7 Releases published by 2 people

• v4.53.2-modernbert-decoder-preview ModernBERT Decoder (based on v4.53.2)

  published Jul 16, 2025

• v4.53.3 Patch release v4.53.3

  published Jul 22, 2025

• v4.53.2-Ernie-4.5-preview Ernie-4.5 and Ernie-4.5 MoE (based on v4.53.2)

  published Jul 23, 2025

• v4.54.0 v4.54.0: Kernels, Transformers Serve, Ernie, Voxtral, LFM2, DeepSeek v2, ModernBERT Decoder...

  published Jul 25, 2025

• 4.54.1 Patch release 4.54.1

  published Jul 29, 2025

• v4.55.0 v4.55.0: New openai GPT OSS model!

  published Aug 5, 2025

• 4.55.0-GLM-4.5V-preview GLM-4.5V preview based on 4.55.0

  published Aug 11, 2025

364 Pull requests merged by 145 people

• feat: extract rev in attn_implementation kernels via @

  #40009 merged Aug 11, 2025

• [GPT Big Code] Fix attention scaling

  #40041 merged Aug 11, 2025

• chore: standardize DeBERTa model card

  #37409 merged Aug 11, 2025

• Fix time_spent in notification_service.py.

  #40081 merged Aug 11, 2025

• added Textnet fast image processor

  #39884 merged Aug 11, 2025

• Fix repo consistency

  #40077 merged Aug 11, 2025

• guard on model.eval when using torch.compile + FSDP2

  #37413 merged Aug 11, 2025

• Remove deprecated cache-related objects

  #40035 merged Aug 11, 2025

• fix: move super().__init__ after vision_config init in Mistral3Config

  #40063 merged Aug 11, 2025

• [gemma3] update conversion key mapping

  #39778 merged Aug 11, 2025

• [qwen-vl] fix beam search with videos

  #39726 merged Aug 11, 2025

• fix: resolve triton version check compatibility on windows

  #39986 merged Aug 11, 2025

• unpin torchcodec==0.5.0 and use torch 2.8 on daily CI

  #40072 merged Aug 10, 2025

• Update HuBERT model card according to template

  #39742 merged Aug 10, 2025

• Revert "fix notification_service.py about time_spent"

  #40044 merged Aug 8, 2025

• GLM-4.5V Model Support

  #39805 merged Aug 8, 2025

• fix notification_service.py about time_spent

  #40037 merged Aug 8, 2025

• Bnb failling tests

  #40026 merged Aug 8, 2025

• Tie weights recursively on all submodels

  #39996 merged Aug 8, 2025

• [core] Refactor the Cache logic to make it simpler and more general

  #39797 merged Aug 8, 2025

• Fix missing None default values for Gemma3n model in get_placeholder_mask (#39991)

  #40024 merged Aug 8, 2025

• Harmonize past_key_value to past_key_valueS everywhere

  #39956 merged Aug 8, 2025

• Fix an annoying flaky test

  #40000 merged Aug 8, 2025

• Higgs modules_to_not_convert standardization

  #39989 merged Aug 8, 2025

• Fix broken image inference for Fuyu model

  #39915 merged Aug 8, 2025

• pin torchcodec==0.5.0 for now with torch 2.7.1 on daily CI

  #40013 merged Aug 7, 2025

• Update expected output values after #39885 (part 2)

  #40015 merged Aug 7, 2025

• Raising error when quantizing a quantized model

  #39998 merged Aug 7, 2025

• docs: fix duplication in 'en/optimizers.md'

  #40014 merged Aug 7, 2025

• unpin torch<2.8 on circleci

  #40012 merged Aug 7, 2025

• FA2 can continue generation from cache

  #39843 merged Aug 7, 2025

• Fix default values of getenv

  #39867 merged Aug 7, 2025

• Fix HGNetV2 Model Card and Image Classification Pipeline Usage Tips

  #39965 merged Aug 7, 2025

• fix: remove CHAT_TEMPLATE import in tests for deepseek-vl

  #40003 merged Aug 7, 2025

• Fix missing video inputs for PerceptionLM.

  #39971 merged Aug 7, 2025

• Fix int4 quantized model cannot work with cpu

  #39724 merged Aug 7, 2025

• Update expected output values after #39885 (part 1)

  #39990 merged Aug 7, 2025

• Fix consistency

  #39995 merged Aug 7, 2025

• Fix return typehint for decoder and annotate inv_freq

  #39610 merged Aug 7, 2025

• Bump transformers from 4.48.0 to 4.53.0 in /examples/tensorflow/language-modeling-tpu

  #39967 merged Aug 7, 2025

• Fix gemma3n feature extractor's incorrect squeeze

  #39919 merged Aug 7, 2025

• [Idefics] fix device mismatch

  #39981 merged Aug 7, 2025

• Various test fixes for AMD

  #39978 merged Aug 7, 2025

• Support input_embeds in torch exportable decoders

  #39836 merged Aug 7, 2025

• [superglue] Fixed the way batch mask was applied to the scores before match assignment computation

  #39968 merged Aug 7, 2025

• Gemma3 fixes

  #39960 merged Aug 7, 2025

• Modular fix: remove the model name in find_file_type

  #39897 merged Aug 6, 2025

• chore: update Deformable_Detr model card

  #39902 merged Aug 6, 2025

• [bugfix] fix flash_attention_2 unavailable error on Ascend NPU

  #39844 merged Aug 6, 2025

• Fix fix_and_overwrite mode of utils/check_docstring.py

  #39369 merged Aug 6, 2025

• remove triton_kernels dep with kernels instead

  #39926 merged Aug 6, 2025

• fix glm4v image process

  #39964 merged Aug 6, 2025

• fix typo

  #39936 merged Aug 6, 2025

• Fix grammatical error in MoE variable name: expert_hitted → expert_hit, hitted_experts → hit_experts

  #39959 merged Aug 6, 2025

• docs: fix typo in 'quantization-aware training'

  #39904 merged Aug 6, 2025

• Enable gpt-oss mxfp4 on older hardware (sm75+)

  #39940 merged Aug 6, 2025

• Fix MXFP4 quantizer validation to allow CPU inference with dequantize option

  #39953 merged Aug 6, 2025

• [docs] ko toc fix

  #39927 merged Aug 6, 2025

• circleci: pin torch 2.7.1 until torchcodec is updated

  #39951 merged Aug 6, 2025

• Fix CI: Tests failing on CPU due to torch.device('cpu').index being None

  #39933 merged Aug 6, 2025

• Avoid utils/check_bad_commit.py failing due to rate limit (requesting api.github.com)

  #39918 merged Aug 5, 2025

• [CI] post-GptOss fixes for green CI

  #39929 merged Aug 5, 2025

• gpt_oss last chat template changes

  #39925 merged Aug 5, 2025

• Add GPT OSS model from OpenAI

  #39923 merged Aug 5, 2025

• 🌐 [i18n-KO] Translated cache_explanation.md to Korean

  #39535 merged Aug 5, 2025

• Export SmolvLM

  #39614 merged Aug 5, 2025

• Update object_detection.md

  #39909 merged Aug 5, 2025

• run model debugging with forward arg

  #39905 merged Aug 5, 2025

• Revert "remove dtensors, not explicit (#39840)"

  #39912 merged Aug 5, 2025

• Fix aria tests

  #39879 merged Aug 5, 2025

• Fix eval thread fork bomb

  #39717 merged Aug 5, 2025

• Replace video_fps with fps in tests

  #39898 merged Aug 5, 2025

• Fix misleading WandB error when WANDB_DISABLED is set

  #39891 merged Aug 5, 2025

• Avoid aliasing in cond's branches for torch 2.8

  #39488 merged Aug 5, 2025

• Remove unnecessary CUDA sync in qwen2_5_vl

  #39870 merged Aug 5, 2025

• fix test_working_of_tp failure of accelerate ut

  #39828 merged Aug 5, 2025

• [Exaone4] Fixes the attn implementation!

  #39906 merged Aug 5, 2025

• Reorder serving docs

  #39634 merged Aug 5, 2025

• chore: update DETR model card

  #39822 merged Aug 4, 2025

• Add support for ModernBertForMultipleChoice

  #39232 merged Aug 4, 2025

• send some feedback when manually building doc via comment

  #39889 merged Aug 4, 2025

• Update cohere2 vision test

  #39888 merged Aug 4, 2025

• [DOCS] : Improved mimi model card

  #39824 merged Aug 4, 2025

• Fix link to models in README

  #39880 merged Aug 4, 2025

• Better return type hint for AutoModelForCausalLM and AutoModelForImageTextToText

  #39881 merged Aug 4, 2025

• Set torch.backends.cudnn.allow_tf32 = False for CI

  #39885 merged Aug 4, 2025

• Replace Tokenizer with PreTrainedTokenizerFast in ContinuousBatchProcessor

  #39858 merged Aug 4, 2025

• Rework add-new-model-like with modular and make test filenames coherent

  #39612 merged Aug 4, 2025

• Fix quant docker for fp-quant

  #39641 merged Aug 4, 2025

• Fix attn_implementation setter for models with backbone_config

  #39855 merged Aug 4, 2025

• Add support for including in-memory videos (not just files/urls) in apply_chat_template

