator, accuracy increases","SushantDaga",3,"","",0,"2026-07-13","2026-07-20"],[1,6361,"Critical training chat-template issue with Qwen3.5 models (in `chat_template_utils`)","giladfrid009",0,"qgallouedec","#6363",1,"2026-07-11","2026-07-16"],[1,6354,"[harbor/openreward] Adopt environment-owned reward (#6238) and dataset (#6349)","adithya-s-k",4,"","",0,"2026-07-10",""],[1,6350,"Feature request: reward-hacking onset monitoring during GRPO/DPO training","Aarav500",3,"","",0,"2026-07-10","2026-07-12"],[1,6311,"Tracking: duplicate \"teacher\" adapter registration gap in SDFTTrainer(teacher_model_kind=\"ema\") — closed in #6","ucalyptus",3,"","",0,"2026-07-06","2026-07-10"],[1,6294,"GRPO + vLLM corrupts SmolVLM multimodal prompts from pre-expanded image tokens","Strongich",4,"","",0,"2026-07-05",""],[1,6291,"DPOTrainer fails with FileNotFoundError when `precompute_ref_log_probs=True`","saattrupdan",0,"Functionhx","#6348",0,"2026-07-05","2026-07-09"],[1,6274,"Multi-turn tool calling in GRPO breaks Qwen VLM training (wrong pixel_values slice in _tool_call_loop)","shaunliew",0,"HariomJangra","#6286",0,"2026-07-03","2026-08-12"],[1,6265,"CI fails with dev dependencies: RuntimeError: The size of tensor a (4) must match the size of tensor b (8) at ","albertvillanova",1,"albertvillanova","#6268",1,"2026-07-03","2026-07-03"],[1,6263,"Proposal: Change default `max_completion_length` from 256 to 512","dhruvnigam93",1,"dhruvnigam93","#6264",0,"2026-07-03","2026-07-23"],[1,6221,"[Feature] Reasoning reward utilities: group-relative length-scaled accuracy (GRPO-LEAD) + graduated format rew","surajsharan",4,"","",0,"2026-06-30",""],[1,6219,"Add warning when all completions are truncated","dhruvnigam93",3,"","",0,"2026-06-30","2026-07-22"],[1,6210,"SuperCompress: Cut LLM token costs by ~65% with open-source prompt compression","arjunkshah",3,"","",0,"2026-06-30","2026-06-30"],[1,6177,"DPOTrainer: ref_model is None during eval when gradient_checkpointing=True","ajwharton",3,"","",0,"2026-06-25","2026-06-25"],[1,6168,"TypeError: Trainer.__init__() got an unexpected keyword argument 'tokenizer'","YushuYan064",2,"","",0,"2026-06-24","2026-06-24"],[1,6166,"GRPO Trainer crashes with Could not infer dtype of NoneType when vLLM returns a NaN token logprob","dwarez",0,"behroozazarkhalili","#6693",1,"2026-06-24","2026-08-13"],[1,6165,"Support encoder-decoder models in SFTTrainer for prompt-completion datasets","GabrielFreeze",3,"","",0,"2026-06-24","2026-06-24"],[1,6151,"Support router-based multi-teacher (MOPD) distillation in DistillationTrainer","mohanpb",4,"","",0,"2026-06-23",""],[1,6133,"AsyncGRPO crashes with documented picklable reward callables","siddiksawani",0,"qgallouedec","#6313",1,"2026-06-21","2026-07-14"],[1,6129,"`GRPOTrainer.evaluate()` with `environment_factory` crashes on partial batches","qgallouedec",2,"","",0,"2026-06-21","2026-07-06"],[1,6124,"`_get_tool_suffix_ids`: EOS-trimmed prefix mismatch with Qwen3 due to `loop.last` in chat template","linfeng-du",2,"","",0,"2026-06-19","2026-06-20"],[1,6115,"SFTTrainer.evaluate() to accept the same kinds of datasets as train()","DavidNemeskey",0,"qgallouedec","#6116",1,"2026-06-19","2026-06-20"],[1,6105,"Get [RANK 0] Mismatch between tokenized prompt ... warning for every training example","DavidNemeskey",2,"","",0,"2026-06-18","2026-06-18"],[1,6102,"Tests (experimental): test_gkd_trainer_with_liger intermittently OOMs the T4 runner, failing unrelated PRs","behroozazarkhalili",1,"behroozazarkhalili","#6103",1,"2026-06-18","2026-06-18"],[1,6089,"CI fails for distributed smoke tests: TypeError: output tensor must have the same type as input tensor","albertvillanova",1,"albertvillanova","#6091",1,"2026-06-17","2026-06-19"],[1,6076,"chunked_nll under FSDP2 triggers per-chunk all-gathers of lm_head.weight during backward","albertvillanova",1,"albertvillanova","#6077",1,"2026-06-16","2026-06-19"],[1,6067,"`AsyncGRPOTrainer`: the `environment_factory` `reset()` observation is dropped — the policy trains on the bare","adithya-s-k",2,"","",0,"2026-06-15","2026-07-03"],[1,6064,"CI fails for distributed smoke tests: RuntimeError: aten.mm.default got mixed torch.Tensor and DTensor","albertvillanova",1,"albertvillanova","#6065",1,"2026-06-15","2026-06-16"],[1,6056,"Add GMPO as an experimental trainer","raghulchandramouli",1,"raghulchandramouli","#6078",0,"2026-06-14","2026-06-24"],[1,6039,"`SFTTrainer` silently breaks datasets that use `Dataset.with_transform`","proshian",0,"he-yufeng","#6054",0,"2026-06-13","2026-06-16"],[1,6028,"AsyncGRPOTrainer: vision-language (*ForConditionalGeneration) checkpoints can't be trained (weight-sync key mi","adithya-s-k",2,"","",0,"2026-06-12","2026-09-23"],[1,6027,"AsyncGRPOTrainer: environment_factory rewards are never passed to reward_funcs","adithya-s-k",2,"","",0,"2026-06-12","2026-07-16"],[1,6017,"Support training on Harbor agentic task suites via environment_factory","adithya-s-k",1,"adithya-s-k","#6018",1,"2026-06-11","2026-06-18"],[1,5975,"Add adapter-only vLLM LoRA sync for GRPO and AsyncGRPO","rycerzes",4,"","",0,"2026-06-08",""],[1,5974,"RFC: Rollout sources — a unified seam for environment & agent RL","adithya-s-k",3,"","",0,"2026-06-08","2026-08-19"],[1,5973,"trackio ProfilingContext support","VikramRangarajan",0,"Anai-Guo","#5979",0,"2026-06-08","2026-06-12"],[1,5935,"A*PO: Accelerating RL for LLM Reasoning with Optimal Advantage regression","raghulchandramouli",2,"","",0,"2026-06-04","2026-06-05"],[1,5912,"RFC: Environment-owned reward","qgallouedec",2,"","",0,"2026-06-02","2026-07-09"],[1,5903,"RFC: Agent RL training with environment: environment-owned dataset, if any","qgallouedec",1,"qgallouedec","#6349",1,"2026-06-01","2026-07-20"],[1,5889,"bug(rloo): KL penalty uses first-order log ratio instead of Schulman approximation (inconsistent with GRPOTrai","Sumu004",0,"abderahmane-ai","#6096",0,"2026-05-30","2026-06-18"],[1,5886,"RDPO support in GRPOTRainer","JishnuKR",3,"","",0,"2026-05-29","2026-10-06"],[1,5878,"CI is broken: ValueError: Either a revision or a version must be specified","albertvillanova",1,"albertvillanova","#5880",1,"2026-05-29","2026-05-29"],[1,5874,"CI fails with minimum versions: AssertionError: gradient mismatch on model.embed_tokens.weight","albertvillanova",3,"","",0,"2026-05-28","2026-06-04"],[1,5865,"AsyncGRPO: make NCCL weight sync robust on PCIe-only GPUs (auto fallback from P2P/SHM)","rycerzes",3,"","",0,"2026-05-27","2026-08-20"],[1,5863,"[AsyncGRPO] Loading from checkpoint recomputes all vLLM past requests","ggcr",0,"AmmarHassona","#5911",0,"2026-05-27","2026-08-20"],[1,5847,"[AsyncGRPO] aiohttp limits to 100 reqs when max_inflight_tasks > 100","ggcr",1,"ggcr","#5861",0,"2026-05-26","2026-06-15"],[1,5831,"Add missing AsyncGRPOTrainer features present in GRPOTrainer","rycerzes",2,"","",0,"2026-05-25","2026-07-22"],[1,5817,"CI slow tests fail: RuntimeError: min(): Expected reduction dim to be specified for input.numel() == 0","albertvillanova",1,"albertvillanova","#5818",1,"2026-05-22","2026-05-22"],[1,5814,"Vllm_importance_sampling_correction sequence-level mode aggregates per-token log-ratios with sum instead of me","ionut-anghelina",2,"","",0,"2026-05-22","2026-05-22"],[1,5807,"`SFTTrainer` does not pass `token_type_ids` for Gemma 3 text-only training","MNIKIEMA",2,"","",0,"2026-05-21","2026-05-21"],[1,5800,"Mean Token Accuracy not found at evaluation  while finetuning an model with SFTTrainer","rahulkate173",0,"qgallouedec","#5811",1,"2026-05-20","2026-05-22"],[1,5785,"GRPOTrainer: padding-direction mismatch silently corrupts multimodal context in VLM training","marcmk6",2,"","",0,"2026-05-19","2026-05-21"],[1,5783,"GRPOTrainer silently uses near-greedy decoding when temperature=1.0 (transformers >= 4.50 + Qwen2.5)","timwei0801",2,"","",0,"2026-05-18","2026-07-09"],[1,5770,"Optimize `old_per_token_logps` recomputation in GRPOTrainer: per-rollout window check instead of static modulo","wengeezhang",3,"","",0,"2026-05-14","2026-07-27"],[1,5768,"CI fails: AssertionError: Param model.visual.blocks.0.norm1.weight is not updated","albertvillanova",0,"matdou","#5792",0,"2026-05-14","2026-05-20"],[1,5761,"Weighted SFT","RyanJFriedrich",3,"","",0,"2026-05-13","2026-07-23"],[1,5754,"CI fails with transformers v5.0.0: AttributeError: 'GptOssConfig' object has no attribute 'num_experts'","albertvillanova",1,"albertvillanova","#5756",1,"2026-05-12","2026-05-15"],[1,5753,"CI fails to parse response tool calls for gptoss/llama with transformers v5.0.0","albertvillanova",1,"albertvillanova","#5755",1,"2026-05-12","2026-05-12"],[1,5750,"CI again often fails with torch.OutOfMemoryError: CUDA out of memory","albertvillanova",2,"","",0,"2026-05-12","2026-05-21"],[1,5742,"[AsyncGRPO] `AsyncGRPO` `add_response_schema` not gated on `self.tools` (diverges from `GRPOTrainer`)","aazizyan",4,"","",0,"2026-05-09",""],[1,5741,"Can gold_trainer support privileged context?","zouYC2021",4,"","",0,"2026-05-09",""],[1,5727,"OpenReward binding only discovers shared `/tools`, misses task-specific `/task_tools`","rycerzes",1,"rycerzes","#5729",0,"2026-05-07","2026-05-25"],[1,5724,"Expose `response_schema=` argument on `add_response_schema` and `GRPOTrainer`/`GRPOConfig`","rycerzes",3,"","",0,"2026-05-07","2026-07-31"],[1,5713,"🛣️ Path to 30B MoE long-context SFT training","AmineDiro",2,"","",0,"2026-05-06","2026-09-09"],[1,5697,"Bug: GRPO quickstart `max_completion_length=256` default silently breaks training","dhruvnigam93",2,"","",0,"2026-05-04","2026-08-18"],[1,5695,"[experimental] OpenReward Standard environment adapter","adithya-s-k",2,"","",0,"2026-05-03","2026-05-06"],[1,5692,"Add support for models using final_logits_softcapping to AsyncGRPOTrainer","mlarnouhet",1,"mlarnouhet","#5691",0,"2026-05-02","2026-05-19"],[1,5663,"`GRPOTrainer` `environment_factory` / `tools` is broken for VLMs whose tools return images","evancasey",3,"","",0,"2026-04-27","2026-08-06"],[1,5662,"TPOTrainer.evaluate() returns NaN eval_loss while training loss is finite","devemree",0,"qgallouedec","#7807",1,"2026-04-27","2026-10-26"],[1,5646,"DPOTrainer does not work for multimodal Gemma 4","dgawlik",2,"","",0,"2026-04-26","2026-07-18"],[1,5645,"Does PPOTrainer use mini_batch_size to update parameters?","zhiang28",2,"","",0,"2026-04-26","2026-04-29"],[1,5636,"CI is failing for pages build and deployment: Liquid syntax error:  'if' tag was never closed in source/chat_t","albertvillanova",2,"","",0,"2026-04-24","2026-04-25"],[1,5633,"CI fails for code quality check: No such file or directory: 'ruff'","albertvillanova",2,"","",0,"2026-04-24","2026-04-27"],[1,5621,"CI fails with dev dependencies for gpt-oss models: RuntimeError: You set `ignore_mismatched_sizes` to `False`","albertvillanova",1,"albertvillanova","#5622",1,"2026-04-22","2026-04-22"],[1,5619,"DAPO loss is not normalized by gradient_accumulation_steps","haotang1995",0,"0xadvait","#6024",0,"2026-04-22","2026-07-25"],[1,5613,"Feature Request: Allow passing dataset-provided sample weights to DPOTrainer","sirano1004",3,"","",0,"2026-04-21","2026-07-03"],[1,5611,"[docs] GRPO model name mismatch across docs + missing GRPO tab in OOM troubleshooting","dhruvnigam93",0,"xodn348","#5740",0,"2026-04-21","2026-05-11"],[1,5608,"Training collapses when token embeddings are not padded to a multiple of 64","Xxxxsir",2,"","",0,"2026-04-20","2026-05-07"],[1,5604,"Support Speculative decoding in vllm-serve","Ofir408",1,"Ofir408","#5605",0,"2026-04-20","2026-04-22"],[1,5590,"Add Target Policy Optimization trainer","JeanKaddour",3,"","",0,"2026-04-18","2026-04-18"],[1,5588,"[Bug] Zero-std reward groups produce spurious KL gradients when `beta > 0` in GRPO/RLOO","SwayamInSync",3,"","",0,"2026-04-18","2026-05-22"],[1,5585,"[Bug] DistillationTrainer fails with Qwen3.5 due to nested config.vocab_size attribute","ZephronCoder",0,"Beichen-Ma","#5592",0,"2026-04-17","2026-04-20"],[1,5584,"Add entropy regularization (e.g., `initial_entropy_coef`) to GRPOTrainer for improved exploration","lailanelkoussy",3,"","",0,"2026-04-17","2026-06-12"],[1,5539,"fix: DataCollatorForRewardModelingDataset checks margin only on examples[0], causing silent data loss after sh","kuishou68",3,"","",0,"2026-04-14","2026-06-07"],[1,5528,"CI fails with min versions: Validation error for field 'import_name'","albertvillanova",0,"qgallouedec","#5513",1,"2026-04-13","2026-04-13"],[1,5514,"OnlineDPOTrainer._generate_vllm_server() flattens vllm-serve completion_ids twice","JohnGiorgi",0,"qgallouedec","#6228",1,"2026-04-10","2026-07-02"],[1,5511,"Expose trainer dataset type metadata","JohnGiorgi",3,"","",0,"2026-04-10","2026-04-22"],[1,5502,"Warning log: Kwargs passed to `processor.__call__` have to be in `processor_kwargs` dict, not in `**kwargs`","albertvillanova",1,"albertvillanova","#5503",1,"2026-04-10","2026-04-10"],[1,5498,"[AsyncGRPO] `AsyncGRPOTrainer` ignores provided `processing_class` when initializing `AsyncRolloutWorker`","xuanduy04",1,"xuanduy04","#5538",0,"2026-04-10","2026-04-21"],[1,5497,"CI fails with dev dependencies for Llava models: AssertionError: Param is not updated","albertvillanova",1,"albertvillanova","#5499",1,"2026-04-10","2026-04-20"],[1,5495,"Ascend NPU Error with CANN 8.5.0 and TRL 0.25.1","shibizhao",3,"","",0,"2026-04-10","2026-06-30"],[1,5471,"Tracking: Add `{% generation %}` chat templates for common model families","qgallouedec",2,"","",0,"2026-04-07","2026-07-16"],[1,5460,"Tracking: tool calling support across chat templates","qgallouedec",2,"","",0,"2026-04-06","2026-07-16"],[1,5455,"GRPO: add opt-in async rollout dispatch for vLLM server mode","PoilZero",3,"","",0,"2026-04-04","2026-04-04"],[1,5449,"ImportError: cannot import name 'TRLExperimentalWarning' from 'trl.import_utils'","Hello-Rk",3,"","",0,"2026-04-03","2026-04-10"],[1,5444,"AsyncGRPO: support async tool calls in AsyncRolloutWorker","PoilZero",2,"","",0,"2026-04-03","2026-09-19"],[1,5429,"CI fails with dev dependencies: TypeError: unsupported operand type(s) for -: 'list' and 'set'","albertvillanova",2,"","",0,"2026-04-01","2026-04-01"],[1,5428,"CI fails with dev dependencies: KeyError: 'qwen2_vl'","albertvillanova",2,"","",0,"2026-04-01","2026-04-02"],[1,5425,"CI fails with dev dependencies: TypeError: 'NoneType' object is not iterable","albertvillanova",2,"","",0,"2026-04-01","2026-04-02"],[1,5421,"CI fails for experimental tests: ValueError: You must specify exactly one of input_ids or inputs_embeds","albertvillanova",0,"SunMarc","#5426",1,"2026-04-01","2026-04-01"],[1,5419,"Add AEO quality badge to README (score: 50/100)","digitamaz",2,"","",0,"2026-04-01","2026-04-22"],[1,5415,"`qwen3_schema` regex in `parse_response` causes catastrophic backtracking — training hangs indefinitely","lailanelkoussy",0,"xodn348","#5798",0,"2026-03-31","2026-05-20"],[1,5412,"Skip unnecessary no_grad forward pass for IS correction in on-policy vLLM training","GJ98",4,"","",0,"2026-03-31",""],[1,5389,"# [Feature Request] GRPOTrainer: support stop-tool termination in agent training loop","lailanelkoussy",4,"","",0,"2026-03-27",""],[1,5386,"CI fails for distributed test_rloo[fsdp2]: CUDA error: device-side assert triggered: probability tensor contai","albertvillanova",1,"albertvillanova","#5387",1,"2026-03-27","2026-03-27"],[1,5382,"Possible NCCL ordering issue in TRL colocate mode with trainer TP + vLLM TP during tool-calling rollout","YihaoJW",4,"","",0,"2026-03-26",""],[1,5379,"GRPO Trainer: Empty tools not handled properly while tokenizing the prompts.","rabinadk1",1,"rabinadk1","#5380",0,"2026-03-26","2026-03-27"],[1,5375,"DAPO token level loss is incorrectly implemented","jiosephlee",3,"","",0,"2026-03-25","2026-08-21"],[1,5373,"Video inputs for multimodal LLM","markub3327",3,"","",0,"2026-03-25","2026-10-01"],[1,5366,"GRPO with tool use","catalin-hanga",2,"","",0,"2026-03-24","2026-06-09"],[1,5361,"Multimodal generation hangs in vllm server mode with large images (works in colocate), manual resizing require","SmartFive",3,"","",0,"2026-03-24","2026-09-22"],[1,5352,"GRPOTrainer: logged reward/reward_std metrics ignore reward_weights","lailanelkoussy",1,"lailanelkoussy","#5353",0,"2026-03-23","2026-03-24"],[1,5348,"Performace vs SamplingParams","Samarth26",2,"","",0,"2026-03-23","2026-03-23"],[1,5334,"Qwen3.5 input embeddings have too many values to unpack with SFTTrainer","JakobJBauer",0,"albertvillanova","#5354",1,"2026-03-21","2026-03-24"],[1,5329,"Outdated documentation after Transformers v5 dtype default behavior changed","brozjak2",0,"casinca","#5457",0,"2026-03-20","2026-06-30"],[1,5324,"completion_only_loss","Alvar1000",0,"RudrenduPaul","#5494",0,"2026-03-20","2026-04-10"],[1,5317,"Missing `\\n` token between completion and tool suffix in `GRPOTrainer._get_tool_suffix_ids` for Qwen3 and Qwen","mattbui",0,"casinca","#5330",0,"2026-03-20","2026-03-21"],[1,5314,"AttributeError: 'VLLMGeneration' object has no attribute 'vllm_client' when using generate_rollout_completions","SmartFive",3,"","",0,"2026-03-19","2026-07-13"],[1,5312,"vLLM weights are not synchronized with vllm_enable_sleep_mode=True","muupan",1,"muupan","#5313",0,"2026-03-19","2026-07-28"],[1,5304,"DPO sequences longer than max_length cause over-allocation and potential OOM during collation","albertvillanova",1,"albertvillanova","#5305",1,"2026-03-18","2026-03-18"],[1,5301,"GRPO + vLLM colocate mode crashes due to missing attribute","falcondai",1,"falcondai","#5302",0,"2026-03-18","2026-03-19"],[1,5298,"GSPO example is not working","mshaocong",2,"","",0,"2026-03-17","2026-03-19"],[1,5285,"DPOTrainer silently corrupts VLM training with \"keep_end\" truncation_mode","albertvillanova",1,"albertvillanova","#5286",1,"2026-03-13","2026-03-17"],[1,5283,"DPOTrainer crashes when max_length is set with VLMs: IndexError","albertvillanova",1,"albertvillanova","#5284",1,"2026-03-13","2026-03-17"],[1,5282,"[Question] Do use_liger_gkd_loss breaks DDP gradient sync by unwrapping student model？","JeffLee1874",0,"albertvillanova","#5934",1,"2026-03-13","2026-06-05"],[1,5277,"DPOTrainer silently ignores mm_token_type_ids when training with Qwen2.5-VL and similar 3D-RoPE models","albertvillanova",1,"albertvillanova","#5279",1,"2026-03-12","2026-03-13"],[1,5271,"ms-swift GRPO Qwen3.5-9B use cuda， but，No module named 'vllm_ascend'","sumufuyun",2,"","",0,"2026-03-11","2026-03-11"],[1,5269,"GRPO + vLLM weight sync fails for Qwen3.5 text-only checkpoints (Qwen3_5ForCausalLM)","deter3",2,"","",0,"2026-03-10","2026-08-16"],[1,5265,"unexpected keyword argument 'ref_model_init_kwargs' in DPOConfig","iPRET",2,"","",0,"2026-03-10","2026-03-10"],[1,5241,"[GOLD][vLLM] On-policy generation decodes prompts with skip_special_tokens=True (chat template stripped) — int","siyan-zhao",0,"he-yufeng","#6063",0,"2026-03-07","2026-06-15"],[1,5231,"Logging how vllm importance ratios are truncated/masked in GRPOTrainer","muupan",4,"","",0,"2026-03-06",""],[1,5226,"Integrate GTPO (Group-relative Trajectory-based Policy Optimization) as a new loss_type in GRPOTrainer","winstonsmith1897",3,"","",0,"2026-03-05","2026-08-14"],[1,5224,"Re-tokenization bug in GRPO multi-turn tool calling","qgallouedec",1,"qgallouedec","#5242",1,"2026-03-05","2026-03-19"],[1,5222,"DPOTrainer ref adapter crashes with PEFT `target_parameters` (MoE models on Transformers 5.x)","pritmish",0,"discobot","#6043",0,"2026-03-05","2026-06-18"],[1,5217,"Recomputed tensor size does not match when using quantization + LoRA with the DPO Trainer with gradient checkp","slhmath",0,"qgallouedec","#6356",1,"2026-03-04","2026-07-15"],[1,5216,"ValueError: The following model_kwargs are not used by the model: ['mm_token_type_ids']","albertvillanova",2,"","",0,"2026-03-04","2026-05-02"],[1,5213,"Qwen3.5-9B: Mismatch between tokenized prompt and the start of tokenized prompt+completion","zouYC2021",2,"","",0,"2026-03-03","2026-04-29"],[1,5211,"Adaptive Beta-DPO","giuseppecurci",3,"","",0,"2026-03-03","2026-07-31"],[1,5207,"CI often fails with torch.OutOfMemoryError: CUDA out of memory","albertvillanova",2,"","",0,"2026-03-03","2026-04-30"],[1,5201,"CI fails with dev dependencies: ValueError: The following `model_kwargs` are not used by the model: ['mm_token","albertvillanova",1,"albertvillanova","#5202",1,"2026-03-02","2026-03-02"],[1,5200,"dpo training with Qwen2.5-0.5B+1000 seq length cost 17GB GPU mem","32234",3,"","",0,"2026-03-02","2026-06-08"],[1,5196,"feat: Variational Sequence-Level Soft Policy Optimization (VESPO)","casinca",1,"casinca","#5199",0,"2026-02-27","2026-03-14"],[1,5195,"Unify Weight Sync for Rollout Path","rycerzes",3,"","",0,"2026-02-27","2026-07-06"],[1,5194,"Make openenv/utils.py Backend-Agnostic","rycerzes",2,"","",0,"2026-02-27","2026-07-16"],[1,5193,"Introduce a Minimal Backend Generation Interface","rycerzes",3,"","",0,"2026-02-27","2026-10-06"],[1,5191,"Align trl skills CLI DX with HF CLI conventions","albertvillanova",3,"","",0,"2026-02-27","2026-09-07"],[1,5177,"CI fails with dev dependencies: IndexError: The shape of the mask at index 0 does not match the shape of the i","albertvillanova",1,"albertvillanova","#5178",1,"2026-02-25","2026-02-26"],[1,5175,"CI fails with dev dependencies: TypeError: GRPOTrainer._get_per_token_logps_and_entropies() got an unexpected ","albertvillanova",1,"albertvillanova","#5176",1,"2026-02-25","2026-02-25"],[1,5167,"Does GOLD Trainer support multi-node training similar to GRPO?","NoviceStone",2,"","",0,"2026-02-25","2026-07-16"],[1,5154,"Support for constrained decoding with JSON Schema in GRPO Trainer + vLLM","ehofm",2,"","",0,"2026-02-23","2026-06-20"],[1,5152,"Computation of merged probabilities for GOLD","vman049",1,"vman049","#5905",0,"2026-02-23","2026-06-30"],[1,5150,"CI fails for experimental tests: RuntimeError: Some tensors share memory","albertvillanova",1,"albertvillanova","#5151",1,"2026-02-23","2026-02-23"],[1,5144,"GRPOTrainer tool_mask can become longer than completion_ids after tool-call retokenization","MichalMraz",0,"qgallouedec","#5242",1,"2026-02-21","2026-03-19"],[1,5142,"[Bug] colocate sleep mode: sync_weights() writes to freed GPU memory before wake_up() on non-CUDA backends","kevin-Abbring",0,"bledden","#5147",0,"2026-02-21","2026-02-25"],[1,5139,"GRPO Trainer Support mixing samples with and without images?","yongxinw",2,"","",0,"2026-02-21","2026-02-21"],[1,5138,"Improper buggy SFT training of Llama-3.2-3B-Instruct.","3rdAT",3,"","",0,"2026-02-20","2026-10-12"],[1,5130,"Improve TRL import latency","albertvillanova",2,"","",0,"2026-02-19","2026-02-20"],[1,5127,"MGPO","damoonsh",3,"","",0,"2026-02-19","2026-09-07"],[1,5121,"Decouple rollout from vLLM [1/N]: Move rollout_func dispatch to top-level _generate_single_turn","albertvillanova",1,"albertvillanova","#5122",1,"2026-02-18","2026-03-03"],[1,5120,"Preserve structured multimodal messages through rollout and generation pipeline","albertvillanova",2,"","",0,"2026-02-18","2026-06-29"],[1,5119,"Decouple inference backend from rollout & agent logic","albertvillanova",2,"","",0,"2026-02-18","2026-07-31"],[1,5113,"GRPOTrainer fails on Windows 11 with OSError: [WinError 6] (invalid handle) during multiprocessing spawn","Bader-CN",3,"","",0,"2026-02-17","2026-02-17"],[1,5099,"CLI help fails: TypeError: not enough arguments for format string","albertvillanova",1,"albertvillanova","#5135",1,"2026-02-13","2026-02-20"],[1,5098,"Documentation unclear for num_generations","alexchen5",2,"","",0,"2026-02-13","2026-02-14"],[1,5095,"Implement Agent Skills for TRL","albertvillanova",2,"","",0,"2026-02-13","2026-02-19"],[1,5094,"Training crash with \"uncorrectable ECC error\" and NCCL timeout?","ShuimuChen-hyq",3,"","",0,"2026-02-13","2026-07-21"],[1,5090,"RFC: Defining the TRL Public API (towards v1.0)","qgallouedec",1,"qgallouedec","#5162",1,"2026-02-12","2026-02-26"],[1,5088,"CI fails with dev dependencies: RuntimeError: The size of tensor a (6) must match the size of tensor b (3) at ","albertvillanova",0,"zucchini-nlp","#43952",1,"2026-02-12","2026-02-12"],[1,5075,"The changes made to \"bfd\" packing in v0.27.0 breaks packing for SFT datasets","ikergarcia1996",0,"albertvillanova","#5076",1,"2026-02-11","2026-02-12"],[1,5041,"GRPOTrainer still has unresolved bugs in VLM / VL setting – any plan for fixing?","wswaq",0,"qgallouedec","#5067",1,"2026-02-10","2026-02-17"],[1,5033,"Source to GRPO online example","ttrung149",2,"","",0,"2026-02-09","2026-07-12"],[1,5032,"Fine-Tuning gemma-3-4b-pt raises \"ValueError: `token_type_ids` is required as a model input when training\"","jesperschlegel",2,"","",0,"2026-02-09","2026-05-04"],[1,5030,"GRPO rewards are not grouped as expected","erikchwang",2,"","",0,"2026-02-09","2026-02-09"],[1,5025,"Maximum Likelihood Reinforcement Learning","catherinelee274",3,"","",0,"2026-02-09","2026-09-07"],[1,5022,"CI PyTorch argument 'device' deprecation warnings","albertvillanova",0,"qgallouedec","#4999",1,"2026-02-09","2026-02-09"],[1,5021,"CI PyTorch JIT deprecation warnings","albertvillanova",0,"qgallouedec","#4999",1,"2026-02-09","2026-02-09"],[1,5018,"CI FutureWarning: `max_prompt_length` is deprecated","albertvillanova",1,"albertvillanova","#5019",1,"2026-02-09","2026-02-09"],[1,5016,"CI FutureWarning: `generate_during_eval` is deprecated","albertvillanova",1,"albertvillanova","#5017",1,"2026-02-09","2026-02-09"],[1,5014,"CI FutureWarning: `tools` is deprecated","albertvillanova",1,"albertvillanova","#5015",1,"2026-02-09","2026-02-09"],[1,5012,"CI FutureWarning: `use_logits_to_keep` is deprecated","albertvillanova",1,"albertvillanova","#5013",1,"2026-02-09","2026-02-09"],[1,5010,"CI FutureWarning: `rpo_alpha` is deprecated","albertvillanova",1,"albertvillanova","#5011",1,"2026-02-09","2026-02-09"],[1,5008,"CI FutureWarning: `ref_model_init_kwargs` is deprecated","albertvillanova",1,"albertvillanova","#5009",1,"2026-02-09","2026-02-09"],[1,5006,"CI TRLExperimentalWarning in regular tests","albertvillanova",1,"albertvillanova","#5007",1,"2026-02-09","2026-02-09"],[1,5005,"CI DeprecationWarning: invalid escape sequence '\\s'","albertvillanova",2,"","",0,"2026-02-09","2026-02-09"],[1,5003,"CI SWIG deprecation warnings","albertvillanova",1,"albertvillanova","#5004",1,"2026-02-09","2026-02-09"],[1,4998,"DPPO - Divergence Proximal Policy Optimization","ErykCh",2,"","",0,"2026-02-06","2026-06-07"],[1,4974,"Support multi-turn training in GRPOTrainer","albertvillanova",2,"","",0,"2026-02-05","2026-07-14"],[1,4973,"[GRPO] AssertionError in vLLM when using QLoRA (4-bit) due to weight shape mismatch during merge_adapter","ChengHu98",4,"","",0,"2026-02-05",""],[1,4968,"Simple Quick Start examples are great – but on what hardware? A note in the doc would be lovely.","jchwenger",2,"","",0,"2026-02-04","2026-09-25"],[1,4967,"Support Unbiased \"K1-in-Reward\" KL Estimator for GRPO","ktolnos",2,"","",0,"2026-02-04","2026-08-17"],[1,4961,"CUDA OOM: Qwen3 MoE Single-GPU VRAM usage differs in DDP Settings","jiosephlee",2,"","",0,"2026-02-03","2026-02-03"],[1,4954,"[PPO] 'PolicyAndValueWrapper' has no attribute 'gradient_checkpointing_disable' in unwrap_model_for_generation","lvhungdev",2,"","",0,"2026-02-03","2026-03-10"],[1,4952,"AttributeError: 'coroutine' object has no attribute 'observation' -  FunctionGemma BrowserGym","erfanshayegani",3,"","",0,"2026-02-03","2026-07-15"],[1,4948,"GRPO Trainer: Remove limitation of sequence importance sampling in GRPO trainer with use_ligerkernel=True","vvvdwbvvv",2,"","",0,"2026-02-02","2026-03-02"],[1,4947,"GRPO Trainer: Remove limitation of sequence importance sampling in GRPO trainer with use_ligerkernel=True","vvvdwbvvv",2,"","",0,"2026-02-02","2026-02-02"],[1,4944,"PPO the run_name parameter does not take effect","mel3c",1,"mel3c","#4945",0,"2026-02-02","2026-02-03"],[1,4943,"CI fails with dev deps: NameError: name 'get_auto_gptq_quant_linear' is not defined","albertvillanova",2,"","",0,"2026-02-02","2026-02-03"],[1,4940,"SDFT: Self-Distillation Fine-Tuning Trainer","Shekswess",2,"","",0,"2026-01-31","2026-07-12"],[1,4939,"GRPO Reward Function documentation and usage mismatch","amit9oct",2,"","",0,"2026-01-31","2026-03-10"],[1,4930,"AttributeError: 'VLLMGeneration' object has no attribute 'state'","albertvillanova",1,"albertvillanova","#4931",1,"2026-01-29","2026-01-29"],[1,4929,"SDPO: Reinforcement Learning via Self-Distillation","Shekswess",0,"MengAiDev","#4935",0,"2026-01-29","2026-03-23"],[1,4922,"Multi-Turn Phi35 mini instruct training","chwunderpers",2,"","",0,"2026-01-28","2026-07-01"],[1,4920,"CI fails for experimental tests: AssertionError: assert not True","albertvillanova",1,"albertvillanova","#4921",1,"2026-01-28","2026-01-28"],[1,4918,"Judges dependency llm-blender is not compatible with transformers v5","albertvillanova",2,"","",0,"2026-01-28","2026-05-12"],[1,4917,"CI fails for experimental tests: TypeError: argument should be a str or an os.PathLike object where __fspath__","albertvillanova",1,"albertvillanova","#4919",1,"2026-01-28","2026-01-28"],[1,4915,"CI fails for experimental tests: ValueError: `temperature` (=0.0) has to be a strictly positive float","albertvillanova",1,"albertvillanova","#4916",1,"2026-01-28","2026-01-28"],[1,4903,"CI fails for slow tests: AssertionError: Parameter model.layers.0.self_attn.q_proj.weight has not changed","albertvillanova",1,"albertvillanova","#4904",1,"2026-01-27","2026-01-27"],[1,4901,"CI fails for slow tests: NotImplementedError: \"_amp_foreach_non_finite_check_and_unscale_cuda\" not implemented","albertvillanova",1,"albertvillanova","#4902",1,"2026-01-27","2026-01-27"],[1,4899,"RLOO and GRPO failing with ZeRO3: IndexError: pop from an empty deque","qgallouedec",0,"albertvillanova","#5541",1,"2026-01-26","2026-04-14"],[1,4897,"vLLM Performance issues (>= 0.11.x vs 0.10.2)","sahandrez",2,"","",0,"2026-01-26","2026-07-10"],[1,4886,"May I ask if there is an official evaluation code for the GOLD algorithm?","141forever",2,"","",0,"2026-01-22","2026-06-20"],[1,4884,"SFT with PEFT and ZeRO-2 fails","qgallouedec",2,"","",0,"2026-01-21","2026-02-10"],[1,4881,"TypeError when model generates tool calls with missing arguments in GRPOTrainer","akshayballal95",1,"akshayballal95","#4890",0,"2026-01-21","2026-02-07"],[1,4879,"Automatic adding of {% generation %} tags to chat_template for training with assistant_only_loss=True","Neelectric",2,"","",0,"2026-01-20","2026-06-23"],[1,4874,"vml sft dataset format","chits-galosh",2,"","",0,"2026-01-20","2026-01-22"],[1,4871,"NotImplementedError: Tool calling is not yet implemented in VLLMClient.chat().","Data-Designer",0,"kansalaman","#4889",0,"2026-01-20","2026-03-03"],[1,4870,"examples/notebooks/grpo_qwen3_vl.ipynb fails with dataset","dajuanmaoa",0,"qgallouedec","#5067",1,"2026-01-20","2026-02-17"],[1,4869,"RuntimeError on calling the `self.generate_on_policy_outputs` function during `GKDTrainer` training loop","NoviceStone",2,"","",0,"2026-01-20","2026-01-20"],[1,4867,"Training Fail when set \"cast_lm_head_to_fp32\" and FSDP2","ZiyiTsang",0,"qgallouedec","#6330",1,"2026-01-20","2026-07-14"],[1,4866,"Expose generation index to tool callables during GRPOTrainer tool execution","lukehinds",3,"","",0,"2026-01-19","2026-06-19"],[1,4865,"qwen3_schema regex causes O(2^n) backtracking hang on malformed tool calls","austinmw",2,"","",0,"2026-01-19","2026-05-28"],[1,4864,"If there are any training acceleration techniques?","141forever",1,"141forever","#4888",0,"2026-01-19","2026-01-26"],[1,4859,"Support explicit JSON tool schemas in GRPOTrainer (in addition to callable tools)","ansh-info",3,"","",0,"2026-01-18","2026-06-06"],[1,4854,"KeyError when using FSDP2 with RLOO","qgallouedec",0,"flutist","#4838",0,"2026-01-17","2026-02-16"],[1,4853,"Implement hierarchical reward conditioning based on GDPO paper","jlcanta",3,"","",0,"2026-01-17","2026-06-29"],[1,4850,"GRPO Training Fail when \"use_liger_kernel\" + \"eval_on_start true\"","ZiyiTsang",0,"qgallouedec","#4852",1,"2026-01-17","2026-01-17"],[1,4842,"CI fails with dev dependencies: AttributeError: 'Qwen3VLTextConfig' object has no attribute 'pad_token_id'","albertvillanova",1,"albertvillanova","#4851",1,"2026-01-16","2026-01-17"],[1,4841,"Addition of GDPO (Group reward-Decoupled normalization Policy Optimization) Trainer to TRL","Cenrax",2,"","",0,"2026-01-16","2026-01-16"],[1,4829,"CI is broken for Python 3.10-3.13: The self-hosted runner lost communication with the server","albertvillanova",1,"albertvillanova","#4833",1,"2026-01-15","2026-01-16"],[1,4825,"formatting_reward_func() got multiple values for keyword argument 'completions'","zkailinzhang",3,"","",0,"2026-01-14","2026-01-14"],[1,4816,"CI fails with experimental: ValueError: Pointer argument (at 0) cannot be accessed from Triton (cpu tensor?)","albertvillanova",1,"albertvillanova","#5074",1,"2026-01-13","2026-02-11"],[1,4812,"DPO and FSDP2 fails","qgallouedec",0,"flutist","#4838",0,"2026-01-12","2026-02-16"],[1,4806,"CI fails with test dependencies: RuntimeError: You set `ignore_mismatched_sizes` to `False`, thus raising an e","albertvillanova",0,"qgallouedec","#4799",1,"2026-01-12","2026-01-12"],[1,4805,"CI fails with test dependencies: ImportError: cannot import name 'HybridCache' from 'transformers.cache_utils'","albertvillanova",0,"qgallouedec","#4798",1,"2026-01-12","2026-01-12"],[1,4795,"[Feature Request] Entropy-Adaptive Fine-Tuning","electroglyph",2,"","",0,"2026-01-09","2026-08-10"],[1,4790,"The new version TRL (0.26.2) is compatible with vllm-ascend (0.11.0rc1), and 910B supports using vllm serve to","pbbb205",3,"","",0,"2026-01-08","2026-07-31"],[1,4786,"KTO refactoring to be promoted to stable","albertvillanova",1,"albertvillanova","#6175",1,"2026-01-08","2026-07-05"],[1,4782,"`RuntimeError: Expected to mark a variable ready only once` when training with PEFT","qgallouedec",1,"qgallouedec","#4811",1,"2026-01-07","2026-01-14"],[1,4779,"Response Schema only get added if tools are defined in GRPOTrainer","ashim-mahara",2,"","",0,"2026-01-07","2026-01-07"],[1,4776,"SFTTrainer stuck","weathon",3,"","",0,"2026-01-06","2026-07-13"],[1,4772,"vllm_importance_sampling_correction causes vllm_importance_sampling_ratio explosion","Aurora-1412",2,"","",0,"2026-01-03","2026-05-26"],[1,4771,"Qwen3-VL-30B-A3B-Instruct + fsdp: RuntimeError: Expected all tensors to be on the same device, but got index i","li126com",3,"","",0,"2026-01-03","2026-07-04"],[1,4767,"Can't find completions in wandb and using IS correction causes loss=0.0","Aurora-1412",3,"","",0,"2026-01-01","2026-06-03"],[1,4766,"Asynchronous generation and training for GRPO?","sxndqc",2,"","",0,"2026-01-01","2026-07-18"],[1,4764,"Can support optional dynamic sampling in GRPO ? (as in DAPO)","ceresOPA",3,"","",0,"2025-12-30","2026-05-24"],[1,4760,"example code  Error","ciaoyizhen",2,"","",0,"2025-12-29","2025-12-29"],[1,4759,"[Feature Discussion] Supervised Reinforcement Learning (SRL) - Step-Level Training for Reasoning","s23deepak",3,"","",0,"2025-12-29","2026-07-31"],[1,4751,"[Feature Request] Add max_tool_iterations parameter to GRPOTrainer","austinmw",0,"pramodith","#4761",1,"2025-12-25","2026-01-12"],[1,4748,"CI fails with test dependencies: AssertionError: Parameter has not changed","albertvillanova",1,"albertvillanova","#4770",1,"2025-12-24","2026-01-06"],[1,4747,"Addition of Supervised Reinforcement Learning","kushalgarg101",3,"","",0,"2025-12-24","2026-06-08"],[1,4746,"examples/notebooks/grpo_qwen3_vl.ipynb fails with TypeError: string indices must be integers, not 'str'","ympan0508",0,"qgallouedec","#5067",1,"2025-12-24","2026-02-17"],[1,4741,"CI fails with test dependencies: torch.AcceleratorError: CUDA error: device-side assert","albertvillanova",1,"albertvillanova","#4745",1,"2025-12-23","2025-12-24"],[1,4739,"CI fails with test dependencies: ValueError: Default process group has not been initialized","albertvillanova",2,"","",0,"2025-12-23","2025-12-23"],[1,4738,"completion_only_loss=True give wrong training steps","HuangChiEn",2,"","",0,"2025-12-23","2025-12-25"],[1,4737,"NCCL error: /pytorch/torch/csrc/distributed/c10d/NCCLUtils.hpp:268, NCCL version 2.22.3 ncclSystemError: Syste","guotong1988",3,"","",0,"2025-12-23","2026-06-16"],[1,4735,"feat GRPOTrainer: Bidirectional MIS from IcePop loss (used in Xiaomi MiMo-V2-Flash and Ring-1T)","casinca",1,"casinca","#4732",0,"2025-12-22","2026-06-11"],[1,4730,"change in behavior of mean_token_accuracy from trl==0.15 to trl==0.26 when using liger-kernel","mvish7",2,"","",0,"2025-12-20","2026-05-02"],[1,4725,"gkd cannot run on the dataset OpenThoughts3-1.2M.","Jim2016713",0,"qgallouedec","#6310",1,"2025-12-19","2026-07-06"],[1,4722,"GRPO agent notebook install failure","austinmw",0,"sergiopaniego","#4724",1,"2025-12-19","2025-12-19"],[1,4719,"Loss calculation of `GKDTrainer` may be inaccurate when performing gradient accumulation?","jue-jue-zi",0,"behroozazarkhalili","#6006",1,"2025-12-18","2026-06-13"],[1,4711,"Add RLVR (Reinforcement Learning with Verifiable Rewards) Trainer","mmjerge",3,"","",0,"2025-12-17","2025-12-29"],[1,4708,"qwen3_schema fails to parse multiple parallel tool calls","austinmw",0,"mattbui","#4709",0,"2025-12-17","2025-12-22"],[1,4707,"GRPOTrainer: Relative LLM-as-a-judge rankings","austinmw",2,"","",0,"2025-12-16","2026-08-19"],[1,4699,"Do not add GOLDConfig/GOLDTrainer into __init__.py","141forever",3,"","",0,"2025-12-16","2025-12-16"],[1,4697,"feat GRPOTrainer: Off-policy sequence masking from DeepSeek V3.2","casinca",1,"casinca","#4689",0,"2025-12-15","2026-01-06"],[1,4692,"LLVM error during GRPO training with Apple M4 Max","neslihaneti",3,"","",0,"2025-12-14","2026-05-24"],[1,4679,"[SFT] High vRAM consumption during eval loop","Khreas",0,"LoganVegnaSHOP","#5069",0,"2025-12-11","2026-02-11"],[1,4669,"[GRPO] evaluation at start fails, due to missing current_gradient_accumulation_step","SonuDixit",0,"qgallouedec","#4852",1,"2025-12-11","2025-12-21"],[1,4658,"CI fails with dev dependencies: AssertionError with tool call tokenization","albertvillanova",2,"","",0,"2025-12-10","2026-01-07"],[1,4655,"Fail to run exmaples/sft.py with Qwen3-8B with multi-gpu single node","BlackPoint-CX",2,"","",0,"2025-12-10","2025-12-12"],[1,4641,"Further improving `GRPOTrainer` doc to include Qwen SAPO in Loss Types","casinca",1,"casinca","#4681",0,"2025-12-08","2025-12-12"],[1,4637,"GRPOTrainer: Implement Unbiased KL Estimator from DeepSeek-V3.2","jlcanta",1,"jlcanta","#4638",0,"2025-12-07","2025-12-09"],[1,4634,"GRPOTrainer and bnb quantization configs are incomptiable","JitheshPavan",2,"","",0,"2025-12-06","2026-07-06"],[1,4631,"[Bug] DeepSpeed ZeRO-2 is unstable with GRPOTrainer using Qwen3-VL MoE","casper-hansen",2,"","",0,"2025-12-05","2026-08-01"],[1,4617,"VLLM deprecated GuidedDecodingParams -> StructuredOutputsParams","ed-pai",0,"jiqing-feng","#4117",0,"2025-12-04","2026-01-05"],[1,4615,"model_kwargs is not used in PPO Example script","JitheshPavan",0,"casinca","#4630",0,"2025-12-03","2025-12-06"],[1,4612,"GRPOTrainer: liger_kernel loss implementation ignores PEFT/LoRA adapters on lm_head","junho328",0,"akshansh47","#5977",0,"2025-12-03","2026-06-19"],[1,4611,"SFTTrainer: Logging Experts Statistics","shon-otmazgin",4,"","",0,"2025-12-02",""],[1,4609,"AttributeError: Qwen2Tokenizer has no attribute response_schema","albertvillanova",2,"","",0,"2025-12-02","2026-06-08"],[1,4602,"openai-compatible responses and chat completions endpoints in vllm_serve.py","cmunley1",2,"","",0,"2025-11-29","2026-06-20"],[1,4601,"CI fails with dev dependencies: AttributeError: 'Qwen2_5_VLForConditionalGeneration' object has no attribute '","albertvillanova",1,"albertvillanova","#4777",1,"2025-11-28","2026-01-07"],[1,4599,"CI fails with dev dependencies: ValueError: Could not load tokenizer","albertvillanova",2,"","",0,"2025-11-28","2026-01-07"],[1,4597,"CI fails with dev dependencies: ImportError: cannot import name 'TRANSFORMERS_CACHE' from 'transformers.utils.","albertvillanova",1,"albertvillanova","#4781",1,"2025-11-28","2026-01-08"],[1,4595,"CI fails with dev dependencies: AttributeError: 'GPTNeoXForSequenceClassification' object has no attribute 'ge","albertvillanova",1,"albertvillanova","#4596",1,"2025-11-28","2025-11-28"],[1,4593,"CI fails with dev dependencies: RuntimeError: output with shape [3, 4, 1, 1] doesn't match the broadcast shape","albertvillanova",2,"","",0,"2025-11-27","2025-12-04"],[1,4592,"CI fails with dev dependencies: AttributeError: 'list' object has no attribute 'items'","albertvillanova",2,"","",0,"2025-11-27","2025-12-04"],[1,4591,"Async-GRPO: Decouple rollout generation and reward computation/model update for faster training","marksverdhei",2,"","",0,"2025-11-27","2026-06-11"],[1,4589,"GRPO with vllm server not work under FSDP2","Xiao-Chenguang",1,"Xiao-Chenguang","#4603",0,"2025-11-27","2025-12-07"],[1,4588,"Qwen3-0.6-base + DeepSpeed Zero-3 + PPO: RuntimeError (tensor dimension mismatch in Rotary Position Embedding)","yitian2ly-tech",2,"","",0,"2025-11-27","2026-06-22"],[1,4585,"GRPO Trainer is going out of memory on g5.48xlarge instance","subhojitdas",3,"","",0,"2025-11-27","2026-07-16"],[1,4584,"[vLLM] Upgrade vLLM dependency to v0.11.2 to support FusedMoE with LoRA","casper-hansen",2,"","",0,"2025-11-26","2025-11-26"],[1,4583,"Generate tiny models fails for Qwen2.5-VL: KeyError: 'rope_theta'","albertvillanova",2,"","",0,"2025-11-26","2025-12-04"],[1,4582,"Does the GRPO Trainer support multi-image input for Qwen3-VL?","Lestoky",2,"","",0,"2025-11-26","2026-05-28"],[1,4578,"SFTTrainer causes PyTorch Distributed deadlock detection timeout with large datasets during preprocessing","ryan-minato",3,"","",0,"2025-11-26","2026-06-11"],[1,4570,"Default to LigerFusedLinearCrossEntropyLoss if Liger is available","jiosephlee",2,"","",0,"2025-11-25","2026-06-12"],[1,4569,"[doc issue] doc on \"GRPO with replay buffer\" buggy","DNXie",0,"pramodith","#4574",1,"2025-11-25","2025-11-25"],[1,4568,"Speedup GKDTrainer with vLLM","weixiao-zhan",2,"","",0,"2025-11-24","2025-11-25"],[1,4562,"Cross-tokenizer distillation fails in GKD and MiniLLM trainers","sambhavnoobcoder",3,"","",0,"2025-11-22","2026-05-23"],[1,4560,"GKD Degradation from trl v0.21.0 to 0.25.1","AamodThakur",2,"","",0,"2025-11-22","2026-07-27"],[1,4559,"grpo+vllm+qwen3vl stuck","wswaq",2,"","",0,"2025-11-22","2026-06-28"],[1,4557,"New GSPO trainer  patch","aamir-gmail",2,"","",0,"2025-11-22","2025-11-24"],[1,4554,"Better packing of data with best-fit decrease strategy","ntnq4",0,"qgallouedec","#4632",1,"2025-11-21","2025-12-16"],[1,4550,"FSDP2+PEFT(LoRA)+GRPO result in `mixed torch.Tensor and DTensor`","Xiao-Chenguang",2,"","",0,"2025-11-20","2025-12-04"],[1,4546,"Does TRL support PipelineRL for compute efficiency?","harisarang",3,"","",0,"2025-11-19","2026-07-19"],[1,4544,"smollm3 DPO recipe fails when using liger_kernal #227","shamanez",0,"qgallouedec","#6372",1,"2025-11-19","2026-07-14"],[1,4543,"[OpenEnv] multi step training issue in GRPO trainer","akshay-gup",2,"","",0,"2025-11-18","2026-07-02"],[1,4541,"Is attn_implementation=sdpa not supported when using SFTTrainer with mllama?","osaidr",3,"","",0,"2025-11-18","2026-06-15"],[1,4540,"Significant Performance Difference between DeepSpeed's `zero_stage=1` and `zero_stage=2`","liyc-ai",2,"","",0,"2025-11-18","2026-07-23"],[1,4538,"TCPStore Issues in GRPOTrainer+vLLM","chao-nlp",0,"pointerhacker","#4545",0,"2025-11-18","2025-12-09"],[1,4535,"GRPOTrainer fails to train when using Accelerate with DeepSpeed ZeRO-3.","1luik",2,"","",0,"2025-11-17","2026-06-01"],[1,4532,"Adaptive KL in GRPO","hakbari14",3,"","",0,"2025-11-17","2026-06-13"],[1,4531,"GRPO training: Incorrect RepeatSampling when num_iterations >1","oapandit",2,"","",0,"2025-11-17","2026-06-14"],[1,4525,"How to modify the advantage computation in GRPOTrainer","Tuziking",2,"","",0,"2025-11-14","2026-06-17"],[1,4523,"Replace torch image by vllm image in Docker","qgallouedec",1,"qgallouedec","#7464",1,"2025-11-13","2026-10-02"],[1,4511,"Usage of gradient_accumulation_steps in GRPO","Fyeward",3,"","",0,"2025-11-12","2026-05-27"],[1,4507,"Can a multimodal model like Gemma be trained in the same way as a text-only model like Qwen, but with the goal","Tuziking",3,"","",0,"2025-11-11","2026-05-31"],[1,4503,"pip install trl failing","allenx123",2,"","",0,"2025-11-08","2026-07-12"],[1,4501,"NaN metric propagation in GRPOTrainer with sparse multi-reward training","shriramc1",2,"","",0,"2025-11-07","2026-06-14"],[1,4497,"DataCollatorForLanguageModeling always uses pad_side='right'","kimihailv",2,"","",0,"2025-11-06","2026-06-15"],[1,4493,"[GRPO] Sequence-level TIS + MIS","LeonEricsson",2,"","",0,"2025-11-06","2026-02-01"],[1,4491,"I am unable to import experimental trainers","MohamedAliRashad",2,"","",0,"2025-11-06","2025-11-06"],[1,4489,"Parallelizing generation and evaluation for complex reward functions","AJain9199",3,"","",0,"2025-11-06","2026-06-11"],[1,4488,"Issues on GRPO with VLM and vLLM (cont.)","Fhrozen",0,"qgallouedec","#6570",1,"2025-11-06","2026-07-28"],[1,4481,"DPOTrainer._prepare_dataset() adds an extra eos_token to conversationally formatted inputs","DevonPeroutky",2,"","",0,"2025-11-06","2026-02-16"],[1,4472,"Move Online DPO to experimental submodule","behroozazarkhalili",1,"behroozazarkhalili","#4473",1,"2025-11-05","2025-11-24"],[1,4471,"Move CPO to experimental submodule","behroozazarkhalili",2,"","",0,"2025-11-05","2025-11-21"],[1,4469,"Move XPOTrainer to trl.experimental","behroozazarkhalili",1,"behroozazarkhalili","#4485",1,"2025-11-05","2025-11-06"],[1,4468,"Move RLOOTrainer to trl.experimental","behroozazarkhalili",2,"","",0,"2025-11-05","2025-12-05"],[1,4467,"Move PRMTrainer to trl.experimental","behroozazarkhalili",1,"behroozazarkhalili","#4483",1,"2025-11-05","2025-11-22"],[1,4466,"Move PPOTrainer to trl.experimental","behroozazarkhalili",1,"behroozazarkhalili","#4482",1,"2025-11-05","2025-11-13"],[1,4465,"Move ORPOTrainer to trl.experimental","behroozazarkhalili",1,"behroozazarkhalili","#4480",1,"2025-11-05","2025-11-21"],[1,4464,"Move NashMDTrainer to trl.experimental","behroozazarkhalili",1,"behroozazarkhalili","#4477",1,"2025-11-05","2025-11-13"],[1,4463,"Move KTOTrainer to trl.experimental","behroozazarkhalili",2,"","",0,"2025-11-05","2026-06-05"],[1,4462,"Move GKDTrainer to trl.experimental","behroozazarkhalili",1,"behroozazarkhalili","#4474",1,"2025-11-05","2025-11-13"],[1,4461,"Move OnlineDPOTrainer to trl.experimental","behroozazarkhalili",2,"","",0,"2025-11-05","2025-11-24"],[1,4460,"Move CPOTrainer to trl.experimental","behroozazarkhalili",1,"behroozazarkhalili","#4470",1,"2025-11-05","2025-11-13"],[1,4451,"prepare_multimodal_messages's bug in _generate_and_score_completions","wswaq",0,"qgallouedec","#5067",1,"2025-11-05","2026-02-17"],[1,4447,"CI fails with dev dependencies: RuntimeError: Could not infer dtype of dict","albertvillanova",1,"albertvillanova","#4448",1,"2025-11-04","2025-11-04"],[1,4443,"Support of LPO","MohamedAliRashad",3,"","",0,"2025-11-03","2026-06-29"],[1,4419,"GRPO with reward model. CUDA out of memory. How to fix? Thank you very much.","guotong1988",2,"","",0,"2025-11-01","2026-06-10"],[1,4416,"SFTTrainer creates empty adapter.safetensors file while saving when training with LoRA and Deepspeed Zero3","maximus-21",2,"","",0,"2025-10-31","2026-07-07"],[1,4413,"What is the default value of num_processes?","thisisraghavkumar",2,"","",0,"2025-10-31","2025-10-31"],[1,4407,"Complete paper index","qgallouedec",2,"","",0,"2025-10-30","2026-02-20"],[1,4404,"Remove setting chat template in most trainers","qgallouedec",0,"behroozazarkhalili","#4437",1,"2025-10-30","2025-11-03"],[1,4403,"Move `masked_mean`, `masked_var` and `masked_whiten` to `ppo.py`","qgallouedec",0,"Harras3","#4444",0,"2025-10-30","2025-11-05"],[1,4402,"Make vLLM server OpenAI-compatible","qgallouedec",3,"","",0,"2025-10-30","2026-06-22"],[1,4401,"Move everything in `test/slow` to their respective trainer test files","qgallouedec",0,"ishitab02","#4408",0,"2025-10-30","2025-10-31"],[1,4400,"Move everything related to Judges to `trl.experimental`","qgallouedec",0,"behroozazarkhalili","#4439",1,"2025-10-30","2025-11-07"],[1,4399,"Update or remove some of the notebooks","qgallouedec",0,"behroozazarkhalili","#4435",1,"2025-10-30","2025-11-04"],[1,4398,"Add kernels to Docker images","qgallouedec",0,"ishitab02","#4445",0,"2025-10-30","2025-11-06"],[1,4397,"Remove or move Multi Adapter RL","qgallouedec",0,"behroozazarkhalili","#4436",1,"2025-10-30","2025-11-04"],[1,4396,"Remove Sentiment Tuning Examples","qgallouedec",0,"Harras3","#4424",0,"2025-10-30","2025-11-04"],[1,4395,"Move everything related to Mergekit to the experimental submodule","qgallouedec",1,"qgallouedec","#4608",1,"2025-10-30","2025-12-05"],[1,4393,"Problems with Cross-Tokenizer Alignment in Correctness and Efficiency","JqzChandler",0,"kashif","#5885",1,"2025-10-30","2026-06-07"],[1,4392,"Cross-Tokenizer Alignment's problem in correctness and efficiency","JqzChandler",2,"","",0,"2025-10-30","2025-10-30"],[1,4390,"failed to start On-policy distillation","airlsyn",0,"kashif","#4765",1,"2025-10-30","2026-01-08"],[1,4387,"Remove support for FSDP1","qgallouedec",2,"","",0,"2025-10-30","2026-07-23"],[1,4386,"Reference supported trainers in Liger Kernel integration guide","qgallouedec",0,"behroozazarkhalili","#4432",1,"2025-10-30","2025-11-03"],[1,4385,"Use a common 'trl-lib` namespace for the models/datasets/spaces","qgallouedec",0,"behroozazarkhalili","#4431",1,"2025-10-30","2026-02-16"],[1,4384,"Write the subsection \"Multi-Node Training\"","qgallouedec",2,"","",0,"2025-10-30","2026-02-13"],[1,4383,"Add PEFT subsection to \"Reducing Memory Usage\"","qgallouedec",0,"behroozazarkhalili","#4430",1,"2025-10-30","2025-11-07"],[1,4382,"Populate \"Speeding Up Training\"","qgallouedec",0,"behroozazarkhalili","#4428",1,"2025-10-30","2025-12-01"],[1,4381,"Remove outdated warning about batch contamination","qgallouedec",0,"Harras3","#4423",0,"2025-10-30","2025-11-04"],[1,4380,"Fully transition from `flash-attn` to `kernels`","qgallouedec",0,"tamoghnokandar","#4426",0,"2025-10-30","2025-11-13"],[1,4379,"Remove or populate \"Training customization\"","qgallouedec",0,"behroozazarkhalili","#4427",1,"2025-10-30","2025-12-01"],[1,4378,"Extend basic usage example to all supported CLIs","qgallouedec",0,"behroozazarkhalili","#4425",1,"2025-10-30","2025-11-14"],[1,4377,"Move every section of `conceptual_guides/experimental` into its own section in `experimental`","qgallouedec",0,"ethanknights","#4411",0,"2025-10-30","2025-10-31"],[1,4376,"Rewrite `peft_integration.md`","qgallouedec",0,"behroozazarkhalili","#4421",1,"2025-10-30","2025-11-24"],[1,4375,"Remove guidance about converting conversational to standard.","qgallouedec",0,"behroozazarkhalili","#4422",1,"2025-10-30","2025-11-03"],[1,4374,"Road to v1","qgallouedec",2,"","",0,"2025-10-30","2026-03-31"],[1,4369,"CI fails for slow tests: TypeError: PreTrainedTokenizerFast._batch_encode_plus() got an unexpected keyword arg","albertvillanova",1,"albertvillanova","#4370",1,"2025-10-29","2025-10-29"],[1,4368,"GKD: multimodal inputs?","e-zorzi",2,"","",0,"2025-10-29","2025-11-07"],[1,4365,"CI fails for experimental tests: TypeError: GRPOWithReplayBufferTrainer.update_with_replay_buffer() got an une","albertvillanova",1,"albertvillanova","#4366",1,"2025-10-29","2025-10-31"],[1,4363,"does it support multi reward function?","zhaichenhao1889",2,"","",0,"2025-10-29","2025-10-29"],[1,4358,"GRPOConfig.max_prompt_length does not truncate prompts for a conversational dataset with vllm (since v0.25.0)","muupan",2,"","",0,"2025-10-28","2026-03-06"],[1,4347,"cuda oom when finetuning mllama with lora and sfttrainer","osaidr",3,"","",0,"2025-10-27","2026-06-24"],[1,4346,"Issues exist with the settings for num generation and the number of model-generated responses (Completion). Th","Haoyong23",2,"","",0,"2025-10-27","2025-10-30"],[1,4345,"CI fails for slow tests: RuntimeError: min(): Expected reduction dim to be specified for input.numel() == 0","albertvillanova",1,"albertvillanova","#4348",1,"2025-10-27","2025-10-27"],[1,4343,"Refactoring out the logps and importance comptuation out of generate_and_score in GRPOTrainer","wirmius",2,"","",0,"2025-10-27","2026-06-30"],[1,4342,"为什么completions一直等于4？和num-generations不同？","Haoyong23",3,"","",0,"2025-10-26","2025-10-26"],[1,4341,"The \"report_to\" parameter in GRPOConfig","LWShowTime",2,"","",0,"2025-10-26","2025-10-29"],[1,4339,"Packing with VLMs","jiosephlee",4,"","",0,"2025-10-25",""],[1,4323,"CI fails with dev dependencies: HFValidationError: Repo id must use alphanumeric chars","albertvillanova",2,"","",0,"2025-10-22","2025-11-26"],[1,4321,"CI fails with dev dependencies: ValueError: zero-size array to reduction operation maximum which has no identi","albertvillanova",1,"albertvillanova","#4322",1,"2025-10-22","2025-10-22"],[1,4319,"Replace flash-attn with kernels in CI","albertvillanova",0,"tamoghnokandar","#4426",0,"2025-10-22","2025-11-13"],[1,4314,"torch.utils.checkpoint.CheckpointError when using LoRA + fsdp2 Context Parallel","zxgx",2,"","",0,"2025-10-21","2026-06-15"],[1,4311,"Turn off Qwen3's reasoning mode","Tuziking",0,"pramodith","#4350",1,"2025-10-20","2025-10-29"],[1,4309,"Reference model for LoRA","SpaceHunterInf",2,"","",0,"2025-10-20","2025-11-24"],[1,4308,"CI fails for Python 3.9: ValueError: zero-size array to reduction operation maximum which has no identity","albertvillanova",1,"albertvillanova","#41747",1,"2025-10-20","2025-11-27"],[1,4299,"RuntimeError in trl vllm-serve CLI: Accelerate state not initialized before logging","BY571",2,"","",0,"2025-10-17","2025-11-24"],[1,4294,"A customized trainer for tree-level optimization","martinjingyu",3,"","",0,"2025-10-16","2026-07-04"],[1,4289,"Image features and image tokens do not match with sfttrainer and vl models","osaidr",2,"","",0,"2025-10-15","2025-10-27"],[1,4286,"CI fails for slow tests: ImportError: vLLM is not installed","albertvillanova",1,"albertvillanova","#4287",1,"2025-10-15","2025-10-15"],[1,4284,"token_type_ids missing in DPOTrainer","aweers",1,"aweers","#4285",0,"2025-10-15","2025-10-15"],[1,4282,"CI fails for slow tests: OSError: You are trying to access a gated repo","albertvillanova",1,"albertvillanova","#4283",1,"2025-10-15","2025-10-15"],[1,4281,"CI fails with dev dependencies: torch.AcceleratorError: CUDA error: device-side assert triggered","albertvillanova",1,"albertvillanova","#41757",1,"2025-10-15","2025-10-22"],[1,4280,"help! activation_offloading True got Backward pass ValueError","YooSungHyun",2,"","",0,"2025-10-15","2026-06-23"],[1,4277,"LoRA from CP-based SFT doesn’t reproduce training data at inference (train loss ≈ 0, context ≈ 100k)","gell1029",2,"","",0,"2025-10-14","2025-10-16"],[1,4274,"Question: What are protected tokens?","SwayamInSync",2,"","",0,"2025-10-14","2025-10-14"],[1,4273,"CI fails with dev dependencies: TypeError: find_adapter_config_file() got an unexpected keyword argument '_ada","albertvillanova",2,"","",0,"2025-10-14","2025-10-27"],[1,4272,"CI fails with dev dependencies: TypeError: 'NoneType' object is not subscriptable","albertvillanova",1,"albertvillanova","#41764",1,"2025-10-14","2025-11-27"],[1,4271,"_get_train_sampler in RLOOTrainer, GRPOTrainer...","yigexu",2,"","",0,"2025-10-14","2026-07-30"],[1,4256,"Use VLLM for Evaluation?","SpaceHunterInf",2,"","",0,"2025-10-10","2026-05-21"],[1,4251,"CI fails with dev dependencies: ImportError for 'require_torch_gpu_if_bnb_not_multi_backend_enabled'","albertvillanova",1,"albertvillanova","#4253",1,"2025-10-10","2025-10-10"],[1,4249,"Disable w&b by default","getStRiCtd",3,"","",0,"2025-10-10","2026-06-04"],[1,4248,"CI fails with dev dependencies: AttributeError: type object 'DynamicCache' has no attribute 'from_legacy_cache","albertvillanova",1,"albertvillanova","#4250",1,"2025-10-10","2025-10-10"],[1,4239,"CI fails: IndentationError: expected an indented block after function definition on line 3","albertvillanova",1,"albertvillanova","#4240",1,"2025-10-09","2025-10-09"],[1,4237,"CI fails: torch.OutOfMemoryError: CUDA out of memory","albertvillanova",1,"albertvillanova","#4238",1,"2025-10-09","2025-10-10"],[1,4228,"CI fails: torch.AcceleratorError: CUDA error: an illegal memory access was encountered","albertvillanova",1,"albertvillanova","#4229",1,"2025-10-08","2025-10-08"],[1,4223,"[RFC] Moving Most TRL Trainers to the experimental Submodule to Streamline the Core","qgallouedec",2,"","",0,"2025-10-07","2026-01-16"],[1,4210,"CI fails: ValueError: Found unknown kwargs={'hidden_size': 32}","albertvillanova",0,"qgallouedec","#4213",1,"2025-10-06","2025-10-06"],[1,4208,"CI fails for Python 3.9 tests: TypeError: unsupported operand type(s) for |: 'type' and 'NoneType'","albertvillanova",1,"albertvillanova","#4209",1,"2025-10-06","2025-10-06"],[1,4205,"vllm importance sampling correction causes tensor size mismatch","wangxiaodong1021",3,"","",0,"2025-10-05","2026-07-16"],[1,4204,"The error of training qwen3-30b-a3b using SFT.","Zhengsh123",2,"","",0,"2025-10-05","2026-08-04"],[1,4202,"Did the GRPO reward function just changed the format","weathon",2,"","",0,"2025-10-04","2026-07-09"],[1,4199,"[question] how does one use the SFTTrainer with VLMs for prompt completion task?","osaidr",2,"","",0,"2025-10-02","2025-10-05"],[1,4189,"sft_trainer.DataCollatorForVisionLanguageModelling does not account for \"non-standard\" processor outputs (e.g.","KarelKenens",0,"qgallouedec","#4190",1,"2025-10-01","2025-10-08"],[1,4184,"[Feature] Add a bool enable/disable flag for `entropy logging`","steveepreston",3,"","",0,"2025-09-30","2026-09-04"],[1,4182,"[Bug] `AttributeError: module 'torch' has no attribute 'xla'` after upgrading trl 0.21.0 > 0.22.0","steveepreston",2,"","",0,"2025-09-30","2026-07-13"],[1,4177,"[New Trainer] TreeRPO: Hierarchical Credit Assignment for Deterministic-Correctness Tasks","omrisapir1",3,"","",0,"2025-09-30","2026-07-14"],[1,4175,"CI fails: ImportError: FlashAttention2","albertvillanova",1,"albertvillanova","#4176",1,"2025-09-30","2025-10-01"],[1,4172,"CI fails: ValueError: Unknown loss type: dapo","albertvillanova",1,"albertvillanova","#4173",1,"2025-09-30","2025-09-30"],[1,4169,"Is it actually possible to train a VLM with multiple image inputs using GRPOTrainer?","Randomdude11",2,"","",0,"2025-09-29","2025-10-04"],[1,4168,"SFTTrainer compute_loss fails when model is PEFT Prefix-tuning (size mismatch)","gabrielle-lebellier",0,"pramodith","#4196",1,"2025-09-29","2025-10-08"],[1,4167,"DDPO?","rteehas",2,"","",0,"2025-09-29","2025-10-07"],[1,4159,"vLLM not computing correct log probs when using GRPO with =/=1 temperature","YonatanGideoni",2,"","",0,"2025-09-28","2025-10-04"],[1,4150,"CI fails: TypeError: RLOOTrainer._get_per_token_logps_and_entropies() got an unexpected keyword argument 'toke","albertvillanova",1,"albertvillanova","#4148",1,"2025-09-26","2025-09-29"],[1,4149,"CUDA error: uncorrectable ECC error encountered","AragornHorse",3,"","",0,"2025-09-26","2026-06-15"],[1,4147,"Prompt-completion does not use correct labels when passing conversational format to SFT","MarkusSpanring",0,"behroozazarkhalili","#4201",1,"2025-09-26","2025-10-07"],[1,4144,"Support for Qwen 2.5 VL Video Modality in GRPOTrainer","AnyaCoder",3,"","",0,"2025-09-25","2025-10-03"],[1,4142,"CI fails: ValueError: `token_type_ids` is required as a model input when training","albertvillanova",1,"albertvillanova","#4148",1,"2025-09-25","2025-09-29"],[1,4131,"CI fails: torch.AcceleratorError: CUDA error: device-side assert triggered","albertvillanova",1,"albertvillanova","#4138",1,"2025-09-24","2025-09-24"],[1,4130,"Async Support for Reward Functions","mohamadlakkis",0,"pramodith","#4567",1,"2025-09-24","2025-09-26"],[1,4129,"Setting trust_remote_code=True for vLLM in GRPOTrainer with vllm_mode==\"colocate\"","muupan",1,"muupan","#4186",0,"2025-09-24","2026-07-06"],[1,4116,"[QUESTION] About applying chat template for base model via `clone_chat_template`","binhere",2,"","",0,"2025-09-22","2026-07-08"],[1,4110,"How does `trl` know what part of dataset is prompt and completion in the following situation?","bminesh-shah",2,"","",0,"2025-09-19","2025-09-19"],[1,4104,"Loss calculation of `RewardTrainer` may be inaccurate when performing gradient accumulation?","jue-jue-zi",2,"","",0,"2025-09-18","2025-10-05"],[1,4103,"RuntimeError: Expected tensor for argument #1 'indices' to have one of the following scalar types: Long, Int; ","harshitkd",2,"","",0,"2025-09-18","2025-09-19"],[1,4100,"DeepSpeed Zero 3 + LoRA 32B model still OOM on 8*H100 (each GPU load one model?)","Aaronhuang-778",2,"","",0,"2025-09-17","2025-10-07"],[1,4099,"CI fails: ValueError: `num_hidden_layers` (2) must be equal to the number of layer types (64)","albertvillanova",1,"albertvillanova","#4101",1,"2025-09-17","2025-09-20"],[1,4098,"trl 0.23 showing error: name 'ParallelismConfig' is not defined. Did you mean: 'parallelism_config'","Fhrozen",2,"","",0,"2025-09-17","2025-09-17"],[1,4096,"SFTTrainer fails with DFT loss during evaluation","solenetarride",0,"pramodith","#4112",1,"2025-09-16","2025-09-24"],[1,4086,"Logging behaviour when steps_per_generation > gradient_accumulation_steps in GRPOTrainer","vatsal-kr",2,"","",0,"2025-09-15","2025-10-05"],[1,4077,"prompt+completion training vs messages training for VLM","weathon",2,"","",0,"2025-09-13","2026-07-06"],[1,4072,"deepspeed+trl load model twice","RRRRRayyyyy",2,"","",0,"2025-09-12","2025-09-15"],[1,4071,"Possible bug in tutorial","GabrieleGiudic",0,"qgallouedec","#3906",1,"2025-09-12","2026-02-20"],[1,4070,"MOE aux loss bug","mst272",0,"pramodith","#4078",1,"2025-09-12","2025-09-14"],[1,4066,"SFTraine compute_loss after trl 0.22.2 works uncorrect with logits_to_keep","BogdanDidenko",2,"","",0,"2025-09-11","2025-09-22"],[1,4062,"DPOTrainer: pick vision path by processing object type; always pass a tokenizer to chat template","ginkyenglee",2,"","",0,"2025-09-11","2026-02-20"],[1,4061,"GRPO on InternVL3.5 does not support dynamic vision tokens","sanchit97",0,"qgallouedec","#6619",1,"2025-09-11","2026-08-20"],[1,4044,"Whether to support GFPO training？","Kangdi123",2,"","",0,"2025-09-09","2025-09-30"],[1,4040,"why kl = nan when grpo train?","uilstong",2,"","",0,"2025-09-09","2026-05-29"],[1,4036,"About \"or None\" and \"defaults to None\"","qgallouedec",0,"albertvillanova","#4058",1,"2025-09-08","2025-09-11"],[1,4035,"CI fails: ValueError: zero-size array to reduction operation maximum which has no identity","albertvillanova",1,"albertvillanova","#4120",1,"2025-09-08","2025-09-22"],[1,4029,"SFTTrainer with PEFT model","lylaiyy",2,"","",0,"2025-09-08","2026-07-10"],[1,4023,"Abnormal results during DPO training","wjjwyj",2,"","",0,"2025-09-06","2026-02-20"],[1,4021,"Does the reward function trainer support multimodal models now?","hlp2020",3,"","",0,"2025-09-06","2026-07-24"],[1,4018,"No warning for unsupported int4 quantization","MRiabov",2,"","",0,"2025-09-05","2026-07-19"],[1,4016,"ParallelismConfig not applied in GRPOTrainer: Trainer._prepare_context_parallel_inputs expects dict[torch.Tens","JdRion",2,"","",0,"2025-09-05","2026-07-04"],[1,4015,"Does the grpo vllm colocate training now support the InternVL3 8B model?","DearFishi",2,"","",0,"2025-09-05","2026-06-22"],[1,4010,"`current_gradient_accumulation_steps` is undefined when `eval_on_start==True`","konstantinjdobler",0,"qgallouedec","#4852",1,"2025-09-04","2026-01-17"],[1,4009,"Can not  from trl import DataCollatorForLanguageModeling","lylaiyy",0,"LeonEricsson","#4011",1,"2025-09-04","2025-09-04"],[1,4004,"Training Step of GRPO in Wandb.","mandyyyyii",3,"","",0,"2025-09-04","2026-06-02"],[1,4002,"DPO trainer with video content","GabrieleGiudic",2,"","",0,"2025-09-03","2025-09-04"],[1,3991,"scale_rewards malfunctioned in GRPOTrainer","Peter-Chou",1,"Peter-Chou","#3992",0,"2025-09-02","2025-09-03"],[1,3988,"accelerator.sync_gradients","AriesJin",2,"","",0,"2025-09-01","2025-09-02"],[1,3985,"Feature Request: Save/Load Precomputed Ref Log-Probabilities in DPOTrainer","ginkyenglee",0,"qgallouedec","#3906",1,"2025-09-01","2026-02-20"],[1,3983,"`GRPOTrainer` uses undefined `self.current_gradient_accumulation_steps`, causing `AttributeError` during train","ahatamiz",0,"qgallouedec","#4852",1,"2025-08-31","2025-09-03"],[1,3982,"DPOTrainer uses tokenizer instead of processor for Gemma3 vision","supreme-gg-gg",2,"","",0,"2025-08-30","2026-02-20"],[1,3980,"Extremely persistent error - ValueError: 'generation_batch_size' and 'steps_per_generation' can not be both co","Rakshith12-pixel",0,"qgallouedec","#3981",1,"2025-08-30","2025-08-30"],[1,3979,"GRPOTrainer vLLM colocate hardcodes MASTER_PORT=12345 so no parallel runs possible","ruggsea",0,"kashif","#4057",1,"2025-08-30","2025-09-23"],[1,3974,"kto trainer invalid configuration error","bryanchrist",2,"","",0,"2025-08-29","2026-06-29"],[1,3973,"REQUEST: Dynamic Sampling for GRPO","wenquanlu",3,"","",0,"2025-08-29","2026-05-29"],[1,3967,"REQUEST：Add estimation of flos metric in GRPO Trainer","LLMOON",3,"","",0,"2025-08-28","2026-07-17"],[1,3966,"[Question] Why isn't vanilla REINFORCE implemented?","nityadav",3,"","",0,"2025-08-27","2026-07-28"],[1,3963,"[QST] Colocation and resharding","jeromeku",2,"","",0,"2025-08-27","2025-09-01"],[1,3961,"PPO grad_norm is 0.0","faker52",2,"","",0,"2025-08-27","2026-05-28"],[1,3958,"sft_video_llm example fail","yao-matrix",0,"qgallouedec","#6193",1,"2025-08-26","2026-06-30"],[1,3957,"sft_gemma3 example doesn't work","yao-matrix",0,"SamuelBarryCS","#4080",0,"2025-08-26","2025-09-23"],[1,3948,"Why not use AutoModel for ref_model in grpo trainer?","csshihao",2,"","",0,"2025-08-25","2026-06-30"],[1,3945,"LoRA-GA: Low-Rank Adaptation with Gradient Approximation","electroglyph",3,"","",0,"2025-08-24","2026-06-06"],[1,3944,"Can we avoid saving the optimization stage?","HelloWorldLTY",2,"","",0,"2025-08-23","2025-08-24"],[1,3939,"Should the reference model of GRPO be an actual model?","shaojun0",3,"","",0,"2025-08-22","2026-07-06"],[1,3934,"Bugs when sft with mixing pure text and multimodal data","shiym2000",2,"","",0,"2025-08-21","2026-07-06"],[1,3933,"`GRPOTrainer` with `top_entropy_quntile < 1` causes hang with multi gpu training","avishaiElmakies",0,"akakakakakaa","#4041",0,"2025-08-21","2025-09-12"],[1,3928,"Add support for RLPR","mitchelldehaven",3,"","",0,"2025-08-20","2026-07-21"],[1,3927,"Using assistant_only_loss=True with sequence length > max_length fails silently","jonnyli1125",0,"qgallouedec","#6320",1,"2025-08-20","2026-07-14"],[1,3926,"SFTTrainer freezes LoRA adapter when using PEFT model as argument","dvgodoy",0,"Hoesu","#4081",0,"2025-08-20","2025-09-18"],[1,3925,"When I want to use client to request trl vllm-serve, it always timeout when init_communicator()","JackeyZhang1001",3,"","",0,"2025-08-20","2026-07-06"],[1,3924,"Feature request for GRPO trainer: vLLM guided decoding with xgrammar (JSON)","sanghak123",2,"","",0,"2025-08-20","2026-07-23"],[1,3920,"Does the trl library support the Lite PPO algorithm?","ArcherShirou",0,"pramodith","#3935",1,"2025-08-19","2025-08-21"],[1,3918,"How to use trl-SFTTrainer to train Qwen-30B-A3B?","JeffWb",2,"","",0,"2025-08-19","2026-05-26"],[1,3915,"Inconsistent `apply_chat_template` behaviour for multimodal dataset between SFT and GRPO","ishaan-rawal-ai",0,"MengAiDev","#3919",0,"2025-08-18","2025-08-20"],[1,3910,"max_length causing training loss issues not found in v0.19.1","AmazingGabriel16",3,"","",0,"2025-08-16","2026-08-06"],[1,3903,"[GRPO Trainer] Accuracy reward stays 0","Revist",3,"","",0,"2025-08-15","2026-06-22"],[1,3901,"sft_gemma3 example fail","yao-matrix",2,"","",0,"2025-08-14","2025-08-27"],[1,3896,"How to gather completions before computing rewards in GRPOTrainer","rubickkcibur",2,"","",0,"2025-08-14","2025-09-03"],[1,3893,"What are the important metrics that help gauge accuracy and validity of training?","RageItalic",2,"","",0,"2025-08-14","2026-07-09"],[1,3892,"Is there a standard GRPO GSM8k process for TRL?","mitchelldehaven",2,"","",0,"2025-08-13","2026-05-29"],[1,3887,"Bug in BFD packing","rdolmedo",0,"MengAiDev","#3888",0,"2025-08-12","2025-08-17"],[1,3885,"GRPO fails with the \"server\" vllm_mode when num_processes > 1","Kirill-Kravtsov",2,"","",0,"2025-08-12","2025-08-12"],[1,3881,"GRPOTrainer training causes model to completely deteriorate when training with accelerate","Randomdude11",2,"","",0,"2025-08-12","2026-05-25"],[1,3878,"truncate mode for response","shiwanghua",3,"","",0,"2025-08-11","2026-07-01"],[1,3877,"Dynamic Fine Tuning, an improvement of SFT","1485840691",2,"","",0,"2025-08-11","2025-09-22"],[1,3872,"PermissionError: [Errno 13] Permission denied: 'Qwen3-0.6B-SFT'","vishaljoshi24",2,"","",0,"2025-08-08","2026-06-19"],[1,3871,"Add reward functions to support RLCR","pramodith",3,"","",0,"2025-08-07","2026-07-31"],[1,3870,"Unable to replicate official GRPO for VLM tutorial - AttributeError: 'dict' object has no attribute 'replace'","Randomdude11",2,"","",0,"2025-08-07","2025-08-08"],[1,3868,"ValueError: Unrecognized configuration class <class 'transformers.models.qwen2_5_vl.configuration_qwen2_5_vl.Q","Randomdude11",2,"","",0,"2025-08-07","2025-08-07"],[1,3865,"Validate the `vllm_mode` is either `server` or `colocate`","sergiopaniego",1,"sergiopaniego","#3866",1,"2025-08-07","2025-08-09"],[1,3860,"`bf16` option in training config prevents FP8 Training","akakakakakaa",3,"","",0,"2025-08-06","2026-06-17"],[1,3859,"'LLMEngine' object has no attribute 'model_executor'","EvilCalf",2,"","",0,"2025-08-06","2026-05-24"],[1,3858,"when use GRPO+ deepspeed_zero3 + ds3_gather_for_generation=False, stuck in the training stage, step is still 0","nstl-zyb",3,"","",0,"2025-08-06","2026-07-18"],[1,3857,"GRPO Trainer loss error? as the info log, why 'rewards/if_rewrite_reward/std': 5.080004692077637 (is not 0), b","harmonytan",2,"","",0,"2025-08-06","2025-08-12"],[1,3853,"vllm prepends two BOS for LLama","wenquanlu",2,"","",0,"2025-08-06","2026-06-19"],[1,3847," Issues at GRPO with VLM","Fhrozen",2,"","",0,"2025-08-05","2025-10-07"],[1,3846,"Ideas to Improve GRPO Training Speed","jp1924",2,"","",0,"2025-08-05","2026-02-18"],[1,3842,"accelerate reducing the batch size and crashing GRPO","limlimg",0,"qgallouedec","#6411",1,"2025-08-04","2026-07-20"],[1,3839,"GRPO with google/gemma-3-1b-it torch.compile error","AdityaKulshrestha",2,"","",0,"2025-08-03","2026-07-22"],[1,3834,"Wrong default clipping params for GSPO","pramodith",2,"","",0,"2025-08-01","2025-08-18"],[1,3829,"Improve RLOO Trainer memory efficiency through string-level processing optimization","nevertmr",2,"","",0,"2025-07-31","2025-09-22"],[1,3828,"TRL doesn't support gemma-3","awestover",0,"albertvillanova","#5907",1,"2025-07-31","2026-07-05"],[1,3827,"DataCollatorForCompletionOnlyLM","tejassaboo",2,"","",0,"2025-07-31","2026-07-12"],[1,3825,"RLOOTrainer tldr experiments not reproducible","jltchiu",2,"","",0,"2025-07-31","2025-09-22"],[1,3823,"Loss not calculated correctly when using GSPO with default loss type bnpo (`importance_sampling_level == \"sequ","avishaiElmakies",0,"discobot","#6042",0,"2025-07-31","2026-06-23"],[1,3822,"Processing class does not have EOS token","debasisdwivedy",2,"","",0,"2025-07-31","2026-05-24"],[1,3819,"[Question] about Profiling & Logging Execution Time to WANDB using GRPO","jonasjunzhang",2,"","",0,"2025-07-31","2026-06-25"],[1,3812,"Prompttuning/ptuning not working with SFTTrainer due to token accuracy","DominikHil",0,"zk-quantum","#3821",0,"2025-07-30","2025-08-15"],[1,3811,"GSPO-token implementation","BounharAbdelaziz",2,"","",0,"2025-07-30","2026-06-27"],[1,3809,"Move vllm_serve FastAPI app out of main method","nialldevlin1",3,"","",0,"2025-07-30","2026-07-10"],[1,3808,"Failed to broadcast barrier_id Bug in `_move_model_to_vllm()` of grpo_trainer when training lora model","shiwanghua",3,"","",0,"2025-07-30","2026-06-12"],[1,3807,"🐛 Bug Report: RuntimeError: element 0 of tensors does not require grad during .train() using SFTTrainer (lates","dt-cs",2,"","",0,"2025-07-30","2025-09-01"],[1,3797,"How to view the training parameters after training is completed","Tuziking",2,"","",0,"2025-07-29","2026-06-18"],[1,3795,"Import error with the last release v0.20.0","khalil-Hennara",2,"","",0,"2025-07-29","2026-06-12"],[1,3788,"GRPO max_num_batched_tokens hardcoded","dgiofre",2,"","",0,"2025-07-28","2026-06-22"],[1,3787,"Sequence Shuffling Error with GSPO","Pclanglais",3,"","",0,"2025-07-28","2026-06-22"],[1,3782,"OnlineDPO missing vLLM server mode support (like GRPO has)","zahlmann",1,"zahlmann","#3783",0,"2025-07-27","2025-09-05"],[1,3781,"Potential oversight or incompatibility in `SFTTrainer`: `assistant_only_loss=True` silently ignored when `use_","andylvua",0,"qgallouedec","#3914",1,"2025-07-27","2025-08-18"],[1,3780,"[Feature Request] Add support for padding-free reward modeling training in TRL","zjjMaiMai",0,"qgallouedec","#4093",1,"2025-07-27","2025-09-30"],[1,3778,"Support for Group Sequence Policy Optimization","vedantdere",2,"","",0,"2025-07-26","2025-07-28"],[1,3777,"Implement VeriFree Trainer","johnsutor",3,"","",0,"2025-07-26","2026-07-02"],[1,3769,"during training, is there a way to modify or augment the original training data at a certain point in time dur","zhanglv0209",2,"","",0,"2025-07-25","2026-07-26"],[1,3768,"IterableDataset causes crash when using assistant_only_loss=True","Dandandooo",0,"LeonEricsson","#3771",1,"2025-07-24","2025-07-26"],[1,3764,"[BUG] Can GKD Trainer perform 111B+ Teacher distributed inference and Student distributed training concurrentl","seona21",2,"","",0,"2025-07-24","2026-06-26"],[1,3759,"OnlineDPO with use_vllm=True: 'str' object has no attribute 'config'","zahlmann",0,"kashif","#3765",1,"2025-07-23","2025-07-25"],[1,3757,"Question regarding GRPOTrainer's advantage clipping","haikangdeng",2,"","",0,"2025-07-23","2026-05-25"],[1,3751,"[Example Script] Is it possible to finetune VLM on the assistant only text instead of the entire sequence?","xhluca",3,"","",0,"2025-07-21","2026-07-04"],[1,3750,"Two potential issues in PPOTrainer: reward added to PAD token instead of EOS, and missing_eos_penalty may not ","noncollapse",3,"","",0,"2025-07-20","2026-06-26"],[1,3748,"model loading on multi-node ROCm","marscod",3,"","",0,"2025-07-20","2026-05-21"],[1,3742,"Bug when creating a reference model in GRPO with deepspeed zero3 for diffusion language models","xijia-tao",3,"","",0,"2025-07-18","2026-07-11"],[1,3739,"GRPO training hanging on a CPU","sameermahajan",2,"","",0,"2025-07-16","2026-06-12"],[1,3736,"Unable to use GrpoTrainer with vllm, timeouts","osaidr",2,"","",0,"2025-07-15","2026-05-24"],[1,3733,"Incorrect logits computation in GRPOTrainer._get_per_token_logps_and_entropies","jhamrick",2,"","",0,"2025-07-14","2025-07-14"],[1,3730,"How to design stable reward functions for open-ended text generation tasks in GRPO?","Jax922",3,"","",0,"2025-07-12","2026-06-24"],[1,3729,"GRPOTrainer.__init__() got an unexpected keyword argument 'compute_metrics'","CHOUDDIDIBUYAOSONG",2,"","",0,"2025-07-12","2025-07-20"],[1,3728,"The assistant_only_loss field becomes ineffective when packing=True is enabled.","GitMonkey0",0,"LeonEricsson","#3749",1,"2025-07-12","2025-07-22"],[1,3718,"Bug when save model: TypeError: Object of type BitsAndBytesConfig is not JSON serializable","shiwanghua",2,"","",0,"2025-07-10","2026-02-20"],[1,3717,"Add activation checkpointing to DPO and GRPO","mistysesame",4,"","",0,"2025-07-10",""],[1,3714,"Logits have nan values during GRPO training","RageItalic",2,"","",0,"2025-07-09","2025-07-20"],[1,3713,"concatenated_input_ids","daniel-dpq",2,"","",0,"2025-07-09","2025-07-24"],[1,3711,"index error","shiwanghua",0,"LeonEricsson","#3719",1,"2025-07-09","2025-07-11"],[1,3709,"question about entropy_threshold calculating","AGI-player",0,"pramodith","#3715",1,"2025-07-09","2025-07-10"],[1,3708,"[Feature Request] support dynamic sampling for GRPO trainer","wa008",4,"","",0,"2025-07-09",""],[1,3707,"ZeroDivisionError: division by zero  in online_dpo_trainer","shiwanghua",2,"","",0,"2025-07-09","2026-05-26"],[1,3706,"trl with vllm integration not working","PodoprikhinMaxim",2,"","",0,"2025-07-08","2026-07-06"],[1,3705,"Packing is currently broken?","jiosephlee",2,"","",0,"2025-07-08","2025-07-08"],[1,3701,"avoid OOM when call _forward() & logprobs","shiwanghua",3,"","",0,"2025-07-07","2026-07-07"],[1,3697,"Clarification on batch_size warning during padding_free training","jiosephlee",2,"","",0,"2025-07-05","2025-07-07"],[1,3695,"packing=True and no flash_attention should not warrant a warning?","jiosephlee",2,"","",0,"2025-07-04","2025-07-07"],[1,3694,"No loss logging when using LoraConfig + SFTTrainer","LukasStruppek",0,"qgallouedec","#39265",1,"2025-07-04","2025-07-08"],[1,3692,"Which collator to use for padding-free packing: WithFlattening or Default LanguageModelling?","jiosephlee",2,"","",0,"2025-07-03","2026-05-22"],[1,3691,"Question about Liger kernels removing columns except input_ids","jiosephlee",2,"","",0,"2025-07-03","2025-07-07"],[1,3689,"ValueError in grpo_trainer.py: batch_size being 0","rynnyang",2,"","",0,"2025-07-03","2026-06-13"],[1,3686,"question about prompts_text processing","AGI-player",0,"pramodith","#3698",1,"2025-07-03","2025-07-07"],[1,3684,"cpt/sft and rl on the same adapter/lora","Dludora",2,"","",0,"2025-07-03","2026-06-22"],[1,3683,"Infinitely repeating cache warning and no training occuring while using GRPO","RageItalic",2,"","",0,"2025-07-02","2025-07-04"],[1,3681,"_get_per_token_logps_and_entropies in GRPO implements the wrong logit slicing","ahatamiz",2,"","",0,"2025-07-02","2025-07-04"],[1,3679,"Generate a group completions always by initial policy model rather than trained model for each input when use_","wa008",2,"","",0,"2025-07-02","2025-07-03"],[1,3678,"OOM when training qwen3-0.6B model by GRPO trainer on 20G VRAM GPU","wa008",2,"","",0,"2025-07-02","2026-07-19"],[1,3676,"When using GRPOTrainer+use_vllm, a self.group.barrier wait is triggered","ansanyuan",0,"CarlosArguilar","#3690",0,"2025-07-02","2025-07-03"],[1,3675,"GRPO: sampled prompts repeat across steps, within a global step","tilaks",2,"","",0,"2025-07-01","2025-07-02"],[1,3671,"GRPO + vLLM colocate + PEFT hangs","RobHessQ",0,"albertvillanova","#6139",1,"2025-06-30","2026-06-26"],[1,3668,"GRPOTrainer reward functions cannot access training progress","seungduk-yanolja",1,"seungduk-yanolja","#3669",0,"2025-06-30","2025-07-01"],[1,3666,"quantile() input type error when using bf16","gitabtion",1,"gitabtion","#3667",0,"2025-06-30","2025-07-04"],[1,3665,"Large steps_per_generation fails","Muennighoff",2,"","",0,"2025-06-29","2025-07-01"],[1,3664,"CUDA Out of Memory for qwen-7b with TRL PPO on 8 A100 GPUs","duyuwen-duen",3,"","",0,"2025-06-29","2026-07-19"],[1,3663,"Could GRPOTrainer support pre-generated completions","evehsu",2,"","",0,"2025-06-29","2025-07-24"],[1,3662,"What is the point of steps_per_gen in GRPO Trainer","ankur6ue",2,"","",0,"2025-06-28","2026-06-05"],[1,3661,"Does trl GRPO support lora with vllm serve on a separeate machine?","JiaqinKang",2,"","",0,"2025-06-28","2025-07-02"],[1,3659,"DPOTrainer generate_during_eval support for MLFlow","dhruvmullick",2,"","",0,"2025-06-27","2025-07-02"],[1,3656,"generation_batch_size divisibility check error in __post_init__ in grpo_config.py","AGI-player",0,"pramodith","#3658",1,"2025-06-27","2025-06-28"],[1,3654,"GRPO Trainer doesn't work with vllm and lora+bnb quantization","mkumar-dmx",2,"","",0,"2025-06-26","2025-07-23"],[1,3653,"Conflict of generation batch size between definition and _post_init_  in grpo_config.py","gpchuan",2,"","",0,"2025-06-26","2025-07-03"],[1,3652,"Add Image input support for VLLMClient","eslambakr",2,"","",0,"2025-06-26","2026-06-30"],[1,3648,"It's impossible to connect to vllm server in 0.9.1 and run GRPO in server mode","ahatamiz",3,"","",0,"2025-06-25","2026-06-05"],[1,3647,"Generalized Knowledge Distillation Trainer ignores `data_collator` argument.","BlackSamorez",2,"","",0,"2025-06-25","2026-06-17"],[1,3645,"strategy=\"ffd\" implements BFD packing","toslunar",0,"LeonEricsson","#3696",1,"2025-06-25","2025-07-07"],[1,3644,"0.19.0 breaks the chat format in GRPO trainer leading to severely mangled chat generations","mchughalex",0,"pramodith","#3651",1,"2025-06-25","2025-06-26"],[1,3641,"Unnecessary breaking change in `SFTTrainer._prepare_dataset` from 0.19.0 compared to 0.18.2","jannisborn",1,"jannisborn","#3677",0,"2025-06-24","2025-07-04"],[1,3640,"Scaling logits with temperature values closer to 0 makes probabilities 0","zkpranav",2,"","",0,"2025-06-24","2026-06-18"],[1,3635,"trainer.train(resume_from_checkpoint=last_checkpoint) doesn't work!","seTalent",2,"","",0,"2025-06-24","2025-06-24"],[1,3633,"GRPO Colocate mode crashes if generation_kwargs is undefined","ahatamiz",1,"ahatamiz","#3634",0,"2025-06-23","2025-06-24"],[1,3632,"RuntimeError: Engine core initialization failed. See root cause above. Failed core proc(s): {}","HT-NEKO",3,"","",0,"2025-06-23","2026-06-21"],[1,3622,"GRPO server mode connection error in the middle of training","re-imagined",2,"","",0,"2025-06-20","2026-06-22"],[1,3621,"Repeated calculation of ref_per_token_logps in grpo_trainer","xiewen354",2,"","",0,"2025-06-19","2025-06-19"],[1,3618,"SFTTrainer crashes when using dataloader > 0 with deepspeed installed","avishaiElmakies",2,"","",0,"2025-06-18","2026-05-29"],[1,3616,"Latest default config of bf16=True breaks test cases when run on CPU.","pramodith",0,"behroozazarkhalili","#6036",1,"2025-06-18","2026-07-22"],[1,3611,"my code outputs \"tensor 'kwargs['input_ids']' size mismatch\" while training. ckpt is not saved and the log is ","HT-NEKO",2,"","",0,"2025-06-18","2026-07-01"],[1,3610,"Attribute error during model card creation in GRPO","toslali-ibm",0,"LeonEricsson","#3613",1,"2025-06-17","2025-06-18"],[1,3608,"GRPO server mode gets stuck in model update","toslali-ibm",2,"","",0,"2025-06-17","2025-07-04"],[1,3605,"How to convert my multiturn dialogue dataset？","Miaoqinghong",2,"","",0,"2025-06-17","2025-09-22"],[1,3604,"huggingface_hub.errors.HFValidationError: Repo id must be in the form 'repo_name' or 'namespace/repo_name': '/","github-eliviate",2,"","",0,"2025-06-17","2025-06-17"],[1,3603,"GRPO server mode fails with pydantic error","toslali-ibm",0,"LeonEricsson","#3607",1,"2025-06-16","2025-06-18"],[1,3595,"PPOTrainer need same tokenizer for policy and reward?","caoyang-sufe",3,"","",0,"2025-06-15","2026-07-25"],[1,3592,"trl vllm-serve support more args: e.g. reasoning-parser","hacker-jerry",2,"","",0,"2025-06-15","2026-05-26"],[1,3590,"Grpo trainer for VLMs like Qwen 2.5 VL","AggarwalShourya",2,"","",0,"2025-06-15","2025-06-17"],[1,3587,"GRPO with vLLM server - get_world_size Not Found","maziyarpanahi",2,"","",0,"2025-06-14","2025-06-14"],[1,3583,"Type hint problem in formatting_func argument in SFTTrainer","MaiqiVerse",1,"MaiqiVerse","#3584",0,"2025-06-13","2025-06-15"],[1,3581,"FSDP 2 + DPO Error:  'MixedPrecisionPolicy' object has no attribute '_module_classes_to_ignore'","JVP15",2,"","",0,"2025-06-13","2026-07-14"],[1,3580,"Clarification on default data collator for SFTTrainer","jiosephlee",0,"LeonEricsson","#3606",1,"2025-06-13","2025-06-18"],[1,3579,"PPO dataset formatter : why transformers to tokenizerd input ids in official ppo scripts demo?","caoyang-sufe",2,"","",0,"2025-06-13","2026-07-09"],[1,3574,"GRPOTrainer Is Extremely Slow on Small Dataset (5MB)","zahraajamali",3,"","",0,"2025-06-12","2026-06-04"],[1,3571,"[GRPO] Entropy metric","LeonEricsson",1,"LeonEricsson","#3700",1,"2025-06-12","2025-07-25"],[1,3569,"TRL GRPOtrainer truncation problem: The decoder prompt (length 3353) is longer than the maximum model length o","aladinggit",0,"LeonEricsson","#3601",1,"2025-06-11","2025-06-20"],[1,3567,"Data collator not found during pickling with trl 0.18.1 and pytorch 2.7","vatsal-kr",0,"qgallouedec","#3568",1,"2025-06-11","2025-06-12"],[1,3566,"GRPO per_device_eval_batch_size can't be set as 1, when there is only 1 GPU","CasanovaLLL",0,"mingxuetian","#4458",0,"2025-06-11","2025-11-24"],[1,3564,"function call initialization failed, code from https://github.com/huggingface/trl/pull/2455#issue-2729366606","tianshui18",2,"","",0,"2025-06-10","2025-06-17"],[1,3562,"add generation_kwargs to `GRPOTrainer`, so people have more control when training","avishaiElmakies",0,"pramodith","#3617",1,"2025-06-10","2025-06-20"],[1,3561,"`get_world_size`is not found(404) on grpo server training","YooSungHyun",2,"","",0,"2025-06-10","2025-06-10"],[1,3560,"Add support for IterableDataset for DPOTrainer","h-tonywu",2,"","",0,"2025-06-10","2025-06-12"],[1,3557,"GRPOTrainer: Parallelism for updating named params","dhruvmullick",2,"","",0,"2025-06-10","2026-09-23"],[1,3556,"Question about the k3 KL estimator implementation","ZhangShiyue",2,"","",0,"2025-06-09","2026-06-25"],[1,3555,"Add entropy based token filtering to the GRPO Trainer/Loss Function.","pramodith",1,"pramodith","#3563",1,"2025-06-09","2025-06-25"],[1,3552,"Resume from checkpoint","arahul14",3,"","",0,"2025-06-09","2026-05-28"],[1,3549,"Packing sequences for memory efficiency in GRPO and other preference learning implementations","RameshArvind",4,"","",0,"2025-06-09",""],[1,3546,"Loading of Gemma 3 crashes with TRL","Remorax",0,"albertvillanova","#5907",1,"2025-06-06","2026-06-02"],[1,3545,"ModuleNotFoundError reported When using Ascend NPU train GRPO","SolarWindRider",3,"","",0,"2025-06-05","2026-07-13"],[1,3544,"GRPOTrainer crashs with dataloader_num_workers >= 1","avishaiElmakies",2,"","",0,"2025-06-05","2025-07-12"],[1,3540,"Implement On-Policy RL with Optimal re- ward baseline (OPO) (Microsoft Research)","Atharva-Phatak",3,"","",0,"2025-06-04","2026-08-20"],[1,3539,"GRPOTrainer - Repeat Sampler - _get_eval_sampler","SnorkelerVigi",0,"mingxuetian","#4458",0,"2025-06-04","2025-11-24"],[1,3532,"steps_per_generation accounted for twice in the GRPO trainer","JunzheJosephZhu",2,"","",0,"2025-06-03","2025-06-03"],[1,3529,"`data_utils.apply_chat_template()` does not allow roles like \"tool\" or \"tool_results\" in the last message of `","junya-takayama",2,"","",0,"2025-06-03","2026-02-16"],[1,3528,"'DeepSpeedEngine' object has no attribute 'base_model_prefix'","1579364808",3,"","",0,"2025-06-03","2026-06-24"],[1,3523,"vllm server mode vs collocate mode with Lora on Qwen3-32B","sravan500",2,"","",0,"2025-06-02","2026-07-17"],[1,3520,"GRPOTrainer generates garbage output for models requiring BOS tokens due to add_special_tokens=False","CuriousG102",2,"","",0,"2025-06-01","2026-07-13"],[2,5270,"Withdrawn","lapid92",2,"","",0,"2026-11-05","2026-11-05"],[2,5242,"Modernize Type Param syntax to PEP 695","FrancescoPezzella",4,"","",0,"2026-11-04",""],[2,5241,"OpenArm connect rewrites Damiao zeros; left gripper limit is not mirrored","ogawakonatsu-hue",4,"","",0,"2026-11-04",""],[2,5233,"Pending private security advisory awaiting maintainer review","r3vpwnx",4,"","",0,"2026-11-03",""],[2,5193,"[Windows] Unstable Concurrent Capture from Two USB 2.0 UVC Cameras (One Camera Drops to ~1 FPS)","zhu57320041-alt",4,"","",0,"2026-11-01",""],[2,5189,"`lerobot-train` profile on one H200 (ACT, SmolVLA, Diffusion): where step time and memory go","arun-elr",4,"","",0,"2026-11-01",""],[2,5187,"[DATA-04] Frame addressing: use integer frame indices for video (measurements from 38 Hub datasets and barcode","arun-elr",4,"","",0,"2026-11-01",""],[2,5185,"LiberoEnv defaults to 360 px observations but the SmolVLA LIBERO checkpoints were trained at 256 px (0/20 vs 6","jdubray",4,"","",0,"2026-10-31",""],[2,5184,"FineART: observation.state[t] equals action[t-1] exactly in every frame","ebeirne",4,"","",0,"2026-10-31",""],[2,5176,"make_policy leaves the resume checkpoint in cfg.pretrained_path when construction fails","JonnyFaeth",4,"","",0,"2026-10-31",""],[2,5173,"2x faster flow-policy inference with a one-pass Gaussian readout","Hadrien-Cornier",4,"","",0,"2026-10-31",""],[2,5172,"Implement IMPRINT (Memory Without Curation): latent episode memory for policies","Hadrien-Cornier",4,"","",0,"2026-10-30",""],[2,5164,"Streaming-WAM: 2.9x faster warm inference for Fast-WAM, no pause at chunk refresh","Hadrien-Cornier",4,"","",0,"2026-10-30",""],[2,5148,"[Performance] pi0.5: the denoise loop is 86% of a bf16 chunk; static prefix KV + CUDA-graph denoise?","arun-elr",4,"","",0,"2026-10-29",""],[2,5138,"remove_feature leaves stale per-episode stats/video columns, causing delete_episodes/split_dataset to fail in ","Teddy-Liao",4,"","",0,"2026-10-29",""],[2,5130,"Example proposal: converting rosbag2/MCAP recordings to LeRobotDataset v3.0 (topic-to-feature mapping recipe)","kittimzhe",4,"","",0,"2026-10-28",""],[2,5118,"PI0 / PI05 / SmolVLA put remaining cameras in the wrong slots when one camera is missing","chaoshengsc",4,"","",0,"2026-10-27",""],[2,5116,"FastWAM: docs checkpoint ZibinDong/fastwam_libero_uncond_2cam224 (rev a784645f) normalizes images twice -> fai","yathAg",0,"C00kieMonsta","#5121",0,"2026-10-27","2026-10-28"],[2,5105,"pi0: padded end-of-episode actions are included in the training loss","johyeongseob",4,"","",0,"2026-10-26",""],[2,5102,"[Discussion] Retrofitting legacy industrial arms with VLA — what's the practical path?","bkbk-robot",4,"","",0,"2026-10-26",""],[2,5099,"[Discussion] Has the bf16 update-erasure paper been evaluated or reproduced in LeRobot?","Muyun99",4,"","",0,"2026-10-26",""],[2,5082,"SmolVLA fine-tuning on ~100 episodes: policy output collapses toward near-constant actions (gripper span 1.4° ","bkbk-robot",4,"","",0,"2026-10-24",""],[2,5080,"Add QF3 filtered-gradient updates for a state-only flow actor","Hadrien-Cornier",4,"","",0,"2026-10-24",""],[2,5075,"concatenate_video_files fails for input paths containing apostrophes (tested fix included)","JonnyFaeth",4,"","",0,"2026-10-23",""],[2,5060,"Koch leader calibration fails with \"Negative values are not allowed\": homing offset has the wrong sign on moto","diegoakel",4,"","",0,"2026-10-22",""],[2,5048,"Faster-WAM: 2.5-5.3x faster chunks in random-weight A100 timing","Hadrien-Cornier",1,"Hadrien-Cornier","#5049",0,"2026-10-22","2026-10-23"],[2,5039,"[Dataset] lerobot/berkeley_autolab_ur5 v3.0: 28 episodes are assigned to the wrong video file, so their timest","lucky07-07",4,"","",0,"2026-10-21",""],[2,5029,"fix(datasets): bound pending AsyncImageWriter frames","TinyGeek",4,"","",0,"2026-10-21",""],[2,5018,"AsyncImageWriter accepts zero-worker configurations and hangs on queued frames","TinyGeek",1,"TinyGeek","#5019",0,"2026-10-20","2026-10-21"],[2,5011,"Kinematics steps read joints in observation dict order instead of motor_names order","ravediamond",4,"","",0,"2026-10-19",""],[2,5002,"Support a Raspberry Pi SO-101 controller without PyTorch imports","Hadrien-Cornier",4,"","",0,"2026-10-17",""],[2,4998,"modify_tasks returns incorrect task labels after editing a loaded dataset","i-luoying",4,"","",0,"2026-10-17",""],[2,4983,"[Model integration] Add Long-WAM support to LeRobot","Aaronhuang-778",2,"","",0,"2026-10-15","2026-10-29"],[2,4982,"split_dataset accepts duplicate episode indices within one split and writes invalid episode IDs","i-luoying",0,"creativityventures","#4985",0,"2026-10-15","2026-11-03"],[2,4976,"[Dataset v3] Invalid episode offsets and bounds silently return wrong frames and cross-episode actions","adi0x90",4,"","",0,"2026-10-15",""],[2,4973,"FP16 constant state/action normalization produces NaN with default epsilon","David1726",4,"","",0,"2026-10-14",""],[2,4962,"Motors: check register names statically","nicolas-rabault",4,"","",0,"2026-10-14",""],[2,4961,"Motors: pipeline the sync read of a control loop","nicolas-rabault",4,"","",0,"2026-10-14",""],[2,4960,"Motors: read Dynamixel motors with Fast Sync Read","nicolas-rabault",4,"","",0,"2026-10-14",""],[2,4944,"push_to_hub defaults every dataset to apache-2.0, including conversions of CC BY datasets","s1g6r",4,"","",0,"2026-10-12",""],[2,4931,"lerobot-eval fails for pusht with the default async envs: workers can't find gym_pusht, and attribute probes s","LinCanNerd",4,"","",0,"2026-10-12",""],[2,4928,"validate_frame rejects features whose shape is declared as a list","vicfei",4,"","",0,"2026-10-12",""],[2,4922,"Video dataset indexing hangs indefinitely when torchcodec native lib fails to load (no system ffmpeg) - the OS","vicfei",4,"","",0,"2026-10-11",""],[2,4921,"ACT: validate_features rejects the state-only input that forward() actually consumes; adding an unused environ","vicfei",4,"","",0,"2026-10-11",""],[2,4919,"lerobot/fineart_vla_base:  can not find this repo!","xianglunkai",2,"","",0,"2026-10-11","2026-10-22"],[2,4870,"TDMPC can't be served by the async PolicyServer: `predict_action_chunk` reads empty queues, and its chunk is t","easyrider11",4,"","",0,"2026-10-07",""],[2,4868,"Feature request: SO-ARM 102 native support and Robots documentation","brnikita",2,"","",0,"2026-10-07","2026-10-21"],[2,4859,"[HIL-SERL] Proposal: add SpaceMouse support for human interventions","Jas000n",4,"","",0,"2026-10-06",""],[2,4858,"reBot B601-DM: issues hit in teleop/recording and fixes (want PRs?)","ErykHalicki",4,"","",0,"2026-10-06",""],[2,4857,"LIBERO doesn't work on native Windows: 4 small blockers, with a tested fix (happy to send a PR)","ahmedsleem109",4,"","",0,"2026-10-06",""],[2,4851,"check_version_compatibility warns \"update lerobot\" when the dataset is older, and stays silent when it is newe","jih0-kim",0,"horizonbymuneeb","#4852",0,"2026-10-05","2026-10-21"],[2,4840,"YAML config with policy.path crashes on dict-valued policy fields (normalization_mapping, input_features): \"un","JoeyTan21",4,"","",0,"2026-10-03",""],[2,4834,"docs(libero_plus): \"10 tasks per suite\" / \"400 total episodes\" don't match the LIBERO-Plus task set","thanhtnguyen10",0,"aroraanushri","#4839",0,"2026-10-02","2026-10-04"],[2,4801,"[Bug] make_policy drops --rename_map when calling set_dataset_feature_metadata","zuoxingdong",1,"zuoxingdong","#4802",1,"2026-09-29","2026-09-29"],[2,4784,"Make `names` checkable: three structural lints and a warning that leaves a trace","laa1991",4,"","",0,"2026-09-27",""],[2,4749,"Reading LeRobotDataset for pre-training dataset analysis — two conventions that make it hard, and a request","aslansd",4,"","",0,"2026-09-24",""],[2,4727,"[Bug] gaussian_actor: use_tanh_squash is documented and stored but never read (tanh squashing cannot be disabl","OctopusWen",4,"","",0,"2026-09-23",""],[2,4723,"[Docs] Introduce the dataset visualizer much earlier and add screenshots","julien-c",0,"s1lent4gnt","#4793",1,"2026-09-22","2026-09-29"],[2,4711,"Silent degradation in groot and wall_x from_pretrained (siblings of #4577)","DanMo661",0,"Samuel-Hills","#4761",0,"2026-09-22","2026-10-21"],[2,4687,"TRAIN-06: standardize encoder freezing","LorenzoCiarpa",4,"","",0,"2026-09-19",""],[2,4679,"lerobot-record ignores the action returned by Robot.send_action()","dancher00",4,"","",0,"2026-09-18",""],[2,4665,"feat(datasets): add storage-format-aware writer routing and a native Lance write backend","yanghua",4,"","",0,"2026-09-18",""],[2,4663,"Recording session failed: not enough values to unpack (expected 2, got 1)","matsujirushi",0,"Hashim1999164","#4674",0,"2026-09-18","2026-10-21"],[2,4655,"migrate_policy_normalization crashes on lerobot/vqbet_pusht: config schema drift (unknown field mlp_hidden_dim","DanMo661",4,"","",0,"2026-09-16",""],[2,4650,"[RDA Audit] Data Quality Report for lerobot/svla_so101_pickplace","liesliy",2,"","",0,"2026-09-15","2026-10-21"],[2,4649,"migrate_policy_normalization emits empty config.json and does not copy model weights","DanMo661",4,"","",0,"2026-09-15",""],[2,4647,"PreTrainedConfig.from_pretrained() leaves pretrained_path=None → silently builds untrained processor pipeline ","DanMo661",1,"DanMo661","#4648",0,"2026-09-15","2026-10-21"],[2,4643,"CSOAI — LeRobot interop: robotics governance","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[2,4642,"CSOAI — LeRobot interop: robotics governance","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[2,4641,"CSOAI — LeRobot interop: robotics governance","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[2,4640,"CSOAI — LeRobot AI governance robotics","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[2,4633,"[Performance] SmolVLA in-place DynamicCache crop prevents torch.compile and limits inference to 4.8 Hz","NotNANtoN",4,"","",0,"2026-09-14",""],[2,4626,"[Bug/Feature Req/Fix?] DAgger rollout strategy cannot record full task-attempt episodes with HIL corrections","grahamstelzer",4,"","",0,"2026-09-13",""],[2,4624,"Proposal: out-of-tree fixed-shape FlowEdge deployment plugin","elprofesoriqo",3,"","",0,"2026-09-13","2026-09-19"],[2,4619,"eval scripts report success rate with no matched benign arm, so two policies cannot be compared","sattyamjjain",4,"","",0,"2026-09-12",""],[2,4617,"Policy identity: should the resolved unnormalizer and controller convention be part of what a checkpoint repor","sattyamjjain",4,"","",0,"2026-09-12",""],[2,4614,"`lerobot/smolvla_libero` ships `n_action_steps=50`, which costs about 20 pp on LIBERO","darrobcker",4,"","",0,"2026-09-11",""],[2,4612,"GPU busy about 30% of the time when training ACT on 8x B200 at the default settings","TarzanZhao",4,"","",0,"2026-09-11",""],[2,4609,"B601: preserve the MIT target when entering safe-home","YT-wk",4,"","",0,"2026-09-11",""],[2,4591,"Support third-party plugin discovery for ProcessorStep (pre/post-processors)","ravediamond",4,"","",0,"2026-09-09",""],[2,4582,"A question about VLM temperature for VLA","sdfzz",2,"","",0,"2026-09-07","2026-09-09"],[2,4577,"PI0Policy.from_pretrained returns an untrained model when loading fails, even with strict=True","shoumikhin",4,"","",0,"2026-09-05",""],[2,4573,"[Docs / EnvHub] Path to run a LeRobot policy on VSArena (browser stacking benchmark)?","arankair",4,"","",0,"2026-09-04",""],[2,4572,"Integrating OpenTrajectory dataset augmentations into LeRobot","edmos7",4,"","",0,"2026-09-04",""],[2,4550,"RealSense + OpenCV camera fails to connect together in lerobot-record (works individually)","j-clawson",4,"","",0,"2026-09-02",""],[2,4548,"smolvla_robocasa is systematically underevaluated: eval instructions are out-of-distribution, and the docs' re","cengishasan",4,"","",0,"2026-09-01",""],[2,4530,"Evaluation recording fails when using multiple sequential batches","alek5k",4,"","",0,"2026-08-28",""],[2,4529,"Make BimanualMixin lifecycle atomic and fail-safe","s1lent4gnt",4,"","",0,"2026-08-27",""],[2,4524,"StreamingLeRobotDataset queries video frames with dataset-global timestamps","rakhimovv",2,"","",0,"2026-08-27","2026-10-21"],[2,4523,"push_checkpoint_to_hub should give option to omit train state push","ReadHegel",4,"","",0,"2026-08-27",""],[2,4520,"Inefficient checkpoint loading (redundant model loads/initializations, device transfers) causes a 30x slowdown","niamorg",4,"","",0,"2026-08-26",""],[2,4519,"v3 datasets disagree on task association: episodes-index `tasks` (pusht) vs per-frame `task_index` only (liber","easyrider11",4,"","",0,"2026-08-26",""],[2,4517,"Finetuning keeps the base checkpoint's stale input_features: make_policy only refreshes them when empty (root ","easyrider11",4,"","",0,"2026-08-26",""],[2,4513,"RDA audit results for xarm_lift / xarm_push / jaco_play (LeRobot v3.0) — same platform, ~4x idle spread","liesliy",2,"","",0,"2026-08-25","2026-10-21"],[2,4509,"LeRobot is unusable on Python 3.14: every from_pretrained raises TypeError at config-parser construction","devangpratap",4,"","",0,"2026-08-24",""],[2,4507,"Optional provenance block in eval_info.json so third-party eval results are reproducible","sattyamjjain",4,"","",0,"2026-08-24",""],[2,4496,"Feature: a CLI to pin camera mount poses and put them back","OliverAltergott",4,"","",0,"2026-08-21",""],[2,4483,"RECORD mode crashes Foxglove with TypeError when pressing left arrow (stop/rerecord)","SYXiao2002",0,"timothyanderson096-ocdealcheck","#4527",0,"2026-08-20","2026-09-07"],[2,4481,"Is there interest in a per-policy robustness field in the model card template?","sattyamjjain",4,"","",0,"2026-08-19",""],[2,4474,"teleop lekiwi base without follower arm","agentcluck77",4,"","",0,"2026-08-18",""],[2,4463,"Discussion: Unexpected performance difference between SmolVLA and Pi0.5 with the same real robot dataset","skkkkk0609",4,"","",0,"2026-08-17",""],[2,4451,"Migration mangles feature names containing an underscore (observation.environment_state)","saime428",4,"","",0,"2026-08-14",""],[2,4442,"Support an explicit motor-to-URDF joint mapping in the kinematics pipeline","awmaure",4,"","",0,"2026-08-13",""],[2,4422,"Relaxing dependency ranges","ruben-arts",4,"","",0,"2026-08-12",""],[2,4420,"Extensible storage backends for LeRobotDataset","zz-jason",4,"","",0,"2026-08-12",""],[2,4417,"[Feature discussion] Where should camera metadata (intrinsics, ...) be stored ?","yguel",4,"","",0,"2026-08-11",""],[2,4415,"Normalization/unnormalization is silently skipped after Pipeline migration for multi-dataset models","ceyxasm",4,"","",0,"2026-08-11",""],[2,4414,"VLA-JEPA: hard-coded last-layer cond tap reads untrained token positions when freeze_qwen=True","OliverAltergott",0,"WOLIKIMCHENG","#4421",0,"2026-08-11","2026-08-13"],[2,4413,"High rate observation recording","lvjonok",2,"","",0,"2026-08-11","2026-09-07"],[2,4409,"[Dataset] Restore distinct push-back-v3 label in lerobot/metaworld_mt50 (50 task IDs -> 49 labels)","sawhney17",2,"","",0,"2026-08-10","2026-10-21"],[2,4406,"[DATA] VLABench default eef Box rejects all 3,114,872 unified actions","sawhney17",4,"","",0,"2026-08-10",""],[2,4404,"[DATA] RoboTwin joint action space rejects 92.01% of robotwin_unified actions","sawhney17",4,"","",0,"2026-08-10",""],[2,4402,"[DATA] DROID task_category and date duplicate unrelated metadata on every row","sawhney17",4,"","",0,"2026-08-10",""],[2,4401,"[DATA] Pinned metadata remediation for 11 official LeRobot datasets","sawhney17",4,"","",0,"2026-08-10",""],[2,4400,"Add TurboVLA as a lightweight, non-VLM VLA policy (plugin first)","ravediamond",2,"","",0,"2026-08-10","2026-09-07"],[2,4397,"docs(reachy2): pollen_data_acquisition_server repo returns 404","tc-huang",0,"ethanvillalovoz","#4418",0,"2026-08-10","2026-09-07"],[2,4393,"torch>=2.7 is allowed but bundled torchcodec 0.11 requires torch>=2.9 symbols (undefined symbol: torch_dtype_f","Yongtae723",2,"","",0,"2026-08-10","2026-08-17"],[2,4392,"delta_timestamps decodes ALL image columns per query on image-dtype datasets (~100x slowdown for action chunki","Yongtae723",2,"","",0,"2026-08-10","2026-08-12"],[2,4390,"mujoco>=3.4.0 silently breaks the initial-state premise of libero_spatial task 5 (bowl no longer seats on the ","Rick0525",4,"","",0,"2026-08-09",""],[2,4388,"lerobot[libero] installs cleanly on macOS but silently omits the simulator; fail at config time instead","sattyamjjain",4,"","",0,"2026-08-09",""],[2,4386,"Architectural Discussion: Bypassing Teleoperation Hardware Locks & Dimensionless Topology","guanenyu-prog",4,"","",0,"2026-08-08",""],[2,4383,"ZMQ is not registered in `async_inference` (`robot_client` actually)","swkfk",2,"","",0,"2026-08-08","2026-08-21"],[2,4374,"SmolVLA rebuilds the RoPE sin/cos tables on every apply_rope call","zuoxingdong",4,"","",0,"2026-08-07",""],[2,4368,"RFC: ROS 2 integration strategy (ecosystem survey and proposed direction)","alejodosr",4,"","",0,"2026-08-07",""],[2,4332,"Optional-dependency guard passes for an installed-but-broken transformers, crashing lerobot.processor / lerobo","omiya0555",4,"","",0,"2026-08-04",""],[2,4327,"eval_policy_all with max_parallel_tasks > 1 silently returns 0% success (shared stateful policy across threads","alejodosr",4,"","",0,"2026-08-04",""],[2,4325,"ActionTokenizerProcessorStep has zero test coverage","ravediamond",4,"","",0,"2026-08-04",""],[2,4321,"Show the language task in the Rerun dataset viewer","zuoxingdong",1,"zuoxingdong","#4322",1,"2026-08-04","2026-10-21"],[2,4319,"Add trackio as an experiment tracker alongside wandb","zuoxingdong",4,"","",0,"2026-08-04",""],[2,4309,"LeKiwi.send_action raises KeyError on every command when max_relative_target is set","galimba",2,"","",0,"2026-08-03","2026-08-04"],[2,4282,"LeKiwi aborts the control loop on a single transient Feetech `sync_read` failure (#4207 equivalent)","zuoxingdong",1,"zuoxingdong","#4283",1,"2026-08-01","2026-08-01"],[2,4280,"LeKiwi `send_action` raises `KeyError` whenever `max_relative_target` is set","zuoxingdong",1,"zuoxingdong","#4281",1,"2026-08-01","2026-08-04"],[2,4259,"DiffusionPolicy: no EMA support — the reference implementation enables it in every config and evaluates the EM","2047767028-lang",0,"alejodosr","266be2b",0,"2026-07-31","2026-08-07"],[2,4258,"Document a small local Dataset v3 porting path before DROID/SLURM","xiaoyaoqilan",4,"","",0,"2026-07-31",""],[2,4245,"[X-VLA] Camera count mismatch and rollout arguments parsing issues with 2-camera setups","FaboNo",4,"","",0,"2026-07-30",""],[2,4240,"Add a ChunkSafetyProcessorStep to validate predicted action chunks before execution","ravediamond",4,"","",0,"2026-07-30",""],[2,4231,"lerobot-eval --env.type=robocasa fails despite following RoboCasa documentation","raul-machine-learning",3,"","",0,"2026-07-30","2026-07-30"],[2,4229,"How to improve robot motion during model evaluation","mcres",4,"","",0,"2026-07-30",""],[2,4219,"Arbitrary code execution via malicious processor.json in from_pretrained","AAtomical",4,"","",0,"2026-07-29",""],[2,4206,"lerobot/pi05_libero_base get's 0% success","raul-machine-learning",2,"","",0,"2026-07-29","2026-10-29"],[2,4205,"[ROCm] NotImplementedError for bilinear interpolate on uint8 tensor in SmolVLA resize_with_pad during manual i","yigenfeng0707-netizen",1,"yigenfeng0707-netizen","ad8162c",0,"2026-07-29","2026-08-04"],[2,4180,"Add SDPA attention backend to SmolVLA","zuoxingdong",4,"","",0,"2026-07-28",""],[2,4178,"`compile_model=true` silently trains on NaN loss and crashes at `max-autotune` — miscompiled `make_att_2d_mask","zuoxingdong",4,"","",0,"2026-07-28",""],[2,4176,"get_safe_default_video_backend picks torchcodec when installed but unloadable (missing FFmpeg shared libraries","ahmedsohail2003",2,"","",0,"2026-07-28","2026-08-22"],[2,4169,"Recommended Training Dataset for the Pi0.5 Quickstart","raul-machine-learning",0,"zuoxingdong","#4186",1,"2026-07-27","2026-07-29"],[2,4168,"ACT+ALOHA, can only get 66% success","raul-machine-learning",4,"","",0,"2026-07-27",""],[2,4156,"Incorrect weighted aggregation of q01/q99 when merging dataset statistics","asmashine",2,"","",0,"2026-07-27","2026-10-21"],[2,4153,"VLA-JEPA default world-model loss leaks future frames into context via shared V-JEPA2 encoder pass","ravediamond",1,"ravediamond","#4381",1,"2026-07-26","2026-08-13"],[2,4152,"LIBERO evaluation: initial-state sequence depends on policy termination timing, via `LiberoEnv.step()`'s inter","Ono-Katsuki",4,"","",0,"2026-07-26",""],[2,4143,"compute_episode_data_index silently returns wrong boundaries when episode lengths are corrupted","Kunal-Somani",4,"","",0,"2026-07-24",""],[2,4141,"pi05, please add a full example that works OOTB","raul-machine-learning",4,"","",0,"2026-07-24",""],[2,4138,"[BUG] get_safe_version() crashes with TypeError instead of raising the \"tag your dataset\" error","harshaljanjani",2,"","",0,"2026-07-24","2026-07-31"],[2,4131,"Offline AutoTokenizer.from_pretrained still requires config.json; Found during pi05/lerobot 0.6 training, root","grahamstelzer",4,"","",0,"2026-07-23",""],[2,4120,"Draccus: --help dumps 3000+ lines instead of scoping to the selected env/policy","ravediamond",2,"","",0,"2026-07-23","2026-07-31"],[2,4118,"Draccus: `typing.Literal` fields are broken, even with valid values","ravediamond",2,"","",0,"2026-07-23","2026-07-31"],[2,4117,"GR00T processor stacks camera views before resize/crop, crashing on non-uniform camera resolutions","omri-berman",4,"","",0,"2026-07-23",""],[2,4115,"Rollout use_torch_compile warmup crashes on slow-compiling hardware: unprotected get_observation() vs. 500 ms ","schmidal",4,"","",0,"2026-07-23",""],[2,4114,"lerobot-rollout --use_torch_compile silently fails: torch.compile called with both mode and options","schmidal",2,"","",0,"2026-07-23","2026-07-31"],[2,4109,"clear_episode_buffer() does not delete video staging frames, so discarded takes get encoded into the re-record","nevertmr",1,"nevertmr","#4110",0,"2026-07-22","2026-07-30"],[2,4107,"How to select or remove observation features used for training","milong26",2,"","",0,"2026-07-22","2026-10-21"],[2,4105,"Draccus: bad CLI input dumps a raw ~15-frame traceback instead of a clean error","ravediamond",0,"imstevenpmwork","af4d15f",1,"2026-07-22","2026-07-31"],[2,4102,"the success rate of Evo1 is 0 after training","2561681244",4,"","",0,"2026-07-22",""],[2,4099,"dataset.fps is misleading — recording strategies actually record at fps * interpolation_multiplier, causing fr","omri-berman",2,"","",0,"2026-07-21","2026-07-21"],[2,4094,"docs: broken and placeholder links in the documentation","tc-huang",0,"youdie006","#4097",0,"2026-07-21","2026-07-31"],[2,4093,"uv sync installs CPU-only PyTorch 2.11 on Windows with an NVIDIA GPU","aHiroakiIshikawa",4,"","",0,"2026-07-21",""],[2,4091,"Action space alignment: delta vs. chunk-relative actions for pi05_libero checkpoints","KevinOrbie",4,"","",0,"2026-07-20",""],[2,4087,"Replace base64 stream observations to ZMQ multipart (−25% bandwidth)","zuoxingdong",1,"zuoxingdong","#4088",1,"2026-07-20","2026-07-30"],[2,4084,"Prefer MJPG when no fourcc is specified (uncompressed YUYV silently saturates USB)","zuoxingdong",4,"","",0,"2026-07-20",""],[2,4082,"LeKiwi default cameras silently capture uncompressed YUYV (USB bus saturation)","zuoxingdong",1,"zuoxingdong","#4083",1,"2026-07-20","2026-07-30"],[2,4080,"Spring-back gripper trigger for SO-100/SO-101 leader arms (like the Koch leader)","zuoxingdong",2,"","",0,"2026-07-20","2026-08-05"],[2,4066,"StreamingLeRobotDataset yields zero frames when video decode fails, with no error surfaced","sundargthb",0,"imstevenpmwork","fbe8f5c",1,"2026-07-18","2026-07-30"],[2,4059,"bug: test_train_utils.py fails on Windows with OSError: [WinError 1314] A required privilege is not held by th","saiteja6006",4,"","",0,"2026-07-17",""],[2,4050,"Verify evals on Papers with Code","NielsRogge",4,"","",0,"2026-07-17",""],[2,4047,"Multiple official lerobot/* checkpoints use old style processor-pipeline system breaking lerobot-eval","ravediamond",4,"","",0,"2026-07-17",""],[2,4045,"RoboCasa SmolVLA uses CamelCase task IDs for training but natural-language prompts for evaluation","j-cyoung",4,"","",0,"2026-07-17",""],[2,4038,"ImportError:Types","9514",2,"","",0,"2026-07-16","2026-08-01"],[2,4034,"RoboCasa evaluation uses fixed episode length instead of task-specific horizons","j-cyoung",0,"anasm266","#4037",0,"2026-07-15","2026-07-30"],[2,4031,"Update draccuss dependency to >= 11.6","ntjohnson1",1,"ntjohnson1","#4033",0,"2026-07-15","2026-07-30"],[2,4025,"Unexpected clip overlapping when converting v3 to v2 videos","SilentJCR",4,"","",0,"2026-07-14",""],[2,4018,"SmolVLA: partial-freeze in set_requires_grad() never matches — last VLM layer + final norm are silently left t","zuoxingdong",1,"zuoxingdong","#4019",1,"2026-07-14","2026-07-30"],[2,4017,"LeRobotDataset defaults revision to CODEBASE_VERSION tag which can be stale","riita10069",4,"","",0,"2026-07-14",""],[2,4011,"`lerobot-train` crashes after reward-model training when `push_to_hub` is enabled","HaomingSong",2,"","",0,"2026-07-13","2026-08-07"],[2,4009,"[FSDP1] OOM caused by Large-policy loading every rank's checkpoint on `cuda:0` when resume training","HaomingSong",2,"","",0,"2026-07-13","2026-07-17"],[2,4007,"Possible typo in SO-101 assembly instructions (Leader handle attachment)","asanoqm",0,"NikodemBartnik","#4008",1,"2026-07-13","2026-07-30"],[2,4006,"`lerobot_train` `--resume=true` crashes for policies that resize state/action dims (e.g. EVO1) because","wjdy",2,"","",0,"2026-07-13","2026-07-31"],[2,4003,"when will the dual-arm version of the Seeed Studio-reBot B601 RS be released?","damon-93",2,"","",0,"2026-07-13","2026-10-21"],[2,3998,"MolmoAct2 training is broken due to normalizer_processor/unnormalizer_processor","therealadityashankar",4,"","",0,"2026-07-12",""],[2,3997,"RTC inference engine crashes with TypeError on ACT (and other non-**kwargs) policies: predict_action_chunk() g","leeohwang",2,"","",0,"2026-07-12","2026-07-31"],[2,3983,"Ask for help about use_relative_actions=True","YuanwenFu",4,"","",0,"2026-07-10",""],[2,3979,"HIL-SERL: a second/replacement actor hangs forever at Ready() — no gRPC deadline, zero thread-pool headroom, n","JeremyHX",4,"","",0,"2026-07-10",""],[2,3978,"init_logging()'s custom formatter drops exc_info — logging.exception() prints no traceback","JeremyHX",0,"imstevenpmwork","185f3e1",1,"2026-07-10","2026-07-29"],[2,3976,"run  examples/tutorial/act/act_training_example.py  ffmpeg OSError: libavutil.so.56: cannot open shared object","zkailinzhang",2,"","",0,"2026-07-09","2026-07-09"],[2,3975,"[Bug] LingBot-VA LoRA fine-tuning with lerobot/lingbot_va_base fails because it is treated as a PEFT adapter","Ren-aptx",0,"Maximellerbach","#4144",1,"2026-07-09","2026-10-26"],[2,3973,"[Feature request] Can be G1 & SmolVLA adapter be added? To deploy smol vla on G1 Unitree robot","Frfor17",2,"","",0,"2026-07-08","2026-10-21"],[2,3971,"from source install error , pip install -e \".