  #39494 merged Aug 4, 2025

• Use comment to build doc on PRs

  #39846 merged Aug 4, 2025

• Refactor label name handling for PEFT models in Trainer class

  #39265 merged Aug 4, 2025

• Improve is_wandb_available function to verify WandB installation

  #39875 merged Aug 4, 2025

• remove dtensors, not explicit

  #39840 merged Aug 1, 2025

• Allow TrackioCallback to work when pynvml is not installed

  #39851 merged Aug 1, 2025

• fix: deprecate plot_keypoint_matching and make visualize_keypoint_matching for all Keypoint Matching models

  #39830 merged Aug 1, 2025

• Add fast image processor Janus, Deepseek VL, Deepseek VL hybrid

  #39739 merged Aug 1, 2025

• Fix responses add tests

  #39848 merged Aug 1, 2025

• Update ux cb

  #39845 merged Aug 1, 2025

• [WIP] Add MM Grounding DINO

  #37925 merged Aug 1, 2025

• Export private symbols

  #39729 merged Aug 1, 2025

• [attn_implementation] remove recursive, allows custom kernels with wrappers

  #39823 merged Aug 1, 2025

• [VLMs] split out "get placeholder mask" to helper

  #39777 merged Aug 1, 2025

• Fix tp cb

  #39838 merged Aug 1, 2025

• Fix bad markdown links

  #39819 merged Jul 31, 2025

• Fix broken links

  #39809 merged Jul 31, 2025

• [cohere2 vision] move doc to multimodal section

  #39820 merged Jul 31, 2025

• Update documentation for Cohere2Vision models

  #39817 merged Jul 31, 2025

• [Model] Cohere2 Vision

  #39810 merged Jul 31, 2025

• [docs] fix korean docs yet again

  #39813 merged Jul 31, 2025

• feat(tokenization): add encode_message to tokenize messages one by one

  #39507 merged Jul 31, 2025

• fix: providing a tensor to cache_position in model.generate kwargs always crashes because of boolean test

  #39300 merged Jul 30, 2025

• Add callback to monitor progress in whisper transcription

  #37483 merged Jul 30, 2025

• Update mT5 model card

  #39702 merged Jul 30, 2025

• chore: update cohere2 (Command R7B) model card

  #39604 merged Jul 30, 2025

• standardized BARThez model card

  #39701 merged Jul 30, 2025

• Fix re-compilations for cross attention cache

  #39788 merged Jul 30, 2025

• Simplify conditional code

  #39781 merged Jul 30, 2025

• Fix an invalid condition

  #39762 merged Jul 30, 2025

• fix chameleonvision UT failure

  #39646 merged Jul 30, 2025

• Super tiny update

  #39727 merged Jul 30, 2025

• more info in model_results.json

  #39783 merged Jul 30, 2025

• [ASR pipline] fix with datasets 4.0

  #39504 merged Jul 30, 2025

• enable static cache on vision encoder decoder

  #39773 merged Jul 30, 2025

• Fix Evolla and xLSTM tests

  #39769 merged Jul 30, 2025

• Don't set run_name when none

  #39695 merged Jul 30, 2025

• Standardize CLAP model card format

  #39738 merged Jul 29, 2025

• docs: Update EfficientLoFTR documentation

  #39620 merged Jul 29, 2025

• Fix OmDet test after arg deprecation

  #39766 merged Jul 29, 2025

• Remove python3.7 reference from doc link

  #39706 merged Jul 29, 2025

• [docs] Ko doc fixes after toc update

  #39660 merged Jul 29, 2025

• Fix Cache.max_cache_len max value for Hybrid models

  #39737 merged Jul 29, 2025

• fix(trainer): Correct loss scaling for incomplete gradient accumulation steps

  #39659 merged Jul 29, 2025

• 🌐 [i18n-KO] Translated how_to_hack_models.md to Korean

  #39536 merged Jul 29, 2025

• 🌐 [i18n-KO] Translated perf_train_gpu_one.md to Korean

  #39552 merged Jul 29, 2025

• 🌐 [i18n-KO] Translated pipeline_gradio.md to Korean

  #39520 merged Jul 29, 2025

• 🌐 [i18n-KO] Translated tokenizer.md to Korean

  #39532 merged Jul 29, 2025

• 🌐 [i18n-KO] Translated tvp.md to Korean

  #39578 merged Jul 29, 2025

• 🌐 [i18n-KO] Translated albert.md to Korean

  #39524 merged Jul 29, 2025

• 🌐 [i18n-KO] Translated main_classes/peft.md

  #39515 merged Jul 29, 2025

• [modenbert] fix regression

  #39750 merged Jul 29, 2025

• add libcst to extras["testing"] in setup.py

  #39761 merged Jul 29, 2025

• Fix version issue in modeling_utils.py

  #39759 merged Jul 29, 2025

• Enable xpu allocator on caching_allocator_warmup

  #39654 merged Jul 29, 2025

• Support loading Qwen3 MoE GGUF

  #39638 merged Jul 29, 2025

• Fix GPT2 with cross attention

  #39754 merged Jul 29, 2025

• Avoid OOM when other tests are failing

  #39758 merged Jul 29, 2025

• AMD disable torchcodec

  #39757 merged Jul 29, 2025

• Use --gpus all in workflow files

  #39752 merged Jul 29, 2025

• Apply several ruff SIM rules

  #37283 merged Jul 29, 2025

• Fix mamba regression

  #39728 merged Jul 29, 2025

• Update IMPORTANT_MODELS list

  #39734 merged Jul 29, 2025

• update GemmaIntegrationTest::test_model_2b_bf16_dola again

  #39731 merged Jul 29, 2025

• Fix: add back base model plan

  #39733 merged Jul 29, 2025

• [Fix] import two missing typos in models/__init__.py for typo checking

  #39745 merged Jul 29, 2025

• fix cache inheritance

  #39748 merged Jul 29, 2025

• extend more trainer test cases to XPU, all pass

  #39652 merged Jul 29, 2025

• BLIPs clean-up

  #35560 merged Jul 29, 2025

• Add Fast Segformer Processor

  #37024 merged Jul 28, 2025

• Superpoint fast image processor

  #37804 merged Jul 28, 2025

• Fix AMD dockerfile for audio models

  #39669 merged Jul 28, 2025

• Fix cache-related tests

  #39676 merged Jul 28, 2025

• Fix Layer device placement in Caches

  #39732 merged Jul 28, 2025

• Fix Qwen2AudioForConditionalGeneration.forward() and test_flash_attn_kernels_inference_equivalence

  #39503 merged Jul 28, 2025

• skip Glm4MoeModelTest::test_torch_compile_for_training

  #39670 merged Jul 28, 2025

• Update QAPipelineTests::test_large_model_course after #39193

  #39666 merged Jul 28, 2025

• mllama outputs refactor

  #39643 merged Jul 28, 2025

• Remove all expired deprecation cycles

  #39725 merged Jul 28, 2025

• [CI] Add Eric to comment slow ci

  #39601 merged Jul 28, 2025

• PATCH: add back n-dim device-mesh + fix tp trainer saving

  #39693 merged Jul 28, 2025

• Add self-hosted runner scale set workflow for mi325 CI

  #39651 merged Jul 28, 2025

• [configuration] remove redundant classmethod

  #38812 merged Jul 28, 2025

• update ernie model card

  #39657 merged Jul 28, 2025

• [processors] add tests for helper fn

  #39629 merged Jul 28, 2025

• xpu optimization for generation case

  #39573 merged Jul 28, 2025

• fix(tokenization): check token.content for trie

  #39587 merged Jul 28, 2025

• Fix missing initialization of FastSpeech2Conformer

  #39689 merged Jul 28, 2025

• fix missing model._tp_size from ep refactor

  #39688 merged Jul 26, 2025

• More robust tied weight test

  #39681 merged Jul 25, 2025

• Add padding-free to Granite hybrid moe models

  #39677 merged Jul 25, 2025

• Fix tied weight test

  #39680 merged Jul 25, 2025

• fix break for ckpt without _tp_plan

  #39658 merged Jul 25, 2025

• Add EXAONE 4.0 model

  #39129 merged Jul 25, 2025

• Support typing.Literal as type of tool parameters or return value

  #39633 merged Jul 25, 2025

• Add ep

  #39501 merged Jul 25, 2025

• bad_words_ids no longer slow on mps

  #39556 merged Jul 25, 2025

• Add xlstm model

  #39665 merged Jul 25, 2025

• Use auto_docstring for perception_lm fast image processor

  #39679 merged Jul 25, 2025

• fix: HWIO to OIHW

  #39200 merged Jul 25, 2025

• Fix auto_docstring crashing when dependencies are missing

  #39564 merged Jul 25, 2025

• Add support for DeepseekAI's DeepseekVL

  #36248 merged Jul 25, 2025

• Add missing flag for CacheLayer

  #39678 merged Jul 25, 2025

• Add evolla rebase main

  #36232 merged Jul 25, 2025

• update expected outputs for whisper after #38778

  #39304 merged Jul 25, 2025

• fix kyutai tests

  #39416 merged Jul 25, 2025

• Fixes the BC

  #39636 merged Jul 25, 2025

• Delete bad rebasing functions

  #39672 merged Jul 25, 2025

• [Ernie 4.5] Post merge adaptations

  #39664 merged Jul 25, 2025

• [CI] revert device in test_export_static_cache

  #39662 merged Jul 25, 2025

• Fix ModernBERT Decoder model

  #39671 merged Jul 25, 2025

• 🚨[Fast Image Processor] Force Fast Image Processor for Qwen2_VL/2_5_VL + Refactor

  #39591 merged Jul 25, 2025

• Rename huggingface_cli to hf

  #39630 merged Jul 25, 2025

• fix(voxtral): correct typo in apply_transcription_request

  #39572 merged Jul 25, 2025

• make fixup

  #39661 merged Jul 25, 2025

• [docs] fix ko cache docs

  #39644 merged Jul 25, 2025

• Make pytorch examples UV-compatible

  #39635 merged Jul 25, 2025

• revert change to cu_seqlen_k and max_k when preparing from position_ids

  #39653 merged Jul 25, 2025

• Fix: explicit not none check for tensors in flash attention

  #39639 merged Jul 25, 2025

• [attention] fix test for packed padfree masking

  #39582 merged Jul 25, 2025

• Add owlv2 fast processor

  #39041 merged Jul 25, 2025

• revert behavior of _prepare_from_posids

  #39622 merged Jul 24, 2025

• [Voxtral] values for A10 runners

  #39605 merged Jul 24, 2025

• [timm] new timm pin

  #39640 merged Jul 24, 2025

• Fix EfficientLoFTR model id in tests

  #39621 merged Jul 24, 2025

• Update recent processors for vLLM backend

  #39583 merged Jul 24, 2025

• [Docs] Translate audio_classification.md from English to Spanish

  #39513 merged Jul 23, 2025

• standardized YOLOS model card according to template in #36979

  #39528 merged Jul 23, 2025

• Feature/standardize opt model card

  #39568 merged Jul 23, 2025

• 🔴 Fix EnCodec internals and integration tests

  #39431 merged Jul 23, 2025

• Fix DAC integration tests and checkpoint conversion.

  #39313 merged Jul 23, 2025

• Move openai import

  #39613 merged Jul 23, 2025

• Transformers serve VLM

  #39454 merged Jul 23, 2025

• Fix important models CI

  #39576 merged Jul 23, 2025

• Fix typos and grammar issues in documentation and code

  #39598 merged Jul 23, 2025

• Allow device_mesh have multiple dim

  #38949 merged Jul 23, 2025

• enable triton backend on awq xpu

  #39443 merged Jul 23, 2025

• [idefics3] fix for vLLM

  #39470 merged Jul 23, 2025

• fix moe routing_weights

  #39581 merged Jul 23, 2025

• FP-Quant support

  #38696 merged Jul 23, 2025

• Rename supports_static_cache to can_compile_fullgraph

  #39505 merged Jul 23, 2025

• [Trackio] Allow single-gpu training and monitor power

  #39595 merged Jul 23, 2025

• Generic task-specific base classes

  #39584 merged Jul 23, 2025

• Fix DynamicCache and simplify Cache classes a bit

  #39590 merged Jul 23, 2025

• Mask2former & Maskformer Fast Image Processor

  #35685 merged Jul 23, 2025

• 🎯 Trackio integration

  #38814 merged Jul 22, 2025

• [WIP] Add OneformerFastImageProcessor

  #38343 merged Jul 22, 2025

• Fix link in "Inference server backends" doc

  #39589 merged Jul 22, 2025

• Torchdec RuntimeError catch

  #39580 merged Jul 22, 2025

• [Paged-Attention] Handle continuous batching for repetition penalty

  #39457 merged Jul 22, 2025

• updated mistral3 model card

  #39531 merged Jul 22, 2025

• Update docs/source/ko/_toctree.yml

  #39516 merged Jul 22, 2025

• [cache refactor] Move all the caching logic to a per-layer approach

  #39106 merged Jul 22, 2025

• General weight initialization scheme

  #39579 merged Jul 22, 2025

• Add AMD GPU expectations for LLaVA tests

  #39486 merged Jul 22, 2025

• Kernels flash attn

  #39474 merged Jul 22, 2025

• Add AMD expectations to Mistral3 tests

  #39481 merged Jul 22, 2025

• [docs] Create page on inference servers with transformers backend

  #39550 merged Jul 22, 2025

• [docs] update attention implementation and cache docs

  #39547 merged Jul 22, 2025

• Add AMD test expectations to DETR model

  #39539 merged Jul 22, 2025

• feat: add support for gradient checkpointing for TimmWrapperModel and TimmWrapperForImageClassification

  #39287 merged Jul 22, 2025

• Fixes needed for n-d parallelism and TP

  #39562 merged Jul 22, 2025

• Bump AMD container for 2.7.1 PyTorch

  #39458 merged Jul 22, 2025

• Add EfficientLoFTR model

  #36355 merged Jul 22, 2025

• [gemma3] fix bidirectional image mask

  #39396 merged Jul 22, 2025

• Update OLMoE model card

  #39344 merged Jul 21, 2025

• Update modernbertdecoder docs

  #39453 merged Jul 21, 2025

• [CI] Fix post merge ernie 4.5

  #39561 merged Jul 21, 2025

• [Fast image processors] Improve handling of image-like inputs other than images (segmentation_maps)

  #39489 merged Jul 21, 2025

• [Ernie 4.5] Add ernie text models

  #39228 merged Jul 21, 2025

• Refactor embedding input/output getter/setter

  #39339 merged Jul 21, 2025

• 🌐 [i18n-KO] Translated perf_infer_gpu_multi.md to Korean

  #39441 merged Jul 21, 2025

• [Fast image processor] refactor fast image processor glm4v

  #39490 merged Jul 21, 2025

• fix ndim check of device_mesh for TP

  #39538 merged Jul 21, 2025

• Refactor MambaCache to modeling_mamba.py

  #38086 merged Jul 21, 2025

• Fix Docstring of BarkProcessor

  #39546 merged Jul 21, 2025

• use the enable_gqa param in torch.nn.functional.scaled_dot_product_at…

  #39412 merged Jul 21, 2025

• Fix missing initializations for models created in 2023

  #39239 merged Jul 21, 2025

• Raise TypeError instead of ValueError for invalid types

  #38660 merged Jul 21, 2025

• Fix pylint warnings

  #39477 merged Jul 21, 2025

• Fix Qwen Omni integration test

  #39553 merged Jul 21, 2025

• 🚨🚨🚨 [Trainer] Enable average_tokens_across_devices by default in TrainingArguments

  #39395 merged Jul 21, 2025

• Rename _supports_flash_attn_2 in examples and tests

  #39471 merged Jul 21, 2025

• Fix the check in flex test

  #39548 merged Jul 21, 2025

• Fix bad tensor shape in failing Hubert test.

  #39502 merged Jul 21, 2025

• GLM-4 Update

  #39393 merged Jul 21, 2025

• [qwen2 vl] fix packing with all attentions

  #39447 merged Jul 21, 2025

• [gemma3] support sequence classification task

  #39465 merged Jul 21, 2025

• Fix placeholders replacement logic in auto_docstring

  #39433 merged Jul 18, 2025

• Update SAM/SAM HQ attention implementation + fix Cuda sync issues

  #39386 merged Jul 18, 2025

• Improve @auto_docstring doc and rename args_doc.py to auto_docstring.py

  #39439 merged Jul 18, 2025

• Add fast image processor SAM

  #39385 merged Jul 18, 2025

• Fix BatchEncoding.to() for nested elements

  #38985 merged Jul 18, 2025

• [gemma3] Fix do_convert_rgb in image processors.

  #39438 merged Jul 18, 2025

• [chat template] return assistant mask in processors

  #38545 merged Jul 18, 2025

• [dependencies] Update datasets pin

  #39500 merged Jul 18, 2025

• Slack CI bot: set default result for non-existing artifacts

  #39499 merged Jul 18, 2025

• 🚨🚨 Fix and simplify attention implementation dispatch and subconfigs handling

  #39423 merged Jul 18, 2025

• [doc builder job] temporary pyarrow pin

  #39496 merged Jul 18, 2025

• Add voxtral

  #39429 merged Jul 18, 2025

• Fix typing order

  #39467 merged Jul 17, 2025

• Add unified logits_to_keep support to LLMClass

  #39472 merged Jul 17, 2025

• [serve] Add speech to text (/v1/audio/transcriptions)

  #39434 merged Jul 17, 2025

• Update integration_utils.py

  #39469 merged Jul 17, 2025

• fix: ImageTextToTextPipeline handles user-defined generation_config

  #39374 merged Jul 17, 2025

• Enable some ruff checks for performance and readability

  #39383 merged Jul 17, 2025

• Fix convert_and_export_with_cache failures for GPU models

  #38976 merged Jul 17, 2025

• Update GemmaIntegrationTest::test_model_2b_bf16_dola

  #39362 merged Jul 17, 2025

• fix a comment typo in utils.py

  #39459 merged Jul 17, 2025

• Use newer typing notation

  #38934 merged Jul 17, 2025

• Fix tests due to breaking change in accelerate

  #39451 merged Jul 17, 2025

• fix max_length calculating using cu_seq_lens

  #39341 merged Jul 17, 2025

• fix(pipelines): QA pipeline returns fewer than top_k results in batch mode

  #39193 merged Jul 17, 2025

• Corrections to PR #38642 and enhancements to Wav2Vec2Processor __call__ and pad docstrings

  #38822 merged Jul 16, 2025

• create ijepa modelcard (ref : PR #36979 ).