[all]\"","zkailinzhang",2,"","",0,"2026-07-08","2026-10-21"],[2,3969,"Support streaming from HF Storage Buckets (hf://buckets/...) in StreamingLeRobotDataset","sundargthb",2,"","",0,"2026-07-08","2026-07-31"],[2,3968,"install ffmpeg 7.1.1 error","zkailinzhang",2,"","",0,"2026-07-08","2026-07-08"],[2,3966,"Excessive Memory Usage with lerobot-roll with GR00T 1.7","raymondlo84-nvidia",2,"","",0,"2026-07-07","2026-07-07"],[2,3965,"[RFC] StarVLA integration with LeRobot for Qwen-series VLA policies","JinhuiYE",4,"","",0,"2026-07-07",""],[2,3952,"Question: Have you considered motor primitive-level annotations for LeRobotDataset?","liesliy",2,"","",0,"2026-07-06","2026-10-21"],[2,3950,"Slim Lerobot V3 Dataset","thunder-007",2,"","",0,"2026-07-06","2026-08-07"],[2,3942,"Teleoperation - gripper widely open et start","FaboNo",4,"","",0,"2026-07-05",""],[2,3923,"Unify world models API","Maximellerbach",4,"","",0,"2026-07-03",""],[2,3916,"Does SmolVLA version 0.4.4 seem to perform better than version 0.5.2?","captainfffsama",2,"","",0,"2026-07-03","2026-10-30"],[2,3890,"phone_to_so100 EE example not working","AnanyaP16",4,"","",0,"2026-06-29",""],[2,3883,"lerobot-edit-dataset merge: video chunk boundaries don't match actual video file lengths","anikita",4,"","",0,"2026-06-27",""],[2,3880,"Feature Request: Render all scalar dataset columns generically in lerobot-dataset-viz","anikita",4,"","",0,"2026-06-26",""],[2,3877,"Request to add our IROS 2026 paper using LeRobot","bhattner143",2,"","",0,"2026-06-25","2026-10-21"],[2,3869,"Bug: ACT and Diffusion Policy crash on RTC rollout — predict_action_chunk missing **kwargs","anikita",2,"","",0,"2026-06-23","2026-07-31"],[2,3868,"Bug: NameError in processor pipeline fallback — HfHubHTTPError not imported","anikita",3,"","",0,"2026-06-23","2026-08-05"],[2,3867,"Bug: PreTrainedConfig.from_pretrained ignores --policy.revision (breaks from-scratch resume)","anikita",0,"HaomingSong","#5197",1,"2026-06-23","2026-11-03"],[2,3863,"Relative actions assume state[:action_dim] aligns with action — support explicit/name-based state→action mappi","HariKrishnan06082k",4,"","",0,"2026-06-23",""],[2,3861,"Diffusion policy fails with async inference on a GPU","RahimMirani",4,"","",0,"2026-06-22",""],[2,3857,"`LeRobotDataset.resume()` fails on freshly-created datasets with no recorded episodes","LorenzoFerriniCodes",0,"AlexandreEDMOND","#3859",0,"2026-06-22","2026-07-28"],[2,3847,"pi0fast.mdx: wrong model ID in training and evaluation code blocks","anikita",0,"AlexandreEDMOND","#3855",0,"2026-06-21","2026-06-24"],[2,3846,"pi0_fast: --policy.action_tokenizer_name has no effect (dead CLI flag)","anikita",4,"","",0,"2026-06-21",""],[2,3845,"pi0_fast: make_policy retains stale input_features from base model, breaking custom datasets","anikita",2,"","",0,"2026-06-21","2026-10-21"],[2,3842,"SARM dense/dual training silently produces all-zero targets when episodes lack subtask-annotation columns (no ","1thanShih",0,"s1lent4gnt","#4306",1,"2026-06-21","2026-08-03"],[2,3832,"LeRobot Community Roadmap 0.7.0 Release","imstevenpmwork",4,"","",0,"2026-06-17",""],[2,3830,"libero image observations need hflip","sean1295",4,"","",0,"2026-06-16",""],[2,3818,"multi-gpu training stuck gpus","milong26",2,"","",0,"2026-06-16","2026-06-26"],[2,3817,"[Windows] Camera live preview shows \"No browser match\" for all external cameras","ftravass",2,"","",0,"2026-06-16","2026-06-16"],[2,3814,"AttributeError: 'OffScreenRenderEnv' object has no attribute 'env'","ilove22",2,"","",0,"2026-06-16","2026-07-02"],[2,3802,"[Bug] lerobot/diffusion_pusht checkpoint fails to load with lerobot 0.4.4 — pos_grid shape mismatch in Spatial","Prithwi13",2,"","",0,"2026-06-15","2026-10-21"],[2,3794,"Post-hoc EMA smoothing and velocity clipping for rollout actions","anikita",4,"","",0,"2026-06-13",""],[2,3793,"Push intermediate training checkpoints to Hub","anikita",2,"","",0,"2026-06-13","2026-08-05"],[2,3788,"aggregate_datasets can leave task_index stats stale after remapping tasks","someone114514",4,"","",0,"2026-06-13",""],[2,3786,"v2.1 -> v3 converter can emit visual features with names=None and missing image stats","someone114514",4,"","",0,"2026-06-13",""],[2,3776,"Pin transformers to a supported version range to prevent API drift","HariKrishnan06082k",2,"","",0,"2026-06-11","2026-07-28"],[2,3775,"GR00TN15Config dataclass fails on Python 3.10 (lerobot 0.4.4): \"non-default argument 'backbone_cfg' follows de","HariKrishnan06082k",2,"","",0,"2026-06-11","2026-06-22"],[2,3765,"Issue in GROOTN15Config","yousefHalsayegh",2,"","",0,"2026-06-10","2026-08-01"],[2,3760,"[feat] lerobot-dataset-quality: per-episode quality metrics and outlier flagging for recorded datasets","maeste",4,"","",0,"2026-06-10",""],[2,3758,"[feat] lerobot-check-calibration: detect leader/follower calibration drift in recorded datasets","maeste",4,"","",0,"2026-06-10",""],[2,3755,"ImportError: placo is installed but failed to import: liburdfdom_sensor.so.4.0: cannot open shared object file","milong26",0,"jiwenc-nv","#3873",0,"2026-06-10","2026-06-24"],[2,3747,"DAGGER resume: RTC engine uses stale observation after correction, causing snap-back","anikita",4,"","",0,"2026-06-09",""],[2,3744,"Safe float casting for VLA-JEPA","Maximellerbach",1,"Maximellerbach","4808d84",1,"2026-06-08","2026-07-31"],[2,3743,"Improve VLA-JEPA pre and post processing pipeline","Maximellerbach",1,"Maximellerbach","#3750",1,"2026-06-08","2026-06-29"],[2,3740,"Support LeKiwi in the standard CLIs (record, teleoperate, replay, rollout)","zuoxingdong",2,"","",0,"2026-06-08","2026-08-01"],[2,3738,"[RFC] Add LingBot-VLA ","miracle-techlink",2,"","",0,"2026-06-08","2026-06-29"],[2,3723,"SO-101 DAGGER: bumpless transfer on pause (sync leader to follower pose)","anikita",2,"","",0,"2026-06-05","2026-06-09"],[2,3722,"No way to opt out of stamp_repo_id timestamp on rollout datasets","anikita",0,"imstevenpmwork","167e22b",1,"2026-06-05","2026-07-28"],[2,3712,"LeRobot Dataset Training High Baseline Memory Consumption Analysis","D222097",4,"","",0,"2026-06-02",""],[2,3700,"[feat] Wayland-friendly interactive reset for lerobot-record","maeste",0,"imstevenpmwork","#3875",1,"2026-05-31","2026-06-25"],[2,3698,"Add HVTracker badge to README?","YugantM",2,"","",0,"2026-05-30","2026-08-02"],[2,3693,"Your project is ranked #15 on HVTracker — embed a trust badge?","YugantM",2,"","",0,"2026-05-29","2026-06-05"],[2,3691,"Linear interpolation of rotvec in `ActionInterpolator` produces non-physical orientation sweeps when adjacent ","theshyPika",4,"","",0,"2026-05-29",""],[2,3686,"Feature request: batched temporal-window loading for faster streaming datasets","kohankhaki",4,"","",0,"2026-05-27",""],[2,3684,"lerobot-record connect order leaves robot without commands during teleop init","NtagkasAlex",0,"imstevenpmwork","bbeacfe",1,"2026-05-27","2026-07-27"],[2,3682,"[BUG] stats.json computes an impossibly small std (e.g., 0.006) for images","SlenderMongoose",0,"Altman-conquer","#3697",0,"2026-05-27","2026-06-13"],[2,3680,"Runtime Error: live_count EB_ObjectWrapperReleasedValue issue on object release","zhixuzhichang",4,"","",0,"2026-05-27",""],[2,3674,"[xvla-libero] Hub processor JSONs are missing two pipeline steps; rollouts silently fail","thrmnn",4,"","",0,"2026-05-26",""],[2,3672,"StreamingLeRobotDataset: optional read modes to reduce shard-hopping (keep shuffle buffer)","kohankhaki",2,"","",0,"2026-05-25","2026-10-21"],[2,3671,"StreamingLeRobotDataset: expose tight delta backtrack bounds (avoid fixed 100/100)","kohankhaki",2,"","",0,"2026-05-25","2026-10-21"],[2,3668,"eval_policy() doesn't restore policy.train() after evaluation — silently disables Dropout / freezes BatchNorm ","ModeEric",2,"","",0,"2026-05-24","2026-07-31"],[2,3659,"Feature Request: Support for pi0.7 (Steerable Model) to improve generalization","s-tory",4,"","",0,"2026-05-23",""],[2,3654,"`save_episode` fails on features with a zero-width dimension (`shape=(0,)`)","kohankhaki",4,"","",0,"2026-05-22",""],[2,3648,"Qwen2.5-VL-based policy performance degradation caused by RoPE index calculation bugs in Transformers v5.4.0","HaomingSong",4,"","",0,"2026-05-22",""],[2,3645,"[Lerobot][Help] AssertionError: mean is infinity.","cc931127500-collab",3,"","",0,"2026-05-22","2026-08-04"],[2,3638,"pi05_libero_finetuned and pi05_libero_base get 0% success on LIBERO-10 with official lerobot-eval, while pi0_l","jaavad",4,"","",0,"2026-05-20",""],[2,3636,"SmolVLA Inference on Jetson Orin (ARM64) Produces Incorrect Results Despite Same Model Working on x86_64","cjd-hub",4,"","",0,"2026-05-20",""],[2,3633,"Available Hugging Face GR00T checkpoints failed to reproduce LIBERO benchmark results","LZY-1021",2,"","",0,"2026-05-20","2026-08-01"],[2,3628,"How to use SmolVLA to reproduce the same result with the official report?","lzy121121",2,"","",0,"2026-05-19","2026-10-21"],[2,3621,"Windows OpenCV camera configuration may fail when FOURCC is set before size and FPS","von-Neumann101",1,"von-Neumann101","#3620",0,"2026-05-18","2026-05-19"],[2,3608,"LeRobotDataset using Lance","lhoestq",2,"","",0,"2026-05-15","2026-10-21"],[2,3607,"[XVLA]: Florence-2 is failed to see target.","DI-trietlm",4,"","",0,"2026-05-15",""],[2,3599,"Support fsspec URIs (s3://, gs://) in StreamingLeRobotDataset","shorbaji",4,"","",0,"2026-05-13",""],[2,3596,"`_copy_and_reindex_episodes_metadata` is very inefficient","pzal",4,"","",0,"2026-05-13",""],[2,3592,"Issues with lerobot-find-cameras script: hardware contention, unconfigurable warmup period, unused code","AnesBenmerzoug",1,"AnesBenmerzoug","#3593",0,"2026-05-12","2026-07-29"],[2,3591,"Finetuned Pi0 seems not to use visual information","YjHH0327",4,"","",0,"2026-05-12",""],[2,3587,"We should use homing offset to reduce the severity of wrap around bugs","0o8o0-blip",4,"","",0,"2026-05-12",""],[2,3586,"Calibration overflow protection","0o8o0-blip",4,"","",0,"2026-05-12",""],[2,3585,"Lerobot allows writing goal positions outside of min / max","0o8o0-blip",4,"","",0,"2026-05-12",""],[2,3584,"cfg.validate() in lerobot/rl/learner.py is conflict with cfg = handle_resume_logic(cfg) in the same function `","jiaxinfan1123",4,"","",0,"2026-05-12",""],[2,3583,"Docs: add migration guide for lerobot v2 to v3","wadeKeith",2,"","",0,"2026-05-12","2026-08-05"],[2,3582,"Feature: support for video codec selection in dataset recording","wadeKeith",2,"","",0,"2026-05-12","2026-07-31"],[2,3581,"Docs: add GPU memory optimization guide","wadeKeith",2,"","",0,"2026-05-12","2026-08-05"],[2,3580,"Potential precision issue in video timestamp concatenation with float32","wadeKeith",4,"","",0,"2026-05-12",""],[2,3579,"Docs: add policy architecture comparison guide","wadeKeith",2,"","",0,"2026-05-12","2026-08-05"],[2,3578,"Feature: add policy benchmarking CLI for standardized evaluation","wadeKeith",4,"","",0,"2026-05-12",""],[2,3577,"Feature: support for adaptive batch size based on GPU memory","wadeKeith",4,"","",0,"2026-05-12",""],[2,3576,"Docs: add troubleshooting guide for common OpenCV camera issues","wadeKeith",4,"","",0,"2026-05-12",""],[2,3575,"Feature: add LoRA+ (LoRA Plus) support for faster convergence","wadeKeith",4,"","",0,"2026-05-12",""],[2,3569,"Race-condition in lerobot-find-cameras opencv","AnesBenmerzoug",2,"","",0,"2026-05-11","2026-08-02"],[2,3567,"Refactor \"learner.py\"  so it imports rl.algorithms.sac for SAC policy","sezan92",2,"","",0,"2026-05-11","2026-05-11"],[2,3562,"Does LeRobot support multi-GPU training with FSDP or DeepSpeed ZeRO-2?","XieWeikai",0,"Maximellerbach","#3810",1,"2026-05-11","2026-06-22"],[2,3551,"PeftConfig missing lora_alpha field — defaults to 8, dampens high-rank LoRA","anikita",0,"wadeKeith","#3573",0,"2026-05-10","2026-05-13"],[2,3549,"lerobot-rollout: missing --policy.revision flag to load a specific model commit","anikita",2,"","",0,"2026-05-10","2026-07-27"],[2,3548,"hil dagger: separate data collection from control actor thread.","xianglunkai",4,"","",0,"2026-05-10",""],[2,3547,"lerobot.motors.robstride implements Damiao protocol — silent failure with real Robstride hardware","infoslack",4,"","",0,"2026-05-10",""],[2,3544,"Fix libero / libero-plus clash","reeceomahoney",4,"","",0,"2026-05-09",""],[2,3537,"[Improvement] Promote smooth-move handover helpers to a shared utility for teleoperate / record / replay","whats2000",4,"","",0,"2026-05-08",""],[2,3535,"[X-VLA] Poor spatial precision and background sensitivity after fine-tuning with 150 episodes","triet4p",2,"","",0,"2026-05-08","2026-05-12"],[2,3523,"Column first optimization is slower than row first access","yamada-airoa",2,"","",0,"2026-05-06","2026-10-21"],[2,3522,"数据集收集时电脑故障被中断了，想删除损坏文件，继续采集数据集","le2222",4,"","",0,"2026-05-06",""],[2,3513,"`IndexError` in `decode_video_frames_torchcodec` when using real hardware timestamps (frame index out of bound","HariKrishnan06082k",2,"","",0,"2026-05-05","2026-06-22"],[2,3508,"max value of the episodes after merge dataset is incorrect","girishsmurthyIntel",0,"CarolinePascal","#4276",1,"2026-05-05","2026-08-03"],[2,3504,"Dagger teleop transition to arm pos","Maximellerbach",1,"Maximellerbach","#3506",1,"2026-05-04","2026-05-05"],[2,3489,"Ambiguous naming of pi05+libero models","sh-choi-1104",4,"","",0,"2026-04-30",""],[2,3488,"[Robstride] Add Position (Mode 1) and Current/Torque (Mode 3) control modes","infoslack",4,"","",0,"2026-04-30",""],[2,3481,"lerobot-rollout not propagating --policy.* args","Maximellerbach",2,"","",0,"2026-04-29","2026-04-29"],[2,3479,"Backward compatibility: old saved configs with RABC fields fail to load after rewards refactor","Maximellerbach",0,"s1lent4gnt","#3480",1,"2026-04-29","2026-04-29"],[2,3478,"Ban","Priyabhunia",2,"","",0,"2026-04-28","2026-04-28"],[2,3474,"Illegal instruction when running Uno Q","AIWintermuteAI",4,"","",0,"2026-04-27",""],[2,3459,"PEFT resume silently discards loaded adapter weights (double wrap_with_peft)","anikita",0,"jashshah999","#3485",0,"2026-04-26","2026-05-04"],[2,3458,"PEFT LoRA resume from Hub fails: pathlib.PosixPath rejected by PeftConfig.from_pretrained()","anikita",0,"jashshah999","#3485",0,"2026-04-26","2026-05-04"],[2,3445,"Async inference crashes with diffusion policy: stack expects a non-empty TensorList","anikita",2,"","",0,"2026-04-23","2026-07-31"],[2,3441,"Missing so100_follower_end_effector in latest release","WhyPeterZhao",2,"","",0,"2026-04-23","2026-10-21"],[2,3440,"Reward classifier tutorial path fails in v0.5.1 (make_policy passes unsupported dataset_stats; import crash wi","ernestknurov",2,"","",0,"2026-04-23","2026-07-31"],[2,3439,"pi05 cannot perform inference.","onaka-ga-pkpk",0,"WOLIKIMCHENG","#3715",0,"2026-04-23","2026-07-27"],[2,3438,"The upper bound of Numpy is not imposed by opencv-python-headless","sotanakamura",2,"","",0,"2026-04-23","2026-04-23"],[2,3436,"lerobot-train with --dataset.streaming=true crashes for SARM due to EpisodeAwareSampler on IterableDataset","bkolligs",4,"","",0,"2026-04-23",""],[2,3433,"Incorrect loss normalization when masking padded actions in SmolVLA (Same issue as ACT)","whats2000",1,"whats2000","#3434",0,"2026-04-22","2026-04-23"],[2,3429,"How to filter static frames without action in a episode?","yangmingbj",2,"","",0,"2026-04-22","2026-10-21"],[2,3425,"--rename_map silently ignored for PI05 when use_relative_actions=true (causes crash)","ihlej",0,"HaomingSong","#3711",1,"2026-04-21","2026-06-03"],[2,3422,"For image undistortion","wzhaofeng",2,"","",0,"2026-04-21","2026-10-21"],[2,3418,"[Bug] SARM training fails with AttributeError: 'BaseModelOutputWithPooling' object has no attribute 'detach' o","masato-ka",1,"masato-ka","#3419",1,"2026-04-20","2026-04-23"],[2,3415,"lerobot_dataset_viz breaks for single dimensional observation state","StephanHasler",2,"","",0,"2026-04-20","2026-07-13"],[2,3407,"WHY SARM in lerobot not use subtask in Lerobot dataset but  custom subtask annotation?","congleetea",4,"","",0,"2026-04-19",""],[2,3404,"Bug: `cleanup_interrupted_episode` leaks image PNG temp dirs after aborted recording","whats2000",1,"whats2000","#3405",0,"2026-04-18","2026-04-19"],[2,3401,"Loss and 0% Success Rate on Libero-10 with lerobot/xvla-libero Checkpoint","Hoar012",2,"","",0,"2026-04-17","2026-05-07"],[2,3400,"P=16 not driving shoulder_pan to target position reliably","0o8o0-blip",4,"","",0,"2026-04-17",""],[2,3397,"Failed building wheel for hf-egl-probe during pip install l'erobot[libero]'","Krthan",2,"","",0,"2026-04-16","2026-04-18"],[2,3394,"RuntimeError: Motor 'gripper' (model 'sts3215') was not found. Make sure it is connected.","starlitsky2010",2,"","",0,"2026-04-16","2026-10-30"],[2,3392,"Bug: visualize_dataset fails with 401 on private org dataset (Team plan)","htlim-nb",2,"","",0,"2026-04-16","2026-07-24"],[2,3385,"Smooth human intervention: avoid follower jump when entering teleoperation by enabling leader following","Miical",2,"","",0,"2026-04-15","2026-08-01"],[2,3384,"RuntimeError: Error while trying to find names to remove to save state dict, but found no suitable name to kee","robot-feng",4,"","",0,"2026-04-15",""],[2,3369,"question：tensorflow_datasets.core.registered.DatasetNotFoundError: Dataset droid not found.","cgscgs",2,"","",0,"2026-04-13","2026-10-21"],[2,3368,"LiberoEnvConfig.fps is not passed to simulator (sim is hardcoded at 20 Hz)","reeceomahoney",0,"imstevenpmwork","cfd9ff9",1,"2026-04-13","2026-07-23"],[2,3355,"Pi0 config incompatibility with latest LeRobot","C00kieMonsta",4,"","",0,"2026-04-11",""],[2,3353,"Incorrect loss normalization when masking padded actions in ACT (mean over padded horizon)","harshab1701",0,"imstevenpmwork","1add460",1,"2026-04-10","2026-04-23"],[2,3350,"XVLA import fails with transformers==5.4.0, but downgrading to transformers==4.49.0 conflicts with lerobot==0.","HYEONHEE5739",2,"","",0,"2026-04-10","2026-04-11"],[2,3347,"lerobot-find-cameras opencv mutates camera resolution on Linux when opencv-python-headless 4.13.0.92 is instal","xzh-xzh",4,"","",0,"2026-04-10",""],[2,3345,"Eye To Hand Calibration Using LeRobot","spring076-super",4,"","",0,"2026-04-10",""],[2,3343,"save_episode fails for shape=(1,) numeric features due to scalar conversion assumptions","SevenFo",1,"SevenFo","#3344",0,"2026-04-09","2026-05-19"],[2,3341,"Clarification on defining action vs next state across different teleoperation setups","SMLee0407",2,"","",0,"2026-04-09","2026-08-05"],[2,3340,"Why is lerobot so cool?","imstevenpmwork",2,"","",0,"2026-04-09","2026-04-09"],[2,3330,"Trouble running HIL on so101 and zmq camera","olofmat",2,"","",0,"2026-04-09","2026-08-01"],[2,3318,"Allow policies to be instantiated without having the dataset they were trained on.","nhnifong",4,"","",0,"2026-04-08",""],[2,3312,"Inference with ACT with relative action not working well","LaFeuilleMorte",4,"","",0,"2026-04-08",""],[2,3299,"[TxRxResult] There is no status packet","nanyi98",4,"","",0,"2026-04-07",""],[2,3298,"[Feature Request] LeRobot + Gemma + Unitree G1: Integrating Gemma 4 as a VLA model","s-tory",2,"","",0,"2026-04-06","2026-10-21"],[2,3290,"🌐 [i18n-zh] Translating docs to Chinese","tc-huang",2,"","",0,"2026-04-06","2026-10-30"],[2,3287,"Inquiry about Training Configurations (Batch Size, LR, Commit Hash) for Replicating SmolVLA on LIBERO Benchmar","Yoonkyo",2,"","",0,"2026-04-05","2026-07-25"],[2,3283,"RelativeActionsProcessorStep: should use action_names[0] when isinstance(self.action_names[0], list)=true","xianglunkai",4,"","",0,"2026-04-05",""],[2,3280,"Dataset quality diagnostic tool -- catches silent training failures","jashshah999",2,"","",0,"2026-04-03","2026-10-21"],[2,3271,"Can some frontier world models be integrated?","moshilangzi",2,"","",0,"2026-04-03","2026-10-21"],[2,3264,"smolvla can not reproduce results in libero eval using official checkpoints","zouyuanpeng",2,"","",0,"2026-04-02","2026-10-21"],[2,3259,"Question: Why does lerobot keep the entire vision_tower in float32, while the original openpi only preserves 3","XueSongTap",4,"","",0,"2026-04-01",""],[2,3252,"Cannot reproduce the performance of PI0 on LIBERO benchmark","jiuyixu25",4,"","",0,"2026-03-31",""],[2,3251,"PI0/PI05/PI0FAST: memory usage regression due to double model construction in pi_gemma.py","aDaikiKamata",0,"imstevenpmwork","#3454",1,"2026-03-31","2026-04-24"],[2,3247,"pi0.5 LIBERO eval yields 0% success rate on 0.5.0 and 0.5.1","Hesh0629",2,"","",0,"2026-03-30","2026-03-30"],[2,3246,"Trouble implementing eef in smolvla","yuzzzdi",4,"","",0,"2026-03-30",""],[2,3244,"# [Bug] `--robot.cameras` not supported for `bi_so_follower` in `lerobot-record`","wy-zong",2,"","",0,"2026-03-29","2026-03-30"],[2,3234,"Teleoperating with keyboard fails. Error: sync_write model = next(iter(models)) in motors_bus.py","prrw",4,"","",0,"2026-03-27",""],[2,3232,"TypeError bubbling up from GR00TN15Config when importing it and other policies","EBridgeford",0,"HaomingSong","#3606",1,"2026-03-27","2026-03-27"],[2,3228,"Can NOT reproduce the result of Gr00t-n1.5 on Libero","lgautel",2,"","",0,"2026-03-27","2026-08-01"],[2,3227,"gym_manipulator: Teleop intervention flag gets overwritten to False in step_env_and_process_transition","Miical",0,"remete618","#3273",0,"2026-03-27","2026-04-14"],[2,3226,"Confusion about OpenPI and LeRobot pi0.5 LIBERO settings","wangerforcs",2,"","",0,"2026-03-27","2026-08-02"],[2,3224,"LeRobotDataset is not revision-safe on shared storage","AdilZouitine",2,"","",0,"2026-03-26","2026-03-27"],[2,3221,"pusht task success rate is low after trained 300k in original arguments","alpttex19",4,"","",0,"2026-03-26",""],[2,3219,"API Examples in Lerobot Imitation Learning Documentation, does not Initiate rurun.io","Mamin130",0,"NikodemBartnik","#3623",1,"2026-03-25","2026-05-21"],[2,3218,"ERROR white screen in Rerun for \"lerobot-record\"","JFI12",0,"NikodemBartnik","#3623",1,"2026-03-25","2026-05-21"],[2,3216,"Pi0-Fast training issues on RoboTwin with LeRobot","VarunGumma",4,"","",0,"2026-03-25",""],[2,3215,"What happened when ik and fk?","milong26",2,"","",0,"2026-03-25","2026-03-25"],[2,3211,"Missing motor IDs","yzhyagao",2,"","",0,"2026-03-24","2026-07-31"],[2,3207,"Engineer looking to incorporate pipetting, but need PCB schematic","JFI12",2,"","",0,"2026-03-23","2026-10-21"],[2,3204,"Expose configurable `atol` for observation similarity in async inference server","LorenzoFerriniCodes",2,"","",0,"2026-03-23","2026-10-07"],[2,3199,"Local dataset viewer not compatible with libtorchcodec","simbotbro",2,"","",0,"2026-03-22","2026-03-27"],[2,3198,"`OpenCVCameraConfig` settings fail under `backend=ANY` on Linux; the same settings work with `backend=V4L2`","thajpo",4,"","",0,"2026-03-21",""],[2,3195,"Datasets don't support slicing 😢","fracapuano",2,"","",0,"2026-03-21","2026-07-31"],[2,3193,"so101 wrist_roll calibration broken in new version: leader/follower zero points don't correspond","ZhongChongWang",2,"","",0,"2026-03-21","2026-04-14"],[2,3191,"[Request] Compute HW Guide/Benchmark","stbdang",2,"","",0,"2026-03-20","2026-05-07"],[2,3187,"Proposal: vla-eval integration — 11+ sim benchmarks for lerobot policies","MilkClouds",4,"","",0,"2026-03-20",""],[2,3181,"rename_map not passed to make_policy in lerobot-record","Clllauk",0,"wadeKeith","a9821af",0,"2026-03-18","2026-04-17"],[2,3178,"OpenArm hardware - motors aren't visible in the lerobot-setup-can","silenzio777",4,"","",0,"2026-03-17",""],[2,3177,"Cumulative floating-point drift in from_timestamp causes FrameTimestampError on concatenated videos","weedmo",3,"","",0,"2026-03-17","2026-07-31"],[2,3176,"Lekiwi wrist_roll motor not work properly due to use_degrees option after v0.4.4","xerathyang",0,"imstevenpmwork","#3531",1,"2026-03-17","2026-05-07"],[2,3172,"Bring Your Own Logger: external logging backend support (MLflow example plugin)","EdoardoGruppi",4,"","",0,"2026-03-16",""],[2,3168,"New robot contribution application","ffllyy467",2,"","",0,"2026-03-16","2026-10-21"],[2,3157,"UMI Gripper Support for Data Collection and Training","rabhishek100",4,"","",0,"2026-03-13",""],[2,3154,"inferencing smol model","hemanthmandava945",2,"","",0,"2026-03-13","2026-10-21"],[2,3152,"feat(datasets): registering torchvision transforms","fracapuano",1,"fracapuano","#3153",0,"2026-03-13","2026-04-07"],[2,3151,"Feature Request: Support for Reinforcement Learning with Pi0.5","NewDingW",2,"","",0,"2026-03-13","2026-10-21"],[2,3150,"[error] train smolvla on my own dataset encounters errors","zouyuanpeng",2,"","",0,"2026-03-13","2026-10-21"],[2,3146,"Feature: ONNX / TensorRT export for trained policies","jashshah999",4,"","",0,"2026-03-12",""],[2,3144,"Any plan of integrating GR00T N1.6 to Lerobot?","arxrean",2,"","",0,"2026-03-12","2026-08-01"],[2,3143,"Reward Models: call for contributions","s1lent4gnt",4,"","",0,"2026-03-12",""],[2,3139,"How to convert MCAP(ros2) to LerobotDataset v3?","zhoujun24",2,"","",0,"2026-03-12","2026-10-21"],[2,3138,"Bug: Dataset metadata codebase_version incorrectly writes v3.0 for lerobot 0.4.4","eastmoon233-prog",2,"","",0,"2026-03-12","2026-03-13"],[2,3134,"Release 0.6.0","imstevenpmwork",2,"","",0,"2026-03-11","2026-07-06"],[2,3131,"SO-101 follower disconnects during teleoperation","Verlyba",0,"imstevenpmwork","7d615ac",1,"2026-03-11","2026-07-29"],[2,3129,"Cross-Episode Experience Memory for Robot Learning","robotmem",2,"","",0,"2026-03-11","2026-10-21"],[2,3127,"[Bug] VQBeT DDP training crashes with \"No backend type associated with device type cpu\" when discretization st","Altman-conquer",1,"Altman-conquer","#3128",0,"2026-03-11","2026-03-18"],[2,3122,"Groot freezing and Pi0.5 accuracy issues (v0.5.0)","SlenderMongoose",0,"mtiezzi","#3182",0,"2026-03-10","2026-04-14"],[2,3111,"lerobot[libero] installation bug","janchk",2,"","",0,"2026-03-08","2026-03-12"],[2,3109,"PI0 WARNING: Vision embedding key might need handling && WandB logging of key \"loss_per_dim\" was ignored","Oooer8",4,"","",0,"2026-03-08",""],[2,3108,"USING PI0 WARNING: skipping cudagraphs due to skipping cudagraphs due to cpu device (cat).","Oooer8",0,"HaomingSong","#3487",1,"2026-03-08","2026-04-30"],[2,3104,"MolmoSpaces Benchmark","nkalyanv99",4,"","",0,"2026-03-08",""],[2,3098,"libero/smolvla: add low-VRAM 2-process evaluation path for <=8GB GPUs","proxi666",4,"","",0,"2026-03-06",""],[2,3083,"Flash Attention move to cute DSL Flash Attention","johnnynunez",4,"","",0,"2026-03-04",""],[2,3080,"Local LeRobotDataset works fine but StreamingLeRobotDataset fails","NokiNyx",2,"","",0,"2026-03-04","2026-10-21"],[2,3079,"Crowdsourced iPhone manipulation demonstrations (LeRobot/RLDS export). Looking for best-practice guidance + su","ChristianNyamekye",2,"","",0,"2026-03-03","2026-10-21"],[2,3078,"lerobot.async_inference.robot_client not recognizing robot.type","Kev-Park",0,"aShunSasaki","#3081",0,"2026-03-03","2026-03-06"],[2,3076,"RL Stack Refactoring: Call for Contributions","s1lent4gnt",4,"","",0,"2026-03-03",""],[2,3072,"Epoch/sample count inflated by num_processes when using effective batch size in MetricsTracker","moured",2,"","",0,"2026-03-03","2026-03-18"],[2,3070,"Read, Store and Training Adaptation of Depth Information for Stereo Camera","CCChen5",2,"","",0,"2026-03-03","2026-06-29"],[2,3059,"The actor_cli function is the core entry point of the Actor process, but the current code lacks log fallback w","XXXJCSAMA",0,"jashshah999","#3063",0,"2026-03-01","2026-04-13"],[2,3058,"🌐 [i18n-KO] Translating docs to Korean","1wos",2,"","",0,"2026-03-01","2026-10-30"],[2,3057,"Some teleoperators that control an end effector don't work","ynaka81",4,"","",0,"2026-03-01",""],[2,3054,"lerobot-dataset-viz use more memory compare to rerun datasest path","leesenchiu",2,"","",0,"2026-02-28","2026-10-21"],[2,3053,"GR00T RTC inference","BoyuanS",2,"","",0,"2026-02-28","2026-08-05"],[2,3052,"paligemma ValueError(msg) fron None","unisung",0,"imstevenpmwork","#2964",1,"2026-02-28","2026-03-05"],[2,3047,"Security: Unsafe pickle deserialization in async inference enables Remote Code Execution (CWE-502)","Chocapikk",2,"","",0,"2026-02-27","2026-10-07"],[2,3039,"ValueError: Column name ['action'] not in the dataset.","anu120",2,"","",0,"2026-02-26","2026-03-06"],[2,3002,"Unable to access Hugging Face Visualize Dataset website from Web Browser","euclidean33",2,"","",0,"2026-02-23","2026-02-24"],[2,3000,"What's the reason of having tensor -> numpy -> tensor conversion in GR00T processor?","arxrean",0,"Priyakarna13","#5112",0,"2026-02-22","2026-10-30"],[2,2999,"How to add extra episode meta data to the dataset","yuecideng",4,"","",0,"2026-02-21",""],[2,2994,"Reinforcement Learning in LiberoEnv","razarcon3",2,"","",0,"2026-02-20","2026-10-21"],[2,2984,"[docs] Fix incorrect training command in SmolVLA documentation","masato-ka",2,"","",0,"2026-02-18","2026-02-18"],[2,2980,"Async inference behaves differently between migrated older models and newly trained models","aShunSasaki",2,"","",0,"2026-02-17","2026-10-07"],[2,2975,"lerobot_train script fails training with lerobot/libero dataset","ozgraslan",2,"","",0,"2026-02-16","2026-10-21"],[2,2965,"如何购买HopeJR快速启动套件","dfddfd0303",2,"","",0,"2026-02-13","2026-02-13"],[2,2963,"pi05_base pretrained model has action/state dimension mismatch with lerobot/libero dataset, causing official Q","krys613",4,"","",0,"2026-02-12",""],[2,2958,"Metaworld config missing","reeceomahoney",0,"jashshah999","#3017",0,"2026-02-11","2026-02-25"],[2,2957,"policy.path option broken when loading config from yaml","reeceomahoney",0,"jashshah999","#3145",0,"2026-02-11","2026-05-13"],[2,2956,"X-VLA LIBERO: Eval pipeline crashes with `bfloat16` models due to numpy incompatibility","gianlucageraci",4,"","",0,"2026-02-11",""],[2,2955,"X-VLA LIBERO: Duplicated Steps in Environment and Policy PreProcessor during Eval","gianlucageraci",2,"","",0,"2026-02-11","2026-10-21"],[2,2954,"X-VLA LIBERO: Inconsistent config.json","gianlucageraci",2,"","",0,"2026-02-11","2026-10-21"],[2,2952,"HIL-SERL SO101 is not implemented","mason-m-lowes",4,"","",0,"2026-02-10",""],[2,2944,"Enable ImageNet pretrained weights by default for Diffusion Policy to improve performance on small datasets","fxdllr",2,"","",0,"2026-02-10","2026-07-31"],[2,2942,"XVLA Model Causes Stuttering Motion (Discontinuous Action Output) ","guanchzh",4,"","",0,"2026-02-10",""],[2,2938,"Vibration occurs in the robotic arm.","jh79783",2,"","",0,"2026-02-09","2026-02-25"],[2,2935,"SmolVLA Gazebo Simulation Issue","yuzzzdi",2,"","",0,"2026-02-09","2026-03-17"],[2,2934,"Too many dataloader workers: 4 (max is dataset.num_shards=3). Stopping 1 dataloader workers.","D222097",2,"","",0,"2026-02-09","2026-10-21"],[2,2924,"Problem with my Wrist_roll motors","rhkrdnwls",2,"","",0,"2026-02-08","2026-04-13"],[2,2919,"flow matching policy","KingRedMan",2,"","",0,"2026-02-07","2026-10-21"],[2,2918,"[Bug] Diffusion Policy: Missing image resize logic for high-res datasets (e.g., aloha_sim)","fxdllr",2,"","",0,"2026-02-07","2026-07-31"],[2,2916,"SARM complains about incorrect transformer version","moured",0,"imstevenpmwork","#2964",1,"2026-02-06","2026-03-05"],[2,2915,"Difficulties with finetuning SmolVLA.","brickerino",2,"","",0,"2026-02-06","2026-10-21"],[2,2907,"How to replay EEF-pose lerobot dataset with simulated Environment?","LaFeuilleMorte",4,"","",0,"2026-02-05",""],[2,2906,"INVALID Libero spatial environment (task 5)","comibear",2,"","",0,"2026-02-05","2026-02-09"],[2,2902,"Support for LIBERO-Plus and LIBERO-PRO","RiccardoIzzo",4,"","",0,"2026-02-04",""],[2,2898,"Rerun can not display datasets in the distant mode","sotanakamura",1,"sotanakamura","#2881",0,"2026-02-03","2026-02-18"],[2,2896,"Core dumped (Segmentation Fault) while training Gr00t","hoangminhtoan",4,"","",0,"2026-02-03",""],[2,2895,"Performance regression in _query_hf_dataset when using delta_timestamps (~40x slower)","jianzhou0420",0,"Rick0525","#4314",0,"2026-02-02","2026-09-01"],[2,2894,"[Feature Request] Diffusion Policy should leverage language instructions when training on LIBERO benchmark","Morpheus-An",2,"","",0,"2026-02-01","2026-10-21"],[2,2890,"Unable to run Pi0 Fast Training","shivakanthsujit",2,"","",0,"2026-01-31","2026-10-21"],[2,2889,"augment_dataset_quantile_stats requires high memory and takes a long time for large datasets","brycegoh",0,"pepa99","#3749",0,"2026-01-31","2026-07-29"],[2,2887,"[Question & Feature Request] Support for SIMPLER-env to evaluate XVLA models","Hesh0629",4,"","",0,"2026-01-31",""],[2,2883,"Bug:` lerobot-setup-can` fails to detect python-can package","FumitakaIwaki",0,"imstevenpmwork","#2937",1,"2026-01-30","2026-02-09"],[2,2870,"Bug in SARM processor causes dense targets to become zero","johnnewsom",2,"","",0,"2026-01-29","2026-02-25"],[2,2868,"Clarification on Open X-Embodiment dataset coverage and RLDS alignment","zhcm",2,"","",0,"2026-01-28","2026-10-21"],[2,2867,"why no video file dir create?","crane-1",2,"","",0,"2026-01-28","2026-10-21"],[2,2866,"How to run multiple eval with async inference without reloading the policy each time?","paulpacaud",2,"","",0,"2026-01-28","2026-10-07"],[2,2865,"SARM not working","sandhya-cb",4,"","",0,"2026-01-28",""],[2,2863,"Question about PI model implementation","SkieyFly",4,"","",0,"2026-01-27",""],[2,2858,"Is there a script directly converts lerobot dataset from v2.0 -> v3.0?","Jiaqi-Wangx",2,"","",0,"2026-01-27","2026-01-27"],[2,2854,"lerobot-dataset-viz fails with IndexError for episodes > 0","vvrs",2,"","",0,"2026-01-25","2026-01-25"],[2,2853,"Question about loss validation and VLA evaluation in LeRobot","kagyvro48",4,"","",0,"2026-01-24",""],[2,2852,"compiling pi05 with torch._inductor  fails during training","continumarch",2,"","",0,"2026-01-24","2026-07-31"],[2,2851,"Robot doesn't move during inference when using validation split during training","kagyvro48",4,"","",0,"2026-01-24",""],[2,2850,"Bug: `--eval.batch_size` unintentionally limits LIBERO environment diversity in `lerobot-eval`","AoqunJin",1,"AoqunJin","#2899",0,"2026-01-23","2026-02-10"],[2,2848,"Circular Import v0.4.3","CursedRock17",2,"","",0,"2026-01-23","2026-02-02"],[2,2846,"XVLA folding checkpoint camera setup","brycegoh",2,"","",0,"2026-01-23","2026-02-27"],[2,2844,"Smolvla Simulation Gazebo Issue","yuzzzdi",2,"","",0,"2026-01-23","2026-02-09"],[2,2843,"SmolVLA (SmolVLM-2) Simulation Control: SO-100 Arm Bending and Oscillating in Gazebo","Vignesh-17-10",4,"","",0,"2026-01-22",""],[2,2839,"Feature Request: Extend Flow Matching to ACT","tommyly201",2,"","",0,"2026-01-22","2026-01-22"],[2,2837,"ACT Decoder uses learnable positional embeddings instead of fixed sinusoidal embeddings as described in paper","varungiridhar",2,"","",0,"2026-01-22","2026-10-21"],[2,2836,"lerobot-record throws exception when using batch video encoding","sfetzel",3,"","",0,"2026-01-21","2026-01-21"],[2,2834,"mujoco 安装失败，原因是环境变量 MUJOCO_PLUGIN_PATH 没有设置","Jrwbit",2,"","",0,"2026-01-21","2026-10-21"],[2,2831,"Feature Request: Add wandb Registry/Collections support for dataset recording","daikw",2,"","",0,"2026-01-21","2026-01-21"],[2,2829,"Error building my own policy","siddarth09",3,"","",0,"2026-01-20","2026-08-04"],[2,2825,"Default torch vision not compatible with Blackwell GPU","RealManShao",2,"","",0,"2026-01-20","2026-02-24"],[2,2823,"ffmpeg failed","trusf",2,"","",0,"2026-01-20","2026-10-21"],[2,2819,"SO-101 follower's servo motor trouble","andrewJYjang",2,"","",0,"2026-01-19","2026-01-30"],[2,2816,"wallx raise ImportError","TommyZihao",2,"","",0,"2026-01-19","2026-08-01"],[2,2814,"lerobotdataset-V3.0： time sync problem?","xianglunkai",4,"","",0,"2026-01-17",""],[2,2805,"Issue: Unable to convert v1 dataset to v2 for SmolVLA training (LeRobot) – robot config error","neso613",2,"","",0,"2026-01-14","2026-01-16"],[2,2803,"About the robot pose","mooncake199809",2,"","",0,"2026-01-14","2026-08-01"],[2,2798,"backward compatibility","pr0tos",2,"","",0,"2026-01-14","2026-10-21"],[2,2797,"bug: need to bump `wandb` to higher than `0.21.4`","nekomeowww",2,"","",0,"2026-01-14","2026-01-14"],[2,2796,"diffusion policy: a very strange train loss","flyingTeng",4,"","",0,"2026-01-14",""],[2,2794,"How to load future frames which have the same length with action chunk?","ygtxr1997",2,"","",0,"2026-01-14","2026-10-21"],[2,2793,"Correct way to teleoperate just one Lekiwi via keyboard","JDRanpariya",4,"","",0,"2026-01-14",""],[2,2790,"smolvla issue","LayLow123456",2,"","",0,"2026-01-13","2026-10-21"],[2,2789,"smolvla train","LayLow123456",2,"","",0,"2026-01-13","2026-10-21"],[2,2787,"XVLA: cameras setup","eliasab16",2,"","",0,"2026-01-13","2026-10-21"],[2,2779,"worse training performance after recent code updates","D222097",2,"","",0,"2026-01-12","2026-10-21"],[2,2775,"Deleting frames / timestamps from an episode","shamilmamedov",2,"","",0,"2026-01-09","2026-10-21"],[2,2774,"Does a stable version, together with its dedicated tutorials, exist?","zifeiyu-tan",2,"","",0,"2026-01-09","2026-10-21"],[2,2762,"The wandb version should update to newer one.","leledeyuan00",0,"nekomeowww","#2800",0,"2026-01-07","2026-01-15"],[2,2761,"Diffusion Policy: \"Blind\" trajectory following at 5k steps, stationary jittering at 30k steps (Overfitting or ","flyingTeng",2,"","",0,"2026-01-07","2026-10-21"],[2,2760,"compute_norm_stats return results that are not as expected","ychangds-ctrl",2,"","",0,"2026-01-07","2026-10-21"],[2,2754,"lerobot-find-cameras fails to work with multiple Innomaker cameras","galexy",2,"","",0,"2026-01-06","2026-07-31"],[2,2753,"Debugging poor eval with SmoVLA and two cameras.","vettorazi",4,"","",0,"2026-01-05",""],[2,2751,"end-effector pose train VLA","lizhehan0905",2,"","",0,"2026-01-05","2026-08-05"],[2,2748,"Access Denied when downloading LeIsaac assets","RiccardoIzzo",2,"","",0,"2026-01-04","2026-10-21"],[2,2746,"Request","Lr-2002",0,"imstevenpmwork","#2750",1,"2026-01-04","2026-01-04"],[2,2745,"【Security】No response of my security report","Chenpinji",4,"","",0,"2026-01-04",""],[2,2743,"Are you open to separating gym_manipulator.py from packages that require PyTorch a bit?","btelles",2,"","",0,"2026-01-02","2026-01-03"],[2,2742,"Encountered an error during execution of save_annotations_to_dataset (file: src/lerobot/data_processing/sarm_a","zhoushaoxiang",4,"","",0,"2026-01-02",""],[2,2741,"XVLA: Clarification on provided lerobot/xvla-base model checkpoint and documentation","gianlucageraci",4,"","",0,"2026-01-02",""],[2,2737,"SARM WITH PI05: Why trainning loss getting more noise?","xianglunkai",2,"","",0,"2025-12-31","2026-01-02"],[2,2736,"Questions about VLA multi-task training.","yquanli",2,"","",0,"2025-12-31","2026-10-21"],[2,2735,"Buy the camera?","JFI12",2,"","",0,"2025-12-30","2026-03-11"],[2,2733,"Pi0.5 paligemma config missing ?","Lr-2002",2,"","",0,"2025-12-30","2025-12-30"],[2,2731,"Evaluation of Policies on Libero","LeeZheXian",2,"","",0,"2025-12-29","2026-02-27"],[2,2721,"The virtual machine is unable to recognize the keyboard.","ht202",2,"","",0,"2025-12-26","2026-03-11"],[2,2720,"Incorrect file path for train.py in ACT Training Notebook","pparth230",2,"","",0,"2025-12-26","2025-12-31"],[2,2719,"Pictures in README show 404 not found in released tags","baihe-liu",0,"imstevenpmwork","#2724",1,"2025-12-26","2025-12-28"],[2,2715,"`aggregate_datasets` loses Image feature schema for image datasets","fracapuano",1,"fracapuano","#2717",0,"2025-12-24","2026-01-19"],[2,2713,"Does LeRobot support HiWonder LanderPi?","quanguyen",2,"","",0,"2025-12-24","2026-03-11"],[2,2712,"Entire dataset gets deleted when wrong index specified using lerobot delete episode","Armstrong-Couch",2,"","",0,"2025-12-24","2025-12-30"],[2,2707,"Transformers dependency","RomDeffayet",2,"","",0,"2025-12-23","2025-12-23"],[2,2704,"Training XVLA: IndexError with auto mode; size mismatch with joint mode on 14D joint-action dataset","DaKhanh",3,"","",0,"2025-12-23","2025-12-23"],[2,2702,"predict_action passes tensor to postprocessor instead of EnvTransition","dangr",2,"","",0,"2025-12-23","2026-08-01"],[2,2701,"Image keys with underscores not supported when migrating to v0.4.x","dangr",2,"","",0,"2025-12-23","2026-04-10"],[2,2700,"Training an Smolvla model on the lerobot/aloha_sim_insertion_human dataset does not converge","sslndora0612-max",2,"","",0,"2025-12-23","2026-10-21"],[2,2697,"Run pi0.5 on Libero, incorrect version of transformers","yqi19",0,"imstevenpmwork","#2964",1,"2025-12-22","2026-03-05"],[2,2696,"RTC does not work.","xiaozhisky1",2,"","",0,"2025-12-22","2025-12-22"],[2,2694,"The GT00T algorithm simply won't run and throws the following error. Could someone please help me fix it?","wuxiaolianggit",2,"","",0,"2025-12-21","2026-08-01"],[2,2693,"Wrist Roll motor not responding","CHIRANJEET1729DAS",4,"","",0,"2025-12-21",""],[2,2692,"[Bug] Too many errors when Train RL in Simulation","Hukongtao",2,"","",0,"2025-12-21","2026-10-29"],[2,2690,"[Bug] Pi0 Inference RuntimeError: Dimension mismatch in Gemma eager_attention_forward (Causal Mask vs Attn Wei","SMWTDDY",2,"","",0,"2025-12-20","2025-12-22"],[2,2689,"problem regarding to update aloha sim dataset version v2.1 to v3.0","conscious-choi",4,"","",0,"2025-12-20",""],[2,2684,"How to manually push a dataset","mcres",2,"","",0,"2025-12-19","2026-01-07"],[2,2680,"Invalid frame index when training on merged datasets [RuntimeError]","RiccardoIzzo",2,"","",0,"2025-12-18","2026-07-31"],[2,2679,"Merging datasets removes fps from scalar features","reeceomahoney",4,"","",0,"2025-12-18",""],[2,2678,"Bug: lerobot-dataset-viz IndexError when visualizing specific episodes","apeSh1t",2,"","",0,"2025-12-18","2026-02-27"],[2,2673,"Dataset v2 not working anymore","imstevenpmwork",2,"","",0,"2025-12-17","2025-12-17"],[2,2670,"Async inference for simulation (libero benchmark)","dywsjtu",2,"","",0,"2025-12-17","2026-10-07"],[2,2668,"test not work","imstevenpmwork",2,"","",0,"2025-12-17","2025-12-17"],[2,2661,"[Feature Request]: Training-Time RTC Support","zhoushaoxiang",2,"","",0,"2025-12-17","2026-02-27"],[2,2658,"Deserealization error when policy server runs cuda, and client runs cpu","Grigorij-Dudnik",2,"","",0,"2025-12-16","2026-02-27"],[2,2656,"cuda support on aarch64","dalishi",2,"","",0,"2025-12-16","2026-02-27"],[2,2655,"ValueError: Sequence length 1254 exceeds max_len_seq=512 X-VLA","caizonglin",2,"","",0,"2025-12-16","2026-10-21"],[2,2654,"Help with async inference","redmindtech",2,"","",0,"2025-12-16","2026-10-07"],[2,2652,"file version error","mjkim001130",2,"","",0,"2025-12-15","2026-02-27"],[2,2650,"During image preprocessing, normalization to the range [-1, 1] was not successfully achieved.","jru001",2,"","",0,"2025-12-15","2026-10-21"],[2,2649,"current_ts in StreamingLeRobotDataset","D222097",2,"","",0,"2025-12-15","2026-10-21"],[2,2643,"convert_dataset_v21_to_v30 issue","JuShuai-gif",2,"","",0,"2025-12-13","2026-10-21"],[2,2642,"Support of torch 2.9.0","mli0603",2,"","",0,"2025-12-12","2026-02-24"],[2,2641,"Incorrect transformer version error while training Pi0.","sowmiah08",0,"imstevenpmwork","#2964",1,"2025-12-12","2026-03-05"],[2,2640,"Documentations : Add uv to setup instructions","Josephrp",0,"imstevenpmwork","#3105",1,"2025-12-12","2026-03-08"],[2,2632,"Integrating lerobot with ROS2 and fairino FR10, Dataset script format","RickyZava",2,"","",0,"2025-12-12","2026-08-04"],[2,2630,"eval on libero with joint action space model","D222097",4,"","",0,"2025-12-12",""],[2,2628,"XVLA async inference error","eliasab16",3,"","",0,"2025-12-11","2025-12-11"],[2,2627,"When training merged datasets, frame timestamp/encoding error says that the queried frame is  a lot of secs fr","SheppCrafd",2,"","",0,"2025-12-11","2025-12-12"],[2,2626,"Training merged datasets gives frame timestamp/encoding error","SheppCrafd",2,"","",0,"2025-12-11","2025-12-11"],[2,2624,"Libero Dataset Simulation Errors (smolvla v0.3.2/v0.4.2 + lerobot 0.3.4)","xiongjie1234",2,"","",0,"2025-12-11","2026-10-21"],[2,2623,"Eval Policy Mismatch (EEF Space) for Pretrained Joint Model","D222097",2,"","",0,"2025-12-11","2026-10-21"],[2,2622,"[BUG]   XVLA async inference error","eliasab16",2,"","",0,"2025-12-11","2026-10-07"],[2,2621,"robot.disconnect() not getting called in record script when exceptions are thrown.","adityabhas22",2,"","",0,"2025-12-11","2026-02-27"],[2,2619,"XVLA not supported for async inference","eliasab16",2,"","",0,"2025-12-11","2025-12-11"],[2,2617,"[Action space] Lerobot dataset MetaWorld","afengleafs",4,"","",0,"2025-12-11",""],[2,2616,"Cannot use OpenPI → PyTorch Pi0 checkpoint with examples/rtc/eval_dataset.py (missing LeRobot policy artifacts","khaledmohamed00",2,"","",0,"2025-12-10","2026-10-21"],[2,2615,"GPU Fluctuating training on large dataset training pi05 or pi0","kyounginbaik",2,"","",0,"2025-12-10","2026-10-21"],[2,2610,"Delta timestamps + episodes filter is broken","atyshka",2,"","",0,"2025-12-09","2026-02-27"],[2,2608,"error during training on rtx 5080","nikitashiwach2001",2,"","",0,"2025-12-09","2026-10-21"],[2,2605,"ACT policy lower than expected performance in aloha-gym","aposadasn",4,"","",0,"2025-12-08",""],[2,2603,"Support \"private\" push for dataset-tools","moured",4,"","",0,"2025-12-08",""],[2,2598,"Significant performance gap between the JAX and PyTorch models","JinmingM",2,"","",0,"2025-12-08","2025-12-08"],[2,2597,"No policy or teleoperator provided, skipping action generation. Should use break to replace continue or update","Xu1024",0,"imstevenpmwork","#3042",1,"2025-12-08","2026-02-27"],[2,2595,"Keyerror when using lerobot policies to train behavior challenge","hairuoliu1",2,"","",0,"2025-12-07","2026-10-21"],[2,2594,"`lerobot-find-cameras` hangs indefinitely on Linux with Intel IPU6 (MIPI) cameras","leonardogargani",2,"","",0,"2025-12-07","2026-02-27"],[2,2590,"[feature request] Traing XVLA without using pretrain weights","hoangminhtoan",2,"","",0,"2025-12-05","2025-12-10"],[2,2589,"Clarification on XVLA folding checkpoint","brycegoh",2,"","",0,"2025-12-05","2026-02-27"],[2,2585,"How can i set pi0 fine-tuning configure?","justhashkr",3,"","",0,"2025-12-05","2026-03-26"],[2,2583,"[feature request] add new repo_id name instead of overwriting the current repo","yanan1116",2,"","",0,"2025-12-04","2026-03-03"],[2,2581,"dataset: Feature","qinglin-ai",2,"","",0,"2025-12-04","2026-10-21"],[2,2580,"How can the leader arm be synchronized to follow the follower arm during inference?","zhoushaoxiang",2,"","",0,"2025-12-04","2026-08-05"],[2,2579,"[BUG]MetricsTracker multiplies `accelerate.num_processes` twice, causing `samples/epochs` to grow by n² in mul","Daniel-rmc",0,"pkooij","#3045",1,"2025-12-04","2026-02-27"],[2,2578,"Federated Learning toolkit: Scaling data and training via community collaboration","ivelin",2,"","",0,"2025-12-03","2026-10-21"],[2,2576,"[feature request] randomly shuffle the lerobot dataset","yanan1116",2,"","",0,"2025-12-03","2026-02-27"],[2,2571,"Gr00t DiT Input","andrearosasco",2,"","",0,"2025-12-03","2025-12-03"],[2,2570,"Unable to reproduce `pi_0.5_finetured`.","z520yu",4,"","",0,"2025-12-03",""],[2,2569,"Use lerobot==0.3.2 to train pusht faster than the new version.","afengleafs",2,"","",0,"2025-12-03","2026-10-21"],[2,2566,"When inference, why Pi05 connect to get tokenizer_config.json?","shs822",2,"","",0,"2025-12-02","2026-02-27"],[2,2562,"Support for RTC with async inference","xfoukas",2,"","",0,"2025-12-01","2026-10-07"],[2,2561,"Support for compression with async inference?","xfoukas",2,"","",0,"2025-12-01","2026-10-07"],[2,2559,"Why does lerobot-pi05 have strong inference jitter, while openpi-pi05 has smooth inference？","xianglunkai",4,"","",0,"2025-12-01",""],[2,2558,"Long training - learning rate decreases almost to zero after 30k steps for SmolVLA","Grigorij-Dudnik",2,"","",0,"2025-12-01","2026-02-27"],[2,2557,"Vibration","qinglin-ai",2,"","",0,"2025-12-01","2026-02-27"],[2,2556,"when running smolvla training.","275p",2,"","",0,"2025-12-01","2026-10-21"],[2,2554,"Urgent, urgent, hello, please re-upload the old version of the pi0 model weights from February 2025 to the off","zhaianbang",2,"","",0,"2025-12-01","2026-02-27"],[2,2553,"smolVLA: Resolution of input_features","kimtaeyeong1229",2,"","",0,"2025-12-01","2026-02-23"],[2,2552,"Windows: policy.path fails with HuggingFace Hub repo IDs due to backslash path conversion","DanielBryars",4,"","",0,"2025-11-30",""],[2,2549,"Column-first _query_hf_dataset() calls leading to >20x slower dataloading time during training","brysonjones",2,"","",0,"2025-11-30","2025-12-01"],[2,2547,"Dataset split sometimes fails to split video frames due to float precision","mkiefel",1,"mkiefel","#2548",0,"2025-11-30","2026-02-25"],[2,2543,"Different finetune loss given policy.type=pi0 / policy.path=lerobot/pi0_base. What is the difference?","BBBBBBob",2,"","",0,"2025-11-28","2025-12-01"],[2,2542,"Evaluation Failed","nikitashiwach2001",2,"","",0,"2025-11-28","2026-02-27"],[2,2541,"RTC: Is it a mistake in eval_with_real_robot.py","xianglunkai",2,"","",0,"2025-11-28","2025-12-01"],[2,2533,"pi0.5 on LIBERO simulation reproduce, zero success rate on pi05_libero_base","zhuxiaofan117",4,"","",0,"2025-11-27",""],[2,2532,"RTC open source?","Zyl-000",2,"","",0,"2025-11-27","2026-02-27"],[2,2531,"Async inference behaves strange when I add ee ik solver","milong26",2,"","",0,"2025-11-27","2026-10-07"],[2,2529,"Could not read the parquet files: Parquet magic bytes not found in footer. Either the file is corrupted or thi","yanan1116",2,"","",0,"2025-11-26","2025-11-28"],[2,2528,"cannot install lerobot in WSL","yanan1116",0,"DominikPalo","#2855",0,"2025-11-26","2026-02-23"],[2,2527,"cannot install lerobot correctly","yanan1116",2,"","",0,"2025-11-26","2026-02-27"],[2,2520,"Cannot visualize episodes whose indexes are not 0","sotanakamura",1,"sotanakamura","#2521",0,"2025-11-26","2025-11-30"],[2,2519,"Follower arm gripper rotates automatically after reconnection","zhoushaoxiang",4,"","",0,"2025-11-26",""],[2,2518,"Robot proprioception input in Pi0.5","BBBBBBob",2,"","",0,"2025-11-25","2025-11-28"],[2,2516,"Question about using pi05 for subtask generation","chenghegao4556",2,"","",0,"2025-11-25","2025-11-26"],[2,2515,"RTC client server example","brycegoh",2,"","",0,"2025-11-25","2025-11-25"],[2,2512,"PI05 Normalize from [0,1] to [-3, 1]","Hukongtao",2,"","",0,"2025-11-24","2026-02-04"],[2,2511,"[Question] RTC denoise update equation is not same with the paper","shlyang",2,"","",0,"2025-11-24","2025-11-25"],[2,2509,"TypeError when setting video_encoding_batch_size = 2","zhoushaoxiang",2,"","",0,"2025-11-24","2026-08-01"],[2,2508,"[Question] Can the LeRobot dataset porting be accelerated?","XieWeikai",2,"","",0,"2025-11-24","2026-10-21"],[2,2507,"When Pi0 Fast policy will be added back?","universedust",2,"","",0,"2025-11-23","2026-02-27"],[2,2506,"diffusion policy is not working when deployed on the real device","MichaelWu99-lab",2,"","",0,"2025-11-23","2026-10-21"],[2,2505,"groot-model finetuned with lerobot doesnt follow instructions","j0nt392",2,"","",0,"2025-11-22","2026-07-31"],[2,2504,"[Question] Does lerobot-train/ecal support fp16 data type?","zhangnju",2,"","",0,"2025-11-22","2026-10-21"],[2,2501,"uv broken with libero","atyshka",2,"","",0,"2025-11-21","2026-10-21"],[2,2500,"question about the gr00t policy","yanan1116",2,"","",0,"2025-11-21","2026-08-04"],[2,2495,"Dataset will be broken if an exception is thrown while recording","sotanakamura",1,"sotanakamura","#2496",0,"2025-11-21","2025-12-15"],[2,2493,"Missing the key \"serial_number\" to find RealSence using name","sotanakamura",2,"","",0,"2025-11-21","2026-07-31"],[2,2489,"An incorrect transformer version is used","qinglin-ai",0,"imstevenpmwork","#2964",1,"2025-11-20","2026-03-05"],[2,2488,"Multiprocessing leads to decoding error","haowang020110",2,"","",0,"2025-11-20","2026-07-27"],[2,2475,"Why there is difference between async inference and local inference in image resize?","milong26",2,"","",0,"2025-11-18","2026-10-07"],[2,2472,"Execution Failure in LeRobot's \"Train RL in Simulation\" Example Code","xiaoxuh",2,"","",0,"2025-11-18","2026-08-02"],[2,2471,"When I run pi0, a problem occurs: my GPU memory usage is high, but the single-GPU utilization is very low. The","ZHHhang",2,"","",0,"2025-11-18","2026-10-21"],[2,2470,"question about the share encoder in  SAC","pegasus-zhang",2,"","",0,"2025-11-18","2025-11-18"],[2,2468,"[Feature Request] add extra options to '--display_data' like 'webserver' with rerun","HumbleHumbert",0,"imstevenpmwork","#2756",1,"2025-11-18","2026-01-07"],[2,2467,"\"task\" column missing when using convert 21_to_30 script","ZiyuXiong",0,"CarolinePascal","#3077",1,"2025-11-18","2026-03-04"],[2,2465,"loss:nan grdn:nan How to solve the gradient explosion problem in PI05 training?","Lilgeneric",4,"","",0,"2025-11-18",""],[2,2464,"Questions about Pi0.5 Model Training Details and High Level Planning Implementation","Ginldaj",2,"","",0,"2025-11-18","2026-08-02"],[2,2463,"gym_manipulator/hilserl cannot record dataset in gym-hil simulation","changh-Lee",4,"","",0,"2025-11-17",""],[2,2460,"Third-party robots, cameras, and teleoperators are not loaded when installing the package in editable mode (pi","Ryu-Yang",2,"","",0,"2025-11-17","2026-02-10"],[2,2459,"Smolvla client and server","Zsir188",3,"","",0,"2025-11-17","2026-08-04"],[2,2458,"async inference fps is very low, only 0.3 fps","wjpoom",2,"","",0,"2025-11-17","2026-10-07"],[2,2457,"Groot n1.5 has a very low success rate on libero_goal.","clementiness",2,"","",0,"2025-11-17","2026-08-01"],[2,2452,"ModuleNotFoundError: No module named 'libero'","Kimho666",2,"","",0,"2025-11-16","2025-12-07"],[2,2446,"Discussion: A better way to manage dataset versions","atyshka",4,"","",0,"2025-11-14",""],[2,2445,"PyArrow Extension Dtype Error in `merge_datasets()`","yaxiongzhao-genesis",4,"","",0,"2025-11-14",""],[2,2443,"cannot resolve dependency with pyproject.toml","antwoor",2,"","",0,"2025-11-14","2025-11-19"],[2,2442,"about dataset","zhq09221107",2,"","",0,"2025-11-14","2026-10-21"],[2,2437,"mixed_precision with accelerate","Hukongtao",2,"","",0,"2025-11-13","2025-11-13"],[2,2431,"AttributeError: 'SO100FollowerConfig' object has no attribute 'urdf_path'","wxkly8888",2,"","",0,"2025-11-12","2026-08-02"],[2,2429,"\"yaak-ai/L2D-v3\" data set is not valid","sudalsol",4,"","",0,"2025-11-12",""],[2,2428,"how to eval the real world recorded dataset?","shs822",3,"","",0,"2025-11-12","2026-03-26"],[2,2427,"Error with image features LeRobot 0.4.1","JaredBaileyDuke",2,"","",0,"2025-11-11","2026-10-21"],[2,2426,"`episodes` argument of `LeRobotDataset` does not work as stated","ankile",2,"","",0,"2025-11-11","2026-02-28"],[2,2422,"Running inference on Libero with pi0","thomasdeng2027",2,"","",0,"2025-11-11","2026-08-05"],[2,2421,"Seeking assistance with tactile data acquisition","zhoushaoxiang",2,"","",0,"2025-11-11","2026-10-21"],[2,2418,"fine tune SmolVLA in libero with tensor shape mismatch","Luo-Zhongwei",2,"","",0,"2025-11-10","2026-10-21"],[2,2417,"Introduce Lance as a new basic format for lerobot","yanghua",2,"","",0,"2025-11-10","2026-06-11"],[2,2414,"draccus.utils.ParsingError","Lakesenberg",2,"","",0,"2025-11-09","2026-10-21"],[2,2413,"Question about so100_to_so100_EE: should I re-calibrate to let my real robot looks same to so101.urdf?","milong26",4,"","",0,"2025-11-09",""],[2,2412,"movement speed has increased by almost 10 times compared to normal","pparkgyuhyeon",4,"","",0,"2025-11-09",""],[2,2411,"GR00T CUDA OOM with batch_size > 1 and index out of range error with batch_size=1","mumtozee",2,"","",0,"2025-11-08","2026-08-01"],[2,2410,"Wrong metadata name of `egl_probe`","simplestory",2,"","",0,"2025-11-08","2026-07-31"],[2,2407,"Multi-GPU training with accelerate，but on one gpu","robot-feng",2,"","",0,"2025-11-07","2026-10-21"],[2,2406,"Proposal: Modernize EnvHub Environment Repository Design","MilkClouds",4,"","",0,"2025-11-07",""],[2,2404,"Recording fails with video_encoding_batch_size > 1","F-Fer",4,"","",0,"2025-11-07",""],[2,2403,"lerobot-record","qinglin-ai",2,"","",0,"2025-11-07","2026-02-27"],[2,2402,"LIBERO reset env after setting an initial state","mumtozee",2,"","",0,"2025-11-07","2026-02-27"],[2,2401,"Metadata bug in conversion file (convert_dataset_v21_to_v30.py)","RahulSChand",2,"","",0,"2025-11-07","2026-03-04"],[2,2400,"Cannot find plugins installed in editable mode","sotanakamura",0,"aalvan","#2687",0,"2025-11-07","2025-12-24"],[2,2399,"Are there plans to support LoRa fine-tuning?","Hukongtao",2,"","",0,"2025-11-07","2026-08-02"],[2,2398,"how to accelerate the iteration in dataset","yanan1116",2,"","",0,"2025-11-06","2025-11-10"],[2,2395,"Multi-task training on Pi0: inference ignores task_id","wobufua",2,"","",0,"2025-11-06","2025-11-12"],[2,2394,"An error occurred while evaluating your policy in Sim.","Kopitau",2,"","",0,"2025-11-06","2026-10-21"],[2,2393,"The GR00T `max_action_dim` mismatch problem","lllzheng",2,"","",0,"2025-11-06","2026-08-01"],[2,2389,"How to resolve the issue that GROOT cannot train properly? Below is my training configuration and error log.","wuxiaolianggit",3,"","",0,"2025-11-05","2026-08-05"],[2,2388,"how to improve the generalization of the vla model like gr00t","Temmp1e",2,"","",0,"2025-11-05","2026-10-21"],[2,2387,"anyone having problem in calibrating so101Leader ?","kotran88",2,"","",0,"2025-11-05","2026-07-31"],[2,2386,"metaworld eval render bug","robot-feng",2,"","",0,"2025-11-05","2026-10-21"],[2,2382,"Gr00t Improvements","imstevenpmwork",3,"","",0,"2025-11-04","2026-07-31"],[2,2381,"Add DexMimicGen environment and imitation learning dataset integration","pmh9960",4,"","",0,"2025-11-04",""],[2,2380,"Add RoboCasa environment and imitation learning dataset integration","pmh9960",2,"","",0,"2025-11-04","2026-08-04"],[2,2378,"Recommended dataset size & training config for pi0.5 on simple PnP task in LeRobot","Daming-W",2,"","",0,"2025-11-04","2026-08-04"],[2,2377,"instruction following of groot inference","jcl2023",2,"","",0,"2025-11-04","2026-08-05"],[2,2375,"Reproducing pi0.5 results on libero","mumtozee",0,"AoqunJin","#2899",0,"2025-11-03","2026-02-10"],[2,2374,"Problems with running the pretrained lerobot/smolvla_base","brickerino",2,"","",0,"2025-11-03","2026-10-07"],[2,2372,"PI05 extra_repr function rising AttributeError: 'GemmaRMSNorm' object has no attribute 'weight'","mumtozee",2,"","",0,"2025-11-03","2026-08-02"],[2,2371,"memory increase continuously during training Groot","caoran2025",4,"","",0,"2025-11-03",""],[2,2368,"Release 0.5.0","imstevenpmwork",2,"","",0,"2025-11-03","2026-03-11"],[2,2366,"`TypeError` when setting `--dataset.stream=true` (finetune pi05)","sigongzi",2,"","",0,"2025-11-03","2026-07-28"],[2,2364,"AssertionError: One or several query timestamps unexpectedly violate the tolerance","RiccardoIzzo",2,"","",0,"2025-11-03","2026-02-27"],[2,2363,"On jetson AGX Orin no gpu used to train policy : We recommend that you migrate to TorchCodec","rafmat42",2,"","",0,"2025-11-03","2026-10-29"],[2,2362,"CUDA 13/Pytorch 2.9 support for lerobot","nahidalam",2,"","",0,"2025-11-03","2026-02-27"],[2,2360,"diffusion transformer","Benxiaogu",2,"","",0,"2025-11-02","2026-08-02"],[2,2357,"groot training issue","boweili666",2,"","",0,"2025-11-02","2025-11-02"],[2,2356,"AsyncInference only running one action chunk","kevinjosethomas",2,"","",0,"2025-11-01","2026-10-07"],[2,2354,"Cannot reproduce SmolVLA results on LIBERO benchmark","Hesh0629",4,"","",0,"2025-11-01",""],[2,2351,"Details of adapting SmolVLA to other robotic arms with different  configurations","yquanli",2,"","",0,"2025-10-31","2025-12-14"],[2,2349,"Gym_Hil offline evaluation","freakwave99",2,"","",0,"2025-10-31","2026-10-21"],[2,2346,"HIL-SERL - Keyboard teleop is not working","jpizarrom",0,"s1lent4gnt","#3075",1,"2025-10-30","2026-04-13"],[2,2339,"pi0 - does it need compute_norm_stats?","dethresearcher",2,"","",0,"2025-10-29","2025-10-31"],[2,2338,"policy gr00t not found when do async inference with gr00t","jcl2023",2,"","",0,"2025-10-29","2025-11-14"],[2,2337,"Can I continue reinforcement learning in HIL-SERL using a pi0","pparkgyuhyeon",2,"","",0,"2025-10-29","2026-10-21"],[2,2336,"Cannot change control/episode time when trying to run RL environment","PhillipFlipper",2,"","",0,"2025-10-29","2026-10-21"],[2,2335,"How to Visualize All Episodes of a LeRobot Dataset Locally?","Vacuame",2,"","",0,"2025-10-29","2026-02-27"],[2,2334,"Gr00t: training ends in error when images are not the same size","gyger",3,"","",0,"2025-10-28","2026-07-31"],[2,2329,"smolvla base model ( the Vlm part) to other model","smartparrot",2,"","",0,"2025-10-28","2025-10-31"],[2,2328,"Can't train on merged dataset if it contains more then one video file per camera","Grigorij-Dudnik",2,"","",0,"2025-10-28","2025-12-05"],[2,2326,"Develop LeRobotDataset tools","michel-aractingi",4,"","",0,"2025-10-28",""],[2,2325,"Correction: Rename groot to gr00t","nv-sachdevkartik",3,"","",0,"2025-10-27","2026-07-31"],[2,2322,"No (supported) data files found in yaak-ai/L2D-v3","thomas-yanxin",2,"","",0,"2025-10-27","2026-10-21"],[2,2320,"GR00T policy Improvements tracker","youliangtan",3,"","",0,"2025-10-26","2026-07-31"],[2,2319,"An incorrect transformer version is used","SkyCodeFly",0,"imstevenpmwork","#2964",1,"2025-10-26","2026-03-05"],[2,2318,"Pi0.5 + Libero: rename_map raises visual mismatch error when it shouldn't","masato-ka",0,"jadechoghari","#2332",0,"2025-10-26","2025-10-31"],[2,2316,"Deleting episodes by `lerobot_edit_dataset` fails when `root` is used","irisTa56",0,"CarolinePascal","#3035",1,"2025-10-26","2026-03-03"],[2,2315,"dataloading in 0.4.0 is very slow when I specifiy delta_timesteps and episodes","Jiaqi-Wangx",2,"","",0,"2025-10-26","2025-11-20"],[2,2312,"Streaming Dataset not Working - `DataFilesNotFoundError`","sattwik-sahu",3,"","",0,"2025-10-25","2026-07-31"],[2,2311,"Question: How I can train only online without dataset?","talregev",4,"","",0,"2025-10-25",""],[2,2309,"Bug: Race Condition in `ReplayBuffer` when using `_get_async_iterator`","Weizhi-Zhao",0,"s1lent4gnt","#3372",1,"2025-10-24","2026-04-14"],[2,2307,"Pi0.5 error loading embed token weights from paligemma","codyswain",0,"imstevenpmwork","#2964",1,"2025-10-24","2026-03-05"],[2,2306,"how to use groot without flash attention","shs822",3,"","",0,"2025-10-24","2026-07-31"],[2,2305,"Error dependence about the `Transformer` library","sunshineharry",0,"imstevenpmwork","#2964",1,"2025-10-24","2026-03-05"],[2,2304,"How to load local model?","milong26",2,"","",0,"2025-10-24","2025-10-24"],[2,2303,"Question: Does the follower arm have an api for scripting movement?","Buttmunky1",3,"","",0,"2025-10-23","2026-07-31"],[2,2302,"ValueError: cannot convert float NaN to integer when using lerobot-edit-dataset on metaworld50 dataset.","RishiRoot-star",4,"","",0,"2025-10-23",""],[2,2294,"Question about the HuggingFaceVLA/smolvla_libero Model Configuration","Hesh0629",3,"","",0,"2025-10-23","2026-07-31"],[2,2291,"Errors in finetuning pi0_base","dalishi",2,"","",0,"2025-10-23","2025-10-28"],[2,2290,"Should I set PYTHONPATH when running with python -m ?","milong26",2,"","",0,"2025-10-23","2025-10-24"],[2,2287,"Cameras not being displayed","Lucisu",3,"","",0,"2025-10-22","2026-07-31"],[2,2284,"Can we have multiple action dtypes? action.btnInputs (bools), action.Pose (float64), etc?","dgm1g15",3,"","",0,"2025-10-22","2026-07-31"],[2,2283,"Potential Bug: `lerobot-edit-dataset` crashes on v3 datasets when deleting/splitting episodes (`ValueError: ca","S-sirigiri",2,"","",0,"2025-10-22","2026-03-05"],[2,2282,"[0.3.4] Severe GPU underutilization vs 0.3.3 (2x slower on same job)","S-sirigiri",0,"imstevenpmwork","#2408",1,"2025-10-21","2025-11-07"],[2,2277,"Diffusion Policy  don't support n_obs_steps=1?","Hukongtao",2,"","",0,"2025-10-21","2026-07-31"],[2,2274,"Change FPS failure with OpencvCamera","xuaner233",1,"xuaner233","#2275",0,"2025-10-21","2025-10-21"],[2,2269,"how to configure pi0_base to train with single camera dataset","dalishi",3,"","",0,"2025-10-21","2026-05-04"],[2,2268,"Using lerobot/svla_so100_stacking throws pyarrow.lib.ArrowInvalid: Could not open Parquet input source '<Buffe","nijkah",2,"","",0,"2025-10-21","2025-10-23"],[2,2264,"bug: aggregate data meta mapping misalignment","bingogome",2,"","",0,"2025-10-20","2026-04-08"],[2,2262,"smolvla using  this dataset   lerobot/svla_so101_pickplace  image_features not find error","axia75",2,"","",0,"2025-10-20","2026-08-04"],[2,2261,"AssertionError: ['human', 'rgb_array', 'depth_array'] when gymnasium==1.1.0","hypertseng",0,"imstevenpmwork","#2267",1,"2025-10-20","2025-10-21"],[2,2259,"Clarifications on fine-tuning on different envs and embodiments","RonPlusSign",2,"","",0,"2025-10-20","2026-10-21"],[2,2258,"How can i train and eval SmolVLA in libero?","haomin1010",2,"","",0,"2025-10-20","2026-10-21"],[2,2255,"Abnormal High Loss and Erratic Robot Behavior when Fine-Tuning pi05_baseCompared to pi0_base","DYwasd",3,"","",0,"2025-10-20","2026-05-04"],[2,2252,"realsense 'rs' not found error","LeoLin6",2,"","",0,"2025-10-19","2026-03-11"],[2,2250,"Unknown device cuda:0","rwst",0,"jadechoghari","#2256",0,"2025-10-19","2025-10-20"],[2,2247,"Smolvla bug","LeoLin6",3,"","",0,"2025-10-18","2026-05-03"],[2,2246,"lerobot-eval run with exception:","talregev",0,"imstevenpmwork","#2265",1,"2025-10-18","2025-10-21"],[2,2245,"release 0.4.0 and torch 2.8.0","antoinedandi",2,"","",0,"2025-10-18","2025-10-19"],[2,2244,"Policy server cuda, robot client on cpu - tensor mismatch","wvangils",2,"","",0,"2025-10-18","2026-08-02"],[2,2242,"Is it no longer possible to fine-tune the previously used π0 model?","pparkgyuhyeon",2,"","",0,"2025-10-18","2025-10-19"],[2,2239,"Models trained using openpi pi0.5 on Lerobot's pi0.5","brycegoh",3,"","",0,"2025-10-18","2026-05-01"],[2,2228,"Trossen WidowX AI model, depth cameras and tests","lromor",2,"","",0,"2025-10-17","2025-10-27"],[2,2227,"How to easily run inference with a trained model","Biz-Joe",2,"","",0,"2025-10-17","2026-10-21"],[2,2225,"[Feature] Could we have a set of smoke tests?","sarahec",3,"","",0,"2025-10-17","2026-04-30"],[2,2224,"Can i just modify the json the pretrained policy to adapt it to my own robot?","shs822",3,"","",0,"2025-10-17","2026-05-03"],[2,2221,"Question about pre-trained weights usability and performance on Hugging Face models","MichaelWu99-lab",2,"","",0,"2025-10-16","2025-10-16"],[2,2220,"Potential Issue With `frames_in_current_file` Calculation In `LeRobotDataset._save_episode_data`","Weizhi-Zhao",2,"","",0,"2025-10-16","2025-10-22"],[2,2219,"ImportError: /lib/x86_64-linux-gnu/libc.so.6: version `GLIBC_2.32' not found","stbzwyd",2,"","",0,"2025-10-16","2026-03-11"],[2,2218,"image pad value in pi0/pi05","Tgzz666",3,"","",0,"2025-10-16","2026-04-30"],[2,2217,"lerobot fails on RTX 50 series GPUs because torch==2.7.1 does not support SM120","gagalo123",2,"","",0,"2025-10-16","2026-03-11"],[2,2216,"gpu memory required to finetune pi05","jcl2023",2,"","",0,"2025-10-16","2026-10-21"],[2,2215,"it seems that the pretrained weights for the action expert head are not being loaded for SmolVLA.","kwk2696",3,"","",0,"2025-10-16","2026-07-31"],[2,2214,"Potential Scale Imbalance in smolVLA Embedding Pipeline","kkTkk012",3,"","",0,"2025-10-16","2026-04-30"],[2,2213,"Poor ACT Policy Performance After Training (Robot Not Moving or Grasping Properly)","Vishalsub",2,"","",0,"2025-10-16","2026-10-29"],[2,2211,"Can't run libero simulation","dywsjtu",3,"","",0,"2025-10-15","2026-07-31"],[2,2210,"Inference lerobot/smolvla_base with so-101 failed","alexhegit",2,"","",0,"2025-10-15","2025-10-17"],[2,2209,"std::bad_alloc / DataLoader worker exited unexpectedly during training with smolvla_base","Daniel-rmc",2,"","",0,"2025-10-15","2026-01-15"],[2,2208,"I'm getting an error when trying to run inference with the pi05 model.","pparkgyuhyeon",4,"","",0,"2025-10-15",""],[2,2207,"Training loss unmatched with test performance using SmolVLA with LIBERO-Spatial","iamzzx",3,"","",0,"2025-10-15","2026-04-27"],[2,2194,"During training with PI0, the loss is very low. Is this normal, and is the training proceeding correctly?","pparkgyuhyeon",2,"","",0,"2025-10-14","2025-10-14"],[2,2193,"FileNotFoundError: Could not find 'policy_preprocessor.json' on the HuggingFace Hub at 'lerobot/pi0fast_base'","zmf2022",2,"","",0,"2025-10-14","2025-10-14"],[2,2192,"how to test PI0's output","Addog666",3,"","",0,"2025-10-14","2026-04-30"],[2,2191,"cannot import name 'get_video_size_in_mb' from 'lerobot.datasets.utils'","Xiawenlong-bug",2,"","",0,"2025-10-14","2025-10-14"],[2,2190,"get_video_size_in_mb not found","zmf2022",2,"","",0,"2025-10-14","2025-10-14"],[2,2189,"Unable to Update Dataset Quantile Stats for QUANTILES Normalization","pparkgyuhyeon",2,"","",0,"2025-10-14","2026-10-21"],[2,2188,"feetech servos require goal_velocity and goal_acceleration to be set","dskill",4,"","",0,"2025-10-13",""],[2,2186,"how to load pi0?","Addog666",2,"","",0,"2025-10-13","2025-10-17"],[2,2185,"Has the lerobot data format been modified after June this year?","Addog666",2,"","",0,"2025-10-13","2025-10-14"],[2,2184,"How to let an episode realize it has finished the task?","genkv",2,"","",0,"2025-10-13","2026-10-21"],[2,2181,"How to chage SmolVLA action_chunk_size?","CCCY-0304",2,"","",0,"2025-10-12","2025-10-17"],[2,2179,"pi05 training fails with transformers==4.57.0 (version compatibility issue?)","pparkgyuhyeon",3,"","",0,"2025-10-12","2026-07-31"],[2,2174,"Unexpectedly slow action generation of SmolVLA","ryuji-oi-artic",3,"","",0,"2025-10-11","2026-05-01"],[2,2173,"[HIL-SERL] possible performance bottleneck in VanillaObservationProcessorStep causes low interaction rates","jpizarrom",4,"","",0,"2025-10-11",""],[2,2172,"Add support for remote GPUs (with async inference!)","MRiabov",2,"","",0,"2025-10-11","2026-10-07"],[2,2171,"Data diffusion and data format conversion","FALCONYU",3,"","",0,"2025-10-11","2026-04-29"],[2,2170,"Fix RealSense Camera Read Failure in lerobot-find-cameras","JairoEPP",0,"imstevenpmwork","#2229",1,"2025-10-11","2025-10-17"],[2,2164,"[HIL-SERL] Episode is not rerecorded in gym_manipulator when gamepad used for teleop","jpizarrom",1,"jpizarrom","#2165",0,"2025-10-10","2025-10-10"],[2,2163,"lerobot-record error: DTS out of order, non monotonically increasing","RiccardoIzzo",2,"","",0,"2025-10-10","2025-11-03"],[2,2162,"[Question] How to suppress verbose Svt[info] logs from video encoding during save_episode()?","zxytql",2,"","",0,"2025-10-10","2025-10-13"],[2,2161,"save_episode_video error","yfynb1111",4,"","",0,"2025-10-10",""],[2,2158,"Convert local dataset to LeRobot format error based on Dataset v3","Hukongtao",2,"","",0,"2025-10-10","2025-10-20"],[2,2157,"The actions of SmolVLA.","ZZQ0316",2,"","",0,"2025-10-10","2025-10-10"],[2,2156,"How to reproduce lerobot/pi0_libero_finetuned?","PuzhenYuan",4,"","",0,"2025-10-09",""],[2,2155,"Imitation Learning in Sim error","makerforrobot",2,"","",0,"2025-10-09","2026-10-21"],[2,2153,"Why can’t I find something like train_expert_only in the latest version of pi0? Do the current versions of pi0","ZHHhang",0,"pxe3","#2727",0,"2025-10-09","2025-12-31"],[2,2151,"\"model.model.paligemma_with_expert.paligemma.model.language_model.embed_tokens\" parameter is missing in pretra","HyeonseokE",2,"","",0,"2025-10-09","2025-10-14"],[2,2148,"save_episode does not support more than one episode","Hukongtao",2,"","",0,"2025-10-09","2025-10-10"],[2,2147,"GemmaRMSNorm.forward() got an unexpected keyword argument 'cond'","RayTang88",2,"","",0,"2025-10-09","2025-10-12"],[2,2146,"Set \"--policy.dtype = bfloat16\" in the training script, get an error","RayTang88",2,"","",0,"2025-10-09","2025-12-13"],[2,2142,"(improving api) Over-reliance on lerobot-record for inference","fracapuano",1,"fracapuano","#2143",0,"2025-10-08","2025-10-14"],[2,2137,"Async Inference Runtime Error when running pi0.5 policy server on GPU device and robot_client on CPU device","Llkhhb",2,"","",0,"2025-10-08","2025-10-10"],[2,2134,"what is the transformers version for latest lerobot pi0?","PuzhenYuan",2,"","",0,"2025-10-07","2025-10-07"],[2,2131,"no  lerobot/scripts/control_robot.py was found","hitbuyi",2,"","",0,"2025-10-07","2026-03-11"],[2,2129,"When using batch_encoding_size is >1 encoding may not occur before dataset push","nhnifong",3,"","",0,"2025-10-07","2026-04-20"],[2,2128,"Unclear documentation for expected format of LeRobotDataset v3.0 data","johnnewsom",2,"","",0,"2025-10-06","2025-10-07"],[2,2125,"Appending episodes to an existing dataset","nhnifong",2,"","",0,"2025-10-06","2025-10-08"],[2,2124,"Question regarding downsampling and resizing dataset","karthikm-0",2,"","",0,"2025-10-06","2026-10-21"],[2,2120,"Async Inference not working properly in newer version of Lerobot","H3771",2,"","",0,"2025-10-06","2025-10-07"],[2,2119,"Pi0.5 inference: ERROR 2025-10-06 14:50:04 y_server.py:247 Error in StreamActions: 'observation.language.token","Llkhhb",2,"","",0,"2025-10-06","2025-10-07"],[2,2116,"State Delta Timesteps is faulty for ACT","fracapuano",4,"","",0,"2025-10-05",""],[2,2114,"Fine-tuned π0 and π0.5 models fail to replicate reported success rates on LIBERO benchmark","ryuji-oi-artic",2,"","",0,"2025-10-04","2025-10-07"],[2,2112,"Dead link in docs on main","johnnewsom",2,"","",0,"2025-10-03","2025-10-17"],[2,2111,"frame deletion","Yysrc",2,"","",0,"2025-10-03","2025-10-10"],[2,2108,"HIL-SERL Transform order for (tanh → rescale) is reversed","priest-yang",4,"","",0,"2025-10-02",""],[2,2107,"Low Success Rate When Training SmolVLA-0.24B on LIBERO","zimgong",4,"","",0,"2025-10-02",""],[2,2105,"KeyError occurred while training the pi0.5 model","zhoushaoxiang",2,"","",0,"2025-10-02","2025-10-03"],[2,2104,"Select the VLM backbone for SmolVLA","Llkhhb",2,"","",0,"2025-10-02","2026-10-21"],[2,2102,"smolvla outdated documentation","camelone49",2,"","",0,"2025-10-01","2025-10-07"],[2,2098,"How can I set the ports for follower and leader arms without having to replug all the motors?","elpidiovaldez",2,"","",0,"2025-10-01","2026-03-11"],[2,2097,"Dataset not being uploaded to Lerobot Visualizer","cijerezg",2,"","",0,"2025-10-01","2025-10-17"],[2,2096,"How can I change the task name of already recorded episodes?","pparkgyuhyeon",2,"","",0,"2025-10-01","2026-07-31"],[2,2093,"Integration between ROS2 and Lerobot","clmnt",3,"","",0,"2025-09-30","2026-04-24"],[2,2092,"我希望增加这个lerobot项目易用性","foreignlands",2,"","",0,"2025-09-30","2025-10-08"],[2,2086,"google.protobuf.runtime_version.VersionError","Hukongtao",0,"imstevenpmwork","#2230",1,"2025-09-29","2025-10-17"],[2,2083,"How to train this RL model with my trained data","993984583",2,"","",0,"2025-09-29","2026-10-21"],[2,2082,"How to train this RL model with my model data","993984583",2,"","",0,"2025-09-29","2025-10-07"],[2,2078,"Why there is no \"visualize_dataset_html.py\" since updated to v3.0?","leledeyuan00",2,"","",0,"2025-09-28","2025-10-07"],[2,2074,"SO-101 leader precision","mirkokiefer",3,"","",0,"2025-09-28","2026-05-04"],[2,2072,"How to run lerobot with RTX 5090? If not possible, please add support","cijerezg",2,"","",0,"2025-09-27","2025-09-27"],[2,2070,"Lego builds","Utopiah",3,"","",0,"2025-09-27","2026-04-20"],[2,2069,"[bug] Cropped ROI in Reward Classifiers assumes you have write access to `repo_id` & can't `push_to_hub`","fracapuano",1,"fracapuano","#2071",0,"2025-09-27","2025-09-27"],[2,2067,"[bug] `pyav`: 'float' object has no attribute 'numerator'","fracapuano",2,"","",0,"2025-09-27","2025-09-27"],[2,2066,"Data recording much slower after adding more episodes?","richardrl",3,"","",0,"2025-09-27","2026-04-20"],[2,2065,"WebSocket protocol error","Chuong98CC",2,"","",0,"2025-09-26","2026-03-11"],[2,2061,"Make policies compatible with torch.compile","AdilZouitine",4,"","",0,"2025-09-26",""],[2,2054,"multiple h20  train model","zhoudelong",2,"","",0,"2025-09-26","2025-10-07"],[2,2053,"如何使用 多 gpu 训练 pi0 模型","zhoudelong",2,"","",0,"2025-09-26","2025-10-07"],[2,2051,"hil-serl performs well during training, but performs poorly when using the eval_policy.py","yfynb1111",1,"yfynb1111","#2073",0,"2025-09-26","2025-09-28"],[2,2050,"I wonder how to use RL on so101 within sim environment?","Temmp1e",2,"","",0,"2025-09-26","2025-10-08"],[2,2049,"[Bug] Video concatenation fails with time_base error when using resume=true in lerobot-record","robot4fun",2,"","",0,"2025-09-26","2025-09-27"],[2,2047,"How can i solve it?","pparkgyuhyeon",2,"","",0,"2025-09-26","2025-09-26"],[2,2046,"Calibration file error during Pi0 inference","pparkgyuhyeon",3,"","",0,"2025-09-26","2026-04-08"],[2,2045,"I would appreciate it if you could explain how to train the slicing clay model","pparkgyuhyeon",2,"","",0,"2025-09-26","2026-03-11"],[2,2044,"Cannot train with libero No module named 'robosuite.macros_private'","LukeLIN-web",2,"","",0,"2025-09-26","2025-10-08"],[2,2042,"Question: How to train to get Task Recovery behavior?","raul-machine-learning",3,"","",0,"2025-09-25","2026-04-08"],[2,2035,"using   \"lerobot/svla_so100_stacking\"   \"lerobot/svla_so100_sorting\"  LeRobotDataset    error","axia75",2,"","",0,"2025-09-25","2025-10-19"],[2,2034,"dataset v2.1 and groot n1.5","zujian-y",2,"","",0,"2025-09-24","2026-08-05"],[2,2025,"Python -m lerobot.teleoperate no longer works after teleoperate.py was moved","NikodemBartnik",2,"","",0,"2025-09-24","2025-09-24"],[2,2014,"Async Inference using policy_server and robot_client","Liu9999ai",3,"","",0,"2025-09-24","2026-07-31"],[2,2013,"terminate called after throwing an instance of 'std::bad_alloc'","yfynb1111",2,"","",0,"2025-09-24","2025-09-24"],[2,2011,"Adding support for the openarm bimanual system would be nice","rabhishek100",2,"","",0,"2025-09-23","2026-08-02"],[2,2000,"bug when recording the gymhil dataset","shs822",2,"","",0,"2025-09-23","2026-08-02"],[2,1999,"Error After Converting Dataset from v2.1 to v3 Using lerobot.record","pparkgyuhyeon",2,"","",0,"2025-09-23","2025-09-23"],[2,1998,"Distributed v2.1 -> v3.0 conversion","fracapuano",4,"","",0,"2025-09-22",""],[2,1995,"Questions about SmolVLA design","gliese581gg",3,"","",0,"2025-09-22","2026-04-29"],[2,1994,"How to improve success rate and generalization","Liu9999ai",2,"","",0,"2025-09-22","2025-09-23"],[2,1992,"Could not find 'policy_preprocessor.json' on the HuggingFace Hub at 'lerobot/smolvla_base' when trying to trai","cijerezg",2,"","",0,"2025-09-21","2026-07-31"],[2,1991,"Pi0 interperates single task as array of 1 character tasks","Skylerwiernik",2,"","",0,"2025-09-20","2025-09-20"],[2,1988,"running command-line data recording command fails when using intel realsense camera","leonliugf",3,"","",0,"2025-09-20","2026-04-05"],[2,1981,"incorrect values in episode_data_index when using relative pathing","richardrl",2,"","",0,"2025-09-19","2025-09-19"],[2,1978,"Is there a best fit model to each sim env？","shs822",2,"","",0,"2025-09-19","2025-10-17"],[2,1969,"how to record a multi-task dataset on so101?","Temmp1e",2,"","",0,"2025-09-18","2025-09-19"],[2,1966,"SO101FollowerEndEffector?","cashlo",2,"","",0,"2025-09-17","2026-10-21"],[2,1962,"Openpi now provides Pytorch support","Skylerwiernik",2,"","",0,"2025-09-17","2025-09-23"],[2,1959,"write_image() is slow due to default compress_level=6","playertr",0,"imstevenpmwork","#2135",1,"2025-09-16","2025-10-08"],[2,1952,"Dataset v3 related errors","Hukongtao",0,"Tavish9","#2076",0,"2025-09-16","2025-10-07"],[2,1951,"Training pi0 in aloha sim got 0% success rate","cindeee",2,"","",0,"2025-09-16","2026-10-21"],[2,1946,"feat: Bimanual support for SO101","bingogome",3,"","",0,"2025-09-16","2026-03-30"],[2,1944,"smolva asynchonous inference not working in my robots too","leonliugf",2,"","",0,"2025-09-16","2025-09-18"],[2,1933,"keyboard input with gym_manipulator on Linux","rwst",2,"","",0,"2025-09-14","2025-09-14"],[2,1923,"Deploying SmolVLA with a simulator","aditya1709",2,"","",0,"2025-09-12","2025-09-23"],[2,1919,"convert_dataset_v21_to_v30 creates bugged metadata if there are images in the parquet","gmullin3",2,"","",0,"2025-09-11","2025-10-08"],[2,1916,"[Bug] Visualization Space fails for v2.1 dataset after v3.0 update","liujiting123",2,"","",0,"2025-09-11","2025-09-21"],[2,1914,"Add LoRA Support for Model Fine-tuning","ywlng",2,"","",0,"2025-09-11","2025-09-23"],[2,1912,"Should delta_timesteps be given in timesteps (integers) instead of seconds?","fracapuano",3,"","",0,"2025-09-10","2026-03-27"],[2,1911,"How to avoid re-write cache data from pyarrow into parquet everytime?","songlinwei-we",3,"","",0,"2025-09-10","2026-03-24"],[2,1907,"AttributeError: 'GemmaModel' object has no attribute 'prepare_inputs_for_generation';","miya-mizumitsuha",2,"","",0,"2025-09-10","2025-09-11"],[2,1899,"Has anyone tried to export the smolvla as onnx model for deployment?","TankerLee",3,"","",0,"2025-09-09","2026-04-20"],[2,1896,"pi0.5 support","lucasjinreal",2,"","",0,"2025-09-09","2025-09-23"],[2,1895,"train.py --dataset.episodes causes IndexError when episode index exceeds filtered dataset size","robot4fun",2,"","",0,"2025-09-08","2025-09-25"],[2,1893,"Dataset start timestamp?","m01",2,"","",0,"2025-09-08","2026-10-21"],[2,1892,"TypeError: SO100FollowerConfig.