  #39354 merged Jul 16, 2025

• Improve grammar and clarity in perf_hardware.md

  #39428 merged Jul 16, 2025

• fix cached file error when repo type is dataset

  #36909 merged Jul 16, 2025

• Fix indentation bug in SmolVLM image processor causing KeyError

  #39452 merged Jul 16, 2025

• Updated Megatron conversion script for gpt2 checkpoints

  #38969 merged Jul 16, 2025

• [CI] Fix partially red CI

  #39448 merged Jul 16, 2025

• Fixes #39204: add fallback if get_base_model missing

  #39226 merged Jul 16, 2025

• make the loss context manager easier to extend

  #39321 merged Jul 16, 2025

• Remove something that should have never been there

  #38254 merged Jul 16, 2025

• Fix processor tests

  #39450 merged Jul 16, 2025

• [Bugfix] [Quantization] Remove unused init arg

  #39324 merged Jul 16, 2025

• Better typing for model.config

  #39132 merged Jul 16, 2025

• Fix typo in generation configuration for Janus model weight conversion

  #39432 merged Jul 16, 2025

• Responses API in transformers serve

  #39155 merged Jul 16, 2025

• [cache] make all classes cache compatible finally

  #38635 merged Jul 16, 2025

• docs: add missing numpy import to minimal example

  #39444 merged Jul 16, 2025

• Remove runtime conditions for type checking

  #37340 merged Jul 16, 2025

• Add StableAdamW Optimizer

  #39446 merged Jul 16, 2025

• add test scanner

  #39419 merged Jul 16, 2025

• Fix missing definition of diff_file_url in notification service

  #39445 merged Jul 16, 2025

• Add cosine_with_min_lr_schedule_with_warmup_lr_rate scheduler in Trainer

  #31870 merged Jul 16, 2025

• Change log level from warning to info for scheduled request logging in ContinuousBatchProcessor

  #39372 merged Jul 16, 2025

• Defaults to adamw_torch_fused for Pytorch>=2.8

  #37358 merged Jul 16, 2025

• Fix L270 - hasattr("moe_args") returning False error

  #38715 merged Jul 16, 2025

• [chat template] add a testcase for kwargs

  #39415 merged Jul 16, 2025

• Fixed a bug calculating cross entropy loss in JetMoeForCausalLM

  #37830 merged Jul 16, 2025

• Remove double soft-max in load-balancing loss. Fixes #39055 .

  #39056 merged Jul 16, 2025

• [Core] [Offloading] Fix saving offloaded submodules

  #39280 merged Jul 16, 2025

• [autodocstring] add video and audio inputs

  #39420 merged Jul 16, 2025

• Responses API (to be merged into #39155)

  #39338 merged Jul 16, 2025

• CI workflow for performed test regressions

  #39198 merged Jul 16, 2025

• docs: update LightGlue docs

  #39407 merged Jul 15, 2025

• docs: update SuperGlue docs

  #39406 merged Jul 15, 2025

• [vlm] fix loading of retrieval VLMs

  #39242 merged Jul 15, 2025

• handle training summary when creating modelcard but offline mode is set

  #37095 merged Jul 15, 2025

• Remove residual quantization attribute from dequantized models

  #39373 merged Jul 15, 2025

• Remove deprecated audio utils functions

  #39330 merged Jul 15, 2025

• Fix bugs in pytorch example run_clm when streaming is enabled

  #39286 merged Jul 15, 2025

• Fix bugs from pipeline preprocessor overhaul

  #39425 merged Jul 15, 2025

• refactor: remove set_tracer_provider and set_meter_provider calls

  #39422 merged Jul 15, 2025

• Fix invalid property

  #39384 merged Jul 15, 2025

• set document_question_answering pipeline _load_tokenizer to True

  #39411 merged Jul 15, 2025

• Ignore extra position embeddings weights for ESM

  #39063 merged Jul 15, 2025

• support loading qwen3 gguf

  #38645 merged Jul 15, 2025

• Add ModernBERT Decoder Models - ModernBERT, but trained with CLM!

  #38967 merged Jul 15, 2025

• Fix typo in /v1/models output payload

  #39414 merged Jul 15, 2025

• [refactor] set attention implementation

  #38974 merged Jul 15, 2025

• Fix/siglip2 pooling comment

  #39378 merged Jul 14, 2025

• Update phi4_multimodal.md

  #38830 merged Jul 14, 2025

• [Docs] Fix typo in CustomTrainer compute_loss method and adjust loss reduction logic

  #39391 merged Jul 14, 2025

• Use np.pad instead of np.lib.pad.

  #39346 merged Jul 14, 2025

• 🚨 Totally rewrite how pipelines load preprocessors

  #38947 merged Jul 14, 2025

• Remove do_reduce_labels Argument from model initialization in run_semantic_segmentation_no_trainer

  #39322 merged Jul 14, 2025

• Fix Lfm2 and common tests

  #39398 merged Jul 14, 2025

• Deprecate AutoModelForVision2Seq

  #38900 merged Jul 14, 2025

• [Qwen2.5-VL] Fix torch.finfo() TypeError for integer attention_mask_tensor

  #39333 merged Jul 14, 2025

• [BLIP] remove cache from Qformer

  #39335 merged Jul 14, 2025

• [shieldgemma] fix checkpoint loading

  #39348 merged Jul 14, 2025

• Fix overriding Fast Image/Video Processors instance attributes affect other instances

  #39363 merged Jul 12, 2025

• update docker file to use latest timm (for perception_lm)

  #39380 merged Jul 12, 2025

190 Pull requests opened by 128 people

• Add Apertus

  #39381 opened Jul 12, 2025

• Fix: Docker Build Vulnerable to Malicious Package Installation Attack in docker/custom-tokenizers.dockerfile

  #39394 opened Jul 14, 2025

• No repeat kv

  #39402 opened Jul 14, 2025

• Add Vocos model

  #39403 opened Jul 14, 2025

• Add a unit test for BartModel to compare eager, sdpa on one particular set of inputs

  #39435 opened Jul 15, 2025

• Fix logger warnings in Gemma model test files

  #39449 opened Jul 16, 2025

• Add eurobert

  #39455 opened Jul 16, 2025

• Fix quantized model initialization for int8 dtypes

  #39456 opened Jul 16, 2025

• Skipping `initialize_weights` when model is quantized

  #39464 opened Jul 17, 2025

• README: Update Bert Japanese model card

  #39466 opened Jul 17, 2025

• Fix quantized model dispatch with device_map='auto'

  #39468 opened Jul 17, 2025

• Fix Bark failing tests

  #39478 opened Jul 17, 2025

• Add model arcinstitute state

  #39480 opened Jul 17, 2025

• Bye bye env vars, keep everything as configs

  #39483 opened Jul 17, 2025

• Add Whole Word Masking and Padding Strategy to DataCollatorForLanguageModeling

  #39485 opened Jul 17, 2025

• Update CTRL model card with improved usage examples and documentation notes

  #39487 opened Jul 17, 2025

• Fix: Skip weight initialization for quantized int8 models

  #39491 opened Jul 17, 2025

• [Voxtral] nit + pin correct mistral common version

  #39493 opened Jul 18, 2025

• Make sure Moshi is exportable with static cache

  #39506 opened Jul 18, 2025

• [WIP] :broom: :broom: :broom: Get set decoder cleanup

  #39509 opened Jul 18, 2025

• 🌐 [i18n-KO] Translated `compressed_tensor.md` to Korean

  #39517 opened Jul 19, 2025

• 🌐 [i18n-KO] Translated `models.md` to Korean

  #39518 opened Jul 19, 2025

• 🌐 [i18n-KO] Translated `main_classes/processors.md` to Korean

  #39519 opened Jul 19, 2025

• build: Add fast image processor tvp

  #39529 opened Jul 20, 2025

• Add Beit3 model

  #39534 opened Jul 20, 2025

• Add Muon optimizer implementation and integration

  #39541 opened Jul 20, 2025

• 🌐 [i18n-KO] Translated feature_extractors.md to Korea

  #39544 opened Jul 21, 2025

• [WIP] try to relax the tie_weights method

  #39555 opened Jul 21, 2025

• 🌐 [i18n-KO] Translated `imageprocessor.md` to Korean

  #39557 opened Jul 21, 2025

• 🌐 [i18n-KO] Translated `main_classes/deepspeed.md` to Korean

  #39559 opened Jul 21, 2025

• fix load_model_end = true work when save_steps < eval_steps

  #39560 opened Jul 21, 2025

• 🌐 [i18n-KO] Translated `vision-encoder-decoder.md` to Korean

  #39563 opened Jul 21, 2025

• 🌐 [i18n-KO] Translated `auto_docstring.md` to Korean

  #39571 opened Jul 22, 2025

• feat(autoformer): Improve ValueError for insufficient sequence length

  #39574 opened Jul 22, 2025

• 🌐 [i18n-KO] Translated `vitpose.md` to Korean

  #39575 opened Jul 22, 2025

• 🌐 [i18n-KO] Translated `pipelines.md` to Korean

  #39577 opened Jul 22, 2025

• [`Ernie 4.5`] Ernie VL models

  #39585 opened Jul 22, 2025

• WIP, reference modeling

  #39588 opened Jul 22, 2025

• Add Fast Image Processor for ImageGPT

  #39592 opened Jul 22, 2025

• 🌐 [i18n-KO] Translated 'xclip.md' to Korean

  #39594 opened Jul 22, 2025

• Fix: check TrainerState file exists before loading during resume

  #39599 opened Jul 23, 2025

• [video processors] decode only sampled videos -> less RAM and faster processing