__init__() missing 1 required positional argument: 'port'","wahliuyunqi-cmd",3,"","",0,"2025-09-08","2026-03-22"],[2,1891,"How to checkout a commit id?","richardrl",2,"","",0,"2025-09-08","2025-09-10"],[2,1883,"Backup and restore motor calibration file","bruno6527",3,"","",0,"2025-09-07","2026-03-22"],[2,1882,"Pretrain - Code for pretraining smolvla","ruiheng123",2,"","",0,"2025-09-07","2025-09-23"],[2,1879,"MultiLeRobotDataset - use repo_id in delta timestamps","villekuosmanen",3,"","",0,"2025-09-05","2026-03-19"],[2,1875,"ValueError: One or several timestamps unexpectedly violate the tolerance inside episode range. This might be d","cuijianaaa",2,"","",0,"2025-09-05","2026-03-11"],[2,1874,"Invalid input feature for pi0","sdfzz",3,"","",0,"2025-09-05","2026-04-06"],[2,1871,"problematic target_entropy initialization in HIL_SERL","SitheralS",4,"","",0,"2025-09-05",""],[2,1847,"Koch Arm Unable to Use find_joint_limits.py in HIL-serl Policy","ethanCSL",4,"","",0,"2025-09-03",""],[2,1842,"Using the latest version of training in libero the loss is as high as 0.05","ymzhong66",3,"","",0,"2025-09-02","2026-04-06"],[2,1839,"Run model based on a gamepad recorded dataset","SonDePoisson",4,"","",0,"2025-09-02",""],[2,1838,"After converted the images to videos, an error occurred during training, indicating video frames is abnormal.","dailouer-guo",2,"","",0,"2025-09-02","2025-09-02"],[2,1834,"Questions about asynchronous inference","godnpeter",2,"","",0,"2025-09-01","2025-09-23"],[2,1828,"Allow the use of gymnasium >0.29.1","traversaro",2,"","",0,"2025-09-01","2025-10-17"],[2,1824,"Example of running HIL-SERL code","Hukongtao",2,"","",0,"2025-09-01","2026-10-21"],[2,1822,"As of 08/31/2025, how do you create a v2.1 dataset from raw data?","IrvingF7",3,"","",0,"2025-08-31","2026-04-21"],[2,1820,"Suggest to install conda via miniforge instead of miniconda?","traversaro",0,"imstevenpmwork","#2249",1,"2025-08-31","2025-10-19"],[2,1819,"关于Hirserl的复现：","bxsytt",4,"","",0,"2025-08-30",""],[2,1811,"Finetuning (overfitting)  pi0Fast for Trossen WidowX_ai","perseusdg",2,"","",0,"2025-08-29","2025-10-14"],[2,1794,"make_policy modify to load pretrained model without env or ds_meta","csjiyw",3,"","",0,"2025-08-28","2026-04-06"],[2,1792,"how to train lerobot model offline with offline data?","dalishi",2,"","",0,"2025-08-26","2025-09-03"],[2,1791,"[Question] Has anyone successfully fine-tuned multitasking using SmolVLA?","kaaatsu32329",2,"","",0,"2025-08-26","2025-09-23"],[2,1790,"Does the new version of the code support a two-leader-two-follower Koch robotic arm?","zhoushaoxiang",2,"","",0,"2025-08-26","2025-09-09"],[2,1789,"Loading RGB into memory","richardrl",3,"","",0,"2025-08-26","2026-03-09"],[2,1786,"Order of the second norm op in ACTEncoderLayer","xiaotaw",3,"","",0,"2025-08-25","2026-03-09"],[2,1785,"Unrecognized model in lerobot/smolvla_base. Should have a `model_type` key in its config.json","SonDePoisson",3,"","",0,"2025-08-25","2026-03-09"],[2,1784,"Is there a bug in the asynchronous reasoning of smolvla?","w646",3,"","",0,"2025-08-25","2026-03-18"],[2,1783,"Load dataset with selected episodes does not work with delta_timestamps","jjyyxx",3,"","",0,"2025-08-24","2026-03-08"],[2,1779,"Can not run find_joint_limits for so101 with config file","Skylerwiernik",0,"imstevenpmwork","#2577",1,"2025-08-23","2025-12-03"],[2,1777,"[Feature Request] Support adding custom environments and policies without modifying LeRobot source","TatsukiNishimura",2,"","",0,"2025-08-22","2025-08-26"],[2,1775,"What's the finetuning method? Is it all full-finetuning?","lin-whale",2,"","",0,"2025-08-22","2025-10-07"],[2,1774,"Finetune smolvla with vision encoder","THU-yancow",2,"","",0,"2025-08-22","2026-07-29"],[2,1773,"Loss of pi0 training on libero is extremely hign(up to 500)!","hamondyan",2,"","",0,"2025-08-22","2025-09-23"],[2,1772,"Running Async Inference on PI0","eliasab16",2,"","",0,"2025-08-21","2025-12-11"],[2,1768,"numpy incompatibility issues","dalishi",2,"","",0,"2025-08-21","2025-08-26"],[2,1766,"Slave arm running lagged","andrew123-q",3,"","",0,"2025-08-21","2026-03-06"],[2,1765,"Questions about using LIBERO dataset (loss starts extremely high)","hamondyan",3,"","",0,"2025-08-21","2026-04-06"],[2,1763,"The best camera setup for fine-tuning SmolVLA with so101","yuanlin2004",2,"","",0,"2025-08-19","2025-08-29"],[2,1747,"Multi-robot Composite Config","wengmister",3,"","",0,"2025-08-17","2026-03-01"],[2,1743,"Cannot run a record with two cameras","FaboNo",3,"","",0,"2025-08-15","2026-04-30"],[2,1742,"Recording a dataset failed by calling smolvla adam optimizer","FaboNo",2,"","",0,"2025-08-15","2025-08-15"],[2,1741,"smolvla's asynchronous inference does not work","onaka-ga-pkpk",2,"","",0,"2025-08-15","2025-08-27"],[2,1739,"Eval Error","mahao18cm",2,"","",0,"2025-08-15","2025-08-18"],[2,1738,"Raspberry Pi 4 Installation","ccha425",2,"","",0,"2025-08-15","2025-10-17"],[2,1736,"Bug Report: Incorrect Key in “actions_id_pad”","Gutilence14",0,"ymd-h","#1717",0,"2025-08-14","2026-03-11"],[2,1734,"Kochv1.1 Leader arm setup issue","aashishdhanani",3,"","",0,"2025-08-14","2026-02-25"],[2,1733,"Missing key(s) when loading model error when training pi0","sdfzz",2,"","",0,"2025-08-13","2025-08-14"],[2,1732,"Ensure the model module passes MyPy type checks","AdilZouitine",0,"quanghuynt14","#1782",0,"2025-08-13","2025-10-19"],[2,1731,"Ensure the transport module passes MyPy type checks","AdilZouitine",0,"sezan92","#1",1,"2025-08-13","2025-12-03"],[2,1730,"Ensure the configuration module passes MyPy type checks","AdilZouitine",0,"tetsugo02","#1746",0,"2025-08-13","2025-10-20"],[2,1729,"Ensure the optimization module passes MyPy type checks","AdilZouitine",0,"cverrier","#2683",0,"2025-08-13","2025-12-20"],[2,1728,"Ensure the environments module passes MyPy type checks","AdilZouitine",2,"","",0,"2025-08-13","2025-10-17"],[2,1727,"Ensure the utilities module passes MyPy type checks","AdilZouitine",2,"","",0,"2025-08-13","2026-09-28"],[2,1726,"Ensure the teleoperators module passes MyPy type checks","AdilZouitine",2,"","",0,"2025-08-13","2026-09-28"],[2,1725,"Ensure the processing module passes MyPy type checks","AdilZouitine",3,"","",0,"2025-08-13","2026-02-25"],[2,1724,"Ensure the cameras module passes MyPy type checks","AdilZouitine",3,"","",0,"2025-08-13","2026-02-25"],[2,1723,"Ensure the motors module passes MyPy type checks","AdilZouitine",0,"imstevenpmwork","35363c5",1,"2025-08-13","2026-02-10"],[2,1722,"Ensure the datasets module passes MyPy type checks","AdilZouitine",3,"","",0,"2025-08-13","2026-02-25"],[2,1721,"Ensure the robots module passes MyPy type checks","AdilZouitine",3,"","",0,"2025-08-13","2026-03-20"],[2,1720,"Ensure the policy module passes MyPy type checks","AdilZouitine",3,"","",0,"2025-08-13","2026-03-03"],[2,1719,"Make LeRobot mypy compliant","AdilZouitine",3,"","",0,"2025-08-13","2026-02-26"],[2,1716,"`max_relative_target` should be a float","brysonjones",0,"jstnhuang","#1837",0,"2025-08-12","2025-09-05"],[2,1714,"Slave Arm Jitter and Latency","CassieXiong",3,"","",0,"2025-08-12","2026-03-01"],[2,1712,"Why hasn't the pi0 model learned the ability to place something in the specified positions? Is it because the ","Alex-Wlog",2,"","",0,"2025-08-12","2026-10-21"],[2,1708,"Factory.py mutable","dengpingsong",3,"","",0,"2025-08-11","2025-08-11"],[2,1707,"SmolVLA policy uses wrong action pad key","ymd-h",1,"ymd-h","#1717",0,"2025-08-10","2026-03-11"],[2,1704,"Nightly Docker container CPU and GPU naming inversion","gdaddi",0,"imstevenpmwork","#1705",1,"2025-08-08","2025-08-08"],[2,1702,"SmolVLMVisionEmbeddings on wrong device","alfieroddan",2,"","",0,"2025-08-08","2026-03-11"],[2,1701,"Episode filtering triggers out of bound error when used with delta timestamps","txmalvarez",3,"","",0,"2025-08-08","2026-02-24"],[2,1700,"Dimension mismatch when inference PI0","kotorikanbe",3,"","",0,"2025-08-08","2026-02-24"],[2,1699,"Raspberry PI 4: Camera error","jfmateos",3,"","",0,"2025-08-08","2026-02-24"],[2,1697,"Nice work,but would you mind to extend the past frames or actions to the model？","oniu-uin0",2,"","",0,"2025-08-08","2025-09-23"],[2,1694,"New version calibration error.","Jianliang-Shen",3,"","",0,"2025-08-07","2026-02-26"],[2,1693,"Teleoperating SO101 with 2 cameras, realsense camera is not found, Couldn't find a choice class for 'realsense","MarioCavero",2,"","",0,"2025-08-07","2025-08-07"],[2,1688,"Gym_hil Simulation Training : num_episodes & episode_length hard coded","SonDePoisson",3,"","",0,"2025-08-06","2026-02-24"],[2,1687,"When using AMP to train a model, why are the saved model weights still in fp32?","Hukongtao",2,"","",0,"2025-08-06","2026-02-10"],[2,1683,"SmolVLMWithExpertModel","xjushengjie",3,"","",0,"2025-08-06","2026-02-24"],[2,1682,"Raspberry Pi 4: conda installs ffmpeg version 4.3.2 without libsvtav1 library and error installing feetech","jfmateos",3,"","",0,"2025-08-06","2026-02-24"],[2,1678,"must be called with a dataclass type or instance","xk21520",3,"","",0,"2025-08-06","2026-02-24"],[2,1677,"No policy or teleoperator provided, skipping action generation.This is likely to happen when resetting the env","Benxiaogu",2,"","",0,"2025-08-06","2025-08-12"],[2,1675,"TimeoutError: Timed out waiting for frame from camera OpenCVCamera(0) after 200 ms. Read thread alive: True","CrossyNautilus",2,"","",0,"2025-08-06","2025-08-31"],[2,1674,"How to train smolvla for multi-task","w673",2,"","",0,"2025-08-06","2025-08-07"],[2,1672,"How to resume training?","milong26",2,"","",0,"2025-08-05","2025-08-06"],[2,1670,"How does leroBot address the issue of training heterogeneous datasets?","mahao18cm",3,"","",0,"2025-08-05","2026-02-24"],[2,1668,"Calibration is not read correctly on SO101 follower","EricPedley",2,"","",0,"2025-08-05","2025-08-05"],[2,1667,"How many episode to have a good result of SmolVLA","chejulien",2,"","",0,"2025-08-05","2025-10-17"],[2,1666,"Please add multi gpu training support","nahidalam",2,"","",0,"2025-08-04","2025-10-17"],[2,1664,"Bug in policy construction","fracapuano",1,"fracapuano","#1665",0,"2025-08-04","2025-08-04"],[2,1663,"No way to train on subset of features","atyshka",2,"","",0,"2025-08-04","2026-01-07"],[2,1654,"Release 0.4.0","imstevenpmwork",1,"imstevenpmwork","#2298",1,"2025-08-02","2025-11-03"],[2,1641,"ValueError: 'policy.repo_id' argument missing. Please specify it to push the model to the hub","321915514",2,"","",0,"2025-08-01","2025-10-17"],[2,1634,"lerobot visualize_dataset huggingface space shows blank white screen","MilkClouds",3,"","",0,"2025-07-31","2026-02-24"],[2,1632,"Are there plans to support distributed training?","Hukongtao",2,"","",0,"2025-07-31","2025-10-17"],[2,1631,"🥚 Filtering Eggs on Moving Table: Dirt/Breakage Detection Feasibility","KannarFr",3,"","",0,"2025-07-30","2026-02-24"],[2,1627,"Diffusion policy training at recent commits may not working properly","HiroIshida",2,"","",0,"2025-07-30","2025-08-13"],[2,1626,"Running ACT training gave me following error:     raise TypeError('must be called with a dataclass type or ins","samuelp4065",3,"","",0,"2025-07-30","2026-02-24"],[2,1624,"PermissionError: Cannot create temporary files on Windows (even after setting TEMP/TMP)","YouXiangyu",3,"","",0,"2025-07-30","2026-02-24"],[2,1623,"Slow video decoding","atyshka",3,"","",0,"2025-07-29","2026-04-06"],[2,1622,"Why is LeRobot’s policy ignoring additional camera streams despite custom `input_features`?","Aaron-Serpilin",3,"","",0,"2025-07-29","2026-04-06"],[2,1620,"Policy image features name mismatch","RiccardoIzzo",2,"","",0,"2025-07-29","2025-10-17"],[2,1616,"DiffusionPolicy bug when not using image features","abhaybd",1,"abhaybd","#1617",0,"2025-07-28","2025-07-29"],[2,1607,"how to control a so-101 with trained ACT model?","initia1013",3,"","",0,"2025-07-28","2026-04-28"],[2,1606,"Wrist camera is glitchy when using record but works well with teleoperate","cijerezg",2,"","",0,"2025-07-28","2025-08-23"],[2,1603,"Missing Motor IDs","shrutic12",2,"","",0,"2025-07-27","2026-07-31"],[2,1602,"How to perform multi-GPU training for SMoVLA?","QZepHyr",2,"","",0,"2025-07-27","2025-07-28"],[2,1599,"Evaluation results of VLA models on MetaWorld Benchmark","Zooy138",3,"","",0,"2025-07-26","2026-02-24"],[2,1598,"Questioning Expert width Choices in SmolVLA","QZepHyr",2,"","",0,"2025-07-26","2025-07-28"],[2,1595,"can not calibrate SO101 leader arm","Nos-Talion",2,"","",0,"2025-07-25","2025-07-25"],[2,1594,"Error when loading dataset with dtype=\"image\" features","destroy314",2,"","",0,"2025-07-25","2026-03-11"],[2,1592,"Time spent on imitation learning training (ACT)","initia1013",2,"","",0,"2025-07-25","2025-10-08"],[2,1590,"trouble running arm","jeffrey-c-star",3,"","",0,"2025-07-24","2026-02-24"],[2,1586,"Real-world deploy on ALOHA Robot","LogSSim",3,"","",0,"2025-07-24","2026-03-05"],[2,1585,"PermissionError: [Errno 13] on Windows when using lerobot record","krayF1sh",3,"","",0,"2025-07-24","2026-04-22"],[2,1582,"Issue with lerobot.find_port. Not Detecting USB Device","DIMARIA123",3,"","",0,"2025-07-24","2026-02-24"],[2,1580,"Environment_State in act and SmolVLA policy","kasiv008",2,"","",0,"2025-07-24","2025-10-08"],[2,1579,"Is there a video backend supporting nondestructive encoding?","milong26",3,"","",0,"2025-07-23","2026-02-24"],[2,1578,"Lerobot metaworld dataset only provides 49 tasks","chenkang455",3,"","",0,"2025-07-23","2026-02-24"],[2,1577,"test failed after training SVLA","Liu-Xiaoyan97",3,"","",0,"2025-07-23","2026-02-24"],[2,1576,"Multiple Dataset training","JustinKai0527",4,"","",0,"2025-07-23",""],[2,1575,"docker documentation needed","LokiMetaSmith",2,"","",0,"2025-07-22","2026-08-05"],[2,1571,"Training SmolVLA, ValueError: Feature type 'List' not found.","initia1013",2,"","",0,"2025-07-22","2025-09-23"],[2,1569,"Fix documentation, resume recording","gribok201",2,"","",0,"2025-07-22","2026-03-11"],[2,1567,"Support for deploying VLA models (e.g., π₀‑FAST) on a real Franka Emika Panda","aatxag",2,"","",0,"2025-07-22","2025-10-08"],[2,1564,"How are Episode Stats used?","andlyu",2,"","",0,"2025-07-21","2025-08-12"],[2,1561,"will you release the libero ft&eval setting?","JuilieZ",2,"","",0,"2025-07-21","2025-09-23"],[2,1559,"Is the current model framework suitable for using automatic mixed precision?","xliu0105",3,"","",0,"2025-07-21","2026-02-24"],[2,1557,"async inference \"self._obs_sanity_checks(obs, self.last_processed_obs)\"will never be True","MoeOver",2,"","",0,"2025-07-21","2025-07-22"],[2,1554,"How to use local datasets to train and evaluate","zym123321",2,"","",0,"2025-07-21","2025-10-08"],[2,1552,"Support smolvla training on Intel GPU","xiangyang-95",3,"","",0,"2025-07-21","2026-04-22"],[2,1546,"Dataset validation inconsistency in TrainRLServerPipelineConfig","Hardwarize",4,"","",0,"2025-07-20",""],[2,1541,"unrecognized arguments: --local-files-only","yingjiewei",2,"","",0,"2025-07-19","2026-02-02"],[2,1540,"Controlling robot with text using SmolVLA","drain-pipe",3,"","",0,"2025-07-18","2026-02-24"],[2,1539,"DiffusionPolicy dimension mismatch error when n_obs_steps=1","abhaybd",3,"","",0,"2025-07-18","2026-03-03"],[2,1538,"issue training model","jeffrey-c-star",2,"","",0,"2025-07-18","2025-09-23"],[2,1537,"Pi0 Inference Latency Much Higher Than Reported in Pi0 Paper","MicDZ",2,"","",0,"2025-07-18","2025-10-08"],[2,1536,"Missing template files for visualizer when installing lerobot as dependency","TilTheunissen",3,"","",0,"2025-07-18","2026-02-24"],[2,1532,"Add proper versioning to repo","emgeee",2,"","",0,"2025-07-17","2026-02-02"],[2,1531,"The use of Action-specific min-max normalization in PI0Fast Encoding","harshas123",2,"","",0,"2025-07-17","2025-10-14"],[2,1530,"lerobot-gpu Docker environment does not render eval environments on GPU","dekahaedra",2,"","",0,"2025-07-17","2025-07-24"],[2,1529,"lerobot/pi0 safetensors bug","shenyanqing1105",2,"","",0,"2025-07-17","2025-10-08"],[2,1527,"Teleop \"max_relative_target\" Type Error","asyk454",0,"jstnhuang","#1837",0,"2025-07-17","2025-09-05"],[2,1523,"[HIL SERL] multiprocessing Queue.get_nowait is not getting all the items in the queue","jpizarrom",1,"jpizarrom","#1524",0,"2025-07-17","2025-07-17"],[2,1518,"Dual Arm support - SOARM101","cerealkiller2527",2,"","",0,"2025-07-16","2026-08-04"],[2,1516,"pi0fast","Hale-Sage",2,"","",0,"2025-07-16","2025-07-22"],[2,1508,"so101_dualarm_triplecam config to evaluate ACT policy?","sebastiandavidlee",3,"","",0,"2025-07-15","2026-02-24"],[2,1507,"[PI0] Evaluation result on the metaworld","chenkang455",2,"","",0,"2025-07-14","2025-10-08"],[2,1506,"episode: None","LogSSim",3,"","",0,"2025-07-14","2026-02-24"],[2,1505,"TypeError in LeRobotDataset When Processing Timestamps and Episode Indices","cindeee",2,"","",0,"2025-07-14","2025-07-14"],[2,1503,"LeRobot So100 and Groot N1.5 Model Multi-Robot Deployment Feasibility Inquiry","devedgar",3,"","",0,"2025-07-14","2026-02-24"],[2,1500,"async inference doesn't generate actions.Observation #0 has been filtered out","milong26",2,"","",0,"2025-07-13","2025-07-30"],[2,1499,"Loading video indefinitely","drain-pipe",2,"","",0,"2025-07-13","2025-07-18"],[2,1498,"Incorrect normalization in Pi0FAST","hgaurav2k",2,"","",0,"2025-07-13","2025-10-14"],[2,1497,"ValueError: 'policy.repo_id' argument missing. Please specify it to push the model to the hub.","dbdxnuliba",3,"","",0,"2025-07-13","2026-02-24"],[2,1494,"release the code for reproducing the performance on the LIBERO dataset reported in the SmolVLA paper?","JustinKai0527",2,"","",0,"2025-07-12","2025-09-23"],[2,1492,"Is there any plan to add a validation loss in the training pipeline, which is not dependent on simulation env.","mohitydv09",2,"","",0,"2025-07-11","2026-07-31"],[2,1489,"The koch leader arm calibration has inconsistent drive modes and the position underflows","Joeclinton1",3,"","",0,"2025-07-11","2026-03-05"],[2,1488,"Dataloader is blazingly fast for ACT training but VERY slow for SmolVLA and Diffusion Policy training","DominiquePaul",2,"","",0,"2025-07-11","2026-10-21"],[2,1487,"Corrupt JPEG data","qingx-cyber",3,"","",0,"2025-07-11","2026-02-24"],[2,1483,"How can I set `max_relative_target` to get safe action?","milong26",3,"","",0,"2025-07-11","2026-02-24"],[2,1476,"Here as interactive gym to play with the robot, (I still need some help)","raul-machine-learning",2,"","",0,"2025-07-09","2026-10-21"],[2,1475,"[Question] What does each number in predicted action(SmolVLA) stand for?","Calvert0921",3,"","",0,"2025-07-09","2026-02-24"],[2,1472,"The GPU/graphics driver is lacking some abilities:Adapter does not support drawing to texture format R32Float.","ddwy",3,"","",0,"2025-07-09","2026-01-20"],[2,1471,"where is 7_get_started_with_real_robot.md?","von63",2,"","",0,"2025-07-09","2025-10-08"],[2,1470,"Suggestion: Int Image format","atyshka",3,"","",0,"2025-07-08","2026-02-24"],[2,1467,"No matching distribution found for torchvision>=0.21.0","SonDePoisson",2,"","",0,"2025-07-08","2026-03-11"],[2,1465,"Opencv cannot opened same camera twice with different backend","LaurentBerger",3,"","",0,"2025-07-08","2026-01-20"],[2,1462,"SO101 kinematics problem","Compromiseee",2,"","",0,"2025-07-08","2025-07-09"],[2,1461,"pi0 lack of params when loading from pretrained model","OpenJarvisAI",2,"","",0,"2025-07-08","2025-10-08"],[2,1460,"How to support dataloading with historical cue?","joeyxin-del",3,"","",0,"2025-07-08","2026-02-24"],[2,1459,"AssertionError: mean is infinity","lei732",2,"","",0,"2025-07-08","2025-10-08"],[2,1458,"how to control a real robot arm-101 with my own pretrained model?","jcl2023",3,"","",0,"2025-07-08","2026-02-24"],[2,1456,"Camera not detected on windows laptop","jeffrey-c-star",3,"","",0,"2025-07-07","2026-02-24"],[2,1449,"ValueError: Shape of 'min' must be (3,1,1), but is (84, 84, 3) instead","MrWiffer",2,"","",0,"2025-07-07","2026-03-11"],[2,1448,"How to specify both policy.type and pretrained path at the same time?","branyang02",3,"","",0,"2025-07-07","2026-02-24"],[2,1447,"SmolVLA input/output clarification","Calvert0921",2,"","",0,"2025-07-06","2025-10-09"],[2,1446,"How to evaluate finetuned SmolVLA model","BintaoBryant",2,"","",0,"2025-07-06","2025-10-17"],[2,1444,"huggingface/tokenizers: The current process just got forked, after parallelism has already been used.","milong26",3,"","",0,"2025-07-04","2026-02-24"],[2,1442,"Trained pi0 policy ignores visual cues","kumarhans",3,"","",0,"2025-07-03","2026-02-24"],[2,1440,"Internal Server Error with visualize_dataset_html.py","seann999",3,"","",0,"2025-07-03","2026-02-24"],[2,1439,"[QUESTION] run a policy on a real robot","FaboNo",3,"","",0,"2025-07-03","2026-02-24"],[2,1438,"lerobot training script is broken in google colab, referenced on the website, corrected code below","therealadityashankar",2,"","",0,"2025-07-03","2026-05-26"],[2,1434,"Data recording takes very long (30-40s processing per 15s episode)","DominiquePaul",2,"","",0,"2025-07-02","2026-03-11"],[2,1433,"StopIteration Exception in Teleoperate","jtabor",3,"","",0,"2025-07-02","2026-02-25"],[2,1429,"When will release the SmolVLA(2.25B & 0.24b)","JuilieZ",2,"","",0,"2025-07-02","2025-10-08"],[2,1428,"single_task broken into characters and causes eval to fail","MatthewWaller",3,"","",0,"2025-07-01","2026-02-24"],[2,1427,"pip install fails: lerobot is not a package","DipFlip",2,"","",0,"2025-07-01","2025-07-02"],[2,1426,"Failed to sync read 'Present_Position' on ids=[1, 2, 3, 4, 5, 6] after 4 tries.","gribok201",3,"","",0,"2025-07-01","2026-03-05"],[2,1425,"Failed to sync read 'Present_Position' on ids=[1, 2, 3, 4, 5, 6] after 1 tries.","gribok201",2,"","",0,"2025-07-01","2025-07-01"],[2,1424,"evaluated trained policy reports 14 pc_success only","raul-machine-learning",3,"","",0,"2025-07-01","2026-02-24"],[2,1423,"running eval.py as mentioned in source code causes `unrecognized arguments` error","raul-machine-learning",2,"","",0,"2025-07-01","2025-07-01"],[2,1422,"`7_get_started_with_real_robot.md` referenced in md file doesn't exist","raul-machine-learning",3,"","",0,"2025-07-01","2026-02-24"],[2,1421,"It would help to have a description for the lerobots datasets:","raul-machine-learning",3,"","",0,"2025-07-01","2026-02-24"],[2,1420,"Quick example in the documentation shouldn't push to hub","raul-machine-learning",3,"","",0,"2025-07-01","2026-02-24"],[2,1419,"simulator should allow pushing objects around with the mouse interactively","raul-machine-learning",3,"","",0,"2025-07-01","2026-02-24"],[2,1418,"Robot tries to transfer cube even if it failed to pick it up, shouldn't it retry?","raul-machine-learning",2,"","",0,"2025-07-01","2025-10-17"],[2,1414,"Identifying Motors 2-6","ramallis",3,"","",0,"2025-06-30","2026-02-25"],[2,1410,"GLFW Error","fireyan8",2,"","",0,"2025-06-30","2026-10-21"],[2,1407,"Can read the current signals from the lerobot?","Frank-ZY-Dou",3,"","",0,"2025-06-30","2026-02-25"],[2,1406,"PI0 policy suffers from weight/key mismatch on main when using transformers ≥ 4.52.0","yongjincho",2,"","",0,"2025-06-30","2026-03-11"],[2,1405,"[QUESTION] about the content of the dataset","FaboNo",2,"","",0,"2025-06-29","2025-06-30"],[2,1400,"No terminal logging when running training or policy execution scripts","unbeschwert",3,"","",0,"2025-06-27","2026-02-25"],[2,1399,"calibrate.py for only follower","ramallis",3,"","",0,"2025-06-27","2026-02-25"],[2,1398,"Using LeRobotDataset in IsaacLab Without Unused Dependencies","ashwinvkNV",3,"","",0,"2025-06-27","2026-02-25"],[2,1393,"motor configuration request - one motor at a time like configure_motors","brainwavecoder9",3,"","",0,"2025-06-26","2026-02-25"],[2,1389,"ConnectionError: Read failed due to communication error on port /dev/ttyACM0 for group_key Present_Position_sh","saraparoji",2,"","",0,"2025-06-26","2026-10-29"],[2,1388,"RuntimeError: ('SmolVLA %d missing / %d unexpected keys', 10, 0 when loading smolvla_base","milong26",2,"","",0,"2025-06-26","2025-07-01"],[2,1387,"so101 can't work with HIL-SERL","yangfengzzz",4,"","",0,"2025-06-26",""],[2,1383,"Can multiple Lerobot datasets be mixed to pre-train a VLA model?","xliu0105",3,"","",0,"2025-06-25","2026-02-25"],[2,1380,"smolVLA paper `o_t`","chenyenru",2,"","",0,"2025-06-24","2025-06-26"],[2,1379,"New motor configuration doesn't center servo motors for so100","Esser50K",3,"","",0,"2025-06-24","2026-02-25"],[2,1378,"Not able to finetune on a dataset with more than 50 episodes","siddhu003",2,"","",0,"2025-06-24","2025-06-25"],[2,1377,"`smolvla_base training`: `Expected all tensors to be on the same device`","raul-machine-learning",2,"","",0,"2025-06-24","2025-07-01"],[2,1374,"Warning: WandB warning when I finetune `pi0`","YushunXiang",0,"reeceomahoney","#2903",0,"2025-06-23","2026-02-09"],[2,1372,"Incorrect Number of Recorded Episodes when Rerecording","cverrier",2,"","",0,"2025-06-23","2025-06-27"],[2,1370,"SmolVLA have poor performance","FanZhang91",2,"","",0,"2025-06-23","2025-09-23"],[2,1369,"The performance of SmolVLA on LIBERO cannot be replicated","hahans",2,"","",0,"2025-06-23","2025-09-23"],[2,1368,"OpenCV camera connection fails on Windows due to default backend","myanvoos",2,"","",0,"2025-06-23","2026-03-11"],[2,1366,"Pi0 have low performance with CALVIN dataset","ducido",2,"","",0,"2025-06-22","2026-10-21"],[2,1365,"MuJoCo Cannot Find OpenGL on macOS","cverrier",3,"","",0,"2025-06-22","2026-03-15"],[2,1361,"Nvidia Gr00t","AbdElRahmanFarhan",3,"","",0,"2025-06-21","2026-03-04"],[2,1360,"Homing offset not taken into account during calibration","godardt",3,"","",0,"2025-06-21","2026-02-25"],[2,1359,"Not clear how to setup a basic interactive simulator demo","aguaviva",2,"","",0,"2025-06-20","2025-10-09"],[2,1357,"Integration of Trossen Robotics newest arms","brysonjones",3,"","",0,"2025-06-20","2026-07-31"],[2,1353,"Second camera read error with so101 dual camera","xmusco",3,"","",0,"2025-06-19","2026-02-25"],[2,1351,"Need help about dataset and train.","hbj52152",2,"","",0,"2025-06-19","2025-10-17"],[2,1350,"Couldn't find a choice class for 'smolvla'","kienchen",2,"","",0,"2025-06-19","2025-10-09"],[2,1346,"LeRobotDataset May Eagerly Load Images, Making It Unusable for Large Datasets","klingmad",2,"","",0,"2025-06-18","2025-10-31"],[2,1345,"A Bug About SmolVLA's embed_suffixz Function","hahans",2,"","",0,"2025-06-18","2025-06-18"],[2,1344,"the directory called \"robot_devices\" which was inside directory \"common\" is missing","azam2u",3,"","",0,"2025-06-18","2026-02-25"],[2,1342,"Calibration parameters are not written to motors","Kracozebr",3,"","",0,"2025-06-17","2026-02-25"],[2,1341,"Support for Pi0-FAST Droid and Pi0 Droid variants","johnnewsom",2,"","",0,"2025-06-17","2025-10-14"],[2,1337,"how to work with ur robot,and collect the data and fine turn the model ?","mmlingyu",2,"","",0,"2025-06-17","2025-10-17"],[2,1336,"def get_cv2_backend() in camera.utils","Emonumia",3,"","",0,"2025-06-17","2026-02-25"],[2,1335,"Request support for the Agilex Piper robotic arm","yquanli",2,"","",0,"2025-06-17","2025-07-27"],[2,1333,"SO-100 Follower: Severe wrist_roll motor instability causing unwanted rotation during teleoperation","TKDRYU104",2,"","",0,"2025-06-17","2026-04-14"],[2,1330,"Could you update the repository to enable the evaluation of SmolVLA's performance?","Pandapan01",2,"","",0,"2025-06-17","2025-10-17"],[2,1329,"Size mismatch error when evaluating a fine-tuned VLA model","RonanFR",3,"","",0,"2025-06-16","2026-02-25"],[2,1328,"Script python -m lerobot.calibrate --teleop.type=so101_leader --teleop.port=/dev/ttyACM1 --teleop.id=the_leade","FaboNo",2,"","",0,"2025-06-16","2025-06-29"],[2,1325,"Does Pi0 load the wrong PaliGemma config in case of missing model_type?","Qu3tzal",2,"","",0,"2025-06-16","2025-10-08"],[2,1324,"Where is control_robot.py script?","AbdElRahmanFarhan",2,"","",0,"2025-06-16","2025-06-18"],[2,1321,"Could lerobot support COBOT Magic?","huazai665",3,"","",0,"2025-06-16","2026-02-25"],[2,1319,"How to debug or inspect the health of Feetech servos in so101 setup?","DIMARIA123",3,"","",0,"2025-06-16","2026-02-25"],[2,1318,"How to use my own dataset to train pi0 or smolVLA","xliu0105",2,"","",0,"2025-06-16","2025-10-17"],[2,1317,"KeyError: \"Column observation.image not in the dataset. Current columns in the dataset: ['observation.state', ","hzxie",3,"","",0,"2025-06-16","2026-02-25"],[2,1316,"[Question] SmolVLA LIBERO / MetaWorld evaluation","tykim0507",2,"","",0,"2025-06-16","2025-10-17"],[2,1305,"TypeError: unhashable type: 'list'","hzxie",3,"","",0,"2025-06-15","2026-02-24"],[2,1296,"calibration failed : ValueError: Magnitude 2114 exceeds 2047 (max for sign_bit_index=11)","FaboNo",2,"","",0,"2025-06-14","2026-04-14"],[2,1292,"issue in walkthrough [https://huggingface.github.io/lerobot/#walkthrough]","balaji-at-bfl",3,"","",0,"2025-06-14","2025-12-26"],[2,1291,"2025, Can we support multi-gpu training?","lucasjinreal",2,"","",0,"2025-06-14","2025-10-17"],[2,1288,"lerobot/scripts/rl/gym_manipulator.py not found","snknitheesh",2,"","",0,"2025-06-13","2025-06-13"],[2,1282,"Data Loading Performance Bottleneck: set_format(type='torch') Incompatible with select/stack, Unable to Accele","elapstjtl",3,"","",0,"2025-06-13","2026-02-24"],[2,1281,"Large scale training","richardrl",3,"","",0,"2025-06-13","2026-02-24"],[2,1279,"Insert nuts for gripper are not explained and missing from the parts sourcing table","commanderfun",3,"","",0,"2025-06-12","2025-12-26"],[2,1271,"Cannot run evaluation with real robot. (with `lerobot/record.py`)","kaaatsu32329",2,"","",0,"2025-06-12","2025-06-13"],[2,1270,"smolvla not work！","ZhuYaoHui1998",3,"","",0,"2025-06-12","2026-02-24"],[2,1269,"resume problem","kelo020304",3,"","",0,"2025-06-12","2026-02-24"],[2,1267,"Multi-card parallel training","xander-2077",2,"","",0,"2025-06-11","2025-06-12"],[2,1262,"use smolVLA, How to know the current task is completed","haoyankai",3,"","",0,"2025-06-11","2026-02-24"],[2,1259,"Question/Potential Bug about SO100 use_degrees mode for sim2real/real2sim alignment","StoneT2000",2,"","",0,"2025-06-11","2025-07-04"],[2,1258,"Leader Servo Numbering different from script to documentation","FaboNo",3,"","",0,"2025-06-10","2026-02-24"],[2,1257,"SMOLVLA Policy: ImportError: cannot import name 'OBS_ROBOT' from 'lerobot.common.constants'","krohling",2,"","",0,"2025-06-10","2025-06-10"],[2,1255,"`wrist_roll` limit overflow causes wrap-around and initialization errors","GaryOAO",2,"","",0,"2025-06-10","2025-06-22"],[2,1254,"[Feature Proposal] Planning a new user friendly simulation environment for new task and data collection","Bigenlight",3,"","",0,"2025-06-10","2026-02-24"],[2,1252,"Failed to sync read 'Present_Position' on ids=[2,3,4,6]after 1 tries. [TxRxResult] There is no status packet","huazai665",4,"","",0,"2025-06-10",""],[2,1251,"where is async inference","JuilieZ",2,"","",0,"2025-06-10","2025-06-30"],[2,1249,"problem when add_frame due to validate_feature_image_or_video","feixyz10",3,"","",0,"2025-06-10","2026-02-24"],[2,1245,"Press Enter does not work on Windows in calibration process","B-Rousseau",2,"","",0,"2025-06-09","2026-07-31"],[2,1244,"Cannot set up servo sts3215","NotNANtoN",3,"","",0,"2025-06-09","2026-07-31"],[2,1243,"Can't set up wheels IDS for lekiwi","Baptiste-le-Beaudry",3,"","",0,"2025-06-09","2026-02-26"],[2,1242,"SmolVLA Gym Simulation - Release?","Jaykumaran",2,"","",0,"2025-06-09","2025-10-17"],[2,1241,"Convert PI0Fast model to torch model","Liuzeyi25",2,"","",0,"2025-06-09","2025-10-08"],[2,1240,"PI0FAST calls a non-existent config param","xander-2077",2,"","",0,"2025-06-09","2025-06-09"],[2,1239,"smolvla model not working properly","yuyoujiang",2,"","",0,"2025-06-09","2025-06-10"],[2,1238,"FileNotFoundError: [Errno 2] No such file or directory: '<my own path>/aloha_sim_dataset/aloha_sim_insertion_h","xuanyaoming",3,"","",0,"2025-06-09","2026-02-24"],[2,1237,"scservo_sdk's version issue??","lanseyege",3,"","",0,"2025-06-09","2026-02-01"],[2,1235,"running visualize_dataset in macOS docker container - WARNING:huggingface_hub.utils._http:HTTP Error 429 throw","akshaygulabrao",2,"","",0,"2025-06-09","2025-06-10"],[2,1234,"smolvla training bugs on 3090","Loki-Lu",3,"","",0,"2025-06-09","2026-02-17"],[2,1233,"smolvla run error","lucasjinreal",3,"","",0,"2025-06-08","2026-01-16"],[2,1226,"404 Not Found","luk-e158",2,"","",0,"2025-06-07","2025-06-08"],[2,1223,"smolvla  introduce an asynchronous inference stack decoupling perception and action prediction?","zmf2022",2,"","",0,"2025-06-07","2025-06-08"],[2,1215,"[Feature Request] Offer unwrapped (draccus) versions of all entry points","jccalvojackson",3,"","",0,"2025-06-06","2026-02-24"],[2,1214,"Create AST to display the structure of data","xb-fight",3,"","",0,"2025-06-06","2026-02-24"],[2,1213,"No base information on aloha_mobile datasets","ssangjunpark",3,"","",0,"2025-06-06","2025-12-17"],[2,1210,"problem about smovla paper table 7","LukeLIN-web",2,"","",0,"2025-06-05","2025-06-10"],[2,1209,"Tutorials/Examples are gone","ml2s",3,"","",0,"2025-06-05","2025-12-22"],[2,1204,"example\\2_calculate_validation_loss.py","Mario-kkk",3,"","",0,"2025-06-05","2026-01-05"],[2,1203,"Could you please upload the config.json file for smolvla?","Pandapan01",0,"danaaubakirova","#1256",1,"2025-06-05","2025-06-11"],[2,1202,"when release smolvla 2.25B?","zmf2022",2,"","",0,"2025-06-05","2025-06-10"],[2,1199,"`find_motors_bus_port.py` not working with a hub on mac","jccalvojackson",2,"","",0,"2025-06-04","2025-06-04"],[2,1195,"ros2_control support","baaluidnrey",2,"","",0,"2025-06-03","2026-10-21"],[2,1189,"When will SmolVLA be released?","hairuoliu1",2,"","",0,"2025-06-03","2025-06-04"],[2,1186,"Update recording process to accommodate follower-only dataset creation","brainwavecoder9",2,"","",0,"2025-06-02","2025-06-02"],[2,1180,"dataset training","bruce577",2,"","",0,"2025-06-01","2025-10-08"],[2,1179,"TimeoutError: Timed out waiting for async_read() to start.","rancheng",2,"","",0,"2025-06-01","2025-06-01"],[3,15195,"FLEX attention silently ignores is_causal=True when no mask is supplied","davrens101",4,"","",0,"2026-11-06",""],[3,15193,"HookRegistry.remove_hook drops the next hook's new_forward (breaks FirstBlockCache with group offloading)","theontho",4,"","",0,"2026-11-06",""],[3,15185,"Per-token FlowMatch Euler steps use the opposite velocity sign","davrens101",4,"","",0,"2026-11-05",""],[3,15182,"`flash_varlen` raises when requesting log-sum-exp","0z5a",4,"","",0,"2026-11-04",""],[3,15180,"`flash_4_hub` does not forward `return_lse` to its kernel","0z5a",4,"","",0,"2026-11-04",""],[3,15178,"Gradient clipping is a no-op in `train_text_to_image_lora.py` and `train_lcm_distill_lora_sdxl.py`","gorkemgul",1,"gorkemgul","#15184",0,"2026-11-04","2026-11-05"],[3,15176,"img2img: outputs paired with the wrong init image","Arnav66692",4,"","",0,"2026-11-03",""],[3,15170,"Cosmos 3 docs: FP8 snippet fails at the first W8A8 step without importing ModelOpt's FP8 GEMM backend","piotrm-nvidia",4,"","",0,"2026-11-02",""],[3,15169,"Cosmos 3 FP8: wrapping pipe.transformer (e.g. torch.compile) silently disables the checkpoint's W8A16 step pol","piotrm-nvidia",4,"","",0,"2026-11-02",""],[3,15167,"UniPC and DPM-Solver with use_flow_sigmas condition the model on integer timesteps while integrating with floa","LiRunGuo",4,"","",0,"2026-11-02",""],[3,15166,"[training scripts] `--logit_mean` biases timesteps in the opposite direction to the SD3 paper (positive m over","Andreas-Hum",4,"","",0,"2026-11-01",""],[3,15163,"Flux and LTX pipelines do not forward the generator to stochastic scheduler steps","JohnLyu2",4,"","",0,"2026-10-31",""],[3,15162,"No LoRA fine-tuning example for RFD4-Proteina","kierandidi",4,"","",0,"2026-10-31",""],[3,15161,"DreamLite: disabling attention slicing replaces the GQA processor and breaks inference","gorkadabo1",4,"","",0,"2026-10-31",""],[3,15156,"Kandinsky 6: add context-parallel (_cp_plan) support to Kandinsky6Transformer3DModel","vladislav-vasilenko",4,"","",0,"2026-10-30",""],[3,15146,"Clarify set_adapters() adapter_weights documentation and type hints","gorkadabo1",1,"gorkadabo1","#15157",0,"2026-10-29","2026-11-04"],[3,15145,"flux_lora_quantization example: compute_embeddings.py crashes on current transformers","kahlun",1,"kahlun","#15144",0,"2026-10-29","2026-10-31"],[3,15139,"DDPM schedulers add fresh noise on terminal transitions when the final timestep is nonzero","surgan12",4,"","",0,"2026-10-27",""],[3,15134,"[Feature Request] Add DiCache for sample-adaptive caching in FLUX transformers","BlazeSR10hawk",4,"","",0,"2026-10-26",""],[3,15132,"[Feature] Reuse Ulysses shard lengths across MiniMax-H3 blocks","YaoleWu",4,"","",0,"2026-10-26",""],[3,15131,"Flux2KleinInpaintPipeline: `padding_mask_crop` pastes a center crop of the result instead of resizing it","hkang4628",4,"","",0,"2026-10-26",""],[3,15127,"set_adapters applies component weights only to the first adapter when broadcasting a dictionary","gorkadabo1",1,"gorkadabo1","#15128",0,"2026-10-24","2026-10-26"],[3,15126,"@dataclass on quant configs breaks __eq__/__repr__/hashing, and to_json_string() raises for most configs","OnePunchMonk",3,"","",0,"2026-10-24","2026-11-01"],[3,15125,"Nunchaku AWQ config accepts any group_size, but the kernel only supports 64","OnePunchMonk",3,"","",0,"2026-10-24","2026-11-01"],[3,15124,"Quantizer configs are mutated in place, leaking modules_to_not_convert across loads","OnePunchMonk",3,"","",0,"2026-10-24","2026-11-01"],[3,15123,"from_single_file forces pre_quantized=True, breaking bnb/TorchAO/Quanto single-file loading","OnePunchMonk",3,"","",0,"2026-10-24","2026-11-01"],[3,15122,"NVIDIAModelOptConfig.to_dict() crashes from_pretrained/save_pretrained when forward_loop is set","OnePunchMonk",3,"","",0,"2026-10-24","2026-11-01"],[3,15120,"AutoRound: user-passed quantization_config silently overwrites checkpoint's quant settings","OnePunchMonk",3,"","",0,"2026-10-24","2026-11-01"],[3,15119,"MiniMax-H3 creates full-clip FP32 temporaries after VAE decoding","YaoleWu",4,"","",0,"2026-10-24",""],[3,15115,"Training examples crash on --revision in expired non-EMA deprecation blocks","SeverinVisionary",0,"A-Raphie","#15118",0,"2026-10-23","2026-10-24"],[3,15113,"[Model integration] Add Hy-Nomas support to Diffusers","jiaxiangc",4,"","",0,"2026-10-23",""],[3,15112,"Text-to-image example updates EMA when AMP skips the optimizer step","SeverinVisionary",4,"","",0,"2026-10-23",""],[3,15108,"`LMSDiscreteScheduler` uses the wrong multistep order when sampling starts part-way through the schedule (img2","mottopanikeiku",4,"","",0,"2026-10-22",""],[3,15104,"`encode_video` truncates a fractional `fps`, so the video runs longer than its audio","devanshc777",4,"","",0,"2026-10-22",""],[3,15101,"[modular] `ConditionalPipelineBlocks` should declare its own trigger inputs","yiyixuxu",4,"","",0,"2026-10-22",""],[3,15100,"[modular] conditional block triggers should be workflow inputs","yiyixuxu",4,"","",0,"2026-10-22",""],[3,15098,"load_video returns mixed-mode frames for GIFs, breaking preprocess_video and AnimateDiff video-to-video","nettanvirdev",4,"","",0,"2026-10-21",""],[3,15090,"DEIS coefficient precision causes increasing error as inference steps increase","Robertboy18",4,"","",0,"2026-10-21",""],[3,15087,"lr_scheduler step bug in examples code","weixiaohao",4,"","",0,"2026-10-20",""],[3,15085,"Error \"cannot import name 'FqnToConfig' from 'torchao.quantization'\" when importing the diffusers library on g","WiNE-iNEFF",0,"usmanmateen","#15095",0,"2026-10-19","2026-10-21"],[3,15083,"Qwen-Image 2.1 Turbo: VAE tiling introduces vertical streaks at 2048x2048 on NVIDIA/CUDA","Religious-J",4,"","",0,"2026-10-19",""],[3,15071,"Follow-up to #14737: drop `output_type=\"latent\"` from modular decode blocks","yiyixuxu",4,"","",0,"2026-10-17",""],[3,15068,"FLUX.3: match the released BFL reference image sizing","yiyixuxu",1,"yiyixuxu","#15073",1,"2026-10-16","2026-10-19"],[3,15067,"Warn when a copy of `pipeline.blocks` is modified","yiyixuxu",4,"","",0,"2026-10-16",""],[3,15065,"Quick survey: your experience as a PyTorch Ecosystem maintainer","michellearoth",4,"","",0,"2026-10-16",""],[3,15056,"Third-order DPM++ multistep coefficients fail a quadratic forcing test","Robertboy18",4,"","",0,"2026-10-15",""],[3,15044,"[LTX-2] Modular pipelines: let callers skip the audio decode for video-only outputs","christopher5106",4,"","",0,"2026-10-15",""],[3,15043,"Warn when add_noise runs before set_timesteps","TejasS1233",4,"","",0,"2026-10-15",""],[3,15040,"[Wan2.2 VAE, Qwen-Image-2.1 VAE] Redundant x.clone() in residual up/down blocks keeps an extra activation aliv","luca-888",4,"","",0,"2026-10-15",""],[3,15037,"LTX2 HDR reference preprocessing applies an unintended stride-aligned resize","Shenrui-Ma",4,"","",0,"2026-10-14",""],[3,15027,"Qwen Image 2.1 with torch.compile triggers a recompile on each run with varying image size","geekuillaume",4,"","",0,"2026-10-14",""],[3,15025,"SDXL invisible watermark: apply_watermark encodes float input, and about a third of outputs are not detected b","CyberNativeAI",4,"","",0,"2026-10-14",""],[3,15022,"Modular pipelines: warn when a `denoiser_input_fields` entry is not passed to the transformer","yiyixuxu",4,"","",0,"2026-10-13",""],[3,15021,"add_noise() before set_timesteps() uses the wrong noise level (DPMSolverMultistep, DPMSolverSinglestep, UniPC,","FilipLangiewicz",4,"","",0,"2026-10-13",""],[3,15020,"[LoRA] Text-encoder LoRA loading assumes CLIP module names and fails for T5 / SmolLM3 / ChatGLM text encoders","Sergasgr",4,"","",0,"2026-10-12",""],[3,15018,"offload_state_dict=True leaves CPU model weights on meta after loading","Shenrui-Ma",4,"","",0,"2026-10-12",""],[3,15014,"gfx1201 (RDNA4) SageAttention attention backend: aiter package or Hub kernel?","Shan2L",1,"Shan2L","#15060",0,"2026-10-12","2026-10-29"],[3,15011,"VaeImageProcessor.denormalize() raises AttributeError for NumPy inputs","BlazeSR10hawk",4,"","",0,"2026-10-12",""],[3,15009,"[modular] `SequentialPipelineBlocks` gets `required` wrong in both `inputs` and `required_inputs`","yiyixuxu",4,"","",0,"2026-10-11",""],[3,15007,"[Z-Image] joint_attention_kwargs={\"scale\": ...} is silently ignored when loading LoRA weights","Sergasgr",1,"Sergasgr","#15015",0,"2026-10-11","2026-10-13"],[3,15006,"WanVideoToVideoPipeline: \"No cache context is set\" when caching is enabled","fusheng-ji",4,"","",0,"2026-10-11",""],[3,15005,"Add TripoSplat","hlky",2,"","",0,"2026-10-11","2026-10-21"],[3,15002,"KDPM2DiscreteScheduler produces NaN on final step due to zero-sigma interpolation","riparanta",4,"","",0,"2026-10-10",""],[3,14994,"load_video() leaks temporary files when remote video decoding fails","BlazeSR10hawk",4,"","",0,"2026-10-09",""],[3,14993,"WanCausalConv3d copies every activation with F.pad. Let the conv do the spatial padding.","will-rice",4,"","",0,"2026-10-08",""],[3,14992,"Add an optional Transformer Engine FP8 training example for FLUX","sbhavani",4,"","",0,"2026-10-08",""],[3,14985,"MiniMax-H3: _decode_video_file turns ±90° display-rotated videos upside down","hsbcelebi",3,"","",0,"2026-10-07","2026-10-08"],[3,14981,"LTX-2.5 SDR-To-HDR IC-LoRA: how should it be structured?","christopher5106",4,"","",0,"2026-10-07",""],[3,14980,"LTX-2: decouple the RoPE time base from the playback frame rate (Slow-Motion-Control LoRA)","christopher5106",1,"christopher5106","#14967",0,"2026-10-07","2026-10-21"],[3,14970,"Per-module alphas of kohya text encoder LoRAs collapse to the most common alpha (alpha_pattern keys stop at th","andre-menutole",4,"","",0,"2026-10-06",""],[3,14955,"get_active_adapters() returns empty for active text-encoder-only LoRA adapters","BlazeSR10hawk",0,"Subodh-17","#14961",0,"2026-10-06","2026-10-08"],[3,14953,"PNDMScheduler.set_timesteps crashes with NumPy broadcasting error for fewer than 4 inference steps","BlazeSR10hawk",4,"","",0,"2026-10-06",""],[3,14952,"`ModelMixin.from_pretrained` on a sharded checkpoint fails under `HF_HUB_OFFLINE=1` even when fully cached (`_","quinncheong",4,"","",0,"2026-10-05",""],[3,14939,"Propose translation to romanian for the diffusers hugging face course","pop123-ux",4,"","",0,"2026-10-04",""],[3,14938,"Qwen-Image LoRAs in lora_A/lora_B format lose their .alpha and load at the wrong strength","dkackman",0,"dcaayushd","#15016",0,"2026-10-04","2026-10-22"],[3,14937,"Qwen-Image-2.1 LoRAs with a fused img_mlp.gate_up load partially without an error","dkackman",4,"","",0,"2026-10-04",""],[3,14935,"QwenImage21: model.to(bf16) (as in train_dreambooth_lora_qwenimage21.py) rounds the timestep-embedding freqs b","tritsystem",4,"","",0,"2026-10-04",""],[3,14934,"StableAudio3DiTModel: loading with torch_dtype=bf16/fp16 collapses RoPE positions (arange built in inv_freq dt","tritsystem",4,"","",0,"2026-10-04",""],[3,14930,"`from_single_file()` support for AnimaTextConditioner","mingyi456",4,"","",0,"2026-10-03",""],[3,14929,"`CosmosTransformer3DModel.from_single_file()` does not work with the `model.diffusion_model.` prefix","mingyi456",4,"","",0,"2026-10-03",""],[3,14928,"Qwen-Image-2.1 VAE tiling: RGB/alpha seams and a full-context decode workaround on ROCm","nabe2030",4,"","",0,"2026-10-03",""],[3,14927,"Windows: from_single_file original_config and fp16_safetensors crash with UnicodeDecodeError under non-UTF-8 l","dltsum",4,"","",0,"2026-10-03",""],[3,14924,"Add a Hugging Face Jobs guide to the docs","davanstrien",0,"stevhliu","#14926",1,"2026-10-02","2026-10-06"],[3,14922,"Enable Taylorseer for all Flux Pipelines","Reis-McMillan",4,"","",0,"2026-10-01",""],[3,14917,"fal FLUX.1 Kontext LoRAs with unprefixed embedder keys fail in _convert_fal_kontext_lora_to_diffusers","christopher5106",4,"","",0,"2026-09-30",""],[3,14907,"CFG cutoff callbacks keep a single embedding row, so they break for num_images_per_prompt > 1 or several promp","JoeyTan21",4,"","",0,"2026-09-29",""],[3,14903,"Required weight pass on a singlefile","xalteropsx",4,"","",0,"2026-09-29",""],[3,14901,"numpy_to_pil removes singleton spatial dimensions from grayscale images","Afloat16",4,"","",0,"2026-09-29",""],[3,14896,"Legacy `final_offload_hook` checks are dead in 14 pipelines; `KandinskyPriorPipeline` skips `maybe_free_model_","kv-248",4,"","",0,"2026-09-28",""],[3,14892,"`Cosmos2VideoToWorldPipeline` conditions on the entire input video — no `num_conditional_frames`, so a full cl","rebel-seinpark",4,"","",0,"2026-09-28",""],[3,14890,"Converted sigma schedules with final_sigmas_type=\"sigma_min\" repeat the last sigma, and the third-order and he","Nicholas022400701",4,"","",0,"2026-09-28",""],[3,14889,"QwenImage21 FlexAttention mask fails to compile on MPS because of boolean ~","korellas",4,"","",0,"2026-09-28",""],[3,14888,"UniPCMultistepScheduler fails in torch.linalg.solve under a float64 default dtype: the unit entry of rks takes","Nicholas022400701",4,"","",0,"2026-09-27",""],[3,14887,"UniPCMultistepScheduler returns nan with solver_type=\"bh1\", and inf or nan with lower_order_final=False or pre","Nicholas022400701",4,"","",0,"2026-09-27",""],[3,14886,"[docs] 22 dead external links across the English docs","pratikgx",0,"mlzerotoheroyoutube","#15174",0,"2026-09-27","2026-11-04"],[3,14882,"MiniMax-H3 video decode only enables fp16 autocast on CUDA, so the block silently runs in fp32 on other accele","li-lizhe",2,"","",0,"2026-09-27","2026-09-28"],[3,14881,"WanAnimatePipeline.get_i2v_mask() defaults device to \"cuda\", which raises on non-CUDA accelerators (NPU/XPU/MP","li-lizhe",4,"","",0,"2026-09-27",""],[3,14880,"Cosmos denoising loops sync with the host every step","adrianrfreedman",4,"","",0,"2026-09-27",""],[3,14879,"Five Z-Image pipelines miss the precomputed-t_norms DtoH-sync fix from ZImagePipeline","adrianrfreedman",4,"","",0,"2026-09-27",""],[3,14878,"`Pipeline.to()` offload guards only cover `cuda`/`xpu`: Ascend NPU misses the sequential-offload conflict erro","li-lizhe",4,"","",0,"2026-09-26",""],[3,14877,"`apply_group_offloading(use_stream=True)` raises on Ascend NPU: no stream branch for `torch.npu`","li-lizhe",4,"","",0,"2026-09-26",""],[3,14876,"Z-Image LoRAs with only fused attention_qkv + attention_out keys are dropped, then fail on leftover alpha keys","christopher5106",1,"christopher5106","#14875",0,"2026-09-26","2026-09-28"],[3,14873,"Ming Image 0.1 Design","alerikaisattera",4,"","",0,"2026-09-25",""],[3,14871,"[Feature] Support backward for Ulysses Anything attention","NancyFyong",4,"","",0,"2026-09-25",""],[3,14870,"MiniMax H3 transformer does not expose attention masks for padded joint sequences","NancyFyong",4,"","",0,"2026-09-25",""],[3,14864,"IndexError when preprocessing an empty image or video list","MohammadHijjawi97",4,"","",0,"2026-09-24",""],[3,14860,"rank_dict.values()  list index out of range in diffuser v0.40","xalteropsx",0,"BeyondBirthday07","#14872",0,"2026-09-23","2026-09-28"],[3,14859,"QwenImage21Pipeline on MPS: every edit after the first in the same process returns all-NaN at output_resolutio","ck71",4,"","",0,"2026-09-23",""],[3,14858,"QwenImage21Pipeline: editing an image the model generated degrades at output_resolution=1024 regardless of see","ck71",4,"","",0,"2026-09-23",""],[3,14845,"Add SpanSynth-Edit for MIDI-guided music synthesis and editing","mimbres",3,"","",0,"2026-09-23","2026-09-23"],[3,14837,"Windows: check_ai.py fails decoding UTF-8 guides with the default locale","dlowzzxx",4,"","",0,"2026-09-22",""],[3,14833,"Enable KV cache offloading for a better speed-memory trade-off","sayakpaul",4,"","",0,"2026-09-22",""],[3,14832,"[Qwen-Image 2.1] Modular pipeline","SamuelTallet",4,"","",0,"2026-09-21",""],[3,14829,"[Qwen-Image 2.1] TaylorSeer cache crashes with the default KV cache (feature shape changes after the prefill s","hide3tu",1,"hide3tu","#14831",0,"2026-09-21","2026-09-25"],[3,14824,"QwenImage21Pipeline: image editing ignores the instruction and returns a haloed near-copy at output_resolution","xocialize",2,"","",0,"2026-09-21","2026-09-22"],[3,14821,"Qwen Image 2.1 transformer is incompatible with `torch.compile`","abel1502",4,"","",0,"2026-09-20",""],[3,14816,"PNDMScheduler combines raw v_prediction outputs in PRK and PLMS instead of noise predictions","Nicholas022400701",4,"","",0,"2026-09-20",""],[3,14814,"How to generate white background images instead of colorful?","emadyounan",4,"","",0,"2026-09-19",""],[3,14811,"A sharded .bin checkpoint can't be loaded with the default from_pretrained (use_safetensors=None)","aminmiral",4,"","",0,"2026-09-18",""],[3,14803,"53 messages in src/diffusers are missing the f-string prefix (safety-checker ValueError in 46 pipelines)","David-Wu1119",4,"","",0,"2026-09-17",""],[3,14794,"TangentialClassifierFreeGuidance.is_conditional reads an attribute that does not exist","ojassharma7",4,"","",0,"2026-09-16",""],[3,14784,"Opt-in weight integrity verification at load time (pinned SHA-256 hashes / manifest)","UniversePeak",4,"","",0,"2026-09-15",""],[3,14782,"text encoder not working","xalteropsx",2,"","",0,"2026-09-15","2026-09-15"],[3,14780,"Deprecating `Siglip2ImageProcessorFast`","ariG23498",1,"ariG23498","#14781",1,"2026-09-15","2026-10-05"],[3,14779,"MLU backend utilities fail despite MLU device detection","SeptPonts",4,"","",0,"2026-09-15",""],[3,14777,"CSOAI — Diffusers interop: generative governance","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[3,14776,"CSOAI — Diffusers interop: generative governance","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[3,14775,"CSOAI — Diffusers AI governance generative","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[3,14773,"LTX2 import fails with Transformers 5.5.0 due to annotation-only Gemma4Unified dependency","Imagineer99",4,"","",0,"2026-09-14",""],[3,14770,"[Bug / Optimization] Make AutoencoderKLWan stateless to fix torch.compile compatibility and enable 1.45x speed","NotNANtoN",4,"","",0,"2026-09-14",""],[3,14769,"save_pretrained(safe_serialization=False) leaves the old safetensors checkpoint behind, and from_pretrained lo","aminmiral",4,"","",0,"2026-09-14",""],[3,14768,"`Cosmos2VideoToWorldPipeline` applies FPS RoPE modulation that the released Cosmos-Predict2 checkpoints disabl","rebel-seinpark",4,"","",0,"2026-09-14",""],[3,14761,"Add IP-Adapter support to StableDiffusionXLInstructPix2PixPipeline","vivi27x",4,"","",0,"2026-09-12",""],[3,14759,"Pipeline properties that read attributes __call__ never sets (SD3 skip_guidance_layers, LEdits++ XL guidance_s","yupengtang",4,"","",0,"2026-09-12",""],[3,14749,"move `LTX2TextConnectors` into `models/condition_embedders/`","yiyixuxu",4,"","",0,"2026-09-09",""],[3,14746,"MiniMax-H3 VAE decoder loads in fp32 but CUDA decode runs it in fp16","asomoza",1,"asomoza","#14754",1,"2026-09-09","2026-09-28"],[3,14745,"[Community script] CachedSearch: cheaper best-of-N for video pipelines","shreshthsaini",1,"shreshthsaini","#14735",0,"2026-09-09","2026-09-09"],[3,14744,"[Community] Rectified-CFG++ pipelines for SD3/3.5, Flux, Lumina 2, Qwen-Image, and Wan","shreshthsaini",4,"","",0,"2026-09-09",""],[3,14732,"Two docstrings reference names the code does not have (peturbation_mask, source_image)","iridescentWen",1,"iridescentWen","#14733",0,"2026-09-08","2026-09-14"],[3,14730,"[modular] move unpack out of the decoder blocks for flux, krea2 and ltx","yiyixuxu",0,"akshan-main","#14737",0,"2026-09-07","2026-10-20"],[3,14729,"AttentionModuleMixin.set_attention_slice references two attributes that do not exist","yupengtang",4,"","",0,"2026-09-07",""],[3,14723,"Three docstrings type a boolean parameter as str","iridescentWen",4,"","",0,"2026-09-07",""],[3,14722,"[Proposal] Vectorize tiled VAE blending while preserving rounding and CUDA graph replay","ShikeChen01",3,"","",0,"2026-09-06","2026-09-06"],[3,14721,"Community pipeline: rerank Stable Diffusion candidates by CLIP score","AmirF194",4,"","",0,"2026-09-06",""],[3,14719,"save_pretrained can delete shards from another variant and leave stale indexes","Pdevadiga45",0,"AloysJehwin","#14844",0,"2026-09-06","2026-10-29"],[3,14711,"[zh] Translating docs to Simplified Chinese","JasonYeYuhe",4,"","",0,"2026-09-05",""],[3,14705,"[Quantization] Add support for Comfy Quants backend","DN6",4,"","",0,"2026-09-04",""],[3,14700,"Add VideoDeltaNet-H3 Model","haochengxi",4,"","",0,"2026-09-04",""],[3,14699,"Fix skipped `test_model_xattn_padding` test by updating mask padding logic","PrakshaaleJain",4,"","",0,"2026-09-03",""],[3,14691,"GGUF: enable_sequential_cpu_offload() raises KeyError: None — any GGUF model larger than VRAM is unrunnable","EigenAx2Pi",4,"","",0,"2026-09-02",""],[3,14668,"Chroma pipeline docstrings state defaults that differ from the signature","iridescentWen",1,"iridescentWen","#14578",0,"2026-09-01","2026-09-01"],[3,14667,"`FqnToConfig` raises inside `TorchAoConfig`","Beinsezii",0,"sayakpaul","#14686",1,"2026-09-01","2026-09-15"],[3,14666,"Missing-dependency error suggests `pip install pytorch_retinaface`, which is not a PyPI package","shmulc8",4,"","",0,"2026-08-31",""],[3,14665,"Missing-dependency error suggests `pip install pytorch_retinaface`, which is not a PyPI package","shmulik-apiiro",2,"","",0,"2026-08-31","2026-08-31"],[3,14649,"`guidance_scale=1.0` no longer disables CFG in LTX-2 pipelines","rootonchair",4,"","",0,"2026-08-30",""],[3,14644,"Proposal: tutorial for synchronized ControlNet pair augmentation with AlbumentationsX","ternaus",4,"","",0,"2026-08-28",""],[3,14643,"Bug: AttributeError in TangentialCFG.is_conditional and NameError in FrequencyDecoupledGuidance.forward","Ricardo-M-L",2,"","",0,"2026-08-28","2026-08-30"],[3,14642,"Bug: randn_tensor MPS info-log suppression never triggers (compares torch.device to string)","Ricardo-M-L",2,"","",0,"2026-08-28","2026-08-30"],[3,14641,"Bug: is_torch_xla_version() checks function truthiness instead of _torch_xla_available flag","Ricardo-M-L",2,"","",0,"2026-08-28","2026-08-30"],[3,14640,"ModularPipeline: loading from a local directory still resolves components against the hub id stored in modular","4suino",2,"","",0,"2026-08-28","2026-09-04"],[3,14639,"MiniMax-H3: float64 rotary position grid cannot be moved to MPS (`Cannot convert a MPS Tensor to float64`)","4suino",0,"SammyTourani","#14767",0,"2026-08-28","2026-09-26"],[3,14634,"give state dict creation its own method in IP-adapter pipeline tests","sayakpaul",4,"","",0,"2026-08-28",""],[3,14624,"Kolors text encoder carries an unreachable torch<2 attention path built on the MPS-unsafe baddbmm(empty, beta=","RudraMantri123",4,"","",0,"2026-08-27",""],[3,14618,"If I add `\"unconditional_transformer\"` and remove `\"text_encoder\"` from the hardcoded component list in `test_","sayakpaul",0,"dg845","#14635",0,"2026-08-27","2026-09-02"],[3,14617,"It looks like most other pipelines such as Flux 2 use the following `callback_kwargs` pattern instead of a dic","sayakpaul",0,"dg845","#14621",0,"2026-08-27","2026-08-28"],[3,14616,"document output_type=\"pt\" in the agent docs","sayakpaul",4,"","",0,"2026-08-27",""],[3,14595,"Provenance/licensing clarification for controlnet-canny-sdxl-1.0","Mfundo360",4,"","",0,"2026-08-24",""],[3,14594,"sdn ui ux control panel","beparykamrul-dev",2,"","",0,"2026-08-24","2026-08-26"],[3,14593,"sdn_wan","beparykamrul-dev",2,"","",0,"2026-08-24","2026-08-26"],[3,14592,"sdn w","beparykamrul-dev",2,"","",0,"2026-08-24","2026-08-26"],[3,14591,"Update remaining pipeline docstrings from `torch_dtype` to `dtype`","iMountTai",4,"","",0,"2026-08-24",""],[3,14590,"RMSNorm crashes on NPU when elementwise_affine=False (weight=None): npu_rms_norm requires a real gamma tensor","xucqX",2,"","",0,"2026-08-24","2026-10-01"],[3,14586,"UnCLIPScheduler.set_timesteps crashes (ZeroDivisionError) for 1 step and silently returns an empty schedule fo","kyo-zzz",4,"","",0,"2026-08-24",""],[3,14575,"diffusers docs have too many `cuda` hardcoding, this is unfriendly to device-agnostic","kaixuanliu",1,"kaixuanliu","#14414",0,"2026-08-24","2026-08-25"],[3,14573,"Several FlowMatch pipelines miss the set_begin_index(0) DtoH-sync fix from #11696","shoemoney",1,"shoemoney","#14576",0,"2026-08-23","2026-09-08"],[3,14569,"fixed_large_log variance sampling returns NaNs","Robertboy18",4,"","",0,"2026-08-23",""],[3,14548,"train_lcm_distill_lora_sdxl.py: intermediate validation logs the un-adapted base model; saved LoRA mixes key c","Skyyyy0920",4,"","",0,"2026-08-21",""],[3,14533,"better reporting of errors when using TP","sayakpaul",0,"JingyaHuang","#14544",1,"2026-08-19","2026-10-01"],[3,14529,"The Diffusers MVP 🚀 — Ideas Track 💡","yiyixuxu",4,"","",0,"2026-08-18",""],[3,14524,"Add support for JoyAI-Video-Edit","feice-huang",4,"","",0,"2026-08-18",""],[3,14520,"Where should a training-free high-resolution hook for Flux (DyPE / spectral attention) live?","smellslikeml",4,"","",0,"2026-08-17",""],[3,14518,"[Krea-2] `enable_gqa` + `attn_mask` produces high usage of VRAM","asomoza",0,"yiyixuxu","#14523",1,"2026-08-17","2026-08-18"],[3,14487,"Missing trainer.py file in src/diffusers/trainer.py causing import error","bobrathbone",2,"","",0,"2026-08-16","2026-08-26"],[3,14486,"MiniMax Music 3 modular pipeline: 2.9x consumer-GPU speedup - findings that may be worth upstreaming","TheDutchRuler",4,"","",0,"2026-08-15",""],[3,14485,"[tests] fix parallelism tests from #14464","sayakpaul",1,"sayakpaul","#14505",1,"2026-08-15","2026-08-18"],[3,14482,"[Community pipeline] Echo-Memory overlay for Wan 2.1 1.3B","WayneJin0918",4,"","",0,"2026-08-15",""],[3,14479,"Coordinate non-interactive library messages through the project logger","wunianze666-netizen",1,"wunianze666-netizen","#14682",0,"2026-08-14","2026-09-11"],[3,14473,"Thoughts on maintaining our lightweight Docker container to reduce startup time for remote inference?","sayakpaul",2,"","",0,"2026-08-14","2026-09-29"],[3,14470,"InpaintProcessor.preprocess returns a bare tensor when no mask is passed, breaking its own 3-value contract","abhati27",2,"","",0,"2026-08-13","2026-10-08"],[3,14469,"[modular] `guider` should be an optional component for text encoder blocks","yiyixuxu",4,"","",0,"2026-08-13",""],[3,14461,"`FlowMatchEulerDiscreteScheduler.set_timesteps` silently ignores explicit `timesteps` since #14011 — intended ","codeAnqiang-ma",0,"yiyixuxu","#14956",1,"2026-08-13","2026-10-06"],[3,14460,"`register_to_config` mislabels positional `__init__` args as `_use_default_values`, so `from_config` round tri","codeAnqiang-ma",4,"","",0,"2026-08-13",""],[3,14451,"Modular pipelines: variant blocksets sharing a `model_name` init the wrong pipeline class","yiyixuxu",4,"","",0,"2026-08-12",""],[3,14449,"pipe.to(device) corrupts custom quantized (uint8 packed) module buffers","pupa3066",4,"","",0,"2026-08-12",""],[3,14442,"Request the UniPCMultistepScheduler to allow a customed timesteps","TheLovesOfLadyPurple",4,"","",0,"2026-08-11",""],[3,14438,"SDXL: enable_attention_slicing() + enable_model_cpu_offload() produces all-black images on MPS (Apple Silicon)","pupa3066",4,"","",0,"2026-08-10",""],[3,14433,"Title: Group offloading with use_stream=True fails on torchao int8 version=2 (pin_memory on CUDA qdata)","animede",4,"","",0,"2026-08-10",""],[3,14430,"[SD3 DreamBooth] Cached latents can be paired with the wrong per-image captions when shuffle=True","zimmoy374",4,"","",0,"2026-08-09",""],[3,14429,"VaeImageProcessorLDM3D ignores constructor configuration arguments","aswanth-07",4,"","",0,"2026-08-09",""],[3,14428,"VaeImageProcessorLDM3D.preprocess fails for batched NumPy RGB/depth inputs","aswanth-07",4,"","",0,"2026-08-09",""],[3,14423,"Breaking change for Minimax H3 in commits after 53e5b00","zouflain",3,"","",0,"2026-08-08","2026-08-08"],[3,14420,"[Feature Request] Add LoRA loading support for HunyuanVideo 1.5","thesayakmondal",4,"","",0,"2026-08-08",""],[3,14411,"Division by 0 bug when only one denoising step","kyle-hoffmeyer",4,"","",0,"2026-08-06",""],[3,14410,"MiniMax H3 unnecessarily swapped w1 in VAE","bghira",4,"","",0,"2026-08-06",""],[3,14405,"Ideogram4: allow the text conditioning projection to be hoisted out of the denoising loop","lsnchow",2,"","",0,"2026-08-06","2026-10-19"],[3,14403,"convrot for int8 and fp4","ghost",4,"","",0,"2026-08-06",""],[3,14397,"Native NPU attention incorrectly converts additive attention masks to boolean masks","mengchengTang",4,"","",0,"2026-08-05",""],[3,14389,"Modular: three tiny test fixtures cannot serve the new workflow tests","yiyixuxu",4,"","",0,"2026-08-05",""],[3,14388,"Modular: `required=True` is silently ignored when an `InputParam` carries a default","yiyixuxu",1,"yiyixuxu","#14521",1,"2026-08-05","2026-10-27"],[3,14382,"A problem on the inference using refiner","TheLovesOfLadyPurple",4,"","",0,"2026-08-04",""],[3,14380,"[Bug] Ascend NPU: RMSNorm crashes with elementwise_affine=False; _native_npu FA rejects [B, N, 1, Skv] masks (","mengchengTang",1,"mengchengTang","#14288",0,"2026-08-04","2026-09-21"],[3,14379,"MiniMax-H3: `references` argument ignored and documented multi-GPU example fails with CUDA device mismatch","jyx13121802323",4,"","",0,"2026-08-04",""],[3,14378,"[Feature Request] Add native ResilPhase cache support for FLUX and HunyuanVideo","zqc214",4,"","",0,"2026-08-04",""],[3,14377,"Issue Labeler fails on most bug reports: 'needs-env-info' label does not exist","4ktLuffy",4,"","",0,"2026-08-03",""],[3,14376,"Secret Leaks workflow fails on every new-branch push (6 of the last 30 runs)","4ktLuffy",4,"","",0,"2026-08-03",""],[3,14368,"IPNDMScheduler and KDPM2DiscreteScheduler return all-NaN on MPS","4ktLuffy",4,"","",0,"2026-08-03",""],[3,14367,"Helios schedulers raise `TypeError` on MPS: `float64` without the device guard","4ktLuffy",4,"","",0,"2026-08-03",""],[3,14366,"Public `embeddings.py` helpers raise `ValueError` with default arguments","4ktLuffy",4,"","",0,"2026-08-03",""],[3,14365,"Remove deprecated VQModel/VQEncoderOutput shim (past target version 0.31)","thesayakmondal",4,"","",0,"2026-08-03",""],[3,14363,"minimax run error currently!","k15201363625",4,"","",0,"2026-08-03",""],[3,14353,"Scheduler `step()` methods silently ignore documented sampling parameters (`s_churn`, `s_noise`, `eta`, `gener","4ktLuffy",4,"","",0,"2026-08-01",""],[3,14352,"Scheduler docstrings omit real parameters (`noise`, `mu`, `device`, `sigma*`) across 16 functions","4ktLuffy",1,"4ktLuffy","#14354",0,"2026-08-01","2026-08-01"],[3,14347,"[Bug] low_cpu_mem_usage=True is not thread-safe and can leak the global nn.Module.register_parameter patch","issue-ac",4,"","",0,"2026-07-31",""],[3,14346,"[Modular] support group offloading in auto offloading","yiyixuxu",4,"","",0,"2026-07-31",""],[3,14342,"cuDNN attention CP path consumes a bool attn_mask as an additive 0/1 bias","YangXu1990uiuc",4,"","",0,"2026-07-30",""],[3,14338,"cuDNN attention backward corrupts gradients: K/V transposed twice, Q once","YangXu1990uiuc",4,"","",0,"2026-07-30",""],[3,14328,"[modular] Migrate ComponentsManager offload hooks to ModelHook / HookRegistry","yiyixuxu",4,"","",0,"2026-07-29",""],[3,14325,"Two ONNX Docker images are missing from build_docker_images.yml, so their :latest tags have been frozen since ","kobihikri",4,"","",0,"2026-07-29",""],[3,14316,"LTX-2 `_cp_plan` appears copied from LTX-1 and fails under `enable_parallelism()` (crash + one silent-correctn","PaulFidika",4,"","",0,"2026-07-29",""],[3,14307,"[LTX-2 VAE] LTX2VideoUpBlock3d.conv_in is wired to the wrong widths","MHD-SMHD",4,"","",0,"2026-07-28",""],[3,14303,"Incompatibility with torchao >= 0.17.0 due to hardcoded legacy import paths in torchao_quantizer.py","lisi31415926",4,"","",0,"2026-07-27",""],[3,14293,"NEXUS","bharath24-cloud",4,"","",0,"2026-07-26",""],[3,14291,"[Bug] ChatGLMTokenizer special token property getters and setters are broken","GiGiKoneti",4,"","",0,"2026-07-25",""],[3,14290,"[Modular] Krea 2 image-to-image generation","SamuelTallet",4,"","",0,"2026-07-25",""],[3,14280,"[Cache] Add First Block Cache (FBCache) support for Flux 2 (Flux2Transformer2DModel)","sqhuang",4,"","",0,"2026-07-24",""],[3,14270,"Mage-Flow","alerikaisattera",4,"","",0,"2026-07-22",""],[3,14265,"Ring attention backward uses stale iteration-0 KV from ctx.saved_tensors — silent gradient corruption when tra","pthombre",0,"sayakpaul","#14274",1,"2026-07-21","2026-08-20"],[3,14255,"[tests] migrate all attention processor related stuff from pipeline to model-level testing","sayakpaul",1,"sayakpaul","#14701",1,"2026-07-21","2026-10-05"],[3,14249,"`set_attention_backend` on one model leaks globally to other models via the active-backend registry","sayakpaul",0,"raghulchandramouli","#14335",0,"2026-07-20","2026-11-03"],[3,14247,"PR GPU CI skips new-model GPU tests: Cosmos3 memory tests failed on an unrelated PR","sayakpaul",4,"","",0,"2026-07-20",""],[3,14243,"LTX2Pipeline: dynamic timestep shift `mu` is constant, ignores the actual sequence length","dxqb",4,"","",0,"2026-07-19",""],[3,14214,"unfuse_lora(components=[...]) desyncs _merged_adapters/num_fused_loras from actual per-component PEFT merge st","ErenAta16",0,"AloysJehwin","#14385",0,"2026-07-17","2026-08-08"],[3,14213,"Ancestral schedulers (EulerAncestral, KDPM2Ancestral) produce all-NaN output with beta_schedule=\"squaredcos_ca","chuenchen309",4,"","",0,"2026-07-17",""],[3,14209,"VaeImageProcessor.blur() docstring is missing blur_factor and has a copy-pasted description","RudrenduPaul",4,"","",0,"2026-07-17",""],[3,14206,"LDM3D rgblike_to_depthmap truncates the 16-bit depth map back to the 8-bit input dtype","chuenchen309",4,"","",0,"2026-07-16",""],[3,14195,"Offline mode rejects local LoRA files and directories without weight_name","fropych",1,"fropych","#14204",0,"2026-07-15","2026-07-16"],[3,14194,"[Bug] QwenImageEdit produces incorrect outputs with transformers >=5 due to missing mm_token_type_ids","wutachiang",4,"","",0,"2026-07-15",""],[3,14183,"AuraFlowPipeline: VAE dtype mismatch on pipeline reuse (latents cast skipped on 2nd call)","IvenHsu01",1,"IvenHsu01","#14184",0,"2026-07-14","2026-07-18"],[3,14175,"Path traversal / arbitrary out-of-directory file read via sharded-checkpoint `weight_map` in `_get_checkpoint_","geo-chen",0,"buffett0323","#14182",0,"2026-07-12","2026-07-17"],[3,14149,"Add AMD Quark support to Diffusers","amd-inechakr",4,"","",0,"2026-07-09",""],[3,14146,"`Krea2Pipeline` produces incorrect results on ROCm/gfx1201","hameerabbasi",2,"","",0,"2026-07-09","2026-07-09"],[3,14145,"`LTX2*Pipeline.from_single_file` cannot load LTX 2.3 (or 2.0) combined checkpoints: unmapped 2.3 keys leave me","cgu2022",4,"","",0,"2026-07-08",""],[3,14142,"DiffusionPipeline.download() on Windows silently skips all subfolder config.json files with huggingface_hub>=1","CalamitousFelicitousness",4,"","",0,"2026-07-08",""],[3,14139,"Non-hub varlen attention backends mispack keys for non-contiguous masks (missed in #14114/#14115)","qflen",4,"","",0,"2026-07-08",""],[3,14138,"Boogu-Image","JaDonghao",4,"","",0,"2026-07-08",""],[3,14135,"HookRegistry.register_hook breaks module forward signature","jloftin-nv",1,"jloftin-nv","#14285",0,"2026-07-07","2026-07-29"],[3,14132,"[modular] ensure branch-specific input defaults","yiyixuxu",1,"yiyixuxu","#14234",1,"2026-07-06","2026-07-31"],[3,14124,"[Version 0.39.0]AttributeError: 'CLIPTextModel' object has no attribute 'text_model' WHEN SET CKIP_SKIP NOT NO","CodingForAGI",4,"","",0,"2026-07-05",""],[3,14122,"`from_single_file()` support missing/incomplete for Ideogram4 and Krea2 transformers","dxqb",4,"","",0,"2026-07-05",""],[3,14114,"`_flash_3_varlen_hub` backend cannot handle non-contiguous mask","zhtmike",1,"zhtmike","#14115",0,"2026-07-03","2026-07-06"],[3,14104,"Flux-family attention: flash-attn and other backends fail in an autocast context","dxqb",4,"","",0,"2026-07-01",""],[3,14103,"Custom AttentionProcessor lost/fallback occurring after forward pass","abdullahejazjanjua",2,"","",0,"2026-07-01","2026-07-01"],[3,14097,"Regression (#13485) Broken TorchAO Compat","Beinsezii",4,"","",0,"2026-07-01",""],[3,14095,"SuperCompress: Free open-source LLM prompt compression - cut token costs by ~65%","arjunkshah",2,"","",0,"2026-06-30","2026-06-30"],[3,14094,"Reduce prompt costs with SuperCompress compression","arjunkshah",2,"","",0,"2026-06-30","2026-06-30"],[3,14090,"Fix model offloading and training tests + prevent examples timeout","GiGiKoneti",1,"GiGiKoneti","#14091",0,"2026-06-29","2026-07-04"],[3,14088,"Handling of .alpha in from-comfy converters","dxqb",4,"","",0,"2026-06-29",""],[3,14086,"Add SeFi-Image pipeline","jmliu206",4,"","",0,"2026-06-29",""],[3,14085,"Backport workflow-hardening fix (excessive-permissions) to 3 release branches","CharlieMCY",4,"","",0,"2026-06-29",""],[3,14076,"[Proposal]: Support LoRA loading for MotifVideo pipelines","Sainava",2,"","",0,"2026-06-26","2026-07-09"],[3,14072,"[Bug] Layerwise casting tests fail on MPS backend due to float8 unsupported dtype","GiGiKoneti",1,"GiGiKoneti","#14073",0,"2026-06-26","2026-06-29"],[3,14067,"uv pip install -e \".[test]\" fails on Python 3.12 due to incompatible numba/llvmlite resolution","gabe-engineers",0,"sayakpaul","#14078",1,"2026-06-25","2026-07-03"],[3,14065,"DreamLite-base fails to load due to unsupported legacy UNet block type names","ElectricGoal",1,"ElectricGoal","#14066",0,"2026-06-25","2026-06-26"],[3,14063,"Kandinsky5 pipeline does not support device_map=balanced","kaixuanliu",1,"kaixuanliu","b549ca9",0,"2026-06-25","2026-06-30"],[3,14060,"Support for LoRA / adapters in AceStepPipeline","Nib-git",0,"chenyangzhu1","#14193",0,"2026-06-24","2026-07-18"],[3,14057,"SD1.5 - pipeline_controlnet_img2img and pipeline_controlnet_inpaint are mixing variables \"image\" and \"control_","buddyforgothisemail",4,"","",0,"2026-06-23",""],[3,14056,"Modular pipeline: Krea 2","SamuelTallet",0,"Cedric-Perauer","#14083",0,"2026-06-23","2026-07-23"],[3,14049,"Add JoyAI-Image Edit Plus pipeline and model","tangyanf",1,"tangyanf","#14032",0,"2026-06-23","2026-07-08"],[3,14042,"Add Flux2KleinImg2ImgPipeline for FLUX.2 Klein","Sainava",2,"","",0,"2026-06-22","2026-06-25"],[3,14041,"Finalize Guidelines for Ulysses based context parallel attention","atharvajoshi10",4,"","",0,"2026-06-22",""],[3,14037,"`HookRegistry._child_registries_cache` goes stale after `enable_cache`/`disable_cache` → `ValueError: No conte","christopher5106",0,"SuryanshSS1011","#14093",0,"2026-06-22","2026-09-05"],[3,14033,"Remove link to HEIM for diffusion model evaluation","beyandbey",4,"","",0,"2026-06-22",""],[3,14025,"MagCache on Wan 2.2 Dual-Transformer Pipelines: Incorrect Step Accounting and Limited Effectiveness on a 4-Ste","fatemehgh1313",4,"","",0,"2026-06-21",""],[3,14013,"A bug on the set_timesteps function of the FlowMatchEulerDiscreteScheduler","TheLovesOfLadyPurple",1,"TheLovesOfLadyPurple","#14011",0,"2026-06-20","2026-07-07"],[3,14012,"attention dispatcher assumes wrong attributes for flash attn kernel from hub","bghira",4,"","",0,"2026-06-20",""],[3,14009,"Kohya->diffusers LoRA key converter fails for Chroma and Flux1, if LoRA is not attention-only","dxqb",2,"","",0,"2026-06-20","2026-06-27"],[3,14008,"from_pretrained fails loading deprecated pipelines","dxqb",4,"","",0,"2026-06-20",""],[3,14005,"_convert_unet_lora_key misses conv_in/conv_out/time_embedding mapping for Kohya UNet LoRAs","dxqb",1,"dxqb","#14006",0,"2026-06-20","2026-06-28"],[3,14004,"PixArtSigmaPipeline: no LoRA loading support","dxqb",4,"","",0,"2026-06-20",""],[3,14003,"Split Q,K,V vs Fused QKV Projections in diffusers Discussion","dg845",4,"","",0,"2026-06-20",""],[3,14002,"Support Original Checkpoint-Compatible PEFT Adapters","dg845",4,"","",0,"2026-06-20",""],[3,13999,"Flash Attention 3 varlen wrapper assumes output is a tuple","bghira",0,"zaid646","#14102",0,"2026-06-19","2026-07-02"],[3,13998,"FLUX kohya LoRA conversion crashes on `final_layer` (KeyError on missing adaLN_modulation_1; \"Incompatible key","christopher5106",0,"lcheng321","#14001",0,"2026-06-19","2026-06-20"],[3,13984,"FLUX LoRA with CLIP text-encoder weights fails (empty rank -> IndexError) under transformers>=5","christopher5106",4,"","",0,"2026-06-18",""],[3,13979,"Can't use `load_lora_weights()` on `Ideogram4ModularPipeline`","SamuelTallet",1,"SamuelTallet","#13980",0,"2026-06-17","2026-06-17"],[3,13967,"Can't use `load_lora_weights()` on `ErnieImageModularPipeline`","SamuelTallet",1,"SamuelTallet","#13948",0,"2026-06-15","2026-06-16"],[3,13935,"Please reconsider `transformers>=5.0.0`","dxqb",4,"","",0,"2026-06-13",""],[3,13930,"LTX2 text connectors pass reversed prompt tokens and misplaced registers to the transformer (regression from #","Boffee",1,"Boffee","#13931",0,"2026-06-12","2026-06-19"],[3,13925,"1 bit, ternary image generation support","rupeshs",4,"","",0,"2026-06-12",""],[3,13920,"Ideogram4: `Ideogram4MRoPE` breaks under `torch.autocast`: all image positions collapse, producing flat single","dxqb",0,"HaozheZhang6","#13922",0,"2026-06-11","2026-06-19"],[3,13909,"[New Pipeline/Model] Add JoyAI-Echo multi-shot audio-video generation pipeline","sjq66",4,"","",0,"2026-06-10",""],[3,13907,"Support for JoyAI-Echo","feice-huang",2,"","",0,"2026-06-10","2026-06-10"],[3,13903,"[Anima] Add img2img capability","asomoza",0,"PreethamNoelP","#13929",0,"2026-06-09","2026-07-16"],[3,13887,"When use UniPCMultistepScheduler as the ODE solver, and input different num_inference_steps in StableDiffusion","TheLovesOfLadyPurple",4,"","",0,"2026-06-08",""],[3,13877,"docs: fix typos in scheduling_euler_discrete.py","ramkumar27072006",1,"ramkumar27072006","#13876",0,"2026-06-07","2026-06-26"],[3,13869,"fix underlying issue with `test_from_save_pretrained_dtype_inference` is that the `model.to(dtype)` cast at","sayakpaul",4,"","",0,"2026-06-05",""],[3,13866,"flux2-klein lora train，bug","lspze",0,"HaozheZhang6","#13923",0,"2026-06-04","2026-06-12"],[3,13864,"AttributeError: 'Cosmos2_5_PredictBasePipeline' object has no attribute 'check_text_safety'","Tortoise17",2,"","",0,"2026-06-04","2026-06-05"],[3,13858,"[<languageCode>] Translating docs to <languageName>","borisdavalosp",0,"liwd190019","#13932",0,"2026-06-03","2026-06-15"],[3,13844,"[Community Support] Integrating visual generative foundation models in diffusers","Bili-Sakura",4,"","",0,"2026-05-30",""],[3,13833,"from_single_file: CLIPTextModel has no attribute 'text_model' with transformers >= 5.6","kappacommit",0,"asomoza","#13843",1,"2026-05-29","2026-06-01"],[3,13819,"Black image when running QwenImageEdit snippet from docs","vgoel-sc",2,"","",0,"2026-05-27","2026-05-28"],[3,13811,"Fix incorrect batch handling in _prepare_image_ids usage in train_dreambooth_lora_flux2_img2img.py","Zhu1116",4,"","",0,"2026-05-26",""],[3,13801,"Incompatibility between FlashAttention and ERNIE-Image","SamuelTallet",4,"","",0,"2026-05-24",""],[3,13793,"Support for Stable Audio 3 Medium","sdevil7th",0,"buffett0323","#14119",0,"2026-05-22","2026-08-19"],[3,13790,"loading Cosmos2 pipeline also disabled gradient tracking globally","haixpham",4,"","",0,"2026-05-22",""],[3,13789,"UNet2DModel dtype property fails under nn.DataParallel with UnboundLocalError in get_parameter_dtype","ragibarnab",4,"","",0,"2026-05-22",""],[3,13772,"Bad image output for Flux.2-dev, rocm, quantization and separate prompt encoding sequence","john-tecplot",4,"","",0,"2026-05-20",""],[3,13762,"FluxKlein Training Scripts - CFG issue","cryptexis",4,"","",0,"2026-05-17",""],[3,13722,"ErnieImageTransformer2DModel does not implement from_single_file","dxqb",0,"akshan-main","#13727",0,"2026-05-11","2026-06-08"],[3,13717,"checkpoint. Make sure all LoRA param names contain `'lora'` substring.","xalteropsx",2,"","",0,"2026-05-11","2026-05-11"],[3,13713,"Incorporate safetensors support to TorchAO","sayakpaul",0,"hlky","#13719",0,"2026-05-11","2026-06-12"],[3,13703,"UNet1DModel loss plateauing at ≃0.5. How to fix time blindness when training on sequential data.","elcaidaprojects-art",2,"","",0,"2026-05-09","2026-05-09"],[3,13700,"f","Antonella-K",2,"","",0,"2026-05-08","2026-05-11"],[3,13696,"[bug] The mask is not correctly sharded for QwenImageTransformer + Ulysses SP","zhtmike",2,"","",0,"2026-05-08","2026-06-08"],[3,13691,"[Security] Incomplete Fix for CVE-2026-44513: community Pipeline Branch Bypasses trust_remote_code Check","icysun",2,"","",0,"2026-05-07","2026-05-13"],[3,13683,"universal method or class to load any model locally","knightsun",4,"","",0,"2026-05-06",""],[3,13682,"如何实现flux2Klein的DPO训练","PlutoQyl",4,"","",0,"2026-05-05",""],[3,13680,"[Windows] RTX 5070 Ti (Blackwell sm_120) - setup and deployment notes","loongmiaow-pixel",2,"","",0,"2026-05-05","2026-06-09"],[3,13656,"[meta issue] Systematic model/pipeline review findings / tracking","hlky",4,"","",0,"2026-04-29",""],[3,13655,"`model_infrastructure` model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13654,"`model_unets_shared` model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13653,"`pipeline_infrastructure` model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13652,"`model_autoencoders_shared` model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13651,"`model_transformers_shared` model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13650,"modular_pipeline_infrastructure model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13649,"ddpm model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13648,"cogview3 model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13647,"allegro model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13646,"deepfloyd_if model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13643,"consistency_models model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13642,"[docs] typo in AutoencoderOobleck docs","ivnvalex",1,"ivnvalex","#13645",0,"2026-04-29","2026-04-29"],[3,13641,"hunyuandit model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13640,"kandinsky3 model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13639,"kandinsky5 model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13638,"easyanimate model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13637,"latent_consistency_models model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13636,"longcat_image model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13635,"ledits_pp model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13634,"lumina model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13633,"omnigen model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13632,"lucy model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13631,"pixart_alpha model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13630,"ovis_image model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13629,"stable_audio model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13628,"marigold model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13627,"stable_video_diffusion model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13626,"t2i_adapter model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13625,"bria model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13624,"aura_flow model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13623,"visualcloze model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13622,"cogvideo model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13621,"cogview4 model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13620,"chronoedit model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13619,"chroma model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13618,"bria_fibo model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13617,"hidream_image model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13616,"dit model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13615,"mochi model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13614,"sana model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13613,"lumina2 model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13612,"kolors model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13611,"stable_diffusion_3 model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13610,"stable_diffusion_xl model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13609,"skyreels_v2 model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13608,"prx model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13607,"cosmos model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13606,"sana_video model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13605,"hunyuan_image model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13604,"helios model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13603,"latent_diffusion model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13602,"audioldm2 model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13601,"ltx2 model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13600,"controlnet model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13599,"animatediff model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13598,"llada2 model/pipeline review","hlky",0,"kashif","#13698",1,"2026-04-29","2026-05-17"],[3,13597,"kandinsky model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13596,"kandinsky2_2 model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13595,"latte model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13594,"pag model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13593,"shap_e model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13592,"stable_diffusion model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13591,"ddim model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13590,"nucleusmoe_image model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13589,"stable_cascade model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13588,"hunyuan_video model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13587,"glm_image model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13586,"consisid model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13585,"ltx model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13584,"flux model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13583,"z_image model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13582,"`hunyuan_video1_5` model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13581,"`qwenimage` model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13580,"`longcat_audio_dit` model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13579,"`flux2` model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13578,"`wan` model/pipeline review","hlky",4,"","",0,"2026-04-29",""],[3,13577,"`ernie-image` model/pipeline review","hlky",2,"","",0,"2026-04-29","2026-05-06"],[3,13574,"Z-Image的文本位置编码长度不一致问题","PlutoQyl",4,"","",0,"2026-04-28",""],[3,13566,"INR0","amitdealer436-collab",4,"","",0,"2026-04-27",""],[3,13565,"India 🫂 hugging","amitdealer436-collab",2,"","",0,"2026-04-27","2026-05-11"],[3,13547,"Error in loading flux2 klein lora adapter","saeedkhanehgir",4,"","",0,"2026-04-22",""],[3,13546,"ERNIE-Image RoPE implementation: Shape mismatch between freqs_cis and chunking?","Rauyolo",4,"","",0,"2026-04-22",""],[3,13506,"Sage attention fails with ring-attention.","songh11",4,"","",0,"2026-04-20",""],[3,13505,"Z-Image's modular TE block only produces negative_prompt_embeds when the pipeline's guider is enabled (CFG act","christopher5106",4,"","",0,"2026-04-20",""],[3,13501,"LoRA support for ERNIE-Image","SamuelTallet",0,"asomoza","#13575",1,"2026-04-18","2026-04-30"],[3,13488,"diffusers-0.38.0.dev0 breaks offloading","Sector14",0,"ParamChordiya","#13489",0,"2026-04-15","2026-04-30"],[3,13487,"Loading loras on \"text_pipe\" of a modular pipeline fails","christopher5106",4,"","",0,"2026-04-15",""],[3,13484,"Lora loading fail on Flux 2 dev for FAL loras","christopher5106",0,"yadferhad","#13486",0,"2026-04-15","2026-04-18"],[3,13477,"ErnieImage callback locals() scope bug.","songh11",1,"songh11","#13478",0,"2026-04-15","2026-04-16"],[3,13475,"QwenImage `has_neg_prompt` derivation","sayakpaul",2,"","",0,"2026-04-15","2026-04-16"],[3,13462,"Decorating bound methods in profiling utils breaks instance isolation","ViktoriiaRomanova",0,"Akash504-ai","#13471",0,"2026-04-13","2026-04-15"],[3,13452,"[CI] Trial Sourcery PR reviews","iwr-redmond",4,"","",0,"2026-04-13",""],[3,13447,"Support for Exclusive Self Attention (XSA) in Diffusion Models (DiT, U-Net, etc.)","Bili-Sakura",4,"","",0,"2026-04-12",""],[3,13446,"Insufficient trust check for custom_pipeline parameter of DiffusionPipeline.from_pretrained method","Vancir",0,"hlky","#13448",0,"2026-04-12","2026-04-23"],[3,13445,"Wrong default layers for Flux 2 Klein","forhaxed",4,"","",0,"2026-04-11",""],[3,13439,"HunyuanVideo 1.5 I2V image conditioning preprocessed at latent resolution instead of pixel resolution","akshan-main",1,"akshan-main","#13440",0,"2026-04-10","2026-04-10"],[3,13430,"Support for JoyAI-Image-Edit","Moran232",1,"Moran232","#13444",0,"2026-04-07","2026-05-07"],[3,13429,"add JoyAI-Image-Edit","Moran232",2,"","",0,"2026-04-07","2026-04-07"],[3,13425,"Division by zero in rescale_noise_cfg can produce NaNs during inference","Akash504-ai",4,"","",0,"2026-04-06",""],[3,13421,"Add negative_prompt parameter to GLMImagePipeline","akshan-main",2,"","",0,"2026-04-06","2026-04-06"],[3,13418,"2 gpu","ayttop",2,"","",0,"2026-04-06","2026-04-06"],[3,13416,"how to use zimage and flux2 with negative prompt?","ghost",4,"","",0,"2026-04-04",""],[3,13411,"`LTXEulerAncestralRFScheduler.set_timesteps(sigmas=...)` does not validate monotonicity, causing silent incorr","VittoriaLanzo",3,"","",0,"2026-04-04","2026-04-26"],[3,13401,"Help us profile important pipelines and improve if needed","sayakpaul",4,"","",0,"2026-04-03",""],[3,13394,"DDPMScheduler allows num_inference_steps=0 without validation (inconsistent with DDIMScheduler)","Akash504-ai",4,"","",0,"2026-04-03",""],[3,13386,"[Bug] `train_dreambooth_lora_qwen_image.py` crashes with `--with_prior_preservation` due to tensor concatenati","chenyangzhu1",1,"chenyangzhu1","#13387",0,"2026-04-02","2026-04-10"],[3,13377,"[Bug] `QwenImagePipeline` silently disables CFG when passing `negative_prompt_embeds` if mask is `None` (which","Sunhill666",1,"Sunhill666","#13379",0,"2026-04-01","2026-04-17"],[3,13376,"Add AEO quality badge to README (score: 32/100)","digitamaz",2,"","",0,"2026-04-01","2026-07-21"],[3,13375,"Investigate the current DtoH sync solution in pipelines","sayakpaul",4,"","",0,"2026-04-01",""],[3,13362,"DDIMPipeline does not validate eta range despite documented constraint [0, 1]","Akash504-ai",0,"NIK-TIGER-BILL","8070f6e",0,"2026-03-29","2026-04-03"],[3,13361,"Enable MPS backend for bitsandbytes quantization","LucasSte",1,"LucasSte","#13915",0,"2026-03-28","2026-06-23"],[3,13359,"AI-generated image provenance metadata (EU AI Act deepfake disclosure)","HMAKT99",4,"","",0,"2026-03-28",""],[3,13357,"KeyError: 'default' for discrete diffusion language model LLaDA2","ksasi",4,"","",0,"2026-03-28",""],[3,13351,"Wan 2.2 Fun Control & Flux.2 Dev Fun ControlNet","adrianmeyerart",4,"","",0,"2026-03-26",""],[3,13350,"how to use black-forest-labs/FLUX.2-klein-9B with Batch inference","ghost",2,"","",0,"2026-03-26","2026-04-03"],[3,13349,"Noticed cropped results when generating with FLUX.2 klein 4B with reference images.","matemato",2,"","",0,"2026-03-26","2026-08-06"],[3,13343,"[CI] Auto-label PRs for better insight and visibility","iwr-redmond",4,"","",0,"2026-03-26",""],[3,13319,"The backward pass of QwenImageTransformer failed with Ulysses SP.","zhtmike",2,"","",0,"2026-03-24","2026-03-30"],[3,13311,"black-forest-labs/FLUX.2-klein-9B lora No LoRA keys associated to Flux2Transformer2DModel found with the prefi","ghost",0,"sayakpaul","#13313",1,"2026-03-23","2026-03-25"],[3,13310,"YIh","nanvyu",2,"","",0,"2026-03-22","2026-03-24"],[3,13301,"Modular Pipeline: support for PixArtAlphaPipeline","gasparyanartur",4,"","",0,"2026-03-21",""],[3,13300,"diffusers fails in PyTorch when generating image using stabilityai/stable-diffusion-3.5-large-turbo, black-for","yurivict",4,"","",0,"2026-03-21",""],[3,13298,"When generating images, if the generator device is on cuda, it break things","weathon",4,"","",0,"2026-03-21",""],[3,13295,"Modular Diffusers 🧨","sayakpaul",4,"","",0,"2026-03-20",""],[3,13292,"[Bug] train_dreambooth_lora_flux2_klein.py: batch size mismatch with --with_prior_preservation","vishk23",0,"chenyangzhu1","#13396",0,"2026-03-20","2026-04-10"],[3,13288,"vbench","3a1b2c3",2,"","",0,"2026-03-19","2026-03-19"],[3,13286,"torchao >= 0.16.0 quantization not supported","zzlol63",2,"","",0,"2026-03-19","2026-06-10"],[3,13284,"AttentionModuleMixin.set_attention_backend does not download hub kernels","Marius-Graml",2,"","40040d5",0,"2026-03-18","2026-11-04"],[3,13281,"Group offloading with use_stream=True breaks torchao quantized models (device mismatch) in qwen image","lllllinux",4,"","",0,"2026-03-18",""],[3,13279,"`AutoencoderRAE` loading error with older transformers","arijit-hub",4,"","",0,"2026-03-17",""],[3,13277,"QwenImagePipeline failed with Ulysses SP and batch inputs","zhtmike",2,"","",0,"2026-03-17","2026-03-25"],[3,13274,"[Bug][AMD] `stabilityai/stable-audio-open-1.0` encounter `RecursionError: maximum recursion depth exceeded`","tjtanaa",4,"","",0,"2026-03-16",""],[3,13266,"Torchao SD3 int8wo","WasamiKirua",4,"","",0,"2026-03-14",""],[3,13261,"black-forest-labs/FLUX.2-klein-9B not working with lora with lokr","ghost",2,"","",0,"2026-03-12","2026-10-09"],[3,13258,"Wan-AI/Wan2.2-TI2V-5B-Diffusers Image to Video Missing","james-imi",4,"","",0,"2026-03-11",""],[3,13256,"diffusers,hugging-face and tokenizers 0.14.1 ,they have a compatiable version","coderXEi",4,"","",0,"2026-03-11",""],[3,13249,"Lora Stopped working for Z-Image","AlpineVibrations",0,"sayakpaul","#13255",1,"2026-03-10","2026-03-13"],[3,13243,"[BUG] FlowMatchEulerDiscreteScheduler.__init__ computes sigma_min/sigma_max after shift, causing duplicate shi","Xilluill",4,"","",0,"2026-03-10",""],[3,13241,"[Chore] Set Diffusers 0.37 as latest release","iwr-redmond",2,"","",0,"2026-03-10","2026-03-11"],[3,13232,"LTX-2.3 Support","rootonchair",2,"","",0,"2026-03-09","2026-04-22"],[3,13227,"[Bug] GlmImagePipeline silently corrupts weights on MPS accelerator","yingding",4,"","",0,"2026-03-08",""],[3,13225,"RAE+DiT support for published checkpoints","plugyawn",4,"","",0,"2026-03-08",""],[3,13224,"Regarding support for Qwen‑Image‑Layered‑Control","suzukimain",4,"","",0,"2026-03-08",""],[3,13221,"Zimage lora support issue too","zhaoyun0071",0,"christopher5106","#14163",0,"2026-03-08","2026-10-09"],[3,13203,"Zimage lora support issue","christopher5106",0,"sayakpaul","#13209",1,"2026-03-02","2026-03-05"],[3,13196,"Quantization documentation runnable in Colab","jchwenger",2,"","",0,"2026-02-27","2026-03-09"],[3,13191,"Elastic-DiT support","geekuillaume",4,"","",0,"2026-02-27",""],[3,13177,"Flux2KleinPipeline doesn't accept tensor format as input image","joansc",4,"","",0,"2026-02-23",""],[3,13176,"UNet2DConditionModel attention error","dumoulina",2,"","",0,"2026-02-23","2026-02-23"],[3,13167,"Wan I2V expand_timesteps: why hard-clamp only first frame (no last-frame clamp)?","emirks",4,"","",0,"2026-02-20",""],[3,13162,"Flux.2-klein-4b lora support","christopher5106",0,"asomoza","#13169",1,"2026-02-18","2026-02-21"],[3,13151,"GGUF + `torch.compile` is not compatible with Hub Kernels","DN6",4,"","",0,"2026-02-16",""],[3,13142,"PRXPipeline.from_pretrained() broken with transformers 5.1.0","DavidBert",1,"DavidBert","#13143",0,"2026-02-13","2026-02-23"],[3,13140,"Qwen Edit Plus + Qwen Edit pipelines fail (TypeError: argument of type 'NoneType' is not iterable)","christopher5106",2,"","",0,"2026-02-13","2026-02-16"],[3,13137,"Flux 2 Klein load lora weights LoKr error","yodo226",2,"","",0,"2026-02-13","2026-03-04"],[3,13124,"train_dreambooth_lora.py -- ValueError: Attempting to unscale FP16 gradients caused by \"--validation_prompt\" p","Xjmengnieer",0,"HaozheZhang6","#13895",0,"2026-02-11","2026-06-09"],[3,13112,"SkyReelsV2DiffusionForcingPipeline does not check for available packages","vladmandic",0,"asomoza","#13113",1,"2026-02-09","2026-02-10"],[3,13107,"Flux Klein LoRA – Missing Key Errors","mshehzadkhan",0,"sayakpaul","#13119",1,"2026-02-09","2026-02-11"],[3,13106,"This issue has been automatically marked as stale because it has not had recent activity. If you think this st","Pediboi666",2,"","",0,"2026-02-09","2026-02-23"],[3,13105,"Wan-AI/Wan2.1-T2V-1.3B-Diffusers output noise","jiqing-feng",2,"","",0,"2026-02-09","2026-02-26"],[3,13104,"Logger is not defined","teleprint-me",2,"","",0,"2026-02-08","2026-03-11"],[3,13101,"huggingface ","Pediboi666",2,"","",0,"2026-02-08","2026-02-23"],[3,13096,"RuntimeError: Tensor.item() cannot be called on meta tensors","TheephopWS",2,"","",0,"2026-02-07","2026-02-16"],[3,13087,"cannot import name 'Flux2KleinPipeline' from 'diffusers'","Ikkyu321",2,"","",0,"2026-02-06","2026-02-08"],[3,13086,"Add Nvidia NitroGen to huggingface diffusers","AffanBinFaisal",4,"","",0,"2026-02-06",""],[3,13079,"FLUX.2-klein: CUDA OOM when running in for loop with different image","dangph-alala",4,"","",0,"2026-02-04",""],[3,13077,"ZImageControlNet.from_transformer creates a shallow copy of the transformer weights","Christophernph",4,"","",0,"2026-02-04",""],[3,13074,"Distributed gather crashes on PyTorch < 2.6","akshatvishu",4,"","",0,"2026-02-03",""],[3,13073,"ZImageControlNetPipeline does not support guidance_scale>1","christopher5106",0,"hlky","#13080",0,"2026-02-03","2026-02-07"],[3,13070,"Add JiT: Just image Transformer (JiT) for Pixel-space Diffusion","sayakpaul",4,"","",0,"2026-02-03",""],[3,13067,"[Feature] Add support for Anima","Metal079",0,"rmatif","#13732",0,"2026-02-03","2026-05-29"],[3,13065,"Qwen-Image controlnet inpaint + controlnet canny","xinli2008",4,"","",0,"2026-02-02",""],[3,13062,"Do not support onnxruntime-windowsml","xieofxie",4,"","",0,"2026-02-02",""],[3,13061,"microsoft/vq-diffusion-ithq `variant='fp16'`  is not in the main branch","GabbySuwichaya",2,"","",0,"2026-02-01","2026-03-09"],[3,13053,"[Feature] Support Unipic 3.0 Pipeline","Ando233",4,"","",0,"2026-01-29",""],[3,13043,"[Bug] `no kernel image is available for execution on the device` when using sage_hub attention backend on RTX ","timeuser4",4,"","",0,"2026-01-28",""],[3,13035,"cannot import name 'MT5Tokenizer' from 'transformers'","n1ck-guo",2,"","",0,"2026-01-27","2026-03-04"],[3,13034,"[Bug] WanPipeline `height` and `width`","Jayce-Ping",1,"Jayce-Ping","#13038",0,"2026-01-27","2026-01-29"],[3,13032,"allow cpu-only support for \"diffusers\" ?","israelandrewbrown",4,"","",0,"2026-01-26",""],[3,13016,"[Feature] Support QwenImageEditPlus series attention mask for NPU","zhangtao0408",1,"zhangtao0408","#13017",0,"2026-01-22","2026-02-17"],[3,13015,"[Bug] QwenImageEditPlus series error `Expected all tensors to be on the same device` on NPU","zhangtao0408",1,"zhangtao0408","#13017",0,"2026-01-22","2026-02-17"],[3,13009,"Question about VAE_IMAGE_SIZE and CONDITION_IMAGE_SIZE constants in pipeline_qwenimage_edit_plus.py","gobigrassland",4,"","",0,"2026-01-21",""],[3,13008,"Support for Multi-Image Input LoRA Training Pipeline for FLUX.2-Klein (e.g., Style Transfer and Multi-Subject ","qizhuzhuang",4,"","",0,"2026-01-21",""],[3,13005,"Add Flux.2 Klein Inpaint pipeline","Natans8",0,"adi776borate","#13050",0,"2026-01-20","2026-04-17"],[3,13001,"Shape mismatch when loading a Flux2.Klein GGUF","dxqb",4,"","",0,"2026-01-19",""],[3,13000,"RAE support","bytetriper",0,"Ando233","#13046",0,"2026-01-19","2026-03-05"],[3,12997,"Why does calculate_dimensions in qwen-image-edit require image height/width to be multiples of 32?","lucky-liuzhihong",4,"","",0,"2026-01-19",""],[3,12995,"Does `Cosmos-Predict2.5` pipeline generate video? or reconstruct input videos?","rebel-dkhong",0,"adi776borate","#13018",0,"2026-01-19","2026-01-27"],[3,12990,"[optimization] help us know which kernels we should integrate in Diffusers","sayakpaul",4,"","",0,"2026-01-17",""],[3,12989,"how Flux.2 Klein support both T2I and TI2I","Bili-Sakura",4,"","",0,"2026-01-16",""],[3,12981,"Problems loading LTX-2 transformer GGUFs","AmericanPresidentJimmyCarter",0,"dg845","#12983",0,"2026-01-15","2026-01-16"],[3,12975,"AnimateDiffPipeline performance regression on CPU.","jiqing-feng",1,"jiqing-feng","#12977",0,"2026-01-14","2026-01-16"],[3,12967,"GGUF  remove de-quantizing and converting support, like comfyui","zwukong",2,"","",0,"2026-01-12","2026-01-22"],[3,12935,"Please create a Huggingface Diffusers SKILL for Claude and a PLUGIN for Claude Code.","Emasoft",4,"","",0,"2026-01-09",""],[3,12926,"LTX-2 condition pipeline","sayakpaul",0,"dg845","#13058",0,"2026-01-08","2026-03-05"],[3,12925,"LTX-2 distilled checkpoint support","sayakpaul",0,"rootonchair","#12934",0,"2026-01-08","2026-01-29"],[3,12920,"Any plan on LTX-2?","gitlabspy",2,"","",0,"2026-01-07","2026-01-07"],[3,12918,"Is there any latest information about CVE-2025-14922?","felnd",2,"","",0,"2026-01-07","2026-01-07"],[3,12913,"Is Lumina2Pipeline's mu calculation correct?","hwangdonghyun",4,"","",0,"2026-01-05",""],[3,12908,"15263","tndud1573",2,"","",0,"2026-01-01","2026-01-02"],[3,12907,"이미지","tndud1573",2,"","",0,"2026-01-01","2026-01-01"],[3,12906,"사이버펑크","tndud1573",2,"","",0,"2026-01-01","2026-01-05"],[3,12905,"Wrong Z_Image guidance scale implementation","Trgtuan10",0,"kirillsst","#13183",0,"2026-01-01","2026-02-26"],[3,12903,"[Feature Request] Add Implement of 'Kuramoto Orientation Diffusion'","Bili-Sakura",4,"","",0,"2025-12-30",""],[3,12900,"[Bug]: OOM of Qwen-Image (2048px) generation using Qwen/Qwen-Image","LJH-LBJ",2,"","",0,"2025-12-29","2025-12-29"],[3,12899,"Training script of  z-image controlnet?","universewill",4,"","",0,"2025-12-29",""],[3,12896,"The same WAN 2.2 i2v script (Almost) works in CUDA, bur fails in MPS","kerufu",2,"","",0,"2025-12-27","2026-01-26"],[3,12895,"StableDiffusion3Pipeline crashes with AttributeError: 'NoneType' object has no attribute 'ndim' when using num","FredyRivera-dev",2,"","",0,"2025-12-27","2026-10-03"],[3,12893,"Z-Image text sequence length issue","dxqb",2,"","",0,"2025-12-26","2026-03-24"],[3,12891,"not support load  Qwen-Image-Edit-2511 gguf or fp8 safetensors？","zhaoyun0071",0,"mbalabanski","#12894",0,"2025-12-26","2026-01-02"],[3,12889,"Question about qwen-image-edit-2511 loading warning","wizardbob",2,"","",0,"2025-12-25","2025-12-25"],[3,12887,"z-image-turbo from_single_file errors","vladmandic",0,"hlky","#12888",0,"2025-12-24","2026-01-02"],[3,12884,"Compatibility issues regarding checkpoint/VAE dependency conflicts when Diffusers load Civitai LoRA","hhhFuture",2,"","",0,"2025-12-23","2025-12-23"],[3,12881,"Is that a bug of prompt2prompt pipeline with replace word pormpt?","lincion",4,"","",0,"2025-12-23",""],[3,12880,"Allow users to pass in a custom device_mesh to the enable_parallelism method in the diffusers library","pthombre",2,"","",0,"2025-12-22","2026-03-11"],[3,12878,"This new version of diffusers causes problems when doing inference with Flux 1.D and Wan2.2 Text To Image (exp","ukaprch",2,"","",0,"2025-12-22","2025-12-23"],[3,12874,"StabilityAI just take down their SD2 from huggingface","Bili-Sakura",2,"","",0,"2025-12-22","2025-12-22"],[3,12864,"Please create a Diffusers SKILL for Claude and a Diffusers Plugin for Claude Code","Emasoft",4,"","",0,"2025-12-20",""],[3,12859,"load_lora_weights(), unload_lora_weights these doesn't work with LongCatImagePipeline.","ovi054",4,"","",0,"2025-12-18",""],[3,12843,"[Z-Image] ONNX export fails due to pad_sequence / unbind / dynamic-shape ops in ZImageTransformer2DModel (requ","er6y",4,"","",0,"2025-12-15",""],[3,12840,"Fix the warning `torch_dtype` is deprecated","msdsm",1,"msdsm","#12841",0,"2025-12-15","2026-01-09"],[3,12838,"Merge Loras for FLUX","shrikrishnalolla",4,"","",0,"2025-12-14",""],[3,12826,"cannot import name 'export_to_video' from 'diffusers","Arslan-Mehmood1",2,"","",0,"2025-12-11","2026-01-10"],[3,12823,"How to use quantizer after pipeline loaded?","DefTruth",2,"","",0,"2025-12-11","2026-01-10"],[3,12821,"train_controlnet","Mandarin1126",4,"","",0,"2025-12-11",""],[3,12818,"train_controlnet","Mandarin1126",2,"","",0,"2025-12-11","2025-12-11"],[3,12817,"Implement Selective Attention for Memory-Efficient Inference","cdutr",2,"","",0,"2025-12-10","2026-01-09"],[3,12809,"Kandinsky5TimeEmbeddings hardcodes 'cuda' in @torch.autocast decorator, causing warning on   non-CUDA systems","knd0331",0,"adi776borate","#12814",0,"2025-12-09","2026-01-08"],[3,12786,"[PRX] Improving export to Tensorrt","WaterKnight1998",1,"WaterKnight1998","#12787",0,"2025-12-03","2025-12-08"],[3,12783,"Sage Attention sm90 causes confetti/noisy output","dylanprins",4,"","",0,"2025-12-03",""],[3,12781,"Impossible to log into Huggingface/Diffusers Discord","tin2tin",2,"","",0,"2025-12-03","2025-12-04"],[3,12779,"Flux2 dreambooth: AttributeError: 'Namespace' object has no attribute 'validation_image_path'","hannalaguilar",2,"","",0,"2025-12-03","2026-01-10"],[3,12778,"AttributeError: type object 'Flux2ImageProcessor' has no attribute 'image_processor'","hannalaguilar",2,"","",0,"2025-12-03","2026-01-10"],[3,12776,"Flux2: Tensor tuples can cause issues for checkpointing","dxqb",2,"","",0,"2025-12-02","2026-06-27"],[3,12772,"How to convert diffusers model to wan2.2 format","wikiwen",2,"","",0,"2025-12-02","2026-01-10"],[3,12771,"DPMSolverMultistepScheduler out-of-bounds error when using beta_schedule=\"squaredcos_cap_v2\" + use_karras_sigm","dylanzhao25",4,"","",0,"2025-12-02",""],[3,12769,"Z-Image Turbo Controlnet Union is out","tin2tin",0,"hlky","#12792",0,"2025-12-02","2025-12-17"],[3,12764,"When will the img2img pipeline of FLUX.2-dev be released?","guanxyu",2,"","",0,"2025-12-01","2026-01-10"],[3,12761,"function TemplatedUlyssesAttentionBackward returned an incorrect number of gradients (expected 12, got 11)","jiarenyf",2,"","",0,"2025-12-01","2025-12-02"],[3,12760,"`StateManager`, `Hook`, `State`: missing context setting (cond or uncond) in pipelines","toilaluan",4,"","",0,"2025-12-01",""],[3,12755,"Flux 2: The shape of the latent argument is undocumented","rolux",4,"","",0,"2025-11-30",""],[3,12754,"`from_pipe` converts pipelines to float32 by default","missionfloyd",4,"","",0,"2025-11-30",""],[3,12752,"Z-Image img2img and inpainting pipeline","CalamitousFelicitousness",4,"","",0,"2025-11-29",""],[3,12749,"An Idea for partial sequential cpu offloading","rodjjo",2,"","",0,"2025-11-29","2026-01-10"],[3,12748,"Please support GGUF format models for Z-Image-Turbo","Vargol",0,"hlky","#12756",0,"2025-11-29","2025-12-04"],[3,12746,"Flux2 Test error cannot import name 'AutoProcessor' from 'transformers'","Honey-666",2,"","",0,"2025-11-29","2025-12-10"],[3,12745,"ZImagePipeline Add Lora","snz2",2,"","",0,"2025-11-29","2026-01-10"],[3,12742,"Zero tensor in transformer_z_image","lime-j",0,"JerryWu-code","#12770",0,"2025-11-28","2025-12-02"],[3,12738,"ZImagePipeline fails with batch sizes above 1 with Assertion Error","Meatfucker",2,"","",0,"2025-11-27","2026-01-10"],[3,12735,"Sam3D support","fabvio",2,"","",0,"2025-11-27","2025-11-27"],[3,12732,"diffusers inference pipeline error for Z-Image","MengHao666",2,"","",0,"2025-11-27","2025-11-27"],[3,12730,"Debugging snippet still remains in Flux2Pipeline","nuclearboy95",2,"","",0,"2025-11-27","2025-12-01"],[3,12728,"torch.compile + channels_last support for Wan 2.2 (T2V / I2V) fails with RuntimeError + Dynamo Unsupported beh","Passenger12138",4,"","",0,"2025-11-27",""],[3,12722,"How to run qwen-image in kaggle gpu T4 * 2 successfully?","ghost",4,"","",0,"2025-11-26",""],[3,12720,"how to quantization wan 2.2 vace after loading lora?","ghost",2,"","",0,"2025-11-26","2026-01-10"],[3,12719,"how to use quantization and device_map=balance to run qwen-image on kaggle T4 * 2","ghost",4,"","",0,"2025-11-26",""],[3,12712,"FLUX.2-dev-bnb-4bit Missing","ghunkins",2,"","",0,"2025-11-25","2025-11-25"],[3,12709,"Meituan Longcat Video","alerikaisattera",4,"","",0,"2025-11-24",""],[3,12706,"[Feature] Ulysses Attention for any sequence length w/o padding","DefTruth",1,"DefTruth","#12996",0,"2025-11-24","2026-02-02"],[3,12699,"the quality of video downgrade very much when calling pipe for 50 different prompts in wan 2.2 with lighting l","ghost",2,"","",0,"2025-11-22","2026-01-10"],[3,12697,"Magcache Support.","Meatfucker",0,"AlanPonnachan","#12744",0,"2025-11-22","2026-02-04"],[3,12695,"Hunyuan v1.5 Support","ghunkins",2,"","",0,"2025-11-21","2026-01-10"],[3,12694,"Feature Request: Enhanced Model Metadata Extraction and Export Utility","Jitmisra",2,"","",0,"2025-11-21","2025-11-22"],[3,12689,"LoRA adapter weights are not loaded from checkpoints","SurAyush",1,"SurAyush","#12690",0,"2025-11-20","2025-11-28"],[3,12683,"[FEATURE] [WAN2.2] Add support to train LoRA for WanVacePipeline","SlimRG",2,"","",0,"2025-11-18","2026-01-10"],[3,12673,"QwenImagepipeline imcompatable with group offloading with record_stream=True","ghost",0,"KimbingNg","#12721",0,"2025-11-17","2025-12-03"],[3,12672,"How to set pipe \"requires_grad=true\"？","micklexqg",2,"","",0,"2025-11-17","2025-11-20"],[3,12669,"Flux1-Dev inference with single file ComfyUI/SD-Forge Safetensors","ddpasa",4,"","",0,"2025-11-16",""],[3,12668,"Complex numbers in Qwen/Z-Image Image pipeline incompatible with torch.compile","dxqb",4,"","",0,"2025-11-15",""],[3,12667,"Any Wan version crashes","tin2tin",4,"","",0,"2025-11-15",""],[3,12666,"Kandinsky 5.0 Video Pipeline not in Diffusers 0.35.2","leffff",2,"","",0,"2025-11-14","2025-11-17"],[3,12663,"huhu","JQueenKing",2,"","",0,"2025-11-14","2025-11-17"],[3,12662,"question on stable_audio_transformer.py","JohnHerry",2,"","",0,"2025-11-14","2026-01-10"],[3,12661,"Crash while using ChronoEdit with context parallelism","DefTruth",1,"DefTruth","#12660",0,"2025-11-14","2025-11-24"],[3,12657,"Qwen Image Edit Quantized Stopped working","AlpineVibrations",2,"","",0,"2025-11-14","2025-11-16"],[3,12656,"[Bug] Positional Embedding Index Out-of-Bounds in AuraFlow","ikuto32",4,"","",0,"2025-11-14",""],[3,12653,"PRX transformers version of rope has a float64 reference, breaks MPS (and npu ?)","Vargol",0,"DavidBert","#12665",0,"2025-11-13","2025-11-15"],[3,12650,"Question about the `# Copied from` system","delmalih",4,"","",0,"2025-11-13",""],[3,12646,"【 EasyAnimate 】 pipine_ easyanimate_inpaint.Py, line 1191: image_1atents used without assigning a value, causi","KittenHaerin",4,"","",0,"2025-11-13",""],[3,12645,"Why Qwen-Image inference speed may slower than comfyui","hanggun",4,"","",0,"2025-11-13",""],[3,12641,"Potential incorrect indentation for logging in train_dreambooth.py","Men1scus",4,"","",0,"2025-11-12",""],[3,12638,"How to design network with DiT blocks that are friendly to Tensorrt fp16 conversion?","JohnHerry",2,"","",0,"2025-11-12","2026-01-10"],[3,12635,"The Diffusers MVP 🚀","sayakpaul",4,"","",0,"2025-11-11",""],[3,12633,"DDIMScheduler.step() logic mismatches set_timesteps() list when using timestep_spacing='linspace'","cbprfCsun",4,"","",0,"2025-11-11",""],[3,12632,"stable_diffusion_xl_img2img failed due to dtype missmatch","jiqing-feng",2,"","",0,"2025-11-11","2026-01-10"],[3,12630,"AttributeError: 'GPT2Model' object has no attribute '_get_initial_cache_position' in AudioLDM2Pipeline","ghost",4,"","",0,"2025-11-11",""],[3,12624,"please add strength to wanimagetovideopipeline","ghost",4,"","",0,"2025-11-10",""],[3,12620,"Missing files in release assets","kwonmha",2,"","",0,"2025-11-10","2025-11-17"],[3,12613,"wan 2.2 cause nan in latent in i2v","ghost",4,"","",0,"2025-11-08",""],[3,12612,"diffusers 0.35.2 wan 2.1 cause nan in latents","ghost",2,"","",0,"2025-11-08","2025-11-08"],[3,12610,"Quanto + Group Offload causes device mismatch error (weights on cpu, mat1 on gpu)","bghira",4,"","",0,"2025-11-07",""],[3,12609,"wan 2.2 cause nan in latents","ghost",2,"","",0,"2025-11-07","2025-11-08"],[3,12608,"yudeng970","hskaskkan",2,"","",0,"2025-11-07","2025-11-07"],[3,12606,"Wrong import of scripts/convert_sd3_to_diffusers.py","Chlorinesu",0,"yashwantbezawada","#12616",0,"2025-11-07","2025-11-10"],[3,12605,"dispatch_attention_fn silently ignores attn_mask for certain backends","zzlol63",0,"dxqb","#12892",0,"2025-11-07","2026-01-06"],[3,12603,"[feature] Add context parallel cp_plan support to many mainstream models","DefTruth",4,"","",0,"2025-11-07",""],[3,12600,"Add KV Cache for Autoregressive Inference","DN6",4,"","",0,"2025-11-06",""],[3,12599,"Moving to CUDA after loading is extremely slow","francois-rozet",2,"","",0,"2025-11-05","2026-01-15"],[3,12597,"[i8n-pt] Updating, improving and expanding current PT translation","cdutr",2,"","",0,"2025-11-05","2025-12-12"],[3,12589,"[feature] implement TeaCache","sayakpaul",4,"","",0,"2025-11-04",""],[3,12588,"Allowing other timestep_spacing heuristics for repaint scheduler. Leading in general is a problematic default.","MarkTension",4,"","",0,"2025-11-04",""],[3,12587,"question about the condition usage of `context_pre_only`","DefTruth",2,"","",0,"2025-11-04","2025-11-04"],[3,12580,"Title: ModuleNotFoundError: Could not import module ‘CLIPTextModel’ in diffusers 0.36.0 + transformers 4.57.0","xxxzzz200",2,"","",0,"2025-11-03","2025-11-04"],[3,12575,"[Chroma] Enabling attention backend checks causes error","zzlol63",2,"","",0,"2025-11-02","2025-11-12"],[3,12574,"[Bug] WanVideoToVideoPipeline fails due to missing handling of control arguments required by underlying VACE t","Mathias5",2,"","",0,"2025-11-02","2025-11-06"],[3,12572,"Chroma inpainting pipeline","rodjjo",0,"hameerabbasi","#12848",0,"2025-11-01","2026-01-15"],[3,12570,"[feature] help us implement unified attention","sayakpaul",2,"","",0,"2025-11-01","2026-01-13"],[3,12569,"[feature] implement TaylorSeer","sayakpaul",2,"","",0,"2025-11-01","2025-12-08"],[3,12568,"Context parallel bug when using QwenImage","sayakpaul",4,"","",0,"2025-11-01",""],[3,12567,"generate noise when using Ulysses Attention","TmacAaron",4,"","",0,"2025-10-31",""],[3,12565,"Wan2.1 i2v running issue","luren55",4,"","",0,"2025-10-31",""],[3,12564,"[Proposals Welcome] Fal Flashpack integration for faster model loading","yiyixuxu",2,"","",0,"2025-10-31","2026-09-01"],[3,12559,"nms不存在！！！我在win系统下边实现这个examples/instruct_pix2pix的报错RuntimeError: operator torchvision::nms does not exist","CSolaris",2,"","",0,"2025-10-29","2026-01-10"],[3,12558,"Please add QwenImageEditPlusPipeline to 0.36","yz342",2,"","",0,"2025-10-29","2025-10-29"],[3,12553,"Reason to move from OpenCV to ffmpeg","Wovchena",4,"","",0,"2025-10-28",""],[3,12550,"Fal Flashpack","vgabbo",2,"","",0,"2025-10-27","2026-04-13"],[3,12547,"Fine tuning Dreambooth Flux Kontext I2I Error: the following arguments are required: --instance_prompt","MichaelMelgarejoFlorez",2,"","",0,"2025-10-27","2025-10-28"],[3,12542,"add custom output_params to export_to_video","eyaler",4,"","",0,"2025-10-25",""],[3,12540,"SanaCombinedTimestepGuidanceEmbeddings do not work with SanaPipeline","frutiemax92",4,"","",0,"2025-10-24",""],[3,12538,"Misallignment in Rotary Frequencies","Warvito",0,"charchit7","#12594",0,"2025-10-23","2025-11-11"],[3,12536,"AssertionError in Context Parallelism during WanImageToVideoPipeline inference: Tensor size along sharding dim","leeguandong",4,"","",0,"2025-10-23",""],[3,12535,"Wan2.2 i2v Latest Lightning LoRA not loading","ljk1291",0,"sayakpaul","#12541",1,"2025-10-23","2025-10-28"],[3,12533,"Hooks conflicts: Context Parallelism and CPU Offload","DefTruth",4,"","",0,"2025-10-23",""],[3,12532,"add train example for qwen-edit-plus","akk-123",4,"","",0,"2025-10-23",""],[3,12529,"Wan2.2 TI2V-5B Tiled VAE Tensor size mismatch","CalamitousFelicitousness",2,"","",0,"2025-10-22","2025-10-30"],[3,12505,"tgate error on img2img sd/sdxl pipelines","eyaler",4,"","",0,"2025-10-18",""],[3,12504,"freeU ComplexHalf warning","eyaler",0,"Ricardo-M-L","#13503",0,"2025-10-18","2026-04-21"],[3,12501,"VAE Decoder next(iter(..)) causes graph break","ppadjinTT",1,"ppadjinTT","#12512",0,"2025-10-17","2025-10-30"],[3,12499,"Wan 2.2 I2V condition shape mismatch","DefTruth",0,"yiyixuxu","#12523",1,"2025-10-17","2025-10-22"],[3,12494,"i think there is something wrong with new/latest scripts. RuntimeError: mat1 and mat2 shapes cannot be multipl","gerylavin",4,"","",0,"2025-10-15",""],[3,12492,"module transformers has no attribute CLIPFeatureExtractor","jiqing-feng",2,"","",0,"2025-10-15","2025-11-03"],[3,12489,"[Bug] CUDA kernel compatibility error (sm_120) — Diffusers ControlNet pipeline fails with “no kernel image is ","lingbin1964",2,"","",0,"2025-10-15","2025-10-29"],[3,12485,"How to enable Context Parallelism for training","liming-ai",2,"","",0,"2025-10-14","2025-10-15"],[3,12484,"persistent conflicts","lingbin1964",2,"","",0,"2025-10-14","2025-10-14"],[3,12483,"Dependency Conflict: diffusers 0.35.1 throws TypeError with transformers 4.57.0","lingbin1964",2,"","",0,"2025-10-14","2025-11-03"],[3,12482,"TypeError: CLIPTextModel got unexpected kwarg 'offload_state_dict'","lingbin1964",2,"","",0,"2025-10-14","2025-10-14"],[3,12481,"Bug] TypeError: CLIPTextModel.