  #39600 opened Jul 23, 2025

• feat: add `is_fast` to ImageProcessor

  #39603 opened Jul 23, 2025

• HunYuan opensource

  #39606 opened Jul 23, 2025

• Chat schemas

  #39609 opened Jul 23, 2025

• Fix FSDP v1 bug: trainer incorrectly uses an unwrapped model

  #39617 opened Jul 23, 2025

• fix tensor device when loading state dict

  #39623 opened Jul 24, 2025

• Fix: allow Union[str, dict, None] fields like deepspeed to be passed via CLI

  #39625 opened Jul 24, 2025

• [serve] Add speech-to-text

  #39631 opened Jul 24, 2025

• fix dead NVIDIA link

  #39632 opened Jul 24, 2025

• 🌐 [i18n-KO] Translated `deepseek_v3.md` to Korean

  #39649 opened Jul 24, 2025

• Fix loss scaling and token aggregation to use only data parallel group

  #39674 opened Jul 25, 2025

• [BugFix]: Support dict and config file path for deepspeed

  #39675 opened Jul 25, 2025

• Fix issue #39191 respect accelerate config to disable torch.dynamo compilation

  #39683 opened Jul 25, 2025

• Allow custom hf_quantizer in from_pretrained

  #39690 opened Jul 26, 2025

• fix misspelled issues

  #39691 opened Jul 26, 2025

• use untyped storage for dtensors due to deprecation

  #39697 opened Jul 26, 2025

• Fix exaone4 layer_types ZeroDivision/TypeError when sliding_window_pattern is None/"LLLG"

  #39698 opened Jul 26, 2025

• Fix Causality Handling in Flash Attention to Support Bidirectional Attention

  #39707 opened Jul 27, 2025

• 🌐[i18n-bn] Introduce Bengali version of Transformers documentation

  #39708 opened Jul 27, 2025

• 🌐 [i18n-KO] Translated `attention_interface.md` to Korean

  #39712 opened Jul 27, 2025

• 🌐 [i18n-KO] Translated `main_classes/optimizer_schedules.md` to Korean

  #39713 opened Jul 27, 2025

• 🌐 [i18n-KO] Translated `main_classes/backbones.md` to Korean

  #39714 opened Jul 27, 2025

• Fix SigLIP2 documentation model/processor mismatch

  #39718 opened Jul 28, 2025

• [Feat] Adding Intern-S1

  #39722 opened Jul 28, 2025

• handle multimodal models with tp_plan on the text_config

  #39735 opened Jul 28, 2025

• [Tests] [Bugfix] Make weights tied for `dynamic_tied_weights` test

  #39740 opened Jul 28, 2025

• Fix HfArgumentParser to filter out dict types from Union

  #39741 opened Jul 28, 2025

• Audio encodings now match conv2d weight dtype in Gemma3nAudioSSCPConvBlock

  #39743 opened Jul 29, 2025

• 🌐 [i18n-KO] Translated `text-to-speech.md` to Korean

  #39751 opened Jul 29, 2025

• Fix rope_deltas corruption in Qwen2.5VL during CFG generation

  #39756 opened Jul 29, 2025

• [Draft] Add Llasa TTS family of models

  #39760 opened Jul 29, 2025

• Improve Gemma3n model and tests

  #39764 opened Jul 29, 2025

• Stop using `from_legacy_cache` as Cache initialization

  #39765 opened Jul 29, 2025

• Benchmarking improvements

  #39768 opened Jul 29, 2025

• [Bugfix] Fix `AutoModel.from_pretrained(..., quantization_config=None)` regression

  #39770 opened Jul 29, 2025

• Fix missing initializations for models created in 2022

  #39772 opened Jul 30, 2025

• Use `dtype` instead of `torch_dtype` everywhere!

  #39782 opened Jul 30, 2025

• fix mllama integration tests

  #39785 opened Jul 30, 2025

• Fix pil dependency torch extra

  #39790 opened Jul 30, 2025

• Served models handle with nested content

  #39792 opened Jul 30, 2025

• Fix DAC conversion script

  #39793 opened Jul 30, 2025

• Fix ProphetNet forward to handle tuple encoder_outputs

  #39794 opened Jul 30, 2025

• [pipelines] text-to-audio pipeline standardization

  #39796 opened Jul 30, 2025

• Mistral: Add support for interleaved attention

  #39799 opened Jul 30, 2025

• [WIP] Add EdgeTAM

  #39800 opened Jul 30, 2025

• fix: qwen 25vl rope if item is masked

  #39802 opened Jul 30, 2025

• Enable SIM rules

  #39806 opened Jul 31, 2025

• 🌐 [i18n-KO] Translated `bamba.md` to Korean

  #39807 opened Jul 31, 2025

• 🌐 [i18n-KO] Translated `gpt2.md` to Korean

  #39808 opened Jul 31, 2025

• [chat template] update when "push_to_hub"

  #39815 opened Jul 31, 2025

• Refactor vit-like models

  #39816 opened Jul 31, 2025

• Support MetaCLIP 2

  #39821 opened Jul 31, 2025

• [serve] guard imports

  #39825 opened Jul 31, 2025

• Add MetaCLIP 2

  #39826 opened Jul 31, 2025

• [serve] allow array `content` inputs for LLMs

  #39829 opened Jul 31, 2025

• refactor(modeling_llama): make RotaryEmbedding default path explicit

  #39831 opened Jul 31, 2025

• add step3v in VLMS

  #39837 opened Aug 1, 2025

• [WIP] RoPE refactor

  #39847 opened Aug 1, 2025

• Fix DeepSpeed mixed precision precedence over Accelerate defaults

  #39856 opened Aug 1, 2025

• WIP: Initial support for bnb 4bit on any nn.Parameter

  #39859 opened Aug 1, 2025

• 🌐 [i18n-KO] Translated grounding-dino.md to Korean

  #39861 opened Aug 2, 2025

• Update model card for gpt neox japanese

  #39862 opened Aug 2, 2025

• 🌐 [i18n-KO] Translated `chat_extras.md` to Korean

  #39863 opened Aug 2, 2025

• 🌐 [i18n-KO] Translated `gemma3.md` to Korean

  #39865 opened Aug 2, 2025

• make sure model.save_pretrained has the correct is_main_process

  #39866 opened Aug 2, 2025

• Update README.md

  #39869 opened Aug 3, 2025

• fix: Catch correct ConnectionError for additional_chat_templates

  #39874 opened Aug 3, 2025

• FP-Quant NVFP4 and Python 3.9 support

  #39876 opened Aug 3, 2025

• Remove deprecated max_size parameter from ConditionalDetrImageProcessor

  #39883 opened Aug 4, 2025

• 🌐 [i18n-KO] Translated `perf_train_gaudi.md` to Korean

  #39886 opened Aug 4, 2025

• 🌐 [i18n-KO] Translated `jamba.md` to Korean

  #39890 opened Aug 4, 2025

• [docs] Add reference to HF-maintained `custom_generate` collections

  #39894 opened Aug 4, 2025

• Add Videoprism

  #39895 opened Aug 4, 2025

• [model] Support MiniCPM-V 4.0

  #39899 opened Aug 5, 2025

• 🌐 [i18n-KO] Translated `fp_quant` to Korean

  #39901 opened Aug 5, 2025

• 🌐 [i18n-KO] Translated clipseg.md to Korean

  #39903 opened Aug 5, 2025

• Update dynamic attnt setter for multimodals

  #39908 opened Aug 5, 2025

• 🌐 [i18n-KO] Translated `tiny_agents.md` to Korean

  #39913 opened Aug 5, 2025

• 🌐 [i18n-KO] Updated ko/perf_train_cpu.md

  #39917 opened Aug 5, 2025

• 🌐 [i18n-KO] Updated ko/perf_train_special.md

  #39920 opened Aug 5, 2025

• 🌐 [i18n-KO] Translated `attention_interface.md` to Korean

  #39922 opened Aug 5, 2025

• Add chat template tests

  #39924 opened Aug 5, 2025

• Fix hidden torchvision>=0.15 dependency issue

  #39928 opened Aug 5, 2025

• Add missing special token properties to MistralCommonTokenizer

  #39930 opened Aug 5, 2025

• Registers StaticCache serialization functions for torch.export.export

  #39931 opened Aug 5, 2025

• Fix whisper `return_language` with `return_timestamp=word`

  #39938 opened Aug 5, 2025

• fixing image_utils.py todo

  #39941 opened Aug 6, 2025

• fix llama issue

  #39942 opened Aug 6, 2025

• Add back `_tp_plan` attribute

  #39944 opened Aug 6, 2025

• Add pytest marker: `torch_compile_test` and `torch_export_test`

  #39950 opened Aug 6, 2025

• Use torch._check instead of a test to make the model Gemma3 exportable

  #39962 opened Aug 6, 2025

• Add Keypoint Matcher pipeline

  #39970 opened Aug 6, 2025

• Causal loss for `ForConditionalGeneration`

  #39973 opened Aug 7, 2025

• [bugfix] Fix tensor device in Idefics2, Idefics3, and SmolVLM

  #39975 opened Aug 7, 2025

• Fix Qwen3 MoE GGUF architecture mismatch

  #39976 opened Aug 7, 2025

• Fix cross-attention masking before residual connection

  #39979 opened Aug 7, 2025

• Fix setting attention for multimodal models

  #39984 opened Aug 7, 2025

• Add a VGGT(Visual Geometry Grounded Transformer) model compatible with huggingface transfromers

  #39987 opened Aug 7, 2025

• Update Glm4V processor and add tests

  #39988 opened Aug 7, 2025

• Default to dequantize if cpu in device_map for mxfp4

  #39993 opened Aug 7, 2025

• chore: Add type hints to import_utils.py module

  #39994 opened Aug 7, 2025

• make sure position_ids are passed in for causal mask creation for gpt-oss

  #39997 opened Aug 7, 2025

• allow TP to work in ND-parallel with fsdp cpu ram efficient loading

  #39999 opened Aug 7, 2025

• [`Flash Attention`] Fix flash attention integration

  #40002 opened Aug 7, 2025

• Fix PerceptionLM image preprocessing for non-tiled image input.

  #40006 opened Aug 7, 2025

• 🚨 Use lru_cache for sine pos embeddings MaskFormer

  #40007 opened Aug 7, 2025

• Fixes for EncoderDecoderCache

  #40008 opened Aug 7, 2025

• 🌐 [i18n-KO] Translated `optimizers.md` to Korean

  #40011 opened Aug 7, 2025

• [WIP] Fix naive for loops for MoE models resulting in sub 20% downstream MFU for training with trl, e.t.c (Qwen3, Deepseek V3, Ernie 4.5, GLM 4.5, Dots1)

  #40016 opened Aug 7, 2025

• Feat/add gpt oss sequence classification

  #40019 opened Aug 8, 2025

• [fix] batch inference for llava_onevision

  #40021 opened Aug 8, 2025

• fix: resolve dropout type error in DogeDecoder

  #40022 opened Aug 8, 2025

• Add support for SDPA for OWLViT and OWLv2

  #40023 opened Aug 8, 2025

• Add amd runners to run-slow command

  #40027 opened Aug 8, 2025

• Revert FA2 kwargs construction

  #40029 opened Aug 8, 2025

• Update boxes expectations for OWLViT test

  #40030 opened Aug 8, 2025

• Add model card for MobileViT

  #40033 opened Aug 8, 2025

• Fix error on importing unavailable torch.distributed

  #40038 opened Aug 8, 2025

• New DynamicSlidingWindowLayer & associated Cache

  #40039 opened Aug 8, 2025

• Add GptOssForSequenceClassification for GPT-OSS models

  #40043 opened Aug 8, 2025

• (small) fix conditional for input_ids and input_embeds in marian

  #40045 opened Aug 8, 2025

• Update wavlm.md to match new model card template

  #40047 opened Aug 8, 2025

• Standardize BARTpho model card: badges, new examples, fixed broken im…

  #40051 opened Aug 9, 2025

• Auto-log parallelism info to wandb.config using HF Accelerate

  #40055 opened Aug 9, 2025

• updated visualBERT modelcard

  #40057 opened Aug 9, 2025

• GGUF Qwen2VL

  #40058 opened Aug 9, 2025

• Fix Inefficient GELU implementation in GPT2

  #40059 opened Aug 9, 2025

• Avoid CUDA stream sync

  #40060 opened Aug 10, 2025

• 🌐 [i18n-KO] Translated `vitdet.md` to Korean

  #40061 opened Aug 10, 2025

• 🌐 [i18n-KO] Translated `videomae.md` to Korean

  #40064 opened Aug 10, 2025

• Delay float32 upcast in ForCausalLMLoss after filtering ignore_index

  #40065 opened Aug 10, 2025

• Change Qwen2RMSNorm to RMSNorm from PyTorch

  #40066 opened Aug 10, 2025

• Add missing arguments to class constructors

  #40068 opened Aug 10, 2025

• Remove _prepare_flash_attention_from_position_ids

  #40069 opened Aug 10, 2025

• initializing branch and draft PR

  #40074 opened Aug 11, 2025

• Skipping pytree registration in case fsdp is enabled

  #40075 opened Aug 11, 2025

• rm pytorch-triton dependency

  #40076 opened Aug 11, 2025

• Update notification service MI325

  #40078 opened Aug 11, 2025

• [WIP] Collated reports

  #40080 opened Aug 11, 2025

• Removes DoLa decoding strategy

  #40082 opened Aug 11, 2025

• Fix regression in mllama vision encoder

  #40083 opened Aug 11, 2025

• remove sequence parallel in llama4

  #40084 opened Aug 11, 2025

• `decoding_method` argument in generate

  #40085 opened Aug 11, 2025

• build unittest for `ViTImageProcessorFast`

  #40086 opened Aug 11, 2025

• DOCS: Add missing space in SECURITY.md

  #40087 opened Aug 11, 2025

• Fix RuntimeError when loading quantized models with int8 weights (#39366)

  #40090 opened Aug 11, 2025

• Replace `logger.warning` with `logger.warning_once` in `GradientCheckpointingLayer`

  #40091 opened Aug 12, 2025

• Optimize LlamaAttention by fusing QKV projections

  #40092 opened Aug 12, 2025

• fix(modeling_utils): correct initialization of missing and mismatched…

  #40093 opened Aug 12, 2025

180 Issues closed by 57 people

• Whisper v-3 pipeline requiring a lot of memory when setting return_timestamps="word"

  #27834 closed Aug 11, 2025

• Incorrect scaling of Gemma embeddings in float32 regime

  #38702 closed Aug 11, 2025

• 🐛 Bug Report: Accelerate config to disable torch dynamo is ignored by transformers automatic compilation

  #39191 closed Aug 11, 2025

• Inconsistant `input_feature` length and `attention_mask` length in `WhisperFeatureExtractor`

  #39214 closed Aug 11, 2025

• [Mistral3] attn_implementation not applied to vision_tower.config in Mistral3Config due to init order

  #40062 closed Aug 11, 2025

• Instantiating `google/gemma-3-4b-pt` with AutoModelForSequenceClassification Reports Unitialized Model

  #39763 closed Aug 11, 2025

• `num_beams` > 1 leads to exception for Qwen2.5VL (Qwen family or all VLM models?)

  #39723 closed Aug 11, 2025

• Triton version check compatibility on windows

  #39985 closed Aug 11, 2025

• apply_rotary_pos_emb_flashatt failed during triton jit compilation 'constexpr' object has no attribute 'bit_length'

  #39167 closed Aug 11, 2025

• Whisper `.generate()` function not respecting `max_new_tokens` or `max_length`

  #36183 closed Aug 10, 2025

• Gemma2 fall back to cpu execusion when attn_implementation='flash_attention_2'

  #39188 closed Aug 10, 2025

• Previous PRs introduced a bug on Accumulated Gradients Losses

  #40052 closed Aug 9, 2025

• Incorrect word timestamps and word repetitions with Whisper-Large-v3-turbo model

  #37248 closed Aug 9, 2025

• Pretrainedtokenizerfast Segmentation fault

  #39099 closed Aug 9, 2025

• New release 4.53.0 breaks HF trainer/model

  #39111 closed Aug 9, 2025

• Gradient accumulation steps for Vision Languge model

  #39123 closed Aug 9, 2025

• Not capable of exporting Mistral to ONNX format with the use of caching

  #39162 closed Aug 9, 2025

• Error when loading gguf file

  #40040 closed Aug 9, 2025

• Weights not tied when loading `from_pretrained` with a wrapped model

  #39900 closed Aug 8, 2025

• `TypeError: 'builtins.safe_open' object is not iterable` in `load_pytorch_state_dict_in_tf2_model `

  #40028 closed Aug 8, 2025

• Major issues with transformers version causing rubbish generations with Gemma3 family using vllm

  #40017 closed Aug 8, 2025

• AttributeError: 'BitsAndBytesConfig' object has no attribute 'get_loading_attributes' with transformers 4.55.0

  #39939 closed Aug 8, 2025

• Gemma3n get_placeholder_mask issue

  #39991 closed Aug 8, 2025

• flash-attn cannot perform deterministic computation

  #39982 closed Aug 8, 2025

• [DeepSeek-V3] Different rotary embedding implementation between DeepSeek-AI and Transformers

  #39687 closed Aug 8, 2025

• ModernBertUnpaddedRotaryEmbedding __init__ error

  #39934 closed Aug 7, 2025

• video_inputs are not passed to perception_lm

  #40004 closed Aug 7, 2025

• Flash Attention fails with non aligned position_ids

  #39814 closed Aug 7, 2025

• `convert_deepseek_vl_weights_to_hf.py` not included in v4.55.0 release.

  #39966 closed Aug 7, 2025

• [Gemma3N] Audio processing issue

  #39911 closed Aug 7, 2025

• v4.55.0 Idefics3 RuntimeError Tensors on different devices

  #39947 closed Aug 7, 2025

• [gpt‑oss] eager_attention_forward not using sliding-window attention for GPT‑OSS models

  #39954 closed Aug 7, 2025

• Finetune `gpt-oss-20b` with `mxfp4` quantization

  #39969 closed Aug 6, 2025

• Fix grammatically incorrect variable name "expert_hitted" → "expert_hit" in MoE implementation

  #39955 closed Aug 6, 2025

• transformers serve doesn't handle OPTIONS http method

  #39932 closed Aug 6, 2025

• 454545

  #39864 closed Aug 6, 2025

• ImportError: cannot import name 'GenerationMixin' from 'transformers.generation'

  #38442 closed Aug 6, 2025

• Streaming mode support on HF vs kyutai-labs for the mimi model

  #38535 closed Aug 6, 2025

• enable GraniteMoeHybridIntegrationTest in UT

  #38542 closed Aug 6, 2025

• Llama4 inference encounter unsupported op in dynamo ?