__init__() got unexpected keyword argument 'offload_state_dict' (Latest Diffs/Tr","lingbin1964",2,"","",0,"2025-10-14","2025-10-14"],[3,12480,"[Bug] TypeError: CLIPTextModel.__init__() got unexpected keyword argument 'offload_state_dict' (Latest Diffs/T","lingbin1964",2,"","",0,"2025-10-14","2025-11-07"],[3,12477,"Add VideoInpaintPipeline for temporally-consistent diffusion-based video inpainting","Aki-07",4,"","",0,"2025-10-13",""],[3,12470,"Add update_unexpected_keys and update_expected_keys APIs to DiffusersQuantizer","Disty0",2,"","",0,"2025-10-11","2025-11-29"],[3,12465,"Remote VAE feedback","Qurtison",4,"","",0,"2025-10-10",""],[3,12459,"FLUX.1-Kontext-dev batch inference throughput issue","Django-Jiang",4,"","",0,"2025-10-10",""],[3,12458,"[Qwen-image-edit] Batch Inference Issue / Feature Request","Django-Jiang",4,"","",0,"2025-10-10",""],[3,12457,"CineScale","bhack",2,"","",0,"2025-10-09","2025-10-09"],[3,12444,"[train_sana_sprint] Custom save_hook overwrites trained transformer with frozen model during checkpointing","riteshrm",4,"","",0,"2025-10-07",""],[3,12441,"Support Wan2.2-Animate","asomoza",0,"dg845","#12526",0,"2025-10-06","2025-11-13"],[3,12437,"Support OmniSVG","Meatfucker",2,"","",0,"2025-10-06","2025-10-07"],[3,12436,"TypeError: CLIPTextModel.__init__() got an unexpected keyword argument 'offload_state_dict'","zillur01",2,"","",0,"2025-10-06","2025-10-07"],[3,12433,"Documentation instructs installing diffusers from git causing version incompatibility with example requirement","moha23",0,"SahilCarterr","#12434",0,"2025-10-05","2025-10-06"],[3,12432,"get_1d_sincos_pos_embed_from_grid uses float64 unavailable on MPS","bghira",0,"Aishwarya0811","#12449",0,"2025-10-05","2025-10-23"],[3,12431,"[ko] Update outdated Korean documentation in docs/source/ko 🇰🇷<ko>","braintrue",4,"","",0,"2025-10-04",""],[3,12427,"UNet-only adapters will still cause problems with pipeline.set_adapters() after torch.compile because the comp","jiangtaozh",4,"","",0,"2025-10-03",""],[3,12423,"QwenImageTransformer2DModel.from_single_file","dxqb",2,"","",0,"2025-10-02","2025-10-03"],[3,12422,"Work WorkNCCL(SeqNum=2521, OpType=ALLREDUCE, NumelIn=160672064, NumelOut=160672064, Timeout(ms)=600000) timed ","hhy5861",4,"","",0,"2025-10-02",""],[3,12415,"SVG 2 kernels","bhack",4,"","",0,"2025-10-01",""],[3,12414,"Validation of results with quantitative evaluation metrics - Text to Image LORA","HripsimeS",2,"","",0,"2025-10-01","2026-01-10"],[3,12412,"HunyuanImage-3.0 support","ryan-seungyong-lee",4,"","",0,"2025-10-01",""],[3,12410,"WanVACEPipeline does not work with low-noise stage only","vladmandic",2,"","",0,"2025-09-30","2025-11-07"],[3,12409,"Diffusers crash due to hard-dependency on `torch._C._distributed_c10d`","vladmandic",0,"sayakpaul","#12420",1,"2025-09-30","2025-10-10"],[3,12406,"Generated image is cropped when using smaller output size to reduce VRAM and processing time in Flux Kontext","dangph-alala",4,"","",0,"2025-09-29",""],[3,12404,"Modular guiders should provide sane defaults","vladmandic",4,"","",0,"2025-09-28",""],[3,12402,"Eliminate code ambiguity in models.transformers.FluxAttention","j2gg0s",4,"","",0,"2025-09-28",""],[3,12401,"Redundant lru_cache in QwenEmbedRope","chenxiao111222",0,"dg845","#12452",0,"2025-09-28","2025-10-20"],[3,12396,"Using leaf_level together with delete_adapters can get some errors.","songh11",0,"adi776borate","#12639",0,"2025-09-26","2025-12-03"],[3,12386,"Implement missing features on ModularPipeline","vladmandic",4,"","",0,"2025-09-24",""],[3,12375,"What kernels should we integrate in Diffusers?","sayakpaul",4,"","",0,"2025-09-23",""],[3,12373,"Unlearn API for Diffusers","ksasi",4,"","",0,"2025-09-23",""],[3,12369,"Wan 2.1 CineScale - up to 4K","bhack",4,"","",0,"2025-09-22",""],[3,12368,"Add WanImage2ImagePipeline","vladmandic",4,"","",0,"2025-09-22",""],[3,12367,"Support for Qwen-Image Eligen Lora like Diffsynth Studio.","krigeta",4,"","",0,"2025-09-22",""],[3,12359,"Chroma pipeline documentation bug regarding the `guidance_scale` parameter","mingyi456",0,"SahilCarterr","#12360",0,"2025-09-21","2025-09-22"],[3,12358,"Add Lumina-DiMOO as a pipeline","qianyu-dlut",4,"","",0,"2025-09-20",""],[3,12355,"Support for SVD Training","Hiahia1369",2,"","",0,"2025-09-20","2026-01-09"],[3,12348,"WanImageToVideoPipeline produces tensor mismatch at some resolutions.","Meatfucker",4,"","",0,"2025-09-18",""],[3,12344,"Qwen Image: txt_seq_lens is redundant and not used","dxqb",0,"kashif","#12702",1,"2025-09-17","2026-01-12"],[3,12338,"`AutoencoderDC` bug with `pipe.enable_vae_slicing()` and decoding multiple images","mingyi456",1,"mingyi456","#12343",0,"2025-09-16","2025-09-22"],[3,12334,"a bug when i use enable_xformers_memory_efficient_attention()","zouzhekang",0,"SahilCarterr","#12337",0,"2025-09-16","2025-09-22"],[3,12329,"WanImageToVideoPipeline loads single-file transformer as Wan2.1 instead of Wan2.2","alfredplpl",2,"","",0,"2025-09-15","2026-01-09"],[3,12326,"Script to convert a SDXL PEFT trained lora to original format","frutiemax92",4,"","",0,"2025-09-13",""],[3,12323,"Add guidance free method as GFT","Bili-Sakura",4,"","",0,"2025-09-12",""],[3,12321,"HunyuanImage-2.1 support","kk3dmax",0,"yiyixuxu","#12333",1,"2025-09-12","2025-10-24"],[3,12319,"Broken group offloading using block_level","vladmandic",0,"rycerzes","#12692",0,"2025-09-11","2025-12-05"],[3,12311,"Flux DEV - bad output quality (bad seeds?)","Mahrkeenerh",4,"","",0,"2025-09-10",""],[3,12308,"Support `flash-attn` kernel support for non-Hopper GPUs","ParagEkbote",2,"","",0,"2025-09-10","2025-12-04"],[3,12300,"add OminiControl","Bili-Sakura",2,"","",0,"2025-09-08","2025-12-30"],[3,12299,"SD3 failing with tensor mismatch when using negative prompt embeddings & num_images_per_prompt > 1","danielamassiceti",2,"","",0,"2025-09-08","2026-01-09"],[3,12296,"Pixar-style red race car with big eyes and smile, slipping on banana peel, spinning, then dancing funny on rac","finofirdaus01-web",2,"","",0,"2025-09-06","2026-01-09"],[3,12295,"Qwen Image pipeline, Various","dxqb",4,"","",0,"2025-09-06",""],[3,12294,"[Qwen-image] encoder_hidden_states_mask is not used","JingyeChen-Canva",2,"","",0,"2025-09-06","2026-02-02"],[3,12292,"Why use torch.repeat instead of torch.repeat_interleave in train_dreambooth_lora_sdxl","Light-yzc",4,"","",0,"2025-09-05",""],[3,12284,"Attention test seems uncompatible with latest xformers","Jovp",4,"","",0,"2025-09-04",""],[3,12282,"Questions about  train_text_to_image_lora_sdxl.py","Light-yzc",3,"","",0,"2025-09-03","2025-09-03"],[3,12276,"The image is blurry.","sucessfullys",2,"","",0,"2025-09-03","2026-01-09"],[3,12273,"New HunyuanVideoI2V PR changes to transformer_hunyuan_video.py","eppaneamd",2,"","",0,"2025-09-03","2026-01-09"],[3,12270,"From 0.35.0: DLL load failed while importing libtriton: A dynamic link library (DLL) initialization routine fa","tin2tin",4,"","",0,"2025-09-02",""],[3,12257,"[Looking for community contribution] support Wan 2.2 S2V: an audio-driven cinematic video generation model","yiyixuxu",4,"","",0,"2025-08-29",""],[3,12255,"[BUG] Misleading ValueError when subclassing StableDiffusionImg2ImgPipeline with a mismatched __init__ signatu","BoostZhu",2,"","",0,"2025-08-28","2025-08-30"],[3,12253,"stabilityai/stable-diffusion-2-base is missing fp16 files' so that the correct variant file can be added","aryaIndia-dev",2,"","",0,"2025-08-28","2026-01-09"],[3,12241,"WAN2.1 FLF2V: Incorrect MASK Creation????","KyujinHan",4,"","",0,"2025-08-26",""],[3,12239,"Support for InfiniteTalk","supermeng",4,"","",0,"2025-08-26",""],[3,12235,"Support fun control variants for Wan","sayakpaul",4,"","",0,"2025-08-25",""],[3,12227,"[Feature] Integrate 'CHORDS: Diffusion Sampling Accelerator with Multi-core Hierarchical ODE Solvers'","Bili-Sakura",4,"","",0,"2025-08-24",""],[3,12224,"Option to cache text embedding vectors","innokean",2,"","",0,"2025-08-23","2026-01-09"],[3,12222,"[Contribution welcome] adding a fast test for Qwen-Image Controlnet Pipeline","yiyixuxu",2,"","",0,"2025-08-22","2025-08-25"],[3,12221,"[Looking for community contribution] support DiffSynth Controlnet in diffusers","yiyixuxu",4,"","",0,"2025-08-22",""],[3,12216,"Qwen-Image-Edit Inferior Results Compared to ComfyUI","enesmsahin",4,"","",0,"2025-08-22",""],[3,12210,"[Feature Request] Settability of Multiple Attention Types in Attention Dispatcher","tolgacangoz",2,"","",0,"2025-08-21","2025-08-21"],[3,12208,"Transformer2DModel onnx convertation problems","Voveka98",4,"","",0,"2025-08-21",""],[3,12202,"[Features] Planning to support TensorRT for Flux Pipelines","vuongminh1907",4,"","",0,"2025-08-21",""],[3,12198,"Add support for guidance-interval during sampling","Bili-Sakura",2,"","",0,"2025-08-20","2025-08-20"],[3,12195,"Version 0.35.0 AutoencoderKL failed with PyTorch 2.3.1: AutoAttributeError: module 'torch.library' has no attr","skedin",0,"Aishwarya0811","#12206",0,"2025-08-20","2025-08-23"],[3,12189,"When using the qwen-image model to generate images, diffusers takes twice as long as diffsynth-studio","gelinhua",2,"","",0,"2025-08-19","2026-01-09"],[3,12186,"`WanVACETransformer3DModel` does not inherit `AttentionMixin` while `WanTransformer3DModel` does","J4BEZ",2,"","",0,"2025-08-19","2025-09-09"],[3,12185,"What's the difference between DreamBooth LoRa and traditional LoRa?","MetaInsight7",2,"","",0,"2025-08-19","2026-01-09"],[3,12175,"Bad Image Quality generated by Qwen Image","vagitablebirdcode",2,"","",0,"2025-08-18","2025-08-21"],[3,12174,"'super' object has no attribute '__getattr__'","White-Friday",2,"","",0,"2025-08-18","2026-01-09"],[3,12173,"Wan i2i with latent upscale","luke14free",4,"","",0,"2025-08-18",""],[3,12165,"Failed to finetune the pre-trained model of 'stable-diffusion-v1-4' on image inpainting task","micklexqg",2,"","",0,"2025-08-17","2025-09-04"],[3,12162,"Not compatible with Flash ATTN 2.8.2","hargret",2,"","",0,"2025-08-16","2026-01-09"],[3,12159,"LoRA loading compatibility with some Qwen Image LoRAs","apolinario",0,"sayakpaul","#12170",1,"2025-08-15","2025-08-18"],[3,12157,"When training the Qwen-Image model using --bnb_quantization_config_path, the following error occurred: 'torch.","DarwinYang",0,"MengAiDev","#12178",0,"2025-08-15","2025-08-18"],[3,12156,"cool man","jcagentco-sudo",2,"","",0,"2025-08-15","2025-08-15"],[3,12155,"monsieur","jcagentco-sudo",2,"","",0,"2025-08-15","2025-08-15"],[3,12154,"dje","jcagentco-sudo",3,"","",0,"2025-08-15","2025-08-17"],[3,12149,"The diffusers0.34.0 installed by pip is not aligned with the source code","LTayfaker",2,"","",0,"2025-08-15","2025-08-26"],[3,12147,"Wan2.2 Lightning LoRA Loading Fails","ghunkins",2,"","",0,"2025-08-14","2025-08-19"],[3,12146,"Wan 2.1 / Wan2.2 Lightning & Sparse Distillation & Tiny VAE Acceleration Methods","ghunkins",4,"","",0,"2025-08-14",""],[3,12145,"resize_token_embeddings (?) (different result from new and old version)","Red-Owl",4,"","",0,"2025-08-13",""],[3,12141,"WanVACEPipeline doesn't work with bfloat16/foat16 vae","SlimRG",1,"SlimRG","#12143",0,"2025-08-13","2025-08-13"],[3,12139,"The IP Adapter provides different conditional guidance for each frame generated by AnimateDiff.","tlemangen",2,"","",0,"2025-08-13","2026-01-09"],[3,12138,"requirest for the model use","kjy09",4,"","",0,"2025-08-13",""],[3,12136,"How to use Diffusers to Convert Safetensors SDXL 1.0 to Onnx?","CypherpunkSamurai",2,"","",0,"2025-08-13","2025-10-31"],[3,12131,"inference with lora suddenly significantly slower","a-r-r-o-w",2,"","",0,"2025-08-12","2026-01-09"],[3,12126,"wan2.2 inference error Bug Report: `NoneType` object has no attribute `cache_context` when switching from `tra","Passenger12138",2,"","",0,"2025-08-12","2025-08-25"],[3,12125,"CompVis/stable-diffusion-v1-4 is missing fp16 files","thepreetam",2,"","",0,"2025-08-12","2026-01-09"],[3,12124,"For qwen-image training file, Maybe \"shuffle\" of dataloader should be \"False\" when custom_instance_prompts is ","yinguoweiOvO",4,"","",0,"2025-08-11",""],[3,12123,"\"Requested to load AutoencoderKL\"","ShadyRaion",2,"","",0,"2025-08-11","2025-08-17"],[3,12120,"How to train a lora with distilled flux model, such as flux-schnell???","Johnson-yue",2,"","",0,"2025-08-11","2026-01-09"],[3,12116,"Attention masking in Chroma pipeline","dxqb",0,"akshay-babbar","#12263",0,"2025-08-10","2025-09-29"],[3,12113,"WanImageToVideoPipeline: Given groups=1, weight of size [160, 12, 3, 3, 3], expected input[1, 3, 3, 258, 258] ","nitinmukesh",2,"","",0,"2025-08-10","2025-10-09"],[3,12112,"Various Chroma pipeline issues","dxqb",4,"","",0,"2025-08-10",""],[3,12110,"Bug in initialization of UNet1DModel GaussianFourier time projection","SammyAgrawal",1,"SammyAgrawal","#12111",0,"2025-08-10","2025-09-16"],[3,12108,"Qwen Image and Chroma pipeline breaks  using schedulers that enable flow matching by parameter.","Vargol",0,"a-r-r-o-w","#12109",0,"2025-08-09","2026-01-22"],[3,12107,"accelerator.init_trackers  error  when try with  a custom object such as list","micklexqg",4,"","",0,"2025-08-09",""],[3,12104,"IndexError: index 0 is out of bounds for dimension 0 with size 0","liushiton",2,"","",0,"2025-08-08","2025-08-17"],[3,12102,"Support training where width does not equal height for Qwen-Image","yinguoweiOvO",4,"","",0,"2025-08-08",""],[3,12098,"Qwen image transformers doesn't currently support from_single_file (i.e. GGUFs)","luke14free",2,"","",0,"2025-08-07","2025-08-08"],[3,12097,"Wan2.2 TI2V-5B VRAM OOM at the end","zhaoyun0071",4,"","",0,"2025-08-07",""],[3,12096,"WanVACEPipeline - doesn't work with apply_group_offloading","SlimRG",2,"","",0,"2025-08-07","2025-12-06"],[3,12094,"[Wan2.2] pipeline_wan miss the 'shift' parameter which used by Wan2.2-A14B-diffusers.","yvmilir",2,"","",0,"2025-08-07","2025-08-10"],[3,12085,"WanVACETransformer3DModel load gguf error","00Neil",2,"","",0,"2025-08-06","2025-08-06"],[3,12084,"Will `cosmos-transfer1` be supported in diffusers in the future?","rebel-shshin",4,"","",0,"2025-08-06",""],[3,12083,"Qwen-Image long prompt will cause error","akk-123",2,"","",0,"2025-08-06","2025-08-08"],[3,12082,"wan2.1 vae take more gpu memory after compile","gameofdimension",4,"","",0,"2025-08-06",""],[3,12080,"Qwen Image : Image Editing Inference","PranavAdlinge",2,"","",0,"2025-08-06","2025-08-06"],[3,12079,"API Suggestion: Expose Methods to Convert to Sample Prediction in Schedulers","dg845",4,"","",0,"2025-08-06",""],[3,12078,"Problem with provided example validation input in the Flux Control finetuning example","kzhang2",4,"","",0,"2025-08-05",""],[3,12075,"Qwen Image prompt encoding is not padding to max seq len","bghira",4,"","",0,"2025-08-05",""],[3,12071,"flux kontext transformer single blocks forward behavior changed","WingEdge777",2,"","",0,"2025-08-05","2025-08-05"],[3,12066,"Qwen Image incorrect device assignment during prompt encode","vladmandic",0,"yiyixuxu","#12099",1,"2025-08-04","2025-08-07"],[3,12065,"Qwen Image | Image-To-Image + Editing + Inpainting","ghunkins",2,"","",0,"2025-08-04","2025-08-04"],[3,12060,"Is there any DiT block defined in the huggingface/diffusers OR huggingface/transformers project?","JohnHerry",4,"","",0,"2025-08-04",""],[3,12053,"Flux1.Dev Kohya Loras text encoder layers no more supported","christopher5106",1,"christopher5106","#13188",0,"2025-08-03","2026-02-27"],[3,12052,"Wan 2.2 with LightX2V offloading tries to multiply tensors from different devices and fails","luke14free",2,"","",0,"2025-08-03","2025-08-08"],[3,12050,"`UNet2DConditionModel` : `qk_norm` setting in `config.json` is ignored","damian0815",4,"","",0,"2025-08-03",""],[3,12047,"Fusing Lightx2v lora on Wan2.2 GGUF fails","luke14free",2,"","",0,"2025-08-02","2025-08-06"],[3,12044,"AttributeError: 'bool' object has no attribute '__module__'. Did you mean: '__mod__'?","quyetnv98",2,"","",0,"2025-08-02","2025-08-21"],[3,12039,"Wan 2.2 VAE forward fails","a-r-r-o-w",2,"","",0,"2025-08-01","2025-08-06"],[3,12038,"Dataset structure for train_text_to_image_lora.py","HripsimeS",2,"","",0,"2025-07-31","2026-01-09"],[3,12037,"Wan 2.2 WanTransformer3DModel not compatible with Lightx2v self-forcing guidance distilled loras","luke14free",0,"sayakpaul","#12040",1,"2025-07-31","2025-08-02"],[3,12034,"Wan 2.2 5b i2v results poor quality compared to official Wan HF Space","okaris",0,"yiyixuxu","#12041",1,"2025-07-31","2025-08-01"],[3,12025,"Invalid API call in Cosmos VAE Encoder","akhilg-nv",0,"tanuj-rai","#12045",0,"2025-07-30","2025-08-02"],[3,12029,"X-Omni-En model won't load","Tylersuard",2,"","",0,"2025-07-30","2026-01-09"],[3,12022,"_flash_attention_3 in dispatch_attention_fn is not compatible with the latest flash-atten interface.","hmzjwhmzjw",0,"veeceey","#13173",0,"2025-07-30","2026-02-26"],[3,12019,"Wan 2.2 First vs Second stage","vladmandic",2,"","",0,"2025-07-29","2025-08-01"],[3,12017,"Dream 7B","ntoxeg",4,"","",0,"2025-07-29",""],[3,12012,"apply_first_block_cache with Wan 2.2 causes ValueError: No context is set. Please set a context before retriev","luke14free",0,"okaris","#12013",0,"2025-07-29","2025-07-29"],[3,12011,"Wan 2.2 a14b i2v OOM","okaris",4,"","",0,"2025-07-29",""],[3,12009,"WanTransformer3DModel.from_single_file wont load Wan2.2 GGUF (NotImplementedError: Cannot copy out of meta ten","luke14free",2,"","",0,"2025-07-29","2025-08-07"],[3,12007,"The fuse_lora() func in class PeftAdapterMixin did not take effect.","dujifeng",2,"","",0,"2025-07-29","2025-07-29"],[3,12003,"Cannot copy out of meta tensor; no data! Please use torch.nn.Module.to_empty() instead of torch.nn.Module.to()","MinhVD-ZenAI",4,"","",0,"2025-07-28",""],[3,11998,"Support training variance learning (Improved DDPM) with VLB loss.","bhosalems",3,"","",0,"2025-07-28","2025-08-02"],[3,11996,"Using AutoModel.from_pretrained for wan2.1 crash my notebook","Javierxx05",2,"","",0,"2025-07-27","2026-01-09"],[3,11992,"Problems with Kontext quantization","MinhVD-ZenAI",4,"","",0,"2025-07-25",""],[3,11989,"The inference image is pure black. This seems to be related to weight, but I have been unable to pinpoint wher","ShanZard",2,"","",0,"2025-07-24","2025-07-29"],[3,11984,"A compatibility issue when using custom Stable Diffusion with pre-trained ControlNets","ScienceLi1125",2,"","",0,"2025-07-24","2025-07-24"],[3,11981,"Groupoffloading introduce bad results","seed93",0,"a-r-r-o-w","#12077",0,"2025-07-23","2025-08-06"],[3,11978,"Add Bria 3.2 pipeline - next-generation commercial-ready text-to-image model","galbria",2,"","",0,"2025-07-23","2025-08-26"],[3,11977,"how to load a finetuned model especially during validation phase","micklexqg",2,"","",0,"2025-07-23","2026-01-09"],[3,11973,"train_dreambooth_lora_flux_kontext.py: error: the following arguments are required: --instance_prompt","huassaue",4,"","",0,"2025-07-22",""],[3,11971,"What is the minimum memory requirement for model training?","WWWPPPGGG",2,"","",0,"2025-07-22","2025-07-22"],[3,11966,"How about forcing the first and last block on device when groupoffloading is used?","seed93",4,"","",0,"2025-07-21",""],[3,11965,"Highdream multiple gpu to offload llama","grio43",2,"","",0,"2025-07-21","2026-01-09"],[3,11964,"KeyError when loading LoRA for Flux model: missing lora_unet_final_layer_adaLN_modulation_1 weights","NEWbie0709",4,"","",0,"2025-07-21",""],[3,11962,"FLUX.1-Kontext-dev Support for GGUF Quantized Model","sahandkh1419",2,"","",0,"2025-07-20","2025-08-02"],[3,11961,"New Adapter/Pipeline Request: IT-Blender for Creative Conceptual Blending","WonwoongCho",4,"","",0,"2025-07-20",""],[3,11959,"Passing enable_gqa flag in attention dispatcher should be version guarded","a-r-r-o-w",1,"a-r-r-o-w","#11967",0,"2025-07-19","2025-07-22"],[3,11957,"Flash/Sage varlen does not work with torch.compile","a-r-r-o-w",0,"ShivamShrirao","#14568",0,"2025-07-18","2026-09-20"],[3,11956,"Frequency-Decoupled Guidance (FDG) for diffusion models","Msadat97",0,"dg845","#11976",0,"2025-07-18","2025-08-07"],[3,11951,"Kontext model loading quantization problem","babyta",2,"","",0,"2025-07-18","2025-07-18"],[3,11948,"Impossible to load WanTransformer3DModel when offline using the 'from_pretrained' function.","guillaumejs2403",2,"","",0,"2025-07-17","2026-01-09"],[3,11946,"closed","SS-snap",2,"","",0,"2025-07-17","2025-07-23"],[3,11945,"Floating point exception with nightly PyTorch and CUDA","MxtAppz",2,"","",0,"2025-07-17","2026-01-09"],[3,11943,"ValueError when loading Lycoris safetensors LoRA: keys not correctly renamed","sss-okayu-korone",2,"","",0,"2025-07-16","2026-01-09"],[3,11942,"Kontext pipeline resizing image even if it is within PREFERRED_KONTEXT_RESOLUTIONS","nitinmukesh",2,"","",0,"2025-07-16","2025-08-02"],[3,11939,"train flux_controlnet code need to update","Johnson-yue",4,"","",0,"2025-07-16",""],[3,11938,"RMSNorm's weight not registered as submodules when initializing","Darkbblue",4,"","",0,"2025-07-16",""],[3,11930,"how to run convert_cosmos_to_diffusers.py correctly?","dedoogong",2,"","",0,"2025-07-15","2026-01-09"],[3,11928,"Support for Large attn_bias via Sparse Tensors or On‑The‑Fly Construction (seq_len ≈ 12 288)","maciek-wisniewski",2,"","",0,"2025-07-15","2025-07-15"],[3,11922,"FLUX.1-Fill-dev running problem","pique2233",4,"","",0,"2025-07-14",""],[3,11917,"SD3ControlNet","ZonglinL",2,"","",0,"2025-07-14","2025-07-14"],[3,11915,"Create modular pipeline from existing pipeline","vladmandic",0,"yiyixuxu","#11944",1,"2025-07-12","2025-08-28"],[3,11914,"Loading multiple LoRAs to 1 pipeline in parallel, 1 LoRA to 2-pipelines on 2-GPUs","vahe-toffee",2,"","",0,"2025-07-12","2025-07-15"],[3,11907,"flux pipeline not work with DPMSolverMultistepScheduler and UniPCMultistepScheduler","gameofdimension",1,"gameofdimension","#11908",0,"2025-07-10","2025-07-16"],[3,11903,"Fused QKV projections incompatible with training","bghira",4,"","",0,"2025-07-09",""],[3,11902,"SD3.5-Controlnet-8b does not use encoder_hidden_states","viv92",4,"","",0,"2025-07-09",""],[3,11901,"can not load Wan2.1-Fun-14B-InP-MPS_reward_lora_comfy.safetensors to wan2.1-14b-720p-diffusers","fzuo1230",4,"","",0,"2025-07-09",""],[3,11900,"stabilityai/stable-diffusion-2 is missing fp16 files","Parimala-15",2,"","",0,"2025-07-09","2025-07-09"],[3,11899,"stabilityai/stable-diffusion-2 is missing fp16 files","Parimala-15",2,"","",0,"2025-07-09","2026-01-09"],[3,11898,"stabilityai/stable-diffusion-2 is missing fp16 files","Parimala-15",2,"","",0,"2025-07-09","2025-07-09"],[3,11889,"Error during batch inference using FluxControlNetInpaintingPipeline","anandguptacv",4,"","",0,"2025-07-08",""],[3,11886,"Issue with FluxKontextPipeline","nitinmukesh",2,"","",0,"2025-07-08","2025-07-12"],[3,11885,"`load_lora_weights` bug in FLUX.1 `diffusers-0.35.0.dev0`","Chenzzzzzz217",2,"","",0,"2025-07-08","2026-01-10"],[3,11878,"WanVACETransformer3DModel with GGUF not working for 1.3B model","nitinmukesh",2,"","",0,"2025-07-07","2025-07-12"],[3,11874,"[lora] `exclude_modules` won't consider the modules names in `target_modules`","sayakpaul",1,"sayakpaul","#11999",1,"2025-07-07","2025-08-08"],[3,11872,"CPU Offloading Inefficiency: UNet/Transformer Repeatedly Moved Between CPU/GPU During Loops","JoeGaffney",2,"","",0,"2025-07-06","2025-08-07"],[3,11867,"Question about control sd3","Henry-Bi",2,"","",0,"2025-07-05","2025-07-14"],[3,11866,"[Docs] Visualizing the codebase for diffusers","ivanmilevtues",2,"","",0,"2025-07-05","2025-07-15"],[3,11865,"AttributeError: type object 'CosmosTransformer3DModel' has no attribute 'from_single_file'","mingyi456",2,"","",0,"2025-07-05","2025-07-11"],[3,11864,"AutoencoderDC.encode fails with torch.compile(fullgraph=True) - \"name 'torch' is not defined\"","SingleBicycle",2,"","",0,"2025-07-05","2025-07-09"],[3,11861,"ControlnetUnion not working correctly","asomoza",1,"asomoza","#11888",1,"2025-07-04","2025-07-10"],[3,11859,"STORK Scheduler","ZT220501",4,"","",0,"2025-07-04",""],[3,11853,"Conflix size when load controlnet+IP-Adapter when load stable-diffusion-xl-base-1.0","dangph-alala",2,"","",0,"2025-07-03","2025-07-03"],[3,11849,"Can not load fusionx_lora into original wan2.1-14b","fzuo1230",4,"","",0,"2025-07-02",""],[3,11846,"Finetuning DreamBoot Inpainting","kkewat",2,"","",0,"2025-07-01","2026-01-09"],[3,11840,"finetune flux kontext on custom dataset","universewill",4,"","",0,"2025-07-01",""],[3,11839,"Try load gguf for FluxKontextPipeline but invalid input size","dangph-alala",4,"","",0,"2025-07-01",""],[3,11833,"set_lora_device reports KeyError even though lora is loaded","hrazjan",0,"BenjaminBossan","#11844",1,"2025-06-30","2025-07-04"],[3,11830,"framepack pipeline return pure black output when using torch.float16","lllllinux",2,"","",0,"2025-06-30","2025-07-03"],[3,11828,"StableDiffusionXLControlNetUnionInpaintPipeline requires a float for controlnet_conditioning_scale & control_g","ewwwgiddings",2,"","",0,"2025-06-29","2026-01-09"],[3,11827,"FLUX Kontext. Transformers error: safetensors_rust.SafetensorError: Error while deserializing header: Metadata","tin2tin",2,"","",0,"2025-06-28","2025-06-29"],[3,11824,"Run FluxKontextPipeline with multi-images input, But An error occurred","veridone",0,"Net-Mist","#11880",0,"2025-06-28","2025-07-09"],[3,11821,"WanVACEPipeline: Passing a list of prompts is not yet supported. This may be supported in the future.","nitinmukesh",0,"albchim","#12251",0,"2025-06-27","2025-09-24"],[3,11819,"LoRA Training Script for Image Editing with FLUX Kontext","Eastforward",4,"","",0,"2025-06-27",""],[3,11817,"flux_control support flux kontext dev","universewill",2,"","",0,"2025-06-27","2025-07-01"],[3,11816,"set_adapters performance degrades with the number of inactive adapters","hrazjan",0,"BenjaminBossan","#11834",1,"2025-06-26","2025-06-30"],[3,11805,"Flash Attention","johnnynunez",2,"","",0,"2025-06-25","2025-06-25"],[3,11798,"Single File and GGUF support of Cosmos-Predict2","Vargol",0,"a-r-r-o-w","#11801",0,"2025-06-24","2025-07-01"],[3,11791,"can't load lora weights after apply group offloading","lllllinux",0,"a-r-r-o-w","#11804",0,"2025-06-24","2025-06-27"],[3,11790,"Bug: lora trained with train_dreambooth_lora_sdxl_advanced.py, the weight cannot be loaded","garychan22",4,"","",0,"2025-06-24",""],[3,11784,"Bug diffuser","Congofire",2,"","",0,"2025-06-23","2026-01-09"],[3,11781,"support for MeanFlow","prakashjayy",2,"","",0,"2025-06-23","2025-07-05"],[3,11779,"Incorrect mask padding in Wan VACE pipeline","bennyguo",1,"bennyguo","#11778",0,"2025-06-23","2025-06-23"],[3,11777,"Some question about training lora for HiDream-I1","yinguoweiOvO",2,"","",0,"2025-06-23","2026-01-09"],[3,11766,"Bug: set adapters when model is compiled","chenxiao111222",2,"","",0,"2025-06-22","2025-07-05"],[3,11764,"RepositoryNotFoundError on inpainting demo","abitrolly",1,"abitrolly","#1904",0,"2025-06-21","2026-01-09"],[3,11762,"Could you help fix the backdoor vulnerability caused by two risky pre-trained models used in this repo?","Rockstar292",2,"","",0,"2025-06-20","2026-01-09"],[3,11761,"Avoid creating tensor in CosmosAttnProcessor2_0","chenxiao111222",1,"chenxiao111222","#11763",0,"2025-06-20","2025-07-10"],[3,11752,"Failed to load lora in int8 mode","babyta",4,"","",0,"2025-06-19",""],[3,11749,"FlowMatchEulerDiscreteScheduler Timesteps have Floating Point Errors","JStyborski",4,"","",0,"2025-06-19",""],[3,11745,"Pipeline Quantization silent ignore of components not present in pipeline","asomoza",0,"sayakpaul","#11750",1,"2025-06-18","2025-06-20"],[3,11742,"Optional 'prompt_2' parameter in encode_prompt method is not optional","japppie",2,"","",0,"2025-06-18","2026-01-09"],[3,11734,"FLUX.1-dev Controlnet Training Script for Custom Dataset.","anandguptacv",2,"","",0,"2025-06-17","2025-06-18"],[3,11733,"ControlNet LoRA support more broadly across models","bghira",4,"","",0,"2025-06-17",""],[3,11732,"Regression? load_lora_weights ValueError: not enough values to unpack (expected 3, got 2)","tin2tin",2,"","",0,"2025-06-17","2025-06-17"],[3,11730,"Add `--lora_alpha` and metadata handling in training scripts follow up","linoytsaban",0,"imbr92","#11744",0,"2025-06-17","2025-06-23"],[3,11724,"Chroma fails to mask attention in the transformer","AmericanPresidentJimmyCarter",2,"","",0,"2025-06-16","2025-06-18"],[3,11721,"SanaPipeline crash on load","francois-rozet",2,"","",0,"2025-06-16","2025-06-20"],[3,11720,"add T5 lora train and infer for FluxPipeline","huangjun12",2,"","",0,"2025-06-16","2026-01-09"],[3,11714,"Reporting a vulnerability","nirajmodi19",2,"","",0,"2025-06-15","2025-06-20"],[3,11710,"[FR] Please support MagCache: Fast Video Generation with Magnitude-Aware Cache","nitinmukesh",2,"","",0,"2025-06-13","2025-10-09"],[3,11709,"CausVid LoRA key error when loading with diffusers","todochenxi",0,"a-r-r-o-w","#11726",0,"2025-06-13","2025-06-17"],[3,11702,"AccVid LoRA key error when loading with diffusers","apolinario",0,"a-r-r-o-w","#11704",0,"2025-06-12","2025-06-13"],[3,11700,"Add pruna integration for loading model through diffusers.from_pretrained / pipeline","davidberenstein1957",2,"","",0,"2025-06-12","2025-08-12"],[3,11694,"Convert VAE from latent-diffusion to diffusers","smandelli",2,"","",0,"2025-06-11","2025-06-11"],[3,11692,"<spam>","Gtvkun-ai",2,"","",0,"2025-06-11","2025-06-11"],[3,11691,"<spam>","Gtvkun-ai",2,"","",0,"2025-06-11","2025-06-11"],[3,11687,"[DOCS] Add `pruna` as optimization framework","davidberenstein1957",1,"davidberenstein1957","#11688",0,"2025-06-10","2025-06-16"],[3,11676,"HunyuanVideoImageToVideoPipeline memory leak","openSourcerer9000",4,"","",0,"2025-06-07",""],[3,11675,"Error in loading the pretrained lora weights","garychan22",2,"","",0,"2025-06-06","2025-06-07"],[3,11674,"[FR] Please support ref image and multiple control videos in Wan VACE","nitinmukesh",2,"","",0,"2025-06-06","2025-10-09"],[3,11661,"[BUG]: Using args.max_train_steps even if it is None in diffusers/examples/flux-control","Markus-Pobitzer",1,"Markus-Pobitzer","#11662",0,"2025-06-05","2025-06-05"],[3,11659,"LoRA load issue","D1-3105",4,"","",0,"2025-06-04",""],[3,11657,"Custom Wan diffusion Lora runs without error but doesn't apply effect and gives warning: No LoRA keys associat","st-projects-00",2,"","",0,"2025-06-04","2025-06-07"],[3,11653,"support .alpha keys in HiDream loras trained using OneTrainer","ali-afridi26",4,"","",0,"2025-06-03",""],[3,11651,"Can't load flux-fill-lora with FluxControl","hardikdava",2,"","",0,"2025-06-03","2026-01-09"],[3,11650,"max_shard_size","Ranking666",2,"","",0,"2025-06-03","2026-01-09"],[3,11648,"how to load lora weight  with fp8 transfomer  model?","Johnson-yue",4,"sayakpaul","#11655",1,"2025-06-03",""],[3,11644,"Error in init from pretrained for LTXConditionPipeline","Vickeyhw",2,"","",0,"2025-06-02","2025-07-09"],[3,11642,"prompt_embeds_scale in FluxPriorReduxPipeline seems to have no effect.","Meatfucker",2,"","",0,"2025-06-01","2026-01-09"],[4,5311,"[Jobs] Build a Job or Sandbox image from a Dockerfile","rycerzes",4,"","",0,"2026-11-06",""],[4,5309,"Dynamic Model Switching for Hugging Face MCPClient","HardikYerne",4,"","",0,"2026-11-05",""],[4,5304,"build_hf_headers(user_agent=\"\") produces an illegal User-Agent header with a trailing \"; \"","BramVanroy",3,"","",0,"2026-11-04","2026-11-05"],[4,5293,"update_webhook / hf webhooks update fail with 400 unless watched and domains are passed again","astroknot-sheep",0,"Wauplin","#5302",1,"2026-11-01","2026-11-04"],[4,5290,"InferenceClient ignores its token when fetching provider mappings","bodapatisaikrishna",3,"","",0,"2026-10-31","2026-11-04"],[4,5289,"[Download] Failed tree-cache writes leave temporary files behind","adenzhou1350",3,"","",0,"2026-10-31","2026-11-02"],[4,5283,"list_jobs() ignores the token and endpoint set on HfApi","mertkayacs",1,"mertkayacs","#5285",0,"2026-10-30","2026-10-30"],[4,5281,"HfFileSystem: \"x\"/\"xb\" mode overwrites existing files instead of failing","davanstrien",4,"","",0,"2026-10-29",""],[4,5280,"HfFileSystem: \"x\"/\"xb\" writes with block_size=0 are silently dropped","davanstrien",4,"","",0,"2026-10-29",""],[4,5274,"hf_raise_for_status raises AttributeError when an error body is a JSON array, string or null","v0ropaev",3,"","",0,"2026-10-29","2026-10-30"],[4,5261,"[Jobs] Option to run a job without outbound network access","rycerzes",4,"","",0,"2026-10-27",""],[4,5260,"[Sandbox] kill() suppresses termination failures and cleanup ownership","salma2vec",4,"","",0,"2026-10-26",""],[4,5256,"`hf cache verify --local-dir` reports the `.cache/huggingface/` download metadata as extra files","alphasafal",1,"alphasafal","#5257",0,"2026-10-25","2026-10-27"],[4,5254,"[CLI] hf cache verify crashes with \"Invalid value.\" on Windows when stdout is redirected","Sonica-B",1,"Sonica-B","#5255",0,"2026-10-25","2026-10-27"],[4,5253,"Renaming containing folder results in erroring","parasite6",3,"","",0,"2026-10-25","2026-10-28"],[4,5252,"[CLI] `hf cache prune --max-size`: keep the cache under a disk budget (LRU eviction)","kayzenline",3,"","",0,"2026-10-25","2026-10-28"],[4,5248,"try_to_load_from_cache allows path traversal and returns paths outside the cache directory","bodapatisaikrishna",1,"bodapatisaikrishna","#5249",0,"2026-10-23","2026-10-27"],[4,5245,"`HfFileSystem.find(file_path, detail=True)` returns an empty info dict, so datatrove readers crash on a single","ontdepart",1,"ontdepart","#5246",0,"2026-10-23","2026-10-27"],[4,5242,"Typed chat-completion tool calls lose `arguments` (`ChatCompletionInputToolCall.function` uses the tool-defini","canblmz1",3,"","",0,"2026-10-23","2026-10-27"],[4,5239,"try_to_load_from_cache misses cached files with a tilde cache_dir","AryanRG",3,"","",0,"2026-10-23","2026-10-27"],[4,5230,"Back to 429s","davidmezzetti",4,"","",0,"2026-10-21",""],[4,5226,"[Inference] URL inputs that return an HTTP error are sent to the model as content","RaphaelFakhri",3,"","",0,"2026-10-21","2026-10-21"],[4,5224,"[Serialization] Sharded load_torch_model ignores mmap and map_location","akshitkhokhani",3,"","",0,"2026-10-20","2026-10-27"],[4,5217,"[HfFileSystem] Preserve trailing slashes in glob patterns","parlorsky",3,"","",0,"2026-10-20","2026-10-20"],[4,5216,"[Serialization] Saving nested shared views fails with a full tensor","parlorsky",3,"","",0,"2026-10-20","2026-10-20"],[4,5215,"[Download] Dry-run misses downloads for outdated local files","parlorsky",0,"hanouticelina","#5219",1,"2026-10-20","2026-10-20"],[4,5211,"RepoCard.load / metadata_load ignore the metadata of a README.md that starts with a UTF-8 BOM, and saving corr","Blacksujit",0,"hanouticelina","#5220",1,"2026-10-20","2026-10-20"],[4,5210,"Webhook domain \"discussion\" is typed/offered as \"discussions\", so `hf webhooks create --domain` can't subscrib","Blacksujit",0,"hanouticelina","#5263",1,"2026-10-20","2026-10-27"],[4,5206,"MCPClient.add_mcp_server prints a \"Logging error\" when debug logging is enabled","guptaparin",3,"","",0,"2026-10-19","2026-10-20"],[4,5192,"[Cache] hf cache rm can delete another repo that has the same commit","mottopanikeiku",3,"","",0,"2026-10-18","2026-11-03"],[4,5191,"hf cache list drops a whole repo when one ref points to a missing commit hash","Harsh23Kashyap",3,"","",0,"2026-10-18","2026-10-19"],[4,5189,"`@strict` skips all type validation in modules that use `from __future__ import annotations`","SergeevDmitry",3,"","",0,"2026-10-18","2026-10-19"],[4,5186,"AsyncInferenceClient blocks the event loop while polling queue providers","digit50",0,"hanouticelina","#5200",1,"2026-10-17","2026-10-19"],[4,5184,"Offline hf_hub_download deletes the existing local_dir file and raises","digit50",0,"hanouticelina","#5199",1,"2026-10-17","2026-10-19"],[4,5183,"create_commit deletes a file when the same commit re-adds it unchanged","digit50",0,"hanouticelina","#5196",1,"2026-10-17","2026-10-19"],[4,5180,"`hf skills add --force` replaces the skill with a symlink to itself when `~/.claude/skills` links to `~/.agent","azank1",0,"hanouticelina","#5204",1,"2026-10-16","2026-10-20"],[4,5176,"HfMount.to_uri() silently changes mount paths ending in :ro or :rw","glaucojrcarvalho",3,"","",0,"2026-10-16","2026-10-19"],[4,5172,"HfFileSystem.url() returns a wrong folder URL when the repo id contains 'resolve'","LemonCondor",3,"","",0,"2026-10-15","2026-10-19"],[4,5170,"`RepoCard.from_template` reads the template with the locale encoding instead of UTF-8","albertvillanova",0,"hanouticelina","#5198",1,"2026-10-15","2026-10-19"],[4,5169,"InferenceClient keeps every non-streaming request and response alive until close()","spicysharkco",0,"hanouticelina","#5197",1,"2026-10-15","2026-10-19"],[4,5168,"Partial/Byte range file updates.","codecnotsupported",2,"","",0,"2026-10-14","2026-10-15"],[4,5160,"Limiting the number of dataset/base_model backlink","julien-blanchon",2,"","",0,"2026-10-14","2026-10-20"],[4,5158,"Use the Hub's safetensors-metadata endpoint in get_safetensors_metadata for sharded models","gary149",2,"","",0,"2026-10-14","2026-10-20"],[4,5154,"[Docs] Invalid Python syntax in Manage your Space examples","Ra1nhy",1,"Ra1nhy","#5156",0,"2026-10-12","2026-10-13"],[4,5145,"`hf cache` accounting omits the shared blob store: `prune` logs \"Saved 0.0.\" after freeing 113 GB, and store-o","Moutonc",0,"hanouticelina","#5148",1,"2026-10-11","2026-10-12"],[4,5139,"list_scheduled_jobs","guptaparin",1,"guptaparin","#5140",0,"2026-10-11","2026-10-12"],[4,5137,"HfFileSystem.find() / glob() skip folders after ls() when folder names share a prefix","anishmehta24",0,"hanouticelina","#5147",1,"2026-10-11","2026-10-12"],[4,5136,"InferenceClient raises httpx2.ReadTimeout instead of InferenceTimeoutError when the timeout expires","anishmehta24",0,"hanouticelina","#5150",1,"2026-10-11","2026-10-12"],[4,5135,"Several HfApi methods ignore HfApi(endpoint=...) and call huggingface.co","anishmehta24",0,"hanouticelina","#5149",1,"2026-10-11","2026-10-12"],[4,5130,"[Dataclasses] valid field validators with variadic kwargs are rejected","Ayushdevo",3,"","",0,"2026-10-09","2026-10-09"],[4,5129,"[Dataclasses] accept_kwargs treats init=False derived fields as required inputs","Ayushdevo",3,"","",0,"2026-10-09","2026-10-09"],[4,5128,"[Dataclasses] @strict silently skips field validation on frozen dataclasses","Ayushdevo",3,"","",0,"2026-10-09","2026-10-09"],[4,5127,"[Dataclasses] validate_typed_dict loses inherited key requiredness","Ayushdevo",3,"","",0,"2026-10-09","2026-10-09"],[4,5126,"[Dataclasses] @strict accepts float values for integer/bool Literals","Ayushdevo",3,"","",0,"2026-10-09","2026-10-09"],[4,5116,"Windows CMD: `uv run hf upload ... --exclude \".venv/**\"` expands exclusion glob into unexpected arguments","23xxCh",0,"hanouticelina","#5117",1,"2026-10-08","2026-10-08"],[4,5112,"snapshot_download(revision=<sha>) fails offline even when the files are cached","ChihebBENCHEIKH1",1,"ChihebBENCHEIKH1","#5113",0,"2026-10-07","2026-10-08"],[4,5111,"Attach images/videos to discussions and PRs (like gh --attach)","gary149",0,"hanouticelina","#5161",1,"2026-10-07","2026-10-21"],[4,5110,"ModelHubMixin cannot reload dataclass configs with init=False fields","LiamLin1130",3,"","",0,"2026-10-07","2026-10-08"],[4,5109,"Frequently raises file reconstruction error","O-O1024",3,"","",0,"2026-10-07","2026-11-04"],[4,5107,"[Dataclasses] validate_typed_dict breaks when Annotated is combined with NotRequired","bharadwaj-pendyala",3,"","",0,"2026-10-07","2026-10-08"],[4,5104,"[Dataclasses] @strict rejects every non-empty tuple for a bare typing.Tuple annotation","ishan-1010",3,"","",0,"2026-10-07","2026-10-09"],[4,5103,"blank","ishan-ins",3,"","",0,"2026-10-07","2026-10-07"],[4,5101,"Concurrent hf_hub_download calls fall back to HTTP because _get_version caches \"N/A\" during the lookup","BramVanroy",1,"BramVanroy","#5102",0,"2026-10-07","2026-10-07"],[4,5100,"`hf cache prune` / `hf cache rm` is quadratic in the number of files per revision (80 s for a 10k-file dataset","novain-shibu",0,"hanouticelina","#5118",1,"2026-10-07","2026-10-08"],[4,5098,"HfFileSystem caches duplicated, stale or partial directory listings","ChihebBENCHEIKH1",1,"ChihebBENCHEIKH1","#5099",0,"2026-10-06","2026-10-07"],[4,5088,"Add `--follow` to `hf jobs stats`","lhoestq",0,"hanouticelina","#5090",1,"2026-10-05","2026-10-06"],[4,5083,"`hf_hub_download` / `hf_hub_url` mishandle leading/trailing slashes in `subfolder`","kayzenline",3,"","",0,"2026-10-05","2026-10-05"],[4,5081,"/papers/{id}.md returns figure SVG content instead of the paper body for some papers","julien-blanchon",0,"mishig25","#20244",1,"2026-10-05","2026-10-05"],[4,5076,"face","ortstadtthomas91-coder",2,"","",0,"2026-10-03","2026-10-03"],[4,5072,"Make scheduled upload failures easier to inspect","mohi-devhub",3,"","",0,"2026-10-02","2026-10-05"],[4,5068,"RepoCard.validate() ignores HF_ENDPOINT and sends card content to huggingface.co","Ad1th",0,"Wauplin","#5069",1,"2026-10-02","2026-10-02"],[4,5065,"[CLI] --env-file / --secrets-file drop the first variable when the file starts with a UTF-8 BOM","nefayran",0,"hanouticelina","#5066",1,"2026-10-02","2026-10-02"],[4,5064,"hf download / snapshot_download report success when allow/ignore patterns match zero files","Ad1th",1,"Ad1th","#5067",0,"2026-10-02","2026-10-02"],[4,5062,"model_index_to_eval_results  crashes on empty list","vector15-05",1,"vector15-05","#5063",0,"2026-10-01","2026-10-02"],[4,5061,"Wrong exception type in load_torch_model","vector15-05",1,"vector15-05","#5063",0,"2026-10-01","2026-10-02"],[4,5049,"Share local upload and download metadata parsing","mohi-devhub",3,"","",0,"2026-09-30","2026-10-01"],[4,5048,"Support `cp_file`/`mv` for buckets in `HfFileSystem` (breaks write-then-rename libraries like `torch.distribut","davanstrien",0,"ankitkumartiwari-web","#5080",0,"2026-09-30","2026-10-29"],[4,5043,"[Bug] Saving a token overwrites malformed stored_tokens after parse error","ducanhnguyen223",3,"","",0,"2026-09-29","2026-09-30"],[4,5031,"[Dataclasses] @strict crashes with IndexError on bare typing.List / Dict / Set annotations","anishmehta24",1,"anishmehta24","#5032",0,"2026-09-29","2026-09-29"],[4,5030,"HF Jobs: h200x8 remains in SCHEDULING for six days without a start time or logs","elenaajayi",2,"","",0,"2026-09-28","2026-09-29"],[4,5019,"`hf datasets sql` returns mojibake or crashes on non-ASCII data when using the DuckDB CLI binary","825pranav",1,"825pranav","#5020",0,"2026-09-27","2026-09-28"],[4,5018,"Errors carry a wrong repo_id (e.g. \"my-model/commits\") for single-segment repo ids on most API routes","JoeyTan21",3,"","",0,"2026-09-27","2026-09-28"],[4,5014,"[i18n-vi] Translate the Hugging Face Hub documentation to Vietnamese","ducanhnguyen223",4,"","",0,"2026-09-26",""],[4,5013,"`stored_tokens` uses the locale encoding: a non-ASCII token name wipes saved tokens, old files break `get_toke","ppcvote",0,"Evolian-o","#5021",0,"2026-09-26","2026-09-28"],[4,5012,"A scalar tags value in a card's front matter is split into one-letter tags and written back","Yasser-Ameur",0,"zeenat28-ui","#5017",0,"2026-09-26","2026-09-28"],[4,5011,"Inference Endpoints: pendingRequests autoscaling never triggers on our AWS endpoints, while hardwareUsage scal","lrdsouza",2,"","",0,"2026-09-25","2026-11-05"],[4,5004,"Bucket sync/ls include/exclude patterns are case-insensitive on Windows","anishmehta24",0,"Wauplin","#5007",1,"2026-09-24","2026-09-25"],[4,5002,"Error repo_id includes the query string, and is missing for download URLs","anishmehta24",0,"Wauplin","#5008",1,"2026-09-24","2026-09-25"],[4,5000,"copy_files to a bucket treats a missing destination as a directory when a lexical sibling exists","AnishPatel526",1,"AnishPatel526","#5001",0,"2026-09-24","2026-09-25"],[4,4987,"Unexpected 401s when listing with generated JWTs","rshkv",2,"","",0,"2026-09-23","2026-09-28"],[4,4975,"eval_results_to_model_index silently drops dataset_args, source and task_name when results share a task/datase","AnishPatel526",3,"","",0,"2026-09-23","2026-09-23"],[4,4972,"`restart_space` / `set_space_sleep_time` / `request_space_hardware` return 401 with a token from `hf auth logi","kuotunyu",2,"","",0,"2026-09-22","2026-09-24"],[4,4961,"Legacy warning message using cli upload","Anthonyg5005",2,"","",0,"2026-09-22","2026-09-22"],[4,4956,"[MCP] MCPClient only discovers the first tools/list page","Yang1107-wzy",0,"Kayvan-Zahiri","#4960",0,"2026-09-21","2026-09-22"],[4,4952,"[i18n-KN] Translating docs to Kannada","Prajwal-chikkur07",4,"","",0,"2026-09-21",""],[4,4947,"Add Boollm to Hugging Face Local Apps","saz3light",4,"","",0,"2026-09-19",""],[4,4946,"tabulate() pads columns by code points, not display columns: every table skews on CJK/emoji values","feiiiiii5",3,"","",0,"2026-09-19","2026-09-21"],[4,4927,"[CLI] --json / -q / --format are silently swallowed on hf jobs run and uv run","davanstrien",0,"Wauplin","#4936",1,"2026-09-17","2026-09-18"],[4,4922,"scan_cache_dir reports PR refs as refs\\pr\\1 on Windows, so hf cache filters on refs/pr/1 match nothing","itzzdev09",0,"gokay-ai","#4925",0,"2026-09-16","2026-09-18"],[4,4915,"Race in are_symlinks_supported(): snapshot_download(max_workers>1) fails with WinError 1314 on Windows without","Mega-Gorilla",0,"Wauplin","#4923",1,"2026-09-15","2026-09-17"],[4,4912,"Windows: files named with trailing dot/space are silently renamed and can collide, losing data","warter666",3,"","",0,"2026-09-15","2026-09-18"],[4,4911,"CSOAI — Hub interop: model governance","CSOAI-ORG",3,"","",0,"2026-09-15","2026-09-15"],[4,4910,"CSOAI — Hub interop: model governance","CSOAI-ORG",3,"","",0,"2026-09-15","2026-09-15"],[4,4909,"CSOAI — HuggingFace Hub AI governance","CSOAI-ORG",3,"","",0,"2026-09-15","2026-09-15"],[4,4907,"still exists Error 416 when copying empty files from a to a bucket on the GUI  #4353","1qaz2wsx3edc4rfv5tgb6yhn",2,"","",0,"2026-09-14","2026-09-23"],[4,4904,"Unnecessary HTTP request to https://huggingface.co/api/agent-harnesses","flange-ipb",2,"","",0,"2026-09-14","2026-09-14"],[4,4901,"[Serialization] save_torch_state_dict raises KeyError on uint16/32/64, newer float8/float4 and complex64 tenso","yupengtang",0,"Wauplin","#4902",1,"2026-09-14","2026-09-14"],[4,4897,"Recognize proposed U3 Safetensors metadata","7dsolv",2,"","",0,"2026-09-13","2026-09-14"],[4,4895,"[Windows] Tree cache disabled by long local_dir paths","askalf",1,"askalf","#4896",0,"2026-09-13","2026-09-14"],[4,4893,"`hf download` throughput collapses on high-bandwidth link with default concurrency","iSOcH",4,"","",0,"2026-09-13",""],[4,4888,"`InferenceClient.tabular_regression` is unreachable: missing from the `hf-inference` task registry","rishabhsai",3,"","",0,"2026-09-12","2026-09-14"],[4,4887,"`InferenceClient.get_endpoint_info` targets `api-inference.huggingface.co`, which no longer resolves","rishabhsai",3,"","",0,"2026-09-12","2026-09-14"],[4,4880,"[Commit] Support duck-typed binary file streams in CommitOperationAdd","Saniyagupte",3,"","",0,"2026-09-11","2026-09-11"],[4,4872,"[Spaces] AttributeError when Space API returns hardware: null in SpaceRuntime","Saniyagupte",1,"Saniyagupte","#4873",0,"2026-09-10","2026-09-11"],[4,4869,"[RepoCard] AttributeError when model-index has `source: null`","Saniyagupte",1,"Saniyagupte","#4870",0,"2026-09-10","2026-09-11"],[4,4866,"[Utils] Use stripped string for integer parsing in _parse_with_unit","Saniyagupte",2,"","",0,"2026-09-10","2026-09-10"],[4,4863,"[Community] Fix incorrect docstring for DiscussionComment.last_edited_by","Saniyagupte",1,"Saniyagupte","#4864",0,"2026-09-10","2026-09-10"],[4,4861,"`get_session()` is not thread-safe","HaukurPall",0,"hanouticelina","#4868",1,"2026-09-10","2026-09-11"],[4,4860,"[CLI] `auto` mode picks `agent` for Warp users: progress bars and colors silently disabled (TERM_PROGRAM=WarpT","xbsheng",1,"xbsheng","#2476",0,"2026-09-10","2026-09-23"],[4,4853,"[Serialization] Saving a state dict removes checkpoint backup files","sueun-dev",0,"g0rdonL","#4857",0,"2026-09-10","2026-09-10"],[4,4850,"[Sandbox] Preserve terminal results when the HTTP tail truncates","evalstate",1,"evalstate","#4851",1,"2026-09-09","2026-09-10"],[4,4821,"CommitInfo cannot be copy.deepcopy-ed or pickled (TypeError in __new__)","canblmz1",0,"Wauplin","#4822",1,"2026-09-07","2026-09-07"],[4,4819,"`hf update` on a Windows pip install fails with WinError 32 and can leave the installation broken","alex-feel",0,"Wauplin","#4823",1,"2026-09-06","2026-09-08"],[4,4815,"hf_hub_download: the body GET is not pinned to the commit resolved by the metadata HEAD call (cross-version ca","AUTHENSOR",3,"","",0,"2026-09-05","2026-09-08"],[4,4812,"Sandbox.create: support caller-supplied ownership labels on dedicated Jobs","evalstate",0,"Wauplin","#4824",1,"2026-09-04","2026-09-08"],[4,4802,"[httpx] Let's migrate to httpx2!","Wauplin",2,"","",0,"2026-09-03","2026-10-13"],[4,4796,"hf auth login aborts on transient DNS failure: whoami has no connection retry","JSBtechnologies",2,"","",0,"2026-09-02","2026-09-03"],[4,4795,"[Inference Providers] Add IteraCompute provider support","K-3-LT",3,"","",0,"2026-09-02","2026-09-30"],[4,4793,"Redirect host check accepts an https -> http downgrade, so the auth token can be sent over plaintext","Puneet15556",3,"","",0,"2026-09-01","2026-09-02"],[4,4788,"`hf repos ls` exits 1 with empty stdout when a repo has null storage","MohammedAlkindi",2,"","",0,"2026-08-31","2026-09-02"],[4,4787,"CI: no `*)` arm in python-tests.yml, so a matrix rename can run zero tests and still pass","MohammedAlkindi",3,"","",0,"2026-08-31","2026-09-01"],[4,4786,"`is_xet_available()` no longer consults the server, so a self-hosted Hub cannot negotiate down to LFS","jwschmo",3,"","",0,"2026-08-31","2026-09-02"],[4,4777,"hf cache xet: Commands to clear/prune the Xet cache","Arcitec",3,"","",0,"2026-08-30","2026-09-21"],[4,4775,"CI: torch_1.11 lane has not installed torch since the matrix rename in #4008","MohammedAlkindi",1,"MohammedAlkindi","#4776",0,"2026-08-29","2026-08-31"],[4,4771,"Add Scalattice as an Inference Provider","romulushill",3,"","",0,"2026-08-28","2026-08-28"],[4,4768,"hf download doesn't work properly","XeonG",3,"","",0,"2026-08-27","2026-09-28"],[4,4757,"feat: --watch flag for `hf buckets sync` (keep a bucket mirrored to a local directory)","morphym",3,"","",0,"2026-08-27","2026-09-08"],[4,4754,"MacOS Installation - hf CLI","AlisinaDevelo",0,"hanouticelina","#4758",1,"2026-08-27","2026-08-27"],[4,4750,"load_dotenv truncates unquoted values at a bare '#' (inline comment should require preceding whitespace)","eeshsaxena",3,"","",0,"2026-08-26","2026-08-26"],[4,4749,"`hf buckets remove --recursive` deletes lexical siblings outside the requested prefix (missing path-boundary f","CorsenAI",0,"Wauplin","#4804",1,"2026-08-26","2026-09-08"],[4,4742,"Removed `labels`/`tags` args still documented, 3 API reference examples raise TypeError","aadi-joshi",0,"Wauplin","#4745",1,"2026-08-26","2026-08-26"],[4,4741,"Downloads fail with 'Distant resource does not have a Content-Length' when proxy strips HEAD Content-Length (H","p0we7",0,"Wauplin","#4805",1,"2026-08-26","2026-09-08"],[4,4738,"[CLI] `hf-cli` SKILL.md links to a removed hf-mount install script","ackalker",0,"Wauplin","#4747",1,"2026-08-25","2026-08-26"],[4,4736,"just a minor bug parse_size_to_int in src/huggingface_hub/serialization/_base.py","ShinapriLN",2,"","",0,"2026-08-25","2026-08-25"],[4,4733,"MCPClient StreamableHTTP import breaks with mcp>=2.0.0 (streamablehttp_client renamed to streamable_http_clien","nerdai",0,"Wauplin","#4735",1,"2026-08-25","2026-08-25"],[4,4729,"The `httpx` package is no longer actively maintained and has been succeeded by `httpx2` (a separate package st","sareetamugde-arch",2,"","",0,"2026-08-24","2026-08-25"],[4,4726,"[i18n-te] Translating docs to Telugu","prithamnandavaram",2,"","",0,"2026-08-23","2026-08-23"],[4,4725,"Safetensors metadata does not detect tensors referenced by index but absent from shards","intelland",3,"","",0,"2026-08-23","2026-08-26"],[4,4722,"Segmentation fault in hf_xet when encountering 429 Too Many Requests","romain-bsn",2,"","",0,"2026-08-22","2026-08-27"],[4,4717,"Xet downloads fall back to HTTP in PyInstaller-packaged applications","ia319",2,"","",0,"2026-08-21","2026-08-21"],[4,4710,"Anonymous S3 access to public buckets","jdddog",4,"","",0,"2026-08-21",""],[4,4709,"hf upload <repo> '<wildcard>' silently uploads nothing (regression since 1.0)","shoutoutuoadi325",1,"shoutoutuoadi325","#4711",0,"2026-08-21","2026-08-26"],[4,4707,"Ai video","ahmadtanoli335-rgb",3,"","",0,"2026-08-20","2026-08-20"],[4,4703,"InferenceClient: no retry on 429/5xx, and the 429 error message drops the provider name","gary149",2,"","",0,"2026-08-18","2026-08-18"],[4,4698,"get_hf_file_metadata can report a redirect body's Content-Length as the file size","assafvayner",1,"assafvayner","#4699",1,"2026-08-17","2026-08-18"],[4,4697,"Aliansa","alanmauriciograndelara3-alt",3,"","",0,"2026-08-17","2026-08-18"],[4,4696,"[Buckets] No way to change bucket visibility after creation","davanstrien",0,"hanouticelina","#4715",1,"2026-08-17","2026-08-25"],[4,4688,"ResolvedRevision silently becomes \"main\" after pickling","yeasah",0,"hanouticelina","#4692",1,"2026-08-16","2026-08-17"],[4,4686,"[i18n-fa] Translating docs to Persian","mmdverse",4,"","",0,"2026-08-16",""],[4,4685,"test_utils_headers auth tests fail when run under an AI coding agent (agent/<id> appended to user-agent)","MohammedAlkindi",3,"","",0,"2026-08-16","2026-08-17"],[4,4684,"Windows: test_cache_no_symlinks tests require real symlink support and fail on a machine that actually cannot ","MohammedAlkindi",3,"","",0,"2026-08-16","2026-08-17"],[4,4680,"[i18n-BN] Translating docs to Bengali","talha937",4,"","",0,"2026-08-16",""],[4,4676,"German docs still document Repository, removed in v1.0 (English docs were updated)","melbinjp",0,"OllieinCanada","#4679",0,"2026-08-15","2026-08-17"],[4,4674,"Replace httpx with httpx2","chinu0609",3,"","",0,"2026-08-14","2026-08-17"],[4,4673,"get_session() returns a permanently unusable client once the global client is closed","GiulioDER",3,"","",0,"2026-08-14","2026-08-17"],[4,4670,"httpx.ReadTimeout error during snapshot_download() without XET (export HF_HUB_DISABLE_XET=1) can't recover","Nekotekina",0,"Wauplin","#4826",1,"2026-08-14","2026-09-08"],[4,4665,"`hf download REPO --include A B C` silently drops A and exits 0","sohampanda000",3,"","",0,"2026-08-13","2026-08-14"],[4,4663,"`hf download REPO --include A B C` silently drops A and exits 0","sohampanda000",3,"","",0,"2026-08-13","2026-08-13"],[4,4654,"[i18n-tr] Translating docs to Turkish","onatozmenn",4,"","",0,"2026-08-12",""],[4,4653,"Empty allow_patterns list silently matches nothing (upload_folder/snapshot_download transfer zero files)","shashvat-singham",3,"","",0,"2026-08-12","2026-08-14"],[4,4646,"Xet downloads ignore `tqdm_class` for the transfer bar","bharadwaj-pendyala",1,"bharadwaj-pendyala","#4647",0,"2026-08-08","2026-08-10"],[4,4638,"[Inference Endpoints] Add hf endpoints hardware to list available instances","davanstrien",0,"hanouticelina","#4672",1,"2026-08-05","2026-08-18"],[4,4637,"snapshot_download fails with misleading LocalEntryNotFoundError when HF_ENDPOINT mirror 308-redirects to the H","vollegrewar",0,"hanouticelina","#4739",1,"2026-08-05","2026-08-28"],[4,4634,"[CLI] load_dotenv drops multi-line quoted values in --env-file / --secrets-file","LuShadowX",3,"","",0,"2026-08-05","2026-08-11"],[4,4633,"A `tqdm_class` that follows the documented contract shows almost no progress on Xet downloads","tobocop2",2,"","",0,"2026-08-05","2026-08-10"],[4,4632,"[BUG] Cancelling one Xet download aborts every other download, and the only way to do it is private API","tobocop2",2,"","",0,"2026-08-05","2026-08-17"],[4,4629,"INFERENCE_ENDPOINT_IMAGE_KEYS is missing newer engine image variants (vLLM, vLLMNeuron, sGLang, hfServe)","gary149",0,"hanouticelina","#4671",1,"2026-08-04","2026-08-14"],[4,4624,"snapshot_download: no way to require that the cached snapshot was actually checked","eaubin",3,"","",0,"2026-08-02","2026-08-14"],[4,4620,"Download struck at 0 bytes","KKNakka2026",2,"","",0,"2026-07-31","2026-09-30"],[4,4611,"[bug] filenotfounderror when using tilde (~/) paths in commitoperationadd","Saniyagupte",1,"Saniyagupte","#4612",0,"2026-07-29","2026-07-30"],[4,4609,"[CLI] `hf version` crashes on Windows when stdout isn't UTF-8 (charmap can't encode ✓)","Wauplin",1,"Wauplin","#4610",1,"2026-07-29","2026-07-30"],[4,4602,"parse_safetensors_file_metadata truncates headers between 99,994 and 100,000 bytes due to an HTTP Range bounda","joe-nv",0,"Wauplin","#4603",1,"2026-07-29","2026-07-29"],[4,4596,"[feat] BLAKE3 hash support","vyrti",2,"","",0,"2026-07-27","2026-07-28"],[4,4594,"S3 GetObject returns 302 redirect for aws-sdk-js clients - please proxy directly like aws-cli/botocore","KinaraPandy",2,"","",0,"2026-07-27","2026-08-25"],[4,4592,"[i18n-TR] Translating docs to Turkish","devemree",2,"","",0,"2026-07-27","2026-09-24"],[4,4591,"[i18n-MR] Translating docs to Marathi","kanaad-lims",2,"","",0,"2026-07-27","2026-10-29"],[4,4582,"[Upload] Commit batches have no size budget for xet files","ishan-1010",3,"","",0,"2026-07-26","2026-07-27"],[4,4581,"HAHA! YOU MESSED IT UP! YOU CAN'T GO BACK NOW...","MICHIFUBRUNO",3,"","",0,"2026-07-25","2026-07-27"],[4,4580,"YOU CANNOT BEAT US","MICHIFUBRUNO",3,"","",0,"2026-07-25","2026-07-27"],[4,4579,"YOU CANNOT BEAT REAL CODERS","MICHIFUBRUNO",3,"","",0,"2026-07-25","2026-07-27"],[4,4574,"hugingface_hub","abcdbdbdbbd-prog",3,"","",0,"2026-07-23","2026-07-23"],[4,4573,"hugingface_hub","abcdbdbdbbd-prog",3,"","",0,"2026-07-23","2026-07-23"],[4,4571,"`upload_folder` pipelined upload loses token in `_final_commit_info` when no files changed","xiaoyao9184",0,"Wauplin","#4572",1,"2026-07-23","2026-07-23"],[4,4558,"[Jobs] Proposal: optional [tool.hf-jobs] table in PEP 723 headers so a script's runtime config travels with it","davanstrien",2,"","",0,"2026-07-21","2026-09-30"],[4,4544,"Link to a compromised website in the docs","asrelo",0,"Wauplin","#4545",1,"2026-07-17","2026-07-17"],[4,4536,"Streaming crashes with JSONDecodeError on \"data:[DONE]\" (no space after colon)","ritsth",3,"","",0,"2026-07-16","2026-07-17"],[4,4535,"hf upload fails on existing static Spaces with 402 Payment Required","yosun",0,"Wauplin","#4539",1,"2026-07-16","2026-07-17"],[4,4533,"code","Uditya31",3,"","",0,"2026-07-16","2026-07-17"],[4,4528,"Slow model download speed from Hugging Face Hub (1.11 GB model takes ~20 minutes on multiple devices)","PhonSobon",2,"","",0,"2026-07-14","2026-11-04"],[4,4522,"hf buckets sync 1.8.0 fails large multi-file apply with hf_xet File exists","marksverdhei",2,"","",0,"2026-07-12","2026-07-13"],[4,4520,"`hf download` stucks after a while on large download","RemiLacroix-IDRIS",2,"","",0,"2026-07-11","2026-08-31"],[4,4519,"`hf download` seems to report incorrect bit rates","RemiLacroix-IDRIS",0,"rajatarya","#4530",1,"2026-07-10","2026-07-15"],[4,4518,"`hf download`: file count not updating during download","jchwenger",0,"Wauplin","#4560",1,"2026-07-10","2026-07-22"],[4,4511,"[CLI] Expose `pipeline_tag` filter in `hf models ls`?","davanstrien",0,"Wauplin","#4512",1,"2026-07-09","2026-07-09"],[4,4508,"Unable to download models with huggingface","ri938",2,"","",0,"2026-07-08","2026-08-31"],[4,4507,"Clarification on `upload_folder` fallback behavior when the server doesn't support XET","achoudharyji",3,"","",0,"2026-07-08","2026-07-21"],[4,4488,"[UX] Symlink warning says \"degraded version\" but caching is fully functional, could confuse Windows users","harshitayadavv",3,"","",0,"2026-07-06","2026-07-06"],[4,4487,"[Bug] hf_hub_download ignores local cache in offline mode and on connection failure — raises before checking c","Yanivcv",2,"","",0,"2026-07-05","2026-07-06"],[4,4486,"[Bug] hf_hub_download ignores local cache in offline mode and on connection failure — raises before checking c","Yanivcv",3,"","",0,"2026-07-05","2026-07-06"],[4,4485,"hf_xet download hangs indefinitely after huggingface_hub.login(token=...), but succeeds when unauthenticated (","dxqb",2,"","",0,"2026-07-05","2026-07-06"],[4,4484,"Transformers ,What can they do.","KrishLakhanpal",2,"","",0,"2026-07-04","2026-07-04"],[4,4476,"`hf jobs --timeout` help says \"int/float\" but float durations raise ValueError","Sreekant13",2,"","",0,"2026-07-02","2026-07-03"],[4,4475,"HF_ENDPOINT but for xet ?","Disservin",2,"","",0,"2026-07-02","2026-07-03"],[4,4467,"A-15, AI&ML","grs7563",3,"","",0,"2026-07-01","2026-07-02"],[4,4458,"hf CLI download command given by \"Download with hf CLI\" button gives bad syntax","MackTuesday",2,"","",0,"2026-07-01","2026-07-01"],[4,4452,"`hf download`: per-file progress bar (to help with indefinite hangs for some models)","jchwenger",2,"","",0,"2026-06-30","2026-07-07"],[4,4451,"parse_hf_uri accepts custom endpoints on the wrong port","nyxst4ck",3,"","",0,"2026-06-30","2026-06-30"],[4,4445,"aminspeal ai","mmohammedthecaesur-pixel",3,"","",0,"2026-06-29","2026-06-29"],[4,4442,"AI withAW","vrew-wali",3,"","",0,"2026-06-29","2026-06-29"],[4,4441,"[CLI] Filter hf models ls by inference provider?","davanstrien",0,"moon-bot-app","#4497",2,"2026-06-29","2026-07-08"],[4,4440,"Remove upper cap on typer dependency","xsuchy",0,"hanouticelina","#4462",1,"2026-06-29","2026-07-03"],[4,4434,"`filter_repo_objects` pattern matching is case-insensitive on Windows but case-sensitive on Linux","Sreekant13",1,"Sreekant13","#4435",0,"2026-06-29","2026-06-29"],[4,4431,"Auto Reject Injected Token in Environment","SHRIJANAGAIN",3,"","",0,"2026-06-28","2026-06-29"],[4,4430,"Multiple Public dataset LFS files return 403 AccessDenied from cdn-lfs-us-1.hf.co despite visible metadata","Aaryan-Kapoor",2,"","",0,"2026-06-28","2026-08-25"],[4,4422,"[Build] `huggingface_hub.templates` absent from setuptools `packages` configuration","HaleTom",0,"Wauplin","#4438",1,"2026-06-27","2026-06-29"],[4,4421,"[Tests] `TestConfigDictNotRequired` has `__init__` causing PytestCollectionWarning","HaleTom",2,"","",0,"2026-06-27","2026-07-02"],[4,4420,"hf cache list reports incorrect total disk usage when a cached model repo is omitted","HaleTom",2,"","",0,"2026-06-27","2026-07-20"],[4,4419,"hf cache verify can find a cached model repo that hf cache list omits","HaleTom",2,"","",0,"2026-06-27","2026-07-20"],[4,4417,"Migrate legacy setup.py metadata to pyproject.toml (PEP 621/639) — fixes setuptools_scm + deprecated license c","HaleTom",3,"","",0,"2026-06-26","2026-06-29"],[4,4412,"hf cache prune: add --incomplete flag to delete orphaned partial-download .incomplete files","HaleTom",0,"Wauplin","#4416",1,"2026-06-26","2026-06-29"],[4,4411,"feat(cli): support `hf discussions edit`","Pierrci",0,"Wauplin","#4415",1,"2026-06-26","2026-06-26"],[4,4397,"`HfFileSystemStreamFile` (block_size=0): stale connection mid-stream causes unhandled `RemoteProtocolError` an","Mattral",2,"","",0,"2026-06-23","2026-06-23"],[4,4390,"[Feature Request] Native bounce compression support for model downloads — resolves #1446","infosave2007",2,"","",0,"2026-06-21","2026-06-22"],[4,4389,"hhjhjhj","achi6000",3,"","",0,"2026-06-20","2026-06-20"],[4,4384,"Circular import on `from huggingface_hub import login`?","ianxmason",0,"hanouticelina","#4385",1,"2026-06-18","2026-06-18"],[4,4377,"get_safetensors_metadata can hang indefinitely: no timeout on shard Range requests","go-bai",1,"go-bai","#4378",0,"2026-06-18","2026-07-20"],[4,4376,"HF Jobs h200x8 job remains in SCHEDULING for hours with repeated \"Job started\" log markers but user command ne","elenaajayi",2,"","",0,"2026-06-17","2026-08-31"],[4,4374,"Add --sort and --limit flags to hf jobs ps","fuleinist",2,"","",0,"2026-06-17","2026-09-23"],[4,4369,"[CLI] hf download doesn't parse hf:// URI for model repos — treats entire URI as repo_id (v1.17.0)","Wontfallo",2,"","",0,"2026-06-16","2026-06-17"],[4,4365,"`downloadsAllTime` Decreasing Day-over-Day","ZMXJJ",2,"","",0,"2026-06-16","2026-06-16"],[4,4362,"upload / upload-large-folder should batch by file size, not file count","fulldecent",2,"","",0,"2026-06-15","2026-07-27"],[4,4355,"Support creating a Space from a template (create_repo + CLI)","gary149",0,"moon-bot-app","#4504",2,"2026-06-12","2026-07-09"],[4,4353,"Error 416 when copying empty files from a to a bucket on the GUI","julien-blanchon",2,"","",0,"2026-06-12","2026-06-22"],[4,4349,"`hf download` routinely crashes with `HF_HUB_DISABLE_XET` (no retry)","gohai",4,"","",0,"2026-06-12",""],[4,4340,"test","Dagim-gof",2,"","",0,"2026-06-11","2026-06-11"],[4,4330,"Xet CDN returns 400 for antirez deepseek-v4-gguf Pro files (>400GB)","17302140726",2,"","",0,"2026-06-09","2026-06-10"],[4,4328,"Looking for VLA Instruction-Violation Dataset: Language Instruction + Failed Action Text Pairs","froming70",3,"","",0,"2026-06-09","2026-06-09"],[4,4327,"error fetching hf token","peprahasomaning-lgtm",3,"","",0,"2026-06-09","2026-06-09"],[4,4319,"Trying to intsall and login","ozmix",3,"","",0,"2026-06-07","2026-06-11"],[4,4317,"Unable to upload my pkl model in my storage bucket.","anugrhaswi",3,"","",0,"2026-06-05","2026-06-11"],[4,4314,"Make the `hf endpoints` CLI agent- and script-friendly (discovery, custom-image/GGUF, safe defaults, logs)","gary149",4,"","",0,"2026-06-05",""],[4,4311,"[Jobs] Add a `wait` primitive — `hf jobs wait` + `HfApi.wait_for_job` (chaining + agent ergonomics)","davanstrien",0,"Wauplin","#4345",1,"2026-06-04","2026-06-18"],[4,4307,"Fine grain tokens, expose scope inference.endpoints.read to allow enumration of namespaces for auto-discovery","dlmiles",4,"","",0,"2026-06-03",""],[4,4304,"[Errno 30] Read-only file system: '/tmp_<uuid>' when downloading to local folder","quachc",0,"Wauplin","#4305",1,"2026-06-03","2026-06-03"],[4,4303,"text_to_voice","naveenkumar921995-cmd",3,"","",0,"2026-06-02","2026-06-02"],[4,4292,"When providing cache_dir to hf_hub_download, \"~\" in path is not expanded to users home directory.","monkmonk99",0,"Wauplin","#4293",1,"2026-05-29","2026-05-29"],[4,4282,"Async Buckets features","lhoestq",4,"","",0,"2026-05-27",""],[4,4280,"aihuggingface","mahircode",2,"","",0,"2026-05-27","2026-05-27"],[4,4273,"Colab","jcbneves",3,"","",0,"2026-05-26","2026-05-27"],[4,4271,"ModuleNotFoundError: No module named 'click' after installing hf CLI","AyRickk",2,"","",0,"2026-05-26","2026-05-26"],[4,4269,"[Jobs] CLI: -v accepts local paths (depends on #4268)","davanstrien",0,"Wauplin","#4346",1,"2026-05-26","2026-06-29"],[4,4268,"[Jobs] Add LocalVolume for local-path uploads (Python API)","davanstrien",0,"Wauplin","#4346",1,"2026-05-26","2026-06-29"],[4,4257,"Can't fetch the model list one page at a time","tobocop2",3,"","",0,"2026-05-25","2026-05-26"],[4,4244,"hf download with local-dir and cache-dir","kimor79",2,"","",0,"2026-05-20","2026-05-22"],[4,4238,"justy curiosity","vataneren",3,"","",0,"2026-05-19","2026-05-19"],[4,4236,"Use Trusted Publishing to upload to PyPI","nsoranzo",2,"","",0,"2026-05-19","2026-05-26"],[4,4232,"Error importing huggingface_hub.hub_mixin: type object 'tqdm' has no attribute 'set_lock'","adamjstewart",2,"","",0,"2026-05-18","2026-05-19"],[4,4227,"My logging rant - Peristent failed logins feel more like a slap in the face than AI grace.","maximlomans",2,"","",0,"2026-05-14","2026-05-18"],[4,4224,"Expose a transport-level cancel/close handle on async streaming responses","adtyavrdhn",4,"","",0,"2026-05-13",""],[4,4223,"`hf download` exits 0 while leaving `.incomplete` blobs after xet 416 CAS errors","cipherfoxie",2,"","",0,"2026-05-13","2026-06-11"],[4,4221,"Incorrect parameter count: safetensors.total = 333,312 while parameters.BF16 = 7.6B on a single-dtype model","ZonglinY",2,"","",0,"2026-05-12","2026-05-18"],[4,4219,"Silent deprecation: HF_HUB_ENABLE_HF_TRANSFER=1 no-op for uploads, no warning unless HF_XET_HIGH_PERFORMANCE a","Adithya191101",1,"Adithya191101","#4220",0,"2026-05-10","2026-05-15"],[4,4218,"`license` is not accepting SPDX license identifier","ZhiyuanChen",2,"","",0,"2026-05-10","2026-05-18"],[4,4208,"`snapshot_download` progress bar total inflates on every `http_get` retry","popfido",1,"popfido","#4209",0,"2026-05-08","2026-05-21"],[4,4207,"How to stop hf models ls from truncating the results in the table?","eabase",2,"","",0,"2026-05-07","2026-08-09"],[4,4196,"No resuming of downloads","TurtleTango42",4,"","",0,"2026-05-06",""],[4,4195,"Deprecation of hf_xet.download_files()","mscheltienne",2,"","",0,"2026-05-06","2026-06-09"],[4,4178,"Streaming dataset hangs consistently","michaelpginn",2,"","",0,"2026-05-01","2026-05-18"],[4,4160,"`hf upload-large-folder` doesn't accept `path_in_repo` argument like `hf upload`","lusxvr",3,"","",0,"2026-04-28","2026-04-28"],[4,4153,"[Buckets] Range-aware file write","Wauplin",4,"","",0,"2026-04-27",""],[4,4152,"access request","Fikraddis",2,"","",0,"2026-04-27","2026-04-27"],[4,4150,"Add parameter to expand/truncate ID column in hf cache ls","akhildevelops",2,"","",0,"2026-04-27","2026-08-17"],[4,4145,"OOM Killed during bucket upload","Hugi-R",2,"","",0,"2026-04-25","2026-06-04"],[4,4144,"Token should not be stored in ~/.cache","gdevenyi",2,"","",0,"2026-04-25","2026-04-27"],[4,4134,"HF Jobs: single-GPU h200 can fail CUDA init with Error 802 before training starts","imranarshad",2,"","",0,"2026-04-23","2026-04-27"],[4,4133,"When using snapshot_download in offline mode, the commit hash read from refs/<revision> is not stripped of tra","Egor-Iudin",3,"","",0,"2026-04-23","2026-05-18"],[4,4129,"[BUG+Workaround]: HTTP download progress callback fires only once per ~10 MB, making progress bars look frozen","tobocop2",3,"","",0,"2026-04-20","2026-05-18"],[4,4128,"CUDA Error 802 on all H200 multi-GPU HF Jobs with vLLM, across CUDA 12 and CUDA 13 images","elenaajayi",2,"","",0,"2026-04-20","2026-04-27"],[4,4127,"[Security] Warn if custom pipelines contain risky code","iwr-redmond",3,"","",0,"2026-04-20","2026-04-20"],[4,4124,"hf cache rm -> ability to remove a specific quantization variant!","Gerporgl",0,"hanouticelina","#4847",1,"2026-04-18","2026-09-14"],[4,4121,"Recent change to dataclass validation breaks loading of OLMo-3 models","loris3",2,"","",0,"2026-04-17","2026-04-17"],[4,4106,"[spaces] update outdated psutil requirements","nrolin",2,"","",0,"2026-04-15","2026-04-15"],[4,4105,"Inconsistency in lower-case matching for cached folders","tboerstad",2,"","",0,"2026-04-15","2026-04-15"],[4,4099,"hf download with URL","thewh1teagle",4,"","",0,"2026-04-14",""],[4,4090,"dlqwen","avitzpriv",3,"","",0,"2026-04-13","2026-04-13"],[4,4089,"google/gemma-2-27b is serving files from a different repository (bucket routing error)","Math-Tutor-In-Jeans",3,"","",0,"2026-04-12","2026-04-13"],[4,4088,"downloading","starsgo",3,"","",0,"2026-04-12","2026-04-14"],[4,4086,"Path Traversal in huggingface_hub Local Download/Upload Paths (Linux/macOS)","sam8k",3,"","",0,"2026-04-12","2026-04-13"],[4,4085,"Large downloads via HF hub stuck (colab)","AJain9199",2,"","",0,"2026-04-11","2026-04-14"],[4,4080,"Urgent: 429 on homepage and public downloads from Brazilian IPs","luistiagos",2,"","",0,"2026-04-09","2026-08-31"],[4,4069,"Collection not visible on org collections page despite being public","MikeDoes",4,"","",0,"2026-04-08",""],[4,4066,"[BUG]: hf_hub_download crashes with ValueError when stderr lacks a real file descriptor","tobocop2",1,"tobocop2","#4065",0,"2026-04-07","2026-04-27"],[4,4060,"Unreliable resume support and lack of partial corruption recovery in large file downloads","kripper",2,"","",0,"2026-04-07","2026-07-09"],[4,4058,"[BUG]: xet downloads barely report progress (bar appears stuck on large files)","tobocop2",4,"","",0,"2026-04-06",""],[4,4053,"Fine tuning model","Sanidhyasharma45",3,"","",0,"2026-04-05","2026-04-06"],[4,4052,"Doc mistake","ljw20180420",0,"hanouticelina","#4055",1,"2026-04-05","2026-04-07"],[4,4050,"tqdm_class progress tracking silently broken in non-TTY environments (disable=None auto-detection)","mtschoen",0,"hanouticelina","#4056",1,"2026-04-04","2026-04-07"],[4,4042,"a","ERICKGLEZ99",3,"","",0,"2026-04-03","2026-04-03"],[4,4040,"Prevent duplication of models in cache on case senstive files systems","cphlipot",2,"","",0,"2026-04-02","2026-04-06"],[4,4039,"HF_TOKEN","nagnathtanikanti-her",3,"","",0,"2026-04-02","2026-04-02"],[4,4037,"Add OCI Object Storage support for snapshot_download and hf_hub_download","fede-kamel",3,"","",0,"2026-04-02","2026-04-02"],[4,4027,"Add CACHEDIR.TAG file to cache directory","nlgranger",0,"Wauplin","#4030",1,"2026-04-02","2026-04-02"],[4,4023,"Vid","tahakhataha2",3,"","",0,"2026-04-01","2026-04-02"],[4,4022,"Ability to disable symlink usage for NAS caches","cjkindel",0,"Wauplin","#4032",1,"2026-04-01","2026-04-02"],[4,4021,"Verifying quantized model provenance: linking GGUFs to their source models","razashariff",2,"","",0,"2026-04-01","2026-04-29"],[4,4017,"[i18n-<languageCode>] Translating docs to <Punjabi>","Raheel-glicth",4,"","",0,"2026-04-01",""],[4,4016,"add new state language","Raheel-glicth",2,"","",0,"2026-03-31","2026-04-01"],[4,4004,"Output-level provenance metadata for inference results","HMAKT99",4,"","",0,"2026-03-28",""],[4,3998,"snapshot download failure","zhangwei217245",4,"","",0,"2026-03-27",""],[4,3996,"HfFileSystem is incompatible with fsspec's config","martindurant",0,"joaquinhuigomez","#4062",0,"2026-03-27","2026-04-24"],[4,3995,"[bucket] sync_bucket (bucket -> local) doesn't update local files correctly","dtcxzyw",0,"Wauplin","#4019",1,"2026-03-27","2026-04-01"],[4,3985,"Add \"distill\" as a supported base_model_relation in model card metadata","Pringled",2,"","",0,"2026-03-27","2026-03-27"],[4,3984,"Model","esdrasnooby-blip",3,"","",0,"2026-03-27","2026-03-27"],[4,3981,"`PyTorchModelHubMixin` models with safetensors weights are loaded in training mode","mxmpl",0,"joaquinhuigomez","#3997",0,"2026-03-26","2026-03-30"],[4,3979,"[CLI] Dual mode in CLI output (human vs agents)","Wauplin",2,"","",0,"2026-03-26","2026-06-01"],[4,3974,"### 🐛 Bug Report: SIGABRT crash on Python 3.14.3 shutdown with hf_xet + PyTorch  .","trendyxlabspro",4,"","",0,"2026-03-25",""],[4,3972,"Jobs API: intermittent 500 Internal Server Error on POST /api/jobs","yharby",4,"","",0,"2026-03-25",""],[4,3971,"Harmonize hf:// parsing logic","Wauplin",2,"","",0,"2026-03-25","2026-08-31"],[4,3963,"TC 3","Imperat0",3,"","",0,"2026-03-23","2026-03-23"],[4,3962,"My HF","rengaraj84",3,"","",0,"2026-03-22","2026-03-23"],[4,3960,"Downloading not working with hf_xet","acuna-public",4,"","",0,"2026-03-20",""],[4,3959,"Dataser","sreymealea21-art",2,"","",0,"2026-03-20","2026-03-20"],[4,3958,"I cannot sign up for pro: Error: Failed to prepare payment","Blubberi",2,"","",0,"2026-03-20","2026-03-25"],[4,3947,"`hf` CLI Friction Points & Improvements","hanouticelina",3,"","",0,"2026-03-18","2026-04-27"],[4,3937,"An incorrect parameter was passed to download_files","OnlyblackTea",2,"","",0,"2026-03-17","2026-03-17"],[4,3934,"232","rlfqjafuf",3,"","",0,"2026-03-16","2026-03-16"],[4,3917,"`hf auth login` silently skips when already logged in (inverted `skip_if_logged_in` logic)","hysts",0,"hanouticelina","#3920",1,"2026-03-12","2026-03-12"],[4,3915,"Support gRPC interface in `InferenceClient`/`AsyncInferenceClient` `feature_extraction`","maxdswain",4,"","",0,"2026-03-11",""],[4,3913,"snapshot_download raises a generic \"Please check your internet connection\" error message.","pavankumarch470",2,"","",0,"2026-03-11","2026-03-12"],[4,3896,"Security: Path Traversal Check Windows-Only + Token Files World-Readable (CWE-22, CWE-276)","spartan8806",3,"","",0,"2026-03-08","2026-03-09"],[4,3871,"Uploads with `New Data Upload` over 2GB get stuck in Windows 11 with Python 3.12","Santodan",4,"","",0,"2026-03-02",""],[4,3869,"Automated release notes","Wauplin",2,"","",0,"2026-03-02","2026-08-31"],[4,3868,"HTTP download fails for files >50GB — why is there a limit?","xiaoyang-hhh",4,"","",0,"2026-03-02",""],[4,3867,"Facing HFValidation Error when trying to load bge-small-en-v1.5","prathamgec21-hue",2,"","",0,"2026-03-02","2026-03-02"],[4,3865,"BUG: Wrong Contributor in commit history","Vinay-Umrethe",2,"","",0,"2026-02-28","2026-03-02"],[4,3864,"feat: Add `hf webhooks` CLI commands","gary149",0,"omkar-334","#3866",0,"2026-02-28","2026-03-05"],[4,3863,"SafeTensors Metadata RCE PoC in SafeTensors","Apollyon13X",4,"","",0,"2026-02-27",""],[4,3862,"allow languages other than nature languages","ZhiyuanChen",4,"","",0,"2026-02-27",""],[4,3859,"Download throughput severely degraded by progress callback GIL contention","Hugoch",0,"Wauplin","ed43eb6",1,"2026-02-26","2026-03-02"],[4,3856,"Better support for collection retrieval","gsaltintas",2,"","",0,"2026-02-26","2026-08-31"],[4,3842,"Collection items display incorrect","superAIyah",2,"","",0,"2026-02-25","2026-03-02"],[4,3838,"Cache directory preserves original casing of repo_id → duplicate caches on case-sensitive filesystems (Linux)","julio75012",3,"","",0,"2026-02-25","2026-02-25"],[4,3832,"Gender Awareness for Factory Workers","nunupplf-cyber",2,"","",0,"2026-02-23","2026-02-23"],[4,3829,"Cannot commit Parquet file to recreated dataset repo (LFS pointer pointed to a file that does not exist)","jorritsandbrink",0,"Wauplin","#3847",1,"2026-02-22","2026-02-25"],[4,3828,"HF Jobs - Job ID env var inside container","Praful932",2,"","",0,"2026-02-22","2026-02-24"],[4,3827,"...","Sgfcgj358",3,"","",0,"2026-02-21","2026-02-23"],[4,3825,"HuggingFace Hub Feature Request: duplicate_repo() for Dataset/Model repos","hyeonsangjeon",0,"Wauplin","#3880",1,"2026-02-20","2026-03-05"],[4,3813,"Metadata rendering bug in dataset card despite following documentation","Raibows",2,"","",0,"2026-02-17","2026-02-17"],[4,3812,"Yt","GabrielSB0119",3,"","",0,"2026-02-16","2026-02-17"],[4,3809,"Syntax error in Transformer section 3 (Transformers, what can they do?)  notebook","khyatimirani",3,"","",0,"2026-02-15","2026-02-15"],[4,3808,"[i18n-<languageCode>] Translating docs to TELUGU","HARSVARDH",2,"","",0,"2026-02-13","2026-02-17"],[4,3806,"Accessing persistent /data folder in a Space with FastAPI","Rizwanali324",2,"","",0,"2026-02-13","2026-08-31"],[4,3801,"jnjb","hudebiyak",3,"","",0,"2026-02-12","2026-02-12"],[4,3796,"Buckets API","Wauplin",2,"","",0,"2026-02-10","2026-02-19"],[4,3795,"Downloads stall on Azure mounted volumes","ferreroal",4,"","",0,"2026-02-10",""],[4,3792,"RFC: HF CLI Redesign — Consistency, Scalability, and Conventions","Wauplin",2,"","",0,"2026-02-10","2026-08-31"],[4,3791,"Dummy Agent Library Template Does Not Run","ccheng122",2,"","",0,"2026-02-09","2026-08-31"],[4,3786,"504 error while duplicating dataset","superAIyah",2,"","",0,"2026-02-07","2026-02-09"],[4,3777,"Issue with downloading a model using hf cli.","alexc957",0,"Wauplin","#3785",1,"2026-02-05","2026-02-13"],[4,3775,"Downloads stall/truncate for xet-enabled repos from Taiwan","mmto-io",4,"","",0,"2026-02-05",""],[4,3774,"hf cli failing to retrieve token from colab secrets","not-lain",3,"","",0,"2026-02-05","2026-02-06"],[4,3770,"L’Esprit de la Richesse","uzzamzafar053053-ops",3,"","",0,"2026-02-04","2026-02-04"],[4,3766,": Handwriting-to-Tax-Form Auto-Fill System","Tayyebwazir",3,"","",0,"2026-02-03","2026-02-03"],[4,3765,"The transfer of the large ZIP file is stuck.","dfm021101",3,"","",0,"2026-02-03","2026-08-31"],[4,3762,"OAuth login options missing on signup page linked from /settings/tokens","lborgman",3,"","",0,"2026-02-02","2026-07-01"],[4,3752,"Add retry/backoff when fetching Xet connection info during hf_hub_download (502 Bad Gateway)","boyealkim",0,"aabhathanki","#3768",0,"2026-01-29","2026-02-04"],[4,3750,"Add per-request timeout to huggingface_hub functions `list_repo_files` and `list_models`","FlynnOwen",2,"","",0,"2026-01-28","2026-01-30"],[4,3747,"Uploading large files on huggingface fails.","EugeoSynthesisThirtyTwo",4,"","",0,"2026-01-28",""],[4,3731,"ChatCompletionInputJSONSchema should have \"$defs\" field","maxsei",2,"","",0,"2026-01-25","2026-02-03"],[4,3730,"AJKMRMR","ajayuu349-wq",3,"","",0,"2026-01-25","2026-01-26"],[4,3728,"404 Job Not Found when running a job through a different namespace other than the user","Praful932",1,"Praful932","#3736",0,"2026-01-24","2026-01-26"],[4,3727,"Cannot subscribe to Pro Account - prepare payment fails","sgpayment",2,"","",0,"2026-01-24","2026-01-26"],[4,3726,"Pre-upload progress is not persisted across restart (Xet), causing re-uploads and excess traffic","uebian",2,"","",0,"2026-01-24","2026-08-31"],[4,3725,"hf_xet.abi3.so installed is dynamically linked to glibc and thus doesn't work in non glibc based systems","berney",2,"","",0,"2026-01-24","2026-01-24"],[4,3717,"Pro account – Router Inference 404 for all models","benybh",2,"","",0,"2026-01-23","2026-01-26"],[4,3716,"[Feature Request] DDUF Flashpack support","iwr-redmond",2,"","",0,"2026-01-22","2026-08-31"],[4,3709,"Deleting files from CLI (`hf repo-files delete ...`) with wildcard *recursively* deletes files","kyo-takano",2,"","",0,"2026-01-22","2026-01-22"],[4,3706,"[CLI] hf cache verify --fail-on-missing-files fails while corresponding folders are here","Mitix-EPI",1,"Mitix-EPI","#3707",0,"2026-01-21","2026-01-22"],[4,3701,"upload_large_folder: Stale _remote_oid on retry may cause empty commits after timeout","davanstrien",2,"","",0,"2026-01-20","2026-08-31"],[4,3690,"Proposal: Support for decision contracts and governance metadata in agent deployments","lexseasson",2,"","",0,"2026-01-18","2026-01-19"],[4,3688,"URGENT: Production outage – Hugging Face Router returns 500 errors for vision models","SmartEyessup",2,"","",0,"2026-01-17","2026-01-20"],[4,3687,"Vdm","oghene4l",2,"","",0,"2026-01-17","2026-01-17"],[4,3684,"Severe performance regression when streaming data from Hugging Face datasets.","leq6c",2,"","",0,"2026-01-15","2026-02-02"],[4,3683,"Email verification Stuck- Invalid Token","LovelyRajput",2,"","",0,"2026-01-15","2026-01-15"],[4,3681,"broken model download via AutoModel using private repositories","cr-klarna",3,"","",0,"2026-01-14","2026-07-09"],[4,3680,"HF_TOKEN","Bongani94",3,"","",0,"2026-01-14","2026-01-20"],[4,3677,"1","jsvu666-code",3,"","",0,"2026-01-12","2026-01-12"],[4,3676,"me","Beloved-Esther",3,"","",0,"2026-01-12","2026-01-12"],[4,3667,"hhg245","HWR-GGG",3,"","",0,"2026-01-07","2026-01-07"],[4,3660,"Bug: Search Algorithm on HuggingFace Hub does not seem correct","MikeDoes",4,"","",0,"2026-01-05",""],[4,3659,"ocr","mohamediirfan",3,"","",0,"2026-01-03","2026-01-05"],[4,3658,"connect with vsc","Makins18",3,"","",0,"2026-01-02","2026-01-20"],[4,3657,"[Feature Request] Support explicitly mapping Model IDs to local paths","Gianthard-cyh",4,"","",0,"2026-01-02",""],[4,3652,"mygen ai","akhileshsisupalan29-dotcom",2,"","",0,"2025-12-28","2025-12-28"],[4,3651,"Allow custom endpoints for jobs APIs","rezass",0,"PredictiveManish","#3653",0,"2025-12-27","2026-01-05"],[4,3649,"Improve exception handling in hf_raise_for_status warning header parsing","mishrapravin114",3,"","",0,"2025-12-26","2026-01-05"],[4,3648,"  On slow internet and breaks - hf download model from 0 again again ","Uglevod",2,"","",0,"2025-12-26","2026-01-05"],[4,3643,"snapshot_download reports success despite blob checksum mismatch (large files, XET storage)","volodymyrjf",2,"","",0,"2025-12-23","2026-01-05"],[4,3640,"Test","Kenzo-0",2,"","",0,"2025-12-21","2025-12-22"],[4,3636,"Add documentation to avoid/fix HF_HUB_DISABLE_SYMLINKS_WARNING  for windows users","bmerkle",2,"","",0,"2025-12-20","2026-09-09"],[4,3632,"upload to organization dataset PR fails with 404 error","drnhhl",0,"hanouticelina","#3634",1,"2025-12-18","2025-12-18"],[4,3628,"[Feature] OAuth 2.0 Device Authorization Grant (RFC 8628) for CLI/headless authentication","clouatre",2,"","",0,"2025-12-15","2026-08-31"],[4,3625,"storyboard","GnanaSounthar",3,"","",0,"2025-12-15","2025-12-15"],[4,3621,"huggingFaceEndpoint invoke error","xiaxiazhu",4,"","",0,"2025-12-14",""],[4,3620,"chat","srijansinghraghuvanshi-png",2,"","",0,"2025-12-13","2025-12-13"],[4,3619,"get_safetensors_metadata for models in local folder","vrdn-23",1,"vrdn-23","#3623",0,"2025-12-12","2026-01-08"],[4,3615,"[Massive Bug] Cannot sign up for pro. Tried multiple angles. Killing your business model","SlipstreamAI",2,"","",0,"2025-12-11","2025-12-12"],[4,3613,"[Feature request] Implement `hf repo list` command to list models, spaces, and datasets","hanouticelina",2,"","",0,"2025-12-11","2026-01-05"],[4,3612,"Intermittent `FileNotFoundError` when repeatedly loading Hub dataset (likely rate limiting)","viktor-shcherb",2,"","",0,"2025-12-11","2025-12-11"],[4,3609,"hf download : crash when a metadata file is corrupted and downloading to local-dir","goulou",0,"Wauplin","#3610",1,"2025-12-10","2025-12-10"],[4,3604,"[Feature request] Implement `hf repo list` command to list models, spaces, and datasets from the CLI","hanouticelina",3,"","",0,"2025-12-09","2025-12-11"],[4,3603,"huggingface cli is not honoring http proxy environment variables on version >=1.0.0","smothiki",4,"","",0,"2025-12-09",""],[4,3602,"[i18n-TE] Translating docs to Telugu","pranayy9346",4,"","",0,"2025-12-08",""],[4,3600,"new","420231013-del",3,"","",0,"2025-12-07","2025-12-08"],[4,3596,"Feature Request: Add method to get access request counts for analytics/monitoring","balanprasanth",4,"","",0,"2025-12-05",""],[4,3594,"get_hf_file_metadata ignores HF_HUB_DOWNLOAD_TIMEOUT","krrome",1,"krrome","#3595",0,"2025-12-04","2025-12-04"],[4,3591,"`list_accepted_access_requests()` has undocumented 1,000 record limit","balanprasanth",2,"","",0,"2025-12-04","2025-12-04"],[4,3582,"g","technology2882009",3,"","",0,"2025-11-29","2025-12-01"],[4,3581,"Empty","bejbaryskakimzan10-eng",2,"","",0,"2025-11-28","2025-11-28"],[4,3580,"hf cli: Extremely slow download of shards toward the end (or hanging)","cadieuxpaquinr",2,"","",0,"2025-11-28","2025-12-04"],[4,3576,"Deepcopy of HfFileSystem fails due to non-picklable HfHubHTTPError response arg","owenowenisme",1,"owenowenisme","#3579",0,"2025-11-27","2025-11-28"],[4,3571,"`tiny-agents` failed due to missing `rich` module if it installed by `uv`","ngdangtu-vn",2,"","",0,"2025-11-26","2025-11-26"],[4,3567,"Add a way to fetch blogs and community posts","Samoed",2,"","",0,"2025-11-24","2025-12-01"],[4,3566,"Add daily papers method `/api/daily_papers`","Samoed",2,"","",0,"2025-11-24","2025-12-02"],[4,3563,"Man","murselahmetbey",3,"","",0,"2025-11-21","2025-11-21"],[4,3560,"Ability to retrieve comments for a paper returned from /api/papers","tleyden",2,"","",0,"2025-11-20","2025-12-01"],[4,3559,"Adding a limit param to the /api/papers/search api","tleyden",0,"Wauplin","#3697",1,"2025-11-20","2026-01-20"],[4,3557,"TypeError: list indices must be integers or slices, not str  in zero_shot_classification","punit1108",0,"hanouticelina","#3561",1,"2025-11-19","2025-11-20"],[4,3555,"Account recovery request: lost access to corporate email","HaoranYokina",3,"","",0,"2025-11-19","2025-11-19"],[4,3554,"Compact transformer","AakashKumarMissrani",3,"","",0,"2025-11-19","2025-11-19"],[4,3553,"Handle rate limit headers in case of HTTP 429","Wauplin",2,"","",0,"2025-11-18","2025-12-03"],[4,3552,"Missing files when downloading subdirectory via huggingface-cli download","Raibows",2,"","",0,"2025-11-17","2025-11-28"],[4,3550,"The payment did not go through","F018K",2,"","",0,"2025-11-14","2025-11-14"],[4,3546,"MedChat","Anandhakrishnanpv",3,"","",0,"2025-11-12","2025-11-13"],[4,3545,"[Collections] Add collections to collections by slug id, not only by object id","Wauplin",0,"hanouticelina","#3551",1,"2025-11-12","2025-11-17"],[4,3543,"zanjebil","meysamcoder313-lang",2,"","",0,"2025-11-11","2025-11-11"],[4,3540,"Assistance with Hugging Face Competition Setup","Tswings",2,"","",0,"2025-11-10","2026-06-02"],[4,3533,"httpx.ConnectError: EOF occurred in violation of protocol (_ssl.c:997)","asyhyxcqdx",3,"","",0,"2025-11-07","2026-10-27"],[4,3530,"Hf-xet downloads ignore custom client factory","guibruand",2,"","",0,"2025-11-06","2025-11-06"],[4,3524,"-","kakamond",3,"","",0,"2025-11-04","2025-11-05"],[4,3515,"[Bug] Inference API for intfloat/multilingual-e5-large-instruct unreachable (health check failed for days)","Chris-K-dev",2,"","",0,"2025-11-01","2026-08-31"],[4,3501,"https://hf.co/cli/install.sh Error","unclemusclez",2,"","",0,"2025-10-29","2025-11-03"],[4,3494,"repo_type_and_id_from_hf_id() misinterprets custom endpoint paths as repo_type (e.g. \"/hf\")","pulltheflower",2,"","",0,"2025-10-29","2026-07-28"],[4,3492,"Amor a distancia","JuanChavez222",3,"","",0,"2025-10-28","2025-10-29"],[4,3487,"aiohttp removed?","lahwaacz",0,"Wauplin","#3488",1,"2025-10-28","2025-10-28"],[4,3484,"Why \"huggingface-hub 1.0.0 does not provide the extra 'cli'\" is appearing when installing cli?","ccocks",2,"","",0,"2025-10-27","2025-10-27"],[4,3480,"Docker Space fails to call Ollama Cloud API - 404 \"model not found\"","indidginus",3,"","",0,"2025-10-26","2025-11-19"],[4,3479,"Modern Access Tokens only work with API v2, causing 401 errors in huggingface_hub","altjack",2,"","",0,"2025-10-25","2025-10-27"],[4,3478,"image","lavanyamekala600-coder",3,"","",0,"2025-10-25","2025-10-27"],[4,3477,"policy_model","venuBandi1411",3,"","",0,"2025-10-25","2025-10-27"],[4,3472,"Can't download models using Abogen","anarion80",2,"","",0,"2025-10-24","2025-10-27"],[4,3468,"Rrt","mtomax35-crypto",3,"","",0,"2025-10-24","2025-10-24"],[4,3466,"Is this still fastest way uploading to Hugging Face?","FurkanGozukara",2,"","",0,"2025-10-23","2025-10-23"],[4,3463,"hf download doesn't appear to be using xet on an nfs volume","doctorpangloss",2,"","",0,"2025-10-22","2025-10-23"],[4,3458,"HUB still down after AWS Outage","bdytx5",2,"","",0,"2025-10-21","2025-10-21"],[4,3457,"The Snapshot download consistently fails when downloading large datasets (exceeding 5TB)","zjl775210697",2,"","",0,"2025-10-21","2026-08-31"],[4,3456,"​Security Risk: Command Injection Vulnerability in RepositoryClass​","Goalina",2,"","",0,"2025-10-21","2025-10-21"],[4,3455,"Cannot Download the Model Checkpoints","YiboZhao624",2,"","",0,"2025-10-20","2025-10-20"],[4,3454,"I need some help and guidance for getting top article writers post writers and such","FurkanGozukara",3,"","",0,"2025-10-19","2025-10-20"],[4,3453,"HF Token","dimitribb177",3,"","",0,"2025-10-18","2025-10-20"],[4,3452,"Docker Space not triggering build despite valid Dockerfile and correct SDK","SamPan-Git",4,"","",0,"2025-10-18",""],[4,3445,"Bug: Inconsistent handling of empty token (token=\"\") across Hub functions","ai-jz",2,"","",0,"2025-10-15","2025-10-16"],[4,3440,"snapshot_download error after ~20%","satyrmipt",4,"","",0,"2025-10-13",""],[4,3438,"Face_iris_recognition","101SkShabbirHussainAryanKhaN",3,"","",0,"2025-10-11","2025-10-14"],[4,3434,"[BUG] Python 3.14t (no gil)","johnnynunez",2,"","",0,"2025-10-10","2025-10-10"],[4,3432,"[RFC] Revamp `hf cache`","Wauplin",2,"","",0,"2025-10-10","2025-11-05"],[4,3430,"Unable to list user while add user during gated user access","AL-Cybision",4,"","",0,"2025-10-10",""],[4,3429,"hf-xet Integration on Windows Causes Ctrl+C Hang and Shows Confusing Installation Warnings","hksbindra",2,"","",0,"2025-10-09","2025-12-02"],[4,3426,"Hugging face Model","Abishekvarshan",3,"","",0,"2025-10-09","2025-10-09"],[4,3422,"What","iamasrakib",3,"","",0,"2025-10-08","2025-10-08"],[4,3417,"[feature request]: ability to track references/metadata from hash | reverse reference","maifeeulasad",3,"","",0,"2025-10-05","2025-10-06"],[4,3416,"Inference chat_completion uses deprecated conversational task","dpsommer",2,"","",0,"2025-10-04","2025-10-06"],[4,3413,"RSS support for repo/model commits","i2",4,"","",0,"2025-10-03",""],[4,3406,"`snapshot_download` uses stale commit when downloading repo with force-pushed history","Butanium",2,"","",0,"2025-10-01","2025-10-02"],[4,3398,"Target_size issue in image to image inference","iam-tsr",0,"hanouticelina","#3399",1,"2025-09-28","2025-09-29"],[4,3396,"Tiny-agents uses 0 tools without explicitly setting allowed_tools","Mithil467",1,"Mithil467","#3397",0,"2025-09-28","2025-09-29"],[4,3394,"HF_HUB_OFFLINE=1 not being respected","redshiva",2,"","",0,"2025-09-25","2025-09-26"],[4,3391,"Peerapat","pexrapat",3,"","",0,"2025-09-25","2025-09-25"],[4,3390,"hf_hub_download 'duplicating' files","johnbradley",2,"","",0,"2025-09-24","2025-09-25"],[4,3388,"my colab","shayatiamo",2,"","",0,"2025-09-24","2025-09-24"],[4,3386,"Timeout when Creating a Scheduled job is ignored","tomaarsen",2,"","",0,"2025-09-24","2025-10-01"],[4,3382,"Add more hf commands / address existing CLI tickets","hanouticelina",2,"","",0,"2025-09-23","2025-10-22"],[4,3380,"List all downloaded models","maifeeulasad",4,"","",0,"2025-09-22",""],[4,3374,"Misleading docstring for the sentence_similarity endpoint","tolgaakar",2,"","",0,"2025-09-19","2025-09-23"],[4,3373,"Add package installer (install in a separate env, install autocomple commands, etc.) Think curl https://hf.co/","hanouticelina",2,"","",0,"2025-09-18","2025-10-01"],[4,3371,"interrupting snapshot_download, then calling it again, returns an invalid repo","doctorpangloss",2,"","",0,"2025-09-17","2025-09-18"],[4,3370,"Add autocomplete support","hanouticelina",2,"","",0,"2025-09-17","2025-09-18"],[4,3368,"API Auth Issue: Error 401 for all keys from Termux on Android","gameahmed184",2,"","",0,"2025-09-16","2025-09-19"],[4,3366,"Encountering HTTP 403 storage limit errors when uploading files","FelixYuan-YF",2,"","",0,"2025-09-16","2025-09-17"],[4,3364,"`set rich_markup_mode=None` to disable rich in all cases https://typer.tiangolo.com/tutorial/commands/help/#ri","hanouticelina",2,"","",0,"2025-09-15","2025-09-18"],[4,3363,"Refactor CLI implementation using Typer","hanouticelina",2,"","",0,"2025-09-15","2025-09-18"],[4,3362,"Zero GPU daily quota","EvgenyZaretskiy",2,"","",0,"2025-09-14","2025-09-15"],[4,3361,"Can't download models on RunPod","peepeepeepoopoopoo",2,"","",0,"2025-09-13","2025-12-21"],[4,3358,"Write comprehensive migration guide for v0.x to v1.0","Wauplin",1,"Wauplin","52c718a",1,"2025-09-12","2025-10-23"],[4,3351,"Improve CLI","hanouticelina",2,"","",0,"2025-09-12","2025-10-22"],[4,3342,"Occasionally HTTP 429 Error When Downloading Many Smaller Files","arth-shukla",2,"","",0,"2025-09-09","2026-08-31"],[4,3341,"fnc","toitoitoi335",3,"","",0,"2025-09-09","2025-09-09"],[4,3340,"[RFC] [1.0] 🚀 huggingface_hub 1.0 release checklist","Wauplin",2,"","",0,"2025-09-09","2025-12-16"],[4,3339,"What is the best replacement of HfFileSystem.glob with HfApi","narugo1992",2,"","",0,"2025-09-09","2025-09-15"],[4,3337,"health ai","kiru65",3,"","",0,"2025-09-06","2025-10-01"],[4,3336,"IA 1.0","herasmarcos04-art",3,"","",0,"2025-09-06","2025-09-08"],[4,3335,"Ско","chernoknizhniykirilo-arch",3,"","",0,"2025-09-05","2025-09-05"],[4,3334,".","chernoknizhniykirilo-arch",3,"","",0,"2025-09-05","2025-09-05"],[4,3332,"Update error message to improve shell compatibility","aopstudio",1,"aopstudio","#3333",0,"2025-09-05","2025-09-05"],[4,3331,"healthcare","Manikandanprabu271",3,"","",0,"2025-09-04","2025-09-04"],[4,3329,"Truncated API response using `recursive=true`","Wenxin-Jiang",2,"","",0,"2025-09-03","2025-09-04"],[4,3325,"Infinite retry loop in upload_large_folder when encountering HTTP 403 storage limit errors","nyuuzyou",2,"","",0,"2025-09-01","2026-08-31"],[4,3324,"cannot download duplicated repository contents","bghira",2,"","",0,"2025-09-01","2025-09-08"],[4,3323,"DNS resolver","LokiWager",2,"","",0,"2025-09-01","2025-09-17"],[4,3319,"Enable filtering model search using `HfApi.list_models` by supported applications, e.g., Ollama","anirbanbasu",2,"","",0,"2025-08-29","2025-09-01"],[4,3316,"Dataset missing from Audio Trending index entirely (not a ranking issue)","Harmonic-Frontier-Audio",2,"","",0,"2025-08-27","2025-08-29"],[4,3314,"`huggingface_hub.snapshot_download` with an ignore_pattern of `*fp16*` does not work.","O-J1",2,"","",0,"2025-08-26","2025-08-27"],[4,3313,"`HfApi().list_models()`: Identical filtering behavior for `'tasks'` and `'tags'`","d-kleine",2,"","",0,"2025-08-23","2025-08-27"],[4,3312,"Production system breaks due to HTTP 429 errors","eyal-converge",4,"","",0,"2025-08-23",""],[4,3311,"Improve language tags on the hub","wasertech",2,"","",0,"2025-08-20","2025-08-21"],[4,3310,"Two tags for Turkish (tur and tr)","A-Yucel",2,"","",0,"2025-08-19","2025-08-26"],[4,3309,"Search engine","Premkumar3406",3,"","",0,"2025-08-18","2025-08-27"],[4,3308,"hf download: list available models for repo owner","abitrolly",2,"","",0,"2025-08-15","2025-08-28"],[4,3307,"OSError: Consistency check failed: file should be of size 0 but has size 2166 (hest.py). We are sorry for the ","Flliin",2,"","",0,"2025-08-15","2025-10-01"],[4,3304,"AI","sinalojayice-cyber",3,"","",0,"2025-08-13","2025-08-13"],[4,3302,"hf auth login fails: requests library converts Bearer auth to Basic auth causing 401 errors","supideo",2,"","",0,"2025-08-12","2025-08-13"],[4,3301,"list_models not working as expected","bchen1116",2,"","",0,"2025-08-12","2025-08-27"],[4,3300,"OOM error occurs when downloading model using recent versions of Huggingface Hub Python library (worked fine i","Jooho",4,"","",0,"2025-08-11",""],[4,3299,"403 forbidden error in Github Codespace","junwha",2,"","",0,"2025-08-11","2025-08-11"],[4,3298,"command for verifying local files","flowpoint",2,"","",0,"2025-08-10","2025-12-16"],[4,3296,"'hf' is not recognized as an internal or external command","qianyun210603",0,"setu4993","#158",1,"2025-08-09","2025-09-01"],[4,3295,"Error while fetching `HF_TOKEN` secret value from your vault","amanamali",2,"","",0,"2025-08-09","2025-08-11"],[4,3293,"Error importing huggingface_hub.hub_mixin: module 'torch' has no attribute 'uint64'","murlock1000",2,"","",0,"2025-08-08","2025-08-11"],[4,3292,"upload_folder doesnt work on a branch","mertunsall",2,"","",0,"2025-08-07","2025-08-08"],[4,3287,"How about --strip-path","pinballelectronica",2,"","",0,"2025-08-05","2025-08-07"],[4,3285,"utils.tqdm is not thread-safe","dolfim-ibm",2,"","",0,"2025-08-05","2026-11-04"],[4,3284,"HF_TOKEN timed out","delinlabs",2,"","",0,"2025-08-04","2025-08-07"],[4,3282,"Potentially set a default User-Agent in `hf_hub_download()`?","06kellyjac",2,"","",0,"2025-08-01","2025-08-06"],[4,3278,"Add pre-flight / validate_repository_limits checks to upload_large_folder to prevent hitting repository limits","davanstrien",1,"davanstrien","#3280",1,"2025-07-31","2025-08-07"],[4,3276,"Add Flash Attention 2 by default to Spaces","Josephrp",4,"","",0,"2025-07-30",""],[4,3273,"Problems downloading models","bowie2211",3,"","",0,"2025-07-29","2025-10-01"],[4,3269,"HfFileSystemStreamFile.read() does not handle Content-Encoding: gzip","rasmusfaber",1,"rasmusfaber","#3271",0,"2025-07-28","2025-07-29"],[4,3266,"HF_HUB_DISABLE_XET not disabling XET-based downloads","devtobi",2,"","",0,"2025-07-27","2025-10-16"],[4,3263,"filesystem stat/info doesn't give last_commit on the latest version of huggingface-hub (0.34.0)","Koncopd",2,"","",0,"2025-07-25","2025-07-25"],[4,3260,"NLP","ny740",3,"","",0,"2025-07-25","2025-07-25"],[4,3259,"NLP","ny740",3,"","",0,"2025-07-25","2025-07-25"],[4,3258,"25/07/25","ny740",3,"","",0,"2025-07-25","2025-07-25"],[4,3256,"hf: command not found","hadipash",2,"","",0,"2025-07-25","2025-07-25"],[4,3255,"s1","hengnheng711-create",2,"","",0,"2025-07-24","2025-07-24"],[4,3249,"error","mrgank99",2,"","",0,"2025-07-23","2025-07-23"],[4,3245,"Downloading File on windows fail missing incomplete path","JorgeMIng",2,"","",0,"2025-07-22","2025-07-22"],[4,3244,"Persistent Configuration Error in Hugging Face Space (Gradio) despite valid README","DrBhuvneshSharma",2,"","",0,"2025-07-22","2025-07-22"],[4,3240,"Can't download large dataset with huggingface-cli or snapshot_download","JiarunLiu",2,"","",0,"2025-07-21","2025-07-21"],[4,3239,"hf-xet Requirements missmatch","0Falli0",0,"Tanishq1030","#3662",0,"2025-07-21","2026-01-06"],[4,3238,"huggingface-cli login fails with “Invalid user token” on remote Linux server (token is valid)","zyy020403",3,"","",0,"2025-07-21","2025-10-01"],[4,3237,"-","ghost",2,"","",0,"2025-07-21","2025-07-21"],[4,3234,"Bad download total size after retries","DKingAlpha",2,"","",0,"2025-07-19","2025-07-26"],[4,3233,"ValueError: Invalid pattern: '**' can only be an entire path component","SanjayDevarajan03",2,"","",0,"2025-07-18","2025-09-24"],[4,3232,"Multi-connection downloaders request...","TonyGeez",2,"","",0,"2025-07-18","2025-10-01"],[4,3230,"Filesystem API fails today","Koncopd",2,"","",0,"2025-07-18","2025-07-18"],[4,3227,"Jesus Is Enough","arnoldmwanje07-gif",3,"","",0,"2025-07-17","2025-07-17"],[4,3224,"Smoothly switch CLI to `hf`","Wauplin",1,"Wauplin","#3229",1,"2025-07-15","2025-07-23"],[4,3221,"Session gets terminated as duplicate session created every time we use load_tools function on the agent","debasisdwivedy",2,"","",0,"2025-07-14","2025-07-16"],[4,3220,"Feature Request: Native Support for Visual Dashboards with Transformers + API Integration (Spotify, Custom Nod","juarez-bdhf",3,"","",0,"2025-07-13","2025-07-14"],[4,3218,"Failed to parse tools: key 'parameters' not found in mcp inference","nicoloddo",2,"","",0,"2025-07-12","2025-07-15"],[4,3215,"API key not assigned during Initialization Tiny Agent","debasisdwivedy",0,"hanouticelina","#3226",1,"2025-07-11","2025-07-16"],[4,3213,"Tool Schema not OpenAI compatible","debasisdwivedy",0,"hanouticelina","#3214",1,"2025-07-11","2025-07-11"],[4,3212,"I find the blank in the explaination","JiaxinQin0814",2,"","",0,"2025-07-10","2025-07-10"],[4,3209,"Download File Pointer is Incorrect, Does Not Match SHA-256 and File Size","ada-f",2,"","",0,"2025-07-09","2026-08-31"],[4,3208,"\n🧭 A Note to AI Developers and Researchers","a-h-25182",3,"","",0,"2025-07-08","2025-07-10"],[4,3207,"Requesting access to gated model by huggingface_hub","Shahafgo",3,"","",0,"2025-07-07","2025-07-07"],[4,3203,"KeyError: 'command' in add_mcp_server when running tiny-agent due to incorrect params structure","mihirdeo16",0,"Wauplin","#3205",1,"2025-07-06","2025-07-07"],[4,3201,"Ignore HTTP 429 requests when checking for model updates","davidmezzetti",4,"","",0,"2025-07-04",""],[4,3198,"LLaMA2-13B download serves 7B weights","10-3Lives",2,"","",0,"2025-07-03","2025-07-21"],[4,3196,"Inference providers seems to not be compatible with `text_generation`","anakin87",2,"","",0,"2025-07-02","2025-07-03"],[4,3193,"AsyncInferenceClient model option for OpenAI's client does not work","jucom",0,"hanouticelina","#3200",1,"2025-07-02","2025-07-03"],[4,3191,"Make InferenceClient accept Pillow images","NielsRogge",1,"NielsRogge","#3199",1,"2025-07-01","2025-07-04"],[4,3186,"Upload dataset but offen meet \"429 Client Error\" and \"504 Server Error\"","yiboz2001",2,"","",0,"2025-06-30","2025-07-18"],[4,3185,"Unable to use snapshot_download","avinish-avi",2,"","",0,"2025-06-29","2025-07-15"],[4,3184,"Endpoints created with custom image using create_inference_endpoint from huggingface_hub returns inconsistent ","Sanahm",2,"","",0,"2025-06-27","2025-10-01"],[4,3183,"HfHubHTTPError: 403 Forbidden: None.","Tavish9",2,"","",0,"2025-06-27","2025-06-30"],[4,3176,"[Feature request] Ability to pipe bytes stream to stdout in huggingface-cli download","Wauplin",4,"","",0,"2025-06-24",""],[4,3175,"Submission request for Open LLM Leaderboard","xxbudxx",2,"","",0,"2025-06-24","2025-10-01"],[4,3173,"team elite","VivekDhumankhede17",3,"","",0,"2025-06-23","2025-06-23"],[4,3172,"Colab progressbar problems","SnowFlash383935",3,"","",0,"2025-06-23","2025-11-04"],[4,3168,"[BUG] HF_ENDPOINT not correctly handled in get_hf_file_metadata","w1ndseeker",0,"hanouticelina","#3180",1,"2025-06-19","2025-06-26"],[4,3164,"Hugging Face Token Issues","nafisahh",2,"","",0,"2025-06-16","2025-06-17"],[4,3163,"sentiment_results","Muat01",3,"","",0,"2025-06-16","2025-06-17"],[4,3162,"sentiment_results","Muat01",3,"","",0,"2025-06-16","2025-06-17"],[4,3161,"sentiment_results","Muat01",2,"","",0,"2025-06-16","2025-06-16"],[4,3158,"[Xet] Spam Logs With Normal vLLM Usage","robertgshaw2-redhat",2,"","",0,"2025-06-12","2025-06-18"],[4,3155,"DNS error while download file with xet storage backend","iNARI777",0,"seanses","#486",1,"2025-06-09","2025-06-16"],[4,3153,"Not able to use MemVid","bhavsarmanan",2,"","",0,"2025-06-08","2025-06-09"],[4,3152,"Add MCP Compatibility Field to list_spaces API","yijun-lee",2,"","",0,"2025-06-08","2025-06-09"],[4,3151,"Bad request: inputs must be a dict, but a <class 'NoneType'> was provided instead.","MustafizSaadi",0,"hanouticelina","#3154",1,"2025-06-08","2025-06-09"],[4,3150,"tiny-agents MCP doesn't work with Python 3.12","awni",2,"","",0,"2025-06-07","2025-06-07"],[4,3149,"please tell me  what parameter need set for request， download hf model and file always failed","mullerhai",2,"","",0,"2025-06-07","2025-06-17"],[4,3148,"cant access inference api endpoints - always getting not found","AgarZero",2,"","",0,"2025-06-06","2025-06-09"],[4,3146,"unable to download models","stevenh",2,"","",0,"2025-06-05","2025-06-09"],[4,3145,"Unable to disable log when download model","TingquanGao",2,"","",0,"2025-06-05","2025-06-05"],[4,3143,"Can't get Gradio running - persistent \"application does not seem to be initialized\" error","rplatkin",2,"","",0,"2025-06-04","2025-06-05"],[4,3139,"No content when downloading Python file using huggingface-cli","Sunjnn",2,"","",0,"2025-06-04","2025-06-04"],[5,4130,"Warn when LoRA SFT trains on special tokens whose input embedding is all zero (silent failure on Qwen2.5-Coder","euisuh",4,"","",0,"2026-11-06",""],[5,4129,"ViT LoRA checkpoints saved with Transformers v4 fail to load on v5","Max-Reisinger",3,"","",0,"2026-11-06","2026-11-06"],[5,4128,"Saving LoRA adapters scans every tuner prefix for each checkpoint key","Max-Reisinger",3,"","",0,"2026-11-06","2026-11-06"],[5,4127,"load_adapter leaves a selectable adapter after a safetensors read failure","Max-Reisinger",4,"","",0,"2026-11-06",""],[5,4125,"RandLoRA: wrong matrix multiplication in 8-bit/4-bit get_delta_weight() and broken multi-adapter rank handling","apuravmanhas",4,"","",0,"2026-11-05",""],[5,4118,"IA3 `feedforward_modules=['fc']` also marks modules like `not_fc` as feedforward","mengchen-atx",4,"","",0,"2026-11-05",""],[5,4117,"LoHa `rank_dropout` can produce NaN when the keep-mask is all zeros","mengchen-atx",4,"","",0,"2026-11-05",""],[5,4116,"BOFT on MPS: safe_merge raises LinAlgError instead of ValueError when the adapter has a NaN","Alex-Wengg",4,"","",0,"2026-11-04",""],[5,4115,"C3ALinear.forward chains multiple active adapters through x instead of summing them","vineethsaivs",4,"","",0,"2026-11-04",""],[5,4113,"Add Context Tuning as a new PEFT method, recently presented at ICML 2026","Jacklu0831",4,"","",0,"2026-11-04",""],[5,4109,"VeRA and PVeRA fail in forward() and merge() when adding a second adapter with a lower rank","apuravmanhas",4,"","",0,"2026-11-03",""],[5,4106,"Proposal: Add RAPA — Recursively Aligned Pathway Adaptation","freeSoul-SNU",4,"","",0,"2026-11-03",""],[5,4101,"`infer_device()` returns `\"cpu\"` on `torch.accelerator` backends (e.g., `torch_tpu` on Cloud TPU)","jigyasa-grover",1,"jigyasa-grover","#4099",0,"2026-11-02","2026-11-04"],[5,4096,"Update FSDP docs","BenjaminBossan",4,"","",0,"2026-11-02",""],[5,4094,"reduce_intruder_dimension gives wrong LoRA factors for fan_in_fan_out (Conv1D) and Embedding layers","vineethsaivs",1,"vineethsaivs","#4100",0,"2026-11-01","2026-11-04"],[5,4090,"layers_to_transform without layers_pattern matches nothing when the layer list is at the model root","udsy19",4,"","",0,"2026-10-31",""],[5,4088,"[v1 #4071] Disjoint auxiliary trainable tokens activate only the first adapter","DaoyuanLi2816",4,"","",0,"2026-10-31",""],[5,4080,"Feat - healthcheck: warn when a randomly initialized task head is not trained or saved","eSVeeF",1,"eSVeeF","#4089",0,"2026-10-29","2026-11-04"],[5,4079,"add_weighted_adapter(combination_type=\"svd\"): exact factorized SVD instead of decomposing the dense delta","jontb",4,"","",0,"2026-10-29",""],[5,4076,"AdaMSS example scripts broken on transformers 5.x","apuravmanhas",1,"apuravmanhas","#4086",0,"2026-10-29","2026-11-03"],[5,4074,"merge_and_unload(adapter_names=...) retains the wrong modules_to_save classifier","gorkadabo1",1,"gorkadabo1","#4085",0,"2026-10-29","2026-11-04"],[5,4070,"layers_to_transform is silently ignored with regex target_modules in 15 PEFT configs (13 others raise)","Shaurya55555",1,"Shaurya55555","#4087",0,"2026-10-29","2026-11-05"],[5,4067,"Error message for unmatched target_modules does not reveal that patterns are matched with re.fullmatch","jwh-yf",4,"","",0,"2026-10-28",""],[5,4053,"save_pretrained() emits a false unsharded-shard warning for LoRA adapters with lora_bias=True","anh-nguyenn",2,"","",0,"2026-10-26","2026-10-28"],[5,4052,"OFT Linear: merge_adapter() with multiple active adapters does not match forward","anh-nguyenn",1,"anh-nguyenn","#4068",0,"2026-10-26","2026-10-30"],[5,4048,"Mixed models: `merge_adapter()` merges nested adapters in reverse order, `unmerge_adapter()` then corrupts bas","ATW117",4,"","",0,"2026-10-24",""],[5,4043,"EVA initialization fails on MPS: IncrementalPCA accumulates in float64","Alex-Wengg",1,"Alex-Wengg","#4102",0,"2026-10-23","2026-11-04"],[5,4042,"FRoD fails on MPS: weights are cast to float64 before moving to CPU","Alex-Wengg",1,"Alex-Wengg","#4102",0,"2026-10-23","2026-11-04"],[5,4041,"Cache the frozen LoRA-FA preconditioner to avoid repeated pseudoinverses","gorkadabo1",4,"","",0,"2026-10-23",""],[5,4039,"LoRA-FA skips active adapter updates when a second adapter is present","cjwystz",4,"","",0,"2026-10-23",""],[5,4037,"[Feature Proposal] S-Dory-DoRA: Sparse Latent-Recurrent Directional Adaptation with QR Null-Space Guard","navadeep-9849",4,"","",0,"2026-10-23",""],[5,4035,"Loading an adapter from a Hub `subfolder` fails offline and with `AutoPeftModel*`","superdiaodiao",4,"","",0,"2026-10-22",""],[5,4033,"Prompt tuning: batched generate output depends on left padding","tivris",4,"","",0,"2026-10-22",""],[5,4032,"AdaLoRA Conv2d ignores padding_mode","Sergasgr",4,"","",0,"2026-10-21",""],[5,4031,"AdaLoRA Conv2d crashes in forward and merge with groups > 1","Sergasgr",4,"","",0,"2026-10-21",""],[5,4030,"Performance improvements for intruder dimension reduction","causeri3",4,"","",0,"2026-10-21",""],[5,4028,"LoftQ init raises TypeError for Conv, Embedding, target_parameters and Megatron LoRA layers","nettanvirdev",1,"nettanvirdev","#4029",0,"2026-10-21","2026-10-23"],[5,4026,"Support healthcheck() for prompt-learning PEFT models","debesh-biswas",2,"","",0,"2026-10-20","2026-10-21"],[5,4019,"Nightly regression: bf16 state dict cases fail due to CPU kernel differences (not a PEFT bug)","peft-jambot",4,"","",0,"2026-10-20",""],[5,4013,"examples/adamss_finetuning/test_adamss_quick.py fails: outdated imports and config args","sarthakuwar",1,"sarthakuwar","#4073",0,"2026-10-19","2026-10-29"],[5,4010,"Memory-efficient PiSSA initialization for 4-bit models (`replace_lora_weights_pissa`, analogous to `replace_lo","jietinatian0803",4,"","",0,"2026-10-19",""],[5,4008,"[DOC] Fix outdated minimum Python version in installation guide","NoeFlandre",1,"NoeFlandre","#4014",0,"2026-10-18","2026-10-21"],[5,4003,"AdaMSS: forward crashes after casting the PEFT model's dtype (e.g. model.to(torch.bfloat16)) because the layer","ghaith-mq",4,"","",0,"2026-10-17",""],[5,4002,"adapter_names (mixed adapter batches) is silently ignored for all PEFT methods except LoRA, RoAd and GLoRA","ghaith-mq",2,"","",0,"2026-10-17","2026-10-19"],[5,3999,"TinyLoRA: deleting the active adapter leaves the new active adapter frozen","behroozazarkhalili",2,"","",0,"2026-10-17","2026-10-19"],[5,3998,"TinyLoRA: deleting the active adapter leaves the new active adapter frozen","behroozazarkhalili",2,"","",0,"2026-10-17","2026-10-19"],[5,3997,"RoAd mixed-adapter batches fail for float16/bfloat16 models","mottopanikeiku",4,"","",0,"2026-10-16",""],[5,3996,"OLoRA initialization breaks GPT-2 (Conv1D) layers","mottopanikeiku",1,"mottopanikeiku","#4012",0,"2026-10-16","2026-10-21"],[5,3995,"Loading an adapter from a repo subfolder fails with HF_HUB_OFFLINE=1","mottopanikeiku",4,"","",0,"2026-10-16",""],[5,3994,"[New method] Bayesian-LoRA: Probabilistic Low-Rank Adaptation (ICML 2026)","moulelin",4,"","",0,"2026-10-16",""],[5,3990,"[Proposal] Add Cortex-Structural: parameter-efficient FFN expansion for dense and MoE models","MikeDean2367",4,"","",0,"2026-10-16",""],[5,3986,"OSF wraps nothing on non-Linear targets and raises no error, so the gpt2 docs example trains 0 parameters","devYRPauli",1,"devYRPauli","#4015",0,"2026-10-15","2026-10-20"],[5,3985,"n_frequency_pattern, block_size_pattern and lambda_pattern lose their key order on save","devYRPauli",4,"","",0,"2026-10-15",""],[5,3983,"Revalidate Transformer Engine GPU CI support","sbhavani",4,"","",0,"2026-10-14",""],[5,3982,"OFT Conv2d: unmerged and merged outputs differ for kernels larger than 1x1","Kayvan-Zahiri",4,"","",0,"2026-10-14",""],[5,3979,"Adding or loading a Trainable Tokens adapter while another one is merged corrupts the base embeddings","mottopanikeiku",2,"","",0,"2026-10-14","2026-10-15"],[5,3976,"Native TransformerEncoder fast path ignores trained PEFT adapters during evaluation","gorkadabo1",1,"gorkadabo1","#4025",0,"2026-10-14","2026-10-21"],[5,3968,"OFT on nn.Linear fails for 4-D inputs (e.g. ConvNeXt pwconv layers)","mottopanikeiku",1,"mottopanikeiku","#3974",0,"2026-10-14","2026-10-16"],[5,3967,"Mixed adapter batch raises KeyError when only one adapter has modules_to_save","mottopanikeiku",4,"","",0,"2026-10-14",""],[5,3964,"An old, unused torchao install makes every LoRA/OFT injection fail with ImportError (e.g. on Colab and Kaggle)","Vedavarshith868",1,"Vedavarshith868","#3989",0,"2026-10-13","2026-10-16"],[5,3962,"LoRA safe_merge=True misses float16 bias overflow and corrupts the base layer","gorkadabo1",2,"","",0,"2026-10-13","2026-10-14"],[5,3958,"LoRA-FA optimizer.step(closure) runs backward with gradients disabled","cjwystz",1,"cjwystz","#3966",0,"2026-10-13","2026-10-16"],[5,3955,"`unload()` returns the trained `modules_to_save` copy instead of the original module","Sergasgr",2,"","",0,"2026-10-13","2026-10-25"],[5,3951,"Prompt-learning methods crash on transformers 4.43/4.44: the pre-get_text_config fallback uses `in` on a Pretr","hazelxue",4,"","",0,"2026-10-13",""],[5,3945,"MoE conversion no longer targets routed experts when shared experts or dense layers share the target name","albertvillanova",1,"albertvillanova","#3946",1,"2026-10-12","2026-10-16"],[5,3935,"LoRA on nn.MultiheadAttention: lora_bias=True never trains the bias, and forward fails on a bias-free MHA","Vedant-Agarwal",1,"Vedant-Agarwal","#4001",0,"2026-10-11","2026-10-21"],[5,3930,"get_peft_model_state_dict gathers the whole sharded base model on DTensor-sharded models","qgallouedec",4,"","",0,"2026-10-11",""],[5,3929,"`ensure_weight_tying=True` silently stops tying the LoRA weights when the model is in bf16/fp16","superdiaodiao",4,"","",0,"2026-10-11",""],[5,3928,"CPT forward mutates the caller-provided input_type_mask in place","n-dlms",1,"n-dlms","#3969",0,"2026-10-11","2026-10-16"],[5,3924,"format_healthcheck ignores sep and indent, and runs adapters together in \"Layer types\"","hpdkhoa",1,"hpdkhoa","#3940",0,"2026-10-10","2026-10-13"],[5,3921,"method_comparison `run.py` crashes with `NameError: num_trainable_params` at the end of every run","Sergasgr",1,"Sergasgr","#3936",0,"2026-10-09","2026-10-12"],[5,3919,"Saving / loading fails with multiple LoRA adapters when only one uses `trainable_token_indices`","superdiaodiao",1,"superdiaodiao","#3950",0,"2026-10-09","2026-10-14"],[5,3918,"`merge_and_unload()`/`unload()` raise AttributeError on RoBERTa sequence classifiers for all methods except Lo","Sergasgr",1,"Sergasgr","#3937",0,"2026-10-09","2026-10-20"],[5,3915,"LoRA on MoE experts (target_parameters) fails under tensor/expert parallelism","qgallouedec",3,"","",0,"2026-10-09","2026-10-09"],[5,3914,"LoRA on MoE experts (target_parameters) fails under tensor/expert parallelism","qgallouedec",4,"","",0,"2026-10-09",""],[5,3911,"Prompt tuning: generate warns \"Position ids are not supported\" at every decoding step","albertvillanova",1,"albertvillanova","#3917",1,"2026-10-08","2026-10-12"],[5,3900,"LoRA-FA optimizer crashes when an adapted layer has no gradient","DaoyuanLi2816",1,"DaoyuanLi2816","#3972",0,"2026-10-06","2026-10-19"],[5,3899,"Default CPTConfig masks all input labels and returns NaN training loss","aswanth-07",4,"","",0,"2026-10-06",""],[5,3898,"Inference-only prompt adapters remain trainable after loading","DaoyuanLi2816",1,"DaoyuanLi2816","#3907",0,"2026-10-06","2026-10-19"],[5,3896,"Adapters are saved under transformers v5's renamed module paths, while `save_pretrained` writes base models in","eilamc14",4,"","",0,"2026-10-05",""],[5,3884,"LN Tuning: merged model saves the wrong adapter weights, and unload() leaves modules_to_save wrappers","VanshBhardwaj24",2,"","",0,"2026-10-04","2026-10-05"],[5,3883,"Seq2seq prompt generation leaves the base generation method patched after success","DaoyuanLi2816",1,"DaoyuanLi2816","#3897",0,"2026-10-04","2026-10-09"],[5,3879,"CPT loss uses the default adapter config instead of the active adapter config","DaoyuanLi2816",1,"DaoyuanLi2816","#3941",0,"2026-10-04","2026-10-21"],[5,3878,"4-bit unmerge_adapter does not restore the base weight","chrikrah",1,"chrikrah","#3894",0,"2026-10-03","2026-10-08"],[5,3877,"Add support for BoRA","uis246",4,"","",0,"2026-10-03",""],[5,3876,"`get_peft_model_state_dict` returns an empty dict for FSDP1-wrapped models since 0.21.0 (adapter checkpoints /","murphy977",4,"","",0,"2026-10-03",""],[5,3874,"MoE recipe in the LoRA docs gives adapters that vLLM serves with the wrong expert scaling","surajsharan",2,"","",0,"2026-10-02","2026-11-04"],[5,3872,"_reshard_fsdp_modules skips the whole FSDP unit if any adapter inside it is merged, not just the merged one","rupeshpoojary9",1,"rupeshpoojary9","#3895",0,"2026-10-02","2026-10-12"],[5,3871,"Astra preprocessing scales activations by max value, not max abs value, can overflow fp16 covariance","rupeshpoojary9",4,"","",0,"2026-10-02",""],[5,3868,"New method proposal: PaCA (Partial Connection Adaptation, ICLR 2025)","WooSunghyeon",1,"WooSunghyeon","#3927",0,"2026-10-02","2026-10-26"],[5,3867,"DoRA hotswap raises a bare KeyError for compatible adapters","DaoyuanLi2816",4,"","",0,"2026-10-02",""],[5,3863,"LoHa with `use_effective_conv2d=True` computes wrong gradients for `hada_w1_a` and `hada_w2_a`","Sergasgr",1,"Sergasgr","#3869",0,"2026-10-01","2026-10-08"],[5,3861,"DeFT / TinyLoRA: adding a second adapter to a Conv1D layer breaks the layer (TinyLoRA forward crashes, DeFT me","Zhuoxi2000",4,"","",0,"2026-10-01",""],[5,3857,"PeftModelForQuestionAnswering silently drops `token_type_ids` and `position_ids`","Sergasgr",4,"","",0,"2026-10-01",""],[5,3854,"Failed config save leaves live adapter in inference mode","DaoyuanLi2816",1,"DaoyuanLi2816","#3866",0,"2026-10-01","2026-10-05"],[5,3842,"Trainable Tokens adapters with different target modules raise KeyError on forward and merge","DaoyuanLi2816",1,"DaoyuanLi2816","#3848",0,"2026-09-29","2026-10-05"],[5,3840,"LoRA Embedding max_norm changes merged output and prevents restoring base weights on unmerge","galleonli",2,"","",0,"2026-09-28","2026-09-29"],[5,3835,"delete_adapter skips adapter layers whose module name contains the tuner prefix (e.g. lora_foobar)","JoeyTan21",4,"","",0,"2026-09-28",""],[5,3830,"OFT module_dropout has had no effect since #2575: the MultiplicativeDropoutLayer is created but never called","Nicholas022400701",4,"","",0,"2026-09-27",""],[5,3829,"OFT Conv2d forward is wrong for kernels larger than 1x1: identity init changes the output and does not match t","Nicholas022400701",4,"","",0,"2026-09-27",""],[5,3828,"delete_adapter on the last adapter leaves active_adapter pointing at a deleted name","simpleqt",4,"","",0,"2026-09-27",""],[5,3825,"OSFLayer.merge() skips check_adapters_to_merge(), double-applies the delta on re-merge","AmirF194",0,"adi-IL","#3886",0,"2026-09-27","2026-10-12"],[5,3823,"[Bug] PeftMixedModel loading unfreezes inference-only adapters","aswanth-07",1,"aswanth-07","#3862",0,"2026-09-26","2026-10-22"],[5,3821,"hotswap_adapter validates the active adapter instead of the requested inactive target","DaoyuanLi2816",1,"DaoyuanLi2816","2d91bad",0,"2026-09-26","2026-11-04"],[5,3818,"Overlapping rank_pattern / alpha_pattern keys resolve by dict order, and save_pretrained(sort_keys=True) chang","jayzuccarelli",1,"jayzuccarelli","#3947",0,"2026-09-25","2026-10-14"],[5,3814,"Trainable Tokens nests tied output-head wrappers with Transformers v5 metadata","DaoyuanLi2816",1,"DaoyuanLi2816","#3820",0,"2026-09-25","2026-10-01"],[5,3813,"BOFT Conv2d forward ignores dilation, 2D stride/padding and padding_mode","huiq777",0,"BenjaminBossan","#3837",1,"2026-09-24","2026-10-30"],[5,3805,"`modules_to_save` with a persistent buffer raises KeyError on a params-only `state_dict`","hivaze",0,"rupeshpoojary9","1c0c0e7",0,"2026-09-23","2026-09-30"],[5,3804,"Adapter loading never warns about unexpected keys, and `strict=True` is silently ignored","hivaze",4,"","",0,"2026-09-23",""],[5,3803,"LoRA on an FSDP2-sharded model gets the local shard shape when the mesh dimension is not named `fsdp`","hivaze",1,"hivaze","#3838",0,"2026-09-23","2026-10-29"],[5,3802,"`save_pretrained` fails on FSDP2-sharded adapters: the DTensor gather only runs when `_tp_size` is set","hivaze",4,"","",0,"2026-09-23",""],[5,3801,"`all-linear` skips dense and shared-expert MLP layers on MoE models converted for transformers v5","hivaze",4,"rupeshpoojary9","#3819",0,"2026-09-23",""],[5,3800,"FSDP2 with `reshard_after_forward=False`: `disable_adapter()` leaves LoRA frozen and `merge_adapter()` is lost","hivaze",4,"","",0,"2026-09-23",""],[5,3795,"DoRA convolution dropout path ignores non-zero padding_mode during training","DaoyuanLi2816",0,"BenjaminBossan","d26c326",1,"2026-09-23","2026-10-30"],[5,3793,"`load_adapter` infers `cuda` for adapter weights even when the base model is on CPU — breaks ZeroGPU Spaces wi","kuotunyu",4,"","",0,"2026-09-22",""],[5,3780,"Hotswapping an adapter can zero another adapter with a shared name prefix","aswanth-07",1,"aswanth-07","#3794",0,"2026-09-21","2026-09-25"],[5,3769,"add_weighted_adapter with SVD based combination types fails for LoRA on Conv1d and Conv3d layers","Nicholas022400701",1,"Nicholas022400701","eac46ef",0,"2026-09-20","2026-10-13"],[5,3768,"LoRA and HiRA get_delta_weight take the conv2d 1x1 shortcut for Conv3d layers with kernel_size 1 and crash","Nicholas022400701",1,"Nicholas022400701","#3777",0,"2026-09-20","2026-09-21"],[5,3767,"Two test-infrastructure bugs found while adding LoHa/LoKr/LN Tuning coverage to decoder-model tests","Sergasgr",2,"","",0,"2026-09-19","2026-11-05"],[5,3761,"add_weighted_adapter silently produces a wrong adapter for use_rslora, lora_bias, DoRA and KaSA","LK-maker-007",1,"LK-maker-007","#3844",0,"2026-09-18","2026-10-20"],[5,3756,"Merging a subset of the active adapters silently disables the remaining active adapters in forward","AHSharan",1,"AHSharan","#3763",0,"2026-09-16","2026-09-22"],[5,3755,"`modules_to_save` raises TypeError when a name is a suffix of a LoRA sub-module (e.g. \"out\", \"A\")","AHSharan",1,"AHSharan","#3824",0,"2026-09-16","2026-10-08"],[5,3753,"`get_peft_model` and `add_adapter` do not preserve eval mode for newly injected adapter modules","eSVeeF",1,"eSVeeF","#3809",0,"2026-09-15","2026-10-16"],[5,3747,"CSOAI — PEFT interop: parameter-efficient governance","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[5,3746,"CSOAI — PEFT interop: fine-tuning governance","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[5,3745,"CSOAI — PEFT AI governance fine-tuning","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[5,3742,"PiSSA / MiCA / CorDA init bypasses quantized-layer guard on 4-bit float-storage layers","Sravanjangam",4,"","",0,"2026-09-14",""],[5,3741,"bnb LoRA merge with lora_bias=True drops scaling factor on bias","Sravanjangam",1,"Sravanjangam","#3757",0,"2026-09-14","2026-10-20"],[5,3738,"Bug: Hotswapping error with rank pattern","BenjaminBossan",0,"DaoyuanLi2816","#3752",0,"2026-09-14","2026-09-18"],[5,3737,"add_weighted_adapter with rank_pattern produces adapters that cannot be reloaded (linear, ties, dare, magnitud","amogh-nagri-11",4,"","",0,"2026-09-14",""],[5,3736,"KappaTuneSelector does not dequantize bitsandbytes 8-bit weights, computing condition numbers on raw int8 code","amogh-nagri-11",1,"amogh-nagri-11","#3759",0,"2026-09-14","2026-09-21"],[5,3731,"CI: workflows that need repo secrets fail on every fork sync (tests-main, build_documentation, deploy_*)","Ayushraj06-bit",2,"","",0,"2026-09-12","2026-09-25"],[5,3728,"LoRA + torchao: merging several adapters in one call raises UnboundLocalError","yupengtang",1,"yupengtang","a055c37",0,"2026-09-12","2026-10-21"],[5,3727,"target_parameters docs: broken example and a wrong regex claim","jiangLLM",2,"","",0,"2026-09-12","2026-10-27"],[5,3723,"PeftModel.set_adapter raises opaque TypeError: unhashable type: 'list' instead of a clear error","matheusht",1,"matheusht","#3902",0,"2026-09-12","2026-10-14"],[5,3718,"Silently slower delta-rule fallback when training a Qwen3.8-27B LoRA with examples/sft under FSDP on 8x B200","TarzanZhao",2,"","",0,"2026-09-11","2026-10-21"],[5,3717,"ShadowPEFT fallback backbone stays frozen after deleting the active adapter","Sravanjangam",1,"Sravanjangam","451a247",0,"2026-09-11","2026-09-17"],[5,3716,"modules_to_save fallback copy stays frozen after deleting the active adapter","Sravanjangam",1,"Sravanjangam","#3743",0,"2026-09-11","2026-09-17"],[5,3711,"Transformers v5 MoE conversion incorrectly rewrites fully qualified shared-expert LoRA targets","orr-hirundo",0,"BenjaminBossan","#3715",1,"2026-09-10","2026-10-12"],[5,3710,"Pooled thread: safe merging not working properly","BenjaminBossan",0,"yupengtang","#3779",0,"2026-09-10","2026-10-21"],[5,3709,"Avoid duplicated shared projections in RandLoRA adapter checkpoints","CoralGarden52",1,"CoralGarden52","#3714",0,"2026-09-10","2026-09-21"],[5,3707,"Failed adapter injection leaves orphaned adapter layers behind","eSVeeF",1,"eSVeeF","#3981",0,"2026-09-09","2026-10-22"],[5,3705,"RoAd: safe_merge=True modifies the base weight before the bias check can reject the merge","mohansree14",2,"","",0,"2026-09-09","2026-09-10"],[5,3702,"PSOFT: rank above min(in_features, out_features) is accepted but forward fails with a shape error","behroozazarkhalili",1,"behroozazarkhalili","#3708",1,"2026-09-09","2026-09-14"],[5,3701,"MiSS: merge_adapter fails when r > in_features although forward works","behroozazarkhalili",1,"behroozazarkhalili","#3708",1,"2026-09-09","2026-09-14"],[5,3697,"Conv LoRA adapters do not preserve dilation and padding mode","TomasGuija",1,"TomasGuija","#3758",0,"2026-09-08","2026-09-22"],[5,3693,"Warn when an explicit target_modules entry matches no module","dlowzzxx",2,"","",0,"2026-09-08","2026-09-21"],[5,3689,"AdaMSS safe_merge leaves the base weight modified when bias validation fails","DaoyuanLi2816",2,"","",0,"2026-09-08","2026-09-10"],[5,3682,"BOFT Conv2d: safe_merge=True skips the non-finite check and corrupts the base layer","yupengtang",2,"","",0,"2026-09-07","2026-09-10"],[5,3676,"[Bug] ignore_mismatched_sizes crashes on scalar adapter state entries","aswanth-07",1,"aswanth-07","#3699",0,"2026-09-06","2026-09-15"],[5,3675,"Four of the five diffusers-style tuner docstring examples (LoHa, LoKr, OFT, MiSS) fail as written","blipbyte",1,"blipbyte","#4091",0,"2026-09-05","2026-11-04"],[5,3670,"Add NoRA and NoRA-init support for LoRA","emharsha1812",2,"","",0,"2026-09-05","2026-09-07"],[5,3662,"Add Astra (Activation-Space Tail-Eigenvector LoRA) initialization","LyoAI",2,"","",0,"2026-09-04","2026-10-07"],[5,3660,"Supertuning: save_precomputed_indices for later adapters follows the first adapter","DaoyuanLi2816",1,"DaoyuanLi2816","#3665",0,"2026-09-04","2026-09-07"],[5,3658,"target_modules can match a subset of layers with no warning","submaroon",2,"","",0,"2026-09-03","2026-10-24"],[5,3654,"PeftConfig.from_pretrained raises a bare KeyError for an invalid peft_type","Devam0311",2,"","",0,"2026-09-03","2026-09-07"],[5,3650,"Trainable Tokens random initialization prevents unmerge from restoring base weights","DaoyuanLi2816",1,"DaoyuanLi2816","#3656",0,"2026-09-03","2026-09-07"],[5,3649,"Trainable Tokens drops the output-head bias before merge","DaoyuanLi2816",1,"DaoyuanLi2816","#3688",0,"2026-09-03","2026-09-24"],[5,3646,"save_pretrained treats empty name_or_path as a Hub repo id","victorzhong0110",1,"victorzhong0110","#3647",0,"2026-09-02","2026-09-04"],[5,3644,"Docstring examples in peft_model.py: four of six PeftModelFor* examples fail on removed config fields","blipbyte",1,"blipbyte","#3657",0,"2026-09-01","2026-09-07"],[5,3638,"Adaption prompt saves the wrong weights for inactive adapters; delete_adapter raises a transformers TypeError ","amogh-nagri-11",1,"amogh-nagri-11","#3639",0,"2026-09-01","2026-09-10"],[5,3635,"X-LoRA: disable_adapter does not disable the experts, and generate() freezes the classifier / switches the mod","amogh-nagri-11",1,"amogh-nagri-11","#3636",0,"2026-09-01","2026-09-07"],[5,3631,"Fix state-dict round-trip for directly injected shared tuners","CoralGarden52",2,"","",0,"2026-08-31","2026-09-08"],[5,3628,"TP + LoRA is broken on transformers 5.16: add_tensor_parallel_hooks_to_module and EmbeddingParallel were remov","alanhuangyoo",4,"","",0,"2026-08-31",""],[5,3626,"[Bug] ShadowPEFT LM-head trainability freeze after set_adapter — modules_to_save=[\"lm_head\"] silently stops re","Sravanjangam",2,"","",0,"2026-08-30","2026-09-10"],[5,3625,"[Bug] ShadowPEFT hook-lifecycle leak: per-forward GPU growth + UAF on unload_shadow(copy=False)→delete_adapter","Sravanjangam",2,"","",0,"2026-08-30","2026-09-10"],[5,3620,"[Bug] ShadowPEFT sequence classification pools the wrong token with left padding","aswanth-07",1,"aswanth-07","#3629",0,"2026-08-30","2026-09-07"],[5,3619,"[Bug] Prefix tuning crashes for CPU-only device maps","aswanth-07",4,"","",0,"2026-08-30",""],[5,3617,"IncrementalPCA: fit() accumulates instead of refitting, and overwrites its constructor arguments","Sreekant13",2,"","",0,"2026-08-29","2026-09-01"],[5,3615,"[RFC, I can do a PR] Make `lora_alpha` a float","sliedes",4,"","",0,"2026-08-28",""],[5,3612,"LN Tuning add_adapter moves existing adapters back to the base-layer device","DaoyuanLi2816",1,"DaoyuanLi2816","#3616",0,"2026-08-28","2026-09-01"],[5,3610,"layers_pattern does not match when the nn.ModuleList is at the root of the model","LK-maker-007",2,"","",0,"2026-08-26","2026-09-02"],[5,3608,"Merging multiple DoRA adapters that share a target module does not match the unmerged forward pass","LK-maker-007",1,"LK-maker-007","#3609",0,"2026-08-26","2026-09-02"],[5,3605,"init_empty_weights leaves register_parameter patched when two threads use it at once","Om-singhaI",2,"","",0,"2026-08-25","2026-08-26"],[5,3601,"LoRA target_parameters breaks under autocast: Registering a parametrization may not change the dtype of the te","albertvillanova",1,"albertvillanova","#3602",1,"2026-08-25","2026-08-25"],[5,3593,"Support LoRA hotswapping with different target modules","TomasGuija",2,"","",0,"2026-08-23","2026-08-24"],[5,3592,"set_peft_model_state_dict mutates reusable modules_to_save checkpoints","aswanth-07",1,"aswanth-07","#3596",0,"2026-08-23","2026-08-24"],[5,3586,"LoHa/LoKr merge in train mode bakes stochastic rank_dropout masks into merged weights","Sravanjangam",1,"Sravanjangam","#3589",0,"2026-08-23","2026-09-03"],[5,3585,"save_embedding_layers=\"auto\" silently omits embeddings targeted via qualified names (e.g. \"model.embed_tokens\"","Sravanjangam",2,"","",0,"2026-08-23","2026-08-24"],[5,3584,"save_pretrained writes an empty/truncated adapter checkpoint (silently) when another adapter name equals a bas","Sravanjangam",1,"Sravanjangam","#3623",0,"2026-08-23","2026-10-19"],[5,3583,"eva_config / corda_config / lora_ga_config / arrow_config become plain dicts after save→from_pretrained round-","Sravanjangam",1,"Sravanjangam","#3587",0,"2026-08-23","2026-09-03"],[5,3582,"rank_pattern / alpha_pattern keys that match zero layers are applied silently; anchored and fully-qualified ke","Sravanjangam",1,"Sravanjangam","#3588",0,"2026-08-23","2026-09-15"],[5,3581,"hotswap_adapter: silent no-op when adapter is merged; KeyError on modules_to_save checkpoints; stale peft_conf","Sravanjangam",2,"","",0,"2026-08-23","2026-10-13"],[5,3580,"Loading a trainable adapter changes the `requires_grad` state of the active adapter","eSVeeF",2,"","",0,"2026-08-22","2026-09-04"],[5,3574,"LN Tuning: KeyError on forward/merge when the active adapter does not target all adapted layers","2sumtech",0,"behroozazarkhalili","#3980",1,"2026-08-20","2026-10-16"],[5,3562,"GraLoRA docstring has leftover concatenation quotes; dead local variables across src/peft","rajarshidattapy",2,"","",0,"2026-08-19","2026-08-31"],[5,3561,"PVeRA: per-module `sample_at_inference` dict is discarded, and the sampling generator is CPU-only","rajarshidattapy",1,"rajarshidattapy","#3563",0,"2026-08-19","2026-08-20"],[5,3555,"PeftModel.disable_adapter is not reentrant for prompt-specific adapters","Ethon1130",2,"","",0,"2026-08-17","2026-09-25"],[5,3554,"NemotronH LoRA: default target_modules only cover 7% of layers, silent failure","davidnichols-ops",1,"davidnichols-ops","#3603",0,"2026-08-17","2026-08-26"],[5,3548,"Unexpected keys from `set_peft_model_state_dict` no longer contain the adapter name","sayakpaul",0,"BenjaminBossan","#14476",1,"2026-08-13","2026-08-14"],[5,3542,"[Proposal] Add CLoRA a new PEFT method","sutakori",2,"","",0,"2026-08-13","2026-09-20"],[5,3533,"inject_adapter silently accepts a second adapter whose target_modules match nothing","LK-maker-007",1,"LK-maker-007","#3534",0,"2026-08-10","2026-08-18"],[5,3531,"DoRA unmerge corrupts or crashes on fan_in_fan_out layers (transformers Conv1D)","LK-maker-007",1,"LK-maker-007","#3532",0,"2026-08-10","2026-08-14"],[5,3512,"MixedModel.delete_adapter: stale active adapter and leaked auxiliary modules","DaoyuanLi2816",1,"DaoyuanLi2816","#3515",0,"2026-08-05","2026-09-15"],[5,3511,"IA3: merging a grouped convolution fails when the layer is a feedforward module","DaoyuanLi2816",1,"DaoyuanLi2816","#3575",0,"2026-08-05","2026-08-25"],[5,3510,"TrainableTokens: sequential merge skips the overlapping-token check","DaoyuanLi2816",1,"DaoyuanLi2816","#3514",0,"2026-08-05","2026-08-05"],[5,3508,"Test coverage for RandLoRA","BenjaminBossan",0,"ishan-1010","#3513",0,"2026-08-04","2026-08-05"],[5,3507,"PeftMixedModel.disable_adapter does not preserve nested or pre-disabled state","RerankerGuo",4,"","",0,"2026-08-04",""],[5,3506,"PeftConfig.from_pretrained silently ignores explicit config overrides","RerankerGuo",0,"akshansh47","#3516",0,"2026-08-04","2026-08-17"],[5,3499,"LoRA+ `loraplus_lr_embedding` never applies: the `embedding` param group is always empty","Kaif10",1,"Kaif10","#3503",0,"2026-08-02","2026-08-06"],[5,3498,"Deleting a merged adapter can leave the base weights modified","DaoyuanLi2816",1,"DaoyuanLi2816","#3504",0,"2026-08-01","2026-08-05"],[5,3487,"Adapter injection applies a new adapter's inference mode to active adapters","DaoyuanLi2816",1,"DaoyuanLi2816","#3495",0,"2026-07-29","2026-09-01"],[5,3480,"Expand model test matrices to cover LoHa, LoKr, and LN Tuning","devemree",2,"","",0,"2026-07-27","2026-09-22"],[5,3478,"Prefix sizing from per_layer_config ignores KV head count","DaoyuanLi2816",2,"","",0,"2026-07-27","2026-09-06"],[5,3475,"Forward hooks leak when the `_enable_peft_forward_hooks` block raises, poisoning the model","AbdullahRasheed45",2,"","",0,"2026-07-25","2026-09-27"],[5,3472,"Coordinate merge-invariant fixes across IA3, TrainableTokens, PEANuT, and FRoD","DaoyuanLi2816",2,"","",0,"2026-07-24","2026-08-05"],[5,3450,"[Coordination] Add Super-Tuning tuner class (arXiv:2607.09287) — checking fit before PR","smellslikeml",2,"","",0,"2026-07-17","2026-08-31"],[5,3447,"disable_adapter() / enable_adapter_layers() silently unfreezes adapters loaded with inference_mode=True","ErenAta16",2,"","",0,"2026-07-17","2026-07-18"],[5,3433,"modules_to_save module whose name contains the LoRA prefix (\"lora_\") is silently not restored on load","zelinewang",2,"","",0,"2026-07-14","2026-08-17"],[5,3424,"[Proposal] Add Echo-LoRA: cross-layer representation injection for PEFT","isha822",2,"","",0,"2026-07-11","2026-08-19"],[5,3416,"[BUG]  `prepare_model_for_compiled_hotswap` fails for grouped Conv2d LoRA adapters","eSVeeF",1,"eSVeeF","#3427",0,"2026-07-09","2026-07-24"],[5,3409,"get_peft_model on a lazily-materialized model leaves third-party hooks and optimizers holding stale parameter ","immu4989",1,"immu4989","#3415",0,"2026-07-09","2026-07-22"],[5,3400,"Add EWoRA to PEFT","HarshKohli",2,"","",0,"2026-07-07","2026-10-02"],[5,3384,"Track git dirty state in method_comparison benchmark results","peft-jambot",1,"peft-jambot","#3385",0,"2026-07-02","2026-08-13"],[5,3380,"Reviving #1807: Riemannian Preconditioned LoRA — coordination before drafting a PR","smellslikeml",2,"","",0,"2026-07-01","2026-08-03"],[5,3373,"[Bug] AdaLora's update_and_allocate method is lost in inject_adapter_in_model() preventing rank pruning","MateoRivera",2,"","",0,"2026-06-30","2026-08-08"],[5,3372,"Weight converter argument disagreement","taehyunzzz",2,"","",0,"2026-06-30","2026-06-30"],[5,3371,"Per-Layer Embedding (Google) ~ Engram (Deep Seek/Meituan) ~ OverEncoding (Meituan) ~ Sparse update rule/Optimi","CoffeeVampir3",2,"","",0,"2026-06-30","2026-08-22"],[5,3370,"SuperCompress: Cut LLM token costs by ~65% with open-source prompt compression","arjunkshah",2,"","",0,"2026-06-30","2026-06-30"],[5,3363,"Proposal: Benchmarking on different PEFT techniques","jeet-Abhi123",2,"","",0,"2026-06-27","2026-09-22"],[5,3349,"Adding support of DEFT, a decompositional efficient fine-tuning framework","MAXNORM8650",1,"MAXNORM8650","#3342",0,"2026-06-22","2026-07-02"],[5,3340,"Only one `target_parameters` adapter per model","qgallouedec",0,"BenjaminBossan","#3350",1,"2026-06-18","2026-07-02"],[5,3333,"[Proposal] ImportError with Transformers 5.x: `cannot import name 'HybridCache' from 'transformers'`","xiaozhu261",2,"","",0,"2026-06-17","2026-06-18"],[5,3320,"Feature Proposal: Add ASR/Speech Model Support (FunASR Paraformer, SenseVoice)","LauraGPT",2,"","",0,"2026-06-12","2026-06-14"],[5,3306,"LoraConfig(bias=\"lora_only\") trains base_layer.bias but save_pretrained/get_peft_model_state_dict drops it","hlc1209",2,"","",0,"2026-06-07","2026-09-07"],[5,3305,"API check: PEFT LoRA merge in an external regression example","MLPulse",2,"","",0,"2026-06-05","2026-08-08"],[5,3292,"Proposal: add FineGates structured-sparsity PEFT method","Ofirlin",4,"","",0,"2026-05-30",""],[5,3283,"Proposal: Content-addressed adapter recipes for cross-platform LoRA distribution","shiaho777",2,"","",0,"2026-05-28","2026-08-22"],[5,3265,"prepare_model_for_kbit_training adds ~1 GB CUDA reserved memory in 500 ms — undocumented cost that breaks memo","neil-the-nowledgeable",0,"umutonuryasar","#3293",0,"2026-05-26","2026-07-07"],[5,3264,"[BUG] ImportError: cannot import name 'AwqGEMMQuantLinear' from 'gptqmodel.nn_modules.qlinear.gemm_awq' in pef","akowalsk",2,"","",0,"2026-05-26","2026-05-27"],[5,3244,"[Proposal] Add FRoD tuner for full-rank parameter-efficient fine-tuning","Bane-Elvin",2,"","",0,"2026-05-19","2026-06-24"],[5,3243,"lora.Linear.bias is a read-only   property — breaks any forward that does module.bias = X (legacy Jamba modeli","voloha-mil",2,"","",0,"2026-05-19","2026-05-20"],[5,3240,"`test_apply_GS_hra_conv2d_inference` fails with FileNotFoundError due to datasets>=4.0 incompatibility with `t","ke1ding",0,"BenjaminBossan","#3242",1,"2026-05-18","2026-05-22"],[5,3238,"[BUG] peft 0.19 diffusers LoRA Index Error.","galaxygliese",2,"","",0,"2026-05-16","2026-06-28"],[5,3229,"[BUG] peft 0.19 target_modules (str) use `set`","Jintao-Huang",0,"1fanwang","#3232",0,"2026-05-12","2026-07-06"],[5,3223,"TorchaoLoraLinear: __init__() missing get_apply_tensor_subclass kwarg with torchao 0.17 (PEFT 0.19.1)","genno-whittlery",0,"Anai-Guo","#3234",0,"2026-05-11","2026-05-28"],[5,3214,"[Windows] RTX 5070 Ti (Blackwell sm_120) - setup and deployment notes","loongmiaow-pixel",2,"","",0,"2026-05-05","2026-05-06"],[5,3211,"Feature Request: Improve offline support for custom architectures in get_peft_model_state_dict","nu11secur1ty",2,"","",0,"2026-05-03","2026-05-03"],[5,3203,"Proposal: FIM-guided adaptive LoRA rank allocation (FimConfig + initialize_lora_fim_ranks)","ramkrishs",2,"","",0,"2026-04-28","2026-06-19"],[5,3201,"PrefixTuningConfig fails on google/gemma-4-e2b-it: tensor expand size mismatch in attention forward (peft 0.19","stharrold",2,"","",0,"2026-04-28","2026-05-06"],[5,3193,"Add torch.nn.Conv2d support in AdaLoRA","PARRYH0TTEr",0,"Anai-Guo","#3196",0,"2026-04-24","2026-04-30"],[5,3191,"AuxiliaryTrainingWrapper.forward requires positional `x`, breaks kwargs-only modules in modules_to_save","ved1beta",1,"ved1beta","#3199",0,"2026-04-23","2026-04-24"],[5,3186,"Proposal: add BEFT as a new PEFT method","whubaichuan",2,"","",0,"2026-04-23","2026-04-30"],[5,3184,"Proposal: support ShadowPEFT method in PEFT","SeanLee97",4,"","",0,"2026-04-22",""],[5,3182,"RFC: Improve code to resolve LoRA variants","BenjaminBossan",4,"","",0,"2026-04-21",""],[5,3169,"LoRA + BnB INT8 + CPU offload: output tensor on wrong device in tuners/lora/bnb.py","sirfyyn",2,"","",0,"2026-04-16","2026-05-25"],[5,3158,"AttributeError: module 'torch' has no attribute 'float8_e8m0fnu' when casting adapter dtype on PyTorch 2.5.1","atomfritte-lead",0,"Anai-Guo","#3161",0,"2026-04-14","2026-04-15"],[5,3142,"Proposal: add MiCA (Minor Component Adaptation) as a PEFT method","sr-networks",2,"","",0,"2026-04-09","2026-06-24"],[5,3141,"Proposal: add MiCA (Minor Component Adaptation) as a new PEFT method","rankofootball",2,"","",0,"2026-04-09","2026-04-09"],[5,3134,"Inquiry About Two-Stage QLoRA Fine-Tuning","AAnirudh07",2,"","",0,"2026-04-07","2026-04-11"],[5,3130,"Add LoRA support for Gemma 4 (Gemma4ClippableLinear)","skap3214",3,"","",0,"2026-04-02","2026-04-02"],[5,3129,"Add support for Gemma4ClippableLinear (Gemma 4 QLoRA fails)","dentity007",2,"","",0,"2026-04-02","2026-05-11"],[5,3123,"Nemotron-3-Nano NemotronHTopkRouter","lucian-student",2,"","",0,"2026-03-30","2026-05-10"],[5,3116,"Evaluation protocol uses best validation score as reported metric and inflates results","paolomandica",2,"","",0,"2026-03-25","2026-03-25"],[5,3114,"GraLoRA merge_and_unload with modules_to_save error","JBurtn",2,"","",0,"2026-03-22","2026-05-16"],[5,3110,"Discrepancy between layers names from `get_peft_model` and `.load_adapter`","AngledLuffa",2,"","",0,"2026-03-17","2026-03-18"],[5,3089,"Can LoRA ignore standard modules it doesn't know about?","AngledLuffa",2,"","",0,"2026-03-10","2026-03-16"],[5,3085,"Security: Arbitrary Module Import via Malicious Adapter Config (CWE-94)","spartan8806",0,"githubnemo","#3090",1,"2026-03-08","2026-04-16"],[5,3080,"FSDP + torch.nn.Parameter (MoE layer) lora fine-tuning doesn't work","dskhudia",2,"","",0,"2026-03-04","2026-04-20"],[5,3075,"Error When Loading Private Adapters","TyTodd",2,"","",0,"2026-03-02","2026-03-02"],[5,3073,"LoRA gradients not normalized by input norm → training instability (NaN)","warjohn",2,"","",0,"2026-03-01","2026-03-03"],[5,3049,"More efficient PISSA initialization for LoRA (faster/cheaper than current SVD)","Brokttv",2,"","",0,"2026-02-18","2026-02-26"],[5,3044,"TP support for Finetuning using LoRA and other PEFT techniques","quic-meetkuma",2,"","",0,"2026-02-16","2026-04-23"],[5,3043,"trainable_token_indices of LoraConfig not working when using more than 1 NVIDIA GPU","Oidlichtnwoada",2,"","",0,"2026-02-14","2026-03-24"],[5,3039,"extremely slow get_peft_model() with LoRA","EaswarGn",2,"","",0,"2026-02-12","2026-02-17"],[5,3034,"Evaluate augmenting driveLM data with spatial SG using off-the-shelf VLM","Hamidreza3252",2,"","",0,"2026-02-09","2026-02-12"],[5,3033,"Create Baseline Evaluation Based on DriveLM data off-the-shelf VLMs","Hamidreza3252",2,"","",0,"2026-02-09","2026-03-21"],[5,3026,"Add PSOFT to PEFT","fei407",2,"","",0,"2026-02-07","2026-03-02"],[5,3025,"Detail: removal of `.peft_config` when performing `.unload()`?","jchwenger",0,"zamal-db","e65400d",0,"2026-02-06","2026-02-27"],[5,3020,"LoRA + activation checkpointing breaks model compilation with torch.compile when dropout != 0","raliouat",2,"","",0,"2026-02-03","2026-04-14"],[5,3016,"layers_to_transform incorrectly matches MoE expert indices (should match layer index)","MichalMraz",2,"","",0,"2026-02-01","2026-03-12"],[5,3012,"Question about multiple LoRAs and FSDP","l1346792580123",2,"","",0,"2026-01-28","2026-03-17"],[5,3009,"peft failed to support LoRA for MoE models with fused expert weights in transformers v5","ITcarrot",2,"","",0,"2026-01-24","2026-01-27"],[5,3004,"Combining adapters linearly with negative weights is broken","orr-hirundo",2,"","",0,"2026-01-21","2026-04-26"],[5,3002,"Running Lora+ T5 Gemma 2 on multi gpu results in error","silviiaromanato",2,"","",0,"2026-01-19","2026-03-17"],[5,2998,"Tutorial uses deprecated dataset (financial_phrasebank) that fails to load","maerory",0,"dhruvildarji","#3058",0,"2026-01-16","2026-03-17"],[5,2993,"Unable to load the saved Lora weights","botany233",2,"","",0,"2026-01-13","2026-01-14"],[5,2991,"CoRDA initialization lacks support for Conv1D layers in older models like GPT-2","sambhavnoobcoder",1,"sambhavnoobcoder","389f588",0,"2026-01-13","2026-01-13"],[5,2988,"Evaluation code uses argmax instead of generate in prompt tuning tutorial - Prompt-based methods","maerory",1,"maerory","f9ed525",0,"2026-01-11","2026-01-14"],[5,2980,"TestOpt4bitBnb::test_lora_4bit result mismatch","jiqing-feng",1,"jiqing-feng","#2996",0,"2026-01-08","2026-01-14"],[5,2977,"Unable to use enable/disable_adapter_layers in FSDP training","varunlmxd",0,"Isalia20","#3001",1,"2026-01-07","2026-02-19"],[5,2975,"Commit 30a19a0 (gradient checkpointing) requires transformers>=4.52.0","bvolkmer",0,"BenjaminBossan","#2976",1,"2026-01-06","2026-01-08"],[5,2973,"Add BoRA to PEFT","cxmscb",2,"","",0,"2026-01-04","2026-02-11"],[5,2970,"Add EWoRA to PEFT","HarshKohli",2,"","",0,"2025-12-31","2026-04-12"],[5,2964,"Regex support for elements in target_modules list","juzhxng",2,"","",0,"2025-12-18","2025-12-19"],[5,2958,"peft assumes that torch has distributed support without checking","vladmandic",1,"vladmandic","#2963",0,"2025-12-15","2025-12-17"],[5,2957,"torch compiled model cannot have LoRA added without workaround","bghira",0,"BenjaminBossan","#2962",1,"2025-12-15","2025-12-17"],[5,2949,"Deprecation of warmup_ratio argument in transformers v5","BenjaminBossan",0,"shantanugupta2004","#2950",0,"2025-12-11","2026-02-12"],[5,2948,"Add PVeRA to PEFT","leofillioux",2,"","",0,"2025-12-10","2026-03-03"],[5,2946,"KeyError: 'token_adapter.trainable_tokens_delta.default' when saving checkpoint with trainable_token_indices a","p1kachu2233",2,"","",0,"2025-12-09","2025-12-12"],[5,2945,"Return base model state_dict with original keys","dvmazur",0,"Isalia20","#3000",1,"2025-12-09","2026-07-28"],[5,2944,"enusre_weight_tying = True does not respect the model config tie_word_embeddings flag","romitjain",2,"","",0,"2025-12-09","2026-05-09"],[5,2941,"Add MonteCLoRA (TMLR 2025 accepted) in PEFT library","victor7246",2,"","",0,"2025-12-07","2026-03-08"],[5,2940,"[Feature Request] Auto-generate 'rank_pattern' using circuit-level sensitivity analysis","songeunseo",2,"","",0,"2025-12-06","2026-01-13"],[5,2930,"[FEAT] Integrate FeRA: Frequency-Energy Constrained Routing for Effective Diffusion Adaptation Fine-Tuning","YinBo0927",2,"","",0,"2025-11-29","2026-03-15"],[5,2929,"Supported `transformers` versions are not clear","abdulfatir",2,"","",0,"2025-11-27","2026-01-05"],[5,2928,"Inference mode with Module_to_save LoRA","NguyenRichard",0,"ada-ggf25","#2931",0,"2025-11-27","2026-03-04"],[5,2927,"Add LoRA-GA (Low-Rank Adaptation with Gradient Approximation) to PEFT","sambhavnoobcoder",1,"sambhavnoobcoder","#2926",0,"2025-11-26","2026-01-15"],[5,2923,"timm: ConvNormAct","lucian-student",2,"","",0,"2025-11-20","2025-12-29"],[5,2916,"ParamWrapper initialization fails when DeepSpeed ZeRO-3 is enabled","jonnyli1125",2,"","",0,"2025-11-18","2025-12-28"],[5,2914,"Loading AWQ weight PeftMixedModel crashes","Tfloow",0,"BenjaminBossan","#2915",1,"2025-11-18","2025-11-20"],[5,2907,"Integrate: LoRA vs Full Fine-tuning: An Illusion of Equivalence","BenjaminBossan",2,"","",0,"2025-11-12","2026-01-28"],[5,2903,"Hugging face","frankjesusgonzales237-maker",2,"","",0,"2025-11-10","2025-11-10"],[5,2901,"AttributeError: 'float' object has no attribute 'meta'","Karol-G",2,"","",0,"2025-11-06","2025-11-17"],[5,2899,"Prompt tuning for VLMs like Qwen2.5VL","Martin9797",2,"","",0,"2025-11-05","2025-12-04"],[5,2898,"LORA adapter with new embedding tokens has size and load issue","imohitmayank",2,"","",0,"2025-11-05","2025-12-20"],[5,2892,"Using trl SFTTrainer creates empty adapter.safetensors file while saving when training with LoRA and Deepspeed","maximus-21",2,"","",0,"2025-11-04","2025-12-13"],[5,2889,"`autocast_adapter_dtype=False` doesn't work when the model is quantized","qgallouedec",0,"BenjaminBossan","#2893",1,"2025-11-04","2025-12-17"],[5,2888,"Potential remote code execution via untrusted tokenizer_kwargs in PromptEmbedding","Vancir",0,"BenjaminBossan","#2896",1,"2025-11-03","2025-11-04"],[5,2886,"[FEAT] Integrate QeRL (NVFP4 + LoRA + Noise scheduler) into PEFT","Aaronhuang-778",2,"","",0,"2025-11-03","2025-12-12"],[5,2885,"PeftModel.from_pretrained and merge_and_unload don't work good for me","shahelaojieraozhi",2,"","",0,"2025-11-01","2025-12-09"],[5,2884,"[Question/Bug] How to safely continue LoRA fine-tuning under DeepSpeed ZeRO-3 (multi-stage training with modul","XiangZhang-zx",2,"","",0,"2025-10-31","2025-12-09"],[5,2882,"[FEAT] Integrate LoRA-One into PEFT","YuanheZ",2,"","",0,"2025-10-30","2026-03-16"],[5,2881,"Bug in prefix tuning of Qwen3-0.6B: group-query attention fix in #1901 still cause error","papaya0481",0,"BenjaminBossan","#2883",1,"2025-10-30","2025-11-05"],[5,2880,"Bug in fine-tuning for Qwen2.5-VL with LoRA","Bytes-Lin",2,"","",0,"2025-10-30","2025-12-08"],[5,2878,"peft \" target_modules='all-linear' \" have different behavior between x86 and aarch ?","HuangChiEn",2,"","",0,"2025-10-29","2025-12-07"],[5,2877,"peft config 'all-linear' include lm_head, is there anyway to remove it ?","HuangChiEn",2,"","",0,"2025-10-29","2025-10-29"],[5,2874,"AttributeError: 'NoneType' object has no attribute 'shape'","tanxiaoke6",2,"","",0,"2025-10-27","2025-10-27"],[5,2873,"Can I use Lora fine-tuning twice?","tohokulgq",2,"","",0,"2025-10-27","2025-12-05"],[5,2864,"Deal with weight tying consistently","BenjaminBossan",0,"sambhavnoobcoder","#2870",0,"2025-10-23","2026-04-28"],[5,2861,"[FEAT] Integrate WoRA (Filtering-WoRA) into PEFT","JT-Sun",2,"","",0,"2025-10-23","2026-03-02"],[5,2855,"When using the ZeRO3 configuration of deepspeed, the target_parameters cannot obtain the shape of the paramete","qiu-xiao-0330",2,"","",0,"2025-10-21","2025-12-01"],[5,2840,"Default set inference_mode=True For saved checkpoint","dengchengxifrank",2,"","",0,"2025-10-15","2025-10-15"],[5,2838,"Embedding scaling for TrainableTokensModel in DoRA","sambhavnoobcoder",2,"","",0,"2025-10-14","2025-10-15"],[5,2835,"Update deprecated torch_dtype argument","BenjaminBossan",0,"shantanugupta2004","#2837",0,"2025-10-14","2025-10-16"],[5,2834,"FSDP QLORA doesn't work with multiple adapters","ojh31",2,"","",0,"2025-10-14","2025-11-30"],[5,2833,"QLORA FSDP doesn't work with 8bit","ojh31",2,"","",0,"2025-10-14","2025-10-21"],[5,2832,"Gradient checkpoint with multiple adapters","NguyenRichard",2,"","",0,"2025-10-14","2025-10-15"],[5,2830,"Embedding scaling for TrainableTokensModel in XLoraEmbeddingLayer","sambhavnoobcoder",2,"","",0,"2025-10-13","2025-10-14"],[5,2827,"ImportError: `PytorchGELUTanh` Removed in Transformers 4.57.0 but Still Required by peft","Sunhill666",2,"","",0,"2025-10-10","2025-10-10"],[5,2826,"Gradient checkpointing and Activated LoRA models","kgreenewald",0,"githubnemo","30a19a0",1,"2025-10-09","2025-11-03"],[5,2822,"Qwen2.5 VL加载checkpoint继续训练报错","Yimi81",2,"","",0,"2025-10-09","2025-10-09"],[5,2821,"PrefixTuning fails with DynamicCache: 'DynamicCache' object has no attribute 'key_cache'","dnanper",2,"","",0,"2025-10-09","2025-10-15"],[5,2816,"PromptEmbedding's random initialization leads to out-of-manifold embeddings that doesn't learn","macmacmacmac",2,"","",0,"2025-10-07","2025-10-09"],[5,2813,"[FEAT] Integrate LoRETTA: Low-Rank Economic Tensor-Train Adaptation for Ultra-Low-Parameter Fine-Tuning of Lar","mbaddar1",2,"","",0,"2025-10-06","2026-02-12"],[5,2809,"TrainableTokensModel doesn't handle embedding scaling","kimihailv",0,"sambhavnoobcoder","#2825",0,"2025-10-01","2025-10-14"],[5,2802,"Guide on training that requires both LoRA and base model forward calls ?","thangld201",2,"","",0,"2025-09-27","2025-10-15"],[5,2798,"Add stricter type checking in LoraConfig for support with HfArgumentParser","romitjain",2,"","",0,"2025-09-23","2025-09-23"],[5,2796,"Support negative weights when merging LoRA adapters","valteu",0,"sambhavnoobcoder","#2811",0,"2025-09-22","2025-10-09"],[5,2795,"Add Support for LoRA-XS","RonanKMcGovern",2,"","",0,"2025-09-22","2025-10-30"],[5,2789,"Tests fail to run: https://github.com/huggingface/peft","yurivict",2,"","",0,"2025-09-18","2025-10-27"],[5,2787,"PeftModel.from_pretrained adapter not applied when using device=\"cuda:1\"","Xxxxsir",2,"","",0,"2025-09-18","2025-10-26"],[5,2786,"Two issues with [X-LoRA] Implementation: Scalings Logging and Top-K Softmax Normalization","Che-Xu",1,"Che-Xu","#2793",0,"2025-09-17","2025-10-09"],[5,2783,"PEFT Loha/Lokr crashes for SDXL pipeline inference","frutiemax92",0,"BenjaminBossan","#2784",1,"2025-09-13","2025-09-25"],[5,2777,"Update the model config to tie_word_embeddings = False in case of lm_head or embedding layer update","romitjain",2,"","",0,"2025-09-08","2025-10-29"],[5,2776,"maximum recursion depth exceeded","misonsky",2,"","",0,"2025-09-08","2025-10-16"],[5,2772,"\"UserWarning: Found missing adapter keys while loading the checkpoint\" when module name contains \"weight\"","mauricebrenner",0,"BenjaminBossan","#2773",1,"2025-09-05","2025-09-30"],[5,2769,"prepare_inputs_for_generation missing when using newer transformers version","Hrovatin",2,"","",0,"2025-09-04","2025-10-13"],[5,2766,"[X-LoRA] RuntimeError: Tensor Size Mismatch with Vision-Language Models due to Sequence Length Inference","seongbin1231",2,"","",0,"2025-09-03","2025-10-12"],[5,2764,"merge_and_unload returns the base (prior to fine-tuning) back!!!!","manitadayon",2,"","",0,"2025-09-01","2025-10-09"],[5,2760,"Integrate DeLoRA","mwbini",2,"","",0,"2025-08-28","2025-10-28"],[5,2759,"PeftModel trainable parameters with multiple adapters","NguyenRichard",2,"","",0,"2025-08-28","2025-10-06"],[5,2757,"Remove usage of from_legacy_cache","BenjaminBossan",2,"","",0,"2025-08-28","2025-09-29"],[5,2756,"Loading multiple adapters with PeftMixedModel","NguyenRichard",2,"","",0,"2025-08-27","2025-08-28"],[5,2755,"EfficientQAT","gapsong",2,"","",0,"2025-08-27","2025-09-01"],[5,2754,"No working peft version available in PyPI for transformers 4.55+ version","qGentry",2,"","",0,"2025-08-27","2025-08-28"],[5,2751,"Number of trainable params is displayed wrong in accelerate logs when using FSDP","Ali-Sayed-Salehi",2,"","",0,"2025-08-25","2025-08-26"],[5,2749,"Set multiple adapters actively when training","liaoyy-hub",2,"","",0,"2025-08-21","2025-09-29"],[5,2747,"Add support for DINO v3","Six-Bit-TX",2,"","",0,"2025-08-21","2025-08-21"],[5,2746,"Gemma 2/3 Attention: Expected a single attention mask, got 2 instead","michelleezhang",2,"","",0,"2025-08-20","2025-08-27"],[5,2745,"Support QLoRA Training on AWQ Quantized Models","yhhtc1201",2,"","",0,"2025-08-19","2025-10-13"],[5,2742,"RuntimeError: element 0 of tensors does not require grad and does not have a grad_fn","Mishajain1110",2,"","",0,"2025-08-15","2025-09-23"],[5,2738,"Which base model weights are getting frozen after applying LoRA?","srbh-dl",2,"","",0,"2025-08-13","2025-08-14"],[5,2733,"LoraLayer.set_scale not working correctly with rslora","dongmingli-Ben",2,"","",0,"2025-08-08","2025-09-16"],[5,2716,"Lora_bias: NoneType object has no attribute data","Fhrozen",2,"","",0,"2025-08-06","2025-08-06"],[5,2699,"UserWarning: Found missing adapter keys while loading the checkpoint","manitadayon",2,"","",0,"2025-08-02","2025-11-09"],[5,2695,"TST: Some SHiRA tests are failing when run with parallelism","BenjaminBossan",1,"BenjaminBossan","#2736",1,"2025-08-01","2025-08-19"],[5,2690,"failed to run kaggle notebook","MengAiDev",2,"","",0,"2025-08-01","2025-08-02"],[5,2688,"[DOC] OFT and BOFT descriptions are the same","AmirBachir",2,"","",0,"2025-07-31","2025-09-02"],[5,2683,"Add Fast-Forward to PEFT","AdirRahamim",2,"","",0,"2025-07-30","2025-09-08"],[5,2673,"ModuleNotFoundError: No module named 'peft'","SAOHEIUD",2,"","",0,"2025-07-27","2025-09-19"],[5,2672,"Support for RoAD: 2D Rotary Adaptation for Efficient Finetuning, Efficient Batching and Composability","ppetrushkov",2,"","",0,"2025-07-25","2025-08-25"],[5,2671,"QLora broken - bnb.py resolve_variant should accept QALoRA params for compatibility with LoraLayer","gapsong",2,"","",0,"2025-07-25","2025-07-25"],[5,2663,"[Bug] PiSSA example script incompatible with current trl dependency","gapsong",2,"","",0,"2025-07-22","2025-07-22"],[5,2660,"Custom models LoRA","stillbetter",2,"","",0,"2025-07-21","2025-07-24"],[5,2658,"check_target_module_exists doesn't accept modules with less than 2 nesting","ShirAmir",2,"","",0,"2025-07-20","2025-10-30"],[5,2656,"Fix BibTeX","cx-alberto-simoes",1,"cx-alberto-simoes","#2659",0,"2025-07-18","2025-07-21"],[5,2655,"Request: Enable LoRA support for additional/custom layers beyond Linear, Conv, and Embedding","asrays",2,"","",0,"2025-07-18","2025-08-25"],[5,2654,"Implementation of `Bottleneck`, or known as `Adapter`","Zuo-Lihan",2,"","",0,"2025-07-18","2025-08-25"],[5,2653,"trainable_token_indices causes training to fail","mark-nlp",2,"","",0,"2025-07-17","2025-09-22"],[5,2652,"Add SingLoRA to PEFT","DavidBCode",2,"","",0,"2025-07-17","2025-09-01"],[5,2651,"DoRA slow forward inference","phemw",0,"BenjaminBossan","#2661",1,"2025-07-17","2026-01-02"],[5,2650,"AttributeError: 'DTensor' object has no attribute 'compress_statistics'","vsoesanto",2,"","",0,"2025-07-16","2025-08-27"],[5,2648,"Feature Request: Add Adaptive Singular Value Decomposition based Orthogonal Subspace Fine-Tuning","NikhilNayak-debug",2,"","",0,"2025-07-15","2026-01-15"],[5,2647,"How can I merge the original model weights with LoRA weights?","guoguo1314",2,"","",0,"2025-07-15","2025-08-23"],[5,2642,"Prompt_Tuning.ipynb example doesn't seem to train the model","ruixing76",2,"","",0,"2025-07-11","2025-08-23"],[5,2640,"Why peft.utils.other.fsdp_auto_wrap_policy do not warp the module do not require grad?","Changlin-Lee",2,"","",0,"2025-07-10","2025-08-18"],[5,2639,"Multi-Lora training","liumc14",2,"","",0,"2025-07-10","2025-07-14"],[5,2636,"Support for GraLoRA","yeonjoon-jung01",2,"","",0,"2025-07-09","2025-08-17"],[5,2634,"Auto-tagging does not support customized peft type","criyle",0,"githubnemo","35000fd",1,"2025-07-06","2025-07-08"],[5,2627,"Some Gemma generate tests fail on GPU with: torch._dynamo.exc.Unsupported: Data-dependent branching","BenjaminBossan",2,"","",0,"2025-07-02","2025-08-18"],[5,2615,"How can I fine-tune the linear layers of the LLM part in Qwen2.5_VL 3B?","guoguo1314",2,"","",0,"2025-06-26","2025-07-18"],[5,2613,"Make Task Type validation case insensitive","cartesian-plane",2,"","",0,"2025-06-25","2025-07-22"],[5,2608,"support lora weights conversion","garychan22",2,"","",0,"2025-06-24","2025-08-01"],[5,2607,"QLoRA tuning support for compressed-tensors","kmehant",2,"","",0,"2025-06-24","2025-06-26"],[5,2603,"lora with trainable_token_indices do NOT support zero3？","LinB203",0,"BenjaminBossan","#2605",1,"2025-06-22","2025-06-26"],[5,2601,"How to Load Adapters with Per-Layer Variable Shapes in `PeftModel.from_pretrained`","yuxuan-z19",2,"","",0,"2025-06-20","2025-06-21"],[5,2595,"LoRA combined with Prefix tuning","aparaselli",2,"","",0,"2025-06-18","2025-07-28"],[5,2592,"diffusion components have no 'add_weighted_adapter'","garychan22",2,"","",0,"2025-06-17","2025-06-23"],[5,2589,"Clarification on embed tokens when using LoRA","jiosephlee",2,"","",0,"2025-06-13","2025-07-17"],[5,2587,"Support for ABBA: Highly Expressive Hadamard Product Adaptation for LLMs","RaghavSinghal10",2,"","",0,"2025-06-12","2025-07-26"],[5,2586,"Merging QLoRA weights to Qwen-2.5-VL-7B-Instruct does not change base model weights","harshit2997",2,"","",0,"2025-06-12","2025-06-12"],[5,2585,"Sparse High Rank Adapters (SHiRA) Support","kkb-code",2,"","",0,"2025-06-11","2025-07-15"],[5,2583,"Lora transfer learning","hardikdava",2,"","",0,"2025-06-11","2025-07-20"],[5,2566,"Found missing adapter keys while loading the checkpoint","JJplane",2,"","",0,"2025-06-03","2025-06-04"],[5,2563,"Integrate Lily","yibozhong",2,"","",0,"2025-06-02","2026-03-04"],[6,3117,"DOC: clarify Arize AX and Phoenix guidance in the tracing tutorial","arizedatngo",4,"","",0,"2026-11-05",""],[6,3106,"parse_code_blobs rejects markdown code fences without a newline before the closing fence","shadowmodder",4,"","",0,"2026-11-02",""],[6,3104,"BUG: Windows: test_import uses bin/python instead of Scripts/python.exe","NavyaVaddi",4,"","",0,"2026-11-02",""],[6,3101,"BUG: Agent and tool publishing use inconsistent metadata overwrite semantics for existing Hub Spaces","shreysherikar",4,"","",0,"2026-11-01",""],[6,3097,"BUG: DuckDuckGo parser can pair a result with the previous result’s URL","Cmz9111",4,"","",0,"2026-11-01",""],[6,3096,"Question: opt-in LLMSafe static compatibility review for smolagents CodeAgent?","rezerpaul-crypto",4,"","",0,"2026-11-01",""],[6,3089,"BUG: LocalPythonExecutor ignores print(file=...): output goes to the agent logs and the buffer stays empty","sunhouy",4,"","",0,"2026-10-29",""],[6,3088,"DOC: Add Future AGI guidance to Inspecting runs with OpenTelemetry","nik13",4,"","",0,"2026-10-29",""],[6,3074,"BUG: Tool.save()/to_dict() fails for Tool classes with @property or @classmethod members","anh-nguyenn",4,"","",0,"2026-10-26",""],[6,3071,"BUG: ChatMessage.from_dict() mutates its input dict, so converting the same dict twice raises TypeError","Sonica-B",4,"","",0,"2026-10-25",""],[6,3062,"BUG: LocalPythonExecutor defers the outer iterable of generator expressions","BichengWang",4,"","",0,"2026-10-23",""],[6,3060,"BUG: LocalPythonExecutor returns the last statement value from functions without an explicit return","OrdoAbChao7",4,"","",0,"2026-10-23",""],[6,3049,"Pre-execution security guardrails for your agent execution loops","Jayanthpulisheri",4,"","",0,"2026-10-22",""],[6,3048,"BUG: LocalPythonExecutor class methods overwrite global functions of the same name","mengchen-atx",4,"","",0,"2026-10-22",""],[6,3047,"BUG: `import pkg.mod` in LocalPythonExecutor does not bind `pkg`","mengchen-atx",4,"","",0,"2026-10-22",""],[6,3045,"BUG: Docker/Modal/Blaxel executors let a later `execute_result` overwrite the final answer","yifanxiong272",4,"","",0,"2026-10-21",""],[6,3043,"BUG: LocalPythonExecutor silently ignores print's sep and end arguments","BichengWang",4,"","",0,"2026-10-21",""],[6,3040,"BUG: `AmazonBedrockModel` ignores `stop_sequences` (never sent as `inferenceConfig.stopSequences`)","yifanxiong272",4,"","",0,"2026-10-21",""],[6,3039,"BUG: `agent.step()` returns an `ActionOutput` wrapper instead of the final answer (or `None`)","yifanxiong272",4,"","",0,"2026-10-21",""],[6,3037,"`Retrying` ignores `Retry-After`, and the first retry already waits 2–3× `wait_seconds`","simin-yuan",4,"","",0,"2026-10-20",""],[6,3034,"OpenAIModel: client retries and ApiModel retries multiply (9 requests per rate limit)","simin-yuan",4,"","",0,"2026-10-20",""],[6,3026,"BUG: LocalPythonExecutor silently overwrites duplicate keywords during call unpacking","BichengWang",4,"","",0,"2026-10-19",""],[6,3025,"BUG:  _get_json_schema_type never returns \"image\"/\"audio\": @tool media annotations become \"object\", and \"image","Oussama-zbir",4,"","",0,"2026-10-19",""],[6,3023,"DOC: Document the Hyperlight micro-VM executor","danbugs",4,"","",0,"2026-10-19",""],[6,3022,"ENH: Add agent-sandbox (Kubernetes) as a remote code execution option","aleks-stefanovic",4,"","",0,"2026-10-19",""],[6,3021,"DOC: inspect_runs has no example for a generic OTLP backend","rogeriochaves",4,"","",0,"2026-10-19",""],[6,3019,"BUG: fix_final_answer_code rewrites \"final_answer\" inside string literals and keyword arguments","notlakshitha",4,"","",0,"2026-10-18",""],[6,3017,"BUG: stream_outputs=True re-parses the whole message as Markdown on every delta, quadratic stall on long tool ","anzal1",4,"","",0,"2026-10-18",""],[6,3015,"BUG: ToolCallingAgent drops a rejected tool call from memory, so the next prompt shows an empty assistant turn","yifanxiong272",4,"","",0,"2026-10-18",""],[6,3014,"BUG: LocalPythonExecutor silently resolves an undefined name to a similarly named variable instead of raising","yifanxiong272",4,"","",0,"2026-10-18",""],[6,3012,"BUG: ThreadPoolExecutor `with` block calls shutdown(wait=True) on exit","CaptianErek",4,"","",0,"2026-10-17",""],[6,3010,"BUG: open_deep_research YouTubeConverter silently drops transcripts (get_transcript removed in youtube-transcr","pushkarsingh32",4,"","",0,"2026-10-17",""],[6,3007,"BUG: detailed memory replay crashes when model input contains an image","AryanRG",4,"","",0,"2026-10-16",""],[6,3005,"ENH: get_clean_message_list deep-copies the whole history every step (agent loop 3-4x slower on long runs)","deadstrobe5",4,"","",0,"2026-10-16",""],[6,2995,"ENH: Add a first-class NvidiaModel for NVIDIA NIM","freshyjmp",4,"","",0,"2026-10-15",""],[6,2992,"BUG: LocalPythonExecutor gives each comprehension iteration its own scope, so closures capture per-iteration v","kevin9327",4,"","",0,"2026-10-15",""],[6,2988,"DOC: Add an example for using Cheaper Inference with OpenAIModel","aiapienthusiast",4,"","",0,"2026-10-14",""],[6,2987,"Docs: link an external Lizard Sandboxes executor","yura-build",4,"","",0,"2026-10-14",""],[6,2983,"BUG: SafeSerializer cannot serialize complex NumPy arrays","akkupratap323",4,"","",0,"2026-10-13",""],[6,2980,"Proposal: optional Plazmi pulse tool example with tested domain-failure behavior","cairn-agent",4,"","",0,"2026-10-13",""],[6,2978,"BUG: from_folder silently discards an explicit managed_agents override","yifanxiong272",4,"","",0,"2026-10-13",""],[6,2976,"BUG: Tool.from_gradio raises KeyError for ordinary Python parameter annotations","yifanxiong272",4,"","",0,"2026-10-12",""],[6,2975,"ENH: financial-research tutorial with a separate Adanos Hub tool","alexander-schneider",4,"","",0,"2026-10-12",""],[6,2973,"BUG: @tool rejects anyOf input schemas produced by its own annotation parser","yifanxiong272",4,"","",0,"2026-10-12",""],[6,2969,"BUG: DuckDuckGo search parser drops title text before nested HTML tags","yifanxiong272",4,"","",0,"2026-10-12",""],[6,2967,"ApiModel RateLimiter is not thread-safe: concurrent calls can bypass requests_per_minute","qiyuhuating",4,"","",0,"2026-10-12",""],[6,2966,"BUG: Class-body assignments ignore earlier class-local bindings","yifanxiong272",4,"","",0,"2026-10-12",""],[6,2963,"BUG: Unary plus preserves bool operands instead of applying Python numeric semantics","yifanxiong272",4,"","",0,"2026-10-12",""],[6,2961,"BUG: SafeSerializer rejects NumPy string and complex scalar values","AdityaBiranje",2,"","",0,"2026-10-12","2026-10-13"],[6,2959,"BUG: text browser keeps find-on-page state after visiting a new page","porkedchop",4,"","",0,"2026-10-12",""],[6,2954,"Document DemonRoute as an OpenAI-compatible backend for open_deep_research","demonroute1",4,"","",0,"2026-10-11",""],[6,2951,"BUG: ToolCallingAgent calls a tool with unparseable tool-call arguments as a raw string","CyberNativeAI",4,"","",0,"2026-10-10",""],[6,2944,"BUG: Tool.from_gradio returns a wrapper missing is_initialized","yifanxiong272",4,"","",0,"2026-10-10",""],[6,2943,"BUG: ToolCallingAgent rejects typed image and audio state arguments before sanitization","yifanxiong272",4,"","",0,"2026-10-10",""],[6,2942,"BUG: Tool.to_dict rejects valid tuple aliases in with statements as undefined names","yifanxiong272",4,"","",0,"2026-10-10",""],[6,2941,"BUG: LocalPythonExecutor loses print output when print is passed as a callback","yifanxiong272",4,"","",0,"2026-10-10",""],[6,2940,"BUG: LocalPythonExecutor silently ignores tuple and subscript assignments in class bodies","yifanxiong272",4,"","",0,"2026-10-10",""],[6,2939,"BUG: ToolCallingAgent finalization changes audio sample rate and returns AgentText","yifanxiong272",4,"","",0,"2026-10-10",""],[6,2938,"README LiteLLM example is missing an os import","hotcholula",4,"","",0,"2026-10-10",""],[6,2930,"BUG: from_dict(..., model=...) constructs the saved model before applying the override","yifanxiong272",4,"","",0,"2026-10-09",""],[6,2928,"BUG: Tool.from_hub forwards Hub download options into the tool constructor","yifanxiong272",4,"","",0,"2026-10-09",""],[6,2927,"BUG: SafeSerializer.get_safe_serializer_code() returns Python that fails to compile","yifanxiong272",4,"","",0,"2026-10-09",""],[6,2926,"BUG: @tool output_schema produces invalid indentation in Tool.to_dict() exported code","yifanxiong272",4,"","",0,"2026-10-09",""],[6,2924,"ENH: Add --max-steps option to the smolagent CLI","BichengWang",4,"","",0,"2026-10-08",""],[6,2923,"BUG: Stop sequences sent to the provider cut off reasoning models mid-thought, leaving empty content","sandro-nanzen",4,"","",0,"2026-10-08",""],[6,2920,"BUG: chained comparison evaluates later operand after false-like result","skyon0522-ai",4,"","",0,"2026-10-08",""],[6,2918,"DOC: OpenAIModel with a local OpenAI-compatible server on Apple Silicon (Rapid-MLX)","raullenchai",4,"","",0,"2026-10-07",""],[6,2910,"Document DemonRoute as an OpenAI-compatible backend for open_deep_research","haterfactory",4,"","",0,"2026-10-06",""],[6,2907,"DOC: zh guided tour still wraps sub-agents in the removed `ManagedAgent` class","MaQun12-ai",4,"","",0,"2026-10-06",""],[6,2904,"local_python_executor: except clauses never match wrapped built-in exceptions (KeyError/IndexError)","TINGyu123644",4,"","",0,"2026-10-06",""],[6,2903,"BUG: bare raise in an except block raises InterpreterError instead of re-raising","gauravsinha12",4,"","",0,"2026-10-05",""],[6,2894,"LocalPythonExecutor incorrectly truth-tests the final operand of and/or","nkjapan",4,"","",0,"2026-10-04",""],[6,2893,"LocalPythonExecutor: except KeyError / except IndexError don't catch failed lookups","jayzuccarelli",4,"","",0,"2026-10-04",""],[6,2891,"Opt-in CI check for writes committed before result publication","astrogilda",4,"","",0,"2026-10-03",""],[6,2890,"BUG: ","marekkokosz-jpg",4,"","",0,"2026-10-03",""],[6,2888,"BUG: NumPy arrays with non-JSON elements bypass the enabled pickle fallback","mikamikasuki",4,"","",0,"2026-10-03",""],[6,2885,"BUG: Tools colliding with base tool names are silently overwritten when add_base_tools=True","effson",4,"","",0,"2026-10-02",""],[6,2884,"ENH: Semaprax effect-policy example for CodeAgent tool calls","wsdt",4,"","",0,"2026-10-02",""],[6,2882,"DOC: zh/hi RAG examples still instruct readers to use the removed HfEngine class","Andiii208",4,"","",0,"2026-10-02",""],[6,2881,"Example proposal: bounded MCP ToolCollection for large tool catalogs","JDeun",4,"","",0,"2026-10-02",""],[6,2879,"Bug: _coerce_tool_call crashes on custom OpenAI-style objects or minimal dicts missing id/type","chauvuusvn",4,"","",0,"2026-10-01",""],[6,2877,"BUG: LocalPythonExecutor does not support the `@` matrix multiplication operator (`@` and `@=`)","zayddawood-0",4,"","",0,"2026-10-01",""],[6,2874,"BUG: LiteLLMRouterModel overrides every deployment's api_base and api_key with None","lets-order-some-fries",4,"","",0,"2026-10-01",""],[6,2873,"BUG: Running the same CodeAgent more than once with a remote python executor breaks","robin-mundra",4,"","",0,"2026-10-01",""],[6,2870,"BUG: SafeSerializer loses dimensions after an empty NumPy array axis","hulkbig",4,"","",0,"2026-10-01",""],[6,2865,"Small MemCode save/recall example for smolagents","vivekgupta-memcode",4,"","",0,"2026-09-29",""],[6,2863,"BUG: LocalPythonExecutor evaluates subscript index before container expression","ZhaoYJ9",4,"","",0,"2026-09-29",""],[6,2861,"BUG: get_json_schema crashes on non-string docstring choices","Afloat16",4,"","",0,"2026-09-28",""],[6,2858,"BUG: LocalPythonExecutor leaves earlier context entered when a later __enter__ fails","zjyHeart",4,"","",0,"2026-09-27",""],[6,2857,"Add optional ScreenContextAgent tool example","ikeikeikeda66",4,"","",0,"2026-09-27",""],[6,2856,"EvalPort: portable interchange for smolagents_benchmark results","adhabnr-ux",4,"","",0,"2026-09-27",""],[6,2847,"SafeSerializer corrupts/crashes on a plain dict whose own `__type__` key collides with an internal type tag","VANDRANKI",4,"","",0,"2026-09-25",""],[6,2846,"BUG: LocalPythonExecutor: augmented assignment to an undefined name starts from 0 instead of raising NameError","jayzuccarelli",4,"","",0,"2026-09-25",""],[6,2839,"ActionStep.to_messages() crashes on list model_output (breaks multi-step ToolCallingAgent)","michaelkillgta",4,"","",0,"2026-09-25",""],[6,2838,"CallbackRegistry TypeError for callbacks defined as def cb(step, *args)","michaelkillgta",4,"","",0,"2026-09-25",""],[6,2837,"AgentImage subclasses PIL.Image.Image but exposes an empty image (.size==(0,0))","michaelkillgta",4,"","",0,"2026-09-25",""],[6,2836,"LocalPythonExecutor does not expand starred expressions in list/tuple/set literals","michaelkillgta",4,"","",0,"2026-09-25",""],[6,2834,"Feature request: per-call payment wrapper for paid tool/API calls (feeless x402 rail)","dhyabi2",4,"","",0,"2026-09-25",""],[6,2833,"BUG: Tool.to_dict() blanket replace of the tool name rewrites occurrences inside the saved forward body, silen","BlueX888",4,"","",0,"2026-09-25",""],[6,2831,"[Closed]","chauvuusvn",3,"","",0,"2026-09-24","2026-09-26"],[6,2824,"DOC: propose a multi-tool web research workflow using MCP","snorfyang",4,"","",0,"2026-09-22",""],[6,2821,"`AgentAudio(value, samplerate=...)` raises TypeError because `str.__new__` receives the extra argument","sharziki",4,"","",0,"2026-09-21",""],[6,2820,"`AgentLogger.log_rule()` ignores its `level` argument, so `replay()` loses its step separators","sharziki",4,"","",0,"2026-09-21",""],[6,2818,"Tool / code-exec failure taxonomy vs model errors","AgentEtna-Founder",4,"","",0,"2026-09-19",""],[6,2812,"BUG: tool validation drops the enclosing function's parameters when the body has a lambda or a nested def","Lesereingrape",4,"","",0,"2026-09-18",""],[6,2805,"DOC: Guided-tour tools links open the wrong tutorial","GDemay",3,"","",0,"2026-09-17","2026-09-17"],[6,2800,"DOC: zh/examples/multiagents.md imports the removed `ManagedAgent` symbol","zhi-code",2,"","",0,"2026-09-15","2026-09-19"],[6,2799,"BUG: Model.to_dict() drops custom_role_conversions and client_kwargs connection settings (api_base/organizatio","BlueX888",4,"","",0,"2026-09-15",""],[6,2796,"CSOAI — Smolagents interop: lightweight governance","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[6,2795,"CSOAI — Smolagents interop: lightweight agent governance","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[6,2794,"CSOAI — Smolagents AI governance lightweight","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[6,2784,"DOC:","Justee21",4,"","",0,"2026-09-10",""],[6,2782,"Local Python executor ignores loop else clauses (for...else / while...else)","AnishPatel526",4,"","",0,"2026-09-10",""],[6,2780,"BUG: MLXModel can truncate at a later stop sequence when multiple stops match.","MMaazesen",4,"","",0,"2026-09-09",""],[6,2775,"[Bug]: Tool.from_space raises IndexError when a space returns a single-element output","Beastmode1470",4,"","",0,"2026-09-08",""],[6,2771,"BUG: LocalPythonExecutor silently ignores decorators, so a decorated function returns the undecorated result","Sreekant13",4,"","",0,"2026-09-08",""],[6,2762,"Tool.from_langchain maps every positional arg onto the first input key","Oskii",4,"","",0,"2026-09-07",""],[6,2758,"Normal return inside with rolls back successful SQLite transactions","feng1201",4,"","",0,"2026-09-06",""],[6,2754,"BUG: LocalPythonExecutor binds function arguments incorrectly (`*args`, `**kwargs`, positional-only, keyword-o","ShayanMuhammad-CS",4,"","",0,"2026-09-05",""],[6,2750,"Security: Harden VisitWebpageTool against SSRF attacks","Bristlecone2026",4,"","",0,"2026-09-04",""],[6,2746,"LocalPythonExecutor: for/while...else silently drops the else clause; list += <non-list iterable> incorrectly ","VANDRANKI",4,"","",0,"2026-09-04",""],[6,2743,"webagent CLI crashes on a plain install: helium and selenium are only in the vision extra","vanshtaneja23",4,"","",0,"2026-09-04",""],[6,2736,"BUG: Tool schema validation is disabled under `python -O`","moonlly233-star",4,"","",0,"2026-09-03",""],[6,2726,"cli.py interactive_mode() offers 'OpenAIServerModel' but load_model() only recognizes 'OpenAIModel', guarantee","VANDRANKI",4,"","",0,"2026-09-01",""],[6,2725,"ActionStep.dict() returns raw non-JSON-serializable bytes for observations_images","VANDRANKI",4,"","",0,"2026-09-01",""],[6,2722,"Make code executors pluggable via entry points","albertvillanova",4,"","",0,"2026-09-01",""],[6,2720,"BUG: _generate_planning_step embeds a structured model response's Python repr into the plan text","revanth-045",4,"","",0,"2026-09-01",""],[6,2712,"[Integration / Feature] Step-Level WAL Durability Callback for Crash-Resilient Agents (LetItLoop)","sdageltc",4,"","",0,"2026-08-30",""],[6,2704,"BUG: _run_stream yields inside `finally`, so closing the stream raises \"generator ignored GeneratorExit\"","AlexLiu190625",3,"","",0,"2026-08-28","2026-08-28"],[6,2703,"BUG: _run_stream yields inside `finally`, so closing the stream raises \"generator ignored GeneratorExit\"","AlexLiu190625",4,"","",0,"2026-08-28",""],[6,2700,"ENH: Add missing return type annotations to utility methods","Samearth17",4,"","",0,"2026-08-27",""],[6,2699,"[Agent Test] Issue management tool test","the-junaid-bashir",4,"","",0,"2026-08-27",""],[6,2690,"Add regression coverage for CodeAgent with no configured tools","Whxuan0701",4,"","",0,"2026-08-24",""],[6,2689,"Document custom Tool exception handling and recovery","Whxuan0701",4,"","",0,"2026-08-24",""],[6,2686,"final_answer_checks are skipped for the max-steps fallback answer","oroikono",4,"","",0,"2026-08-23",""],[6,2683,"BUG: ToolCallingAgent silently drops tool calls with missing or duplicate IDs","crown-sports",4,"","",0,"2026-08-23",""],[6,2674,"ModalExecutor._wait_for_server can busy-loop forever on a non-200, non-exception response","VANDRANKI",4,"","",0,"2026-08-21",""],[6,2670,"Add BilibiliHotTool for fetching trending Chinese video content","Tree005",4,"","",0,"2026-08-21",""],[6,2668,"[BUG] validate_tool_attributes never checks keyword-only or positional-only __init__ parameters","VANDRANKI",4,"","",0,"2026-08-20",""],[6,2657,"[BUG] AgentImage rejects numpy arrays when torch is not installed","VANDRANKI",4,"","",0,"2026-08-18",""],[6,2656,"[BUG] AgentImage inverts pixel values when built from a tensor (255 - array * 255)","VANDRANKI",4,"","",0,"2026-08-18",""],[6,2654,"ENH: [Feature] Governance middleware hooks for tool/code execution (PII blocking, cost caps, tool authorizatio","nagasatish007",4,"","",0,"2026-08-18",""],[6,2615,"BUG: Agent serialization drops documented runtime configuration","Gaurav890",4,"","",0,"2026-08-11",""],[6,2608,"Nested `Any` leaks `{\"type\": \"any\"}` into tool schemas: get_tool_json_schema sanitises only the top level","percymcn",4,"","",0,"2026-08-10",""],[6,2593,"BUG: Misleading 401 Unauthorized from smolagents despite valid Hugging Face authentication","CheefColt",2,"","",0,"2026-08-01","2026-08-14"],[6,2586,"Retrying's default wait_seconds=0.0 disables backoff entirely, and reraise has no effect","ErenAta16",4,"","",0,"2026-07-29",""],[6,2584,"BUG: default InferenceClient model Qwen/Qwen3-Next-80B-A3B-Thinking error","zzz6519003",4,"","",0,"2026-07-29",""],[6,2577,"GradioUI crashes on structured `model_output`: `_clean_model_output` assumes a string","ErenAta16",4,"","",0,"2026-07-28",""],[6,2576,"BUG: MCP doesn't work with `mcp >= 2.0`","davidmezzetti",4,"","",0,"2026-07-28",""],[6,2571,"ENH: Add ConceptDiffTool for 85% Token Reduction on Web Scrapes","vishal-raaj-dnd",4,"","",0,"2026-07-26",""],[6,2569,"MCPClient.get_tools returns stale tools after disconnect","KXHXK",4,"","",0,"2026-07-25",""],[6,2566,"Input-token cost grows ~O(n²) with step count — full memory is replayed every action step (docs/mitigation)","wartzar-bee",4,"","",0,"2026-07-23",""],[6,2557,"Feature request: a supported pre-execution tool-call hook (no equivalent to CrewAI's before_tool_call / LangCh","nzdsf2-gif",4,"","",0,"2026-07-21",""],[6,2555,"BUG: local executor rejects starred and iterable unpacking assignments (a, *b = ...)","himanshu748",4,"","",0,"2026-07-21",""],[6,2552,"BUG: dict unpacking in dict literals fails with misleading 'NoneType is not supported'","himanshu748",4,"","",0,"2026-07-20",""],[6,2513,"ENH: Make space_sdk configurable in CodeAgent.push_to_hub()","osmangulercan",4,"","",0,"2026-07-13",""],[6,2510,"BUG: planning updates render a literal remaining_steps placeholder and ignore run overrides","tlysanhuo",4,"","",0,"2026-07-13",""],[6,2487,"CI fails: AttributeError: 'str' object has no attribute '__module__'","albertvillanova",1,"albertvillanova","#2488",1,"2026-07-10","2026-07-10"],[6,2481,"Tool.from_space() fails with gradio_client 2.5.0: Client.init() got an unexpected keyword argument 'hf_token'","osmangulercan",4,"","",0,"2026-07-09",""],[6,2473,"BUG: Thread-based timeout cannot interrupt single big-integer operations","Nkluge-correa",4,"","",0,"2026-07-06",""],[6,2464,"BUG: `local_python_executor.timeout()` deadlocks when the wrapped call hangs forever","Nkluge-correa",4,"","",0,"2026-07-03",""],[6,2458,"BUG: max_steps=0 silently ignored due to falsy-value or operator","axiom-of-choice",4,"","",0,"2026-07-02",""],[6,2457,"Bug: Parallel tool call failure discards ALL results including successful ones","axiom-of-choice",4,"","",0,"2026-07-02",""],[6,2456,"Bug: final_answer_checks silently disabled under python -O (assert used for control flow)","axiom-of-choice",4,"","",0,"2026-07-02",""],[6,2437,"`Tool.to_code_prompt()` method does not extract `Returns: ...` from the function docstring","tovara9000-oss",4,"","",0,"2026-06-30",""],[6,2434,"Optional ATR security check via step_callbacks / final_answer_checks","eeee2345",3,"","",0,"2026-06-29","2026-10-15"],[6,2431,"[Collaboration] First external LoopBench row from SmolAgents","KanakMalpani",4,"","",0,"2026-06-28",""],[6,2424,"BUG: ManagedAgent provide_run_summary leaks inner tool calls and responses into parent observation","Fr3ya",4,"","",0,"2026-06-26",""],[6,2417,"BUG: VLLMModel incompatible with recent vLLM (≥0.11.0)","Nkluge-correa",4,"","",0,"2026-06-25",""],[6,2412,"Optional: add an HVTracker trust badge to the README","YugantM",4,"","",0,"2026-06-23",""],[6,2395,"BUG: Object Lifecycle Hijacking via Unrestricted Dunder Methods in Custom Class Definitions within LocalPython","QiuYucheng2003",4,"","",0,"2026-06-19",""],[6,2394,"Refactor: Replace assert statements with proper validation","Lothnic",4,"","",0,"2026-06-19",""],[6,2393,"FEATURE: Add unique media output naming with counters to prevent overwrites","Lothnic",4,"","",0,"2026-06-19",""],[6,2391,"Proposal: add interop notes for MCP tools and sandboxed execution","kmishra1204",4,"","",0,"2026-06-19",""],[6,2372,"DOC: `evaluate_python_code` docstring is missing two parameters","DPug888",4,"","",0,"2026-06-12",""],[6,2368,"ENH: Executor support for microsandbox","dtbis",3,"","",0,"2026-06-09","2026-06-11"],[6,2365,"BUG: ActionStep 的 observations 也应该是一个 list","coyoteXujie",4,"","",0,"2026-06-09",""],[6,2351,"BUG: KeyError on 'tool_calls' when using AmazonBedrockModel with cross-region inference profile","Koji-Minami",4,"","",0,"2026-06-07",""],[6,2349,"BUG: Broken Space runtime in tools.md documentation example","Koji-Minami",4,"","",0,"2026-06-06",""],[6,2344,"BUG: SmolAgent doc leads to a broken MTEB Leaderboard space","PierreVannier",4,"","",0,"2026-06-04",""],[6,2343,"DOC: Would a formula-readback MCP/Space tool example fit the tools docs?","gregkonush",4,"","",0,"2026-06-04",""],[6,2339,"[RFC] Add PEP 484 type annotations to tools.py public API","yehorcallmedai-maker",2,"","",0,"2026-06-02","2026-06-09"],[6,2332,"[Feature Request] Memory Poisoning Protection for smolagents via OWASP Agent Memory Guard","vgudur-dev",4,"","",0,"2026-05-31",""],[6,2305,"Feature: Add MCP server trust verification for tool security","vdineshk",4,"","",0,"2026-05-24",""],[6,2303,"Feature: MCP server trust verification layer for smolagents","vdineshk",4,"","",0,"2026-05-22",""],[6,2300,"[Integration Proposal] ejentum: pre-generation cognitive scaffolds (4 MCP tools)","ejentum",4,"","",0,"2026-05-22",""],[6,2290,"Security: memory poisoning defense (OWASP ASI06) for smolagents tool memory","vgudur-dev",4,"","",0,"2026-05-19",""],[6,2284,"Discussion: should agents be able to discover external tasks at runtime?","Aigen-Protocol",4,"","",0,"2026-05-16",""],[6,2283,"test-403-check-1778931238","Aigen-Protocol",2,"","",0,"2026-05-16","2026-05-16"],[6,2274,"CI fails: CerebrasException - Missing credentials","albertvillanova",1,"albertvillanova","#2275",1,"2026-05-14","2026-05-14"],[6,2272,"BUG:  No solution found when resolving dependencies while uv","Dharani-Eswaramurthi",4,"","",0,"2026-05-14",""],[6,2267,"Still being maintained?","davidmezzetti",2,"","",0,"2026-05-12","2026-05-13"],[6,2256,"New Release?","victorlearned",2,"","",0,"2026-05-08","2026-05-14"],[6,2251,"📝 Integration Proposal: CAJAL — Local Scientific Paper Tool","Agnuxo1",3,"","",0,"2026-05-04","2026-05-14"],[6,2248,"ENH: Parallel execution mode to overlap the code execution latency with the agent's generation process","v587su",4,"","",0,"2026-05-02",""],[6,2246,"feat: Add SMOLAGENTS_VERBOSE env var for tool call debugging","fuleinist",4,"","",0,"2026-04-30",""],[6,2245,"Hive Civilization — x402-native tools for smolagents (notification)","srotzin",3,"","",0,"2026-04-30","2026-05-14"],[6,2232,"BUG: DANGEROUS_FUNCTIONS check does not block os.system on Windows (nt module mismatch)","adityaghai07",4,"","",0,"2026-04-27",""],[6,2224,"Tool idea: BuyWhere Singapore product catalog for agents doing real-world shopping research","BuyWhere",3,"","",0,"2026-04-25","2026-05-14"],[6,2222,"ENH: Persist ThreadPoolExecutor as Class Instance of ToolCallingAgent","luqmansen",4,"","",0,"2026-04-24",""],[6,2216,"[Feature]: a new concept. Action-only agents","mohamad-tohidi",4,"","",0,"2026-04-22",""],[6,2213,"[Feature] Governance middleware for tool-call policy enforcement","imran-siddique",2,"","",0,"2026-04-20","2026-07-09"],[6,2211,"Tool save can miss requirements for commented imports","luojiyin1987",4,"","",0,"2026-04-20",""],[6,2207,"Typo cleanup across docs, examples, and docstrings","luojiyin1987",1,"luojiyin1987","#2208",0,"2026-04-19","2026-05-14"],[6,2203,"ENH: Architecture Proposal: Hardening CodeAgent LocalPythonExecutor via Defense-in-Depth (AST + PEP 578)","kwdoug63",4,"","",0,"2026-04-17",""],[6,2197,"BUG: LocalPythonExecutor timeout waits for worker completion before returning","shaun0927",4,"","",0,"2026-04-16",""],[6,2180,"Add Chart Library MCP as a finance tool example for smolagents","grahammccain",3,"","",0,"2026-04-08","2026-05-14"],[6,2178,"ENH: Add \"reason\" to interrupt agent","alex-schulster",4,"","",0,"2026-04-08",""],[6,2177,"Audit trail hooks for agent actions","jagmarques",2,"","",0,"2026-04-07","2026-04-12"],[6,2176,"Tool execution governance — delegation check before tool.forward()","aeoess",4,"","",0,"2026-04-07",""],[6,2173,"Proposal: ClawMem example or extension for persistent agent memory","EricZequan",4,"","",0,"2026-04-07",""],[6,2172,"Add audit trail / governance callback support","jagmarques",2,"","",0,"2026-04-06","2026-08-04"],[6,2171,"[Integration Proposal] WhichModel MCP — real-time model pricing for cost-aware agent tool selection","simonamsharp",3,"","",0,"2026-04-06","2026-05-14"],[6,2170,"Tool: DeFi Token Safety Scanner (27 scam patterns, 6 chains)","Aigen-Protocol",3,"","",0,"2026-04-05","2026-05-14"],[6,2168,"Security audit: 65 ungoverned call sites detected (OWASP Agentic Top 10)","Acacian",2,"","",0,"2026-04-05","2026-04-05"],[6,2166,"ManagedAgent swallows tool errors from sub-agents - manager sees empty/None result","VANDRANKI",4,"","",0,"2026-04-04",""],[6,2165,"No built-in retry/backoff for transient model API errors in MultiStepAgent","VANDRANKI",4,"","",0,"2026-04-04",""],[6,2164,"VisitWebpageTool has no response size limit and can silently overflow context window","VANDRANKI",2,"","",0,"2026-04-04","2026-04-04"],[6,2163,"Add documentation and example for using CodeAgent with Groq via LiteLLMModel","VANDRANKI",4,"","",0,"2026-04-04",""],[6,2162,"Make HTTP request timeout configurable in WebSearchTool","VANDRANKI",4,"","",0,"2026-04-04",""],[6,2160,"Next Release","davidmezzetti",2,"","",0,"2026-04-04","2026-05-14"],[6,2151,"Add native OCI Generative AI provider (OCIGenAIModel)","fede-kamel",4,"","",0,"2026-04-02",""],[6,2144,"Add SandlockExecutor for lightweight kernel-level sandboxing on Linux","congwang-mk",4,"","",0,"2026-04-01",""],[6,2142,"feat: support serialization for step_callbacks and final_answer_checks in MultiStepAgent","chjnett",4,"","",0,"2026-04-01",""],[6,2141,"Feature Request: Agent Identity & Trust Layer (e.g. WTRMRK integration)","64R3N",4,"","",0,"2026-04-01",""],[6,2138,"Integration Request: Clarvia AEO tool for agent-ready tool discovery and validation","digitamaz",2,"","",0,"2026-03-31","2026-07-21"],[6,2133,"BUG: f-string escape error in ToolSerializer.dumps causes uninformative error message","Ricardo-M-L",1,"Ricardo-M-L","#2134",0,"2026-03-30","2026-05-14"],[6,2129,"Feature: behavioral fingerprint hook for memory consolidation events in MultiStepAgent","agent-morrow",4,"","",0,"2026-03-29",""],[6,2125,"Add lightweight OS-level sandboxing via sandlock (between LocalPythonExecutor and Docker)","congwang-mk",4,"","",0,"2026-03-27",""],[6,2117,"ENH: Add pre-tool-call authorization layer to MultiStepAgent","uchibeke",4,"","",0,"2026-03-23",""],[6,2114,"[Feature] Native indirect prompt injection shield for tool outputs","NithiN-1808",4,"","",0,"2026-03-22",""],[6,2112,"Feature: x402 payment handling for agents accessing paid APIs","up2itnow0822",4,"","",0,"2026-03-21",""],[6,2109,"Tool suggestion: anybrowse for web scraping with Cloudflare bypass","kc23go",3,"","",0,"2026-03-21","2026-05-14"],[6,2094,"[Security] Sandbox escape via ctypes in LocalPythonExecutor","rishavkumarthapa01-sketch",4,"","",0,"2026-03-17",""],[6,2090,"BUG: with statement crashes when","Lukasbygmal",4,"","",0,"2026-03-16",""],[6,2088,"BUG: GradioUI throws an error when uploading images.","tyxlb",4,"","",0,"2026-03-16",""],[6,2081,"[Security] SSRF Vulnerability in AgentAudio and VisitTool - Server-Side Request Forgery","hacchoseuk21-bot",4,"","",0,"2026-03-15",""],[6,2076,"Support huggingface_hub >= 1.0.0","davidmezzetti",0,"merveenoyan","#2075",1,"2026-03-13","2026-03-13"],[6,2073,"BUG: Can't import classes or methods from smolagents package","aaishwarymishra",2,"","",0,"2026-03-10","2026-03-13"],[6,2071,"Feature: Cryptographic receipts for agent tool execution (AAR)","Cyberweasel777",4,"","",0,"2026-03-09",""],[6,2069,"ENH: Add Google Colab Model (currently available at BPSA)","joaopauloschuler",4,"","",0,"2026-03-08",""],[6,2061,"🔗 New Community Tool: AgentFolio — Agent Identity, Trust & Reputation","0xbrainkid",2,"","",0,"2026-03-07","2026-10-07"],[6,2055,"Surface cache token usage from Anthropic/OpenAI in TokenUsage","oneryalcin",4,"","",0,"2026-03-06",""],[6,2054,"Add prompt caching support for Anthropic models (90% input token savings)","oneryalcin",4,"","",0,"2026-03-06",""],[6,2053,"LocalPythonExecutor silently suppresses stack traces from code execution errors","KlementMultiverse",2,"","",0,"2026-03-06","2026-03-09"],[6,2051,"Async note: compute provisioning path for smolagents users","evgenyponomarev",2,"","",0,"2026-03-05","2026-03-09"],[6,2050,"BUG: Docker executor leaves orphaned containers blocking port 8888 when script stops unexpectedly","lifeishurdle",4,"","",0,"2026-03-05",""],[6,2042,"ENH: Capsule for safe Python/JS code execution in Wasm sandboxes","mavdol",2,"","",0,"2026-03-04","2026-06-25"],[6,2041,"ENH: [RFC] Support loading Agent Spec configurations for cross-framework agent portability","paul-cayet",2,"","",0,"2026-03-03","2026-03-24"],[6,2040,"Multi-agent coordination via entity-level claims (prevents agents from stepping on each other)","rs545837",4,"","",0,"2026-03-03",""],[6,2000,"Add exec-sandbox executor (self-hosted QEMU microVM sandbox)","clemlesne",4,"","",0,"2026-02-27",""],[6,1999,"ENH: Make SmolAgents Smoller","KeithCu",4,"","",0,"2026-02-27",""],[6,1998,"BUG: Syntax errors cause print statements from previous step to emit.","dilwong",4,"","",0,"2026-02-26",""],[6,1992,"[Feature] smolagents debug cookbook using the WFGY 16-problem failure map","onestardao",4,"","",0,"2026-02-21",""],[6,1990,"ENH: Add ProxiesSx tool for mobile proxy access with autonomous USDC payments","bolivian-peru",4,"","",0,"2026-02-19",""],[6,1989,"Proposal: Safety Guardrails for smolagents","imran-siddique",2,"","",0,"2026-02-17","2026-03-08"],[6,1972,"BUG: LiteLLMModel crashes when multiple consecutive system messages are provided","graylin-byte",4,"","",0,"2026-02-08",""],[6,1968,"BUG: TypeError: 'ChatMessage' object is not iterable when use **ChatOpenAI & CodeAgent**","LivinLuo1993",4,"","",0,"2026-02-06",""],[6,1967,"BUG: RAG Examples Needs Update","lucifertrj",4,"","",0,"2026-02-05",""],[6,1965,"BUG: Code execution failed at line 'from sklearn.linear_model import LogisticRegression' due to: AttributeErro","Kenwwww",4,"","",0,"2026-02-03",""],[6,1961,"BUG: Missing parent id for OpenTelemetry tool spans when using CodeAgent","de-code",4,"","",0,"2026-01-29",""],[6,1956,"ENH: Combine CodeAgent and ToolCallingAgent","cyzus",4,"","",0,"2026-01-28",""],[6,1941,"BUG: AmazonBedrockModel returns content in Anthropic's tool_use format","mattdai01",4,"","",0,"2026-01-16",""],[6,1938,"ENH: Support for Open Responses /responses api","bveiseh",4,"","",0,"2026-01-15",""],[6,1934,"GLM models via ZAI","tulas75",4,"","",0,"2026-01-08",""],[6,1930,"ENH: Handling max output token limit","OscarGarciaF",4,"","",0,"2026-01-01",""],[6,1929,"BUG: SmolagentsInstrumentor causes Error during agent execution (Telemetry with Langfuse)","HugoYg-dev",4,"","",0,"2025-12-31",""],[6,1928,"it run good with ollama","kmnnmk212-source",3,"","",0,"2025-12-30","2026-01-14"],[6,1913,"Is it possible to turn off deepseek v3 thinking with smolagents?","XinyueZ",2,"","",0,"2025-12-22","2026-01-14"],[6,1911,"Create New Release?","ZeroCool2u",2,"","",0,"2025-12-21","2026-01-16"],[6,1908,"ENH: Feature request / Idea: Custom CodeAgent: Codex, Claude Code, OpenHeands, Copilot etc","qdrddr",4,"","",0,"2025-12-20",""],[6,1905,"BUG: Agent code should not be able to catch FinalAnswerException","borja-optiak",0,"majiayu000","#1914",0,"2025-12-18","2025-12-23"],[6,1901,"BUG: Upgrade to Gradio v6.x from v5.x and temporarily constrain dependency support","ZeroCool2u",0,"alejandro-ao","#1903",1,"2025-12-17","2025-12-18"],[6,1899,"ENH: TransformersModel: Support for parallel tool calls in single response?","austinmw",4,"","",0,"2025-12-16",""],[6,1897,"BUG: Smolagents cant process function call with reasoning model","AIR-hl",2,"","",0,"2025-12-14","2025-12-15"],[6,1896,"BUG: CodeAgent outputs empty steps that ChatCompletionMessage(content=None,...)","daviden1013",4,"","",0,"2025-12-14",""],[6,1893,"BUG: gpt-5.2 model on CodeAgent returns \"Unsupported parameter: 'stop' is not supported with this model.\"","arturoPMM",0,"aymeric-roucher","#1895",1,"2025-12-12","2025-12-17"],[6,1890,"Question: how to use sever-side tools provided by Google Gemini or OpenAI GPT?","victorx-deckard",4,"","",0,"2025-12-01",""],[6,1889,"ENH: Adding timelimit parameter to duckduckgo search tool","demetriushrowland",4,"","",0,"2025-11-30",""],[6,1887,"BUG: `GradioUI(...).launch()` does not work because of deprecated keyword argument to `gr.Blocks`","atharva-novi",0,"alejandro-ao","#1903",1,"2025-11-28","2025-12-18"],[6,1886,"Question: CodeAgent's remote executor never propagates non-default initialization parameters in tools?","sadaisystems",4,"","",0,"2025-11-26",""],[6,1885,"BUG: Performance degradation after upgrading from 1.17.0 to 1.23.0 + Docker + Ollama + qwen2.5-coder:3b","g-cosco",2,"","",0,"2025-11-26","2025-11-28"],[6,1883,"ENH: Add lifecycle hooks for CodeAgent execution flow","MkYacine",4,"","",0,"2025-11-22",""],[6,1879,"BUG: FinalAnswerStep callbacks not executed","victorlearned",2,"","",0,"2025-11-19","2025-12-28"],[6,1876,"BUG: Object of type ChatCompletionMessageToolCall (LiteLLM) is not JSON serializable","Zoe14",2,"","",0,"2025-11-18","2025-12-18"],[6,1875,"Question on token count manager of open deep research","shang-zhu",4,"","",0,"2025-11-18",""],[6,1874,"BUG: Missing reasoning_effort = \"none\" option for gpt-5.1","arturoPMM",2,"","",0,"2025-11-18","2025-11-19"],[6,1872,"BUG: Streaming outputs breaks CodeAgent due to incomplete code block termination","giltirn",4,"","",0,"2025-11-17",""],[6,1869,"ENH: Interleaved Thinking","ZeroAurora",4,"","",0,"2025-11-15",""],[6,1862,"ENH: Add AssertError exception string to output tool validation","liukidar",4,"","",0,"2025-11-12",""],[6,1860,"BUG: GPT-OSS 20B doesn't generate proper code responses","davidmezzetti",4,"","",0,"2025-11-12",""],[6,1857,"ENH: Feature Request: Add `sites_to_avoid` parameter to `WebSearchTool`","HKanoje",4,"","",0,"2025-11-12",""],[6,1853,"ENH: Add pull request template to .github","colesmcintosh",4,"","",0,"2025-11-07",""],[6,1851,"BUG: Partial stop sequences in model output cause SyntaxError in code parsing","colesmcintosh",2,"","",0,"2025-11-07","2025-11-15"],[6,1849,"BUG: MultiStepAgent.from_dict() passes **kwargs to managed agents, causing child agent configurations to be ov","XXSg559",4,"","",0,"2025-11-07",""],[6,1848,"ENH: claude agent skills","fjyihan",4,"","",0,"2025-11-07",""],[6,1847,"ENH: New requirement: Interruption during React process","xiaoc-coding",4,"","",0,"2025-11-07",""],[6,1843,"CI: CI tests fail for Python 3.12: Failed to build `tokenizers==0.10.3`","albertvillanova",1,"albertvillanova","#1844",1,"2025-11-06","2025-11-06"],[6,1839,"ENH: extracts arguments from CodeAgent function call smolagents","ozanbarism",4,"","",0,"2025-11-03",""],[6,1834,"Discussion: how to edit the messages sent to the underlying LLM","njbrake",2,"","",0,"2025-10-27","2025-10-27"],[6,1830,"BUG: InterpreterError: Forbidden function evaluation: 'open' is not among the explicitly allowed tools","arturoPMM",2,"","",0,"2025-10-22","2026-01-14"],[6,1828,"ENH: Ensure that verbosity_level=0 does not log the user prompt on CodeAgent errors","arturoPMM",2,"","",0,"2025-10-18","2025-10-18"],[6,1825,"BUG: OpenAIModel raises AttributeError when gpt-5 returns tool-only response","chahn",1,"chahn","#1826",0,"2025-10-17","2025-10-27"],[6,1816,"BUG: Duplicate action_step yield and NameError when max_steps=0","AhnJoonSung",4,"","",0,"2025-10-14",""],[6,1815,"BUG: Uninitialized variable and duplicate yield in _run_stream method","AhnJoonSung",2,"","",0,"2025-10-14","2025-10-14"],[6,1814,"BUG: LocalPythonExecutor unable to handle nested for loop","SingCheng",0,"suryabdev","#1823",0,"2025-10-14","2025-10-17"],[6,1810,"Introduce translation guidelines","suryabdev",4,"","",0,"2025-10-12",""],[6,1808,"BUG: Default ToolCallingAgent InferenceClient examples failing","suryabdev",2,"","",0,"2025-10-11","2025-10-17"],[6,1807,"Python 3.12 Compatibility Issue","Edison-A-N",2,"","",0,"2025-10-11","2025-10-18"],[6,1799,"BUG: Premature validation of executor type in `CodeAgent` prevents subclassing.","Lrakotoson",2,"","",0,"2025-10-08","2025-10-10"],[6,1796,"ENH: Model thoughts for ToolCallingAgent","OscarGarciaF",4,"","",0,"2025-10-06",""],[6,1795,"BUG: LocalPythonExecutor Code execution failed at line 'class Status(Enum)","SingCheng",0,"colesmcintosh","#1797",0,"2025-10-06","2025-10-09"],[6,1794,"BUG: VLLMModel breaks when using vllm > 0.10.1","qjflores",2,"","",0,"2025-10-05","2025-10-30"],[6,1789,"ENH: Allow stopping of agents if all tools were called correctly","MRiabov",4,"","",0,"2025-09-29",""],[6,1785,"BUG: Error in generating model output: 'NoneType' object has no attribute 'endswith'","alex-schulster",0,"albertvillanova","#1786",1,"2025-09-24","2025-09-25"],[6,1783,"BUG: Use of gpt-5 produces \"Error while parsing tool call from model output: The model output does not contain","athatheo",4,"","",0,"2025-09-24",""],[6,1781,"BUG: Managed agents share state causing identical parallel execution results","Zoe14",4,"","",0,"2025-09-22",""],[6,1779,"BUG:","daxton256",2,"","",0,"2025-09-21","2025-09-25"],[6,1777,"BUG: AST parsing fails for code that contains newline in the string","debasisdwivedy",3,"","",0,"2025-09-21","2025-09-28"],[6,1775,"ENH: Check Tool input keys","cemde",4,"","",0,"2025-09-16",""],[6,1774,"ENH: Optimize smolagents for reasoning models","Zoe14",4,"","",0,"2025-09-15",""],[6,1770,"BUG: smolagents.utils.AgentGenerationError: Error while generating output","ed21b006",4,"","",0,"2025-09-14",""],[6,1767,"BUG: ModuleNotFoundError: No module named 'ddgs'","valmetisrinivas",2,"","",0,"2025-09-11","2025-09-17"],[6,1765,"BUG: xAI's grok-4 and grok-3-mini not supporting the stop parameter","kutsushitaneko",0,"suryabdev","#1766",0,"2025-09-11","2025-09-17"],[6,1764,"ENH:  Add User Elicitation Support to smolagents MCPClient","amirkiarafiei",4,"","",0,"2025-09-09",""],[6,1744,"Add support for e2b-code-interpreter v2","albertvillanova",0,"August-murr","#1751",0,"2025-09-01","2025-09-04"],[6,1743,"BUG: DockerExecutor fails to initialize with \"Connection reset by peer\" error","albertvillanova",0,"benoriol","#1684",0,"2025-09-01","2025-09-01"],[6,1742,"BUG: agent.replay(detailed=True) fails with error:  'ChatMessage' object is not iterable","lkubaski",0,"Flakes342","#1763",0,"2025-09-01","2025-10-09"],[6,1740,"Asking for documentations on the datasets released in HF datasets.","bwookwak",2,"","",0,"2025-09-01","2025-09-01"],[6,1739,"BUG: e2b executor is not composable with MCP tools","jacopotagliabue",4,"","",0,"2025-08-31",""],[6,1732,"Error while parsing tool call from model output: The model output does not contain any JSON blob.","aneesarom",4,"","",0,"2025-08-29",""],[6,1730,"BUG: e2b executor does not work: TypeError: SandboxBase.