  #38118 closed Aug 6, 2025

• Misleading WandB error when WANDB_DISABLED=True and report_to="wandb" are both set

  #39878 closed Aug 5, 2025

• pytorch_utils.py > isin_mps_friendly > RuntimeError: Expected elements.dtype() == test_elements.dtype() to be true, but got false.

  #37423 closed Aug 5, 2025

• Inefficient memory resharding in attention layer

  #39072 closed Aug 5, 2025

• Inefficient default GELU implementation in GPT2

  #39073 closed Aug 5, 2025

• facebook/dinov2-with-registers-giant does not appear to have a file named pytorch_model.bin, model.safetensors, tf_model.h5, model.ckpt or flax_model.msgpack.

  #39075 closed Aug 5, 2025

• AttributeError: 'HfTrainerDeepSpeedConfig' object has no attribute 'is_zero3'

  #39081 closed Aug 5, 2025

• Why `lm-head` weight still exists with `"tie_word_embeddings": true`

  #39812 closed Aug 4, 2025

• Iwin Transformer: Hierarchical Vision Transformer using Interleaved Windows

  #39704 closed Aug 4, 2025

• ValueError: Max cache length is not consistent across layers

  #39877 closed Aug 4, 2025

• Allow video objects (np array etc.) in apply_chat_template (not just paths or urls)

  #36560 closed Aug 4, 2025

• Exception while inference Qwen2VL and Qwen2VL, assert module.weight.shape[1] == 1

  #38665 closed Aug 4, 2025

• model.generate custom encoder and decoder outputs/inputs

  #39871 closed Aug 3, 2025

• Vision Encoder-Decoder fails with LLaMA decoder due to missing cross-attention implementation

  #34674 closed Aug 3, 2025

• Only with newest version (4.52.4): from_pretrained() esm.embeddings.position_embeddings.weight missing

  #39038 closed Aug 3, 2025

• pytorch version 1.8.1 compatibility

  #39049 closed Aug 3, 2025

• TypeError: couldn't find storage object Float8_e4m3fnStorage - which version is needed for this?

  #39409 closed Aug 2, 2025

• 'Mistral3Model' object has no attribute 'prepare_inputs_for_generation'

  #39007 closed Aug 2, 2025

• Not able to use flash attention with torch.compile with model like BERT

  #39017 closed Aug 2, 2025

• Tool-Calling Model (ToolACE-2-Llama-3.1-8B) Responds with Irrelevant Tool message on General Question

  #39833 closed Aug 1, 2025

• Add MM Grounding DINO

  #37744 closed Aug 1, 2025

• Using Gemma3n with text-only generation requires image dependencies

  #39169 closed Aug 1, 2025

• Qwen2-VL err

  #39818 closed Jul 31, 2025

• Option to tokenize messages one after the other

  #39417 closed Jul 31, 2025

• [rank0]: ValueError: Your setup doesn't support bf16/gpu.

  #39716 closed Jul 31, 2025

• Blip model got performance regression on compile mode after refactor cache.

  #39774 closed Jul 30, 2025

• BioGPT Implementation Bug Report

  #39776 closed Jul 30, 2025

• tokenizer decode decode with timestamp fails for extended vocabulary

  #35330 closed Jul 30, 2025

• How to streaming output audio of Qwen2.5-omni-7b

  #37570 closed Jul 30, 2025

• Significant WER Increase with Whisper Chunking Compared to Long-Form Transcription

  #38347 closed Jul 30, 2025

• Transformers version causing my finetuned model to hallucinate

  #38378 closed Jul 30, 2025

• `load_balancing_loss_func` doesn't support 4D attention mask

  #38910 closed Jul 30, 2025

• Max cache length issue with Gemma 3

  #39711 closed Jul 29, 2025

• Loss is incorrectly scaled in Trainer during the last step with gradient accumulation when the final batch is smaller than accumulation steps.

  #38837 closed Jul 29, 2025

• ModernBERT has been totally destroyed by PR #38974 and #38838

  #39747 closed Jul 29, 2025

• Support loading Qwen3 MoE GGUF

  #39721 closed Jul 29, 2025

• [XPU] Model get OOM when loading models

  #39627 closed Jul 29, 2025

• encoder decoder model compile failed after refactor cache

  #39746 closed Jul 29, 2025

• _supports_static_cache disappear

  #39744 closed Jul 29, 2025

• [exaone4] ZeroDivisionError/TypeError when sliding_window_pattern is None/"LLLG" and _attn_implementation stays None (4.54.0 & main)

  #39696 closed Jul 29, 2025

• device mismatch error when using `SlidingWindowLayer`.

  #39730 closed Jul 28, 2025

• AddedToken should check content on `_update`

  #39586 closed Jul 28, 2025

• Checkpointing broken for classifier training multi-gpu

  #38925 closed Jul 28, 2025

• vlmm 0.10.0 load baidu/ERNIE-4.5-300B-A47B-Base-PT error

  #39719 closed Jul 28, 2025

• [i18n-<languageCode>] Translating docs to <عربي>

  #38381 closed Jul 27, 2025

• Not installable on arm64 due to jaxlib upper bound

  #36611 closed Jul 27, 2025

• KeyError in Llama-4-Maverick-17B-128E-Instruct-FP8 Inference with Offloading

  #38281 closed Jul 27, 2025

• ImportError: DLL load failed while importing _safetensors_rust: The specified module could not be found

  #38479 closed Jul 27, 2025

• Contribute to Transformers on windows natively without WSL

  #38601 closed Jul 27, 2025

• Error when create ModernBert model with flash attention TypeError: RotaryEmbedding.__init__() got an unexpected keyword argument 'pos_idx_in_fp32'

  #38843 closed Jul 27, 2025

• Reproducibility Issue of Siglip2 with Blackwell Architecture GPUs (RTX 5090)

  #38874 closed Jul 27, 2025

• The wrong config parameter found in src/transformers/models/qwen2_5_vl/configuration_qwen2_5_vl.py.

  #38889 closed Jul 27, 2025

• CRITICAL ISSUE REPORT! GEMMA 3 1B CANNOT RUN!

  #39686 closed Jul 26, 2025

• text-generation extremely slow with large `bad_words_ids` list

  #39512 closed Jul 25, 2025

• Does Gemma 3 need positions ids to be 1-indexed explicitly?

  #39023 closed Jul 25, 2025

• Add Deepseek-VL

  #36110 closed Jul 25, 2025

• Grammatical error in the "Loading model's" page

  #39018 closed Jul 25, 2025

• Inference API Returning 404

  #39650 closed Jul 25, 2025

• Backwards incompatible change in returned hidden states

  #39558 closed Jul 25, 2025

• Typo in `apply_transcrition_request` method name

  #39530 closed Jul 25, 2025

• video_auto_processing.py breaks everything

  #38846 closed Jul 25, 2025

• Should `compute_metrics` only run on the main process when doing DDP?

  #38851 closed Jul 25, 2025

• VoxtralForConditionalGeneration import error

  #39611 closed Jul 24, 2025

• `Trainer._save()` May Incorrectly Save Empty Model State (safetensors)

  #38686 closed Jul 24, 2025

• `gemma-3-1b-it` with `use_cache=True` and `past_key_values` throws `RuntimeError: CUDA error: device-side assert` error

  #39593 closed Jul 24, 2025

• Wandb isn't logging config in offline mode

  #38968 closed Jul 23, 2025

• The similarity between image and text in siglip2 is very low

  #39597 closed Jul 23, 2025

• Does Qwen_2_5_VL support variable length attention computation?

  #38007 closed Jul 23, 2025

• Have to import cv2 and pop up window frist, or else it stuck forever

  #38139 closed Jul 23, 2025

• CI skipped failures tracking issue

  #38820 closed Jul 23, 2025

• "ValueError: Predictions and/or references don't match the expected format." error

  #39510 closed Jul 22, 2025

• Clarification on Recent Changes to Loss and Gradient Accumulation

  #39567 closed Jul 22, 2025

• Add EfficientLoFTR model

  #36354 closed Jul 22, 2025

• Gemma3 bidirectional mask for image tokens isn't reaching attention forward

  #39389 closed Jul 22, 2025

• Is the new Intel–Weizmann speculative decoding algorithm integrated into Transformers?

  #39545 closed Jul 21, 2025

• Enabling `average_tokens_across_devices` by default in Trainer

  #39392 closed Jul 21, 2025

• T5Gemma problem with tokenizer(?)

  #39521 closed Jul 21, 2025

• Causal mask is not compatible with Qwen2-VL when using padding-free training

  #39400 closed Jul 21, 2025

• KeyError: 'llava_qwen2'

  #39533 closed Jul 21, 2025

• Add Gemma 3 For Sequence Classification

  #36755 closed Jul 21, 2025

• Expected all tensors to be on the same device, but found at least two devices

  #37545 closed Jul 21, 2025

• DynamicCache results in too many torch recompiles after 4.51

  #37908 closed Jul 21, 2025

• Confusion about num_labels and problem_type in classification logic 🐛

  #38219 closed Jul 21, 2025

• Silent Overwrite of Custom Optimizer When Using DeepSpeed with Transformers Trainer

  #38753 closed Jul 21, 2025

• DTensor issues when running Llama4ForConditionalGeneration with tensor parallel.