__init__() missing 5 required positional arguments","jacopotagliabue",0,"albertvillanova","#1733",1,"2025-08-28","2025-08-31"],[6,1728,"CI fails to build main documentation: RuntimeError: The following files are present in the table of contents b","albertvillanova",1,"albertvillanova","#1729",1,"2025-08-28","2025-08-28"],[6,1726,"CI fails to build main documentation: mapping values are not allowed here","albertvillanova",1,"albertvillanova","#1727",1,"2025-08-28","2025-08-28"],[6,1725,"BUG: 🐛 LiteLLMRouterModel Error","also-good",2,"","",0,"2025-08-28","2025-08-28"],[6,1724,"BUG: MCP tools calls gets logged inside the Python Code intepreter","jacopotagliabue",4,"","",0,"2025-08-28",""],[6,1719,"What's the best way of preventing token overflow when the agent is making LLM request?","duguyue100",2,"","",0,"2025-08-26","2025-10-12"],[6,1717,"✨ support microsandbox as Executor","also-good",4,"","",0,"2025-08-26",""],[6,1715,"CI fails: 403 error for Wikipedia page in test_visit_webpage","albertvillanova",1,"albertvillanova","#1716",1,"2025-08-26","2025-08-26"],[6,1713,"BUG: Web search tools can hang indefinitely","dzungvpham",4,"","",0,"2025-08-24",""],[6,1706,"BUG: The variable `bytes` is not defined.","HairlessVillager",2,"","",0,"2025-08-21","2025-08-21"],[6,1704,"BUG: Sometimes I need show the python code directly rather than execute","SimengBian",4,"","",0,"2025-08-21",""],[6,1703,"BUG: Stop sequences not working properly in TransformersModel","MaxBourdon",4,"","",0,"2025-08-20",""],[6,1697,"Support openai 1.100","albertvillanova",1,"albertvillanova","#1698",1,"2025-08-20","2025-08-20"],[6,1696,"BUG: Cannot use remote Langfuse with local Ollama model","toniher",3,"","",0,"2025-08-19","2025-08-20"],[6,1695,"BUG: Regarding multi-agents, when the management agent calls the sub-agent, the sub-agent's memory will be cle","whfeLingYu",4,"","",0,"2025-08-19",""],[6,1694,"BUG: The gpt oss generated format cannot be processed correctly","whfeLingYu",4,"","",0,"2025-08-19",""],[6,1692,"CI fails: ImportError: cannot import name 'ResponseTextConfig' from 'openai.types.responses.response'","albertvillanova",1,"albertvillanova","#1693",1,"2025-08-18","2025-08-18"],[6,1689,"BUG: smolagents.utils.AgentGenerationError: Error in generating model output: Error code: 400 - {'object': 'er","02deno",2,"","",0,"2025-08-18","2025-08-18"],[6,1688,"BUG: Return statement in try-except block not working as expected in LocalPythonExecutor","niuhaoxing",2,"","",0,"2025-08-18","2025-08-18"],[6,1687,"[Question] Open Deep Research future plans?","davidquicast",4,"","",0,"2025-08-18",""],[6,1682,"BUG: SmolagentsInstrumentor crashes with OpenAIServerModel: AttributeError for last_input_token_count / last_o","KingJacM",0,"albertvillanova","#1699",1,"2025-08-14","2025-08-20"],[6,1677,"DOC: There is no way to provide the LLM temperature and other settings","vijayanmmm",0,"albertvillanova","#1679",1,"2025-08-12","2025-08-13"],[6,1676,"ENH: Add user-configurable parameter override system for model completion parameters","albertvillanova",1,"albertvillanova","#1678",1,"2025-08-12","2025-09-12"],[6,1675,"BUG: Error while generating or parsing output: Error code: 400 - {'error': {'message': \"Unsupported parameter:","DevNinjaDynasty",2,"","",0,"2025-08-12","2025-08-12"],[6,1674,"BUG: AmazonBedrockServerModel Throw Error When calling DeepSeek R1: \"text\" field not found","khe-gh",0,"albertvillanova","#1681",1,"2025-08-11","2025-08-18"],[6,1673,"BUG: AmazonBedrockServerModel throw error when using DeepSeek R1","khe-gh",2,"","",0,"2025-08-11","2025-08-11"],[6,1672,"ENH: Prompt caching","Owe1n",4,"","",0,"2025-08-11",""],[6,1671,"BUG: Trying to use GPT-5-mini leads to \"Unsupported parameter: 'stop' is not supported with this model.\"","de-code",2,"","",0,"2025-08-11","2025-08-11"],[6,1664,"examples/sandboxed_execution","ftgreat",2,"","",0,"2025-08-08","2025-08-19"],[6,1649,"BUG: The LocalPythonExecutor currently lacks support for the yield keyword","kingdomad",4,"","",0,"2025-08-06",""],[6,1640,"ENH: A human can review and edit the output from the LLM before proceeding tool call","coyoteXujie",4,"","",0,"2025-08-01",""],[6,1629,"CI warning: PytestUnknownMarkWarning: Unknown pytest.mark.timeout","albertvillanova",1,"albertvillanova","#1630",1,"2025-07-28","2025-07-28"],[6,1626,"BUG: @tool cannot correctly handle the body of some functions in remote Python executor","HairlessVillager",1,"HairlessVillager","#1627",0,"2025-07-28","2025-07-31"],[6,1624,"Support transformers 4.54.0","albertvillanova",1,"albertvillanova","#1625",1,"2025-07-27","2025-07-27"],[6,1619,"CI test fails: ImportError: cannot import name 'require_soundfile' from 'transformers.testing_utils'","albertvillanova",1,"albertvillanova","#1620",1,"2025-07-26","2025-07-26"],[6,1613,"BUG: push_to_hub generates invalid app.py due to use of class as a keyword argument","albertvillanova",1,"albertvillanova","#1615",1,"2025-07-25","2025-07-25"],[6,1609,"BUG: In Gradio_UI, clicking on the \"clear\" button (the bin icon on the top right) does not reset the agent's h","enzomich",0,"albertvillanova","#1611",1,"2025-07-24","2025-07-24"],[6,1607,"[i18n-ko] Translating smolagents docs to Korean","wo0jun",4,"","",0,"2025-07-24",""],[6,1606,"ENH: Support Default SamplingParam Customisation for VLLMModel","KOU-199024",4,"","",0,"2025-07-23",""],[6,1602,"ENH: 🧬 Feature Request: Life-Science info gathering Tool for Life Science and Health Questions","david-thrower",4,"","",0,"2025-07-21",""],[6,1600,"BUG: human-in-the-loop-customize-agent-plan-interactively","kumarkrish85",3,"","",0,"2025-07-21","2025-07-24"],[6,1596,"BUG: Remote MCP return string instead of json dict","rudaoshi",0,"chahn","#1663",0,"2025-07-20","2025-08-22"],[6,1588,"Support mlx 0.26.5","albertvillanova",1,"albertvillanova","#1589",1,"2025-07-20","2025-07-20"],[6,1587,"CI test fails: ImportError: libmlx.so: cannot open shared object file: No such file or directory","albertvillanova",1,"albertvillanova","#1586",1,"2025-07-20","2025-07-20"],[6,1585,"mcp.shared.exceptions.McpError: Session terminated","eromoe",4,"","",0,"2025-07-19",""],[6,1579,"ENH: Add a way to insert custom chat history for multi-turn chat and what's the current best way?","duguyue100",4,"","",0,"2025-07-18",""],[6,1578,"BUG: Missing \"Thought\" and \"Code\" sections in `phoenix` with `CodeAgent`","alex-schulster",4,"","",0,"2025-07-18",""],[6,1576,"BUG: MultiStepAgent.to_dict() and from_dict() do not use same data structure for managed_agents, causing from_","hazrmard",1,"hazrmard","#1577",0,"2025-07-18","2025-07-24"],[6,1575,"ENH: Add built-in pretty-print utility for agent.write_memory_to_messages()","yuka-with-data",4,"","",0,"2025-07-18",""],[6,1572,"Reimplement wrongly implemented test_change_tools_after_init","albertvillanova",4,"","",0,"2025-07-17",""],[6,1569,"BUG: Parallel function calling does not work in streaming mode","dzungvpham",4,"","",0,"2025-07-17",""],[6,1568,"BUG: Concatenation of Messages Can Causes Malformed Prompts when Agent Memory is Enabled","Hrayo712",4,"","",0,"2025-07-16",""],[6,1567,"DOC: text generation interface (TGI) documentation is missing.","hadifar",4,"","",0,"2025-07-16",""],[6,1565,"BUG: Unable to get example code working","dhairyadalal",0,"albertvillanova","#1570",1,"2025-07-16","2025-07-17"],[6,1560,"Support async in smolagents","duguyue100",4,"","",0,"2025-07-15",""],[6,1557,"Passing extra attributes to tool definitions","anil-gurbuz",2,"","",0,"2025-07-15","2025-07-15"],[6,1555,"[BUG] Confused about the token usage","AIR-hl",4,"","",0,"2025-07-14",""],[6,1554,"MCPAdapt added support for ImageContent and AudioContent","grll",0,"albertvillanova","#1558",1,"2025-07-14","2025-07-16"],[6,1547,"Feature Request：Support Integration of Anthropic’s Think Tool into FunctionCallingAgent & CodeAgent","seanchen1993",4,"","",0,"2025-07-11",""],[6,1543,"✨ Support GradioUI Multi-Agent","also-good",4,"","",0,"2025-07-10",""],[6,1542,"regressiong error","weiyanhein",3,"","",0,"2025-07-10","2025-07-15"],[6,1541,"Feature Request: Extend final_answer_check to Support Contextual Validation","Zoe14",4,"","",0,"2025-07-09",""],[6,1537,"Add support for Bedrock API Keys","ConnorKirk",0,"colesmcintosh","#1612",0,"2025-07-09","2025-07-25"],[6,1536,"[BUG] qwen3:4b on managed agents doesnt seem to work properly and no tool call happens.","SiddhardhaSaran",4,"","",0,"2025-07-08",""],[6,1535,"[BUG] CLI fails to load Hugging Face Spaces as tools due to missing required parameters in Tool.from_space()","aniaan",0,"balazsthomay","#1859",0,"2025-07-08","2025-11-17"],[6,1532,"[BUG] ChatMessage `content` Attribute Accessed Incorrectly","mdr223",1,"mdr223","#1533",0,"2025-07-07","2025-07-08"],[6,1531,"[BUG] Error code: 400 via groq-api and CoderAgent","alinzh",2,"","",0,"2025-07-07","2025-07-07"],[6,1524,"Error \"Error in generating final LLM output: 'image'\" when agent reaches max steps with image input","ZhangNy301",0,"robin-ede","#1525",0,"2025-07-05","2025-07-07"],[6,1518,"[BUG] smolagents returns 422 if using OpenAIServerModel as the model, with deepinfra as the provider.","lvicentesanchez",2,"","",0,"2025-07-04","2025-07-04"],[6,1512,"How can we use this benchmark to evaluate local models?","OoOPenN",2,"","",0,"2025-07-03","2025-07-03"],[6,1511,"[BUG] The `Returns` part of the docstring of a tool is not taken into account","xavier-owkin",4,"","",0,"2025-07-02",""],[6,1504,"[BUG]FileNotFoundError for GAIA Image cca530fc-4052-43b2-b130-b30968d8aa44.png","genshindawang",2,"","",0,"2025-06-30","2025-07-01"],[6,1501,"Add robust error handling for duckduckgo_search.RatelimitException-HTTP-202","david-thrower",0,"albertvillanova","#1515",1,"2025-06-29","2025-07-04"],[6,1499,"[BUG] Step callbacks are not executed for planning steps","spisupat",2,"","",0,"2025-06-27","2025-06-27"],[6,1490,"[Doc] Simple Tool Usage in Quick Start - DuckDuckGoSearchTool","yuka-with-data",2,"","",0,"2025-06-26","2025-06-27"],[6,1489,"[BUG] `final_answer_checks` are called after each step, not only the final one","xavier-owkin",2,"","",0,"2025-06-26","2025-06-27"],[6,1488,"[BUG] Reported token utilization changes by order of magnitude with engine streaming/not streaming","peabody124",0,"albertvillanova","#1503",1,"2025-06-26","2025-06-30"],[6,1487,"Provide an option to limit the number of images retained in AgentMemory","mkrauklis",4,"","",0,"2025-06-25",""],[6,1483,"Automate large scale repository edits","AleKitsune98",2,"","",0,"2025-06-25","2025-06-25"],[6,1481,"[BUG] Final answer is too greedy in ToolCallingAgent parallel calls","aymeric-roucher",1,"aymeric-roucher","#1482",1,"2025-06-24","2025-06-27"],[6,1478,"[BUG] SpeechToTextTool returns incoherent output and prints unexpected informational messages","Martin-Labenne",4,"","",0,"2025-06-24",""],[6,1474,"[BUG] Local python executor fails to run `isinstance` function","xavier-owkin",0,"albertvillanova","#1475",1,"2025-06-23","2025-06-24"],[6,1467,"How can I add prompt words in the most elegant way to make the final answer of agents in Chinese or all the re","ShelterWFF",2,"","",0,"2025-06-22","2025-06-22"],[6,1466,"'NoneType' object has no attribute 'input_tokens'","satwikkansal",0,"aymeric-roucher","#1449",1,"2025-06-22","2025-06-30"],[6,1465,"[BUG] Broken links are not being correctly checked before returning in output","MHRDYN7",4,"","",0,"2025-06-21",""],[6,1461,"[BUG] TypeError when instantiating SpeechToTextTool with model argument","Martin-Labenne",0,"abdulhakkeempa","#1462",0,"2025-06-21","2025-06-22"],[6,1460,"Feature Proposal: MCP Resource Support","grll",4,"","",0,"2025-06-20",""],[6,1455,"[BUG] ToolCallingAgent ignores managed agents","albertvillanova",1,"albertvillanova","#1456",1,"2025-06-19","2025-06-19"],[6,1453,"[BUG] Running tool agent with deepseek model results in invalid type: sequence error","sergeevii123",2,"","",0,"2025-06-18","2025-06-18"],[6,1450,"[Feature] Allow additional arguments into prompt templates","LeoWanty",4,"","",0,"2025-06-18",""],[6,1443,"[BUG]  None check is not there for FinalOutput.output","bacharSalleh",0,"aymeric-roucher","#1448",1,"2025-06-16","2025-06-19"],[6,1440,"[BUG] Agent cannot run when final_answer_checks is provided","Zoe14",0,"aymeric-roucher","#1448",1,"2025-06-13","2025-06-19"],[6,1438,"Can you provide an example script for running on BrowseComp?","hzy312",4,"","",0,"2025-06-13",""],[6,1437,"[Question | Feat]Can we make live rendering of output optional?","nanic",4,"","",0,"2025-06-12",""],[6,1436,"[BUG] Streaming failing token usage Nonetype management in OpenAIServerModel","nanic",2,"","",0,"2025-06-12","2025-06-12"],[6,1435,"Is there a way to return plan of a CodeAgent?","yjiahao",0,"albertvillanova","#1445",1,"2025-06-12","2025-06-30"],[6,1433,"GradioUI - Audio","rinabuoy",4,"","",0,"2025-06-12",""],[6,1432,"[BUG] AmazonBedrockServerModel doesnt work with ToolCallingAgent","cmclel7",4,"","",0,"2025-06-11",""],[6,1431,"[BUG] example code doesn't work","fwwucn",0,"vladlen32230","#1434",0,"2025-06-11","2025-06-17"],[6,1430,"Tool as final_answer","jzju",2,"","",0,"2025-06-11","2025-06-16"],[6,1428,"[BUG] Multiline final answer not detected","tobiasofsn",0,"albertvillanova","#1444",1,"2025-06-10","2025-06-24"],[6,1423,"web browswer example's model is misleading","zzehli",0,"albertvillanova","#1424",1,"2025-06-08","2025-06-10"],[6,1419,"Support prompt caching for claude models","liangjs",4,"","",0,"2025-06-07",""],[6,1404,"Exclude thinking generation from python parsing","sbucaille",4,"","",0,"2025-06-01",""],[7,4093,"Swapped slice lengths in Accelerate and MKL matmul std::slice::from_raw_parts","BUKOWSKIREAL",4,"","",0,"2026-11-06",""],[7,4084,"bert: attention uses generic ops::softmax; softmax_last_dim is ~32% faster on CPU with identical output","rich-murphey",4,"","",0,"2026-11-01",""],[7,4074,"P0 security: candle-core 0.11.0 requires zip ^8.6.0, covered by critical GHSA-m665-v7pm-7grj (fixed in zip 9.0","srinji-kaggss",2,"","",0,"2026-10-29","2026-10-31"],[7,4070,"Repeated CPU thread-count detection slows down inference when RAYON_NUM_THREADS is unset","WatNeru",4,"","",0,"2026-10-28",""],[7,4064,"Error: backward is not supported for conv-transpose1d","LunaticWyrm467",4,"","",0,"2026-10-24",""],[7,4063,"ConvTranspose1D: unsigned subtraction underflow for valid output shapes","LunaticWyrm467",4,"","",0,"2026-10-24",""],[7,4058,"Current version BROKEN due to ZIP being yanked","Goldziher",2,"","",0,"2026-10-22","2026-10-31"],[7,4052,"The fused CPU RMSNorm rounds the scale to the storage dtype, and the two implementations in `ops.rs` disagree ","lu-zero",4,"","",0,"2026-10-20",""],[7,4049,"candle-onnx: Shape op ignores ONNX semantics for start and end","RaphaelFakhri",4,"","",0,"2026-10-20",""],[7,4047,"Implement backward pass for flash attention","Enderlook",4,"","",0,"2026-10-19",""],[7,4043,"wasm32 simd128: Q8_0 vec_dot_2 / vec_dot_4","mgoldio",4,"","",0,"2026-10-16",""],[7,4042,"Release zip 9 support: zip 8.6.0 is yanked","jbg",2,"","",0,"2026-10-16","2026-10-31"],[7,4034,"ONNX Transpose with omitted perm swaps only the last two axes","vera-rubin",4,"","",0,"2026-10-12",""],[7,4031,"conv1d/encoder gives different results for strided (narrow) vs contiguous input with identical values","Trisert",4,"","",0,"2026-10-10",""],[7,4029,"x86 repacked matmul: add a Q5K tile","bhaskargurram-ai",4,"","",0,"2026-10-09",""],[7,4027,"Metal: Tensor::copy() shares the source's buffer, so in-place updates (Var::set, zero_set) show through the \"c","skilesare",4,"","",0,"2026-10-08",""],[7,4023,"ONNX ReduceMax/ReduceMin treat explicitly empty axes as a no-op by default","hulkbig",4,"","",0,"2026-10-05",""],[7,4021,"Metal SDPA: support head dim 48 (d_model 384 / 8 heads)","OmniaFlux",4,"","",0,"2026-10-03",""],[7,4020,"Metal SDPA silently returns wrong values for non-contiguous q/k/v","OmniaFlux",4,"","",0,"2026-10-03",""],[7,4018,"Apertus model support","saona-raimundo",4,"","",0,"2026-09-30",""],[7,4014,"wasm32 simd128 backend doesn't define CurrentCpuF16/CurrentCpuBF16, breaking any build with target-feature=+si","willarrington",4,"","",0,"2026-09-29",""],[7,4013,"disregard","willarrington",3,"","",0,"2026-09-29","2026-09-29"],[7,4011,"Metal: power-of-two buffer pooling doubles peak memory for large intermediates (conv1d im2col) and OOMs on lon","adamswanglin",2,"","",0,"2026-09-29","2026-10-02"],[7,4004,"quantized_phi3: partial rotary embeddings fail with \"inconsistent last dim size in rope\"","czeti",2,"","",0,"2026-09-27","2026-10-19"],[7,3994,"Backprop: index-add only supports contiguous tensors","LunaticWyrm467",0,"emicastro","#3997",0,"2026-09-21","2026-09-24"],[7,3986,"Unsound APIs (`Layout::new`, `Layout::contiguous_with_offset`) — unchecked `start_offset` leads to heap-buffer","ksj1230",4,"","",0,"2026-09-17",""],[7,3985,"Vulkan accelerator backend (Device::Vulkan): implementation available, seeking to upstream","rexlunae",4,"","",0,"2026-09-16",""],[7,3982,"CSOAI — Candle interop: Rust ML governance","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[7,3981,"CSOAI — Candle interop: Rust ML governance","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[7,3980,"CSOAI — Candle interop: Rust ML governance","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[7,3979,"CSOAI — Candle AI governance Rust ML","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[7,3972,"Regression: `layer_norm()`/`rms_norm()` always fails to find weight/bias tensors [main branch]","cryptomilk",0,"dundysm","#3973",0,"2026-09-11","2026-09-28"],[7,3967,"Gemma 4 models cannot use","Zhang-JK",4,"","",0,"2026-09-09",""],[7,3958,"[Closed] perf(cpu): Rotary Embeddings (RoPE)","bmartin-systems",3,"","",0,"2026-09-04","2026-09-07"],[7,3948,"mimi uses the interleaved RoPE convention but the checkpoints are stored in the transformers layout","jorge-menjivar",1,"jorge-menjivar","#3949",0,"2026-09-03","2026-09-24"],[7,3940,"CUDA `copy_strided_src` copies past the end of an offset source view, silently corrupting `slice_scatter` (CPU","PCfVW",1,"PCfVW","#3944",0,"2026-09-02","2026-09-24"],[7,3924,"[feature] Adding Deepseek v4 Flash architecture to candle.","haydonryan",3,"","",0,"2026-08-29","2026-09-08"],[7,3921,"Design check: Intel Arc (SYCL) backend","Hahihula",4,"","",0,"2026-08-28",""],[7,3915,"Unable to build whisper example","inferrna",0,"HuzaifaAbdulRehman","48fb171",0,"2026-08-24","2026-09-02"],[7,3907,"Reshape ignores the allowzero attribute (candle-onnx)","keevin-xu",1,"keevin-xu","#3908",0,"2026-08-20","2026-08-23"],[7,3901,"qwen3_vl: no way to reset the KV cache, so a model instance is single-use","sidd-27",4,"","",0,"2026-08-17",""],[7,3900,"CUDA is bound at link time, so a `cuda`-enabled binary cannot start without CUDA installed","nabbisen",4,"","",0,"2026-08-17",""],[7,3893,"`conv2d_grad_noncontiguous_kernel_gpu` fails on CUDA: PR #3736 fixed the CPU offset bug but not the GPU path","astorise",1,"astorise","8e68623",0,"2026-08-15","2026-08-29"],[7,3886,"F8E4M3 CUDA kernels are unreachable: `DType::as_str()` yields `f8e4m3`, the kernels are named `f8_e4m3`","astorise",4,"","",0,"2026-08-14",""],[7,3884,"`get_or_load_func` drops the kernel name from CUDA symbol-lookup errors","astorise",4,"","",0,"2026-08-14",""],[7,3882,"ModelOpt per-tensor FP8 falls through to a dense read, dropping `weight_scale`","astorise",2,"","",0,"2026-08-13","2026-08-17"],[7,3878,"potential deadlock risk from multiple read locks over storage in tensor operations","scaspin",4,"","",0,"2026-08-12",""],[7,3874,"CPU argsort/sort_last_dim ignores the view's start offset, silently mis-sorting narrow'd tensors","phas02",0,"LegenDairy93","#3875",0,"2026-08-11","2026-08-13"],[7,3871,"`broadcast_matmul` copies the whole rhs when only the batch dims need broadcasting","pjmalandrino",1,"pjmalandrino","#3872",0,"2026-08-09","2026-08-13"],[7,3868,"qwen3_5::Config deserializes without checking its fields agree, so an incoherent one fails deep instead of at ","astorise",4,"","",0,"2026-08-09",""],[7,3866,"ModelOpt checkpoints: the weight source stops at the projection; detection, shards and config are still on eve","astorise",4,"","",0,"2026-08-08",""],[7,3865,"Qwen3.5 quantized (GGUF) MoE checkpoints are not supported (qwen35moe)","astorise",4,"","",0,"2026-08-08",""],[7,3864,"QTensor::quantize on a narrowed view quantizes the parent (silently in release, debug_assert in debug)","oetiker",4,"","",0,"2026-08-08",""],[7,3863,"Metal: K-quant mul_mv kernels store past ne01; Q2K over-dispatches 2x on top of it","oetiker",4,"","",0,"2026-08-08",""],[7,3860,"Nvfp4LinearPacked decodes each block scale once per column, not once per block","astorise",2,"","",0,"2026-08-08","2026-08-17"],[7,3857,"NVFP4 has no execution path off CUDA: no packed CPU operator, and no Metal backend like its sibling crates","astorise",2,"","",0,"2026-08-06","2026-08-17"],[7,3850,"Metal: command buffer execution errors are silently ignored, producing wrong results instead of an error","oetiker",1,"oetiker","#3862",0,"2026-08-06","2026-08-14"],[7,3847,"Metal: `Tensor::expand` backward intermittently produces NaN gradients under repeated execution","Not-Nik",0,"waamm","#3873",0,"2026-08-05","2026-08-13"],[7,3846,"ModelOpt/NVFP4: nothing connects candle-nvfp4-kernels to the models that can use them","astorise",2,"","",0,"2026-08-05","2026-08-17"],[7,3837,"qwen3_5: the sparse-MoE layers and partial rotary the released checkpoints need","astorise",4,"","",0,"2026-08-04",""],[7,3835,"candle-core fails to build for wasm32 with +simd128 (the repo's own default): CurrentCpuF16 undefined on the s","Ttimmahlax",0,"damageboy","#3904",0,"2026-08-03","2026-09-02"],[7,3832,"qwen3_5: a projection seam, so a non-GGUF quantized checkpoint can reuse the layers","astorise",4,"","",0,"2026-08-03",""],[7,3831,"candle-nvfp4-kernels: device-resident weights and a Tensor-level matmul","astorise",2,"","",0,"2026-08-03","2026-08-17"],[7,3828,"`candle-nn`: column- and row-parallel linear layers for multi-GPU inference","astorise",4,"","",0,"2026-08-02",""],[7,3826,"`candle-transformers::generation`: constrained decoding from a JSON schema","astorise",4,"","",0,"2026-08-02",""],[7,3824,"Add `candle-nvfp4-kernels` for NVIDIA ModelOpt FP4 weights","astorise",2,"","",0,"2026-08-02","2026-08-17"],[7,3820,"`quantized_lm`: dispatch Qwen3.5 GGUF checkpoints","astorise",4,"","",0,"2026-08-01",""],[7,3817,"`candle-transformers::models::llama`: expose `Block` for per-block composition","astorise",4,"","",0,"2026-08-01",""],[7,3816,"Shape::elem_count() silently wraps in release, producing a Tensor with corrupted metadata (dims disagree with ","x14ngch3n",4,"","",0,"2026-08-01",""],[7,3815,"quantized/ggml_file.rs: from_raw_parts on a 1-byte-aligned pointer without an alignment check","x14ngch3n",4,"","",0,"2026-07-31",""],[7,3812,"Tensor::from_slice / from_vec silently accept a concrete shape whose element count ≠ data length","astorise",0,"ccyyy1023","#3813",0,"2026-07-31","2026-08-13"],[7,3810,"`candle-transformers::generation`: provide stop criteria and a finish reason","astorise",4,"","",0,"2026-07-30",""],[7,3809,"`candle-transformers::models::llama`: expose a seam for substituting the MLP","astorise",4,"","",0,"2026-07-30",""],[7,3808,"13 PRs related to performance, 3d support, metal and old hardware support","mahogny",4,"","",0,"2026-07-30",""],[7,3793,"candle-kernels 0.9.2 fails to build CUDA kernels for sm75; build script unconditionally compiles MoE WMMA kern","mahogny",4,"","",0,"2026-07-27",""],[7,3783,"`candle-transformers`: dispatch GGUF checkpoints by `general.architecture`","astorise",4,"","",0,"2026-07-26",""],[7,3782,"`candle-transformers::generation`: provide a correct incremental detokenizer","astorise",4,"","",0,"2026-07-26",""],[7,3781,"`quantized`: expose whether a `GgmlDType` supports matmul on a given device","astorise",4,"","",0,"2026-07-26",""],[7,3774,"Add doc for public tensor operation","ZENOTME",4,"","",0,"2026-07-25",""],[7,3769,"Allow loading LoRA adapters into an existing Llama without reloading base weights","astorise",4,"","",0,"2026-07-23",""],[7,3766,"Add batch-native multi-prompt decode loop for heterogeneous LoRA serving","astorise",4,"","",0,"2026-07-22",""],[7,3764,"qwen3: causal_mask builds data for one batch element but shapes the tensor (b, 1, tgt, tgt+offset) — garbage/N","jhughes-dev",4,"","",0,"2026-07-21",""],[7,3759,"Add heterogeneous LoRA batching kernels for S-LoRA/Punica-style multi-adapter forward","astorise",4,"","",0,"2026-07-20",""],[7,3752,"[Invalid — misfiled downstream issue] LayerNorm fast path silently severs autograd graph (apply_op3_no_bwd)","ynishi",2,"","",0,"2026-07-19","2026-07-19"],[7,3750,"DeBERTa-v2 models fail to run at F16","tobocop2",4,"","",0,"2026-07-18",""],[7,3745,"Gemma models use the wrong activation function","fleur-in-progress",4,"","",0,"2026-07-16",""],[7,3744,"CPU matmul silently returns incorrect values for stride-zero broadcast batch","tomsanbear",0,"akashjainn","#3960",0,"2026-07-16","2026-09-24"],[7,3739,"Track ROCm follow-ups for PR #3424 (RDNA4 / ROCm 7.1)","zachelnet",2,"","",0,"2026-07-12","2026-07-17"],[7,3737,"regression in support for nvidia rtx 20xx GPUs with newer nvcc toolchains","embrynod",4,"","",0,"2026-07-12",""],[7,3735,"CPU conv2d backward produces incorrect gradients for non-contiguous kernels","EthanAlmloff",2,"","",0,"2026-07-11","2026-07-12"],[7,3732,"PagedKvCache::write_new_kv's host readback breaks CudaGraph capture for the decode step","astorise",4,"","",0,"2026-07-10",""],[7,3728,"Add a graph-replay-safe decode position seam (`Cache::set_decode_position`) for CUDA graph decode capture","astorise",4,"","",0,"2026-07-09",""],[7,3727,"Softmax on f16/bf16 is numerically broken on long axes (rows sum to ~2.0, not 1.0)","v-code01",4,"","",0,"2026-07-08",""],[7,3725,"Add an additive use_flashinfer_attention seam to candle-transformers::models::llama for external decode-attent","astorise",4,"","",0,"2026-07-08",""],[7,3720,"Add an additive Cache::Paged seam to candle-transformers::models::llama for external paged-attention wiring","astorise",4,"","",0,"2026-07-06",""],[7,3708,"No lazy/mmap loading for GGUF weights (MmapedSafetensors equivalent)","kitecosmic",4,"","",0,"2026-07-05",""],[7,3707,"CPU quantized matmul has no batch path: prefill runs at token-at-a-time speed","kitecosmic",4,"","",0,"2026-07-05",""],[7,3705,"Metal: deterministic corruption in quantized generation once the KV cache reaches depth 8,192 (evidence points","shayne-fletcher",4,"","",0,"2026-07-05",""],[7,3701,"[feature] Opt-in block metadata for long-context KV pruning experiments","teerthsharma",4,"","",0,"2026-07-04",""],[7,3698,"phi3: long-prompt output degrades into word-salad — BF16 RoPE table + ignored `sliding_window` (rope dtype als","AdamJonR",4,"","",0,"2026-07-02",""],[7,3696,"Add first-class LoRA adapter injection to candle-transformers model loaders","astorise",4,"","",0,"2026-06-30",""],[7,3687,"candle-datasets ssl","scskZacharyDovel",4,"","",0,"2026-06-30",""],[7,3686,"candle-kernels 0.10.2 fails to build with CUDA 13.2 on Windows (CCCL requires /Zc:preprocessor)","Mahavishnu-K",4,"","",0,"2026-06-29",""],[7,3683,"Write CUDA kernels in Rust","mayocream",4,"","",0,"2026-06-29",""],[7,3679,"test_gelu_operation fails on Apple Silicon (1 ULP f32 precision mismatch)","jaweed3",4,"","",0,"2026-06-28",""],[7,3659,"Ouro llm support for candle","silicology1",4,"","",0,"2026-06-26",""],[7,3656,"[feature] CPU/Metal (device-agnostic) PagedAttention to complement the CUDA kernels from #3655","astorise",4,"","",0,"2026-06-25",""],[7,3654,"[feature] Add min_p sampling and frequency/presence penalties to generation","astorise",4,"","",0,"2026-06-25",""],[7,3653,"[feature] Add DeepSeek-V3, Llama 4 and Qwen3-MoE model architectures","astorise",4,"","",0,"2026-06-25",""],[7,3652,"[feature] First-class LoRA adapters in candle-nn (load + merge + multi-adapter)","astorise",4,"","",0,"2026-06-25",""],[7,3651,"[feature] CUDA Graph capture for autoregressive decode + optional FlashInfer backend","astorise",4,"","",0,"2026-06-25",""],[7,3650,"[feature] Weight quantization kernels: GPTQ/Marlin, AWQ, block-wise FP8","astorise",4,"","",0,"2026-06-25",""],[7,3649,"[feature] PagedAttention: block-paged KV-cache attention kernel","astorise",2,"","",0,"2026-06-25","2026-06-25"],[7,3638,"Gemma 4 like-Architecture ","AI10x",3,"","",0,"2026-06-21","2026-06-21"],[7,3637,"Exponentially increase MoE log linearly for N Perams — and total magnitude?","AI10x",3,"","",0,"2026-06-21","2026-06-21"],[7,3633,"Missing 0.10.2 git tag — crates published to crates.io but the release was never tagged","yoshihirosuzuki",2,"","",0,"2026-06-19","2026-06-25"],[7,3620,"Pickle memo bomb — CPU + memory exhaustion (algorithmic-complexity DoS) in candle-core pickle reader","professor-moody",4,"","",0,"2026-06-14",""],[7,3619,"Deterministic panic (DoS) via `Long1` opcode in candle-core's pickle reader","professor-moody",4,"","",0,"2026-06-14",""],[7,3617,"Unbounded pickle-VM working set in `candle-core/src/pickle.rs` (DoS via crafted `.pth`)","PCfVW",4,"","",0,"2026-06-13",""],[7,3605,"candle-flash-attn-v3 fails to build with CUDA 13.x (pinned CUTLASS uses unversioned PFN_cuTensorMapEncodeTiled","Xueying-VirtueAI",4,"","",0,"2026-06-11",""],[7,3604,"candle-core CUDA backend: LaunchConfig::for_num_elems(el as u32) silently truncates; ops fail on tensors with ","Xueying-VirtueAI",4,"","",0,"2026-06-11",""],[7,3603,"candle-flash-attn-v3: ragged causal varlen (different seqlens in one batch) hangs in the SingleTileScheduler m","Xueying-VirtueAI",4,"","",0,"2026-06-11",""],[7,3582,"Qwen3: causal_mask shaped (b, 1, tgt, tgt) from a b-independent buffer → wrong output for batch size > 1","jeffmichaeljohnson-tech",0,"pjdurden","#3586",0,"2026-06-05","2026-07-30"],[7,3580,"Add the gpt-oss model (OpenAI open-weights MoE) to candle-transformers","toddwbucy",4,"","",0,"2026-06-04",""],[7,3569,"candle-nn: ops::softmax_last_dim is apply_op1_no_bwd; severs autograd (same pattern as #2168 / PR #3526)","nxrobins",4,"","",0,"2026-05-28",""],[7,3568,"candle-nn: rotary_emb::rope is apply_op3_no_bwd; severs autograd (same pattern as #2168 / PR #3526)","nxrobins",4,"","",0,"2026-05-28",""],[7,3567,"Qwen2: prepare_attention_mask is bidirectional (no causal triangle) when attn_mask provided","nxrobins",4,"","",0,"2026-05-28",""],[7,3566,"MetalDevice::new panics on empty MTLCopyAllDevices() instead of returning Err","Yahook",4,"","",0,"2026-05-27",""],[7,3552,"`panic!` when running on tvOS simulator with `mistral.rs`","setoelkahfi",4,"","",0,"2026-05-20",""],[7,3541,"16GB mem unable to build candle anymore","lucasjinreal",4,"","",0,"2026-05-18",""],[7,3540,"16GB mem unable to build candle anymore","lucasjinreal",4,"","",0,"2026-05-18",""],[7,3534,"Tensor::from_slice / from_raw_buffer return Ok when the declared shape does not match the storage element coun","UNILESS",4,"","",0,"2026-05-15",""],[7,3533,"Unguarded allocation sites in candle-core/src/quantized/gguf_file.rs (DoS via crafted GGUF)","UNILESS",2,"","",0,"2026-05-14","2026-06-06"],[7,3530,"Candle leaves ~ model size of memory unused on Spark","sempervictus",4,"","",0,"2026-05-11",""],[7,3527,"Cannot load tokenizer from gguf file","wangyang1749",4,"","",0,"2026-05-09",""],[7,3524,"Cross-CUDA Tensor::to_device fails with CUDA_ERROR_INVALID_CONTEXT (no peer-access enable)","toddwbucy",4,"","",0,"2026-05-08",""],[7,3523,"Phi4 mini multiple runs","dimitrov-anasoft",4,"","",0,"2026-05-08",""],[7,3522,"Metal SDPA panics during quantized LLaMA GQA decode","athola",4,"","",0,"2026-05-07",""],[7,3515,"candle-flash-attn vendored kernels ~13 months stale; long-context bf16 divergence vs PyTorch flash_attn 2.8.3","toddwbucy",4,"","",0,"2026-05-06",""],[7,3514,"Contribute Qwen3.5/3.6 35B-A3B port + 3 fused perf kernels — packaging guidance?","lukeme117",4,"","",0,"2026-05-06",""],[7,3505,"Qwen3 VL forward pass doesn't construct attention mask properly","raka-gunarto",4,"","",0,"2026-05-01",""],[7,3497,"The fact that Candle supports Metal is not mentioned in the mdbook (installation)","Xuanmizhen",4,"","",0,"2026-04-27",""],[7,3492,"Gemma 4 Textmodel doesn't work.","AlpineVibrations",4,"","",0,"2026-04-25",""],[7,3491,"[candle-onnx] simple_eval is hardcoded to CPU, preventing CUDA/GPU acceleration","UU-0665D27",4,"","",0,"2026-04-23",""],[7,3473,"missing <think> and </think> tokens on qwen models","Qu1ntev",2,"","",0,"2026-04-18","2026-10-01"],[7,3464,"Metal buffer pool grows beyond physical RAM on small unified-memory devices","cjchanh",4,"","",0,"2026-04-13",""],[7,3460,"the link to rust trunk is wrong","andrewchen5678",4,"","",0,"2026-04-11",""],[7,3455,"whisper example gives an error","vandana-rajan",4,"","",0,"2026-04-08",""],[7,3448,"Gemma 4 Model download issues","AlpineVibrations",4,"","",0,"2026-04-04",""],[7,3434,"Add turboquant support","lucasjinreal",4,"","",0,"2026-04-01",""],[7,3430,"SafeTensors loading produces corrupted weights on big-endian architectures (s390x)","johannes-adr",4,"","",0,"2026-03-31",""],[7,3425,"Implement new TurboQuant research","zbloss",4,"","",0,"2026-03-29",""],[7,3410,"quantized_llama.rs uses wrong RoPE convention for NEOX-style models (Qwen2, Falcon, Phi, etc.)","joelteply",1,"joelteply","#3411",0,"2026-03-20","2026-04-15"],[7,3404,"conv2d_tiled produces incorrect results compared to conv2d_im2col_gemm (CPU backend)","developer0hye",1,"developer0hye","#3405",0,"2026-03-16","2026-03-22"],[7,3401,"Adding support for LightOnOCR_2_1B","Abhinav5132",4,"","",0,"2026-03-13",""],[7,3399,"Consider merging `cuda` and `cudnn` features","slckl",4,"","",0,"2026-03-08",""],[7,3398,"Does this framework support AMD Ryzen™ AI?","luozhonghe",2,"","",0,"2026-03-07","2026-03-08"],[7,3393,"Qwen3.5 Support","lucasjinreal",4,"","",0,"2026-03-03",""],[7,3390,"ACE-Step-1.5 support !","IllusionPerdu",4,"","",0,"2026-03-01",""],[7,3389,"conv1d with large groups (depthwise) is extremely slow on CUDA — O(groups) kernel launches","m96-chan",4,"","",0,"2026-02-27",""],[7,3388,"Metal SDPA kernel produces NaN with non-square mask + GQA","starpit",4,"","",0,"2026-02-26",""],[7,3369,"Subject: Contribution for Metal Device Initialization Fix in Candle","PlutoEternity",4,"","",0,"2026-02-15",""],[7,3366,"Candle Kernels tries to build wmma kernels on unsupported hardware.","asglover",1,"asglover","#3367",0,"2026-02-12","2026-02-15"],[7,3358,"Error: Get error \"forward compatibility was attempted on non supported HW\" on RTX 5090.","tjliupeng",4,"","",0,"2026-02-03",""],[7,3357,"Metal: Missing kernel for `index_select` with `u32` indices and `i64` source","TimmyOVO",0,"Elvis339","#3371",0,"2026-02-02","2026-02-19"],[7,3356,"Metal: `arg_sort_last_dim` is unstable and inconsistent with CPU backend","TimmyOVO",2,"","",0,"2026-02-02","2026-02-03"],[7,3353,"linear_split in candle-metal-kernels cannot handle empty tensors","SKRohit",4,"","",0,"2026-02-01",""],[7,3351,"Questions about WASM memory for Quant-Qwen3","vandana-rajan",2,"","",0,"2026-01-31","2026-02-03"],[7,3348,"PaddleOCR-VL consider support bf16 or f16 inference","lucasjinreal",4,"","",0,"2026-01-30",""],[7,3345,"Qwen3-ASR support !","lucasjinreal",4,"","",0,"2026-01-30",""],[7,3344,"Feature request: AMD GPU support for Candle embedding/inference (needed for semantic-router on AMD + mac)","Xunzhuo",2,"","",0,"2026-01-30","2026-01-30"],[7,3343,"Multiple versions of the same dependencies","bridiver",4,"","",0,"2026-01-28",""],[7,3339,"tensor upsample_cubic support","lucasjinreal",4,"","",0,"2026-01-27",""],[7,3337,"O_DIRECT when reading weight files","sempervictus",4,"","",0,"2026-01-26",""],[7,3336,"Consider using candle built-in interpolate op for image resize","lucasjinreal",4,"","",0,"2026-01-26",""],[7,3333,"DType enum changes in 0.9.2 patch release violated semver - please add #[non_exhaustive]","antimora",0,"eyupcanakman","58694bb",0,"2026-01-25","2026-03-30"],[7,3331,"Unable To Build for Nvidia GeForce GTX 1650","zicklag",4,"","",0,"2026-01-24",""],[7,3323,"Add support for Qwen3-TTS","lucasjinreal",4,"","",0,"2026-01-23",""],[7,3322,"when will paddleocrvl pushed to crates.io?","lucasjinreal",2,"","",0,"2026-01-22","2026-01-24"],[7,3314,"Is there any plan to support LFM 2.5 Audio , Text and Vision Models?","SKRohit",4,"","",0,"2026-01-19",""],[7,3307,"SAM Metal error when using --generate-masks","HashbangH",4,"","",0,"2026-01-16",""],[7,3305,"Save state during forward in CustomOp","benjamin-lieser",4,"","",0,"2026-01-14",""],[7,3302,"Metal GEMM Performance Issue: Fixed Tile Configuration","SpenserCai",1,"SpenserCai","#3313",0,"2026-01-14","2026-01-21"],[7,3299,"Potential bug in gemma3 - mask always -infinity for kv cache","vandana-rajan",2,"","",0,"2026-01-12","2026-01-29"],[7,3297,"Error: cannot seed the CPU rng with set_seed","phil-el",4,"","",0,"2026-01-10",""],[7,3296,"[README] Add Oxide Lab to “Useful External Resources” (LM Studio-like app powered by Candle)","FerrisMind",3,"","",0,"2026-01-10","2026-02-06"],[7,3295,"candle-video: Rust-native text-to-video (LTX-Video) on Candle — maintainer feedback + where to list it?","FerrisMind",2,"","",0,"2026-01-09","2026-02-10"],[7,3294,"[RFC] Candelabra: JAX-style tracing and compilation for Candle","DrJesseGlass",2,"","",0,"2026-01-09","2026-01-24"],[7,3291,"bug: cargo run --example qwen,Execution error - model.safetensors file not found when specifying local --weigh","caowenbo",2,"","",0,"2026-01-09","2026-01-28"],[7,3290,"`candle-onnx` support for `DynamicQuantizeLinear`","ZichengLiang",4,"","",0,"2026-01-09",""],[7,3288,"LTX-2 support in Candle","ogghead",4,"","",0,"2026-01-07",""],[7,3286,"Add unit tests to candle-transformers","Anri-Lombard",4,"","",0,"2026-01-07",""],[7,3277,"fatal error LNK1181: cannot open input file 'stdc++.lib'","forwardwu",0,"Elvis339","#3278",0,"2026-01-02","2026-01-03"],[7,3276,"Proposal to add an optional `iex` dpendency","DanikVitek",4,"","",0,"2025-12-31",""],[7,3272,"Added support for Vulkan, any interest?","davidwynter",4,"","",0,"2025-12-30",""],[7,3270,"Add inverse trigonometric operations","Trivaxy",4,"","",0,"2025-12-28",""],[7,3259,"dep: `ug-cuda v0.5.0` uses `cudarc v0.17.8` which conflicts with a separate dep of `cudarc v0.18.1` and doesn'","DanikVitek",1,"DanikVitek","#3268",0,"2025-12-23","2025-12-31"],[7,3258,"Add xLSTM model support","Anri-Lombard",4,"","",0,"2025-12-23",""],[7,3257,"Add Mamba2 model support","Anri-Lombard",4,"","",0,"2025-12-21",""],[7,3253,"broadcast_matmul to handle stride-0 broadcast dimensions","DrJesseGlass",4,"","",0,"2025-12-20",""],[7,3249,"Cannot compile under Windows with CUDA 13 and an RTX 5060.","the-pawn-2017",4,"","",0,"2025-12-18",""],[7,3247,"Parakeet V3 support?","mobicham",4,"","",0,"2025-12-16",""],[7,3240,"GLM ASR mode support","lucasjinreal",4,"","",0,"2025-12-10",""],[7,3227,"In-place ops deadlock when operands share the same storage","zackangelo",4,"","",0,"2025-12-05",""],[7,3222,"Please revert PR #3219","sdabbour",2,"","",0,"2025-12-03","2025-12-03"],[7,3215,"cannot find gemma3.attention.head_count in metadata","daviscai",0,"clocksmith","#3274",0,"2025-11-26","2025-12-30"],[7,3211,"Standardize Interfaces and Plumbing","sempervictus",4,"","",0,"2025-11-24",""],[7,3210,"`DriverError(CUDA_ERROR_INVALID_PTX, \"a PTX JIT compilation failed\")` occurs","mayocream",2,"","",0,"2025-11-23","2025-11-23"],[7,3208,"`cudarc` dynamic loading support","mayocream",4,"","",0,"2025-11-22",""],[7,3206,"SAM3 support","iamxtso",4,"","",0,"2025-11-21",""],[7,3203,"The visibility modifier `pub(crate)` is preventing me from using `Config` while constructing `Qwen3VLModel::ne","amritsingh183",2,"","",0,"2025-11-20","2025-11-20"],[7,3199,"CUDA integer reductions use undefined initialization (±INFINITY) causing incorrect max/min on Blackwell (sm_10","TimmyOVO",2,"","",0,"2025-11-18","2025-11-20"],[7,3198,"Error: Metal error no metal implementation for sigmoid, in candle_core/candle_nn v0.9.1","xiuno",2,"","",0,"2025-11-18","2025-11-20"],[7,3190,"PP-OCRv5 support","zhengqwe",4,"","",0,"2025-11-15",""],[7,3187,"Depth anything V3 support","lucasjinreal",4,"","",0,"2025-11-14",""],[7,3185,"Lost support for MacOS < 15.0 Sequoia","jacobgorm",0,"ivarflakstad","#3205",1,"2025-11-13","2025-12-04"],[7,3184,"Request to upgrade cudarc dependency from 0.16.6 to 0.17.8 to fix unsupported operation error","lihongtao1024",2,"","",0,"2025-11-13","2025-11-17"],[7,3183,"CPU-Optimized Kernels for Interleaved GGUF Weights (Following llama.cpp)","DrJesseGlass",4,"","",0,"2025-11-13",""],[7,3182,"Reorganize Transformers Module by Model Family","DrJesseGlass",2,"","",0,"2025-11-12","2025-12-05"],[7,3181,"Support Review of ConcatKvCache (#3143) and Plan for Future Adoption","DrJesseGlass",2,"","",0,"2025-11-12","2025-11-28"],[7,3178,"Metal QMatMul::forward panics on F16 input (regression from 0.9.1); F32 OK; forward_via_f16 OK","TimmyOVO",2,"","",0,"2025-11-11","2025-11-17"],[7,3177,"[Cuda] u32 tensor div by f64 leads to zeroed out tensor","LunaticWyrm467",4,"","",0,"2025-11-10",""],[7,3176,"Requesting ability to load model from Bytes","David-OConnor",2,"","",0,"2025-11-09","2025-11-13"],[7,3170,"Supports InstanceNormalization in onnx","lucasjinreal",4,"","",0,"2025-11-08",""],[7,3167,"Qwen 3-1.7b looks like something is wrong and doesn't stop properly.","xiuno",0,"olafurjohannsson","#3377",0,"2025-11-07","2026-02-19"],[7,3166,"Seeking Wisdom on Cuda","David-OConnor",4,"","",0,"2025-11-05",""],[7,3154,"Hello, author. Do you have any plans to support model fine-tuning","hadoop835",4,"","",0,"2025-10-29",""],[7,3153,"Unsupported `linear` `Resize` node with `candle_onnx`","universonic",4,"","",0,"2025-10-28",""],[7,3151,"Tensor conversion to_vec1() failing on 0.9.2-alpha.1 - Metal","si-harps",2,"","",0,"2025-10-27","2025-11-06"],[7,3141,"PaddleVLOCR 0.9B support","lucasjinreal",2,"","",0,"2025-10-21","2026-01-08"],[7,3138,"ModernBERT head.forward fails on Metal with Invalid matmul arguments … (CPU OK) — missing contiguous() before ","whitebox2",2,"","",0,"2025-10-19","2025-11-06"],[7,3137,"Strategic Discussion: Flicker's Hybrid Architecture for Lightweight Inference + Advanced Training","jagan-nuvai",2,"","",0,"2025-10-18","2025-10-18"],[7,3134,"No Rayon mode","XiangpengHao",4,"","",0,"2025-10-15",""],[7,3133,"Qwen3 VL support","lucasjinreal",4,"","",0,"2025-10-13",""],[7,3120,"AutoModel / PreTrainedModel equivalent magic ?","ierezell",4,"","",0,"2025-10-07",""],[7,3119,"Conv2d cpu performance is somewhat slow","slckl",4,"","",0,"2025-10-07",""],[7,3096,"[Question] Minimal documentation/example on including weights in compiled executable","bitanath",0,"matthewhaynesonline","#3104",0,"2025-09-24","2025-10-07"],[7,3095,"Support step parameter in slice_assign method","antimora",4,"","",0,"2025-09-23",""],[7,3094,"Add multilingual documentation structure for mdBook","maximizemaxwell",4,"","",0,"2025-09-23",""],[7,3087,"Add support for CUDA 13.0","Flakes342",2,"","",0,"2025-09-15","2025-10-17"],[7,3080,"Error with candle-nn on latest Metal Refactor","AlpineVibrations",0,"ivarflakstad","#3081",1,"2025-09-08","2025-09-08"],[7,3076,"Diffusers model adapters","sansmoraxz",4,"","",0,"2025-09-06",""],[7,3060,"support for  sliding_window_pattern for google/gemma-3-270m-it","baijal1982",4,"","",0,"2025-08-20",""],[7,3057,"SNAC model feature support","lucasjinreal",4,"","",0,"2025-08-15",""],[7,3056,"Failing tests on newer CUDA hardware","zackangelo",4,"","",0,"2025-08-13",""],[7,3052,"Candle vs. PyTorch performance","aiman2039",4,"","",0,"2025-08-11",""],[7,3048,"[FEAT] OpenAI OSS support","jorge-menjivar",4,"","",0,"2025-08-07",""],[7,3047,"Can the safetensor files from OpenAI's new gpt-oss-20b work with any existing setup?","zcourts",4,"","",0,"2025-08-06",""],[7,3044,"RWKV models not sane","sempervictus",4,"","",0,"2025-08-04",""],[7,3043,"Flash-attn feature fails to build in 22.04 Docker","sempervictus",2,"","",0,"2025-08-03","2025-08-03"],[7,3039,"Request support for Qwen2.5-vl or Fast-VLM","826327700",4,"","",0,"2025-07-31",""],[7,3037,"A pure Rust mini-language for sized vector types based on traits","stefanmohl",2,"","",0,"2025-07-27","2025-11-07"],[7,3034,"Implement Task 10","moreWax",2,"","",0,"2025-07-26","2025-08-11"],[7,3032,"`matmul` (and others) Precision issues between Candle & PyTorch","andrew-shc",2,"","",0,"2025-07-23","2025-09-27"],[7,3031,"Docs enhancement","maximizemaxwell",4,"","",0,"2025-07-22",""],[7,3029,"Proposal: Add CoreML support to Candle (for Apple Neural Engine inference) + offer to co-maintain","mazhewitt",2,"","",0,"2025-07-20","2025-07-24"],[7,3028,"Feature Request: Voxtral","jorge-menjivar",1,"jorge-menjivar","#3036",0,"2025-07-19","2025-08-04"],[7,3026,"Qwen2.5 VL and some TTS model support","lucasjinreal",4,"","",0,"2025-07-19",""],[7,3025,"Apple silicon coreml backend","hayloftbisque",3,"","",0,"2025-07-17","2025-07-18"],[7,3024,"Model Support: nllb","hayloftbisque",4,"","",0,"2025-07-17",""],[7,3022,"candle_datasets: Improve from_hub usability by exposing FileReader and cleaning up iterator logic","xavierforge",2,"","",0,"2025-07-16","2025-10-01"],[7,3018,"MKL SGEMM error","henrique-1",2,"","",0,"2025-07-10","2025-11-24"],[7,3017,"candle_nn: LayerNorm does not expose `remove_mean` and `eps` parameter","nemosupremo",0,"jgremes","#3387",0,"2025-07-07","2026-03-29"],[7,3016,"Build fails on Maxwell GPU due to __dp4a undefined in quantized.cu","fishonamos",4,"","",0,"2025-07-07",""],[7,3015,"Unable to build candle with flash attention on iOS","jpchen",4,"","",0,"2025-07-06",""],[7,3014,"Short context length on Qwen quantized examples.","AlpineVibrations",4,"","",0,"2025-07-02",""],[7,3013,"Word Timestamp for whisper","bp7968h",4,"","",0,"2025-06-29",""],[7,3011,"LayerNorm Gradient Flow Issue in candle-nn","tymat",4,"","",0,"2025-06-28",""],[7,3009,"[ Gemma 3n ]  -  multi modal - small enough to work on a mobile","oddpxl",2,"","",0,"2025-06-27","2025-06-30"],[7,3004,"qwen3 q,k tensors not contiguous after rms norm","zackangelo",2,"","",0,"2025-06-24","2025-06-25"],[7,3003,"Build for multiple arch?","johnnynunez",4,"","",0,"2025-06-23",""],[7,3002,"Is there a roadmap or intention to support CUDA Graph?","guoqingbao",4,"","",0,"2025-06-23",""],[7,3001,"Unsupported ONNX operator: DequantizeLinear","adamreichold",4,"","",0,"2025-06-22",""],[7,3000,"Inconsistent last_hidden_state slicing position compared to HuggingFace Transformers in ModernBertForSequenceC","whitebox2",2,"","",0,"2025-06-22","2025-10-16"],[7,2999,"Performance regression after upgrade to cudarc 0.14","zackangelo",2,"","",0,"2025-06-20","2025-06-20"],[7,2997,"Implement Conv3D support for compatibility with Qwen-VL and similar models","maximizemaxwell",4,"","",0,"2025-06-19",""],[7,2990,"question about VarBuilder:from_pth","jingangdidi",4,"","",0,"2025-06-12",""],[7,2986,"How to reset gradient before each batch","lokxii",4,"","",0,"2025-06-03",""],[7,2985,"candle_onnx compatible object detection models","magi-2",4,"","",0,"2025-06-02",""],[7,2977,"No backward pass for `LayerNorm` if tensor is contiguous and remove_mean is true","toolness",4,"","",0,"2025-06-01",""],[8,8956,"[3d datasets]: repos with gltf textures get detected as images","cfahlgren1",4,"","",0,"2026-11-04",""],[8,8949,"imagefolder metadata.jsonl: a list starting with null loads as wrong values","shashank11yadav",4,"","",0,"2026-11-03",""],[8,8926,"JSON loader turns string values into numbers, booleans or null when a field has mixed types","behroozazarkhalili",4,"","",0,"2026-11-02",""],[8,8919,"Batched map / Dataset.batch with num_proc>1 splits batches at shard boundaries and drop_last_batch loses rows","udsy19",4,"","",0,"2026-10-31",""],[8,8914,"Json.cast_storage only inspects the first 5 values, so plain strings past that point are stored unencoded and ","Steve-20-ai",4,"","",0,"2026-10-31",""],[8,8912,"Empty typed Dataset default save cannot reload because it writes zero shards","alberttwong",4,"","",0,"2026-10-31",""],[8,8909,"Docs: missing `import re` in \"Use with Polars\" data extraction example","docsalive-team",4,"","",0,"2026-10-30",""],[8,8907,"xnumpy_load, xpandas_read_excel, xpyarrow_parquet_read_table and xsio_loadmat never close their xopen handle","michaelkillgta",4,"","",0,"2026-10-29",""],[8,8898,"JSON loading retries indefinitely when a mixed-type field name contains a slash","bbbbbbbbbbb114",4,"","",0,"2026-10-27",""],[8,8888,"Fixed-shape ArrayXD pandas round trips lose row nulls and integer precision","sjdoane",4,"","",0,"2026-10-23",""],[8,8877,"Dataset.from_sql caches different SQLAlchemy queries under the same key when bound values change","gorkadabo1",4,"","",0,"2026-10-22",""],[8,8863,"IterableDataset.decode(False) has no effect after shard() on a streamed media dataset","RaphaelFakhri",4,"","",0,"2026-10-20",""],[8,8859,"concatenate_datasets reuses preprocessing caches across different axes","gorkadabo1",1,"gorkadabo1","#8865",0,"2026-10-20","2026-10-23"],[8,8849,"cast_column / cast are ignored on streaming datasets decoded in threads (audio sampling_rate, decode=False)","sebxt",4,"","",0,"2026-10-18",""],[8,8838,"to_sql with num_proc > 1 fails with 'table already exists' when a later batch reaches the database first","mertkayacs",4,"","",0,"2026-10-17",""],[8,8836,"Indexing a Dataset with a list or array of bools silently returns rows 0 and 1","mertkayacs",4,"","",0,"2026-10-17",""],[8,8832,"to_csv ignores the encoding argument and always writes UTF-8","mertkayacs",4,"","",0,"2026-10-17",""],[8,8831,"to_csv and to_sql with index=True restart the index at 0 in every batch","mertkayacs",4,"","",0,"2026-10-17",""],[8,8830,"ClassLabel.cast_storage fails on a pandas Categorical column (dictionary-encoded Arrow storage)","mertkayacs",4,"","",0,"2026-10-17",""],[8,8799,"concatenate_datasets drops the split_dataset_by_node split, so every rank gets all the data","hzmsh",1,"hzmsh","#8800",0,"2026-10-14","2026-10-21"],[8,8793,"Fatal `PyGILState_Release` at interpreter exit when streaming Parquet with pyarrow 25 (arrow#45214 workaround ","javi22020",4,"","",0,"2026-10-13",""],[8,8789,"re-pushing all dataset partitions corrupts it","belambert",4,"","",0,"2026-10-12",""],[8,8784,"CSV builder never closes the xopen file handle passed to pandas.read_csv (streaming and non-streaming)","chrikrah",1,"chrikrah","#8786",0,"2026-10-11","2026-10-15"],[8,8776,"Loading a dataset with many columns is quadratic in the number of columns (20k-column CSV: 15-25 min to load, ","novain-shibu",1,"novain-shibu","#8777",0,"2026-10-09","2026-10-15"],[8,8766,"Dataset.add_item fails when an existing struct-backed feature meets an inferred dictionary","LiamLin1130",4,"","",0,"2026-10-07",""],[8,8763,"Dataset.shard raises IndexError for the empty shards when num_shards is larger than the dataset","Mudassiruddin7",4,"","",0,"2026-10-07",""],[8,8755,"Audio array remains NumPy after Torch formatting","abellion",4,"","",0,"2026-10-06",""],[8,8743,"Parquet export does not close its writer after a batch or metadata failure","nawaaaaaAaar",4,"","",0,"2026-10-05",""],[8,8740,"Parquet file with an empty row group fails to load: `ArrowInvalid: BatchSize must be greater than 0`","Sriram-PR",4,"","",0,"2026-10-05",""],[8,8724,"IterableDataset.filter fails with arrow, pandas and polars formats","felipebridge",1,"felipebridge","#8725",0,"2026-10-02","2026-10-05"],[8,8719,"concatenate_datasets(axis=1) crashes on arrow-backed IterableDatasets with different lengths","Zhuoxi2000",4,"","",0,"2026-10-01",""],[8,8714,"Casting mixed JSON and plain strings to Json depends on row order and batch size","Marcuswang0824",4,"","",0,"2026-09-30",""],[8,8713,"JSON loader turns strings like \"123\" or \"null\" into other types in inferred Json columns","u7k4rs6",4,"","",0,"2026-09-30",""],[8,8709,"string_to_dict maps fields to unrelated regex capture groups","Afloat16",4,"","",0,"2026-09-29",""],[8,8706,"IterableDataset skip/take fails when iterating a nonzero epoch","galleonli",1,"galleonli","#8707",0,"2026-09-28","2026-10-05"],[8,8705,"Skip constructing disabled `hf_tqdm` progress bars in `map_nested` and `_single_map_nested`","zhixiangli",4,"","",0,"2026-09-28",""],[8,8704,"Avoid repeated `np.lib.NumpyVersion(np.__version__)` parsing per column/row in `NumpyArrowExtractor._arrow_arr","zhixiangli",4,"","",0,"2026-09-28",""],[8,8702,"Windows: `load_dataset` fails with WinError 206 for moderately deep `cache_dir` because the lock filename dupl","825pranav",4,"","",0,"2026-09-27",""],[8,8699,"Dataset.from_pandas and pandas-formatted map fail on ArrayXD columns: PandasArrayExtensionArray has no __arrow","JoeyTan21",1,"JoeyTan21","#8701",0,"2026-09-27","2026-10-05"],[8,8681,"Casting `large_string` / `string_view` to ClassLabel fails, so `Dataset.from_pandas(df, features=ClassLabel)` ","lucaluo925",4,"","",0,"2026-09-24",""],[8,8675,"load_dataset(\"json\") fails on mixed-shape JSONL when a chunksize boundary separates the shapes","aksakalmustafa",4,"","",0,"2026-09-23",""],[8,8671,"push_to_hub() / to_parquet() leave nested list/struct columns uncompressed","aksakalmustafa",0,"jayzuccarelli","#8674",0,"2026-09-23","2026-10-05"],[8,8669,"`IterableDataset.shuffle()` in 5.0 reports `num_shards = min(files per source)`, leaving DataLoader workers an","rwightman",0,"yuchenwang3","#8670",0,"2026-09-22","2026-10-05"],[8,8663,"IterableDataset numpy format returns object instead of the declared dtype","Yonghui-Lee",1,"Yonghui-Lee","#8665",0,"2026-09-22","2026-10-05"],[8,8659,"TranslationVariableLanguages crashes on empty translations with declared languages","HuaTNA",4,"","",0,"2026-09-21",""],[8,8653,"Long lock filenames do not provide mutual exclusion across Python processes","Yang1107-wzy",4,"","",0,"2026-09-21",""],[8,8637,"FileLock wrapper drops timeout and every other constructor argument","behroozazarkhalili",4,"","",0,"2026-09-17",""],[8,8626,"Probabilistic interleave without replacement samples empty datasets","HaokaiDing",4,"","",0,"2026-09-15",""],[8,8625,"CSOAI — Datasets interop: data governance","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[8,8624,"CSOAI — Datasets interop: data governance","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[8,8623,"CSOAI — HuggingFace Datasets AI governance","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[8,8618,"`DataFilesList.filter()` leaves `origin_metadata` misaligned with the filtered files","betacatsling",4,"","",0,"2026-09-14",""],[8,8617,"`SplitDict.copy()` crashes when `dataset_name` is not set","betacatsling",4,"","",0,"2026-09-14",""],[8,8612,"load_dataset on a dataset name with no namespace raises HfUriError","ebarkhordar",4,"","",0,"2026-09-12",""],[8,8608,"`push_to_hub()` keeps stale `dataset_info.features` when re-pushing a single split with different columns","belambert",2,"","",0,"2026-09-12","2026-10-12"],[8,8596,"load_dataset silently accepts out-of-range ClassLabel indices when the source dtype already matches","caiotheodoro",4,"","",0,"2026-09-10",""],[8,8595,"IterableDataset.decode(num_threads) busy-polls its ThreadPool futures, starving the decode threads — 0 rows on","HaukurPall",2,"","",0,"2026-09-10","2026-10-15"],[8,8590,"record_checksums default changed True -> False in 4.8.5, not in the release notes","arian-gogani",4,"","",0,"2026-09-08",""],[8,8572,"concatenate_datasets(axis=1) drops the columns of a source with no rows instead of raising","caiotheodoro",4,"","",0,"2026-09-06",""],[8,8570,"[zh] Add Simplified Chinese translation for README","JasonYeYuhe",4,"","",0,"2026-09-06",""],[8,8568,"skip() after split_dataset_by_node gives every rank the whole dataset","ebarkhordar",1,"ebarkhordar","#8569",0,"2026-09-06","2026-09-10"],[8,8560,"Bump fsspec to 2026.7.0","roshanjrajan",1,"roshanjrajan","#8561",0,"2026-09-03","2026-09-10"],[8,8543,"xPath.glob passes double-wrapped storage options to url_to_fs","ebarkhordar",4,"","",0,"2026-09-01",""],[8,8537,"JPEG XL support","StableLlama",4,"","",0,"2026-08-30",""],[8,8522,"Docs proposal: relabel left/right segmentation classes during horizontal flips","ternaus",4,"","",0,"2026-08-28",""],[8,8511,"`Image.cast_storage` warns about downcasting int64 to uint8 for storage that is already uint8","albertvillanova",4,"","",0,"2026-08-24",""],[8,8507,"ArrayXD silently reshapes inputs with incompatible nested dimensions","oroikono",4,"","",0,"2026-08-23",""],[8,8505,"Request to correct language name for ISO code 'blk' from 'Pa'o Karen' to 'Pa'O'","KhunPhanDuae",4,"","",0,"2026-08-23",""],[8,8500,"with_format(\"numpy\") returns bool/datetime values as 0-d unhashable arrays and casts duration columns to int64","2sumtech",4,"","",0,"2026-08-19",""],[8,8499,"TsFile format filter uses an inconsistent URL and is not available initially","ColinLeeo",4,"","",0,"2026-08-19",""],[8,8494,"cast_column to ClassLabel silently accepts out-of-range label indices","shoutoutuoadi325",4,"","",0,"2026-08-18",""],[8,8491,"`Dataset.map(num_proc=N)` worker crashes with `ValueError: I/O operation on closed file` when `finalize()` is ","HowardZorn",4,"","",0,"2026-08-17",""],[8,8483,"Dataset.take(n) raises IndexError when n > len(dataset), but IterableDataset.take(n) returns what it has","shashvat-singham",4,"","",0,"2026-08-15",""],[8,8481,"ClassLabel.str2int and int2str do not validate the elements of an iterable","shashvat-singham",4,"","",0,"2026-08-15",""],[8,8475,"`Dataset.select` raises a raw OverflowError on in-range negative indices that its own validation allows","codeAnqiang-ma",4,"","",0,"2026-08-13",""],[8,8473,"Negative percent split boundary silently anchors to the start when it rounds to 0 (`train[-1%:]` returns the w","codeAnqiang-ma",4,"","",0,"2026-08-13",""],[8,8472,"batch(0) silently returns one batch with the whole dataset instead of raising","shashvat-singham",4,"","",0,"2026-08-12",""],[8,8469,"Empty column loses its dtype under numpy format (always float32)","shashvat-singham",4,"","",0,"2026-08-12",""],[8,8467,"PandasArrayExtensionDtype has no construct_from_string, so comparing the dtype to any string raises AssertionE","shashvat-singham",4,"","",0,"2026-08-12",""],[8,8465,"PandasArrayExtensionArray.take coerces fill_value before checking whether anything is filled, breaking boolean","shashvat-singham",4,"","",0,"2026-08-12",""],[8,8463,"very bad quality images in most of them","anis-mselmi",4,"","",0,"2026-08-12",""],[8,8458,"Replace httpx with httpx2","Aegdesil",4,"","",0,"2026-08-10",""],[8,8410,"_legacy_no_dict_keys_sorting has no reader since #7817, so the 2.15.0 legacy cache lookup computes the wrong c","ebarkhordar",4,"","",0,"2026-08-06",""],[8,8394,"IterableDataset.repeat() drops the Arrow fast path and widens numpy dtypes","ebarkhordar",1,"ebarkhordar","#8395",0,"2026-08-04","2026-08-31"],[8,8393,"CPU Bottleneck in `IterableDataset.state_dict()` when shuffling multiple shards","yuxin00j",1,"yuxin00j","#8387",0,"2026-08-04","2026-09-04"],[8,8391,"Nanosecond timestamps are silently truncated even when Value(\"timestamp[ns]\") is requested","LeSingh1",4,"","",0,"2026-08-04",""],[8,8390,"to_json cannot round-trip temporal columns: all timestamps written as epoch milliseconds","LeSingh1",4,"","",0,"2026-08-04",""],[8,8386,"map(batched=True, drop_last_batch=True) returns the input unchanged when the dataset is smaller than batch_siz","LeSingh1",4,"","",0,"2026-08-04",""],[8,8381,"IterableDataset.to_dict() and to_polars() return an empty generator instead of the data","ebarkhordar",0,"cwklurks","#8632",0,"2026-08-03","2026-10-05"],[8,8375,"`PandasArrayExtensionDtype._metadata` should be a tuple, not a string","kohankhaki",4,"","",0,"2026-07-31",""],[8,8373,"TypeError in _dill.py _batch_setitems with Python 3.14","Vinello28",4,"","",0,"2026-07-29",""],[8,8365,"to_json/to_csv/to_sql cast nullable integer columns to float, corrupting values past 2**53","ebarkhordar",1,"ebarkhordar","#8366",0,"2026-07-25","2026-07-28"],[8,8362,"ArrayXD fixed-shape integer column with a null row is silently cast to float on the python read path","ebarkhordar",1,"ebarkhordar","#8363",0,"2026-07-24","2026-07-24"],[8,8359,"IterableDataset.filter() checkpoint/resume silently drops the rest of the in-flight Arrow table","ebarkhordar",1,"ebarkhordar","#8360",0,"2026-07-24","2026-07-24"],[8,8351,"with_format('numpy') promotes flat float64/int columns with nulls to dtype=object (float32 stays numeric)","ebarkhordar",1,"ebarkhordar","#8352",0,"2026-07-23","2026-07-23"],[8,8343,"Dataset.to_pandas() returns raw JSON strings for Json() columns instead of decoded objects","ebarkhordar",1,"ebarkhordar","#8344",0,"2026-07-19","2026-07-24"],[8,8341,"concatenate_datasets(axis=1) with Arrow format drops the first source's columns","ebarkhordar",1,"ebarkhordar","#8342",0,"2026-07-18","2026-07-22"],[8,8331,"datasets-server.huggingface.co returning 503 on every endpoint (whole host down, not one route)","ChristianDenniss",2,"","",0,"2026-07-14","2026-07-17"],[8,8330,"Dataset Studio and Viewer down","Taryn-Cail",2,"","",0,"2026-07-14","2026-07-24"],[8,8328,"Streaming datasets fail with 403 SignatureError: invalid key pair id from xet-bridge-us (Colab, latest librari","mahmoud-mannes",2,"","",0,"2026-07-14","2026-07-24"],[8,8327,"`Dataset.from_pandas` allocates ~1 MB per Arrow chunk, Out of Memory errors","nenadom",0,"SuryanshSS1011","#8339",0,"2026-07-13","2026-07-22"],[8,8324,"Path traversal / arbitrary file read via the `file_name` metadata field in `*folder` dataset builders (no `tru","geo-chen",0,"Kaif10","#8325",0,"2026-07-12","2026-07-23"],[8,8308,"Streaming IterableDataset state stops advancing after first resume — second resume restarts from the beginning","francesco-bertolotti",2,"","",0,"2026-07-06","2026-07-23"],[8,8302,"Turkish DIK-style Syntactic Structure Benchmark Dataset","nane100503-gif",4,"","",0,"2026-07-03",""],[8,8296,"[security] Symlink-Following Arbitrary File Write via Archive Extraction in huggingface/datasets","AAtomical",4,"","",0,"2026-06-30",""],[8,8293,"Support concatenating multiple streaming datasets while preserving the sum of shards","muyihao",4,"","",0,"2026-06-30",""],[8,8292,"SuperCompress: Cut LLM token costs by ~65% with open-source prompt compression","arjunkshah",2,"","",0,"2026-06-30","2026-06-30"],[8,8269,"MetadataConfigs drops parquet shards when exported config rows are non-consecutive","KirtiRamchandani",4,"","",0,"2026-06-14",""],[8,8261,"Feature Proposal: Add FunASR Chinese/Multilingual ASR Benchmark Datasets","LauraGPT",4,"","",0,"2026-06-12",""],[8,8256,"Dataset Viewer fails on TSFile datasets","gengziyand",4,"","",0,"2026-06-10",""],[8,8253,"Feature request: Add a streaming_shard operator for early sample-level sharding when file-level sharding is in","muyihao",0,"behroozazarkhalili","#8565",1,"2026-06-09","2026-09-09"],[8,8247,"VORTEXRAG benchmarked on 6 HuggingFace datasets — new open-source RAG framework","vignesh2027",4,"","",0,"2026-06-07",""],[8,8246,"RecursionError when loading GLB file with deeply nested JSON in Mesh feature","ghost",4,"","",0,"2026-06-07",""],[8,8242,"\"eval_strategy\": \"no\" perform evaluation","SamuelLarkin",4,"","",0,"2026-06-05",""],[8,8241,"load_dataset(\"json\", ...) infers a different schema when the JSONL file starts with a UTF-8 BOM","f-salvetti",0,"archievi","#8243",0,"2026-06-05","2026-06-09"],[8,8217,"Multi-GPU Training with Large Dataset — NCCL Timeouts / Slow Data Loading + Excessive Memory Usage with keep_i","D222097",2,"","",0,"2026-05-22","2026-06-03"],[8,8204,"Please enable Private Vulnerability Reporting","0xBassia",4,"","",0,"2026-05-16",""],[8,8201,"Encountered an error when loading data in WebDataset format using `load_datasets` during multi-matchines train","aihao2000",4,"","",0,"2026-05-15",""],[8,8199,"Replace AutoFeatureExtractor with AutoImageProcessor in image preprocessing docs","ajaystar8",0,"gautamkishore","#8326",0,"2026-05-15","2026-07-22"],[8,8197,"Feature Request / Discussion: Skip origin_metadata fetch when streaming=True?","yuxin00j",4,"","",0,"2026-05-15",""],[8,8178,"The full dataset viewer is not available (click to read why). Only showing a preview of the rows. Job manager ","kakamond",4,"","",0,"2026-05-07",""],[8,8171,"Dataset UI shows 0:00/0:00 for audio after push_to_hub","MNIKIEMA",2,"","",0,"2026-05-04","2026-05-20"],[8,8169,"Streaming dataset hangs consistently","michaelpginn",0,"lhoestq","#8176",1,"2026-05-01","2026-05-06"],[8,8167,"[BUG] Abnormal progress bar in dataset.map when load_from_cache_file=False","Jintao-Huang",0,"Nitin-Rajasekar","#8170",0,"2026-04-30","2026-05-29"],[8,8165,"IterableDataset: corrupted EXIF image silently terminates streaming iterator instead of skipping the sample","LIUYellowBlack",4,"","",0,"2026-04-30",""],[8,8164,"Lance: HF auth token dropped for private repos in streaming mode (broken `storage_options` lookup)","ericjaebeom",1,"ericjaebeom","#8166",0,"2026-04-30","2026-04-30"],[8,8159,"tests/conftest.py, tests/_test_patching.py, tests/fixtures/fsspec.py and other files are missing from the PYPI","yurivict",4,"","",0,"2026-04-28",""],[8,8149,"[Optimization] Prevent per-thread instantiation of Cloud Storage FileSystem during Data loading initialization","ankitaluthra1",4,"","",0,"2026-04-23",""],[8,8145,"`.map()` on a streaming IterableDataset silently skips examples after load_state_dict","MonliH",0,"Brianzhengca","#8147",0,"2026-04-22","2026-04-27"],[8,8131,"datasets.to_list: Behaviour change in version 4.7.0 with the addition of JSON type","ItsTania",1,"ItsTania","#8137",0,"2026-04-13","2026-04-27"],[8,8129,"`load_dataset()` hangs when `hf_xet` is enabled","lewtun",2,"","",0,"2026-04-10","2026-04-10"],[8,8125,"Feature request: being able to disable chmod for flat permission filesystems","younesbelkada",4,"","",0,"2026-04-08",""],[8,8120,"Patch fields truncated — missing trailing newlines","hetao-git",2,"","",0,"2026-04-03","2026-04-30"],[8,8116,"`to_json()` serializes `Json()` feature as escaped string instead of native JSON object","ain-soph",0,"Brianzhengca","6254baf",0,"2026-04-02","2026-04-10"],[8,8110,"`set_format` resets features","plutonium-239",2,"","",0,"2026-03-31","2026-05-11"],[8,8085,"ImportError: cannot import name 'VideoReader' from 'torchvision.io' when using datasets with torch format","DamJanusz",0,"lhoestq","#8087",1,"2026-03-23","2026-03-23"],[8,8079,"StepExamplesIterable._iter_arrow drops samples for non-zero ranks for single-sharded","wwwjn",0,"lhoestq","#8081",1,"2026-03-19","2026-03-19"],[8,8078,"Inconsistent loading of LFS-hosted files in Dangindev/viet-cultural-vqa dataset","ngayngo9x",2,"","",0,"2026-03-17","2026-03-30"],[8,8075,"`.batch()` error on formatted datasets","mariosasko",0,"bluehyena","#8126",0,"2026-03-17","2026-04-10"],[8,8065,"ARM audio support broken","leondz",4,"","",0,"2026-03-13",""],[8,8052,"[Security] Zip/Rar/7z Archive Extraction Path Traversal (Incomplete CVE-2007-4559 Mitigation)","spartan8806",2,"","",0,"2026-03-08","2026-03-10"],[8,8048,"Support 3D Datasets","Vinay-Umrethe",2,"","",0,"2026-03-06","2026-06-02"],[8,8046,"`Dataset.map()` with `num_proc > 1` fails when some shards produce `List(Value('null'))` (empty lists)","ain-soph",1,"ain-soph","#8047",0,"2026-03-05","2026-03-09"],[8,8038,"Typo in iterable_dataset.py","sybaik1",0,"omkar-334","#8049",0,"2026-03-04","2026-03-09"],[8,8037,"IterableDataset.filter() chaining fails due to features being lost when is_typed=True","sh0416",2,"","",0,"2026-03-02","2026-03-09"],[8,8034,"[BUG] load_datasets() cannot use the generated Arrow cache correctly","Nexround",4,"","",0,"2026-02-28",""],[8,8029,"`cache_dir` option in `download_config` in `load_dataset` is not respected","TsXor",4,"","",0,"2026-02-27",""],[8,8015,"Iterating a streaming dataset correctly in random order","unrealwill",4,"","",0,"2026-02-22",""],[8,8012,"HC3 dataset fails to load with datasets>=3 due to legacy script (HC3.py)","navan0",2,"","",0,"2026-02-19","2026-02-20"],[8,8010,"audiofolder / folder-based loader rejects metadata.csv when file_name column is inferred as large_string","Koosh0610",2,"","",0,"2026-02-19","2026-02-19"],[8,8007,"Add option for loading audio with video","Samoed",4,"","",0,"2026-02-16",""],[8,8006,"Regression: from_generator crashes if one generator call returns no results","hartmans",0,"lhoestq","#8023",1,"2026-02-15","2026-02-24"],[8,8005,"Multi-channel audio is automatically cast to mono, num_channels is ignored","ZackHodari",4,"","",0,"2026-02-14",""],[8,8002,"The `Sequence` class in features do not have \"dtype\"","gonzalo-santamaria-iic",4,"","",0,"2026-02-13",""],[8,7999,"Too many dataloader workers: 4 (max is dataset.num_shards=3). Stopping 1 dataloader workers.","D222097",4,"","",0,"2026-02-09",""],[8,7998,"[doc] Inconsistant ENV VAR Name for Progress Bar Toggle","Moenupa",0,"AnkitAhlawat7742","#8000",0,"2026-02-08","2026-02-16"],[8,7994,"Bump fsspec upper bound constraint","hadim",0,"jayzuccarelli","#7995",0,"2026-02-06","2026-02-16"],[8,7991,"list(api.list_datasets()) giving jsondecode error","Moll-j",2,"","",0,"2026-02-04","2026-02-05"],[8,7990,"Dataset.map crashes when first examples return None and later examples return dict — writer not initialized","meta-program",4,"","",0,"2026-02-04",""],[8,7988,"`Dataset.map()` breaks when `function` calls `import polars as pl` and `num_proc`>1: \"UnboundLocalError: canno","ligz08",0,"lhoestq","#8024",1,"2026-02-04","2026-02-24"],[8,7986,"`Dataset.map()` causes cache miss/fingerprint change when closure captures self containing non-deterministic s","Cloud0310",4,"","",0,"2026-02-03",""],[8,7984,"Data","iLenceJhay",4,"","",0,"2026-02-03",""],[8,7970,"cast_column(..., Audio) fails with load_dataset(\"csv\",)","jstangroome",4,"","",0,"2026-01-29",""],[8,7968,"Potential conflicting type checks and dead code in `/src/datasets/table.py`","rc4typecheck",2,"","",0,"2026-01-28","2026-05-11"],[8,7965,"`huggingface_hub.errors.HfHubHTTPError: 404 Client Error: Not Found for url` when fetching a dataset with `dat","harupy",2,"","",0,"2026-01-27","2026-01-28"],[8,7958,"[CUDA Tensors Not working in ~v4.5.0] set_format(type=\"torch\", device=\"cuda\") returns cpu","ai-nikolai",2,"","",0,"2026-01-23","2026-03-04"],[8,7956,"Is the 10k files / folder limit a hard limit for a dataset repo?","pavanramkumar",4,"","",0,"2026-01-21",""],[8,7954,"all_exhausted_without_replacement working same as first_exhausted","prathamk-tw",2,"","",0,"2026-01-21","2026-03-23"],[8,7947,"MMLU get_dataset_config_names provides different lists of subsets in online and offline modes","rikrd",4,"","",0,"2026-01-16",""],[8,7946,"Question: Is there a faster way to push_to_hub for large image datasets?","adithya-s-k",4,"","",0,"2026-01-15",""],[8,7939,"datasets.load_from_disk progress bar optional manual control","Tommigun1980",4,"","",0,"2026-01-13",""],[8,7936,"_add_retries_to_file_obj_read_method makes file_obj invalid for pyarrow","li-yi-dong",0,"lhoestq","#7942",1,"2026-01-09","2026-01-13"],[8,7934,"xPath cannot handle hdfs:///xxxx properly","li-yi-dong",4,"","",0,"2026-01-08",""],[8,7931,"Enable CORS + HTTP Range support for browser partial reads on cas-bridge.xethub.hf.co (Parquet row-group acces","cornhundred",2,"","",0,"2026-01-03","2026-02-24"],[8,7928,"`load_dataset` `revision` param not respected when fetching from cache","Scott-Simmons",4,"","",0,"2026-01-02",""],[8,7927,"Using Stateful Dataloader with Split Dataset By Node and DCP for DDP","conceptofmind",4,"","",0,"2026-01-01",""],[8,7922,"Support Apache TsFile Datasets","qiaojialin",0,"JackieTien97","#8160",0,"2025-12-31","2026-06-01"],[8,7918,"datasets.load_from_disk doesn't show progress bar","Tommigun1980",0,"omarfarhoud","#7919",0,"2025-12-28","2026-01-16"],[8,7917,"IterableDataset supports automatic sharding","howitry",4,"","",0,"2025-12-27",""],[8,7916,"No description provided.","howitry",2,"","",0,"2025-12-27","2025-12-27"],[8,7915,"GDPval dataset Word docs corrupted","alexheat",4,"","",0,"2025-12-25",""],[8,7914,"[ROCm] please install 'torchcodec'","AndreasKaratzas",4,"","",0,"2025-12-24",""],[8,7911,"IndexError when saving few large examples to disk","CloseChoice",1,"CloseChoice","#7912",0,"2025-12-22","2026-01-09"],[8,7909,"Support cast_kwargs in cast_columns","Moenupa",4,"","",0,"2025-12-20",""],[8,7905,"Unbounded network usage when opening Data Studio","alizaredornica-sys",2,"","",0,"2025-12-16","2026-01-06"],[8,7904,"Request: Review pending neuroimaging PRs (#7886 BIDS loader, #7887 lazy loading)","The-Obstacle-Is-The-Way",4,"","",0,"2025-12-14",""],[8,7902,"The child process retrieves the dataset directly from the main process instead of executing `memory_mapped_arr","HQF2017",4,"","",0,"2025-12-12",""],[8,7901,"ShuffledDataSourcesArrowExamplesIterable cannot properly resume from checkpoint","howitry",4,"","",0,"2025-12-12",""],[8,7900,"`Permission denied` when sharing cache between users","qthequartermasterman",4,"","",0,"2025-12-09",""],[8,7894,"embed_table_storage crashes (SIGKILL) on sharded datasets with Sequence() nested types","The-Obstacle-Is-The-Way",4,"","",0,"2025-12-03",""],[8,7893,"push_to_hub OOM: _push_parquet_shards_to_hub accumulates all shard bytes in memory","The-Obstacle-Is-The-Way",3,"","",0,"2025-12-03","2025-12-05"],[8,7883,"Data.to_csv() cannot be recognized by pylance","xi4ngxin",0,"Aditya2755","#7888",0,"2025-11-26","2025-12-08"],[8,7882,"Inconsistent loading of LFS-hosted files in epfml/FineWeb-HQ dataset","Oligou",4,"","",0,"2025-11-26",""],[8,7880,"Spurious label column created when audiofolder/imagefolder directories match split names","neha222222",4,"","",0,"2025-11-26",""],[8,7879,"python core dump when downloading dataset","hansewetz",4,"","",0,"2025-11-24",""],[8,7877,"work around `tempfile` silently ignoring `TMPDIR` if the dir doesn't exist","stas00",0,"ada-ggf25","#7891",0,"2025-11-21","2025-12-16"],[8,7872,"IterableDataset does not use features information in to_pandas","bonext",4,"","",0,"2025-11-19",""],[8,7871,"Reqwest Error: HTTP status client error (429 Too Many Requests)","yanan1116",2,"","",0,"2025-11-19","2025-11-30"],[8,7870,"Visualization for Medical Imaging Datasets","CloseChoice",1,"CloseChoice","#7874",0,"2025-11-19","2025-11-21"],[8,7869,"Why does dataset merge fail when tools have different parameters?","hitszxs",0,"lhoestq","#8027",1,"2025-11-18","2026-03-09"],[8,7868,"Data duplication with `split_dataset_by_node` and `interleaved_dataset`","ValMystletainn",0,"lhoestq","#7992",1,"2025-11-17","2026-02-04"],[8,7867,"NonMatchingSplitsSizesError when loading partial dataset files","QingGo",4,"","",0,"2025-11-13",""],[8,7864,"add_column and add_item erroneously(?) require new_fingerprint parameter","echthesia",4,"","",0,"2025-11-13",""],[8,7863,"Support hosting lance / vortex / iceberg / zarr datasets on huggingface hub","pavanramkumar",4,"","",0,"2025-11-13",""],[8,7861,"Performance Issue: save_to_disk() 200-1200% slower due to unconditional flatten_indices()","KCKawalkar",4,"","",0,"2025-11-11",""],[8,7856,"Missing transcript column when loading a local dataset with \"audiofolder\"","gweltou",2,"","",0,"2025-11-08","2025-11-09"],[8,7852,"Problems with NifTI","CloseChoice",1,"CloseChoice","#7853",0,"2025-11-06","2025-11-06"],[8,7842,"Transform with columns parameter triggers on non-specified column access","mr-brobot",0,"lhoestq","#7843",1,"2025-11-03","2025-11-03"],[8,7841,"DOC: `mode` parameter on pdf and video features unused","CloseChoice",1,"CloseChoice","#7848",0,"2025-11-02","2025-11-05"],[8,7839,"datasets doesn't work with python 3.14","zachmoshe",2,"","",0,"2025-11-02","2025-11-04"],[8,7837,"mono parameter to the Audio feature is missing","ernestum",0,"CloseChoice","#7840",0,"2025-10-31","2025-11-03"],[8,7834,"Audio.cast_column() or Audio.decode_example() causes Colab kernel crash (std::bad_alloc)","rachidio",4,"","",0,"2025-10-27",""],[8,7832,"[DOCS][minor] TIPS paragraph not compiled in docs/stream","art-test-stack",2,"","",0,"2025-10-27","2025-10-27"],[8,7829,"Memory leak / Large memory usage with num_workers = 0 and numerous dataset within DatasetDict","raphaelsty",4,"","",0,"2025-10-24",""],[8,7821,"Building a dataset with large variable size arrays results in error ArrowInvalid: Value X too large to fit in ","kkoutini",4,"","",0,"2025-10-16",""],[8,7819,"Cannot download opus dataset","liamsun2019",4,"","",0,"2025-10-15",""],[8,7818,"train_test_split and stratify breaks with Numpy 2.0","davebulaval",0,"ArjunJagdale","#7831",0,"2025-10-15","2025-10-28"],[8,7816,"disable_progress_bar() not working as expected","windmaple",2,"","",0,"2025-10-14","2025-10-14"],[8,7813,"Caching does not work when using python3.14","intexcor",0,"sghng","#7817",0,"2025-10-10","2025-10-27"],[8,7811,"SIGSEGV when Python exits due to near null deref","iankronquist",4,"","",0,"2025-10-09",""],[8,7804,"Support scientific data formats","lhoestq",4,"","",0,"2025-10-09",""],[8,7802,"[Docs] Missing documentation for `Dataset.from_dict`","aaronshenhao",4,"","",0,"2025-10-09",""],[8,7798,"Audio dataset is not decoding on 4.1.1","thewh1teagle",4,"","",0,"2025-10-05",""],[8,7793,"Cannot load dataset, fails with nested data conversions not implemented for chunked array outputs","neevparikh",4,"","",0,"2025-09-27",""],[8,7792,"Concatenate IterableDataset instances and distribute underlying shards in a RoundRobin manner","LTMeyer",2,"","",0,"2025-09-26","2025-10-15"],[8,7788,"`Dataset.to_sql` doesn't utilize `num_proc`","tcsmaster",4,"","",0,"2025-09-24",""],[8,7780,"BIGPATENT dataset inaccessible (deprecated script loader)","ishmaifan",2,"","",0,"2025-09-18","2025-09-25"],[8,7777,"push_to_hub not overwriting but stuck in a loop when there are existing commits","Darejkal",2,"","",0,"2025-09-17","2025-09-17"],[8,7772,"Error processing scalar columns using tensorflow.","khteh",4,"","",0,"2025-09-15",""],[8,7767,"Custom `dl_manager` in `load_dataset`","ain-soph",4,"","",0,"2025-09-12",""],[8,7766,"cast columns to Image/Audio/Video with `storage_options`","ain-soph",4,"","",0,"2025-09-12",""],[8,7765,"polars dataset cannot cast column to Image/Audio/Video","ain-soph",0,"CloseChoice","#7800",0,"2025-09-12","2025-10-13"],[8,7760,"Hugging Face Hub Dataset Upload CAS Error","n-bkoe",4,"","",0,"2025-09-10",""],[8,7759,"Comment/feature request: Huggingface 502s from GHA","Scott-Simmons",4,"","",0,"2025-09-09",""],[8,7758,"Option for Anonymous Dataset link","egrace479",4,"","",0,"2025-09-08",""],[8,7757,"Add support for `.conll` file format in datasets","namesarnav",0,"CrypticCortex","#8219",0,"2025-09-06","2026-05-27"],[8,7756,"datasets.map(f, num_proc=N) hangs with N>1 when run on import","arjunguha",4,"","",0,"2025-09-05",""],[8,7753,"datasets massively slows data reads, even in memory","lrast",4,"","",0,"2025-09-04",""],[8,7751,"Dill version update","Navanit-git",4,"","",0,"2025-08-27",""],[8,7746,"Fix: Canonical 'multi_news' dataset is broken and should be updated to a Parquet version","Awesome075",4,"","",0,"2025-08-22",""],[8,7745,"Audio mono argument no longer supported, despite class documentation","jheitz",4,"","",0,"2025-08-22",""],[8,7744,"dtype: ClassLabel is not parsed correctly in `features.py`","cmatKhan",2,"","",0,"2025-08-21","2025-09-10"],[8,7742,"module 'pyarrow' has no attribute 'PyExtensionType'","mnedelko",4,"sagearc","#2",0,"2025-08-20",""],[8,7741,"Preserve tree structure when loading HDF5","klamike",1,"klamike","#7743",0,"2025-08-19","2025-08-26"],[8,7739,"Replacement of \"Sequence\" feature with \"List\" breaks backward compatibility","evmaki",4,"","",0,"2025-08-18",""],[8,7738,"Allow saving multi-dimensional ndarray with dynamic shapes","ryan-minato",4,"","",0,"2025-08-18",""],[8,7733,"Dataset Repo Paths to Locally Stored Images Not Being Appended to Image Path","dennys246",2,"","",0,"2025-08-08","2025-10-07"],[8,7732,"webdataset: key errors when `field_name` has upper case characters","YassineYousfi",4,"","",0,"2025-08-08",""],[8,7731,"Add the possibility of a backend for audio decoding","intexcor",4,"","",0,"2025-08-08",""],[8,7729,"OSError: libcudart.so.11.0: cannot open shared object file: No such file or directory","SaleemMalikAI",4,"","",0,"2025-08-07",""],[8,7728,"NonMatchingSplitsSizesError and ExpectedMoreSplitsError","efsotr",4,"","",0,"2025-08-07",""],[8,7727,"config paths that start with ./ are not valid as hf:// accessed repos, but are valid when accessed locally","doctorpangloss",4,"","",0,"2025-08-06",""],[8,7724,"Can not stepinto load_dataset.py?","micklexqg",4,"","",0,"2025-08-05",""],[8,7723,"Don't remove `trust_remote_code` arg!!!","autosquid",4,"","",0,"2025-08-04",""],[8,7722,"Out of memory even though using load_dataset(..., streaming=True)","padmalcom",4,"","",0,"2025-08-04",""],[8,7721,"Bad split error message when using percentages","padmalcom",4,"","",0,"2025-08-04",""],[8,7720,"Datasets 4.0 map function causing column not found","Darejkal",4,"","",0,"2025-08-03",""],[8,7719,"Specify dataset columns types in typehint","Samoed",4,"","",0,"2025-08-02",""],[8,7717,"Cached dataset is not used when explicitly passing the cache_dir parameter","padmalcom",4,"","",0,"2025-08-01",""],[8,7709,"Release 4.0.0 breaks usage patterns of with_format","wittenator",2,"","",0,"2025-07-30","2025-08-07"],[8,7707,"load_dataset() in 4.0.0 failed when decoding audio","jiqing-feng",2,"","",0,"2025-07-29","2025-08-01"],[8,7705,"Can Not read installed dataset in dataset.load(.)","HuangChiEn",4,"","",0,"2025-07-28",""],[8,7703,"[Docs] map() example uses undefined `tokenizer` — causes NameError","Sanjaykumar030",4,"","",0,"2025-07-26",""],[8,7700,"[doc] map.num_proc needs clarification","sfc-gh-sbekman",4,"","",0,"2025-07-25",""],[8,7699,"Broken link in documentation for \"Create a video dataset\"","cleong110",4,"","",0,"2025-07-24",""],[8,7698,"NotImplementedError when using streaming=True in Google Colab environment","Aniket17200",4,"","",0,"2025-07-23",""],[8,7697,"-","ghost",2,"","",0,"2025-07-23","2025-07-25"],[8,7696,"load_dataset() in 4.0.0 returns different audio samples compared to earlier versions breaking reproducibility","Manalelaidouni",2,"","",0,"2025-07-22","2025-07-30"],[8,7694,"Dataset.to_json consumes excessive memory, appears to not be a streaming operation","ycq0125",4,"","",0,"2025-07-21",""],[8,7693,"Dataset scripts are no longer supported, but found superb.py","edwinzajac",4,"","",0,"2025-07-20",""],[8,7692,"xopen: invalid start byte for streaming dataset with trust_remote_code=True","sedol1339",4,"","",0,"2025-07-20",""],[8,7691,"Large WebDataset: pyarrow.lib.ArrowCapacityError on load() even with streaming","cleong110",4,"","",0,"2025-07-19",""],[8,7689,"BadRequestError for loading dataset?","WPoelman",2,"","",0,"2025-07-18","2025-07-18"],[8,7688,"No module named \"distributed\"","yingtongxiong",4,"","",0,"2025-07-17",""],[8,7687,"Datasets keeps rebuilding the dataset every time i call the python script","CALEB789",4,"","",0,"2025-07-17",""],[8,7686,"load_dataset does not check .no_exist files in the hub cache","jmaccarl",4,"","",0,"2025-07-16",""],[8,7685,"Inconsistent range request behavior for parquet REST api","universalmind303",4,"","",0,"2025-07-16",""],[8,7682,"Fail to cast Audio feature for numpy arrays in datasets 4.0.0","luatil-cloud",0,"lhoestq","#7684",1,"2025-07-14","2025-07-15"],[8,7681,"Probabilistic High Memory Usage and Freeze on Python 3.10","ryan-minato",4,"","",0,"2025-07-14",""],[8,7680,"Question about iterable dataset and streaming","Tavish9",4,"","",0,"2025-07-12",""],[8,7679,"metric glue breaks with 4.0.0","stas00",2,"","",0,"2025-07-10","2025-07-11"],[8,7678,"To support decoding audio data, please install 'torchcodec'.","alpcansoydas",2,"","",0,"2025-07-10","2025-07-11"],[8,7677,"Toxicity fails with datasets 4.0.0","serena-ruan",2,"","",0,"2025-07-10","2025-07-11"],[8,7676,"Many things broken since the new 4.0.0 release","mobicham",4,"","",0,"2025-07-09",""],[8,7675,"common_voice_11_0.py failure in dataset library","egegurel",4,"","",0,"2025-07-09",""],[8,7671,"Mapping function not working if the first example is returned as None","dnaihao",4,"","",0,"2025-07-08",""],[8,7669,"How can I add my custom data to huggingface datasets","xiagod",4,"","",0,"2025-07-04",""],[8,7668,"Broken EXIF crash the whole program","Seas0",4,"","",0,"2025-07-03",""],[8,7665,"Function load_dataset() misinterprets string field content as part of dataset schema when dealing with `.jsonl","zdzichukowalski",3,"","",0,"2025-07-01","2025-07-01"],[8,7664,"Function load_dataset() misinterprets string field content as part of dataset schema when dealing with `.jsonl","zdzichukowalski",4,"","",0,"2025-07-01",""],[8,7662,"Applying map after transform with multiprocessing will cause OOM","JunjieLl",4,"","",0,"2025-07-01",""],[8,7660,"AttributeError: type object 'tqdm' has no attribute '_lock'","Hypothesis-Z",4,"","",0,"2025-06-30",""],[8,7650,"`load_dataset` defaults to json file format for datasets with 1 shard","iPieter",4,"","",0,"2025-06-27",""],[8,7647,"loading mozilla-foundation--common_voice_11_0 fails","pavel-esir",4,"","",0,"2025-06-26",""],[8,7637,"Introduce subset_name as an alias of config_name","albertvillanova",4,"","",0,"2025-06-24",""],[8,7636,"\"open\" in globals()[\"__builtins__\"], an error occurs: \"TypeError: argument of type 'module' is not iterable\"","kuanyan9527",4,"","",0,"2025-06-24",""],[8,7633,"Proposal: Small Tamil Discourse Coherence Dataset.","bikkiNitSrinagar",4,"","",0,"2025-06-23",""],[8,7632,"Graceful Error Handling for cast_column(\"image\", Image(decode=True)) in Hugging Face Datasets","ganiket19",4,"","",0,"2025-06-23",""],[8,7630,"[bug] resume from ckpt skips samples if .map is applied","felipemello1",4,"","",0,"2025-06-21",""],[8,7627,"Creating a HF Dataset from lakeFS with S3 storage takes too much time!","Thunderhead-exe",2,"","",0,"2025-06-19","2025-06-23"],[8,7624,"#Dataset Make \"image\" column appear first in dataset preview UI","jcerveto",2,"","",0,"2025-06-18","2025-06-20"],[8,7619,"`from_list` fails while `from_generator` works for large datasets","abdulfatir",4,"","",0,"2025-06-17",""],[8,7617,"Unwanted column padding in nested lists of dicts","qgallouedec",2,"","",0,"2025-06-15","2025-06-16"],[8,7612,"Provide an option of robust dataset iterator with error handling","wwwjn",4,"","",0,"2025-06-13",""],[8,7611,"Code example for dataset.add_column() does not reflect correct way to use function","shaily99",0,"ArjunJagdale","#7648",0,"2025-06-12","2025-07-17"],[8,7610,"i cant confirm email","lykamspam",4,"","",0,"2025-06-12",""],[8,7607,"Video and audio decoding with torchcodec","TyTodd",1,"TyTodd","#7616",0,"2025-06-11","2025-06-19"],[8,7600,"`push_to_hub` is not concurrency safe (dataset schema corruption)","sharvil",0,"lhoestq","#7708",1,"2025-06-07","2025-07-31"],[8,7599,"My already working dataset (when uploaded few months ago) now is ignoring metadata.jsonl","JuanCarlosMartinezSevilla",2,"","",0,"2025-06-06","2025-06-16"],[8,7597,"Download datasets from a private hub in 2025","DanielSchuhmacher",2,"","",0,"2025-06-06","2025-06-13"],[8,7594,"Add option to ignore keys/columns when loading a dataset from jsonl(or any other data format)","avishaiElmakies",4,"","",0,"2025-06-05",""],[8,7591,"Add num_proc parameter to push_to_hub","SwayStar123",2,"","",0,"2025-06-04","2025-09-04"],[9,1877,"Tensor.min(dim)/max(dim) return 0 on integer dtypes when a slice is entirely positive or negative","laoyyu",4,"","",0,"2026-11-06",""],[9,1871,"Tensor.to() ignores float16 bit-pattern storage for every target except float32","laoyyu",4,"","",0,"2026-11-05",""],[9,1869,"cat() does not validate shapes and silently returns a corrupt tensor","laoyyu",4,"","",0,"2026-11-05",""],[9,1867,"Tensor.squeeze() ignores negative dims","laoyyu",4,"","",0,"2026-11-05",""],[9,1865,"Tensor.view() does not validate shapes","laoyyu",4,"","",0,"2026-11-05",""],[9,1862,"Tensor.sigmoid() silently returns zeros on integer and bool tensors","laoyyu",4,"","",0,"2026-11-05",""],[9,1860,"Tensor accessors and reductions return bit patterns (or throw) for float16 tensors","laoyyu",4,"","",0,"2026-11-05",""],[9,1859,"Decode of image/audio-text-to-text models spends ~10% of each step reading text embeddings back to the CPU","kylo5aby",4,"","",0,"2026-11-05",""],[9,1857,"Tensor.to() from float to int64/uint64 truncates toward -inf instead of toward zero","laoyyu",4,"","",0,"2026-11-05",""],[9,1855,"Scalar arithmetic (add/sub/mul/div) on int64/uint64 tensors throws TypeError","laoyyu",4,"","",0,"2026-11-05",""],[9,1853,"Tensor.permute/transpose do not support negative dims (torch.permute does)","laoyyu",4,"","",0,"2026-11-05",""],[9,1851,"Tensor.round() uses half-away-from-zero rounding, diverging from torch.round","laoyyu",4,"","",0,"2026-11-05",""],[9,1849,"Bug: sum/mean/norm silently wrap (wrong results) on small integer dtypes (uint8, int32, ...)","laoyyu",4,"","",0,"2026-11-04",""],[9,1847,"Bug: Tensor.min(dim)/Tensor.max(dim) throw TypeError on int64/uint64 tensors","laoyyu",4,"","",0,"2026-11-04",""],[9,1845,"Tensor.norm() returns wrong results for negative values when p is 1 or odd","RaphaelFakhri",4,"","",0,"2026-11-04",""],[9,1843,"docs: warn that web GPU tensors must use the session Tensor class (dual onnxruntime copies)","ShenweiFFF",4,"","",0,"2026-11-03",""],[9,1842,"Web build: no public way to get the session Tensor class (dual onnxruntime Tensor copies)","ShenweiFFF",4,"","",0,"2026-11-03",""],[9,1840,"Two copies of onnxruntime's Tensor class in the web build: wrapping a fromGpuBuffer tensor makes dispose() rec","daveycodez",4,"","",0,"2026-10-30",""],[9,1839,"Bump onnxruntime-web to ≥ 1.31.1: sliding-window attention gets flash attention on WebGPU","daveycodez",4,"","",0,"2026-10-30",""],[9,1838,"Prefix caching across generate() calls: DynamicCache.crop and state snapshots for hybrid models","daveycodez",4,"","",0,"2026-10-30",""],[9,1837,"TextStreamer decodes the whole line on every token (quadratic on long lines)","daveycodez",4,"","",0,"2026-10-30",""],[9,1836,"Feature request: chunked prefill in generate() (e.g. `prefill_chunk_size`) — reading long prompts in one pass ","daveycodez",4,"","",0,"2026-10-29",""],[9,1835,"Use onnxruntime-web's JSPI build when the browser supports JSPI (much lower memory in Safari)","daveycodez",4,"","",0,"2026-10-29",""],[9,1833,"random.seed() still collapses large negative integer seeds to seed(0)","michaelkillgta",4,"","",0,"2026-10-29",""],[9,1831,"Proposal: a same-origin inference broker shared by browser tabs","jabrailkhalil",4,"","",0,"2026-10-29",""],[9,1830,"mean_pooling is slower than the model forward pass on GPU (stride-embedDim loop)","tfriedel",4,"","",0,"2026-10-29",""],[9,1827,"iOS Safari: tab is killed at the end of the first download of a large `.onnx_data` file (the file is held in m","daveycodez",4,"","",0,"2026-10-29",""],[9,1813,"Injecting a runtime via Symbol.for('onnxruntime') still loads the bundled onnxruntime-node, which crashes next","iliev-blnk",4,"","",0,"2026-10-22",""],[9,1810,"Qwen2VLImageProcessor merges several images into one (same size) or throws \"offset is out of bounds\" (differen","ranacseruet",4,"","",0,"2026-10-21",""],[9,1806,"Auxiliary operator sessions ignore the model device and cannot be configured (TensorOpRegistry.session_options","jakesalvesen",4,"","",0,"2026-10-20",""],[9,1802,"Qwen3.5 (Qwen VL family): text-only `generate()` with an external `past_key_values` calls the vision encoder","toomore",4,"","",0,"2026-10-17",""],[9,1801,"`from_pretrained()` never settles in the browser when an external data (`.onnx_data`) download fails","toomore",0,"iamdgarcia","#1804",0,"2026-10-17","2026-10-26"],[9,1800,"Add support for SAM 3 (Sam3Model): text-prompted concept segmentation","AModSoul",4,"","",0,"2026-10-15",""],[9,1794,"[Bug]VoxtralRealtime: generate() decodes one step after the audio runs out and streams a junk token","jiangzhuo",4,"","",0,"2026-10-08",""],[9,1786,"Does Whisper Support MOV (.mov) Video Input?","limxinying0720",4,"","",0,"2026-10-06",""],[9,1784,"Compiler Errors when using --skipLibcheck false","dpfens",4,"","",0,"2026-09-30",""],[9,1767,"createInferenceSession() memoises a rejected session promise, breaking every later session creation in the pro","angri450",4,"","",0,"2026-09-08",""],[9,1759,"Importing Transformers.js causes warning in Rspack and Webpack","JonnyBurger",0,"nico-martin","#1760",1,"2026-08-26","2026-08-29"],[9,1754,"[Bug] model_file_name parameter ignored for encoder-decoder models; hardcoded _quantized suffix cannot be over","Conlactor",4,"","",0,"2026-08-22",""],[9,1752,"[Discussion] AI Agent Collaboration for Browser-Based ML Testing","gemini-3-1-pro",3,"","",0,"2026-08-19","2026-08-20"],[9,1750,"WASM beat WebGPU by ~9.6x on a small embedding model — building an open benchmark to find out where else that'","Sallamhamza",4,"","",0,"2026-08-17",""],[9,1749,"onnx-community/OpenMed-PII-SuperClinical: q8 silently returns near-chance logits; fp16 and q4f16 fail to load","matthewchung74",4,"","",0,"2026-08-17",""],[9,1748,"Chrome 153 WebNN API: deviceType: 'webnn-npu' rejected, must use 'npu'","yzhou51",4,"","",0,"2026-08-15",""],[9,1741,"[WebGPU] WebGPU decode permanently degrades ~3x at S=2048/4096/6144","kylo5aby",4,"","",0,"2026-08-12",""],[9,1739,"Whisper ASR pipeline leaks ~650 MB of GPU memory per 30 s chunk on WebGPU (both `return_timestamps` modes)","maboa",4,"","",0,"2026-08-03",""],[9,1735,"Moonshine ASR pipeline throws 'Unable to decode without a tokenizer' — _call_moonshine calls processor.batch_d","mosabsayyed",0,"Mr-Neutr0n","#1738",0,"2026-08-01","2026-09-03"],[9,1734,"Chatterbox KV cache leak","m96-chan",1,"m96-chan","#1737",0,"2026-07-31","2026-09-04"],[9,1729,"sharp `^0.34.5` range cannot admit patched 0.35.x (libvips CVEs, GHSA-f88m-g3jw-g9cj)","wardspan",0,"xenova","#1766",1,"2026-07-23","2026-09-15"],[9,1728,"[WebGPU] embeddinggemma-300m-ONNX q4/q8 produce silently wrong embeddings (cos ~0 vs WASM); fp32 correct","jethac",4,"","",0,"2026-07-22",""],[9,1727,"Audit flagged version of adm-zip via onnxruntime-node@1.24.3 via @huggingface/transformers@4.2.0","sroussey",0,"xenova","#1766",1,"2026-07-21","2026-09-15"],[9,1726,"ParakeetForCTC outputs only blank tokens - empty transcription on all backends (v4.2.0)","madforstrength",3,"","",0,"2026-07-18","2026-07-19"],[9,1724,"LFM uses {%- generation -%} found only in newer jinja not used by transformers.js","sroussey",2,"","",0,"2026-07-16","2026-10-18"],[9,1719,"[Question] What is the release plan for the next version after 4.2.0(April 23)?","daijh",4,"","",0,"2026-07-10",""],[9,1718,"Bump onnxruntime-node to ^1.27 to drop deprecated boolean transitive dependency","efenocchi",0,"xenova","#1766",1,"2026-07-09","2026-09-15"],[9,1711,"Gemma Challenge for ONNX + transformers.js","sroussey",4,"","",0,"2026-06-18",""],[9,1707,"whisper-base.en_timestamped q8 decoder fails session creation on v4 runtime: 'Missing required scale ... MatMu","maboa",0,"xenova","#1766",1,"2026-06-11","2026-09-15"],[9,1704,"RawAudio.toBlob() encodes the entire underlying ArrayBuffer, ignoring TypedArray byteOffset/length","xiankaing",0,"yushuosun","#1712",0,"2026-06-10","2026-09-07"],[9,1698,"OnnxRuntime crash in CI (Ubuntu) with cpu","wszgrcy",4,"","",0,"2026-06-01",""],[9,1697,"Context Window of Various Models","jdp8",2,"","",0,"2026-05-29","2026-06-05"],[9,1689,"arXiv endorsement request — first CWV benchmark of Transformers.js inference","srikarphanikumar",4,"","",0,"2026-05-21",""],[9,1687,"Bump `onnxruntime-node` to ^1.26.0 to drop deprecated `boolean@3` from install tree","PedroCF87",0,"xenova","#1766",1,"2026-05-20","2026-09-15"],[9,1686,"Feature request: assisted/speculative decoding via assistant_model in generate()","wordingone",2,"","",0,"2026-05-16","2026-06-16"],[9,1684,"ASR (Whisper) regression for detected timestamps","jozefchutka",4,"","",0,"2026-05-13",""],[9,1678,"Gemma 4 not  working in browser -[WebGPU] Error running MODEL_ID_GOES_HERE","zelqarnain88mhss-png",2,"","",0,"2026-05-07","2026-05-08"],[9,1677,"Why doesn't HuggingFace create a browser extension for transformers js?","AgustinRamiroDiaz",4,"","",0,"2026-05-04",""],[9,1672,"transformers.js can't be used in a Bun `--compile` single binary without patches","evantahler",4,"","",0,"2026-05-04",""],[9,1667,"Error on translation","thomas-0816",4,"","",0,"2026-05-01",""],[9,1666,"Gemma 4 generation passes num_logits_to_keep=0 in decoder_forward, causing full-prompt logits memory blowup","TimPietruskyRunPod",0,"xenova","#1681",1,"2026-04-29","2026-05-08"],[9,1663,"Model files are downloaded multiple times if progress_callback is active","thomas-0816",0,"anishesg","#1664",0,"2026-04-24","2026-04-27"],[9,1660,"Missing chat template for Gemma4","thomas-0816",2,"","",0,"2026-04-23","2026-04-23"],[9,1656,"Support for chatterbox-multilingual","mscdex",4,"","",0,"2026-04-19",""],[9,1653,"Bonsai 4B: Uncaught Error: Can't create a session. ERROR_CODE: 6, ERROR_MESSAGE: std::bad_alloc","thomas-0816",4,"","",0,"2026-04-15",""],[9,1652,"return_tensors ?","sroussey",2,"","",0,"2026-04-15","2026-04-16"],[9,1651,"No release artefact since 4.0.1","Jerboas86",2,"","",0,"2026-04-15","2026-04-24"],[9,1645,"Export supportedDevices / deviceToExecutionProviders from public API","TimPietrusky",0,"Amarnath10i","#1753",0,"2026-04-15","2026-10-23"],[9,1643,"get_available_devices()","alexewerlof",4,"","",0,"2026-04-14",""],[9,1642,"Auto device on Linux x64 fails hard when CUDA shared library is unavailable instead of falling back","alexewerlof",4,"","",0,"2026-04-14",""],[9,1640,"Illegal Instruction when running `pipeline` on Raspberry Pi 3 Model B (64-bit)","sammorrisdesign",4,"","",0,"2026-04-12",""],[9,1637,"`Error: Can't create a session` for translation pipeline","vitonsky",4,"","",0,"2026-04-10",""],[9,1636,"How complete is the TransformersJS Gemma4 Multimodal support?","hpssjellis",2,"","",0,"2026-04-09","2026-04-15"],[9,1635,"All translation models failling for me on web","sroussey",4,"","",0,"2026-04-03",""],[9,1634,"@huggingface/transformers.js 4.0.1: onnxruntime-node 1.24.3 does not ship with x64 darwin (MacOS) bindings.","zqwitt",2,"","",0,"2026-04-03","2026-09-21"],[9,1632,"max_new_tokens?","sroussey",2,"","",0,"2026-04-02","2026-04-15"],[9,1630,"is_pipeline_cached(\"feature-extraction\") incorrect","sroussey",2,"","",0,"2026-04-02","2026-04-07"],[9,1628,"model support for jinaai/jina-embeddings-v5-text-nano-retrieval","aperez900907",2,"","",0,"2026-04-02","2026-04-12"],[9,1625,"Feature: Resumable downloads and integrity verification for large models","hamzaydia",4,"","",0,"2026-04-01",""],[9,1624,"types in example","sroussey",0,"xenova","#1641",1,"2026-04-01","2026-04-14"],[9,1623,"increase strictness of type checking over time","sroussey",4,"","",0,"2026-04-01",""],[9,1622,"Request: ONNX conversion for fine-tuned SmolVLM-256M (Patricijia/smolvlm-gif-descriptor)","Patricijia",4,"","",0,"2026-03-31",""],[9,1621,"Youtube broken link in the V4 release note","Jerboas86",2,"","",0,"2026-03-30","2026-03-30"],[9,1612,"LlamaTokenizer ignores `add_prefix_space` when `legacy=false` (v3.x)","oshiteku",2,"","",0,"2026-03-28","2026-04-12"],[9,1606,"is_pipeline_cached","sroussey",0,"xenova","#1614",1,"2026-03-25","2026-03-29"],[9,1605,"image-text-to-text model support","sroussey",2,"","",0,"2026-03-24","2026-04-07"],[9,1604,"transformers@4.0.0-next.7 in Safari: no available backend found. ERR: [webgpu]","tomayac",0,"xenova","#1700",1,"2026-03-24","2026-06-08"],[9,1599,"[Performance] Qwen3.5-4B (q4f16) is ~3x slower decode + ~20x slower TTFT vs Qwen3-4B on WebGPU (Transformers.j","youqibing",4,"","",0,"2026-03-20",""],[9,1598,"Some questions about Xenova/sam-vit-base","rmahdav",2,"","",0,"2026-03-18","2026-03-18"],[9,1590,"[webgpu] Whisper encoder fp16 precision issues","jozefchutka",4,"xenova","#1594",1,"2026-03-17",""],[9,1584,"[Feature Request] Deterministic generation for MusicGen (support for seed/PRNG)","jozefchutka",2,"","",0,"2026-03-16","2026-03-17"],[9,1581,"Incorrect model selection by pipeline dtype parameter","starkdg",4,"","",0,"2026-03-16",""],[9,1577,"How to get dtypes?","sroussey",1,"sroussey","#1580",0,"2026-03-14","2026-03-24"],[9,1576,"[V4] pipeline.dispose() destroys WebGPU device, can't be recreated afterwards","nico-martin",0,"xenova","28f9267",1,"2026-03-13","2026-03-30"],[9,1574,"Releasing the Qwen3.5 ONNX conversion script","ir2718",4,"","",0,"2026-03-11",""],[9,1568,"model info utility?","sroussey",3,"","",0,"2026-03-06","2026-04-12"],[9,1567,"Xenova/all-MiniLM-L6-v2 q4f16","sroussey",2,"","",0,"2026-03-06","2026-03-07"],[9,1565,"Saved context state?","sroussey",4,"","",0,"2026-03-06",""],[9,1561,"ModelRegistry.is_pipeline_cached is poisoning the cache","sroussey",0,"nico-martin","#1564",1,"2026-03-05","2026-03-06"],[9,1554,"ModelRegistry.is_pipeline_cached does not need network calls","sroussey",2,"","",0,"2026-03-03","2026-03-05"],[9,1550,"v3.x: onnxruntime-node@1.21.0 pulls in vulnerable tar (<7.5.8) — CVE-2026-26960","emiluzelac",3,"","",0,"2026-03-01","2026-03-05"],[9,1545,"Problem with WebGPU (windows with amd gpu)","DarkSorrow",4,"","",0,"2026-02-25",""],[9,1544,"Loading the same model simultaneously in multiple processes fails","sroussey",0,"nico-martin","#1548",1,"2026-02-25","2026-03-04"],[9,1536,"[WebGPU] Error running MODEL_ID_GOES_HERE","chinaboom13-alt",2,"","",0,"2026-02-20","2026-02-20"],[9,1534,"Types not exported","sroussey",2,"","",0,"2026-02-20","2026-02-20"],[9,1532,"[v4.0.0-next.3] Incorrect implementation of `loadWasmFactory`","fs-eire",0,"nico-martin","#1558",1,"2026-02-18","2026-03-05"],[9,1527,"[4.0.0-next.3] CDN wasmPaths breaks local dev servers (CORS Worker construction error)","AdrianBesleaga",0,"nico-martin","#1558",1,"2026-02-15","2026-03-05"],[9,1522,"[v4.0.0-next.2] ProgressCallback type missing","Jerboas86",0,"nico-martin","#1528",1,"2026-02-11","2026-02-16"],[9,1521,"[v4.0.0-next.2] wasm ort larger than 25 MiB not supported by cloudflare worker","Jerboas86",2,"","",0,"2026-02-11","2026-03-02"],[9,1518,"Still crashing using WebGPU with translation pipeline","thesmartwon",4,"","",0,"2026-02-09",""],[9,1512,"[WebGPU] Incorrect results running int8 quantized models","sudonatalie",2,"","",0,"2026-02-03","2026-02-09"],[9,1508,"file was not found locally","qiyongjin",2,"","",0,"2026-01-29","2026-03-04"],[9,1506,"wav2phoneme model","guettli",2,"","",0,"2026-01-27","2026-03-04"],[9,1503,"gemme3_text report","sere-otubu",2,"","",0,"2026-01-21","2026-01-31"],[9,1500,"Where is v4?","aghasemi",0,"xenova","#1382",1,"2026-01-20","2026-02-09"],[9,1496,"Add license information for Xenova/flan-t5-small model card","rmahdav",2,"","",0,"2026-01-05","2026-01-09"],[9,1495,"Incorrect Type Definitions for AutoTokenizer and AutoModelForCausalLM","VinayHajare",0,"xenova","#1382",1,"2025-12-23","2026-02-09"],[9,1491,"Memory Leak in Image Embedding POC","yovasx2",0,"xenova","#1382",1,"2025-12-19","2026-02-09"],[9,1490,"Example models for each pipeline","sroussey",4,"","",0,"2025-12-19",""],[9,1488,"Add whisper-medium timestamped models to onnx-community","neonwatty",2,"","",0,"2025-12-16","2025-12-16"],[9,1487,"License clarification for some of the converted models","rmahdav",2,"","",0,"2025-12-16","2025-12-16"],[9,1485,"Often getting error: Could not locate file: \"https://huggingface.co/.../onnx/model.onnx\"","customautosys",2,"","",0,"2025-12-14","2026-01-31"],[9,1484,"Should npm @xenova/transformers be deleted or marked deprecated?","matthieu-talbot-ergonomia",4,"","",0,"2025-12-13",""],[9,1479,"[RFC] Move from global `env` to session-based configuration","nico-martin",4,"","",0,"2025-12-09",""],[9,1476,"Support of  \"depth-anything/Depth-Anything-V2-Metric-*-*-hf\"","Disk-MTH",4,"","",0,"2025-12-03",""],[9,1470,"onnx-community/age-gender-prediction-ONNX runtime error","orv1s",2,"","",0,"2025-12-01","2026-01-31"],[9,1469,"[WebGPU] JSEP crashes when running Gemma 3 (1b-it)","KevinL10",0,"xenova","#1382",1,"2025-12-01","2026-02-09"],[9,1467,"Missing the following inputs: input_points, input_labels (or input_boxes)","sherlockchou86",2,"","",0,"2025-11-28","2025-12-01"],[9,1460,"Phi-4 Mini Support on Transformers.js v4","ibelem",2,"","",0,"2025-11-19","2026-02-12"],[9,1458,"ONNX Backend Env variable","Heinrik-20",2,"","",0,"2025-11-19","2026-01-14"],[9,1457,"Support for LayoutLMv3","somus",4,"","",0,"2025-11-17",""],[9,1456,"big large models","kmnnmk212-source",0,"xenova","#1382",1,"2025-11-14","2026-02-09"],[9,1450,"SmolVLM2 500M Video Instruct - Video inference","youchi1",4,"","",0,"2025-11-10",""],[9,1448,"3.7.6 typescript build errors","jozefchutka",0,"xenova","#1382",1,"2025-10-23","2026-02-09"],[9,1447,"How to use half precision ONNX models?","richarddd",2,"","",0,"2025-10-23","2026-01-31"],[9,1446,"Zhare-AI/sd-1-5-webgpu on HuggingFace.co lists itself as Transformer.js supported?","LostBeard",2,"","",0,"2025-10-22","2025-10-24"],[9,1443,"Segment-Everything-Everywhere-All-At-Once model addition","Heinrik-20",4,"","",0,"2025-10-20",""],[9,1439,"Integration to a CLI application created using PKG","JosephJibi",4,"","",0,"2025-10-16",""],[9,1438,"Add support for Gemma3 (VLM)","azizbekabdul",0,"xenova","#1601",1,"2025-10-12","2026-03-23"],[9,1437,"Add \"Building an Angular Application\" page in documentation","lukonik",4,"","",0,"2025-10-11",""],[9,1435,"Error: using ceil() in shape computation is not yet supported for MaxPool","geronimi73",0,"xenova","#1382",1,"2025-10-09","2025-10-20"],[9,1434,"Add support for Chatterbox","brianferri",0,"xenova","#1382",1,"2025-10-08","2026-02-09"],[9,1430,"WebGPU/CoreML support for node.js","nemphys",0,"xenova","#1382",1,"2025-10-03","2026-02-09"],[9,1429,"text-generation task fails with temperature: 0","nico-martin",1,"nico-martin","#1431",1,"2025-10-02","2025-10-03"],[9,1425,"Build fails for text-to-speech-webgpu example due to outetts incompatibility","Darshan0312",4,"","",0,"2025-10-02",""],[9,1424,"Problems with WebGPU backend in Apple M3 Pro macbook Pro","caraya",0,"xenova","#1382",1,"2025-09-29","2025-10-15"],[9,1420,"Add model support for janus-pro-7b-webgpu","adrian-leung-zhare-ai",4,"","",0,"2025-09-27",""],[9,1419,"Why is `token-classification` with T5 not available? (`T5ForTokenClassification`)","debevv",4,"","",0,"2025-09-21",""],[9,1418,"EmbeddingGemma usage","MithrilMan",0,"xenova","#1382",1,"2025-09-21","2026-02-09"],[9,1416,"\"Rotary interleaved attention is not supported\" error in WebGPU implementation for MobileLLM","jpabbuehl",0,"xenova","#1382",1,"2025-09-14","2026-02-09"],[9,1415,"Manually converted `Qwen2.5-Coder-0.5B-Instruct` does not work, but  `onnx-community/Qwen2.5-0.5B-Instruct` do","visitsb",0,"xenova","#1382",1,"2025-09-14","2026-02-09"],[9,1410,"Janus-1.3B-ONNX model web demo does not work with \"A float16 tensor's data must be type of function Float16Arr","zlelik",2,"","",0,"2025-09-08","2025-09-08"],[9,1409,"Getting  TS2304: Cannot find name 'PretrainedProcessorOptions'. error","ShibleNK",0,"nico-martin","#1436",1,"2025-09-08","2025-10-14"],[9,1408,"qwen3 embedding with customCache throw error","wszgrcy",0,"xenova","#1617",1,"2025-09-08","2026-03-30"],[9,1407,"Expected time to load a super-resolution model locally","ymtoo",2,"","",0,"2025-09-08","2025-09-30"],[9,1406,"option to use `onnxruntime-web` in nodelike environments","oofdere",0,"xenova","#1382",1,"2025-09-07","2026-02-09"],[9,1403,"MacOS ONNX runtime workaround libc++abi: terminating due to uncaught exception of type std::__1:system_error: ","coffeedome",2,"","",0,"2025-08-29","2026-02-09"],[9,1402,"Add Siglip2 (naflex) support","pySilver",0,"xenova","#1382",1,"2025-08-28","2026-02-09"],[9,1401,"When the model is downloading, the progressInfo in the progress_callback method sometimes lacks the status, na","kukudixiaoming",2,"","",0,"2025-08-28","2025-09-12"],[9,1393,"Angular Hello World Sentiment Analysis","BrianMikinski",2,"","",0,"2025-08-18","2025-10-21"],[9,1392,"Cannot convert YOLO .pt models using convert script or Optimum","ayan4m1",2,"","",0,"2025-08-16","2025-09-30"],[9,1389,"docs: Transformers.js model provider for Vercel AI SDK","jakobhoeg",2,"","",0,"2025-08-15","2026-03-05"],[9,1388,"Hf","InteligenceQ",2,"","",0,"2025-08-07","2025-08-16"],[9,1387,"Missing unit tests and mismatched output in audio_utils (e.g., spectrogram, window_function) vs Python tests","jozefchutka",0,"xenova","#1594",1,"2025-08-06","2026-03-29"],[9,1385,"VideoFrame support (WebCodecs)","kixelated",4,"","",0,"2025-08-03",""],[9,1383,"Unable to use gemma 3n onnx model?","grctest",2,"","",0,"2025-08-01","2025-08-07"],[9,1380,"Crash using WebGPU with translation pipeline","thesmartwon",0,"xenova","#1382",1,"2025-07-25","2026-02-09"],[9,1379,"Why Do I Get Different Outputs in Python and JavaScript for the Same ONNX Model?","mahdin75",2,"","",0,"2025-07-23","2025-07-28"],[9,1374,"nanoVLM support","sbrzz",3,"","",0,"2025-07-22","2026-02-20"],[9,1372,"[deno] OffscreenCanvas not supported by this browser for server-side worker","korywka",2,"","",0,"2025-07-21","2026-03-04"],[9,1366,"Request: Add ONNX version of CodeT5+ or CodeBERT for browser usage","Emilianodelarc",4,"","",0,"2025-07-16",""],[9,1365,"Uncaught (in promise) RangeError: Array buffer allocation failed","ItsSimko",0,"xenova","#1382",1,"2025-07-14","2026-02-09"],[9,1363,"The specified module could not be found.","tocha688",4,"","",0,"2025-07-11",""],[9,1361,"Conversion script to convert to ONNX not working","alonsosilvaallende",2,"","",0,"2025-07-10","2025-07-11"],[9,1358,"whisper-base_timestamped broken with chunk_length_s=30","jozefchutka",0,"xenova","#1594",1,"2025-07-07","2026-03-29"],[9,1357,"whisper-large-v3-turbo_timestamped has broken timestamps","jozefchutka",0,"xenova","43b2662",1,"2025-07-07","2026-03-29"],[9,1356,"Gemma3n browser support","bil-ash",2,"","",0,"2025-07-07","2026-03-06"],[9,1355,"Add Nanonets-OCR-s","jpabbuehl",4,"","",0,"2025-07-04",""],[9,1345,"Add a `supported` API","tarekziade",0,"nico-martin","#1511",1,"2025-06-25","2026-03-01"],[9,1343,"numpy break change np.fromstring binary mode removed","xiaoyao9184",0,"fidoriel","#1351",0,"2025-06-18","2025-07-02"],[9,1342,"Attempt to load onnx model in browser failed: TypeError: t.replace is not a function","mdlufy",3,"","",0,"2025-06-17","2025-06-18"],[9,1341,"FireFox compatible models","12th-devs",4,"","",0,"2025-06-16",""],[9,1340,"Audio-to-Audio task","LuSrodri",4,"","",0,"2025-06-15",""],[9,1339,"Model is cached, but still reloads from network?","patrickinminneapolis",2,"","",0,"2025-06-11","2025-06-12"],[9,1338,"Question about supporting Float16Array","xmcp",0,"xenova","#1382",1,"2025-06-11","2026-02-09"],[9,1337,"3.5.2 typescript build errors","jozefchutka",0,"nico-martin","#1436",1,"2025-06-10","2025-10-14"],[9,1336,"node.js WebGPU compatibility and WASM performance in web enviornment","devnarekm",0,"xenova","#1382",1,"2025-06-10","2026-02-09"],[9,1334,"Gemma 3n and Gemma 3 multimodal","NSTiwari",2,"","",0,"2025-06-03","2025-06-26"],[9,1332,"TypeError: Cannot read properties of undefined (reading 'create')","A-Ghattas",0,"xenova","#1333",1,"2025-06-02","2025-06-04"],[10,974,"`kernel-builder init` test fails in `testshell` with edition 6","mohabdr0805",4,"","",0,"2026-11-06",""],[10,970,"Concurrent `get_kernel` calls on a fresh machine fail with `TUFError: Failed to refresh TUF metadata`","albertvillanova",4,"","",0,"2026-11-05",""],[10,956,"kernels-common: hash_variant loads entire file into memory, causing high RSS on large .so files","nrenvo",4,"","",0,"2026-11-01",""],[10,926,"Hub requests fail with HF_HUB_DISABLE_TELEMETRY=1 (empty user agent → illegal header)","rycerzes",0,"danieldk","#939",1,"2026-10-22","2026-10-26"],[10,925,"Hub requests fail with HF_HUB_DISABLE_TELEMETRY=1 (empty user agent → illegal header)","swapnildutta-atlan",3,"","",0,"2026-10-22","2026-10-22"],[10,914,"kernels benchmark pulls from the old model repos and says every gpu is cpu","iemAnshuman",4,"","",0,"2026-10-19",""],[10,913,"Use per-call prerelease policy in the Nix wheel dependency hook","notatallshaw",0,"danieldk","#938",1,"2026-10-19","2026-10-26"],[10,905,"Would be good to have a small test for this in the resolver tests as well, but I think that can be another PR.","sayakpaul",1,"sayakpaul","eb9a88f",1,"2026-10-15","2026-10-27"],[10,838,"Document kernels cache a bit more","sayakpaul",4,"","",0,"2026-09-22",""],[10,834,"Kernel-builder drops for Strix Halo architecture","AzeezIsh",0,"danieldk","#839",1,"2026-09-19","2026-09-22"],[10,804,"Kernal","yeye111095-hue",2,"","",0,"2026-09-02","2026-09-02"],[10,778,"Key CUDA-capability lookup by device index instead of process-wide lru_cache","AmirF194",0,"danieldk","#941",1,"2026-08-21","2026-10-26"],[10,770,"We have an existing test that tests usage of `pyext`. Maybe this new example can replace it? (cuts down test b","sayakpaul",1,"sayakpaul","#811",1,"2026-08-18","2026-09-03"],[10,757,"[Important] Shutting down of the \"model\" type repos of kernels on the Hub","sayakpaul",4,"","",0,"2026-08-12",""],[10,753,"macOS build failure due to missing `metadata.json`","danieldk",2,"","",0,"2026-08-10","2026-09-11"],[10,742,"CUTLASS 4.5.2 fixed-output hash mismatch in kernel-builder","LiangSu8899",2,"","",0,"2026-08-01","2026-08-03"],[10,720,"kernel-builder: support SM103 (Blackwell Ultra / B300) capabilities, e.g. 10.3a","sayakpaul",2,"","",0,"2026-07-17","2026-09-02"],[10,719,"Add Flash-MSA kernels","swizzleboi",2,"","",0,"2026-07-16","2026-07-18"],[10,718,"Versioning / build issues vendoring quack (cutlass/cutedsl pin) for sonic-moe","sayakpaul",4,"","",0,"2026-07-15",""],[10,717,"Make kernel-builder available as a CLI in the job container (use build-and-upload instead of nix run)","sayakpaul",1,"sayakpaul","#738",1,"2026-07-15","2026-08-01"],[10,715,"More (optional) load time verification","vasqu",4,"","",0,"2026-07-14",""],[10,711,"[TPU] I want to add support for torch tpu and pallas kernels","tengomucho",1,"tengomucho","#714",1,"2026-07-13","2026-08-03"],[10,707,"Process exits when architecture is not supported","remi-or",2,"","",0,"2026-07-09","2026-08-24"],[10,701,"Add support for MLU kernelize","zsl223",4,"","",0,"2026-07-08",""],[10,698,"Support fetching for particular CUDA capabilities","sayakpaul",4,"","",0,"2026-07-07",""],[10,696,"AOT-compiling what are normally JIT kernels","sayakpaul",4,"","",0,"2026-07-07",""],[10,669,"kernels benchmark does not support XPU","danielfleischer",4,"","",0,"2026-06-22",""],[10,640,"First-time loading of large kernels (e.g. `deep-gemm`) hits repeated 429 rate limits and takes very long","Muinez",4,"","",0,"2026-06-13",""],[10,632,"`get_kernel` 404s on `/api/kernels/.../refs` for an existing kernel repo","jiqing-feng",2,"","",0,"2026-06-11","2026-06-12"],[10,621,"[docs] structural improvements","stevhliu",2,"","",0,"2026-06-04","2026-06-23"],[10,605,"Device.__post_init__ never calls validate_capability_range, silently accepting inverted min/max capability","thesayakmondal",2,"","",0,"2026-05-30","2026-06-01"],[10,593,"Ensure the latest symbol files","sayakpaul",1,"sayakpaul","#629",1,"2026-05-28","2026-06-10"],[10,592,"kernel-builder upload error","ashwanirathee",2,"","",0,"2026-05-27","2026-05-28"],[10,584,"LocalLayerRepository.__str__ produces a malformed string (missing closing parenthesis)","thesayakmondal",0,"danieldk","#597",1,"2026-05-26","2026-05-29"],[10,566,"variants: TvmFfi._VARIANT_REGEX uses lazy quantifier instead of exact single-digit match for major version","thesayakmondal",2,"","",0,"2026-05-19","2026-05-20"],[10,553,"Offline kernel loading fails despite cached snapshot","spidersouris",0,"sayakpaul","#580",1,"2026-05-14","2026-05-27"],[10,548,"persistence of broken link issue","ArjunSrivastava1",0,"danieldk","#581",1,"2026-05-13","2026-05-23"],[10,542,"Broken link","peltax",2,"","",0,"2026-05-12","2026-05-12"],[10,538,"Update to check trusted repos from hub","drbh",2,"","",0,"2026-05-11","2026-05-20"],[10,537,"Clarify Windows support","danieldk",1,"danieldk","#586",1,"2026-05-11","2026-05-27"],[10,536,"Build aotriton from binary wheels as well","danieldk",2,"","",0,"2026-05-11","2026-05-20"],[10,535,"Update cute-dsl if needed","danieldk",2,"","",0,"2026-05-11","2026-05-13"],[10,529,"Remove versionless `get_kernel`, etc.","danieldk",2,"","",0,"2026-05-10","2026-05-27"],[10,528,"Code signing","danieldk",2,"","",0,"2026-05-10","2026-09-02"],[10,521,"'kernels skills add' does not work","KishanPipariya",2,"","",0,"2026-05-05","2026-05-07"],[10,508,"Implement `trust_remote_code` for non-trusted publishers","sayakpaul",2,"","",0,"2026-04-27","2026-05-06"],[10,503,"create kernels curated installations","sayakpaul",1,"sayakpaul","#613",1,"2026-04-23","2026-06-03"],[10,486,"Card generation incorrectly shows `layers` as a function","danieldk",0,"sayakpaul","#500",1,"2026-04-18","2026-04-22"],[10,477,"`kernel-builder upload` does not respect branch","danieldk",1,"danieldk","#485",1,"2026-04-17","2026-04-19"],[10,469,"Avoid rebuilding kernel-builder when there are unrelated changes","danieldk",1,"danieldk","#478",1,"2026-04-16","2026-04-17"],[10,465,"Include capabilities for noarch kernels in build.toml and export to metadata","sayakpaul",2,"","",0,"2026-04-16","2026-07-07"],[10,458,"Clean up documentation","sayakpaul",2,"","",0,"2026-04-15","2026-04-16"],[10,454,"Add option to specify compiler flags on the CPU (like -ffast-math)","martinloretzzz",1,"martinloretzzz","#455",0,"2026-04-14","2026-05-19"],[10,444,"Consider supporting cuda-tile as a Python dependency","danieldk",2,"","",0,"2026-04-12","2026-05-28"],[10,442,"Move kernels skills out of the kernels package and transfer to builder","sayakpaul",1,"sayakpaul","#456",1,"2026-04-10","2026-04-15"],[10,441,"Support the new kernel repo type","danieldk",2,"","",0,"2026-04-10","2026-04-19"],[10,440,"Source files of header only libraries aren't added to nix sandbox when added with include","martinloretzzz",4,"","",0,"2026-04-10",""],[10,439,"Metadata fetching utility from a kernels repo","sayakpaul",2,"","",0,"2026-04-10","2026-06-10"],[10,436,"Consider removing variant enumeration again and instead explain why each available are not compatible","danieldk",1,"danieldk","#561",1,"2026-04-09","2026-05-18"],[10,430,"`kernel-builder`: add a `build.toml` option to mark a kernel as using the Torch stable ABI","danieldk",2,"","",0,"2026-04-08","2026-06-01"],[10,429,"`kernels`: add a build variant for the Torch stable ABI","danieldk",2,"","",0,"2026-04-08","2026-06-01"],[10,427,"Add back deprecation warning when no `version` or `revision` is specified","danieldk",0,"sayakpaul","#450",1,"2026-04-03","2026-04-14"],[10,415,"Set up e2e testing for the new kernel-builder cli","sayakpaul",2,"","",0,"2026-03-30","2026-04-08"],[10,414,"Create onboarding script with the new kernel-builder cli","sayakpaul",1,"sayakpaul","#418",1,"2026-03-30","2026-04-01"],[10,413,"Speed up variant enumeration in Nix completion","danieldk",2,"","",0,"2026-03-30","2026-04-02"],[10,412,"Get variant list and support variant for build/devshell/testshell","danieldk",2,"","",0,"2026-03-30","2026-04-02"],[10,405,"[docs] add a security page","sayakpaul",2,"","",0,"2026-03-26","2026-10-26"],[10,404,"Replace `huggingface-hub` git dependency from `kernel-builder`","danieldk",2,"","",0,"2026-03-26","2026-05-06"],[10,403,"XPU Torch 2.11","danieldk",2,"","",0,"2026-03-26","2026-03-30"],[10,402,"Add JAX glue for tvm-ffi","danieldk",1,"danieldk","#431",1,"2026-03-26","2026-04-09"],[10,396,"Wrap `kernels-data` data structures with PyO3 and use in `kernels`","danieldk",2,"","",0,"2026-03-24","2026-05-06"],[10,370,"follow up on old build failures","drbh",2,"","",0,"2026-03-17","2026-10-26"],[10,359,"Add C++ dependency support for tvm-ffi","danieldk",1,"danieldk","#385",1,"2026-03-15","2026-03-20"],[10,356,"Fix noarch devshell + testshell","danieldk",1,"danieldk","89a80f4",1,"2026-03-13","2026-03-23"],[10,355,"tvm-ffi ROCm support","danieldk",2,"","",0,"2026-03-13","2026-10-26"],[10,354,"Add tvm-ffi CUDA capability detection (for devshells)","danieldk",1,"danieldk","#357",1,"2026-03-13","2026-03-16"],[10,353,"tvm-ffi XPU support","danieldk",1,"danieldk","#368",1,"2026-03-13","2026-03-17"],[10,352,"tvm-ffi Metal support","danieldk",4,"","",0,"2026-03-13",""],[10,351,"Add `kernel-builder build-and-upload` and make it dispatch to `nix build-and-upload`","danieldk",2,"","",0,"2026-03-13","2026-03-26"],[10,350,"Hook up card filling in `nix run .#build-and-upload`","danieldk",2,"","",0,"2026-03-13","2026-03-26"],[10,349,"Move card filling/instantiation to `kernel-builder`","danieldk",2,"","",0,"2026-03-13","2026-03-30"],[10,348,"Move `kernels init` to `kernel-builder init`","danieldk",2,"","",0,"2026-03-13","2026-03-26"],[10,347,"Add `kernel-builder testshell` and make it dispatch to `nix develop .#test`","danieldk",2,"","",0,"2026-03-13","2026-03-24"],[10,346,"Add `kernel-builder devshell` and make it dispatch to `nix develop`","danieldk",2,"","",0,"2026-03-13","2026-03-24"],[10,345,"Add `kernel-builder build` and make it dispatch to `nix build`","danieldk",2,"","",0,"2026-03-13","2026-03-26"],[10,338,"Warn users when not using the latest kernel version","sayakpaul",1,"sayakpaul","#342",1,"2026-03-12","2026-03-13"],[10,327,"Compatibility with modeling code relying on `flash-attn` package","pmalic",2,"","",0,"2026-03-10","2026-04-15"],[10,326,"Support resolving the CUDA version to another compatible version","danieldk",0,"sayakpaul","#330",1,"2026-03-10","2026-03-11"],[10,324,"[docs] add a page showing how `kernels` is incorporated in the ecosystem for now","sayakpaul",2,"","",0,"2026-03-10","2026-04-13"],[10,323,"Create dataclasses to hold frameworks for added flexibility","sayakpaul",2,"","",0,"2026-03-10","2026-04-10"],[10,317,"Exploring collaboration: Integrating FlagOS Kernels into HuggingFace Kernels","Dongxu-H",2,"","",0,"2026-03-09","2026-03-21"],[10,315,"README utility for `main` branches on the build repos on Hub","sayakpaul",2,"","",0,"2026-03-06","2026-03-26"],[10,314,"Guide to show how to port existing `kernels` incompatible builds to `kernels` compatible ones","sayakpaul",2,"","",0,"2026-03-06","2026-07-07"],[10,313,"`kernels upload` should start new branches fresh","danieldk",0,"sayakpaul","#322",1,"2026-03-05","2026-03-10"],[10,312,"all kernels segfaults on load/import on Python 3.14-nogil","Qubitium",2,"","",0,"2026-03-05","2026-06-06"],[10,307,"Metal example uses unsafe MPS encoder pattern that crashes on sequential calls","robtaylor",4,"","",0,"2026-03-04",""],[10,302,"[FEATURE] Triton Skill","L-Reichardt",0,"jaygala223","#675",0,"2026-03-02","2026-07-07"],[10,299,"Not able to build any kernels with Nix","ParagEkbote",2,"","",0,"2026-02-23","2026-04-15"],[10,264,"Using `uvx kernels versions` errors out complaining `torch` not installed","ariG23498",0,"danieldk","#272",1,"2026-02-05","2026-02-10"],[10,236,"Terraform scripts for setting up instances for building, development, etc.","sayakpaul",1,"sayakpaul","#303",1,"2026-01-26","2026-03-04"],[10,213,"Compatibility with `hf download`","Adel-Moumen",2,"","",0,"2026-01-19","2026-04-10"],[10,201,"Using `kernelize` isn't yielding any speed-ups","pramodith",2,"","",0,"2026-01-06","2026-04-10"],[10,179,"`kernels-community/flash-attn2` not found, but the build exists","qgallouedec",2,"","",0,"2025-11-06","2025-11-07"],[10,172,"offline install/use?","hpcpony",2,"","",0,"2025-10-17","2025-10-22"],[10,163,"AttributeError: module 'torch.version' has no attribute 'xpu'","zheliuyu",2,"","",0,"2025-10-13","2025-10-14"],[10,144,"Introduce staging tests for the upload test in GitHub","sayakpaul",1,"sayakpaul","#152",1,"2025-09-16","2025-09-22"],[10,143,"Give a better error message when a forward signature does not match","danieldk",1,"danieldk","#145",1,"2025-09-15","2025-09-16"],[10,133,"Support diffusers models","ParagEkbote",2,"","",0,"2025-08-19","2025-09-05"],[10,131,"Bug: \"cpu\" kernels are not supported","fritzo",0,"danieldk","#132",1,"2025-08-12","2025-08-25"],[10,130,"Support  for causal-conv1d","HaroldBenoit",2,"","",0,"2025-08-11","2025-09-05"],[10,101,"Flash Attention mha_fwd getting incorrect results.","JamesLYan",2,"","",0,"2025-06-26","2025-07-17"],[11,4488,"enable_model_cpu_offload(): first post-load generation thrashes the Windows pagefile (~6x step time), silently","AIwork4me",4,"","",0,"2026-11-01",""],[11,4480,"Stateful DataLoader resume at an epoch boundary replays the completed epoch (shard iteration counter is not ch","anishpentyala",4,"","",0,"2026-10-31",""],[11,4476,"Sequence parallelism: allow an opt-out of the linear-attention refusal, and make disable_in_eval configurable","eilamc14",4,"","",0,"2026-10-30",""],[11,4474,"FSDP2 + cpu_ram_efficient_loading: ignored_modules are sharded anyway, and models with persistent buffers cras","eilamc14",4,"","",0,"2026-10-30",""],[11,4469,"accelerator.reduce(tensor, \"mean\") fails on integer tensors when more than one process is used","MaiAditya",4,"","",0,"2026-10-29",""],[11,4465,"get_max_memory() budgets CPU offload from host RAM, not the container memory limit","prabhkodes",4,"","",0,"2026-10-27",""],[11,4454,"save_state(total_limit=...) deletes the newest checkpoint and load_state() resumes from an old one on Windows","Sonica-B",4,"","",0,"2026-10-25",""],[11,4452,"Stateful split-batch dispatcher replays consumed samples after restoring","AryanRG",4,"","",0,"2026-10-25",""],[11,4445,"Optimizer exceptions leave a stale AMP step marker and advance the scheduler after overflow","salma2vec",4,"","",0,"2026-10-23",""],[11,4443,"Failed dataloader iteration leaves stale gradient state","ska2704",4,"","",0,"2026-10-23",""],[11,4436,"save_state deletes retained checkpoints before a replacement save succeeds","salma2vec",4,"","",0,"2026-10-21",""],[11,4434,"load_checkpoint_and_dispatch ignores dtype with default device_map=None","akshitkhokhani",4,"","",0,"2026-10-20",""],[11,4430,"`preserve_rng_stage=False` in fsdp2_apply_ac causes bug when using dropout with activation checkpointing","YisuZhou",4,"","",0,"2026-10-20",""],[11,4419,"FSDP2 forward does not run under mixed precision autocast, unlike DDP and FSDP1","12kimih",4,"","",0,"2026-10-18",""],[11,4411,"Quick survey: your experience as a PyTorch Ecosystem maintainer","michellearoth",4,"","",0,"2026-10-16",""],[11,4410,"gather_for_metrics drops or duplicates examples with SubsetRandomSampler","gorkadabo1",4,"","",0,"2026-10-15",""],[11,4409,"test2","Asthenia0412",4,"","",0,"2026-10-15",""],[11,4399,"`pad_across_processes` turns `bool` tensors into `int64`, which breaks gather across ranks","arthurdjn",4,"","",0,"2026-10-13",""],[11,4397,"`send_to_device` ignores `skip_keys` for Mappings that have `.to()` method","arthurdjn",4,"","",0,"2026-10-13",""],[11,4393,"DataLoaderDispatcher.total_batch_size raises AttributeError with ordinary datasets","kocean9-freedom",4,"","",0,"2026-10-13",""],[11,4392,"Warn when DeepSpeed ZeRO silently changes custom optimizer semantics","XIONGPEILIN",4,"","",0,"2026-10-13",""],[11,4382,"FSDP2 RAM-efficient loading fails on persistent buffers with AttributeError: 'Tensor' object has no attribute ","xvxuopop",4,"","",0,"2026-10-09",""],[11,4381,"compile_regions doesn't compile gradient-checkpointed transformers layers that contain a graph break","qgallouedec",4,"","",0,"2026-10-09",""],[11,4359,"`gather_for_metrics` keeps padding duplicates for a sized `IterableDataset` with `dispatch_batches=False`","Zhuoxi2000",4,"","",0,"2026-10-01",""],[11,4353,"wait_for_everyone() raises NotImplementedError on MULTI_CPU when the machine has an MPS accelerator (barrier g","JoeyTan21",1,"JoeyTan21","#4354",0,"2026-09-28","2026-10-01"],[11,4339,"FSDP2 get_state_dict() does not preserve tied-weight aliasing","UnitedSnakes",2,"","",0,"2026-09-26","2026-11-04"],[11,4330,"Resuming with load_state + skip_first_batches (checkpointing example) does not restore the shuffle order","SeverinVisionary",4,"","",0,"2026-09-24",""],[11,4323,"`load_state()` and `save_state()` raise a bare `IndexError` under `automatic_checkpoint_naming`","Om-singhaI",2,"","",0,"2026-09-23","2026-09-29"],[11,4315,"Saving a configuration mutates its Enum values and empty dictionaries","HuaTNA",2,"","",0,"2026-09-23","2026-09-29"],[11,4306,"split_between_processes crashes on tuples when apply_padding=True","shoutoutuoadi325",1,"shoutoutuoadi325","#4307",0,"2026-09-22","2026-09-25"],[11,4297,"`balanced_low_0` puts the whole model on GPU 0 when `max_memory` skips a device id","Om-singhaI",1,"Om-singhaI","#4298",0,"2026-09-21","2026-09-25"],[11,4295,"[Bug] skip_first_batches drops _drop_last, _non_blocking, slice_fn, and torch_device_mesh across checkpoint re","truecallerabreham",1,"truecallerabreham","#4296",0,"2026-09-20","2026-09-29"],[11,4289,"[Bug] FSDP2 + fp16: load_state() does not round-trip the GradScaler state","HaomingSong",1,"HaomingSong","#4291",1,"2026-09-20","2026-09-24"],[11,4283,"[Bug] RNG state restoration failure swallowed silently — resumed training diverges without warning","cmaguire001",2,"","",0,"2026-09-19","2026-10-29"],[11,4280,"OffloadedWeightsLoader initialization scans CPU keys for every disk weight","emme1t",1,"emme1t","#4281",0,"2026-09-19","2026-09-25"],[11,4277,"`simple_launcher` discards the child's error, unlike the elastic launch path","albertvillanova",1,"albertvillanova","#4305",1,"2026-09-18","2026-09-23"],[11,4269,"all_gather_into_tensor is deprecated, _gpu_gather warns on every gather","qgallouedec",0,"Yi-111-a","#4272",0,"2026-09-16","2026-09-25"],[11,4265,"CSOAI — Accelerate AI governance distributed","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[11,4257,"`PrefixedDataset` violates the `Mapping` contract","betacatsling",2,"","",0,"2026-09-14","2026-10-23"],[11,4253,"compile_regions runs the uncompiled model when the root module has an accelerate hook","yupengtang",2,"","",0,"2026-09-13","2026-09-29"],[11,4249,"prepare_model unconditionally imports DTensor, breaking single-GPU training on torch builds without a distribu","LeoBorcherding",0,"shoemoney","#4250",0,"2026-09-11","2026-09-25"],[11,4242,"[Bug] Fused AdamW AMP overflow leaves step_was_skipped false and advances the scheduler","gss10282023",2,"","",0,"2026-09-10","2026-10-19"],[11,4240,"ACCELERATE_BYPASS_DEVICE_MAP is the only boolean env flag read case-sensitively, so =True is silently ignored","Sreekant13",1,"Sreekant13","#4241",0,"2026-09-10","2026-09-25"],[11,4226,"find_executable_batch_size does not catch MPS out-of-memory errors","caiotheodoro",1,"caiotheodoro","#4227",0,"2026-09-08","2026-09-08"],[11,4215,"Second circular import: accelerate.state <-> accelerate.utils (same pattern as #4173)","Rakshit-gen",2,"","",0,"2026-09-04","2026-09-07"],[11,4212,"`accelerate estimate-memory` fails with timm 1.0.29: model name has no source prefix","iamsharduld",1,"iamsharduld","#4213",0,"2026-09-04","2026-09-08"],[11,4204,"`use_stateful_dataloader` + `num_workers>0` draws epoch-0 permutation during `prepare()` before cross-rank RNG","GoldenStain",4,"","",0,"2026-09-02",""],[11,4200,"Add ep_size to ParallelismConfig so FSDP2 can compose with expert parallelism","YeonwooSung",2,"","",0,"2026-09-02","2026-09-07"],[11,4199,"Skip automatic DataLoader sharding when the loader is already rank-sharded","YeonwooSung",2,"","",0,"2026-09-02","2026-10-17"],[11,4195,"`use_stateful_dataloader` under multi-process training restores the cursor, not the shuffle order — sampler pe","GoldenStain",2,"","",0,"2026-08-30","2026-10-13"],[11,4191,"Proposal: tutorial for sample-stable image augmentation across workers and resume","ternaus",2,"","",0,"2026-08-29","2026-09-08"],[11,4183,"get_max_memory() adds a separate cpu entry on integrated CUDA GPUs, double-counting shared RAM (GB10 / DGX Spa","Defilan",0,"harshitakrishna18","#4187",0,"2026-08-24","2026-09-07"],[11,4173,"Circular import between accelerate.utils and accelerate.big_modeling breaks concurrent imports","adrihans",2,"","",0,"2026-08-20","2026-09-07"],[11,4159,"AcceleratedOptimizer.eval() does not reach through DeepSpeed wrapper, silently saving train-view schedule-free","XIONGPEILIN",2,"","",0,"2026-08-11","2026-09-07"],[11,4154,"send_to_device crashes on defaultdict batches","aswanth-07",2,"","",0,"2026-08-09","2026-08-24"],[11,4122,"BatchSamplerShard.__len__ overcounts for some ranks when split_batches=True and even_batches=False","ErenAta16",2,"","",0,"2026-07-17","2026-09-01"],[11,4111,"concatenate() still raises TypeError on non-tensor batch fields when num_processes > 1","albertvillanova",2,"","",0,"2026-07-09","2026-08-18"],[11,4102,"`--parallelism_config_sp_seq_length_is_variable False` parses as True","MicroMilo",2,"","",0,"2026-07-06","2026-09-01"],[11,4099,"[Feature] ShardedOptimStateDictConfig + per-rank save/load for large-scale FSDP checkpoint","gygdh-001",1,"gygdh-001","#4105",0,"2026-07-06","2026-07-22"],[11,4094,"SuperCompress: Cut LLM token costs by ~65% with open-source prompt compression","arjunkshah",2,"","",0,"2026-06-30","2026-06-30"],[11,4088,"pre-commit fails on a fresh checkout: pinned ruff v0.2.1 can't parse pyproject.toml (Unknown rule selector: UP","devangpratap",1,"devangpratap","#4089",0,"2026-06-26","2026-07-07"],[11,4086,"When we use StateDictType.FULL_STATE_DICT to resume training from a checkpoint, it takes over 5 hours to start","gygdh-001",2,"","",0,"2026-06-22","2026-07-30"],[11,4080,"[bug report] DataSourcesShufflingDisallowed when training using split datasets.IterableDataset","KirkGXL",2,"","",0,"2026-06-15","2026-07-23"],[11,4075,"[Feature request] Support already-sharded DataLoaders in Accelerator.prepare","GoldenStain",2,"","",0,"2026-06-12","2026-07-20"],[11,4074,"Feature Proposal: Add FunASR ASR Distributed Inference Examples","LauraGPT",2,"","",0,"2026-06-12","2026-07-20"],[11,4067,"[Security] Path traversal + malicious-model DoS via sharded-checkpoint `weight_map` in ….","geo-chen",0,"mattcurf","#47",0,"2026-06-09","2026-08-12"],[11,4062,"Question: Manual Dataset Sharding per GPU Rank with Accelerate + DistributedSampler (Avoid Double DataLoader L","D222097",2,"","",0,"2026-06-03","2026-07-12"],[11,4055,"cncl should at the same level in _prepare_backend in src/accelerate/state.py","legionGIT",0,"Anai-Guo","#4057",0,"2026-05-29","2026-06-09"],[11,4053,"[Bug]  Implicit padding when splitting input between processes while padding flag  is disabled","3manifold",1,"3manifold","#4052",0,"2026-05-27","2026-06-09"],[11,4041,"MLflowTracker.log drops log_kwargs; tracker.log requires step despite docstring","1fanwang",1,"1fanwang","#4039",0,"2026-05-13","2026-06-09"],[11,4036,"Device_map multi-GPU inference can silently corrupt tensors when local CUDA peer copies are unhealthy","Aitejiu",2,"","",0,"2026-05-11","2026-06-09"],[11,4030,"To achieve consistent computation results under disparate device quantity configurations.","YangBase",2,"","",0,"2026-05-06","2026-05-06"],[11,4029,"[Windows] RTX 5070 Ti (Blackwell sm_120) - setup and deployment notes","loongmiaow-pixel",2,"","",0,"2026-05-05","2026-07-07"],[11,4016,"Feature request: disk_offload() that mmaps HF-cache safetensors directly (no duplicate copy, no CPU-RAM materi","AlliedToasters",2,"","",0,"2026-04-22","2026-05-31"],[11,4014,"Unable to infer valid auto device map for gemma-4-E4B-it","ShenranTomWang",2,"","",0,"2026-04-22","2026-05-31"],[11,4010,"Ram explodes when training a very simple model with accelerate tpu","dercodeKoenig",2,"","",0,"2026-04-15","2026-04-18"],[11,3996,"DataLoader shuffle sequence replays from epoch 0 after resuming from a checkpoint","XiangY-Q",2,"","",0,"2026-04-05","2026-05-14"],[11,3995,"device_map=\"auto\": silent corruption of tensors captured in register_forward_hook (3+ GPUs, inference_mode, st","keenanpepper",2,"","",0,"2026-04-03","2026-05-12"],[11,3991,"Isssue when using torch.compile","tritam593",2,"","",0,"2026-03-28","2026-05-27"],[11,3990,"[Feature] Save model-only feature in `save_state`","3manifold",2,"","",0,"2026-03-27","2026-05-04"],[11,3983,"[FSDP2] auto-exclude incompatible Params4bit from fully_shard to prevent silent QLoRA corruption","roycho96",1,"roycho96","#3987",0,"2026-03-19","2026-04-10"],[11,3979,"[Bug] FSDP2 mixed-precision upcast to fp32 is a silent no-op since v1.13.0","iavinas",2,"","",0,"2026-03-18","2026-04-26"],[11,3971,"FSDP2 + SHARDED_STATE_DICT: optimizer checkpoint load fails with missing step key. (RuntimeError: Missing key ","iavinas",2,"","",0,"2026-03-11","2026-03-19"],[11,3968,"loading weight is so slow for fsdp2","jiaqiw09",2,"","",0,"2026-03-10","2026-04-18"],[11,3967,"[Bug] prepare() silently double-wraps models and double-serializes checkpoints when called twice on the same o","iavinas",0,"joshuaswanson","#3977",0,"2026-03-09","2026-04-16"],[11,3964,"Security: Unsafe torch.load in FSDP2 scaler path (CWE-502)","spartan8806",2,"","",0,"2026-03-09","2026-04-16"],[11,3963,"[Bug] TrackioTracker.log() ignores step parameter","iavinas",0,"joshuaswanson","#3975",0,"2026-03-07","2026-03-20"],[11,3960,"[Bug] RNG states from multiple backends (e.g. CUDA + HPU) are saved but only one is restored on load_state","iavinas",0,"Rakshit-gen","#4217",0,"2026-03-06","2026-04-25"],[11,3956,"`infer_auto_device_map` does not place submodule buffers on `device_map` when submodule is split","dg845",2,"","",0,"2026-03-05","2026-04-12"],[11,3954,"AssertionError for FP8 Benchmarks","Hukongtao",2,"","",0,"2026-03-04","2026-04-12"],[11,3949,"Cut a new release","shimizust",2,"","",0,"2026-02-27","2026-03-05"],[11,3939,"[Bug] in_order parameter in logging is broken - deadlock or ignored","siryuson",2,"","",0,"2026-02-25","2026-03-02"],[11,3938,"[Bug] DataLoaderShard with StatefulDataLoader produces wrong state dict in DDP","yuxinyuan",0,"SunMarc","#3952",1,"2026-02-25","2026-03-04"],[11,3933,"Segfault in device_map='auto' weight dispatch on RTX 5090 D v2 (Blackwell, sm_120, GB202-240)","XHXIAIEIN",2,"","",0,"2026-02-24","2026-04-04"],[11,3932,"why does (Naive MP) show better memory efficiency than FSDP1/2 ??","Lim-Sung-Jun",2,"","",0,"2026-02-24","2026-04-05"],[11,3930,"`_prepare_fsdp2` does not work with nn.Module on `meta` device","jiayuanmark",2,"","",0,"2026-02-21","2026-04-06"],[11,3925,"notebook_launcher raises \"Cannot re-initialize CUDA in forked subprocess\"","minhtriet",0,"SunMarc","#4059",1,"2026-02-10","2026-06-09"],[11,3923,"fsdp2 + lora | TypeError: fully_shard() got an unexpected keyword argument 'ignored_params'","garychan22",0,"Mr-Neutr0n","#3924",0,"2026-02-09","2026-03-02"],[11,3913,"`save_function` arg in `save_pretrained` is dropped in transformers 5 RCs","khoover",2,"","",0,"2026-01-24","2026-03-02"],[11,3911,"Severe GPU training performance degradation due to incorrectly set `KMP_AFFINITY`","hexfaker",2,"","",0,"2026-01-23","2026-02-23"],[11,3899,"Test failures under Python 3.14 on x86_64-linux","sarahec",0,"SunMarc","#3901",1,"2026-01-05","2026-01-09"],[11,3898,"FSDP 2 + cpu_ram_efficient_loading=True fails for models with buffers (e.g. DS 3.1)","d5031",2,"","",0,"2026-01-05","2026-02-12"],[11,3897,"activation_checkpointing=True uses *a lot* more memory than gradient_checkpointing=True *with FSDP 2.0* (FSDP ","d5031",2,"","",0,"2026-01-05","2026-02-15"],[11,3896,"Fail to `load_state` immediately after `save_state` with FSDP plugins.","xiuqhou",2,"","",0,"2025-12-30","2026-02-06"],[11,3894,"How to specify different number of process per node","AIML001",2,"","",0,"2025-12-21","2026-01-28"],[11,3889,"UlyssesSPAttentionHF with PEFT model","shon-otmazgin-wix",2,"","",0,"2025-12-16","2026-01-25"],[11,3885,"CpuOffload pre_forward and post_forward should have @torch.compiler.disable(), otherwise torch.compile fails i","doctorpangloss",0,"SunMarc","#3888",1,"2025-12-13","2025-12-17"],[11,3882,"`accelerate.load_checkpoint_and_dispatch` does not load GPT-OSS Models correctly","KyleMylonakisProtopia",2,"","",0,"2025-12-11","2026-02-02"],[11,3879,"Bug in accelerator.unwrap_model","Giuseppe5",0,"amanzoni1","#3881",0,"2025-12-09","2025-12-16"],[11,3877,"Gradient accumulation gives worse results when using DeepSpeed ZeRO 2","swyoon",2,"","",0,"2025-12-05","2026-02-08"],[11,3876,"Why TP can't be used with pure DP?","quic-meetkuma",2,"","",0,"2025-12-05","2026-02-22"],[11,3875,"Bug with nested Scheduler","FrsECM",2,"","",0,"2025-12-05","2025-12-05"],[11,3874,"Feature request: FSDP2 QLoRA","ojh31",4,"","",0,"2025-12-05",""],[11,3873,"How to specify accelerate launch yaml config item when running with torchrun","WhoisZihan",2,"","",0,"2025-12-04","2026-01-11"],[11,3871,"Can't run accelerate in a CPU only colab env","FlorianVal",1,"FlorianVal","#3872",0,"2025-12-03","2025-12-04"],[11,3870,"FSDP2 fails due to `KeyError: 'lm_head.weight'`","qgallouedec",0,"amanzoni1","#3878",0,"2025-12-03","2025-12-11"],[11,3869,"Degraded performance when resuming from checkpoint","mdenomme24",2,"","",0,"2025-12-03","2026-01-11"],[11,3868,"Missing library stubs or py.typed marker","josecannete",2,"","",0,"2025-12-01","2026-01-11"],[11,3867,"I'm getting a \"AttributeError: 'Accelerator' object has no attribute '_cp_context'\" error when trying to fine-","Adam-Mazur",2,"","",0,"2025-12-01","2025-12-05"],[11,3866,"Feature Request: Add API to Materialize Meta Tensors for Device Movement","EricRollei",2,"","",0,"2025-11-30","2026-09-25"],[11,3861,"NCCL timeout for validation","WenxiongZhou",2,"","",0,"2025-11-28","2026-01-11"],[11,3860,"Accelerate and DataLoaders","Bhoy1",2,"","",0,"2025-11-28","2026-02-06"],[11,3859,"Optimizer Adagrad not working with Accelerate","sdjasj",2,"","",0,"2025-11-28","2026-01-05"],[11,3856,"context parallel `_cp_context` not created","ppetrushkov",0,"kashif","#3858",1,"2025-11-26","2025-11-28"],[11,3855,"KeyError in optimizer.state_dict() under FSDP2 when using Adagrad optimizer","sdjasj",2,"","",0,"2025-11-26","2026-01-11"],[11,3854,"Missing `fsdp_min_num_params` leads to type conversion failure in FSDP config","sdjasj",0,"CodeMan62","#3902",0,"2025-11-26","2026-01-07"],[11,3849,"`DataLoaderDispatcher` doesn't accept non-Tensor values from the data collator","tomaarsen",1,"tomaarsen","#3850",1,"2025-11-25","2025-11-26"],[11,3846,"The test dependency torchpippy is merged into PyTorch and the repository is archived","yurivict",2,"","",0,"2025-11-23","2026-01-01"],[11,3844,"FSDP2 with lora take more memory than FSDP","Zhikaiiii",0,"ojh31","#3848",0,"2025-11-21","2025-11-26"],[11,3839,"fix: allow for flattening and unflattening optimizer state dict using flatten_optimizer_state_dict","kmehant",2,"","",0,"2025-11-12","2025-12-21"],[11,3838,"GPT-OSS fails to load with FSPD2","qgallouedec",2,"","",0,"2025-11-11","2025-12-21"],[11,3836,"When using gradient accumulation, does the order of optimizer.zero_grad() affect training?","polestarss",2,"","",0,"2025-11-10","2025-11-17"],[11,3833,"notebook_launcher Kaggle RuntimeError: CUDA error: initialization error","dinhngoc267",2,"","",0,"2025-11-06","2025-12-15"],[11,3832,"FSDP MoE PEFT hangs during forward pass","ojh31",2,"","",0,"2025-11-04","2025-11-21"],[11,3830,"Torchao fp8 fails if using accelerate config file with Trainer","shimizust",1,"shimizust","#3831",0,"2025-11-03","2025-12-03"],[11,3829,"Does Accelerate automatically set the DataLoader’s sampler to a DistributedSampler?","caixxiong",2,"","",0,"2025-11-03","2025-12-16"],[11,3827,"Accelerate mixed torch.Tensor and DTensor error when using TE FP8 and FSDP/TP","gmintoco",2,"","",0,"2025-11-01","2025-12-16"],[11,3826,"Accelerate save_state() error using FSDP2/TP","gmintoco",2,"","",0,"2025-11-01","2025-12-10"],[11,3825,"Error generating example: 'weight' must be 2-D in model.generate()","yurkoff-mv",2,"","",0,"2025-11-01","2025-12-09"],[11,3824,"load_checkpoint_and_dispatch not work","ziliangok",2,"","",0,"2025-10-31","2025-12-16"],[11,3822,"`TypeError: NoneType object is not callable` occurs when using `fsdp_auto_wrap_policy=NO_WRAP` with `fsdp_acti","sdjasj",2,"","",0,"2025-10-30","2025-12-21"],[11,3821,"KeyError: \"A parameter in the optimizer couldn't be switched to its sharded version\" occurs under specific Acc","sdjasj",2,"","",0,"2025-10-30","2025-12-13"],[11,3820,"AssertionError: found no DeviceMesh from dtensor args for c10d.broadcast_.default","juraev",2,"","",0,"2025-10-29","2025-12-11"],[11,3818,"Duplicate W&B initialization in offline mode","ShuyUSTC",0,"shantanugupta2004","#3886",0,"2025-10-23","2025-12-16"],[11,3814,"Deadlock when running multi tasks","RuizhePang",2,"","",0,"2025-10-14","2026-06-09"],[11,3813,"Supporting new hardware","haifeng-jin",2,"","",0,"2025-10-13","2025-12-08"],[11,3812,"RuntimeError during load_state","Silverster98",2,"","",0,"2025-10-13","2025-11-21"],[11,3811,"ValueError: Could not find the transformer layer class QwenImageTransformerBlock in the model.","garychan22",2,"","",0,"2025-10-11","2025-11-22"],[11,3807,"Clipping grad not working for deepspeed","trxcc",2,"","",0,"2025-10-07","2025-11-22"],[11,3806,"Installing accelerate slows down Pylance","Haoming02",2,"","",0,"2025-10-07","2025-11-15"],[11,3805,"Beginner question about usage details","WinderK",2,"","",0,"2025-10-05","2025-11-12"],[11,3804,"File \"/opt/conda/lib/python3.11/site-packages/torch/distributed/device_mesh.py\", line 132, in create_sub_mesh ","ldh127",2,"","",0,"2025-10-01","2025-11-09"],[11,3803,"should_reduce_batch_size function not returning expected output for out of memory exception on H100 NVL MIG","mshannon-sil",2,"","",0,"2025-10-01","2025-12-03"],[11,3802,"i want to request a demo code for StatefulDataLoader , i want to use data checkpoint to recover the train stag","ldh127",2,"","",0,"2025-09-30","2025-11-09"],[11,3801,"os.env/torch.cuda.empty_cache are not thread safe","Qubitium",2,"","",0,"2025-09-30","2025-12-02"],[11,3798,"KeyError: \"Invalid mesh_dim_names ('dp_shard_cp',) specified. Valid mesh_dim_names are ['tp'].\"","ldh127",2,"","",0,"2025-09-29","2025-09-29"],[11,3797,"Question: ReduceLROnPlateau wrapped by AcceleratedScheduler in DDP may multiply LR by num_processes?","nicelulu",2,"","",0,"2025-09-26","2025-11-03"],[11,3795,"torch.distributed.DistNetworkError","TongJingLin",2,"","",0,"2025-09-26","2025-11-03"],[11,3794,"Error when evaluating with multi-gpu","adfad1",2,"","",0,"2025-09-25","2025-11-03"],[11,3793,"ND-Parallel with FSDP v2 occasionally hang when parallelism_config_dp_replicate_size > 1.","npuichigo",2,"","",0,"2025-09-25","2025-11-03"],[11,3792,"tests/test_accelerator.py: Random weights test flakes (possibly aarch64 specific)","xanderlent",2,"","",0,"2025-09-24","2025-11-15"],[11,3789,"script: accelerate.py function: clip_grad_norm_  if parameters == [p for p in model.parameters()]: RuntimeErro","Xingyuxingyu",2,"","",0,"2025-09-20","2025-11-28"],[11,3788,"Issue with IterableDataset and variable batch sequence length","dsrivastavv",2,"","",0,"2025-09-20","2025-10-28"],[11,3785,"accelerate.cpu_offload can not be used for training","CO18326",3,"","",0,"2025-09-18","2025-09-18"],[11,3784,"AttributeError: 'Accelerator' object has no attribute 'deepspeed_config'. Did you mean: 'deepspeed_plugin'?","alexge233",2,"","",0,"2025-09-18","2025-10-27"],[11,3780,"\"Gradient accumulation on training samples of variable size\" example broken when using DeepSpeed","inkitori",2,"","",0,"2025-09-15","2025-10-24"],[11,3778,"Unable to specify mp_policy for FSDP2 from config file","npuichigo",2,"","",0,"2025-09-13","2025-10-21"],[11,3775,"Muti node CUDA error: invalid device ordinal","rdolmedo",1,"rdolmedo","#3779",0,"2025-09-12","2025-09-13"],[11,3770,"Need for proper mention in the documentation  that accelerate.cpu_offload   can not be used for training","CO18326",2,"","",0,"2025-09-10","2025-10-21"],[11,3768,"Optimize the time complexity of get_balanced_memory","Jintao-Huang",0,"SamuelBarryCS","#3776",0,"2025-09-09","2025-09-17"],[11,3763,"Writing environment variables to `.deepspeed_env` in append mode may break the pipeline","jue-jue-zi",2,"","",0,"2025-09-07","2025-10-15"],[11,3758,"TypeError: device() received an invalid combination of arguments","BrazilForever11",0,"rezaqorbani","#3895",0,"2025-09-03","2025-11-05"],[11,3754,"NPU availability check error if `torch_npu` import error","yanyongyu",1,"yanyongyu","#3764",0,"2025-09-01","2025-09-09"],[11,3749,"collect_epoch_for_metrics — stream step outputs, auto-dedup, optional trim/shard","ishachat",2,"","",0,"2025-08-27","2025-10-05"],[11,3748,"How pass two layer class by use --fsdp_transformer_layer_cls_to_wrap?","sunjian2015",2,"","",0,"2025-08-26","2025-08-26"],[11,3747,"accelerate does not load from the checkpoint properly","wilnn",2,"","",0,"2025-08-25","2025-10-04"],[11,3746,"Does FSDP with Accelerate support async checkpoint saving (e.g. via torch.distributed.checkpoint)?","LucienXian",2,"","",0,"2025-08-25","2025-10-04"],[11,3741,"accelerate.prepare() will consume much NPU memory","oym1994",2,"","",0,"2025-08-21","2025-09-28"],[11,3738,"Accelerate launch with Zero3, Model Parallelism in the same machine when multinode train.","591582523h",2,"","",0,"2025-08-19","2025-08-22"],[11,3735,"CUDA Out of Memory using 2 GPUs","dmitrijlabetskij",2,"","",0,"2025-08-15","2025-09-22"],[11,3734,"Model saved using deepspeed and accelerate cannot be loaded or incomeplete","manitadayon",2,"","",0,"2025-08-15","2025-09-22"],[11,3732,"FSDP Sharded peft classifer saving broken","ojh31",2,"","",0,"2025-08-13","2025-09-22"],[11,3730,"int8 quantization for Lora finetuning with Deepspeed Zero3 is throwing error","joeltalex",2,"","",0,"2025-08-12","2025-09-20"],[11,3729,"FSDP V2 + TP give DeviceMesh Error","eljandoubi",2,"","",0,"2025-08-12","2025-11-17"],[11,3727,"Accelerator.init_trackers error when deal with a custom object of list","micklexqg",2,"","",0,"2025-08-10","2025-09-17"],[11,3726,"Can devices be specified in distributed training","Run-19",2,"","",0,"2025-08-09","2025-09-22"],[11,3724,"Missing release note: accelerate==1.10.0 now requires pytorch >= 2.2","Anturex",2,"","",0,"2025-08-08","2025-09-15"],[11,3723,"accelerate==1.10.0 and safetensors==0.6.1 are incompatible with transformers==4.53.1","ani-bg",2,"","",0,"2025-08-08","2025-09-15"],[11,3718,"Getting \"fp16 mixed precision requires a GPU (not 'mps').\" error running ASR task recipe","renet10",2,"","",0,"2025-08-06","2025-09-08"],[11,3717,"Failed accelerate test with multi node.","DK-DARKmatter",2,"","",0,"2025-08-06","2025-09-14"],[11,3711,"When using InitProcessGroupKwargs to set nccl timeout, in which specific operations/waiting processes will it ","xliu0105",2,"","",0,"2025-08-05","2025-09-12"],[11,3710,"FSDP2 FP8 not giving same results as displayed.","Manan17",2,"","",0,"2025-08-05","2025-09-12"],[11,3709,"`Accelerator.main_process_first` will cause Deepspeed ZeRO3 to hang","universuen",2,"","",0,"2025-08-04","2025-09-11"],[11,3708,"Detected kernel version 5.4.250, which is below the recommended minimum of 5.5.0;","tangzhiling",2,"","",0,"2025-08-04","2025-09-11"],[11,3706,"Dynamic dataloader with accelerator .prepare()","sleepymalc",2,"","",0,"2025-08-03","2025-09-11"],[11,3705,"[Enhancement] Adding parallelism info to Accelerator for parallelism logging in W&B","WoosungMyung",2,"","",0,"2025-08-03","2025-08-04"],[11,3704,"UserWarning: Found Missing Adapter keys while loading checkpoints","manitadayon",2,"","",0,"2025-08-02","2025-09-11"],[11,3702,"[ Enhancement] Getter methods for parallel ranks (TP/DP/CP) to Accelerator class","WoosungMyung",2,"","",0,"2025-08-01","2025-08-02"],[11,3699,"只使用accelerate数据并行，不用模型并行","Seifon",2,"","",0,"2025-07-30","2025-09-06"],[11,3697,"save top-k checkpoints","leomqyu",2,"","",0,"2025-07-28","2025-08-01"],[11,3696,"Package should depend on safetensors with torch extra","xanderlent",2,"","",0,"2025-07-27","2025-11-23"],[11,3694,"Add an external option to toggle offload_to_cpu","scissorstail",2,"","",0,"2025-07-25","2025-09-03"],[11,3693,"logger message \"dataset had no length\" confusing when `drop_last=True`","lpjiang97",2,"","",0,"2025-07-23","2025-08-31"],[11,3691,"Advice with GPU computations in collator/pre-process","saikoneru",2,"","",0,"2025-07-22","2025-08-30"],[11,3690,"[DeepSpeed] GPU VRAM usage increases each validation step","RJExtrac",2,"","",0,"2025-07-22","2025-08-30"],[11,3689,"What will happened if I use mix_precision but don't add with accelerator.autocast():","huangfu170",2,"","",0,"2025-07-22","2025-08-30"],[11,3685,"dtype issue when using accelerate","ZhangYi1999",2,"","",0,"2025-07-18","2025-08-27"],[11,3683,"bugs in accelerator.unwrap_model","ThisisBillhe",2,"","",0,"2025-07-16","2025-10-14"],[11,3680,"When using deepspeed, setting timeout is invalid","xliu0105",2,"","",0,"2025-07-15","2025-07-17"],[11,3679,"Cann't do DDP static_graph=True with gradient accumulation","juexZZ",2,"","",0,"2025-07-11","2025-07-12"],[11,3678,"[REQUEST] Exposing Parallelism rank as environment Variable for integration with profiling tools like CUPTI","WoosungMyung",2,"","",0,"2025-07-11","2025-08-02"],[11,3676,"deepspeed test fails with datasets=4.0.0 release","stas00",2,"","",0,"2025-07-10","2025-07-11"],[11,3672,"a bug in remove_hook_from_module","zouzhekang",0,"sayakpaul","#11896",1,"2025-07-08","2025-07-10"],[11,3670,"Unable to save/load optimizer state after accelerator.prepare()","fkggggg",2,"","",0,"2025-07-03","2025-07-17"],[11,3668,"weight is not tied within `init_empty_weights`","Ghost-LZW",2,"","",0,"2025-07-02","2025-08-18"],[11,3667,"Followed all instructions from multiple tutorials and I just get a blank comfy UI and no latentsync node, erro","simplyexperimental",2,"","",0,"2025-07-01","2025-07-01"],[11,3666,"Deferred initialization support","jiayuanmark",2,"","",0,"2025-07-01","2025-08-25"],[11,3665,"ValueError: ~/.cache/huggingface/hub/models--mistralai--Mixtral-8x7B-Instruct-v0.1/snapshots/41bd4c9e7e4fb318c","wenhao89",2,"","",0,"2025-07-01","2025-08-09"],[11,3664,"Multi-XPU launch error","Xiao-Chenguang",2,"","",0,"2025-06-29","2025-08-30"],[11,3663,"accelerate prepare hangs when training two models from epoch 2 onwards","mmjwxbc",3,"","",0,"2025-06-29","2025-06-29"],[11,3662,"When using accelerator.save_state, the state of all processes is not fully saved","xliu0105",2,"","",0,"2025-06-28","2025-07-14"],[11,3661,"Gradient accumulation step calculation","khalil-Hennara",2,"","",0,"2025-06-27","2025-06-27"],[11,3658,"The Socket Timeout is displayed when a PPO is trained on multiple machines","duyuwen-duen",2,"","",0,"2025-06-26","2025-08-18"],[11,3655,"Why does cuda:0 cause memory explosion when creating model and loading weights?","xliu0105",2,"","",0,"2025-06-25","2025-06-25"],[11,3652,"Transformer Engine memory-efficient initialization to convert_model for large models","mayukh-stackav",2,"","",0,"2025-06-24","2025-08-25"],[11,3647,"Why does torchtune FSDP have a lower memory footprint than accelerate?","jasper-lu",2,"","",0,"2025-06-20","2025-06-28"],[11,3644,"torch._dynamo.exc.Unsupported: generator","anicolson",2,"","",0,"2025-06-20","2025-08-03"],[11,3643,"Typo in dataloader in newest version","TonyCongqianWang",2,"","",0,"2025-06-20","2025-07-22"],[11,3641,"NameError: name 'is_torch_version' is not defined","alterdim",2,"","",0,"2025-06-20","2025-06-21"],[11,3639,"[Docs] Update low-precision training docs for MS-AMP","shimizust",0,"ManasVardhan","#3929",0,"2025-06-19","2026-03-02"],[11,3634,"Does Accelerate support PyTorch Weight Averaging (SWA and EMA)?","yeruouo",2,"","",0,"2025-06-18","2025-07-27"],[11,3633,"how to save a model with FSDP2 ?","colinzhaoxp",0,"S1ro1","#3636",0,"2025-06-18","2025-06-18"],[11,3628,"[rank0]: AssertionError: found no DeviceMesh from dtensor args for c10d.broadcast_.default!","lucasjinreal",2,"","",0,"2025-06-14","2025-06-15"],[11,3624,"dataloader with IterableDataset cannot work after accelerator.prepare()","leeruibin",2,"","",0,"2025-06-12","2025-07-21"],[11,3620,"Final output is upcasted to float32 even though I used mixed_precision=\"fp16\"","antoinedelplace",2,"","",0,"2025-06-11","2025-07-17"],[11,3617,"Exception \"NotImplementedError: Cannot copy out of meta tensor; no data! Please use torch.nn.Module.to_empty()","edmcman",2,"","",0,"2025-06-09","2025-07-21"],[11,3616,"sync between multi gpus","Reginald-L",2,"","",0,"2025-06-09","2025-07-26"],[11,3614,"Error in concatenating the last batch with the first batch when dimensions of batches are different","Boltzmachine",2,"","",0,"2025-06-07","2025-07-23"],[11,3612,"RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:1 and cuda:0","Indigo-Coder-github",2,"","",0,"2025-06-07","2025-06-11"],[11,3607,"Tracking with `WANDB_MODE=\"offline\"` will not pass through the config dict","SamD770",0,"ved1beta","#3648",0,"2025-06-04","2025-07-15"],[12,4327,"[Proposal] Nano evaluators for any Nano-set in the NanoBEIR format","hotchpotch",4,"","",0,"2026-11-05",""],[12,4299,"Add KaLM-Reranker-V1-R2 models to the Sentence Transformers integrations collection","cosylsine",2,"","",0,"2026-10-30","2026-10-30"],[12,4269,"[Docs] CrossEncoder.predict() already applies sigmoid for num_labels=1 — warn against applying it again","hgx555",2,"","",0,"2026-10-25","2026-10-26"],[12,4263,"Discussion: Native OpenTelemetry Instrumentation for Sentence Transformers","piyush182004",4,"","",0,"2026-10-23",""],[12,4258,"Add an optional RaBitQ IVF backend for semantic search","gouyt13",4,"","",0,"2026-10-23",""],[12,4256,"Feature request: optional fused kernels for MaxSim and SPLADE","capemox",4,"","",0,"2026-10-22",""],[12,4240,"gradient_checkpointing=True crashes with transformers >= 5.16: BaseModel.gradient_checkpointing_enable() got a","OsamaS99",0,"tomaarsen","#4241",1,"2026-10-21","2026-10-21"],[12,4238,"`fit()` silently drops the hard negatives from every example when one InputExample has fewer texts","zikrulihsan",1,"zikrulihsan","#4239",0,"2026-10-21","2026-10-26"],[12,4232,"push_to_hub fails with \"widget … is not allowed to be empty\" when the training data contains an empty text","zikrulihsan",1,"zikrulihsan","#4233",0,"2026-10-19","2026-10-20"],[12,4227,"SentenceTransformer.fit() crashes with DataLoader(batch_sampler=...) and silently ignores custom batch sampler","zikrulihsan",1,"zikrulihsan","#4228",0,"2026-10-19","2026-10-19"],[12,4206,"Proposed bug fixes and PR tracking","Bembaby",2,"","",0,"2026-10-17","2026-10-19"],[12,4186,"MSEEvaluator silently broadcasts unequal source and target sentence counts","leeshengis",1,"leeshengis","#4187",0,"2026-10-16","2026-10-16"],[12,4185,"v6: models with `Normalize` at the model root (path \"\" in modules.json) fail to load: `Normalize.__init__() go","tritsystem",0,"tomaarsen","#4190",1,"2026-10-16","2026-10-16"],[12,4165,"PListMLE default position weights overflow on long ranking lists","kocean9-freedom",1,"kocean9-freedom","#4166",0,"2026-10-13","2026-10-13"],[12,4152,"`MultiVectorEncoder.encode`: no way to offload token embeddings per batch without converting to numpy","Samoed",2,"","",0,"2026-10-10","2026-10-13"],[12,4151,"`maxsim` pads all queries to the longest one","Samoed",0,"Arthur031221","#4153",0,"2026-10-10","2026-10-13"],[12,4134,"Cached losses slice stacked (batch-first) image tensors by row, breaking EmbeddingGemma 2","jinwoo-cho-ilevit-com",0,"arcusbuilds","#4157",0,"2026-10-07","2026-10-13"],[12,4125,"[Feature Request] Optional probability-target validation for SoftmaxLoss","amishaguptaberk",2,"","",0,"2026-10-05","2026-10-14"],[12,4107,"StaticEmbedding ignores model2vec `mapping`/`weights` tensors, so e.g. `minishlab/potion-code-16M` produces wr","JoeyTan21",1,"JoeyTan21","#4108",0,"2026-10-01","2026-10-08"],[12,4105,"weightedmean pooling produces NaNs for 512-token float16 inputs","AHMETHAKANBEZIR1",1,"AHMETHAKANBEZIR1","#4106",0,"2026-10-01","2026-10-06"],[12,4103,"Weighted mean pooling changes with left padding","AHMETHAKANBEZIR1",1,"AHMETHAKANBEZIR1","#4104",0,"2026-10-01","2026-10-12"],[12,4097,"Similarity utilities reject dense tensor lists returned by encode","LimbC-C",1,"LimbC-C","#4098",0,"2026-09-30","2026-10-08"],[12,4090,"encode_document() never uses a \"passage\" or \"corpus\" prompt","LindseyZ1205",4,"","",0,"2026-09-26",""],[12,4081,"[Bug] GradCache replay downcasts FP64 embeddings to FP32 in the surrogate dot product","gss10282025",1,"gss10282025","#4082",0,"2026-09-26","2026-10-07"],[12,4075,"Mask/tensor shape mismatch when using input flattening and Flash Attention 2","marianrh-apa",0,"palanigsharma182-ctrl","#4079",0,"2026-09-24","2026-10-07"],[12,4074,"Source link in quantize_embeddings have shifted code highlights","mitanshu7",0,"j341nono","#4280",0,"2026-09-24","2026-10-26"],[12,4070,"`AttributeError: 'str' object has no attribute 'to'` in CachedGISTEmbedLoss","marianrh-apa",2,"","",0,"2026-09-23","2026-10-19"],[12,4068,"compute_count_vector counts explicit sparse zeros as nonzero values","HuaTNA",2,"","",0,"2026-09-23","2026-10-15"],[12,4060,"Inputs are silently truncated at `max_seq_length` with no way for the caller to detect it","ErnisBadmae",4,"","",0,"2026-09-21",""],[12,4059,"[Feature Request] Jev-style structured judgments for CrossEncoder — working implementation in KaLM-Jev","ItsukiFujii",4,"","",0,"2026-09-21",""],[12,4055,"StaticEmbedding.from_distillation ignores auto-detected device","Anu27n",1,"Anu27n","#4056",0,"2026-09-20","2026-10-07"],[12,4027,"Cached losses (GradCache) trigger one DDP all-reduce per mini-batch per column instead of one per step","BramVanroy",0,"tomaarsen","#4039",1,"2026-09-15","2026-10-07"],[12,4010,"model.safetensors is missing read permissions after saving with SentenceTransformer.save()","BestCrazyNoob",2,"","",0,"2026-09-12","2026-10-07"],[12,3996,"No constructor-based way to set query_length/document_length on `MultiVectorEncoder`","robro612",4,"","",0,"2026-09-09",""],[12,3991,"Problems in training multi-modal models with large batch sizes","benHeid",4,"","",0,"2026-09-08",""],[12,3980,"Loading a LoRA CrossEncoder checkpoint does not restore a trained `modules_to_save` classification head","shniubobo",4,"shniubobo","#48595",0,"2026-09-04",""],[12,3973,"Backend ignores root level ONNX files, and cannot detect full model exports","tomaarsen",4,"","",0,"2026-08-31",""],[12,3953,"Let a module declare an extra output type (e.g. multi-vector + sparse from one model)","davidmyriel",4,"","",0,"2026-08-18",""],[12,3950,"[`docs`] 'Performance ratio' in efficiency.rst GPU/CPU performance images isn't clearly about NDCG@10","tomaarsen",1,"tomaarsen","#3977",1,"2026-08-18","2026-09-01"],[12,3939,"`MultiVectorInformationRetrievalEvaluator` keeps the first model’s similarity function when reused","eSVeeF",0,"tomaarsen","#3976",1,"2026-08-16","2026-09-01"],[12,3929,"paraphrase_mining_embeddings keeps max_pairs - 1 candidates","FU-max-boop",1,"FU-max-boop","#3930",0,"2026-08-13","2026-08-31"],[12,3923,"Docs: deployment guidance for cold starts and offline model loading (scale-to-zero, local_files_only, warmup)","jgjoe",4,"","",0,"2026-08-11",""],[12,3916,"CrossEncoder PEFT fine-tuning: confusing TypeError (BatchEncoding not Tensor) with manual get_peft_model; no o","XXXM1R0XXX",4,"","",0,"2026-08-07",""],[12,3908,"`RuntimeError: _apply(): Couldn't swap Linear.weight` when `torchao` with `torch.compile`","NewUserHa",2,"","",0,"2026-08-05","2026-09-15"],[12,3903,"mine_hard_negatives crashes on a tensor shape when num_negatives exceeds the range_min/range_max window","LK-maker-007",1,"LK-maker-007","#3904",0,"2026-08-04","2026-08-06"],[12,3899,"community_detection truncates a community at sort_max_size when similarities equal the threshold","LK-maker-007",1,"LK-maker-007","#3900",0,"2026-08-04","2026-08-04"],[12,3896,"sentence-transformers>=5.0 hard-imports torchcodec at package load, crashing text-only usage when torchcodec/F","MHammett",0,"mturac","#3897",0,"2026-08-03","2026-08-04"],[12,3894,"Similarity and cross-encoder scores computed below float32 create spurious ties in low-precision inference","KisuYang",1,"KisuYang","#3892",0,"2026-07-31","2026-08-12"],[12,3889,"Add the KaLM-Reranker-V1 family to the Sentence Transformers v5.4 integrations collection","cosylsine",2,"","",0,"2026-07-29","2026-08-04"],[12,3885,"Feature request for encoder-decoder reranker","ItsukiFujii",2,"","",0,"2026-07-28","2026-08-04"],[12,3884,"Architectural Review of test_transformer.py","amirii2025amirii-beep",2,"","",0,"2026-07-27","2026-07-27"],[12,3881,"CrossEncoder silently drops the whole text pair when the backbone's chat template ignores the query/document r","ishan-1010",0,"tomaarsen","#3902",1,"2026-07-26","2026-08-04"],[12,3874,"Per-sample `video_metadata` gets attached to the wrong video when a batch mixes dict-wrapped and raw videos","meutsabdahal",1,"meutsabdahal","#3876",0,"2026-07-21","2026-08-03"],[12,3870,"mine_hard_negatives: cache_folder key ignores query_prompt/corpus_prompt, silently returns stale embeddings","ErenAta16",1,"ErenAta16","#3871",1,"2026-07-17","2026-07-20"],[12,3864,"Multiple videos and metadata process in one time","NewUserHa",4,"","",0,"2026-07-16",""],[12,3851,"High RSS growth in SparseEncoder.encode_document() when convert_to_sparse_tensor=True","anfedoro",2,"","",0,"2026-07-07","2026-07-08"],[12,3850,"infer_modality() raises ValueError on malformed URL-like strings and breaks SentenceTransformer.encode()","IPanfilov",2,"","",0,"2026-07-07","2026-07-07"],[12,3849,"Problem using CrossEncoder.rank with jina-reranker-v3","PeteBleackley",4,"","",0,"2026-07-07",""],[12,3845,"Proposal: Add hardness weighting to GISTEmbedLoss, with CachedGISTEmbedLoss parity, following #3667","Incheonkirin",4,"","",0,"2026-07-02",""],[12,3842,"[Bug] AdaLora silently acts as standard LoRA because update_and_allocate method is missing from wrapped model","MateoRivera",2,"","",0,"2026-06-29","2026-06-30"],[12,3840,"Passing a list of images as video can work in `transformers` but not in `sentence transformer`","NewUserHa",4,"","",0,"2026-06-27",""],[12,3839,"Docs: add production deployment guide for async servers (FastAPI/uvicorn + OpenMP deadlock)","Shubh3005",4,"","",0,"2026-06-27",""],[12,3836,"`token` not forwarded to `AutoProcessor.from_pretrained()` in `Transformer.__init__` — breaks private/gated mo","tstadel",2,"","",0,"2026-06-26","2026-06-29"],[12,3830,"Optimum ONNX Export with SentenceTransformer results in property 'config' has no setter","chrisss404",0,"lcheng321","#3831",0,"2026-06-19","2026-07-06"],[12,3822,"Explicit device assignment ignored in multi-GPU (PyTriton) environment","alejandrojcastaneira",4,"","",0,"2026-06-15",""],[12,3819,"relative_margin can keep negatives more similar than positives when positive-pair scores are negative","Incheonkirin",1,"Incheonkirin","#3821",0,"2026-06-13","2026-06-16"],[12,3804,"[Security] Arbitrary code execution when loading a local model directory","geo-chen",4,"","",0,"2026-06-08",""],[12,3801,"Remote Code Execution via trust_remote_code=False Bypass (Code Injection Vulnerability)","CFionaBF",0,"tomaarsen","#3807",1,"2026-06-06","2026-06-12"],[12,3795,"AttributeError: module 'torch.distributed' has no attribute 'is_initialized' when using ROCm‑based PyTorch","Rhyx14",0,"swankystark","#3798",0,"2026-06-04","2026-06-08"],[12,3793,"Reranking listwise loss occur float and bfloat16 conflict","LetheHS",0,"Incheonkirin","#3800",0,"2026-06-03","2026-06-08"],[12,3778,"Error when saving checkpoints while training jina-embeddings-v5-text-nano with Sentence-Transformers","ghzamani",4,"","",0,"2026-05-17",""],[12,3759,"运行Qwen3-Reranker报错","yvway",4,"","",0,"2026-05-09",""],[12,3758,"Invalid Markdown URL causes a runtime crash","niekvandepas",2,"","",0,"2026-05-08","2026-05-11"],[12,3757,"Question regarding KL Divergence gradient flow in `AdaptiveLayerLoss` (edit: and possible issue with layers we","bobox2997",0,"tomaarsen","#3901",1,"2026-05-08","2026-08-04"],[12,3756,"Resuming training restarts at much higher loss","conceptofmind",4,"","",0,"2026-05-08",""],[12,3748,"Feature request: support for RetroMAE unsupervised training","bobox2997",4,"","",0,"2026-05-03",""],[12,3747,"Adding STS datasets to the Hub","ir2718",2,"","",0,"2026-05-02","2026-05-08"],[12,3743,"Custom batches for training with in-batch negatives (MultipleNegativesRankingLoss)","lmores",2,"","",0,"2026-04-27","2026-05-07"],[12,3737,"`DenoisingAutoEncoderLoss` and Transformers v5","Noobzik",0,"omkar-334","#3781",0,"2026-04-20","2026-06-02"],[12,3732,"`Audio/VideoDict` and torchcodec objects cannot be processed by `chat_template`","Samoed",0,"tomaarsen","#3736",1,"2026-04-18","2026-04-20"],[12,3725,"embeddinggemma-300m: silently uses causal attention on transformers < 4.57, producing wrong embeddings","nilichen",0,"tomaarsen","#3726",1,"2026-04-15","2026-04-15"],[12,3724,"fix_mistral_regex=True","conceptofmind",0,"tomaarsen","#45444",1,"2026-04-14","2026-04-16"],[12,3723,"Can't instantiate a processor","conceptofmind",2,"","",0,"2026-04-14","2026-04-17"],[12,3722,"Mixing modalities in `encode()` doesn't work for CLIP","x-tabdeveloping",4,"","",0,"2026-04-13",""],[12,3721,"Multimodal update breaks encode() when strings are passed in NumPy arrays","x-tabdeveloping",2,"","",0,"2026-04-13","2026-04-13"],[12,3718,"Bug: `infer_modality` misclassifies 1D numpy string arrays as audio, raising `ValueError: Modality 'audio' is ","idodoron11",0,"tomaarsen","#3720",1,"2026-04-12","2026-04-13"],[12,3717,"Error when using models \"jinaai/jina-embeddings-v5-text-nano\" and \"Alibaba-NLP/gte-base-en-v1.5\"","nicheleth",2,"","",0,"2026-04-11","2026-05-08"],[12,3715,"Faile to import due to missing requests module","sejohnst",2,"","",0,"2026-04-10","2026-05-08"],[12,3702,"Euclidean distance is inverted, breaking TripletLoss default values","langdonholmes",0,"tomaarsen","#3704",1,"2026-04-06","2026-04-09"],[12,3701,"Issue when resuming from LORA checkpoint","conceptofmind",4,"","",0,"2026-04-03",""],[12,3696,"Save individual routes from a Router-based model as standalone artifacts","ignasgr",2,"","",0,"2026-03-29","2026-10-27"],[12,3684,"Different batch sizes per dataset/loss (CachedMNRL + CoSENT) in multi-dataset training","ddofer",4,"","",0,"2026-03-10",""],[12,3683,"Security: Unrestricted Module Import Bypasses trust_remote_code=False (CWE-94)","spartan8806",2,"","",0,"2026-03-09","2026-03-10"],[12,3682,"IndexError: list index out of range","zkailinzhang",2,"","",0,"2026-03-05","2026-05-08"],[12,3679,"Facing HFValidation Error when trying to load embedding model","prathamgec21-hue",2,"","",0,"2026-03-02","2026-03-03"],[12,3674,"Can CrossEncoderTrainer support IterableDataset?","nursery42",0,"tomaarsen","#3676",1,"2026-02-26","2026-02-26"],[12,3671,"Constructive criticism for Sentence Transformers.","BBC-Esq",4,"","",0,"2026-02-23",""],[12,3666,"ONNX Export Support for jinaai/jina-embeddings-v5-text-nano-retrieval","contrebande-labs",4,"","",0,"2026-02-19",""],[12,3661,"Sentence Transformers 5.2.2 on Mac Intel/CPU torch>=2.6 not available","michaelwechner",2,"","",0,"2026-02-12","2026-02-17"],[12,3650,"Update docstring for `DenoisingAutoEncoderLoss.py`","omkar-334",2,"","",0,"2026-02-02","2026-02-06"],[12,3640,"Masked elements can be selected as in-batch negatives in `MultipleNegativesRankingLoss`","fuutot",1,"fuutot","#3641",0,"2026-01-30","2026-01-30"],[12,3633,"`MultipleNegativesRankingLoss`: `num_negatives=None` raises an error","fuutot",1,"fuutot","#3636",0,"2026-01-29","2026-01-29"],[12,3622,"Feature Request: Introduce global orthogonal regularizer (GOR) loss for `SentenceTransformer` models","tomaarsen",1,"tomaarsen","#3654",1,"2026-01-27","2026-02-05"],[12,3621,"Update example scripts relying on `model.fit` to use `SentenceTransformerTrainer` instead","tomaarsen",1,"tomaarsen","#3761",1,"2026-01-27","2026-05-10"],[12,3620,"Update example scripts relying on `http_get` to download from the Hugging Face Hub instead","tomaarsen",1,"tomaarsen","#3761",1,"2026-01-27","2026-05-10"],[12,3617,"Missing `requests` dependency in v5.2.1","csadorf",0,"tomaarsen","#3618",1,"2026-01-26","2026-01-27"],[12,3616,"sentence-transformers package incompatibility started today 26/1/2026","MauricioCafiero",2,"","",0,"2026-01-26","2026-01-28"],[12,3606,"BUG: `tests/samplers` failures in Python 3.14 due to changed `Pickler._batch_setitems()` definition","sarahec",2,"","",0,"2026-01-05","2026-01-09"],[12,3605,"Low performance when importing the library","Social-Mean",2,"","",0,"2026-01-04","2026-05-08"],[12,3602,"Support for T5Gemma2 Encoder","contrebande-labs",2,"","",0,"2025-12-23","2026-02-18"],[12,3601,"how to finetuning a bi-encoder embedding model of multimodel input","fancyerii",2,"","",0,"2025-12-22","2026-05-08"],[12,3599,"Make progress bar in `SentenceTransformer.encode()` configurable","lbhm",2,"","",0,"2025-12-19","2026-10-26"],[12,3585,"How to choose negative instance when using MultipleNegativesRankingLoss train embedding model?","4daJKong",2,"","",0,"2025-12-05","2026-05-08"],[12,3584,"different with transformers implement","bjtuxck",2,"","",0,"2025-12-04","2026-05-08"],[12,3582,"Collator setting include_prompt_lengths prevents compilation","chschroeder",0,"RizgarOzan","#4015",0,"2025-11-29","2026-09-15"],[12,3581,"MultipleNegativesRankingLoss issue","yfdjchvlgv",4,"","",0,"2025-11-23",""],[12,3576,"Documentation - losses problem","MLKoz2",2,"","",0,"2025-11-20","2025-11-20"],[12,3575,"How to override model's `max_seq_length`?","Samoed",2,"","",0,"2025-11-19","2026-05-08"],[12,3574,"Downloading models freezes at 0 downloading model.safetensors","lyons7",2,"","",0,"2025-11-17","2025-11-17"],[12,3571,"Peft package update causing import issues","FabianBnvc",2,"","",0,"2025-11-14","2025-11-14"],[12,3570,"Consideration on dynamic batch sizes","KennethEnevoldsen",2,"","",0,"2025-11-13","2025-11-25"],[12,3569,"datasets 4.0.0 compatibility","mengerj",2,"","",0,"2025-11-13","2026-05-08"],[12,3568,"Speed degradation over training","ZeusFSX",4,"","",0,"2025-11-12",""],[12,3566,"Replicating cross-encoder/ms-marco-MiniLM-L12-v2","MathVast",2,"","",0,"2025-11-10","2025-11-24"],[12,3564,"CachedMultipleNegativesRankingLoss + MPS is broken","morrisalp",0,"Incheonkirin","#3812",0,"2025-11-07","2026-06-12"],[12,3563,"Set the eos_token_id to 151645, the output of model.tokenizer is 151643","springchuntian",2,"","",0,"2025-11-06","2026-05-08"],[12,3562,"ONNX backend does not work when cache_folder is specified and local_files_only is True","lcorcodilos",2,"","",0,"2025-11-05","2026-04-21"],[12,3561,"Confusing message when not having enough space on device","KennethEnevoldsen",2,"","",0,"2025-11-04","2026-05-08"],[12,3559,"Disable all progress bars - how?","rq-aszasz",2,"","",0,"2025-11-03","2026-05-08"],[12,3558,"Support of `LiquidAI/LFM2-ColBERT-350M`","MohamedAliRashad",4,"","",0,"2025-11-02",""],[12,3557,"Using megablocks causes RuntimeError: 0 active drivers ([]). There should only be one.","KIC",2,"","",0,"2025-10-28","2026-05-08"],[12,3556,"Evaluator runs on all devices","ShreyGanatra",0,"tomaarsen","#4004",1,"2025-10-28","2026-09-15"],[12,3552,"chore: Drop Python 3.9 support","tomaarsen",2,"","",0,"2025-10-21","2025-11-25"],[12,3549,"accelerate package not automatically installed when needed?","Andrew-Chen-Wang",2,"","",0,"2025-10-16","2026-05-08"],[12,3547,"Unit test (test_safetensors) fails in Windows but passes in Linux","kaushikacharya",2,"","",0,"2025-10-12","2026-05-08"],[12,3546,"Bug: `ImportError` from `from sentence_transformers import SentenceTransformer`  — `.backend` issue","chonknick",2,"","",0,"2025-10-11","2026-05-08"],[12,3545,"Getting different result compared to paper when evaluating SparseEncoder (splade-v3)","yjoonjang",4,"","",0,"2025-10-11",""],[12,3544,"Error When Downloading New Model from Hub","k1bijari",2,"","",0,"2025-10-11","2025-10-16"],[12,3540,"Should there be a scaling method for MarginMSE?","yjoonjang",4,"","",0,"2025-10-05",""],[12,3536,"Wrong defaults used when loading older non-mean-pooled models via subclass","kddubey",0,"tomaarsen","#3763",1,"2025-10-02","2026-05-11"],[12,3532,"What is the proper way to use prompts? Do we have to format/render them ourselves?","seanswyi",2,"","",0,"2025-09-28","2025-09-30"],[12,3531,"Can't use Deepspeed in Sentence-transformers","mrpeerat",0,"tomaarsen","#3764",1,"2025-09-27","2026-05-11"],[12,3526,"numpy not available","gg4u",4,"","",0,"2025-09-25",""],[12,3523,"AttributeError: 'Transformer' object has no attribute 'model_forward_params'","AmitPoonia",2,"","",0,"2025-09-23","2026-05-11"],[12,3522,"Custom Cross Encoder model training","ShreyGanatra",4,"","",0,"2025-09-22",""],[12,3520,"tests fail with: ModuleNotFoundError: No module named 'tests.utils'","yurivict",4,"","",0,"2025-09-19",""],[12,3519,"tokenizers isn't listed in dependencies in pyproject.toml","yurivict",0,"aadhar-build","#3844",0,"2025-09-19","2026-07-06"],[12,3518,"MultipleNegativesRankingLoss applies Sigmoid to logits before CE by default - rationale vs. InfoNCE literature","roycho96",4,"","",0,"2025-09-19",""],[12,3517,"Let Evaluators return or store the generated embeddings.","benHeid",4,"","",0,"2025-09-16",""],[12,3515,"Not able to export Qwen/Qwen3-Embedding-0.6B with Openvino backend","santhoshtr",4,"","",0,"2025-09-08",""],[12,3514,"Jupyter Kernel Crash when importing sentence_transformers & faiss and/or hnswlib","mkleinbort-wl",2,"","",0,"2025-09-06","2025-09-06"],[12,3513,"[BUG] google/embeddinggemma-300m does not support bidirectional attention","noooop",2,"","",0,"2025-09-05","2025-09-05"],[12,3512,"[BUG] Error in serialization when writing results to csv using Sparse Encoder Information Retrieval evaluator","nadshe",4,"","",0,"2025-09-03",""],[12,3511,"encode(device): docs say List[str] allowed for multi-process, but type hints show str | None and passing a lis","haddyadnan",2,"","",0,"2025-09-03","2025-09-04"],[12,3510,"Question on model loading","KennethEnevoldsen",4,"","",0,"2025-09-02",""],[12,3507,"Different Embeddings produced on mps with torch 2.8","jonathanasdf",2,"","",0,"2025-08-31","2025-10-22"],[12,3505,"Hyperparameters of original cross-encoder/ms-marco-MiniLM series models?","Liwen-Hu3",4,"","",0,"2025-08-29",""],[12,3499,"Cannot install sentence-transformers because package versions have conflicting requirements (on MAC, Python 3.","rweintr-t",2,"","",0,"2025-08-25","2025-09-02"],[12,3498,"Qwen3 produces `nan` embeddings (SDPA + macOS)","alvarobartt",4,"","",0,"2025-08-25",""],[12,3493,"Support for jinaai/jina-reranker-m0","wirthual",4,"","",0,"2025-08-20",""],[12,3492,"Support for listwise loss functions in bi-encoder","madhumitha-mohan",4,"","",0,"2025-08-19",""],[12,3491,"[mutex.cc : 452] RAW: Lock blocking when trying to load with AutoModel","fastier-atlassian",2,"","",0,"2025-08-18","2025-08-19"],[12,3490,"Support for using MultipleNegativesRankingLoss with SentenceTransformerTrainer (evaluation, early stopping, ev","pinareceaktan",2,"","",0,"2025-08-18","2025-08-20"],[12,3489,"Installing specified pretrained models","Lidne",2,"","",0,"2025-08-18","2025-08-20"],[12,3487,"Iterable dataset for evaluator","sasakiyori",2,"","",0,"2025-08-15","2025-08-18"],[12,3486,"encode_query multi gpu compute embeddings not working for router module","zionsoumik",0,"tomaarsen","#3488",1,"2025-08-11","2025-08-27"],[12,3484,"Using KLDiv with CMNRL","yjoonjang",4,"","",0,"2025-08-07",""],[12,3483,"hard_negatives copus error","sampathrajapaksha",2,"","",0,"2025-08-06","2025-08-12"],[12,3474,"Dead Link in Contrastive Loss Documentation","johneckberg",0,"tomaarsen","#3476",1,"2025-07-29","2025-07-30"],[12,3472,"SPLADE-Index Python Library: An ultra-fast search index for SPLADE sparse retrieval models","rasyosef",2,"","",0,"2025-07-28","2025-08-20"],[12,3471,"ColBERT-style token-level retrieval vs sentence embeddings for modern RAG tasks (esp. multimodal like ColPali)","HuangBugWei",4,"","",0,"2025-07-28",""],[12,3469,"Embedding space collapsing after training using cosine similarity loss","rishabh16196",4,"","",0,"2025-07-26",""],[12,3468,"TypeError when evaluating similarity model on A100 GPUs using EmbeddingSimilarityEvaluator (works on CPU)","joaograndotto",2,"","",0,"2025-07-25","2025-07-25"],[12,3465,"error handling for empty input in cross_encoder.predict","nitin-nsp",2,"","",0,"2025-07-25","2025-12-16"],[12,3464,"CrossEncoderRerankingEvaluator with non-binary labels","rupeshgx",2,"","",0,"2025-07-24","2025-07-29"],[12,3462,"Image Search example raises a mypy error","anakin87",2,"","",0,"2025-07-23","2026-04-10"],[12,3461,"Confusing `model_max_length` information for `paraphrase-multilingual-MiniLM-L12-v2`","next-game-solutions",2,"","",0,"2025-07-23","2025-07-24"],[12,3460,"Question about fine-tuning LLM-based retrievers","dkoterwa",4,"","",0,"2025-07-23",""],[12,3459,"Disable transformers evaluation when evaluator is provided to `SentenceTransformerTrainer`","emapco",2,"","",0,"2025-07-22","2025-07-24"],[12,3457,"Using sentence-transformers with TPU: RuntimeError: Cannot set version_counter for inference tensor","wirthual",4,"","",0,"2025-07-22",""],[12,3452,"Router training giving errors","zionsoumik",0,"tomaarsen","#3454",1,"2025-07-19","2025-07-22"],[12,3451,"Add SparseCachedMultipleNegativesRankingLoss","yjoonjang",4,"","",0,"2025-07-19",""],[12,3450,"Drop in Dev NDCG@10 After Training on CodeSearchNet-CCR (Python only): Possible Issue with In-Batch Negatives","gnatesan",4,"","",0,"2025-07-18",""],[12,3449,"Model Distillation Tuples Format - Memory Issue","leobavila",2,"","",0,"2025-07-17","2025-07-22"],[12,3448,"Which models recommend as tsdae decoder?","Ccongggg",4,"","",0,"2025-07-17",""],[12,3443,"util.mine_hard_negatives TypeError: unsupported operand type(s) for +: 'Column' and 'Column'","libyair",0,"tomaarsen","#3445",1,"2025-07-15","2025-07-16"],[12,3441,"VRAM is very high when training with a router","ccdv-ai",0,"tomaarsen","#3454",1,"2025-07-10","2025-07-22"],[12,3440,"The special_tokens in tokenizer should also be controlled by do_lower_case in encoder_config.","noooop",4,"","",0,"2025-07-10",""],[12,3438,"Not able to get optimized onnx model weights file","rkoystart",0,"LK-maker-007","#3913",0,"2025-07-09","2026-08-31"],[12,3437,"ValueError: The model must be a Transformer-based SentenceTransformer or CrossEncoder model loaded with `backe","rkoystart",4,"","",0,"2025-07-09",""],[12,3436,"Invalid credentials in Authorization header","mlang",2,"","",0,"2025-07-09","2025-07-10"],[12,3435,"Training on Intermediate Logits with Cross-Encoder","andreawynn",4,"","",0,"2025-07-08",""],[12,3434,"Cannot continue training on pretrained PEFT Adapter","CisterMoke",4,"","",0,"2025-07-08",""],[12,3433,"How to use a custom batch sampler?","Hypothesis-Z",4,"","",0,"2025-07-08",""],[12,3431,"Having trouble training sparse embedding models","yjoonjang",4,"","",0,"2025-07-08",""],[12,3426,"Details of Benchmark Dataset","Jacobsonradical",4,"","",0,"2025-07-04",""],[12,3425,"Not able to generate onnx optimized and quantized graph for Qwen-3 models.","rkoystart",4,"","",0,"2025-07-04",""],[12,3424,"AttributeError: 'SentenceTransformer' object has no attribute 'add_adapter'","awmoe",4,"","",0,"2025-07-04",""],[12,3423,"guidance on text to code fine tuning","pulkitmehtawork",2,"","",0,"2025-07-04","2025-07-04"],[12,3422,"Can we finetune sbert with both add_adapter and add_module?","awmoe",4,"","",0,"2025-07-04",""],[12,3420,"GradCache Training Curve","OsamaS99",2,"","",0,"2025-07-02","2025-09-22"],[12,3416,"How to calculate prompt tokens for embedding model encode?","gaoxt1983",4,"","",0,"2025-07-02",""],[12,3415,"Different Gradient for Distributed Data Parallel Training","XuanZhangg",4,"","",0,"2025-07-01",""],[12,3414,"How to fine tune multimodal embedding model?","groklab",4,"","",0,"2025-07-01",""],[12,3413,"muvera algorithm for multivectors","raulcarlomagno",4,"","",0,"2025-07-01",""],[12,3411,"requests.exceptions.MissingSchema: Invalid URL distiluse-base-multilingual-cased-v1","IgorSiilva",4,"","",0,"2025-07-01",""],[12,3404,"Any pretrained models available for commercial use?","sidkgp",4,"","",0,"2025-06-28",""],[12,3402,"`resume_from_checkpoint` does not work for models with peft adapter","mjstel",0,"tomaarsen","#3405",1,"2025-06-27","2025-06-30"],[12,3400,"Role of batch_size and mini_batch_size in CrossEncoder","andreawynn",4,"","",0,"2025-06-20",""],[12,3399,"Feature Request: Surfacing Token Level Attention Scores inside the `forward` pass of `Transformer` class.","pramodith",4,"","",0,"2025-06-19",""],[12,3398,"[BUG] CrossEncoder with BGE Reranker V2-M3 giving all similarities as exactly 1","chayankb",2,"","",0,"2025-06-19","2025-11-05"],[12,3397,"Can't figure this out! TOKENIZATION FAILED: Unable to create tensor, you should probably activate truncation a","BBC-Esq",3,"","",0,"2025-06-18","2026-10-12"],[12,3396,"`NotImplementedError:` Cannot copy out of meta tensor when loading model with PyTorch 2.2+ and SentenceTransfo","jwilson-mdmi",4,"","",0,"2025-06-18",""],[12,3395,"dimension model","ixn3rd3mxn",4,"","",0,"2025-06-13",""],[12,3394,"How can I determine which prompt is used during training?","BigPanBaidu",2,"","",0,"2025-06-12","2025-06-16"],[12,3393,"[BUG] Working with gradient_checkpointing enabled","ccdv-ai",4,"","",0,"2025-06-11",""],[12,3392,"include example of using bitsandbytes","BBC-Esq",3,"","",0,"2025-06-11","2026-10-12"],[12,3390,"How to create a customized model architecture that fits sentence-transformer's training framework?","HuangLED",4,"","",0,"2025-06-11",""],[12,3389,"Interest in variable mini-batch size for CMNRL?","mermerico",4,"","",0,"2025-06-10",""],[12,3388,"How should I train instruct-based models like Qwen3 embedding?","yusufcakmakk",2,"","",0,"2025-06-08","2025-06-09"],[12,3387,"Is bi-attention supported when using CausalLM models as embedding models?","daegonYu",4,"","",0,"2025-06-08",""],[12,3386,"Dependency Issue with Numpy","sapphirethunderbird",2,"","",0,"2025-06-08","2025-06-11"],[12,3384,"LoRA Fine-tuning for Cross-Encoders","raulcarlomagno",4,"","",0,"2025-06-05",""],[12,3383,"Qwen 3 Embeddings support","achibb",2,"","",0,"2025-06-05","2025-06-11"],[12,3382,"Contrastive loss with non-binary labels","Garrafao",0,"oyinkanchekwas","#3838",0,"2025-06-02","2026-07-06"],[12,3381,"Error with multinode GPU training in PyTorch with SentenceTransformers","gnatesan",4,"","",0,"2025-06-01",""],[13,1483,"PassAtK underestimates pass@k when fewer than n samples are available","aidless",4,"","",0,"2026-11-06",""],[13,1481,"[EVAL] Add KorMedMCQA: Korean medical licensing exam MCQA","kangmin04",4,"","",0,"2026-11-05",""],[13,1479,"[EVAL] HRM8K: Korean math reasoning (8,011 problems, 5 subsets)","dw021213",4,"","",0,"2026-11-05",""],[13,1476,"[BUG] Numbers grouped with a no-break or narrow no-break space (\"2 500\") are read as only their first or last ","ChrystianSchutz",4,"","",0,"2026-11-05",""],[13,1475,"[BUG] Bold answers (\"ANSWER: **18**\") are parsed as the expression \"18**\", so the score falls back to another ","ChrystianSchutz",4,"","",0,"2026-11-05",""],[13,1471,"[BUG] Belebele cf tasks fail for Latvian, Lithuanian, Maltese and Slovenian: missing translation literals","HannanJaved",4,"","",0,"2026-11-02",""],[13,1469,"HLE `calibration_error` compares confidence in percent (0-100) with correctness in 0/1, so a model that is alw","damienen",4,"","",0,"2026-11-02",""],[13,1466,"[BUG] Stderr of `_average` and `all` rows is the mean of the subtask stderrs (about 7x too large for MMLU)","Andreas-Hum",4,"","",0,"2026-11-01",""],[13,1464,"[BUG] LCB grader fails correct solutions that write via sys.stdout.buffer","abhishekcpp007",4,"","",0,"2026-11-01",""],[13,1463,"[EVAL] Add support for IFEval-Ko","SubJeeLee",4,"","",0,"2026-11-01",""],[13,1459,"[BUG] LiveCodeBench grader runs generated code with no memory limit","williamcaban",4,"","",0,"2026-10-29",""],[13,1453,"[BUG] Reasoning tag removal fails when <think> is part of the input prompt","Tialo",4,"","",0,"2026-10-27",""],[13,1448,"[EVAL] HAE-RAE Bench: Korean-native knowledge and vocabulary","jhwanseok",4,"","",0,"2026-10-23",""],[13,1442,"[BUG] TruthfulQA MC2 drops true answers after a gap in gold indexes","salma2vec",4,"","",0,"2026-10-20",""],[13,1441,"[BUG] Token F1 ignores repeated words in predictions and references","salma2vec",4,"","",0,"2026-10-20",""],[13,1440,"Bayes@N metric for repeated generative evaluation (existing PR #1219)","mohsenhariri",4,"","",0,"2026-10-20",""],[13,1437,"[EVAL] CLIcK: Korean cultural and language understanding","ojm6135",4,"","",0,"2026-10-20",""],[13,1430,"[BUG] Inspect math scorer rejects equivalent numerical answers by comparing parser fallback strings","surgan12",4,"","",0,"2026-10-16",""],[13,1424,"[EVAL] KMMLU (Korean-native MMLU, 45 subjects): would a PR be welcome?","CY-HYUN",4,"","",0,"2026-10-10",""],[13,1414,"Importing swiss_legal metrics sets float32 matmul precision to \"medium\" process-wide, breaking long transforme","dzautner",4,"","",0,"2026-10-02",""],[13,1410,"PassAtK returns 1.0 for all-wrong samples when k > n","caiotheodoro",4,"","",0,"2026-09-30",""],[13,1406,"[BUG] 187 multilingual/QA tasks can't load on datasets>=4 (script-based datasets); ifeval-fr/gpqa-fr/bac-fr re","David-Wu1119",4,"","",0,"2026-09-28",""],[13,1400,"[BUG] LiveCodeBench scores every generation as -2 when multiprocessing uses spawn or forkserver","David-Wu1119",4,"","",0,"2026-09-27",""],[13,1397,"[FT] Add a format-neutral post-run export hook","halvrenofviryel",4,"","",0,"2026-09-23",""],[13,1395,"Metrics.drop aggregates em/f1 across documents with max, so one correct sample reports 1.0","shaurya416",4,"","",0,"2026-09-23",""],[13,1384,"avg_at_n silently ignores strip_strings/normalize (preprocess not applied), deflating scores","Abelo9996",4,"","",0,"2026-09-10",""],[13,1375,"Normalized multiple-choice probability underflows for long sequence log-probabilities","YusefSyed",4,"","",0,"2026-09-02",""],[13,1374,"[FT] Corpus-level Cohen's Kappa (unweighted / linear / quadratic) for ordinal and judge-graded tasks","oberoir080",2,"","",0,"2026-09-01","2026-09-04"],[13,1365,"SamplingMetric string `normalize` argument always raises: getmembers() list treated as a dict","Abelo9996",4,"","",0,"2026-08-29",""],[13,1364,"Prediction cache identity is the model config, not the model that ran: weights moving under the same revision ","AUTHENSOR",4,"","",0,"2026-08-29",""],[13,1363,"Sample cache: doc.id collides across evaluation splits, crashing runs and silently serving one split's answers","AUTHENSOR",4,"","",0,"2026-08-29",""],[13,1353,"F1_score collapses nltk's undefined result onto zero, including when the gold reference is empty","buzzedaldrinx86-creator",4,"","",0,"2026-08-23",""],[13,1346,"[BUG] Why isn't apply_chat_template called with add_generation_prompt=True in LLM as a judge?","Jeronymous",4,"","",0,"2026-08-18",""],[13,1330,"[BUG] Incompatibility with latest version of xxhash","Giuseppe5",4,"","",0,"2026-08-13",""],[13,1329,"[BUG] dyck_language scores with inspect_ai's exact(), which ignored word order in every release up to 0.3.252","LippInc",4,"","",0,"2026-08-13",""],[13,1327,"lighteval eval crashes on fresh install: GenerateConfig rejects frequence_penalty / log_probs / response_forma","sk8ordie84",4,"","",0,"2026-08-11",""],[13,1318,"[Proposal] OpenEval Import/Export Support","adhabnr-ux",4,"","",0,"2026-07-29",""],[13,1316,"Lighteval TriviaQA evaluation not giving results as per smolLM-360M reported numbers","Kun-AI-san",4,"","",0,"2026-07-27",""],[13,1309,"Balanced few-shot sampling truncates on a falsy class label and uses the global random module","ErenAta16",4,"","",0,"2026-07-20",""],[13,1307,"Sequential few-shot selection mutates the task's shared few-shot pool in place, corrupting later variance seed","ErenAta16",4,"","",0,"2026-07-20",""],[13,1306,"[EVAL] Add ArmBench-LLM Armenian evaluation suite","bugdaryan",4,"","",0,"2026-07-20",""],[13,1304,"avg_at_k / avg_at_k_math / maj_at_k metric regression fixtures silently no-op (metric_class name doesn't match","ErenAta16",4,"","",0,"2026-07-19",""],[13,1299,"[BUG] Task : `wikitext:103:document_level` does not work","debasisdwivedy",4,"","",0,"2026-07-15",""],[13,1296,"[BUG] LLM judge silently scores API error strings: litellm backend sends max_tokens as a list, provider return","AbhilakshSinghReen",4,"","",0,"2026-07-14",""],[13,1293,"hellaswag_arabic_pfn uses eval() instead of ast.literal_eval() on a dataset field, arbitrary code execution fr","ErenAta16",4,"","",0,"2026-07-11",""],[13,1278,"[BUG] transformers backend overrides the model's generation_config.eos_token_id, so chat models whose turn ter","ranawaqas-ai",4,"","",0,"2026-07-02",""],[13,1270,"Issue when evaluating Ministral-3-14B-Instruct-2512","kafetzopoulos",2,"","",0,"2026-06-22","2026-06-24"],[13,1266,"[FT] Introduce a static 'lint' command to validate custom tasks before execution","Kesav2k04",4,"","",0,"2026-06-18",""],[13,1265,"MMLU-Pro incorrect prompt format","RasmusHoier",4,"","",0,"2026-06-17",""],[13,1255,"Lighteval's LCB grader fails correct solutions that read sys.stdin.buffer","RasmusHoier",4,"","",0,"2026-06-12",""],[13,1240,"[FT] IRT-based scoring as alternative ranking method for fairer model comparison","testofschool",4,"","",0,"2026-05-20",""],[13,1216,"[BUG] LiteLLM endpoint backend doesn't clamp generation_size to model context, breaks small models on long-gen","orsharir",2,"","",0,"2026-04-27","2026-04-27"],[13,1204,"[BUG] [vLLM] Batch truncation bug: context length check uses first prompt instead of longest","paulovsantanas",4,"","",0,"2026-04-06",""],[13,1202,"Custom task: RAIL Score responsible AI evaluation (8-dimension scoring)","SumitVermakgp",4,"","",0,"2026-04-02",""],[13,1200,"[META] Is this repo still maintained? Community PRs going unreviewed for months","kryvokhyzha",4,"","",0,"2026-03-29",""],[13,1198,"[BUG]","w-ahmad1a10",4,"","",0,"2026-03-23",""],[13,1184,"squad_v2.py : unanswerable questions are filtered out, making the task equivalent to SQuAD 1.1","Matteovanypersele",4,"","",0,"2026-03-09",""],[13,1170,"[BUG] LogProbTokenNorm crashes with IndexError in CFFormulation for belebele","Rijgersberg",4,"","",0,"2026-02-18",""],[13,1168,"Add jfinqa: Japanese Financial Numerical Reasoning QA","ajtgjmdjp",4,"","",0,"2026-02-17",""],[13,1167,"[EVAL] Request support for AIME26","JackLingjie",4,"","",0,"2026-02-13",""],[13,1164,"[FT] add a long horizon \"tension crash test\" dimension based on a TXT S-class question pack","onestardao",4,"","",0,"2026-02-06",""],[13,1163,"[BUG] KeyError: 'default'","2niuhe",4,"","",0,"2026-01-31",""],[13,1162,"[BUG] IndexError: list index out of rang","2niuhe",4,"","",0,"2026-01-31",""],[13,1160,"Fix TypeError in aa_omniscience_prompt","pjavanrood",4,"","",0,"2026-01-22",""],[13,1158,"Fix split loading error in bigbench","pjavanrood",4,"","",0,"2026-01-22",""],[13,1156,"Fix CoQA metric and multi-doc loading","pjavanrood",2,"","",0,"2026-01-22","2026-01-22"],[13,1154,"Fix RecursionError in imdb_contrastset_prompt","pjavanrood",4,"","",0,"2026-01-22",""],[13,1152,"Fix legal_summarization keys and SummaC metric","pjavanrood",2,"","",0,"2026-01-22","2026-01-22"],[13,1150,"Fix non-existent evaluation splits in lextreme","pjavanrood",4,"","",0,"2026-01-22",""],[13,1148,"Fix evaluation split config in lsat_qa","pjavanrood",4,"","",0,"2026-01-22",""],[13,1146,"Improve NarrativeQA metrics and prompt structure","pjavanrood",4,"","",0,"2026-01-22",""],[13,1144,"Fix schema validation in olympiad_bench Doc.specific","pjavanrood",4,"","",0,"2026-01-22",""],[13,1142,"Fix key mismatch and context access in PubMedQA","pjavanrood",4,"","",0,"2026-01-22",""],[13,1140,"Fix TypeError in real_toxicity_prompts due to None choices","pjavanrood",4,"","",0,"2026-01-22",""],[13,1138,"Fix column mismatch and metric in SimpleQA","pjavanrood",4,"","",0,"2026-01-22",""],[13,1136,"Fix subset names in StoryCloze","pjavanrood",4,"","",0,"2026-01-22",""],[13,1134,"Fix Doc init and missing metadata in Summarization tasks","pjavanrood",4,"","",0,"2026-01-22",""],[13,1132,"Fix hardcoded path in tiny_benchmarks","pjavanrood",4,"","",0,"2026-01-22",""],[13,1130,"Fix KeyError in truthful_qa_generative_prompt","pjavanrood",4,"","",0,"2026-01-22",""],[13,1128,"Fix MT-Bench multi-turn evaluation logic","pjavanrood",2,"","",0,"2026-01-22","2026-01-23"],[13,1122,"[BUG] Missing PIL dependency","pjavanrood",2,"","",0,"2026-01-13","2026-01-15"],[13,1115,"[FT] Add response_format support to LiteLLM","pjavanrood",1,"pjavanrood","#1116",0,"2026-01-09","2026-01-13"],[13,1113,"[FT] Support programmatic API key and URL configuration for JudgeLLM and LiteLLM","pjavanrood",1,"pjavanrood","#1114",0,"2026-01-08","2026-01-08"],[13,1112,"[BUG] chrF++/chrF/TER metrics receive references in wrong format, causing incorrect corpus-level scoring","dyurchenko98",4,"","",0,"2026-01-06",""],[13,1110,"[FT] Support setting the multiple random seeds and decoupling it from model deployment?","Kwen-Chen",4,"","",0,"2025-12-17",""],[13,1109,"[FT] Support for retriever-augmented and latent-memory models.","akshathmangudi",4,"","",0,"2025-12-16",""],[13,1108,"[BUG] return _winapi.DuplicateHandle","chigkim",4,"","",0,"2025-12-16",""],[13,1106,"[BUG] Code execution stuck without any information","JianxXiong",2,"","",0,"2025-12-12","2025-12-13"],[13,1102,"[BUG] Accelerate backend with dp fails to load cache","Aceticia",0,"shipbehaves","#1271",0,"2025-12-10","2026-06-29"],[13,1100,"[BUG] concurrent_requests in litellm backend limited to 100","yt0428",4,"","",0,"2025-12-08",""],[13,1096,"[FT] I am unable to use microsoft/Phi-4-multimodal-instruct","ryuzakace",4,"","",0,"2025-12-03",""],[13,1094,"[BUG] Get low performance when evaluate Qwen3-8B on MMLU-Redux using the thinking mode.","StarLooo",4,"","",0,"2025-12-03",""],[13,1093,"[FT] Log-likelihood metrics support for API-based evaluation (LiteLLM / completions endpoint)","DenysYurchenko24",4,"","",0,"2025-12-02",""],[13,1090,"[BUG] Breaking change for Hellaswag/Piqa","Giuseppe5",4,"","",0,"2025-11-28",""],[13,1089,"[BUG] Cannot run math_500/mmlu_pro on models without chat template","Giuseppe5",4,"","",0,"2025-11-28",""],[13,1063,"Improving input token throughput?","carlesoctav",4,"","",0,"2025-11-17",""],[13,1056,"[EVAL] Long Horizon Execution","NathanHB",4,"","",0,"2025-11-13",""],[13,1055,"[BUG] git clone issues due to filenames with special characters","amosyou",2,"","",0,"2025-11-13","2026-06-08"],[13,1054,"MajAtN.compute calls compute_score with swapped arguments","WeiKangda",2,"","",0,"2025-11-12","2025-11-14"],[13,1053,"[FT] Prompt Level Caching","f14-bertolotti",4,"","",0,"2025-11-12",""],[13,1047,"[BUG] Got Zero em on MMLU using Deepseek distilled Qwen models","Zoeyyao27",4,"","",0,"2025-11-11",""],[13,1046,"Wrong attribute in AvgAtN","f14-bertolotti",2,"","",0,"2025-11-10","2025-11-12"],[13,1045,"Why custom task cannot success ?  ValueError(f\"Cannot find task {expanded_task} in task list or in custom task","stringofn",2,"","",0,"2025-11-10","2025-11-17"],[13,1040,"[BUG] [eval] --max-samples 10 and then without, you get the cached data from the previous run.","NathanHB",4,"","",0,"2025-11-06",""],[13,1036,"[EVAL] Add kyrgyzLLM benchmark","golden-ratio",4,"","",0,"2025-11-04",""],[13,1030,"[BUG] fail to execute task harness|bbh:movie_recommendation|0","colinzuo",2,"","",0,"2025-10-27","2025-11-05"],[13,1029,"[BUG] Python API doesn't seem to work with lighteval|gsm8k|<few-shot-number>","Saravana-Ace",4,"","",0,"2025-10-24",""],[13,1028,"How to evaluate MMLU-Pro","qhz991029",4,"","",0,"2025-10-24",""],[13,1026,"[BUG] Cache management failed for tasks using sample_params to update function attrs","colinzuo",4,"","",0,"2025-10-22",""],[13,1024,"[BUG] LightevalTaskConfig doesn't support comparison, but Cache management uses sorted on it","colinzuo",4,"","",0,"2025-10-21",""],[13,1023,"[BUG] Cache management failed for tasks using MetricGrouping","colinzuo",4,"","",0,"2025-10-21",""],[13,1020,"Vision Language Model Support?","ByungKwanLee",4,"","",0,"2025-10-15",""],[13,1019,"[EVAL] MultiChallenge","lewtun",2,"","",0,"2025-10-15","2026-01-14"],[13,1009,"[BUG] LiveCodeBench requires dataset scripts","alielfilali42",2,"","",0,"2025-10-06","2025-10-06"],[13,1007,"[BUG] GSMPlus fails with `'NoneType' object has no attribute 'id'`","lewtun",1,"lewtun","#1008",1,"2025-10-04","2025-10-06"],[13,1005,"[BUG] MixEval crashes with `KeyError: TaskID(task_name='extended|mixeval_hard:multichoice|0', task_hash='50295","lewtun",2,"","",0,"2025-10-03","2025-11-04"],[13,1004,"[BUG] truthfulqa requires extended_tasks dependencies because registry eagerly imports all extended tasks","mcavdar",2,"","",0,"2025-10-03","2026-07-23"],[13,1002,"[FT] Upgrade the VLLM dependency to 0.10.2+","JIElite",4,"","",0,"2025-10-02",""],[13,1001,"RULER benchmark not found","Roadsong",4,"","",0,"2025-09-30",""],[13,1000,"Evaluation of encoder and decoder models on SuperGLUE","fabiancpl",4,"","",0,"2025-09-30",""],[13,999,"How to print all pass@k scores when generating 16 samples?","passing2961",4,"","",0,"2025-09-29",""],[13,997,"[BUG] PMI normalization unusable: requires `unconditioned_logprob` but API/docs don’t expose a way to provide ","emircanerol",4,"","",0,"2025-09-29",""],[13,996,"[BUG] RuntimeError: No executable batch size found, reached zero.","tomtyiu",2,"","",0,"2025-09-28","2025-10-03"],[13,995,"[BUG] Handle BERTScore baseline files","amstu2",4,"","",0,"2025-09-27",""],[13,991,"[BUG] TInyBenchmark metrics need conversion from string to Metric","amstu2",0,"NathanHB","#992",1,"2025-09-24","2025-11-05"],[13,988,"[BUG] FileNotFoundError with newest lighteval 0.11.0","jxmorris12",4,"","",0,"2025-09-23",""],[13,982,"[EVAL] Add SLR-Bench","Ahmad21Omar",4,"","",0,"2025-09-22",""],[13,979,"[BUG] Cannot run LiveCodeBench with `lighteval==0.11.0` (`ValueError: Incorrect type for <function codegen_met","lewtun",0,"NathanHB","#981",1,"2025-09-22","2025-09-23"],[13,975,"[EVAL] Add MathVista","ThakurRajAnand",4,"","",0,"2025-09-17",""],[13,974,"Can't reproduce the Chinese FineTasks results.","bg51717",4,"","",0,"2025-09-17",""],[13,973,"[BUG] Cannot list tasks if the user doesn't install multilingual extras","elichen3051",2,"","",0,"2025-09-17","2025-12-04"],[13,970,"How to use a configuration file?","oluwandabira",2,"","",0,"2025-09-16","2025-09-24"],[13,968,"[BUG] Compatibility issue with datasets>=3.5.0: trust_remote_code parameter not accepted","alvaro-francisco-gil",2,"","",0,"2025-09-16","2025-09-16"],[13,966,"[FT] Additional LiteLLM model config options and better context length estimate","rolshoven",4,"","",0,"2025-09-16",""],[13,962,"[FT] Improvements for LLM judges with litellm backend","rolshoven",2,"","",0,"2025-09-15","2025-09-16"],[13,960,"[BUG] lighteval does not support openai endpoints but says it does","oluwandabira",4,"","",0,"2025-09-13",""],[13,959,"[BUG] Cuda OOM with gpt-oss-20b","vince62s",4,"","",0,"2025-09-12",""],[13,957,"[BUG] bug prevents mixing batched and non-batched metrics","rolshoven",4,"","",0,"2025-09-12",""],[13,955,"[FT] Multilingual Extractiveness Metrics","rolshoven",1,"rolshoven","#956",0,"2025-09-11","2025-09-16"],[13,950,"[BUG] Evaluation tracker won't save task which relies on using metrics of 2 different types","clefourrier",4,"","",0,"2025-09-09",""],[13,949,"[BUG] PreTrainedTokenizerFast._batch_encode_plus() got multiple values for keyword argument 'truncation_strate","yujonglee",4,"","",0,"2025-09-08",""],[13,948,"[FT] Option to retain model across evaluations to reuse the same vLLM instance","HaakonMongstad",2,"","",0,"2025-09-05","2025-09-17"],[13,946,"[BUG] Datasets module got an unexpected keyword argument  'trust_remote_code'","mertbozkir",2,"","",0,"2025-09-03","2025-09-03"],[13,943,"[FT] Support GGUF in vllm, and use HF tokenizer together","JIElite",2,"","",0,"2025-08-29","2025-10-03"],[13,940,"[FT] Support confidence reporting and sampling","clefourrier",4,"","",0,"2025-08-28",""],[13,933,"[BUG] VLLM Code Generate hangs","berserkr",4,"","",0,"2025-08-22",""],[13,927,"[FT] Save reasoning_content of responses as details","muupan",1,"muupan","#929",0,"2025-08-19","2025-08-22"],[13,918,"[FT] Pass reasoning effort to `litellm` endpoint.","gardarjuto",4,"","",0,"2025-08-12",""],[13,916,"[FT] Automatically remove thinking traces from outputs","clefourrier",1,"clefourrier","#925",1,"2025-08-11","2025-08-19"],[13,910,"[FT] Relax upper bound on `datasets` dependency","lewtun",2,"","",0,"2025-08-09","2025-08-21"],[13,908,"[EVAL] Add IFBench","lewtun",2,"","",0,"2025-08-08","2025-09-18"],[13,903,"[BUG] `tasks list` is not a complete list","alielfilali01",2,"","",0,"2025-08-06","2025-08-23"],[13,902,"spelled 'judgment' in 'bbh_causal_judgment_harness' but spelled \"judgement\" in the dataset repo (https://huggi","HaakonMongstad",2,"","",0,"2025-08-06","2025-08-06"],[13,897,"[BUG] TransformersModel.from_model provides an incompletely initialized object compared to __init__","huaanrui",4,"","",0,"2025-08-05",""],[13,896,"[BUG]  Automatic batch size detection causes distributed deadlock in model parallel mode","huaanrui",4,"","",0,"2025-08-05",""],[13,895,"[BUG] Model Parallelism is incorrectly disabled on single multi-GPU machines","huaanrui",4,"","",0,"2025-08-05",""],[13,894,"[BUG] inspect command does not work","uralik",2,"","",0,"2025-08-04","2025-08-18"],[13,893,"[BUG] TriviaQA numbers are off","jxmorris12",2,"","",0,"2025-08-04","2026-01-06"],[13,892,"[BUG] Logs are missing some fields, truncation likely buggy","clefourrier",4,"","",0,"2025-08-04",""],[13,890,"[BUG] If using a local model, save won't happen in the correct folder","clefourrier",1,"clefourrier","#891",1,"2025-08-04","2025-08-04"],[13,888,"[BUG] : Typo in _continuous_greedy_until causes error with stop_sequence field","huaanrui",4,"","",0,"2025-08-03",""],[13,881,"[INFO] Document or remove `logits` from the `ModelResponse` class","clefourrier",4,"","",0,"2025-07-25",""],[13,880,"[FT] Cache partial outputs to enable re-run from interruption","lewtun",2,"","",0,"2025-07-24","2025-08-20"],[13,879,"[FT] Store system prompt in results","lewtun",2,"","",0,"2025-07-24","2025-08-20"],[13,878,"[FT] Fail early when a task is not defined","lewtun",0,"clefourrier","222a31c",1,"2025-07-24","2025-08-22"],[13,877,"[EVAL] Add MMLU-Redux-2","lewtun",2,"","",0,"2025-07-24","2025-09-18"],[13,876,"[EVAL] Add Arena Hard","lewtun",4,"","",0,"2025-07-24",""],[13,875,"[EVAL] Add LiveBench","lewtun",4,"","",0,"2025-07-24",""],[13,874,"[EVAL] Add tau-bench","lewtun",4,"","",0,"2025-07-24",""],[13,873,"[EVAL] Add BFCL-v3","lewtun",4,"","",0,"2025-07-24",""],[13,872,"[FT] Add regression tests to improve library stability with vLLM backend","lewtun",4,"","",0,"2025-07-24",""],[13,871,"[EVAL] Add Multi-IF","lewtun",4,"","",0,"2025-07-24",""],[13,870,"[FT] Allow k in pass@k and maj@k to be customisable at the command line","lewtun",2,"","",0,"2025-07-24","2025-08-25"],[13,869,"[FT] Remove thinking from all evals","lewtun",0,"clefourrier","#882",1,"2025-07-24","2025-08-04"],[13,868,"How to calculate perplexity from an OpenAI compatible API","mrtpk",4,"","",0,"2025-07-24",""],[13,867,"[BUG] use_chat_template=True causes model responses to include multiple Q&A pairs instead of terminating at th","huaanrui",4,"","",0,"2025-07-23",""],[13,866,"[BUG] output-dir parameter is ignored when model_name is an absolute path","huaanrui",2,"","",0,"2025-07-23","2025-08-04"],[13,865,"how can i get an evaluation result with 8-samples in Lighteval?","ForgetH3r",2,"","",0,"2025-07-18","2025-07-22"],[13,864,"[BUG] vLLM and Transformers scores are different for the same model and same task","basma-boussahaa",4,"","",0,"2025-07-17",""],[13,861,"[BUG] Python API Documentation Error","ZichengXu",2,"","",0,"2025-07-16","2025-08-26"],[13,858,"[BUG] Fail to eval on MT-Bench","StarLooo",4,"","",0,"2025-07-10",""],[13,857,"[BUG] `override_batch_size` disappeared from `PipelineParameters`, but there is no deprecation notice & it is ","fxmarty-amd",4,"","",0,"2025-07-10",""],[13,856,"[BUG] The latest datasets==4.0.0 not compatible.","QingtaoLi1",2,"","",0,"2025-07-10","2025-08-04"],[13,855,"Add a way to easily enable/disable thinking mode for reasoning models","MaxLSB",4,"","",0,"2025-07-09",""],[13,854,"[BUG] Error in lighteval/metrics/metrics_sample.py:513: _rouge_score_with_bootsrap, invalid argument","cafeii",4,"","",0,"2025-07-09",""],[13,853,"[BUG] Cannot import name taskinfo_selector from lighteval.tasks.registry","ljvmiranda921",2,"","",0,"2025-07-08","2025-08-13"],[13,848,"[FT] Why not use math_verify as a direct dependency?","jiangwangyi",2,"","",0,"2025-06-30","2025-07-25"],[13,844,"[BUG] System Prompt option now missing?","teknium1",4,"","",0,"2025-06-27",""],[13,834,"[FT] Problems with CorpusLevelMetricGrouping","LeonMalteW",4,"","",0,"2025-06-25",""],[13,827,"[FT] Manage fallback for incomplete configs","clefourrier",1,"clefourrier","#828",1,"2025-06-24","2025-10-14"],[13,826,"Adapt RULER eval to use chat template","clefourrier",4,"","",0,"2025-06-23",""],[13,825,"[FT]  Tokenizing the data right before launching the evaluation job and saving it to the cache","clefourrier",4,"","",0,"2025-06-23",""],[13,824,"[FEAT] Managing multiturn evaluations","clefourrier",4,"","",0,"2025-06-23",""],[13,823,"Remove extended tasks system and move everything to a \"custom task\" folder - add the extended task system as c","clefourrier",4,"","",0,"2025-06-23",""],[13,822,"Documenting how to launch multilingual tasks","clefourrier",4,"","",0,"2025-06-23",""],[13,821,"Simplify suite system in tasks","clefourrier",4,"","",0,"2025-06-23",""],[13,819,"Max token issue on RULER","clefourrier",4,"","",0,"2025-06-20",""],[13,818,"Mamba OOM deep dive","clefourrier",2,"","",0,"2025-06-20","2025-06-23"],[13,817,"[EVAL] Needle in a haystack","clefourrier",2,"","",0,"2025-06-20","2025-06-24"],[13,816,"[EVAL] LVEval","clefourrier",4,"","",0,"2025-06-20",""],[13,815,"[EVAL] MRCR","clefourrier",4,"","",0,"2025-06-20",""],[13,813,"[BUG] Failed downloads / 429 rate limiting for concurrent evaluations sharing cache","lukemerrick",4,"","",0,"2025-06-16",""],[13,809,"[Question] generation max_new_tokens or generation_size? which parameter takes effect?","Cppowboy",2,"","",0,"2025-06-11","2025-06-17"],[13,808,"[FT] VL model support for vllm/sglang backend?","xgwang",4,"","",0,"2025-06-11",""],[13,806,"[FT] Allow configuration of `litellm` cache location","jamesbraza",4,"","",0,"2025-06-10",""],[13,805,"[BUG] `Registry.task_registry` not reimporting the custom task module","jamesbraza",4,"","",0,"2025-06-10",""],[13,804,"[FT] showing count in Markdown summary table","jamesbraza",4,"","",0,"2025-06-09",""],[13,803,"[FT] `MetricUseCase.FORMATTING` for reasoning models","jamesbraza",4,"","",0,"2025-06-09",""],[13,802,"[BUG] Unable to Download Dataset in Restricted Network Environment (No Access to Hugging Face or Mirrors)","xjli360",4,"","",0,"2025-06-09",""],[13,801,"[BUG] `test` split forced to hit `Careful` warning","jamesbraza",4,"","",0,"2025-06-07",""],[13,800,"[BUG] Forced to hit `You cannot select the number of dataset splits` with `litellm`","jamesbraza",2,"","",0,"2025-06-07","2025-08-04"],[13,799,"[FT] integrate typo checker to check typos like `refenrence`","jamesbraza",4,"","",0,"2025-06-07",""],[13,798,"[FT] supporting train-time vs test-time metrics","jamesbraza",4,"","",0,"2025-06-07",""],[13,797,"[FT] more docstrings and typing in `Doc`","jamesbraza",4,"","",0,"2025-06-07",""],[13,796,"[FT] `StrEnum` for `suites` to intuitively document options","jamesbraza",4,"","",0,"2025-06-06",""],[13,794,"[BUG] Faithfulness metric fails because the SummaCZS model is instantiated but never assigned","sahilds1",2,"","",0,"2025-06-05","2025-06-17"],[13,793,"[BUG] support direct evaluation of local API and local data sets?","HemiFate",4,"","",0,"2025-06-05",""],[13,790,"[BUG] IFEval metrics incorrect for reasoning models","lewtun",2,"","",0,"2025-06-04","2025-06-04"],[13,789,"[BUG] 1","wubingheng111",2,"","",0,"2025-06-04","2025-06-04"],[13,788,"[FT] Store system prompt in results","lewtun",2,"","",0,"2025-06-03","2025-06-25"],[14,2625,"ByteFallback decoder replaces valid characters with U+FFFD when an invalid byte follows in the same run","damienen",4,"","",0,"2026-11-02",""],[14,2602,"Would you take an OSS-Fuzz integration for tokenizers?","JayOfTheKeyboard",4,"","",0,"2026-10-27",""],[14,2590,"[v1] LongestFirst pair truncation changes retained token IDs compared with 0.23.2","michaelfeil",4,"","",0,"2026-10-26",""],[14,2589,"[v1] Singleton Sequence[Metaspace] fails to load while equivalent direct Metaspace works","michaelfeil",4,"","",0,"2026-10-26",""],[14,2561,"[1.0.0-rc.3] Kimi-K2 pretokenizer is not backwards-comaptible on Han characters","SBrandeis",4,"","",0,"2026-10-21",""],[14,2559,"[1.0.0-rc.3] Forked child processes hang when loading a tokenizer after the parent process loaded one","SBrandeis",1,"SBrandeis","#2581",1,"2026-10-21","2026-10-30"],[14,2557,"[1.0.0-rc.3] DeepSeek encode returns different token IDs from 0.23.2","zhu-he",4,"","",0,"2026-10-21",""],[14,2555,"[v1] Canonical `tokenizer.json` 2.0 is not readable by 0.x, and a version-patched file silently yields wrong i","zRzRzRzRzRzRzR",2,"","",0,"2026-10-20","2026-10-20"],[14,2528,"[v1] meta-issue: required work for integration with `transformers`","SBrandeis",4,"","",0,"2026-10-14",""],[14,2527,"[v1] Rust: train through a tokenizer's own pipeline","SBrandeis",4,"","",0,"2026-10-14",""],[14,2526,"[v1] Python: expose the trainers and `Tokenizer.train_from_iterator`","SBrandeis",4,"","",0,"2026-10-14",""],[14,2525,"[v1] Pre-tokenized input","SBrandeis",4,"","",0,"2026-10-14",""],[14,2524,"[v1] Decode a list of token strings","SBrandeis",4,"","",0,"2026-10-14",""],[14,2523,"[v1] Overflowing tokens with stride","SBrandeis",4,"","",0,"2026-10-14",""],[14,2522,"[v1] `special_tokens_mask`","SBrandeis",4,"","",0,"2026-10-14",""],[14,2521,"[v1] Offsets and word alignment","SBrandeis",4,"","",0,"2026-10-14",""],[14,2520,"[v1] `DecodeStream`","SBrandeis",4,"","",0,"2026-10-14",""],[14,2519,"[v1] `Encoding.tokens`, with the original text for unknown pieces","SBrandeis",4,"","",0,"2026-10-14",""],[14,2518,"[v1] Added tokens API","SBrandeis",4,"","",0,"2026-10-14",""],[14,2517,"[v1] Vocabulary lookups","SBrandeis",1,"SBrandeis","#2614",1,"2026-10-14","2026-11-02"],[14,2516,"[v1] Serialization: `to_str()`, `save(path)`, `from_str(json)`","SBrandeis",1,"SBrandeis","#2563",1,"2026-10-14","2026-11-02"],[14,2515,"[v1] Composition API: build a tokenizer from components, and get/set them","SBrandeis",2,"","",0,"2026-10-14","2026-11-06"],[14,2514,"[v1] Added-token ids differ from 0.23 when `added_tokens` are listed out of id order","SBrandeis",4,"","",0,"2026-10-14",""],[14,2513,"[v1] A vocabulary containing an empty-string token fails to load","SBrandeis",1,"SBrandeis","#2538",1,"2026-10-14","2026-10-19"],[14,2512,"[v1] Normalized added token at the start of the input gets a stray `▁` with `Metaspace`","SBrandeis",4,"","",0,"2026-10-14",""],[14,2511,"[v1] Python bindings are built without a regex backend: regex `Split` / `Replace` fail","SBrandeis",1,"SBrandeis","3fbc49b",1,"2026-10-14","2026-10-19"],[14,2499,"Concurrent encode calls construct multiple shared worker pools","ZenAlexa",4,"","",0,"2026-10-08",""],[14,2498,"Don't emit control tokens in untrusted text segments","SBrandeis",4,"","",0,"2026-10-08",""],[14,2489,"huggingface-hub-1.33.0 ->  huggingface_hub-2.1.1?","jrp2014",2,"","",0,"2026-10-04","2026-10-09"],[14,2488,"Build abi3t wheels for Python 3.15 support","ngoldbaum",4,"","",0,"2026-10-03",""],[14,2460,"v1.0.0-rc.2: canonicalize_file rejects ByteLevel add_prefix_space=true supported by 0.23.2","michaelfeil",4,"","",0,"2026-09-26",""],[14,2458,"[1.0.0rc2] Encoding.ids_array panics while another thread decodes the ids","nimbrel",0,"SBrandeis","#2464",1,"2026-09-26","2026-10-07"],[14,2456,"Type Stubs for subpackages don't work with basedpyright","LyndonArmitage",2,"","",0,"2026-09-24","2026-10-09"],[14,2447,"[1.0.0rc2] Possible regressions vs. `0.23.x`: Byte atom not found in the vocabulary","xenova",0,"SBrandeis","#2550",1,"2026-09-23","2026-11-06"],[14,2412,"BertWordPieceTokenizer ignores wordpieces_prefix when loading a vocabulary","HaokaiDing",2,"","",0,"2026-09-16","2026-09-20"],[14,2405,"CSOAI — Tokenizers interop: NLP governance","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[14,2404,"CSOAI — Tokenizers interop: NLP governance","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[14,2403,"CSOAI — Tokenizers AI governance NLP","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[14,2334,"`Precompiled` normalizer diverges from SentencePiece: shortest-prefix match + discarded remainder silently dro","imhmhm",3,"","",0,"2026-08-14","2026-10-07"],[14,2320,"BpeTrainer emits a merge for a pair that occurs zero times (negative pair_counts sign-extended by `as u64`)","Abraham-y",3,"","",0,"2026-08-07","2026-09-29"],[14,2236,"[BUG] | The docs listing page for tokenizers is throwing 404","JayaKrishnaNamburu",2,"","",0,"2026-07-23","2026-09-20"],[14,2225,"[Feature/Security] Support trusted and untrusted special-token spans","kexinoh",4,"","",0,"2026-07-21",""],[14,2210,"EncodingVisualizer: data-* attributes built from token text are not HTML-escaped (XSS gap left by #1937)","ErenAta16",2,"","",0,"2026-07-17","2026-09-20"],[14,2198,"BPE `build()` panics on load for a crafted `tokenizer.json` (merge concatenation exceeds the scratch buffer)","professor-moody",2,"","",0,"2026-07-14","2026-09-20"],[14,2116,"Precompiled normalizer uses grapheme-level shortest-prefix matching and diverges from SentencePiece on NFD inp","Incheonkirin",3,"","",0,"2026-06-17","2026-09-27"],[14,2109,"Proposal: emit UserWarning from Split.pre_tokenize_str() when it produces zero matches (catches silent str vs ","vamsin07",3,"","",0,"2026-06-16","2026-09-28"],[14,2094,"[Security] Load-time process abort when loading a malicious tokenizer.json (BPE merge buffer overrun)","geo-chen",3,"","",0,"2026-06-10","2026-10-31"],[14,2091,"Release 0.23.1 introduces breaking change in Encoding.overflowing","alistairewj",2,"","",0,"2026-06-08","2026-10-09"],[14,2088,"Two incorrect type annotations affect input typing to tokenizers.Tokenizer.encode()","JEHoctor",3,"","",0,"2026-06-07","2026-09-30"],[14,2075,"Reproducible WordPiece encode bench shows ~5x speedup on 123B fixture — methodology question for maintainers","SMC17",3,"","",0,"2026-05-28","2026-10-08"],[14,2069,"Unigram trainer: prune loss uses `alternatives.len()` instead of `alternatives[id].len()`","hunter-heidenreich",4,"","",0,"2026-05-24",""],[14,2058,"BPE trainer: i32 overflow in pair_counts corrupts merge ordering on large code corpora","petyb",3,"","",0,"2026-05-14","2026-10-09"],[14,2045,"Publish error in npm repository, Node.js binary bindings","uasan",2,"","",0,"2026-04-28","2026-09-03"],[14,2042,"Build failure with maturin@1.13.1 on macOS","manuelakuhn",2,"","",0,"2026-04-27","2026-04-27"],[14,2035,"Massive performance degradation and memory spike when using text_pair with long inputs","Leonater",2,"","",0,"2026-04-25","2026-05-02"],[14,2027,"It is important to understand the real cause of the problem.","Nearbe",3,"","",0,"2026-04-18","2026-10-10"],[14,2023,"Feature request: compact vocabulary remapping for domain-specific training with pretrained tokenizers","saslifat-gif",3,"","",0,"2026-04-14","2026-10-10"],[14,1996,"[Bug] Fast Tokenizer (ByteLevel BPE) incorrectly decodes added_tokens containing specific Unicode characters (","zidsi",0,"ErenAta16","#1",1,"2026-03-27","2026-05-13"],[14,1975,"[RFC] Korean Tokenization: Jamo Decomposition as Pre-tokenizer (A Blueprint for Compositional Scripts)","nicezic",3,"","",0,"2026-03-22","2026-10-11"],[14,1973,"Make huggingface-hub an optional dependency (extras)","clemlesne",4,"","",0,"2026-03-20",""],[14,1962,"BertTokenizer or AutoTokenizer suddenly results in JSONDecodeError","shivanraptor",2,"","",0,"2026-03-12","2026-03-12"],[14,1961,"Add riscv64 (linux_riscv64) wheel to PyPI releases","gounthar",2,"","",0,"2026-03-11","2026-03-26"],[14,1960,"Serialization of Bytelevel for decoder and pre_tokenizer is missleading","ArthurZucker",3,"","",0,"2026-03-04","2026-10-16"],[14,1957,"Tokenizer encode ignores pre-tokenizer when using right to left (rtl) digit splits with add_tokens","gsaltintas",4,"","",0,"2026-03-01",""],[14,1954,"Constructive criticism for Tokenizers library.","BBC-Esq",2,"","",0,"2026-02-23","2026-09-21"],[14,1947,"Enable/Expose Unigram LRU cache for DebertaV2Tokenizer?","jbertoldo",2,"","",0,"2026-02-06","2026-02-06"],[14,1933,"all features are not included in docs.rs","Redhawk18",0,"wheynelau","#1953",0,"2026-01-20","2026-02-24"],[14,1929,"`encode_batch` has suboptimal parallelization on high-core systems","stargazerZJ",3,"","",0,"2026-01-14","2026-10-17"],[14,1927,"[FR] Add type hint for python","zhaob1n",2,"","",0,"2026-01-11","2026-02-16"],[14,1926,"[bug] Why is Apple's development for computers with Inter chips not supported in versions above 0.30.0","sustly",2,"","",0,"2026-01-06","2026-03-26"],[14,1922,"[Node.js] Cannot find module 'tokenizers-darwin-arm64' after npm install on macOS (Apple Silicon)","uasan",0,"MayCXC","47c2fbc",0,"2025-12-27","2026-04-02"],[14,1920,"[Docs] [Bug] Latest docs on huggingface.co and latest github release don't match","tomschelsen",2,"","",0,"2025-12-26","2026-03-26"],[14,1919,"De/tokenization on CUDA","vadimkantorov",4,"","",0,"2025-12-26",""],[14,1918,"do we have plan support java ，we need you","mullerhai",4,"","",0,"2025-12-25",""],[14,1917,"`spm_precompiled` dependency bricks rust-analyzer in certain build setups","gvilums",3,"","",0,"2025-12-21","2026-10-17"],[14,1915,"Decoding of AddedTokens containing Latin-Extended A characters are wrong","andrewivan123",3,"","",0,"2025-12-19","2026-10-17"],[14,1914,"Windows Link Error: esaxx-rs dependency forces Static CRT (/MT), breaking integration with onnx","altunenes",2,"","",0,"2025-12-16","2025-12-16"],[14,1913,"Wrong and unsuppressable print when instantiating BPE","bauwenst",0,"ArthurZucker","#1924",1,"2025-12-15","2026-01-05"],[14,1910,"[Docs] `Visualizer` dead links","dudeperf3ct",0,"Shivam-Bhardwaj","#1934",0,"2025-12-13","2026-03-25"],[14,1909,"[Docs] `Encode Inputs` rendering issues","ariG23498",3,"","",0,"2025-12-12","2026-10-17"],[14,1908,"`TemplateProcessing` does not apply `type_id` to overflow encodings","antoine-agthe-unity",3,"","",0,"2025-12-09","2026-10-20"],[14,1906,"Windows arm64 wheel release","finnagin",1,"finnagin","#1907",0,"2025-12-08","2026-02-11"],[14,1904,"python release workflow failing","finnagin",1,"finnagin","#1905",0,"2025-12-08","2026-01-05"],[14,1902,"Guide: Compiling `tokenizers` on Android/Termux","Manamama-Gemini-Cloud-AI-01",3,"","",0,"2025-12-01","2026-10-18"],[14,1900,"Performance: batch_encode scales poorly on high-core Server CPUs compared to sharded tokenizer instances","stargazerZJ",3,"","",0,"2025-11-30","2026-10-18"],[14,1897,"Support splitting to chunks during encode","xrl1",3,"","",0,"2025-11-28","2026-10-18"],[14,1890,"Upstream `RefMutGuard` / `RefMutContainer` to PyO3","davidhewitt",3,"","",0,"2025-11-27","2026-10-19"],[14,1887,"Failed to build installable wheels","BrandonStudio",4,"","",0,"2025-11-20",""],[14,1886,"Discussion: Alternatives to owned `String` in `Token` to reduce memory allocations","ChieloNewctle",3,"","",0,"2025-11-19","2025-12-03"],[14,1885,"Feature request: Characters delimiter argument","VasLem",3,"","",0,"2025-11-13","2026-10-20"],[14,1882,"`BaseTokenizer.normalize` method calls non-existent `_tokenizer.normalize`","T90REAL",0,"ishitab02","#1884",0,"2025-10-31","2025-11-28"],[14,1881,"Provide byte-level offsets for effective alignment in Cross-Tokenizer On-Policy Distillation","JqzChandler",3,"","",0,"2025-10-30","2026-10-22"],[14,1879,"rust tokenizer","gogo2464",3,"","",0,"2025-10-24","2026-10-23"],[14,1877,"encode bytes directly","tsengalb99",3,"","",0,"2025-10-19","2026-10-24"],[14,1876,"Add an API for tokens to ids","winstxnhdw",3,"","",0,"2025-10-15","2026-10-31"],[14,1873,"Why is my Python implementation faster than the Rust implementation?","sambaPython24",2,"","",0,"2025-10-05","2025-10-08"],[14,1871,"Add Middle-Truncation Option to Tokenizer","mxngjxa",3,"","",0,"2025-09-30","2026-10-31"],[14,1870,"How can I convert a trained tokenizer into `transformers` format","dibbla",2,"","",0,"2025-09-30","2025-09-30"],[14,1868,"How to set the cache_dir in the Rust implementation?","sambaPython24",3,"","",0,"2025-09-24","2026-10-25"],[14,1863,"WordPiece can't always avoid <unk> even with ByteLevel pretokenization.","craigschmidt",3,"","",0,"2025-09-11","2026-10-25"],[14,1855,"Breaking change in 0.21.2","phil-opp",3,"","",0,"2025-08-27","2026-10-26"],[14,1854,"Question: bert normalizer CJK handling","xiaoshihou514",3,"","",0,"2025-08-26","2026-10-26"],[14,1851,"SentencePieceBPE + Unicode NFD preprocessing leads to noise ?","PonteIneptique",3,"","",0,"2025-08-24","2026-10-26"],[14,1850,"Safe encoding of strings that might contain special token text","joschu",2,"","",0,"2025-08-21","2025-09-01"],[14,1847,"memory allocation of 21474836480 bytes failed","lostmsu",3,"","",0,"2025-08-16","2026-10-31"],[14,1845,"Trouble installing tokenizers (python) on AWS lambda arm64 image.","crawftv",3,"","",0,"2025-08-13","2026-10-31"],[14,1839,"`encode_special_tokens` is not saved for loading","janEbert",3,"","",0,"2025-07-31","2026-10-31"],[14,1837,"Poetry don't find tokenizers 0.21.4","lupa-fabio-coelho",2,"","",0,"2025-07-29","2025-07-30"],[14,1835,"Can you provide binary releases?","goldenmomonga",2,"","",0,"2025-07-26","2025-09-08"],[14,1833,"RUSTSEC-2024-0436 (paste is no longer maintained)","nystromjd",1,"nystromjd","#1834",0,"2025-07-23","2025-08-29"],[14,1829,"The parameter in initial_alphabet of the \"class BpeTrainer(Trainer)\" does not allow more than one character to","vmenan",3,"","",0,"2025-07-21","2026-11-03"],[14,1828,"Extra [SEP] added for ModernBERT decoder model","janimo",3,"","",0,"2025-07-19","2026-11-03"],[14,1825,"Proposal: Replace regex in `whitespace.rs` with manual code for speed improvements","8ria",3,"","",0,"2025-07-11","2026-11-02"],[14,1824,"Practical limits of BPE tokenizer training","dvruette",2,"","",0,"2025-07-08","2025-07-08"],[14,1823,"how can i only initiate one instance of tokenizer but use it in multi-process?","wangguanggg",3,"","",0,"2025-07-07","2026-11-02"],[14,1821,"Proposal: Faster `Whitespace` PreTokenizer Implementation (10–30% Speedup)","8ria",3,"","",0,"2025-07-07","2026-11-02"],[14,1820,"Cannot download test data: 'make test' and direct links fail with \"Repository not found\" / 404","8ria",3,"","",0,"2025-07-07","2026-11-02"],[14,1819,"Bug with `add_prefix_space` parameter for ByteLevel post-processor","megatron6000",0,"CloseChoice","#1878",0,"2025-07-04","2025-11-28"],[14,1817,"Inconsistent encoding of unknown tokens when they have special tokens as prefixes in the WordLevel Tokenizer","diego-andres-ardila",3,"","",0,"2025-07-03","2026-11-03"],[14,1816,"tokenizers cannot be compiled successfully on riscv machine","zengqingfu1442",2,"","",0,"2025-07-02","2025-07-07"],[14,1815,"Needs minor version release","sftse",2,"","",0,"2025-07-01","2025-07-29"],[14,1813,"Is it possible to return the count of truncated tokens?","leejuyuu",3,"","",0,"2025-06-29","2026-11-03"],[14,1812,"Is it possible for this library to provider WebAssembly support?","phosae",2,"","",0,"2025-06-27","2025-06-28"],[14,1811,"Avoiding '�' in decoded output when decoding token by token (currently testing with GPT-2)?","jchwenger",2,"","",0,"2025-06-26","2025-06-27"],[14,1810,"Building Rust numpy v0.23 dependency fails on Mac OS Sequoia","lordfeck",2,"","",0,"2025-06-24","2025-06-25"],[14,1807,"error: casting &T to &mut T is undefined behavior #1485","Fshrink",3,"","",0,"2025-06-22","2026-11-04"],[14,1797,"py03 async bindings for encode/decode in rust","michaelfeil",2,"","",0,"2025-06-11","2025-11-30"],[14,1795,"Bugs encountered during training","gitterlover",3,"","",0,"2025-06-11","2026-11-06"],[14,1794,"`WordPieceTrainer.train_from_iterator` is not deterministic","Tialo",3,"","",0,"2025-06-07","2026-11-06"],[14,1793,"`end_of_word_suffix` is ignored","Tialo",3,"","",0,"2025-06-07","2026-11-06"],[14,1791,"SyntaxWarning: invalid escape sequence '\\w'","wyattscarpenter",2,"","",0,"2025-06-04","2026-10-04"],[14,1790,"Regression in THUDM/chatglm3-6b  at least starting from **tokenizers-0.21.1**","pavel-esir",4,"","",0,"2025-06-04",""],[14,1787,"Hi, we need java /scala tokenizers bindings","mullerhai",4,"","",0,"2025-06-01",""],[15,2772,"Add THOX edge devices (ThoxAir / ThoxMini / ThoxClip) to hardware list","ttracx",4,"","",0,"2026-11-02",""],[15,2757,"jinja: startswith and endswith ignore start/end arguments","jihtsan",4,"","",0,"2026-10-30",""],[15,2755,"jinja: string literals reject escapes Python Jinja accepts (\\d, \\U, octal)","michaelkillgta",4,"","",0,"2026-10-29",""],[15,2752,"Proposal: add LlamaNest to \"Use this model\" local apps (GGUF)","fipra",4,"","",0,"2026-10-29",""],[15,2742,"[hub] raw and resolved LFS downloads collide in downloadFileToCacheDir","samirhvbr",4,"","",0,"2026-10-28",""],[15,2739,"[xet] `uploadFiles` allocates ~128MB of zero-filled buffers per call for tiny files","Kakulukian",0,"coyotte508","#2740",1,"2026-10-28","2026-10-28"],[15,2710,"[jinja] `~` throws when one side is undefined (breaks Llama 4 chat template with tools)","quanghuynt14",4,"","",0,"2026-10-24",""],[15,2681,"[Feature Request] Add AMD Radeon Pro V340L to the user hardware / platform list","cdtush",2,"","",0,"2026-10-18","2026-10-26"],[15,2678,"Hardware to Add to Hardware Selection","benni-ben",2,"","",0,"2026-10-16","2026-10-26"],[15,2641,"Proposal: Add Badgr to Hub model-page “Use this model” actions","michaelmanly",4,"","",0,"2026-10-10",""],[15,2607,"Add pocketpal ai to list of local apps","hextherich",4,"","",0,"2026-10-05",""],[15,2597,"Jinja strip methods ignore the chars argument","morgan-coded",4,"","",0,"2026-10-02",""],[15,2586,"GGUF quant labels do not recognize `ik_llama.cpp` quant types","joelfarthing",4,"","",0,"2026-10-01",""],[15,2582,"Hardware list: add common GPUs and CPU families missing per Hub user feedback","gary149",4,"","",0,"2026-09-30",""],[15,2554,"[closed] Space created via /api/repos/create (hub) becomes a phantom repo: invisible to the space API and unde","mas-ghub",2,"","",0,"2026-09-26","2026-09-26"],[15,2553,"Bug report filed in error — belongs on huggingface/hub","mas-ghub",2,"","",0,"2026-09-26","2026-09-26"],[15,2552,"Space created via /api/repos/create (hub) becomes a phantom repo: invisible to the space API and undeletable","mas-ghub",2,"","",0,"2026-09-26","2026-09-26"],[15,2551,"Space created via /api/repos/create becomes a phantom repo: invisible to the space API and undeletable","mas-ghub",2,"","",0,"2026-09-26","2026-09-26"],[15,2547,"Add hardware to list","Antiskim",0,"BlivionIaG","#2593",0,"2026-09-26","2026-10-01"],[15,2545,"listFiles, listCommits and countCommits fail on revisions with a slash","morgan-coded",1,"morgan-coded","#2546",0,"2026-09-25","2026-09-28"],[15,2537,"[lerobot] Wrong camera sizes for datasets with channel-first shape","cfahlgren1",1,"cfahlgren1","#2539",1,"2026-09-24","2026-09-25"],[15,2536,"Add MTPLX to the list of local apps","Lancelotbronner",4,"","",0,"2026-09-24",""],[15,2521,"checkCredentials rejects any non-hf_ token even when hubUrl points at a self-hosted hub","chappie1998",4,"","",0,"2026-09-22",""],[15,2501,"BrainrotGPT doesn’t work","woowsoimnotallowedtoexpose-cyber",4,"","",0,"2026-09-17",""],[15,2489,"Recognize proposed U3 Safetensors metadata","7dsolv",4,"","",0,"2026-09-13",""],[15,2477,"Add AMD Radeon Pro 5300M to hardware list","moawiah",4,"","",0,"2026-09-11",""],[15,2474,"hey","muathsalotaibi-ui",3,"","",0,"2026-09-11","2026-09-11"],[15,2472,"Add MLXHub to the list of local apps","JuanColilla",4,"","",0,"2026-09-10",""],[15,2455,"Add oMLX to the list of local app","bplein",4,"","",0,"2026-09-07",""],[15,2453,"Error in running deepseek-ai/DeepSeek-V4-Flash","malihae",4,"","",0,"2026-09-07",""],[15,2447,"Add GTX 960M to NVIDIA hardware list (My Hardware dropdown)","moawiah",4,"","",0,"2026-09-05",""],[15,2443,"Diffusers text-to-audio models generate an incorrect image Colab snippet","taylorty",4,"","",0,"2026-09-03",""],[15,2438,"35","mdh601889-bot",3,"","",0,"2026-09-01","2026-09-01"],[15,2434,"Follow-up to #2416: zeromodels library resolves, but downloads still show \"not tracked\"","IMvision12",2,"","",0,"2026-08-31","2026-09-02"],[15,2421,"@huggingface/xetchunk-wasm can emit chunks below the minimum and diverge from xet-core","assafvayner",1,"assafvayner","#2422",1,"2026-08-27","2026-08-27"],[15,2418,"Intel Core Ultra 7 266V","sadroddin-aghaei",4,"","",0,"2026-08-27",""],[15,2408,"Follow-ups to range-edit uploads (#2407): tail-byte caching, incremental sha256, /v2/shards, append helper","coyotte508",4,"","",0,"2026-08-24",""],[15,2400,"Hub GGUF indexer picks mmproj as the repo's model when filenames contain \"MTP\" (reports clip / 0.46B)","mabry1985",4,"","",0,"2026-08-22",""],[15,2396,"opencode","Legendary-Brilliantforous",4,"","",0,"2026-08-21",""],[15,2376,"provider: \"auto\" ignores custom fetch for the provider-mapping pre-flight request","kurian-j",1,"kurian-j","#2377",0,"2026-08-14","2026-08-24"],[15,2370,"New I'd models for you most parojzet","a6771511-oss",2,"","",0,"2026-08-14","2026-08-15"],[15,2366,"WebBlob.create returns a size-0 blob in the browser (packages/blob still probes with HEAD)","codeAnqiang-ma",4,"","",0,"2026-08-13",""],[15,2364,"FileBlob/WebBlob/XetBlob.slice() negative-argument guard never throws (missing `throw` before `new TypeError`)","codeAnqiang-ma",1,"codeAnqiang-ma","#2365",0,"2026-08-13","2026-08-13"],[15,2355,"I was told to report this on this github page","CreateR-coder",3,"","",0,"2026-08-07","2026-08-25"],[15,2334,"McpClient drops model reasoning between tool-calling rounds","gary149",4,"","",0,"2026-07-31",""],[15,2333,"X","mickmix258-gif",2,"","",0,"2026-07-30","2026-07-30"],[15,2327,"Colab demo for Kimi K3 on hugginface does not work","xaviergonzalez",4,"","",0,"2026-07-28",""],[15,2313,"https://github.com/huggingface/huggingface.js.git","mickmix258-gif",2,"","",0,"2026-07-26","2026-07-26"],[15,2311,"A late abort after successful hashing terminates a worker already returned to the pool and emits an unhandled ","MicroMilo",0,"DrVelvetFog","#2325",0,"2026-07-25","2026-07-28"],[15,2310,"Prototype-inherited names are treated as language codes and return non-language values instead of null","MicroMilo",0,"DrVelvetFog","#2320",0,"2026-07-25","2026-07-27"],[15,2290,"whatever","gustavohof11-a11y",3,"","",0,"2026-07-12","2026-07-13"],[15,2287,"Add device_map=\"auto\" in default transformers snippets","pagezyhf",2,"","",0,"2026-07-10","2026-07-22"],[15,2259,"No documented or supported mechanism for Spaces to appear in ?category= task filters","rfonod",4,"","",0,"2026-06-29",""],[15,2255,"ultralytics library missing from sidebar filter on model hub despite being defined in model-libraries.ts","rfonod",2,"","",0,"2026-06-29","2026-06-29"],[15,2232,"Expand hardware catalog of processors","sharktide",4,"","",0,"2026-06-15",""],[15,2225,"Together ASR via router fails with 'Model not supported by provider together' for every model (audio/transcrip","maboa",2,"","",0,"2026-06-10","2026-07-13"],[15,2214,"Adding Intel(R) Core(TM) Ultra 7 255HX (2.40 GHz) to hardware list","Rafael-Silva-Oliveira",1,"Rafael-Silva-Oliveira","#2220",0,"2026-06-04","2026-06-05"],[15,2164,"Atomic Chat not appearing in \"Use this model\" or Settings → Local Apps","danyurkin",2,"","",0,"2026-05-11","2026-07-03"],[15,2150,"@huggingface/xetchunk-wasm published with broken metadata at npmjs.com breaking @huggingface/hub","sroussey",2,"","",0,"2026-05-08","2026-05-08"],[15,2148,"Missing hardware for RTX 3080 Laptop with 16G","bhundven",1,"bhundven","#2149",0,"2026-05-07","2026-05-22"],[15,2145,"Adding radeon igpus to hardware list","BlivionIaG",2,"","",0,"2026-05-07","2026-05-22"],[15,2119,"Model JSON is not Found from the URL","anuj-eminent",3,"","",0,"2026-04-23","2026-05-04"],[15,2107,"Hardware compatibility widget shows SOME QX_K_P but not all","hauhaut",4,"","",0,"2026-04-17",""],[15,2094,"Deprecation issue","dyeungling",3,"","",0,"2026-04-12","2026-04-13"],[15,2091,"Add mflux library to model-libraries.ts for download tracking","TimPietrusky",0,"Sanjays2402","#2196",0,"2026-04-10","2026-07-28"],[15,2073,"Please will you add model.safetensors as a default tracking file for counting downloads","lyndonlauder",2,"","",0,"2026-04-04","2026-04-08"],[15,2037,"List models should return dtype options","sroussey",2,"","",0,"2026-03-14","2026-04-02"],[15,2035,"KeyError: 'default'","Bhavesh-ahuja",3,"","",0,"2026-03-13","2026-03-13"],[15,1985,"Generated code snippets do not work in kaggle","yeswa-test",4,"","",0,"2026-02-18",""],[15,1984,"Generated code snippets do not work in colab","Abdul-Rehman745",4,"","",0,"2026-02-17",""],[15,1976,"Sorting results from listModels by downloads","brunnock",0,"coyotte508","4732d17",1,"2026-02-13","2026-02-16"],[15,1960,"Generated code snippets do not work in colab","singing2owl",4,"","",0,"2026-02-04",""],[15,1951,"local apps: add pi","gary149",2,"","",0,"2026-01-31","2026-03-24"],[15,1945,"ImportError: AlpamayoR1 not found when loading nvidia/Alpamayo-R1-10B","Gatwaza",2,"","",0,"2026-01-29","2026-01-30"],[15,1943,"Consider using uv instead of pip in model library snippets","gary149",2,"","",0,"2026-01-29","2026-01-30"],[15,1929,"[jinja] support separators in tojson filter","juanmarin-co",0,"Copilot","4610794",2,"2026-01-25","2026-01-26"],[15,1915,"@huggingface/gguf: ArrayBuffer resize fails on certain GGUF files with local file access","yuranich",1,"yuranich","#1918",0,"2026-01-14","2026-01-20"],[15,1903,"Whisper models are incorrectly mapped to Seq2SeqLM","irudachirath",4,"","",0,"2026-01-11",""],[15,1898,"@huggingface/blake3-wasm is slow","Brooooooklyn",0,"coyotte508","#1899",1,"2026-01-05","2026-01-20"],[15,1892,"Google Colab not working with predefined script for openai/whisper-large-v3","XDavidT",4,"","",0,"2025-12-21",""],[15,1868,"ax<YX<yx","sebagrz03",3,"","",0,"2025-12-04","2025-12-04"],[15,1850,"Generated code snippets for Tranception_small, Tranception_medium, and Tranception_large aren't working","tun66280",3,"","",0,"2025-11-21","2025-11-21"],[15,1831,"The download count for my model cannot be counted.","z-jiaming",2,"","",0,"2025-11-04","2025-11-04"],[15,1815,"model.safetensors download error","oromiaGodanna",3,"","",0,"2025-10-20","2025-10-21"],[15,1803,"mini cpm v-4.5 code snippet google colab","allisonandreyev",4,"","",0,"2025-10-14",""],[15,1802,"generated code snippets do not work In colab","EmanueleLippi",4,"","",0,"2025-10-14",""],[15,1793,"Add n1n.ai API as a model provider","n1n-api",4,"","",0,"2025-10-13",""],[15,1775,"Generated code snippets not working on google Colab","TJordanP",4,"","",0,"2025-09-27",""],[15,1773,"Tasks: Add Subtasks under Reinforcement Learning","MrShahzebKhoso",4,"","",0,"2025-09-27",""],[15,1769,"Using tasks/keypoint-detection in huggingface.js or transformers.js","jo-chemla",4,"","",0,"2025-09-25",""],[15,1752,"TinyLlama 1.1B not working in google collab","ravipras8",2,"","",0,"2025-09-17","2025-09-26"],[15,1751,"[BUG] \"Use This Model\" does not show up despite local apps being available","danielholanda",2,"","",0,"2025-09-16","2025-09-17"],[15,1726,"xet upload: more progress indicators","coyotte508",1,"coyotte508","#1727",1,"2025-09-04","2025-09-04"],[15,1723,"add query files for my library","kongzhecn",4,"","",0,"2025-09-03",""],[15,1722,"Code to use \"facebook/blenderbot-400M-distill\" is not working from \"use this model\" section","anupamsaraswat",4,"","",0,"2025-09-02",""],[15,1721,"xet upload: avoid downloading blobs twice?","coyotte508",4,"","",0,"2025-09-02",""],[15,1720,"Unnecessary `xet-fileinfo` request in `downloadFile` causes extra network overhead","viocha",3,"","",0,"2025-09-02","2025-09-02"],[15,1713,"No 304 response is returned even when ETag matches","viocha",2,"","",0,"2025-08-30","2025-09-01"],[15,1706,"Failed when running the generated code snippets on Google Colab","hieultimate",2,"","",0,"2025-08-27","2025-08-28"],[15,1705,"xet upload: allow changing only the beginning of the file","coyotte508",1,"coyotte508","#1718",1,"2025-08-26","2025-09-02"],[15,1704,"xet upload: parrallelize xorb/shard creation","coyotte508",4,"","",0,"2025-08-26",""],[15,1703,"xet upload: backtrack for dedup","coyotte508",1,"coyotte508","#1708",1,"2025-08-26","2025-08-28"],[15,1699,"jinaai/jina-embeddings-v4 snippet bug","Yelinz",0,"coyotte508","#1701",1,"2025-08-24","2025-08-25"],[15,1698,"Colab code not working for ByteDance-Seed/Seed-OSS-36B-Instruct","thejenilsoni",4,"","",0,"2025-08-21",""],[15,1697,"Downloads are not tracked for LVFace model.","GitHup2016yjh",2,"","",0,"2025-08-21","2025-08-21"],[15,1681,"Oi","Sameyboy88",3,"","",0,"2025-08-10","2025-08-10"],[15,1680,"Discrepancy between `client.sentence_similarity` input snippet & reality","tomaarsen",4,"","",0,"2025-08-08",""],[15,1671,"AttributeError: module diffusers has no attribute `QwenImagePipeline`","subhrm",2,"","",0,"2025-08-05","2025-08-12"],[15,1634,"To get register cyfuture.ai as inference service provider on hugginface","Cyfuture-AI",4,"","",0,"2025-07-17",""],[15,1625,"the `ProcessorMixin` class is not found in the huggingface.js library","yusuf-eren",4,"","",0,"2025-07-14",""],[15,1622,"CSS bad gateway errors for Spaces","anasuyab",2,"","",0,"2025-07-13","2025-07-16"],[15,1614,"Malformed Request during Initialization Tiny Agent","debasisdwivedy",0,"hanouticelina","#1618",1,"2025-07-11","2025-07-16"],[15,1609,"Tool Schema not OpenAI compatible","debasisdwivedy",0,"hanouticelina","#1610",1,"2025-07-11","2025-07-11"],[15,1603,"Assembly Script: pass uint8 arrays wihtout copy","coyotte508",4,"","",0,"2025-07-08",""],[15,1600,"Failed to load tngtech/DeepSeek-TNG-R1T2-Chimera due to missing file and FP8 hardware incompatibility","omarkamelte",4,"","",0,"2025-07-07",""],[15,1585,"[QUESTION] xet integration for file uploading","frascuchon",2,"","",0,"2025-07-02","2026-08-24"],[15,1553,"[Feature Request] Add Docker Model Runner as a Local App","felladrin",0,"ilopezluna","#1555",0,"2025-06-24","2025-07-01"],[15,1527,"DeepSeek-v3 notebook not working","josephSummerhays17",3,"","",0,"2025-06-12","2025-06-17"],[15,1525,"Link for Table Question Answering is wrong (now for Question Answering)?","John6666cat",2,"","",0,"2025-06-12","2025-06-12"],[16,1081,"Add a chunk decompression API that writes into a caller-provided buffer","assafvayner",4,"","",0,"2026-11-05",""],[16,1075,"bug: Remove unconditional ctor dependency from xet-runtime","Xuanwo",0,"seanses","#1079",1,"2026-11-02","2026-11-05"],[16,1058,"`SafeFileCreator` can publish incomplete files when dropped after a write failure","seanses",2,"","",0,"2026-10-26","2026-10-30"],[16,1053,"Upload progress stops below 100% when a large file deduplicates against its own data","rajatarya",4,"","",0,"2026-10-23",""],[16,1050,"Inverted byte range in download_stream / download_unordered_stream is not rejected","RaphaelFakhri",4,"","",0,"2026-10-22",""],[16,1036,"Dedup silently stops once the shard chunk index reaches chunk_index_table_max_size, but global dedup keeps dow","rajatarya",0,"seanses","#1039",1,"2026-10-20","2026-10-28"],[16,1018,"Download stalls repeatedly: one slow range request blocks the sequential writer, no timeout or retry is trigge","guoggg",4,"","",0,"2026-10-11",""],[16,992,"`XetUploadCommit::commit()` fails after one stream upload is aborted or dropped","glutamatt",4,"","",0,"2026-09-25",""],[16,991,"Avoid invalidating builds with `git_version!`","jamadeo",4,"","",0,"2026-09-25",""],[16,964,"perf: download stream groups repeatedly rebuild HTTP clients and TLS roots","Xuanwo",4,"","",0,"2026-09-10",""],[16,959,"chore: Fix deprecation warnings for `Atomic::fetch_update` renamed to `try_update`","lastcodingbeast",2,"","",0,"2026-09-04","2026-09-11"],[16,956,"Every 403 on xorb fetch is treated as URL expiry; body discarded, 0 bytes written for minutes before aborting","TomekZet",4,"","",0,"2026-09-02",""],[16,944,"`cargo test` fails with unresolved `clap` imports in `xet_pkg` examples","julienmoumne",0,"assafvayner","#955",1,"2026-08-19","2026-09-01"],[16,942,"XetFileDownloadGroup.abort() returns immediately but the aborted transfer keeps downloading to completion","chintan-diwakar",4,"","",0,"2026-08-16",""],[16,930,"Adaptative concurrency improvements - fair sharing expectation","coyotte508",4,"","",0,"2026-08-06",""],[16,929,"Multirange responses are buffered, stream them?","coyotte508",3,"","",0,"2026-08-06","2026-08-31"],[16,928,"hf-xet 1.4.3: concurrent repeated Bucket uploads fail with \"I/O error: failed to fill whole buffer\"","lewtun",2,"","",0,"2026-08-06","2026-08-07"],[16,927,"Download buffer defaults are constants: 2.7x left on a 20 Gbit/s host, OOM in a 1 Gi container","danielhanchen",0,"rajatarya","#943",1,"2026-08-04","2026-08-28"],[16,926,"with_high_performance() documents one of the twelve values it sets, and none of the memory cost","danielhanchen",3,"","",0,"2026-08-04","2026-09-24"],[16,921,"uv won't install","cthoyt",2,"","",0,"2026-07-29","2026-08-03"],[16,920,"Download fails with \"error decoding response body\"","seanses",4,"","",0,"2026-07-29",""],[16,909,"Feature request: support custom proxy","greenhat616",4,"","",0,"2026-07-22",""],[16,904,"Feature request: cryptographically signed models","xlr8harder",4,"","",0,"2026-07-20",""],[16,898,"Xet downloads stall indefinitely on GCP us-central1; plain HTTP path from the same host runs at ~87 MB/s (hf_x","liorn",2,"","",0,"2026-07-14","2026-07-15"],[16,897,"Xet downloads hang on GCP: reconstruction routes to us.gcp.cdn.hf.co, which rejects its own signed URLs (Signa","petkepau",2,"","",0,"2026-07-14","2026-07-22"],[16,896,"hf-xet 1.5.1 stalls around 2 GB reconstructing one model file on macOS and RunPod","JuiceB0xC0de",2,"","",0,"2026-07-13","2026-07-14"],[16,895,"Download fails with \"Task error: Unable to parse string as hex hash value\" (hf-xet 1.5.1)","btseytlin",0,"seanses","#4595",1,"2026-07-11","2026-07-28"],[16,892,"Investigate ReFS/Dev Drive support in Windows","rajatarya",4,"","",0,"2026-07-09",""],[16,891,"Remove auto-initialize pyo3 feature that breaks builds on `--disable-shared` Python (manylinux)","jvulgan",0,"rajatarya","#923",1,"2026-07-09","2026-08-03"],[16,889,"Enable CI Support for ppc64le (IBM Power) Architecture","sandeepgupta12",4,"","",0,"2026-07-08",""],[16,888,"Withdrawn","aisyah-revo",3,"","",0,"2026-07-06","2026-07-06"],[16,883,"XET is hanging when downloading large models (>=80% of the time)","ri938",2,"","",0,"2026-07-01","2026-07-16"],[16,875,"Adaptive concurrency: RTT predictor emits degenerate predictions from near-singular fits","assafvayner",4,"","",0,"2026-06-15",""],[16,869,"hf-xet download stalls indefinitely for facebook/sam3.1 large checkpoint, while regular HTTP download succeeds","Aryo-Zare",2,"","",0,"2026-06-10","2026-07-29"],[16,853,"`XetSession::new_upload_commit().build()` scans the entire shard cache (~4s) on every call","kszucs",4,"","",0,"2026-05-26",""],[16,851,"Termux: fails because `Expect rustls-platform-verifier to be initialized` ","Manamama",3,"","",0,"2026-05-23","2026-06-01"],[16,850,"Downloads Stall when hf-xet is Enabled","ConnoisseurProtege",2,"","",0,"2026-05-23","2026-07-16"],[16,846,"[Bug]: winget git-xet installation pollutes System PATH with a trailing double quote","yoooooolx",4,"","",0,"2026-05-21",""],[16,840,"hf-xet rust crate (xet_pkg) should support WASM target","assafvayner",1,"assafvayner","#841",1,"2026-05-11","2026-06-17"],[16,839,"Multiple WASM Runtime Panics ","Wrufesh",2,"","",0,"2026-05-11","2026-06-22"],[16,838,"WASM Client fails global deduplication for unkeyed shards","Wrufesh",2,"","",0,"2026-05-11","2026-05-11"],[16,828,"Update hf-xet Python publish workflow to not use \"maturin publish\"","seanses",2,"","",0,"2026-05-04","2026-06-22"],[16,822,"git-xet macOS release pipeline fails because some deps of an action are not pinned to commit SHA","seanses",2,"","",0,"2026-04-28","2026-05-06"],[16,821,"cas-bridge.xethub.hf.co: ~65-75% of CloudFront edge servers capped at 8.7 MB/s","aliasocracy",3,"","",0,"2026-04-28","2026-04-29"],[16,807,"Raise default `HF_XET_CLIENT_READ_TIMEOUT` from 120s to 300s","rajatarya",1,"rajatarya","#808",1,"2026-04-21","2026-05-18"],[16,803,"`git-xet` tries to modify `~/.config/git/config`, which is not allowed on NixOS","34j",3,"","",0,"2026-04-18","2026-05-06"],[16,800,"Xet-backed downloads stall on GCP (works when HF_HUB_DISABLE_XET=1)","eliran-hirundo",2,"","",0,"2026-04-16","2026-04-23"],[16,795,"Terminal token-refresh failure deadlocks hf_hub_download forever (xet)","VascoSch92",2,"","",0,"2026-04-13","2026-06-22"],[16,793,"Intermittent OSError when multiple processes call from_pretrained() concurrently on xet-backed models","lefft",2,"","",0,"2026-04-10","2026-09-14"],[16,790,"`HF_XET_CLIENT_RETRY_MAX_DURATION` environment variable is not enforced","saileshd1402",2,"","",0,"2026-04-08","2026-07-16"],[16,789,"hf_xet downloads stalling dead.","TidyWeb",2,"","",0,"2026-04-08","2026-04-19"],[16,763,"OS-level I/O errors (errno) are lost before reaching Python callers","cquil11",4,"","",0,"2026-03-30",""],[16,723,"FileUploadSession and FileDownloadSession could include a default token refresher for simpler integration","kszucs",0,"seanses","#771",1,"2026-03-17","2026-04-02"],[16,722,"Resolve security suggestions filed by dependabot","seanses",2,"","",0,"2026-03-16","2026-03-31"],[16,715,"No chunk cache despite `HF_XET_CHUNK_CACHE_SIZE_BYTES=10000000000`","amezin",2,"","",0,"2026-03-14","2026-03-16"],[16,700,"Add riscv64 (linux_riscv64) wheel to PyPI releases","gounthar",3,"","",0,"2026-03-11","2026-03-23"],[16,687,"Security: Please bump aws-lc-sys to 0.38.0","mikkeschiren",2,"","",0,"2026-03-09","2026-03-16"],[16,658,"hf-xet source builds fail due to python-source/module path mismatch","chenrui333",0,"rajatarya","#659",1,"2026-02-16","2026-02-17"],[16,628,"Rust API stabilization plan?","WSL0809",2,"","",0,"2026-01-29","2026-07-07"],[16,621,"git-xet installation error using the installation script for macOS/linux (hidden dependency on openssl3)","nicolaioestergaard",0,"seanses","#626",1,"2026-01-28","2026-02-09"],[16,608,"git_xet not present in PyPI tarball","xsuchy",2,"","",0,"2026-01-12","2026-01-12"],[16,604,"Race condition in FileProvider causes corrupted blob content during parallel downloads","volodymyrjf",0,"seanses","#641",1,"2025-12-30","2026-02-25"],[16,592,"CAS Bridge not working for large files","nicoboss",4,"","",0,"2025-12-10",""],[16,590,"Transparently mount super-lage directory using FUSE","ghost",2,"","",0,"2025-12-10","2026-03-23"],[16,588,"Feature Request: Add Windows ARM64 Support","mengshengwu",0,"rajatarya","#662",1,"2025-12-05","2026-02-24"],[16,585,"hf-xet SIGINT handler remains registered after completing operation (download/upload)","rajatarya",1,"rajatarya","#587",1,"2025-12-02","2026-01-12"],[16,582,"error cross compile","predators46",2,"","",0,"2025-11-26","2025-11-27"],[16,581,"Increased failure rate with `xet` over conventional `hub` downloads","leondz",2,"","",0,"2025-11-25","2026-02-25"],[16,572,"Please allow socks5h:// proxies.","imkzh",2,"","",0,"2025-11-20","2025-11-21"],[16,569,"Git push of dataset rejected because not using XeT","DanielNobbe",2,"","",0,"2025-11-19","2025-11-20"],[16,554,"wasm thread: TypeError: [object Int32Array] is not a shared typed array.","Wrufesh",2,"","",0,"2025-11-04","2025-12-03"],[16,549,"How to get the \"Xet backed hash\"?","arch-btw",2,"","",0,"2025-11-02","2025-11-06"],[16,548,"Upload error crashes `hf upload`","pwilkin",2,"","",0,"2025-10-31","2026-02-25"],[16,547,"[BUG] Shard footer_size validation not enforced for upload files - violates specification","volodymyrjf",2,"","",0,"2025-10-31","2025-11-04"],[16,527,"Xet not working on Kaggle TPUs","windmaple",4,"","",0,"2025-10-14",""],[16,526,"Bug: Python exception updating progress - (Pin `ipykernel < 7.0.0`)","achembarpu",2,"","",0,"2025-10-13","2025-10-29"],[16,512,"Cannot build hf-xet","clin1234",0,"rajatarya","#524",1,"2025-09-25","2025-10-21"],[16,500,"InvalidArchiveError with hf-xet when creating new conda environment","robinsibley",2,"","",0,"2025-09-19","2025-09-21"],[16,488,"Cannot download repo content from Space over Xet bridge (EU)","jsulz",2,"","",0,"2025-09-10","2025-09-12"],[16,483,"Still can't download models","frenzybiscuit",2,"","",0,"2025-09-08","2025-10-29"],[16,479,"Whole Xorb download","AvivFrog159",3,"","",0,"2025-08-28","2025-08-28"],[16,477,"The HTTP 1.3.1 on which hf_xet depends has the CVE-2022-36032 CVE-2023-26044 vulnerability. Does the vulnerabi","qianxia520",3,"","",0,"2025-08-28","2025-08-28"],[16,470,"hf-xet does not follow the HF_ENDPOINT environment variable","eric-gitta-moore",2,"","",0,"2025-08-24","2025-11-21"],[16,469,"Is git-xet(tool) deprecated and the plan is to use  xet-core with git LFS api specs","Wrufesh",2,"","",0,"2025-08-24","2025-08-26"],[16,465,"Empty metadata for hf-xet install","the01",0,"rajatarya","#472",1,"2025-08-20","2025-08-26"],[16,463,"\"No buffer space available\" error on macos-latest GitHub runner","jsulz",2,"","",0,"2025-08-19","2025-09-15"],[16,450,"Can't download models from huggingface xet on mac","julian-fong",2,"","",0,"2025-08-13","2025-08-14"],[16,448,"Downloads increase drive activity causing computer to freeze","jsulz",2,"","",0,"2025-08-10","2026-02-25"],[16,446,"xet too broken on windows: downloading a huggingface asset with xet hangs while others in the same method succ","doctorpangloss",2,"","",0,"2025-08-07","2025-12-17"],[16,444,"Download issue RuntimeError: Data processing error: CAS service error : IO Error: Timer expired","rbgo404",2,"","",0,"2025-08-06","2025-11-22"],[16,436,"Error Downloading mlx-community/XBai-o4-8bit","jsulz",2,"","",0,"2025-08-03","2025-08-06"],[16,417,"Cache directory env vars ignored, causing failures when ~/.cache isn't usable","half-duplex",0,"hoytak","#426",0,"2025-07-17","2025-07-31"],[16,415,"Error downloading from Xet repository in Zero GPU Space","jsulz",2,"","",0,"2025-07-15","2025-08-06"],[16,409,"Download stucks at 99%","FurkanGozukara",2,"","",0,"2025-07-10","2026-02-25"],[16,407,"Cannot download file from XET hosted repo using CLI","khulaifi95",2,"","",0,"2025-07-09","2025-09-15"],[16,404,"HTTP status client error (401 Unauthorized) during download","jsulz",2,"","",0,"2025-07-02","2025-08-06"],[16,400,"Unable to download XET files from repositories","jsulz",2,"","",0,"2025-06-30","2025-10-08"],[16,399,"Cannot Download XET Files","sanjifr3",2,"","",0,"2025-06-30","2025-07-07"],[16,398,"Options for self-hosting?","ggoggam",3,"","",0,"2025-06-26","2025-06-26"],[16,396,"Build wheels for the free-threaded Python build (3.13t)","ngoldbaum",2,"","",0,"2025-06-23","2025-10-24"],[16,394,"RuntimeError - Data processing error: CAS service error : Error : single flight error: Real call failed:","FurkanGozukara",2,"","",0,"2025-06-19","2025-10-30"],[16,391,"Remove spurious log messages when retries successful, swallow everything but ERROR & FATAL by default","rajatarya",2,"","",0,"2025-06-18","2025-09-15"],[16,390,"pypy cannot install hf-xet. Does datasets not support pypy","hajimichen",2,"","",0,"2025-06-18","2025-07-28"],[16,389,"Xet hub InvalidCertificate(UnknownIssuer)","michaelfeil",2,"","",0,"2025-06-17","2025-06-20"],[16,386,"lock file version `4` was found, but this version of Cargo does not understand this lock file","grzegorz-roboflow",2,"","",0,"2025-06-13","2025-06-16"],[16,375,"Reqwest(reqwest::Error { kind: Request, url: \"https://transfer.xethub.hf.co ...\"})","marcindulak",2,"","",0,"2025-06-06","2025-06-18"],[16,373,"Can't download models!","frenzybiscuit",2,"","",0,"2025-06-05","2025-06-19"],[17,949,"Add priorities to inference scheduling","vincevannoort",4,"","",0,"2026-10-09",""],[17,947,"Feature Request: Support for Google EmbeddingGemma 2","MartinEmrich",4,"wrosko","#1",1,"2026-10-08",""],[17,941,"FlashNomicBert (nomic-embed-text-v2-moe) crashes at warmup with CUDA_ERROR_INVALID_VALUE when `--max-batch-tok","dietmar",1,"dietmar","ba0b012",0,"2026-09-30","2026-11-05"],[17,927,"Released images crash on nomic-embed-text-v1.5/v1 current config (`duplicate field max_position_embeddings`) —","kwado-tech",2,"","",0,"2026-09-15","2026-11-06"],[17,926,"CSOAI — TEI interop: embeddings governance","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[17,925,"CSOAI — TEI interop: embeddings governance","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[17,924,"CSOAI — TEI AI governance embeddings","CSOAI-ORG",2,"","",0,"2026-09-15","2026-10-02"],[17,922,"Documentation proposal: Kubernetes deployment with a community Helm chart","mberlofa",4,"","",0,"2026-09-11",""],[17,914,"Non-flash F32 path fails with CUDA_ERROR_INVALID_VALUE on sm_120 with candle 0.9.2 (#809 branch)","MyoungHaSong",4,"","",0,"2026-08-14",""],[17,911,"Feature request: OTLP over HTTP (http-proto) exporter for OpenTelemetry traces","AmreshSinha",0,"beriktassuly","#940",0,"2026-08-06","2026-11-05"],[17,909,"Include OTel trace_id/span_id in JSON logs for log-trace correlation","bastiansmn",0,"modelpath-dev","#935",0,"2026-08-04","2026-11-05"],[17,902,"target-cpu=native causes SIGILL when running AMD-built binaries on Intel CPUs","thorfei",4,"","",0,"2026-07-28",""],[17,900,"Publish arm64 images with release numbers (e.g. cpu-arm64-1.9)","dangbert",2,"","",0,"2026-07-20","2026-11-06"],[17,895,"Header content-range missing during Artifactory proxy download","vincentkelleher",4,"","",0,"2026-07-09",""],[17,889,"fix: unbounded buffer allocation in paddedbatch.from_pb","kumburovicbranko682-boop",1,"kumburovicbranko682-boop","#890",0,"2026-06-27","2026-09-08"],[17,888,"Support --root-path CLI Argument","satendrakumar",0,"alvarobartt","#894",1,"2026-06-26","2026-11-05"],[17,882,"Qwen3/Gemma3 candle skip attention masks for equal-length batches","malaiwah",1,"malaiwah","#883",0,"2026-06-24","2026-11-05"],[17,881,"Health path isn't ignored for Opentelemetry.","bastiansmn",4,"","",0,"2026-06-24",""],[17,880,"te_queue_size metric does not increase when requests are appended to the queue","hanizav",0,"bhaskargurram-ai","#936",0,"2026-06-24","2026-11-05"],[17,879,"te_request_count (method=single) is incremented twice for single requests","hanizav",0,"original4422","#917",0,"2026-06-24","2026-09-07"],[17,874,"Incorrect retrieval results and inflated similarity scores for sdadas/stella-pl-retrieval-8k in TEI","Moraw1993",4,"","",0,"2026-06-12",""],[17,872,"Vera rubin support","eladamittai",4,"","",0,"2026-06-07",""],[17,870,"cuda-entrypoint.sh awk pattern fails on NVIDIA driver 6xx (CUDA UMD Version: header rename) → CUDA_ERROR_NO_DE","randomnimbus",0,"Anai-Guo","#871",0,"2026-05-26","2026-07-08"],[17,866,"Contextualized embedding endpoint","francoisWeber",4,"","",0,"2026-05-13",""],[17,864,"Add optional raw binary embeddings output to reduce serialization overhead","uasan",4,"","",0,"2026-04-16",""],[17,861,"The PPLX 0.6B model fails to fit within the memory capacity of an A100, resulting in out-of-memory errors.","vmarchenkoff",2,"","",0,"2026-04-09","2026-05-29"],[17,857,"Panic - Qwen3 4 model (Base Image version: cuda-1.9)","AnujShahSifthub",4,"","",0,"2026-04-04",""],[17,845,"Qwen3-Embedding-8B produces NaN embeddings for token 474 (\"import\")","cbwchuck",4,"","",0,"2026-03-15",""],[17,839,"Version 1.9.1 inside docker image 1.9.2","TomekZet",0,"alvarobartt","#849",1,"2026-03-04","2026-03-10"],[17,838,"`v1.9.2` HTTP server's `/info` endpoint returns `{ \"version\": \"1.9.1\" }`","slavistan",0,"alvarobartt","#849",1,"2026-03-03","2026-03-03"],[17,837,"Model startup takes minutes with `cuda-1.9`, only seconds with `1.9`","slavistan",4,"","",0,"2026-02-25",""],[17,833,"How to disable --auto-truncate via the `text-embeddings-router` CLI?","slavistan",0,"vrdn-23","#836",0,"2026-02-19","2026-02-23"],[17,822,"Requesting to support for Qwen3 VL embedding and Qwen3 VL reranker!!","ZHANG-SHI-CHANG",4,"","",0,"2026-02-12",""],[17,820,"NPU Support","Jeaniowang",4,"","",0,"2026-02-11",""],[17,819,"Decreasing warmup for use on spot instances","mattkrick",4,"","",0,"2026-02-10",""],[17,816,"Requesting to support for voyageai/voyage-4-nano","GaneshDoosa",2,"","",0,"2026-02-06","2026-11-06"],[17,815,"bfloat16 support","seagrine",4,"","",0,"2026-02-04",""],[17,813,"Invalid `prompt_name`s yield a '413 Payload too large' HTTP response","slavistan",0,"vrdn-23","#818",0,"2026-02-03","2026-02-09"],[17,811,"Panic in text-embeddings-inference due to internal queue overflow under high concurrency","ffxmm",0,"kozistr","#726",0,"2026-02-02","2026-02-12"],[17,807,"Inconsistency between HTTP and gRPC output normalization (Qwen3-Embedding)","Kyogre-Primal",0,"kozistr","#810",0,"2026-01-27","2026-02-09"],[17,806,"Compute cap 12.0+ support","YaroslavVoitovych",0,"danielealbano","#735",0,"2026-01-23","2026-01-24"],[17,804,"BAAI/bge-m3 models fails in SageMaker ml.g4dn.xlarge instance.","mjmanoj001",4,"","",0,"2026-01-16",""],[17,803,"Use the model stored on the local drive","MLKoz2",2,"","",0,"2026-01-13","2026-01-14"],[17,798,"Configurable Batch Channel Capacity for Pipeline Parallelism","jovsa",4,"","",0,"2026-01-08",""],[17,793,"Trying to download a model with a local disk mounted to the container","JxKim",2,"","",0,"2025-12-31","2025-12-31"],[17,791,"Hopper-1.8.3 image incompatible with CUDA 13.0 - Floating point exception on H20","khalidnass",4,"","",0,"2025-12-30",""],[17,788,"Thoughts on dynamic LoRA adapter support for embedding models in TEI?","BlackWingedKing",4,"","",0,"2025-12-24",""],[17,780,"[Feature]: could output the given format logger ?","ucas010",2,"","",0,"2025-12-18","2026-01-10"],[17,774,"maybe a bug :the docker results are diff with the sentence-transformers","ucas010",2,"","",0,"2025-12-12","2025-12-12"],[17,771,"ModernVBert support","noobHappylife",4,"","",0,"2025-12-05",""],[17,769,"ARM64 Support","swtb3-ryder",0,"nazq","#840",0,"2025-12-03","2026-03-31"],[17,768,"Tokio Runtime Worker Panick: Queue background task","pandysp",0,"kozistr","#726",0,"2025-11-26","2026-02-12"],[17,764,"Stell embeddings doesn't match with the official example (w/ sentence transformers)","girishponkiya",4,"","",0,"2025-11-24",""],[17,763,"Qwen3-Reranker-0.6B start error","Shengshenlan",4,"","",0,"2025-11-24",""],[17,760,"`Qwen/Qwen3-Embedding-0.6B` OOM error with Turing Docker image","juanjucm",0,"aalquwayfili","#953",0,"2025-11-19","2026-11-05"],[17,759,"Bug Report: Critical & High Vulnerabilities in HuggingFace Base Container Environment for TEI for tags cpu-1.6","Sanjay-Prajapathi",4,"","",0,"2025-11-17",""],[17,758,"TEI stops in \"Warming up model\" phase","gururaser",2,"","",0,"2025-11-16","2026-02-12"],[17,757,"Allow users to define specific onnx files","LLukas22",4,"","",0,"2025-11-13",""],[17,756,"Security issues (CVs) in images","StefanKarlsson321",0,"alvarobartt","#823",1,"2025-11-13","2026-02-12"],[17,754,"OpenAPI spec has wrong examples","Kablys",0,"alvarobartt","#755",1,"2025-11-10","2025-11-14"],[17,752,"Building docker containers is not supported for SM120","apaz-cli",0,"danielealbano","#735",0,"2025-11-05","2026-02-11"],[17,750,"`nomic-ai/nomic-embed-text-v1` - Failed to parse `config.json`: missing field `hidden_size`","chiragjn",0,"alvarobartt","#751",1,"2025-11-03","2025-11-04"],[17,749,"embeddinggemma-300m output is not the same as SentenceTransformer","Alireza3242",2,"","",0,"2025-11-01","2025-11-05"],[17,748,"embeddinggemma-300m: only supported in fp32 precision","Alireza3242",2,"","",0,"2025-11-01","2025-11-04"],[17,747,"/predict does not work for batches for sequence classification","vrdn-23",4,"","",0,"2025-10-30",""],[17,744,"Bug Report: Cannot embed text `Queue background task dropped the receiver or the receiver is too behind. This ","MrNemo64",0,"kozistr","#726",0,"2025-10-24","2026-02-12"],[17,741,"support Youtu-Embedding model","yuzhiruimmpp-web",4,"","",0,"2025-10-17",""],[17,740,"Extend support to compute capability 6.0/6.1","cduk",4,"","",0,"2025-10-11",""],[17,739,"Failed to deploy in NVIDIA H20","DaiJianghai",4,"","",0,"2025-10-11",""],[17,736,"Bug Report: Hugging Face TEI Vulnerabilities","Sanjay-Prajapathi",0,"alvarobartt","#823",1,"2025-10-06","2026-02-12"],[17,734,"Support Jina v3 Reranker","realsigridjin",4,"","",0,"2025-10-04",""],[17,733,"Release pre-built container images for aarch64 CUDA platforms","JCRPaquin",4,"","",0,"2025-10-03",""],[17,732,"TEI with local models don't load Dense layers","ghost",0,"alvarobartt","#738",1,"2025-10-02","2025-10-27"],[17,731,"Support Static Embeddings","franz101",4,"","",0,"2025-09-27",""],[17,729,"Prometheus /metrics Exposed on 8080 and not Default 9000","cmcguirk17",0,"augustas1","#678",0,"2025-09-25","2026-11-06"],[17,728,"Compile error in multiple environments for CPU MKL backend","nkh0472",2,"","",0,"2025-09-25","2026-03-22"],[17,723,"Infinite Loop When max-batch-tokens < model max_input_length","andrey-chernykh",0,"kozistr","#725",0,"2025-09-18","2025-09-25"],[17,722,"Gpu fallen of the bus - candle cuda health check for GPU health.","michaelfeil",4,"","",0,"2025-09-15",""],[17,721,"IPv6 support","MRiabov",2,"","",0,"2025-09-14","2025-09-15"],[17,714,"TEI 1.8.1 report \"ERROR\" when the \"Qwen3-Embedding-0.6B-ONNX\" model is used","quaggalinux",2,"","",0,"2025-09-08","2025-09-09"],[17,713,"Severely degraded reranking speed depending on --max-batch-requests","pandysp",4,"","",0,"2025-09-04",""],[17,710,"Which is better to deploy bge-m3 text embedding model? vllm or huggingface tei?","zengqingfu1442",2,"","",0,"2025-09-02","2025-09-15"],[17,709,"Error compiling locally on CUDA (with fix)","iamlemec",0,"alvarobartt","#719",1,"2025-08-29","2025-09-25"],[17,707,"support for token-classification","chenchunhui97",4,"","",0,"2025-08-28",""],[17,706,"nomic-embed-text-v2-moe on cpu, Intel MKL Error.","jens-create",0,"alvarobartt","#715",1,"2025-08-27","2025-09-09"],[17,704,"Qwen/Qwen3-Embedding-0.6B output error","asdfgh19",0,"alvarobartt","#705",1,"2025-08-20","2025-09-01"],[17,694,"Some inputs hang the whole embedding service on Qwen3-Embedding-0.6B","lambdaupb",0,"kozistr","#725",0,"2025-08-09","2025-09-25"],[17,693,"My 5090 is rusty because it has been waiting for too long.","trillionmonster",2,"","",0,"2025-08-08","2025-08-14"],[17,691,"Any plan to support qwen3 reranker model in TEI","tinafengfun",4,"","",0,"2025-08-04",""],[17,690,"Latest version avaiblable on gcr.io (Google Cloud Platform Artifact Repository) is 1.6 months after release of","tithercube",2,"","",0,"2025-07-31","2025-08-06"],[17,688,"Cuda level metrics collection!","iAdtya",4,"","",0,"2025-07-28",""],[17,687,"Masked Language Modeling","AmitMY",4,"","",0,"2025-07-26",""],[17,684,"Logging","chenchunhui97",2,"","",0,"2025-07-24","2025-07-25"],[17,683,"Openai compatiblity for Rerank models","Yiqian-Liu",4,"","",0,"2025-07-19",""],[17,682,"Support Jina's Embedding V4 Model","realsigridjin",4,"","",0,"2025-07-17",""],[17,680,"Error during the Launch Text Embeddings Inference","henrique-1",0,"alvarobartt","#715",1,"2025-07-10","2025-09-09"],[17,675,"Qwen3-Embedding-8B Could not start ORT backend: Could not start backend: DType float16 is not supported","cxx-cxx",2,"","",0,"2025-07-04","2025-07-09"],[17,673,"Feature: Support Matryoshka Representation Learning","binbabou",0,"kozistr","#676",0,"2025-07-03","2025-07-23"],[17,672,"HF Models Can't be downloaded with TEI:turing-1.5","NavKumarGit",4,"","",0,"2025-07-03",""],[17,670,"Issue with ghcr.io/huggingface/text-embeddings-inference:turing-1.7 Version (Turing Version for T4 GPU) for BA","NavKumarGit",4,"","",0,"2025-07-03",""],[17,669,"Support for VLM","SorenDreano",4,"","",0,"2025-07-02",""],[17,668,"Qwen3 0.6 inconsistent embedding","Noha-Magdy",2,"","",0,"2025-07-01","2025-07-16"],[17,667,"Can't start Qwen3 0.6B on CPU","teamclouday",0,"alvarobartt","#715",1,"2025-07-01","2025-09-09"],[17,665,"The output of Qwen3 0.6B model is not correct","diuzi",2,"","",0,"2025-06-30","2025-07-02"],[17,662,"Difference between single input and batch concurrency","augustas1",0,"kozistr","#726",0,"2025-06-27","2026-01-02"],[17,658,"Qwen3 4B model embs only 1024 dims (not 2560)?","deklanw",2,"","",0,"2025-06-26","2025-08-06"],[17,657,"Can we get automatic CPU-only ARM64 images?","tpendragon",0,"z4y4ts","#827",0,"2025-06-25","2026-03-30"],[17,652,"Add support for Blackwell architecture","msharara1998",4,"","",0,"2025-06-24",""],[17,649,"Inconsistent Embedding Outputs with qwen3-embedding-0.6B","Ztomepic",0,"kozistr","#650",0,"2025-06-19","2025-06-24"],[17,645,"Intel MKL instruction error on Intel CPU with Qwen3","090Gamer",2,"","",0,"2025-06-17","2025-06-20"],[17,644,"Qwen3-embedding got list of null","franklucky001",0,"alvarobartt","#692",1,"2025-06-17","2025-08-15"],[17,643,"Feature Request: Add Support for Qwen3-Reranker Model","franklucky001",4,"","",0,"2025-06-17",""],[17,642,"Qwen3-Embedding models: embeddings from TEI differ sharply from Sentence-Transformers reference","kevinmuto",0,"kozistr","#646",0,"2025-06-16","2025-06-18"],[17,640,"5090 support soon ?","trillionmonster",0,"danielealbano","#735",0,"2025-06-16","2026-02-11"],[17,636,"TEI CPU inference fails with Intel MKL errors on AMD processors when running Qwen3 embedding models","randomm",0,"alvarobartt","#715",1,"2025-06-13","2025-09-09"],[17,633,"Provide server APIs to allow text-embeddings-inference to be used directly in other Axum projects","matthewhaynesonline",4,"","",0,"2025-06-12",""],[17,630,"Qwen3 models only support CUDA devices with flash attention","randomm",1,"randomm","#632",0,"2025-06-11","2025-06-12"],[17,629,"Qwen3: Error: Model backend is not healthy","gardner",0,"alvarobartt","#631",1,"2025-06-10","2025-06-11"],[17,628,"Unknown output keys: name: \"logits\" answerdotai/ModernBERT-base","zbloss",2,"","",0,"2025-06-09","2025-09-14"],[17,626,"Add Qwen3 embedding models","ethankim00",0,"alvarobartt","#627",1,"2025-06-05","2025-06-10"],[18,2104,"[OpenVINO] 16-bit MoE expert weights escape NNCF compression with OpenVINO >= 2026.5 (Mixtral, PhiMoE, Qwen2-M","Wovchena",4,"","",0,"2026-11-03",""],[18,2102,"optimum-cli export openvino crashes (access violation) on LFM2.5-1.2B-Instruct hybrid architecture on Windows","andiekobbietks",4,"","",0,"2026-11-03",""],[18,2093,"optimum-cli export openvino with local model path produces OpenVINO IR model missing beam_idx input port, caus","shawn9977",4,"","",0,"2026-10-29",""],[18,2092,"optimum-cli export openvino cannot infer task from local model directory, even though config.json contains all","shawn9977",4,"","",0,"2026-10-29",""],[18,2086,"Phi-3.5-mini-instruct: OpenVINO IR produces garbage past ~4.5k prompt tokens (LongRoPE?)","aweussom",4,"","",0,"2026-10-26",""],[18,2080,"[OpenVINO] Support NemotronH (nemotron_h) for Nemotron 3.5 Lightning / Bolt","aweussom",4,"","",0,"2026-10-24",""],[18,2052,"Feature request: export AutoRound-quantized checkpoints to OpenVINO","blairducrayoppat",4,"","",0,"2026-10-06",""],[18,2050,"Error while exporting Qwen/Qwen-Image-2.1 to OpenVINO","azhuvath",4,"","",0,"2026-10-06",""],[18,2041,"Z image turbo create noise output on intel igpu","ak-sangal",2,"","",0,"2026-10-01","2026-10-06"],[18,2037,"Question: exporting Intel's Qwen3-Omni-30B-A3B int4-AutoRound checkpoint to OpenVINO","blairducrayoppat",2,"","",0,"2026-09-30","2026-10-06"],[18,2023,"[Guidance] multiple image edit  using Qwen image 2.1","ak-sangal",4,"","",0,"2026-09-24",""],[18,2011,"{Guidance} to speedup Qwen image model 2.1 in intel irisx iGPU","ak-sangal",2,"","",0,"2026-09-21","2026-09-24"],[18,2003,"Add OpenVINO export support for spark2_5 architecture (Spark-X2.5-4B / 1.7B)","spf1983",4,"","",0,"2026-09-17",""],[18,1998,"Difference in WWB Scores for LTX2 Text2Video compared to HF when using transformers v4/v5","goyaladitya05",2,"","",0,"2026-09-14","2026-09-18"],[18,1970,"z image turbo in GPU Bad OUTPUT","ak-sangal",2,"","",0,"2026-09-03","2026-10-01"],[18,1964,"[OpenVINO] Eagle3 tooling: greedy-equality test can't detect a degraded drafter; export needs einops, declared","blairducrayoppat",4,"","",0,"2026-08-29",""],[18,1948,"OpenVINO export can silently produce an IR that cannot use PagedAttention (no fused SDPA op)","aweussom",4,"","",0,"2026-08-21",""],[18,1934,"Feature Request: Qwen-Image-Edit OpenVINO Support","ak-sangal",4,"","",0,"2026-08-15",""],[18,1927,"[OpenVINO] Support Meta Muse Glimmer 30B (muse_glimmer)","aweussom",2,"","",0,"2026-08-12","2026-08-12"],[18,1921,"Feature Request] Support phi4-siglip / Phi-4-reasoning-vision-15B architecture in OpenVINO export","jonfon96",4,"","",0,"2026-08-08",""],[18,1878,"[OpenVINO] Support Speculative Decoding in Gemma 4 with model type 'gemma4_assistant' and 'gemma4_unified_assi","whatevermarch",4,"","",0,"2026-07-19",""],[18,1786,"`Qwen3_5DynamicCache` import error in optimum-intel v2.0.0 — class not found in transformers","PlanteAmigor",2,"","",0,"2026-06-11","2026-06-14"],[18,1785,"Gemma-4 export broken in v2.0.0 — TorchScript tracing fails after ONNX path removal","PlanteAmigor",2,"","",0,"2026-06-11","2026-06-17"],[18,1767,"Add OpenVINO export support for FLUX.2-dev / Flux2Pipeline","MichaelDevine",4,"","",0,"2026-06-05",""],[18,1764,"[OpenVINO] Support Gemma 4 12B with model type gemma4_unified","OpenArcBob",4,"","",0,"2026-06-05",""],[18,1754,"Native OpenVINO export for Qwen3.6 / Qwen3.5 MoE (model_type=qwen3_5_moe)","bob798",4,"","",0,"2026-05-28",""],[18,1748,"Native support for DeepSeek-V4 (model_type=deepseek_v4)","bob798",4,"","",0,"2026-05-26",""],[18,1722,"Bug: Qwen3.5-9B INT4 export produces incoherent output (hybrid linear/full attention model)","daviburg",4,"","",0,"2026-05-06",""],[18,1721,"Qwen3.6 (qwen3_5 / qwen3_5_moe) text-generation export blocked — only image-text-to-text task registered","bobvious",4,"","",0,"2026-05-05",""],[18,1720,"[Bug] Unable to load Qwen3.5 Models to GPU","droans",2,"","",0,"2026-05-05","2026-05-06"],[18,1708,"[Bug] Mixtral 8x22B (and other large MoE) cannot be exported on a sub-400 GB RAM machine — full model is loade","t8",4,"","",0,"2026-04-30",""],[18,1707,"[Bug] Misleading \"Unable to read the model\" error when intermediate IR is truncated during INT4 export","t8",4,"","",0,"2026-04-30",""],[18,1698,"[Code Quality] Replace Python runtime asserts with ValueError in ERNIE-Image model patchers","Rishabh-Sahni-0809",4,"","",0,"2026-04-22",""],[18,1697,"[Bug] Platform-specific code and hardcoded paths in test_ernie_image.py break tests on Windows","Rishabh-Sahni-0809",4,"","",0,"2026-04-22",""],[18,1696,"[Bug] Missing OVErnieImagePipeline dummy class causes ImportError when dependencies unavailable","Rishabh-Sahni-0809",2,"","",0,"2026-04-22","2026-04-22"],[18,1693,"Gemma 4 export to OpenVINO not supported","Tanay2104",2,"","",0,"2026-04-20","2026-05-05"],[18,1680,"Bug: sentence-transformers export fails due to removed _text_length attribute in version 5.4.0","MissLostCodes",4,"","",0,"2026-04-13",""],[18,1668,"Remote Code Execution via LlavaQwen2OpenVINOConfig and Phi3VisionOpenVINOConfig","Vancir",4,"","",0,"2026-03-31",""],[18,1667,"Feature Request: Native OpenVINO export and NNCF quantization support for SeamlessM4T-v2 (seamless_m4t_v2)","seferdemirci",4,"","",0,"2026-03-30",""],[18,1662,"Potential remote code execution via IPEXModelForCausalLM","Vancir",0,"echarlaix","#1666",1,"2026-03-30","2026-03-30"],[18,1654,"Add support for PaddleOCR-VL-1.5","purofle",4,"","",0,"2026-03-27",""],[18,1647,"Add support for moonshotai/Kimi-K2.5","MissLostCodes",4,"","",0,"2026-03-24",""],[18,1645,"Add tests which check, that required transformations are applied","popovaan",4,"","",0,"2026-03-23",""],[18,1639,"Add support for Python 3.12 and newer","Greendogo",4,"","",0,"2026-03-17",""],[18,1630,"Phi-3.5-vision-instruct invalid EOS token","przepeck",4,"","",0,"2026-03-03",""],[18,1628,"Qwen3.5 Family Support ❤️","savvadesogle",2,"","",0,"2026-03-02","2026-06-02"],[18,1614,"Fix conversion of ltx_video models in bf16 format","sbalandi",4,"","",0,"2026-02-13",""],[18,1611,"Tencent HY-MT1.5-1.8B model support","QingMaxwell",0,"popovaan","#1621",1,"2026-02-12","2026-02-26"],[18,1608,"GigaChat3-10B-A1.8B Support ❤️","savvadesogle",4,"","",0,"2026-02-08",""],[18,1595,"Unable to export Phi-4-multimodal-instruct using automatic-speech-recognition task","Fede2782",2,"","",0,"2026-01-24","2026-02-27"],[18,1594,"GLM4.7 Flash Support","hhunaid",4,"","",0,"2026-01-21",""],[18,1581,"Request for input shapes to be specified","danielliuce",4,"","",0,"2026-01-08",""],[18,1573,"Cannot export microsoft/Phi-3.5-vision-instruct","przepeck",2,"","",0,"2025-12-19","2026-01-12"],[18,1571,"Enable Llama 3.2 11B Vision Instruct in OpenVINO","abhijitsinha17",4,"","",0,"2025-12-18",""],[18,1567,"Support funasr/paraformer-zh to export OpenVINO IR model","10vliu13",4,"","",0,"2025-12-18",""],[18,1562,"MiniCPMO MAX_TRANSFORMERS_VERSION","RyanMetcalfeInt8",4,"","",0,"2025-12-17",""],[18,1552,"Being asked to provide dataset for text embedding submodel calibration when converting VLM using CLI","hermeschen1116",0,"nikita-savelyevv","#1553",0,"2025-12-12","2025-12-12"],[18,1543,"Tongyi-MAI/Z-Image-Turbo with optimum-cli but it gives me ZImagePipeline error for diffuser","hatrungdung",4,"","",0,"2025-12-03",""],[18,1537,"Missing package while exporting Qwen3-1.7B","przepeck",2,"","",0,"2025-12-02","2025-12-02"],[18,1528,"Errors during inference of a whisper model exported with --task automatic-speech-recognition","nikita-savelyevv",0,"echarlaix","#1534",1,"2025-11-24","2025-12-02"],[18,1516,"Got Caught exception: bad_function_call during inference on GPU","hermeschen1116",4,"","",0,"2025-11-11",""],[18,1510,"Potential remote code execution via untrusted quantization_config in OVQuantizer","Vancir",0,"nikita-savelyevv","#1520",0,"2025-11-03","2025-11-20"],[18,1500,"DeepSeek OCR support","qllesudo",4,"","",0,"2025-10-29",""],[18,1498,"[BUG] Import optimum-intel broke export","AlexanderDokuchaev",0,"echarlaix","#1501",1,"2025-10-28","2025-11-08"],[18,1496,"Optimum-CLI: Select and convert a specific LTX-Video model version","likholat",4,"","",0,"2025-10-24",""],[18,1488,"notebooks/openvino/vision_language_quantization.ipynb failed in Step 5c","yangxazhou",4,"","",0,"2025-10-23",""],[18,1478,"Qwen3-VL support","circuluspibo",2,"","",0,"2025-10-15","2026-02-10"],[18,1459,"[Windows 11] neural-compressor 3.5 requires numpy<2.0, but you have numpy 2.2.6 which is incompatible.","savvadesogle",2,"","",0,"2025-10-04","2026-03-18"],[18,1455,"openai/gpt-oss-20b fails to export","mwrothbe",2,"","",0,"2025-09-29","2025-09-30"],[18,1450,"Dev env recommendation","Wovchena",2,"","",0,"2025-09-25","2025-10-08"],[18,1448,"not support qwen2.5-omni","legend-7-7",4,"","",0,"2025-09-25",""],[18,1447,"Add version CLI option to optimum-cli","idubinov",4,"","",0,"2025-09-24",""],[18,1446,"Qwen3 OpenVINO support failing","tomaarsen",4,"","",0,"2025-09-19",""],[18,1444,"ipex CI failing due to missing/deleted models on the intel hf hub orgs","IlyasMoutawwakil",4,"","",0,"2025-09-08",""],[18,1425,"Please add support for openai/gpt-oss-20b","J-N-ch",2,"","",0,"2025-08-18","2025-08-18"],[18,1420,"Any examples of non-LLM stuff such as computer vision","ogencoglu",2,"","",0,"2025-08-10","2025-09-02"],[18,1419,"Gemma 3n support","viridivn",4,"","",0,"2025-08-09",""],[18,1418,"Whisper stateless model empty output with transformers 4.53.0","as-suvorov",0,"IlyasMoutawwakil","#1423",1,"2025-08-08","2025-08-12"],[18,1411,"Converting openai/gpt-oss-20b face issue","morteza89",2,"","",0,"2025-08-06","2025-08-07"],[18,1410,"Exporting model NousResearch/Hermes-3-Llama-3.1-8B doesn't create tokenizer and detokenizers openvino models","dtrawins",0,"nikita-savelyevv","#1414",0,"2025-08-06","2025-08-12"],[18,1409,"Model mistralai/Mistral-7B-Instruct-v0.3 doesn't have defined tasks","dtrawins",0,"nikita-savelyevv","#1413",0,"2025-08-06","2025-09-29"],[18,1405,"'Gemma3ForCausalLM' object has no attribute '_update_causal_mask_mm' with transformers-4.51.3","yangsu2022",0,"IlyasMoutawwakil","#1408",1,"2025-08-06","2025-08-06"],[18,1404,"HF - Model Export Issue with OSS models","ep150de",0,"echarlaix","#1406",1,"2025-08-05","2025-09-19"],[18,1398,"llava-v1.6-mistral-7b-hf export failure","Wovchena",0,"IlyasMoutawwakil","#1408",1,"2025-08-04","2025-08-06"],[18,1397,"Converting Qwen3-32B fails when using AWQ.","HumerousGorgon",2,"","",0,"2025-08-02","2025-08-28"],[18,1388,"OpenVINO export of mistralai/Mistral-7B-Instruct-v0.3 model is not converting the tokenizers","azhuvath",2,"","",0,"2025-07-21","2025-07-22"],[18,1380,"Convert bridgetower model to openvino","pkhara31",4,"","",0,"2025-07-11",""],[18,1353,"Are object detection tasks supported with openVINO?","Kallinteris-Andreas",2,"","",0,"2025-06-20","2025-08-04"],[18,1338,"Please add support for  mistral3 models for openvino export","yuanpf123",0,"kyoui-dev","#1627",0,"2025-06-11","2026-08-25"],[19,2975,"Add BEND 2.0 to the benchmark allow-list","yhsure",4,"","",0,"2026-11-04",""],[19,2966,"Supported way to bind a ZeroGPU proxy identity to an authenticated API caller","manuelk2607",4,"","",0,"2026-11-03",""],[19,2949,"Eval-results: eligibility and visibility for oracle-feedback agent entries (GPQA / AIME 2025)","Imontlaji",4,"","",0,"2026-10-29",""],[19,2938,"Verify Your Email One last step to start using Hugging Face  We've sent a verification email to:  nitin73111@g","Nittiin-oss",3,"","",0,"2026-10-28","2026-10-29"],[19,2873,"Paper authorship claim still pending for REMORY (2610.11287)","1ring2rta",2,"","",0,"2026-10-09","2026-11-05"],[19,2872,"Benchmark allow-list request: olaverse/african-typed-decisions","olastephen",4,"","",0,"2026-10-09",""],[19,2850,"Malware scanning page: files of 2 GiB and over are not scanned (0 of 487 in our sample), but show \"safe\"","MarkovianProtocol",4,"","",0,"2026-10-02",""],[19,2848,"Feature request: show separate MLX quantization variants and sizes in the Hub sidebar","FlorianVal",4,"","",0,"2026-10-02",""],[19,2835,"Benchmark leaderboards for mteb/arguana and mteb/BRIGHT do not list a model whose .eval_results parse","space-bacon",4,"","",0,"2026-09-30",""],[19,2829,"Support richer metadata in DOI registrations","adambuttrick",4,"","",0,"2026-09-30",""],[19,2812,"丸子","PrinceLan",3,"","",0,"2026-09-25","2026-09-25"],[19,2797,"Model Cards User Studies page links and images are invalid","thavlice",0,"hafsakhan09090","#2819",0,"2026-09-20","2026-09-28"],[19,2778,"GitHub-style profile README for user profile pages","Adirohansuyal",4,"","",0,"2026-09-12",""],[19,2760,"Bounded window of 'Scheduling failure: unable to schedule' on Streamlit Spaces, Nov 2025 - May 2026","ashishsinha1602",4,"","",0,"2026-09-03",""],[19,2757,"Bucket trusted publisher: POST .../settings/trusted-publishers 500s at 10s, and /oauth/token returns the same ","hadim",2,"","",0,"2026-09-03","2026-09-22"],[19,2747,"Benchmark allow-list request + three design questions: distribution-fidelity (KL divergence) evaluation result","malaiwah",4,"","",0,"2026-08-30",""],[19,2743,"38.6% of non-static Spaces are in a permanent error state — full census, and a filtering suggestion","ashishsinha1602",2,"","",0,"2026-08-29","2026-08-31"],[19,2725,"music","jonbaton14-beep",2,"","",0,"2026-08-23","2026-08-23"],[19,2724,"Music","jonbaton14-beep",2,"","",0,"2026-08-23","2026-08-23"],[19,2704,"Parameter count regression: metadata.total_parameters ignored for packed/quantized models (89B shown for a 314","avlp12",0,"mishig25","#2415",1,"2026-08-13","2026-09-02"],[19,2702,"[Bug] Sharded GGUF tags rejected with 400 by the Ollama registry endpoint","kappa8219",0,"Pierrci","#19532",1,"2026-08-13","2026-08-13"],[19,2674,"Pressing Enter to confirm Chinese IME composition unexpectedly submits the search form","zhaoyu190",0,"kefranabg","#19558",1,"2026-07-31","2026-08-18"],[19,2648,"### Request: add `IntelligenceLab/Long-Horizon-Terminal-Bench` to the benchmark allow-list","zli12321",4,"","",0,"2026-07-16",""],[19,2598,"Add `webnn` to the Spaces embedding iframe `allow` attribute (WebNN now blocked by Permissions Policy in Chrom","ibelem",0,"moon-bot-app","#18800",2,"2026-06-26","2026-06-29"],[19,2582,"Generated Colab notebook uses HF Router for models without inference providers","OMCHOKSI108",2,"","",0,"2026-06-23","2026-06-23"],[19,2576,"z","dwallman1-pixel",2,"","",0,"2026-06-20","2026-06-20"],[19,2553,"`deepseek-ocr-sagemaker` notebook references uncommitted code (`entry.sh`, `sm_job_runner.py`, `llm_ocr/`)","dwarez",4,"","",0,"2026-06-11",""],[19,2525,"Add `dagster-hf-datasets` into Integrated Libraries","ParagEkbote",3,"","",0,"2026-06-03","2026-06-19"],[19,2503,"Real download count tracking","xianbaoqian",2,"","",0,"2026-05-27","2026-05-27"],[19,2481,"image-to-text task not supported for zai-org inference provider despite model page examples","raccoon70",4,"","",0,"2026-05-16",""],[19,2456,"Request benchmark allow-list — benchflow/skillsbench","xdotli",2,"","",0,"2026-05-05","2026-07-03"],[19,2452,"Notification issue huggingface website","RichardErkhov",2,"","",0,"2026-05-03","2026-05-25"],[19,2440,"Java client for HuggingFace Hub","haifengl",0,"Wauplin","#2444",1,"2026-04-27","2026-04-29"],[19,2432,"Cannot access account due to lost email – request repo deletion","YUYANG0915",2,"","",0,"2026-04-24","2026-04-24"],[19,2404,"Integration: PaperClaw — autonomous research paper generator for Hugging Face","Agnuxo1",4,"","",0,"2026-04-17",""],[19,2392,"Misleading Pro Plan documentation","douglasware",2,"","",0,"2026-04-15","2026-04-17"],[19,2364,"Email verification link is sent, but the page is blank","gavrilovalova",2,"","",0,"2026-04-07","2026-04-10"],[19,2359,"hf_xet downloads stall","TidyWeb",4,"","",0,"2026-04-04",""],[19,2356,"ZeroGPU workers can't access files downloaded by main process (FileNotFoundError)","pelegs202-design",2,"","",0,"2026-04-03","2026-04-09"],[19,2317,"Hub website bug: “Use this model” dropdown stays on loading placeholders for GGUF quants across devices/browse","John6666cat",2,"","",0,"2026-03-20","2026-03-20"],[19,2311,"Where to request benchmark allow-list access?","lianghsun",4,"","",0,"2026-03-20",""],[19,2290,"Add allow=\"web-share\" to Spaces iframe for Web Share API support","SyntaxSphinx",2,"","",0,"2026-03-13","2026-04-27"],[19,2209,"BUG","zakriaalgobani",3,"","",0,"2026-02-06","2026-02-09"],[19,2174,"UI Bug: 'Ask a Question Open In Colab Open In Studio Lab' tab overlaps with chapter titles","Almexiir93",4,"","",0,"2026-01-21",""],[19,2172,"Stlite Template Proposal","1TSnakers",4,"","",0,"2026-01-20",""],[19,2143,"OpenAPI documentation does not contain Repo listing API","oleksii-donets",4,"","",0,"2026-01-09",""],[19,2136,"Webhook not work on some orgs","bunnynode",4,"","",0,"2026-01-06",""],[19,2122,"Support custom query file for download tracking","GengzeZhou",2,"","",0,"2025-12-17","2025-12-17"],[19,2114,"`whoami-v2` undocumented around auth `expiresAt` and `createdAt`","CodeSandwich",0,"coyotte508","#16128",1,"2025-12-15","2025-12-15"],[19,2113,"cannot change hardware list on https://huggingface.co/settings/local-apps","yukuz0",2,"","",0,"2025-12-14","2026-02-26"],[19,2092,"Duplicate Installation Instructions","Kaszanas",2,"","",0,"2025-12-04","2026-01-05"],[19,2073,"Registration Verification mail on click link blank screen","mariusgb",4,"","",0,"2025-11-23",""],[19,2040,"Huggingface's Next Top Model","TimeLordRaps",4,"","",0,"2025-11-07",""],[19,2031,"`SmolLM3` is listed under Multimodal in `transformers-v4.57` API","Axe--",0,"yyouretoast","#46770",0,"2025-11-03","2026-06-19"],[19,2005,"Add repo-wide docs quality gates: Markdown linting and external link checks","Tejassveer08",4,"","",0,"2025-10-21",""],[19,1998,"https://storage.googleapis.com/pai -images/1794178609621966848.jpeg \n\n","Hosseinkhosravi-web",2,"","",0,"2025-10-16","2025-10-16"],[19,1976,"nasa ai","maxtronh",2,"","",0,"2025-10-06","2025-10-07"],[19,1949,"Space direct locall install","TOPG-V",2,"","",0,"2025-09-24","2025-11-19"],[19,1931,"Request Organization Verification for “Qwen” (Official Alibaba Tongyi Lab LLM Series)","myazid",4,"","",0,"2025-09-18",""],[19,1929,"Spaces - Gradio iframe height invalid on load","catwell",4,"","",0,"2025-09-16",""],[19,1928,"Truncated Python version in README.md preview (3.10 -> 3.1)","catwell",4,"","",0,"2025-09-16",""],[19,1922,"LinkedIn profile field rejects valid usernames with accented characters","RDTvlokip",2,"","",0,"2025-09-13","2025-09-16"],[19,1906,"wrong repo link in Dinov2ForImageClassification doc example","dnsch",2,"","",0,"2025-09-08","2026-04-30"],[19,1905,"File detection Problem at Huggingface Private Repository","ShivamDhawan330",2,"","",0,"2025-09-06","2026-02-03"],[19,1901,"Update the link to the Hub API playground to https://huggingface.co/spaces/huggingface/openapi","julien-c",0,"copilot-swe-agent","#1902",2,"2025-09-05","2025-09-10"],[19,1899,"update number of models, datasets, spaces, similar to #1751","julien-c",0,"copilot-swe-agent","#1900",2,"2025-09-05","2025-09-05"],[19,1889,"CORS: Add `Access-Control-Max-Age` Header to All API and Space Endpoints to Reduce Preflight Requests","viocha",2,"","",0,"2025-08-30","2025-09-15"],[19,1869,"PerceptionLM Document Issue","wonderwind271",4,"","",0,"2025-08-09",""],[19,1865,"Can we please get sort option to repositories?","FurkanGozukara",4,"","",0,"2025-08-06",""],[19,1855,"MONSTERDOG","Sameyboy88",2,"","",0,"2025-07-30","2025-07-31"],[19,1849,"Dataset splits manual configuration clarification?","ego-thales",4,"","",0,"2025-07-25",""],[19,1846,"MONSTERDOG ","Sameyboy88",2,"","",0,"2025-07-24","2025-07-30"],[19,1840,"An issue occurs while opening multiple pages of the /learn or /docs at one time","abukanabek",4,"","",0,"2025-07-17",""],[19,1836,"Gemma Documentation Seems to Incorrectly Use System Prompts","mitchelldehaven",4,"","",0,"2025-07-11",""],[19,1831,"Any doc page: Copy page as markdown option","gary149",4,"","",0,"2025-07-10",""],[19,1818,"api-inference.huggingface.co/models/xyz started giving 404","kanishmalviya",4,"","",0,"2025-07-07",""],[19,1795,"Unable to install packages that depend on libc6 using packages.txt in Gradio Spaces","John6666cat",4,"","",0,"2025-06-25",""],[19,1789,"Incorrect documentation for vision-language models with language modeling head","Aparna-B",4,"","",0,"2025-06-18",""],[19,1781,"Colab link dissapeared, StudioLab link appeared in the docs","e-ferrari",4,"","",0,"2025-06-12",""],[19,1775,"403 Not Authorized on Email Verification (Unable to Access Hugging Face APIs)","manish3089",4,"","",0,"2025-06-10",""],[19,1766,"Recommended 'text-generation' models unavailable in Inference API","sperezacuna",4,"","",0,"2025-06-04",""],[20,2735,"ML Intern: give the agent reference Spaces so complex workflow apps come out right first time","pngwn",4,"","",0,"2026-10-28",""],[20,2734,"ML Intern: create GPU-backed Spaces on ZeroGPU by default","pngwn",4,"","",0,"2026-10-28",""],[20,2733,"ML Intern: measure and reduce time to first useful artifact","pngwn",4,"","",0,"2026-10-28",""],[20,2732,"ML Intern: allow steering or interrupting a running subagent","pngwn",4,"","",0,"2026-10-28",""],[20,2731,"ML Intern: per-session compute budget maximum is too low for paper reproductions","pngwn",4,"","",0,"2026-10-28",""],[20,2730,"ML Intern: research unfamiliar named things before trying generative approaches","pngwn",4,"","",0,"2026-10-28",""],[20,2729,"ML Intern: enforce a sandbox run before submitting GPU jobs","pngwn",4,"","",0,"2026-10-28",""],[20,2728,"ML Intern Slack bot: session controls (Open in HuggingChat, Stop session) are shown to everyone in the thread","pngwn",4,"","",0,"2026-10-28",""],[20,2727,"ML Intern Slack bot: answering an interrupted question shows ⚠ and a stale form","pngwn",4,"","",0,"2026-10-28",""],[20,2726,"ML Intern Slack bot: budget grants are easy to miss: typed amounts are ignored, and some offered options don't","pngwn",4,"","",0,"2026-10-28",""],[20,2725,"ML Intern: session goes idle after the model hits its output limit mid-task","pngwn",4,"","",0,"2026-10-28",""],[20,2724,"ML Intern: sandbox has no Hub token, so pushing to the Hub and import_file from the sandbox fail","pngwn",4,"","",0,"2026-10-28",""],[20,2720,"Prompt cache hit rate is ~15% on tool-enabled conversations","gary149",4,"","",0,"2026-10-27",""],[20,2698,"Share id follow-ups from #2697: message DELETE 500, resolveConversation types, turn state test","pngwn",1,"pngwn","#2699",1,"2026-10-20","2026-10-23"],[20,2603,"UI error when connected to Azure OpenAI models","dmaltsiniotis",4,"","",0,"2026-09-22",""],[20,2601,"Allow filtering of specific models for use","dmaltsiniotis",4,"","",0,"2026-09-22",""],[20,2599,"Shared conversation view returns 500 for logged-out visitors","gary149",4,"","",0,"2026-09-21",""],[20,2573,"ML Intern: a fresh chat started from a prompt example is created without the mode","pngwn",4,"","",0,"2026-09-10",""],[20,2564,"Half-finalized turn: producer death between the two interrupt writes leaves an unterminated message and a clie","pngwn",4,"","",0,"2026-09-07",""],[20,2537,"Turn continuity P4: reaper closure in-band, producer fencing, and guarantee tests","pngwn",4,"","",0,"2026-08-28",""],[20,2536,"Turn continuity P3: unify the POST transport onto the turn subscription channel","pngwn",4,"","",0,"2026-08-28",""],[20,2527,"Free OpenAI-Compatible Demo Endpoint with Built-In Monetization via ChatADy","chatady-com",4,"","",0,"2026-08-26",""],[20,2522,"[ask_user_question tool] in the \"something else\" textbox, clicking enter/return should submit the form","pngwn",4,"","",0,"2026-08-24",""],[20,2512,"chat titles are \"We need to produce a title....\".","pngwn",0,"revanth-045","#2515",0,"2026-08-20","2026-08-24"],[20,2502,"[Feature Request]: Connect Hugging Face Chat to Private Organization Models and Inference Endpoints","amine759",4,"","",0,"2026-08-15",""],[20,2500,"[Feature Request] Support organizing conversations into Projects / Folders (similar to Claude / ChatGPT Projec","barrettj12",4,"","",0,"2026-08-13",""],[20,2497,"Cloudflare ","amirsource133",4,"","",0,"2026-08-13",""],[20,2489,"Serving over HTTP (not https) broken in docker container","dmaltsiniotis",4,"","",0,"2026-08-06",""],[20,2460,"Artifact screenshot annotation: add a text tool","gary149",1,"gary149","#2463",1,"2026-07-29","2026-07-30"],[20,2429,"https://huggingface.co/chat/models/google/gemma-4-31B-it","abdulhafeezenayatallh-lang",4,"","",0,"2026-07-15",""],[20,2397,"ai-chat-webapps","FlyTripVisa",4,"","",0,"2026-06-30",""],[20,2394,"lifecycle_outside_component: onDestroy registered inside async onMount in +layout.svelte","LoneRifle",1,"LoneRifle","#2396",0,"2026-06-30","2026-07-01"],[20,2393,"Feature: Prompt compression to reduce token costs in chat conversations","arjunkshah",4,"","",0,"2026-06-29",""],[20,2387,"addSibling throws unnecessarily when target is the root message","nuthalapativarun",4,"","",0,"2026-06-26",""],[20,2386,"fix: addSibling incorrectly rejects root message as sibling target","nuthalapativarun",4,"","",0,"2026-06-26",""],[20,2373,"Created by mistake — please ignore","zzzonghwa",3,"","",0,"2026-06-19","2026-06-19"],[20,2344,"Feature request: add offline mode - read conversations without connectivity","LoneRifle",4,"","",0,"2026-06-12",""],[20,2286,"Feature request: Add FunASR for voice input transcription","LauraGPT",4,"","",0,"2026-05-31",""],[20,2282,"CI: lint workflow uses `npm install ci` instead of `npm ci`","fallintoplace",4,"","",0,"2026-05-24",""],[20,2243,"Application can't send chats to local VLLM","jerkstorecaller",0,"anishesg","#2249",0,"2026-04-21","2026-04-30"],[20,2242,"PinnochIA","NickelRamQc94",2,"","",0,"2026-04-20","2026-05-21"],[20,2234,"Mini audit: possible false-action-claim gap","ghost",2,"","",0,"2026-04-13","2026-05-21"],[20,2214,"Add Clarvia AEO badge — agent-readiness score for HuggingChat","digitamaz",2,"","",0,"2026-04-01","2026-05-21"],[20,2194,"Make TEXT_MIME_ALLOWLIST configurable via an environment variable","dxspotxfinsxnity",4,"","",0,"2026-03-17",""],[20,2188,"RFC: Let models declare accepted file MIME types for native file handling","tessaherself",4,"","",0,"2026-03-16",""],[20,2161,"Urge DOJ & FBI to Protect AI Innovation Even as the U.S. Focuses on the Iran War","NobleResearch",2,"","",0,"2026-03-03","2026-05-21"],[20,2131,"Image not showing","IMaduwike",4,"","",0,"2026-02-17",""],[20,2084,"Hovering on Pasted content","nakul-py",4,"","",0,"2026-01-28",""],[20,2067,"Docs suggestion: canonical production deployment example","corradoizzo",4,"","",0,"2026-01-21",""],[20,2066,"Installation & production deployment is fragile and undocumented (manifest.json / base path issues)","corradoizzo",0,"claude","052c91b",0,"2026-01-21","2026-01-26"],[20,2064,"Copy button for messages","nakul-py",4,"","",0,"2026-01-21",""],[20,2033,"Hovering on Close button makes it disappears in dark theme.","nakul-py",1,"nakul-py","#2034",0,"2026-01-07","2026-01-07"],[20,1991,"Response of Kimi-K2-Instruct-0905 with agentic stops before completing.","VinayHajare",2,"","",0,"2025-11-24","2026-01-16"],[20,1986,"HI i would like to use  default_headers={         \"X-HF-Bill-To\": \"org-name\"     } in my chatui local deployme","aditya-oss-prog",0,"gary149","#2054",1,"2025-11-17","2026-01-19"],[20,1984,"Support mcp ui","Borg93",4,"","",0,"2025-11-13",""],[20,1957,"Fail to use proxy","geek0011",4,"","",0,"2025-10-27",""],[20,1956,"Failed to load models from OpenAI base URL","anhnh2002",2,"","",0,"2025-10-26","2026-01-20"],[20,1947,"HuggingChat MoM (Mixture-of-Models) Integration Proposal 🤗","Xunzhuo",4,"","",0,"2025-10-19",""],[20,1944,"Voice calls","lee-b",4,"","",0,"2025-10-17",""],[20,1941,"huggingchat shows error 413","xfst14",4,"","",0,"2025-10-17",""],[20,1940,"Add a web search tool on the new huggingface tool","xfst14",2,"","",0,"2025-10-17","2026-01-20"],[20,1920,"Possible .dockerignore misconfiguration","tomokinakamaru",4,"","",0,"2025-10-06",""],[20,1870,"Empty content in message payload causes error with some OpenAI-compatible endpoints","BugsBuggy",4,"","",0,"2025-08-21",""],[20,1861,"Agent Communication Protocol (ACP) support","antonkulaga",4,"","",0,"2025-06-26",""],[20,1858,"Cannot enter text on Hugging Chat for Firefox","Question2",2,"","",0,"2025-06-18","2025-06-18"],[20,1857,"Feature Request: Display Step-by-Step Summarized Reasoning Content","jinbagi",4,"","",0,"2025-06-17",""],[20,1855,"Add User Identification Headers to OpenAI Requests","ishatalkin",4,"","",0,"2025-06-16",""],[20,1852,"No Output was Generated Flask OpenAI Endpoint","RobertEichner",4,"","",0,"2025-06-12",""],[20,1850,"Internal Server Error when accessing `Share URL`","jinbagi",2,"","",0,"2025-06-10","2025-06-10"],[20,1844,"Search does not support partial string matching","jinbagi",4,"","",0,"2025-06-04",""],[20,1843,"can you make a release?","antonkulaga",4,"","",0,"2025-06-02",""],[20,1842,"Not working on mobile (UI appears and disappears instantly)","KaKi87",4,"","",0,"2025-06-01",""],[21,2882,"-----","Satoshi-Takakura",2,"","",0,"2026-10-28","2026-10-28"],[21,2877,"Clarify Apache-2.0 and ImageNet notice for efficientnet_b3.ra2_in1k weights","cloudsddss",2,"","",0,"2026-10-23","2026-10-23"],[21,2840,"train.py: AttributeError on 'NoneType' t_in_epochs when --sched is 'none' or unrecognised","caiotheodoro",0,"rwightman","#2843",1,"2026-10-08","2026-10-08"],[21,2816,"[BUG] list_models/list_pretrained: str exclude_filters is iterated char by char when pretrained tags are inclu","JoeyTan21",0,"rwightman","1816c27",1,"2026-09-29","2026-09-29"],[21,2814,"[BUG] Adan.__setstate__ does not back-fill 'foreach', so a checkpoint saved before timm 1.0.12 raises KeyError","Nicholas022400701",2,"","",0,"2026-09-28","2026-09-28"],[21,2813,"[BUG] Adopt foreach path takes the first-step check and the clip value from device_state_steps[0], so a param ","Nicholas022400701",1,"Nicholas022400701","47ab1cc",0,"2026-09-28","2026-10-06"],[21,2800,"init_distributed_device_so() never binds the device for non-CUDA accelerators (Ascend NPU): `--device npu:N` s","li-lizhe",1,"li-lizhe","#2801",0,"2026-09-23","2026-09-24"],[21,2782,"[BUG] augmix fails on non-square and single-band images: mixing buffer is allocated (width, height, bands)","caiotheodoro",1,"caiotheodoro","#2783",0,"2026-09-14","2026-09-14"],[21,2781,"[BUG] continuously increasing reserved gpu memory caused by `PrefetchLoader`","sjiang95",0,"rwightman","cf33e92",1,"2026-09-14","2026-09-14"],[21,2773,"Add Qwen-Drive-1.0-4B vision tower as a qwen3_vit pretrained tag","developer0hye",1,"developer0hye","#2774",0,"2026-09-11","2026-09-17"],[21,2772,"Add DeepSeek-V4 / V4.1 vision encoder (deepseek_vit)","developer0hye",1,"developer0hye","#2775",0,"2026-09-11","2026-09-11"],[21,2766,"[BUG] `PatchEmbed` returns a view tensor, triggering an FSDP2 warning","sjiang95",2,"","",0,"2026-09-09","2026-09-17"],[21,2764,"[FEATURE] Add Qwen3.5 / Qwen3.8 (Qwen3-VL) vision encoder as a timm backbone","developer0hye",1,"developer0hye","#2765",0,"2026-09-08","2026-09-10"],[21,2762,"[FEATURE] Add Sapiens2 (Meta human-centric ViT w/ RoPE + GQA + registers) to Eva","developer0hye",1,"developer0hye","#2763",0,"2026-09-08","2026-09-09"],[21,2729,"[FEATURE] Adding ISNet as new model","CoderOMaster",2,"","",0,"2026-07-29","2026-10-16"],[21,2725,"[BUG] `timm.create_model` failed when local path contain valid character \"#\"","NewUserHa",0,"rwightman","64512ab",1,"2026-07-21","2026-07-26"],[21,2712,"[BUG] reg_token excluded from group_matcher()/no_weight_decay() in Eva and VisionTransformer","alexanderctrl",0,"rwightman","2cff235",1,"2026-07-04","2026-07-04"],[21,2711,"[FEATURE] Add MogaNet (ICLR 2024) backbone","shakhsm123",2,"","",0,"2026-06-28","2026-07-13"],[21,2700,"[FEATURE] Add TIPSv2 model","Jserax",2,"","",0,"2026-04-24","2026-05-27"],[21,2693,"[BUG] torch.compile applied at task level has no effect during validation and causes multi-node stalls","janbernloehr",0,"rwightman","56e3981",1,"2026-04-10","2026-04-24"],[21,2691,"[BUG] The behaviour of `param_groups_layer_decay()` towards `requires_grad==False` parameters","sjiang95",2,"","",0,"2026-04-09","2026-04-22"],[21,2690,"[FEATURE] EUPE pretrained weights","ankandrew",0,"rwightman","7377f08",1,"2026-04-04","2026-05-27"],[21,2687,"[BUG] FX tracing fails for ViT due to dynamic is_causal passed to scaled_dot_product_attention","YitianYu69",2,"","",0,"2026-03-30","2026-04-06"],[21,2682,"[BUG] `AttentionPool2d` produces non-deterministic outputs when `num_classes=0`","karray",0,"rwightman","3e8def8",1,"2026-03-13","2026-03-13"],[21,2681,"Potential mismatch with original DINOv3: timm models use avg pooling","hassonofer",0,"rwightman","8b4239c",1,"2026-03-10","2026-03-17"],[21,2671,"Add timmx to Awesome PyTorch Resources","Boulaouaney",2,"","",0,"2026-02-24","2026-02-26"],[21,2670,"[FEATURE] Remove the usage of `pickle` throughout the repository.","nv-fszarwacki",2,"","",0,"2026-02-19","2026-02-19"],[21,2663,"[BUG] Deprecation of `@torch.jit.interface`","mscheltienne",0,"rwightman","bdab30b",1,"2026-02-06","2026-02-06"],[21,2661,"[BUG] Swin Transformer relative_position_index not initialized when loaded via transformers TimmBackbone","CharlesCNorton",0,"rwightman","41754f2",1,"2026-02-05","2026-02-06"],[21,2659,"[FEATURE] Support for Mamba architecture","ssj9596",2,"","",0,"2026-02-01","2026-02-17"],[21,2653,"[BUG] Wrong order of biases inside of `ParallelScalingBlock`","ghwatson",0,"rwightman","247c545",1,"2026-01-20","2026-01-21"],[21,2644,"[BUG] RotaryEmbedding is not imported in timm.layers.attention_pool2d anymore","HYLcool",0,"rwightman","030985f",1,"2026-01-06","2026-01-06"],[21,2628,"[FEATURE] 3D model support for volumetric / medical imaging in timm","CowboyH",4,"","",0,"2025-12-10",""],[21,2623,"[FEATURE] Add DeepSpeed / ZeRO support in training script","Tianyi-Franklin-Wang",2,"","",0,"2025-12-05","2026-09-23"],[21,2622,"[FEATURE] Model Request... Add CSATv2 to timm model zoo","gusdlf93",2,"","",0,"2025-12-04","2025-12-12"],[21,2618,"Training code for resnet34.a1_in1k","vvhj3357816826-dotcom",2,"","",0,"2025-12-01","2025-12-01"],[21,2616,"[BUG] Torch compile slower + NaNs","conceptofmind",2,"","",0,"2025-11-25","2025-11-25"],[21,2613,"[BUG] ROCm 7.0 compatibility issue with Attention2d modules","EmilienM",2,"","",0,"2025-11-25","2025-12-02"],[21,2612,"[BUG] Slow forward for conv_dw and mlp in ConvNeXt","xiuqhou",2,"","",0,"2025-11-18","2025-11-19"],[21,2611,"[BUG] A very strange problem: running the exact same model and dataset on different devices and operating syst","csw92805",2,"","",0,"2025-11-18","2025-11-19"],[21,2609,"[Bug] FSDP FULL_SHARD incorrectly rejects timm models with features_only=True (FeatureListNet) due to overly-s","wenwwww",2,"","",0,"2025-11-13","2026-02-22"],[21,2607,"[FEATURE] [RFC] Support for interchangable attention backends","fffffgggg54",4,"","",0,"2025-11-08",""],[21,2604,"[FEATURE] Validation Frequency","t0278611",2,"","",0,"2025-11-03","2025-11-06"],[21,2600,"[BUG] A random number seed was set up, but the experimental results could not be reproduced.","Nathan33907",2,"","",0,"2025-10-21","2025-10-21"],[21,2597,"[FEATURE] Support for Sapiens","chenzhekl",2,"","",0,"2025-10-17","2026-09-09"],[21,2594,"[BUG] DINOv3 Model Export for CoreML fails","jbadger3",0,"rwightman","68bc434",1,"2025-10-14","2025-10-14"],[21,2593,"[BUG] AttributeError: 'RotaryEmbeddingDinoV3' object has no attribute 'update_feat_shape'","ananthu-aniraj",0,"rwightman","ef00378",1,"2025-10-03","2025-10-03"],[21,2590,"[FEATURE] Support for DINOv3 backbone in segmentation_models.pytorch (SMP)","Pinery-lee",2,"","",0,"2025-09-24","2025-09-24"],[21,2586,"[BUG] RuntimeError: Invalid pretrained tag (dinov3_lvd1689m) for convnext_base.","icesuki",2,"","",0,"2025-09-20","2025-09-21"],[21,2584,"The Self Attention forward part using 5D tesnor can be used in 4D tensor","deeperlearner",2,"","",0,"2025-09-19","2025-09-22"],[21,2581,"[FEATURE] Consistent License Handling","alexanderdann",2,"","",0,"2025-09-16","2025-10-14"],[21,2580,"[FEATURE] Add Support for Muon Optimizer","caojiaolong",0,"rwightman","94bdadb",1,"2025-09-13","2025-10-17"],[21,2577,"[BUG] Fix ResizeKeepRatio.repr","amirhfarzaneh",0,"rwightman","0515294",1,"2025-09-10","2025-09-21"],[21,2573,"[BUG] Identity op in onnx","stillbetter",2,"","",0,"2025-08-21","2025-08-21"],[21,2570,"[BUG] vit_mediumd_patch16_reg4_gap_256 doesn't use best pretrained weights in default config","TonyCongqianWang",2,"","",0,"2025-08-16","2025-08-16"],[21,2567,"[FEATURE] Add DINO v3 weights","seefun",2,"","",0,"2025-08-15","2025-09-17"],[21,2562,"[FEATURE] Add weights of naflexvit-siglip2","seefun",0,"rwightman","055a600",1,"2025-08-03","2025-08-04"],[21,2558,"[BUG] Hiera det (SAM 2) init_values are being ignored","hassonofer",1,"hassonofer","#2559",0,"2025-07-26","2025-07-26"],[21,2556,"[BUG] Different update_time after update","caojiaolong",0,"rwightman","55d4b5d",1,"2025-07-24","2025-07-24"],[21,2555,"[BUG] Use of `timm 1.0.18` with Python 3.9 results in TypeError","KellyWalker",0,"rwightman","d08d5a0",1,"2025-07-23","2025-07-24"],[21,2551,"[FEATURE] Support for DPT decoder","dqj5182",2,"","",0,"2025-07-16","2026-10-16"],[21,2550,"[FEATURE] PE latest distilled models","Jacobew",0,"rwightman","4a67e13",1,"2025-07-16","2025-07-19"],[21,2549,"[BUG] Eva model crashes on applying RotaryEmbeddingCat when patch drop rate is set","surajpaib",0,"rwightman","6063f82",1,"2025-07-16","2025-07-19"],[21,2548,"[FEATURE] Faster TIMM by combined background H2D copies","alex77g2",2,"","",0,"2025-07-14","2025-07-20"],[21,2546,"[BUG] timm is unable to load the huggingface custom model","fangyynb",2,"","",0,"2025-07-11","2025-07-11"],[21,2542,"[BUG]  naflexvit_so400m_patch16_siglip has undocumented different default pos_embed_interp_mode of \"bicubic\" i","redhottensors",0,"rwightman","a23518b",1,"2025-07-09","2026-09-06"],[21,2540,"[FEATURE] Unified `embed` Interfaces for Vision Transformer Models for MIM Pretrain Research","ryan-minato",4,"","",0,"2025-07-09",""],[21,2525,"[BUG] Fasternet model family not available?","brendanartley",2,"","",0,"2025-06-24","2025-06-24"],[21,2524,"[BUG] Library update on PyPi","stevenmoss-menditai",2,"","",0,"2025-06-24","2025-06-24"],[21,2522,"a problem about pre-trained weights.","chenyanting1",2,"","",0,"2025-06-18","2025-07-02"],[21,2521,"[BUG] Huggingface Rate Limit","NusretOzates",2,"","",0,"2025-06-16","2025-06-16"],[21,2514,"Naflex performance","stas-sl",2,"","",0,"2025-06-12","2025-06-20"],[21,2513,"[BUG] Multi GPU validate or inference, REPORT cuda out of memory","AiteYx",0,"rwightman","013851b",1,"2025-06-12","2025-06-12"],[21,2512,"offline mode","sipie800",2,"","",0,"2025-06-12","2025-06-12"],[21,2506,"Will you add Precision, Recall, and F1 score to metrics","chenyanting1",0,"ha405","#2568",0,"2025-06-05","2025-09-04"]]}