  #38803 closed Jul 21, 2025

• Version 4.52.3 leads to error after bundling with pyinstaller

  #38402 closed Jul 20, 2025

• Issue importing models in jupyter notebooks 'No module named transformers.models.ipynb_checkpoints'

  #38726 closed Jul 19, 2025

• T5Gemma returning 0 loss for s2s training

  #39514 closed Jul 19, 2025

• Whisper models appear to be broken with Flash Attention 2

  #38662 closed Jul 18, 2025

• Speculative Decoding(do_sample=False) get different outputs

  #39421 closed Jul 18, 2025

• BarkProcessor voice_preset doesn't work

  #34634 closed Jul 18, 2025

• dataset 4.0.0 , issue with load_dataset loading audio dataset

  #39497 closed Jul 18, 2025

• Gemma3n don't support chat with history

  #39498 closed Jul 18, 2025

• modeling_flax_gemma.FlaxGemmaModule failed with incompatible shapes when running with GemmaConfig

  #39492 closed Jul 18, 2025

• Error for `return_assistant_tokens_mask` in MLLM processor

  #38521 closed Jul 18, 2025

• `get_video_features` in XCLIPModel always returns `pooled_output`

  #38709 closed Jul 18, 2025

• `.to` is not supported for `4-bit` or `8-bit` bitsandbytes models. Please use the model as it is, since the model has already been set to the correct devices and casted to the correct `dtype`.

  #38717 closed Jul 18, 2025

• I can't make sense of this works on Windows but not on Linux AutoModelForCausalLM.from_pretrained

  #39461 closed Jul 17, 2025

• HfArgumentParser cannot parse `str` for local path

  #39462 closed Jul 17, 2025

• breaking changes in ESM model classes

  #39405 closed Jul 17, 2025

• [torch.export] Unhandled FakeTensor Device Propagation for two different devices

  #38975 closed Jul 17, 2025

• QA pipeline prediction generates wrong response when `top_k` param > 1

  #38984 closed Jul 17, 2025

• When will transformers 4.51.4 be released?

  #37812 closed Jul 17, 2025

• [RuntimeError: Expected all tensors to be on the same device, but found at least two devices] when fine-tuning with peft and device_map=auto

  #38687 closed Jul 17, 2025

• CheckpointLoaderSimple ..... Error while deserializing header: InvalidHeaderDeserialization

  #38692 closed Jul 17, 2025

• can't torch.export.export tinyllama model

  #39463 closed Jul 17, 2025

• Missing 4 spaces in SmolVLMImageProcessorFast

  #39442 closed Jul 16, 2025

• ModernBERT for Sequence Classification - issues with finetuning

  #38720 closed Jul 16, 2025

• When creating a Trainer object for a MixedModel, the initialization tries to access attribute get_base_model (which does not exist) rather than model

  #39204 closed Jul 16, 2025

• SigLip2 text pooler output selection

  #39269 closed Jul 16, 2025

• [YosoConfig] Missing `architectures` field

  #39424 closed Jul 16, 2025

• Qwen3 tokenizer wrong offset_mapping

  #39401 closed Jul 16, 2025

• OpenTelemetry Collector Connection error when installing the latest release 4.53.0 during `docker build`

  #39143 closed Jul 16, 2025

• DBRX model passes probabilities and not logits to the load balancer

  #39055 closed Jul 16, 2025

• `verify_tp_plan` function raises an error if a key without '.' is given

  #38419 closed Jul 16, 2025

• Whisper chunking algorithm increases WER

  #37789 closed Jul 16, 2025

• model_type = self._reverse_config_mapping[key.__name__] KeyError: 'Qwen2RMConfig'

  #38517 closed Jul 16, 2025

• TypeError: 'NoneType' object is not iterable in ESM when using DDP training

  #38667 closed Jul 16, 2025

• LlamaAttention forward function type hint is incorrect

  #38739 closed Jul 15, 2025

• `quantization_method` is not cleared after calling `.dequantize()`

  #39295 closed Jul 15, 2025

• Saving model with shared tensors fails on cpu but succeeds on gpu

  #33688 closed Jul 15, 2025

• Mypy errors since v4.51.0

  #37339 closed Jul 15, 2025

• Errors using TinyLlama-1.1B-Chat-v1.0 and DirectML

  #38340 closed Jul 15, 2025

• Pytorch language_modelling example run_clm fails when streaming is enabled

  #39285 closed Jul 15, 2025

• `transformers.utils.metrics` sets global `TracerProvider`

  #39115 closed Jul 15, 2025

• There is no transformers version that can run DeepSeek V3 generate

  #38710 closed Jul 15, 2025

• Support of Qwen3 GGUF model

  #38650 closed Jul 15, 2025

• Latest Transformers release causes CUDA out-of-memory errors during VisionLLM fine-tuning

  #39337 closed Jul 14, 2025

• Paligemma model card needs update

  #38544 closed Jul 14, 2025

• Using resnet-18 in flax

  #39388 closed Jul 14, 2025

• Getting Warnings When Instantiating Object Detection Models Due to Meta Tensor Initialization

  #37615 closed Jul 14, 2025

• 4.52.2 报错Could not import module 'Qwen3ForCausalLM'

  #38291 closed Jul 14, 2025

• Transformers fail to load deepseek-ai/DeepSeek-V3 with vllm

  #38588 closed Jul 13, 2025

• MambaInnerFnBackward

  #38600 closed Jul 13, 2025

• Failed to full fine tuning code5p 2B

  #38602 closed Jul 13, 2025

• Exporting google/gemma-3n-e4b-it language_model (decoder) into ONNX format

  #39328 closed Jul 12, 2025

• Removing the modification of loss value due to rounding off to 4 digits

  #38032 closed Jul 12, 2025

• `AutoModel.from_pretrained(...)` (with explicit `device_map` unset) fails under `with torch.device("meta")` with PyTorch 2.6.0 and 2.7.0

  #38066 closed Jul 12, 2025

• eval_loss not found when training a peft model using trainer.py / losses not retrieved from base model where appropriate

  #38130 closed Jul 12, 2025

• Clarification on default top_k sampling parameter

  #38549 closed Jul 12, 2025

• hidden_states, self_attn_weights = self.self_attn( ValueError: too many values to unpack (expected 2)

  #38554 closed Jul 12, 2025

123 Issues opened by 118 people

• Could not import module 'AutoTokenizer'. Are this object's requirements defined correctly?

  #40089 opened Aug 11, 2025

• Default behavior of llama tokenizers breaks text by removing spaces (round trip is not identity function)

  #40088 opened Aug 11, 2025

• TypeError in DogeDecoderLayer with MoE Configuration when using dropout()

  #40079 opened Aug 11, 2025

• gpt_oss inference activates *all* experts for every token

  #40073 opened Aug 11, 2025

• Issue running model from ImageSegmentationPipeline

  #40071 opened Aug 10, 2025

• Transformer GGUF support philosophy / naive question

  #40070 opened Aug 10, 2025

• [BUG] No umt5 config for GGUF. This is not supported configuration.

  #40067 opened Aug 10, 2025

• Question: How to write a custome tokenizer form scratch

  #40056 opened Aug 9, 2025

• Whisper transcription accuracy improves when last 1600 samples of input audio are muted

  #40054 opened Aug 9, 2025

• Support text classification with GPT-OSS models

  #40050 opened Aug 9, 2025

• Please support loading Qwen 2.5 VL from GGUF

  #40049 opened Aug 9, 2025

• Recent releases break backwards-compatibility with key_cache

  #40046 opened Aug 8, 2025

• Support loading glm4moe GGUF

  #40042 opened Aug 8, 2025

• `plamo-2-1b` broken on latest main

  #40034 opened Aug 8, 2025

• Add Padding Strategy to DataCollatorForLanguageModeling

  #40032 opened Aug 8, 2025

• [gpt-oss] MoE routing bug in the mxfp4 implementation (in distributed setting)

  #40031 opened Aug 8, 2025

• accelerate==1.10.0 and safetensors==0.6.1 are incompatible with transformers==4.53.1

  #40020 opened Aug 8, 2025

• need GptOssForSequenceClassification

  #40018 opened Aug 8, 2025

• Customizable Logit Warping Strategies for Generation

  #40010 opened Aug 7, 2025

• Possible wrong init call

  #40001 opened Aug 7, 2025

• [gpt-oss] Transform checkpoint from safetensors to state dict

  #39992 opened Aug 7, 2025

• CVE fix for v4.37.2 and v4.38.0

  #39983 opened Aug 7, 2025

• FSDP2 not compatible with transformers >= 4.54.0 GenericForTokenClassification

  #39977 opened Aug 7, 2025

• bug in new transformers: 'Florence2ForConditionalGeneration' object has no attribute '_supports_sdpa'

  #39974 opened Aug 7, 2025

• Gemma3 with fp16 in inference (I don't know if this change is working in fine-tune) #BUG FIX

  #39972 opened Aug 6, 2025

• change `dataloader_persistent_workers` default value to `True`

  #39963 opened Aug 6, 2025

• Calling `trainer.evaluate()` before `trainer.train()` with FSDP 2 raises `ValueError: When using FSDP2, a model and optimizer must be passed together to `Accelerator.prepare()...`

  #39961 opened Aug 6, 2025

• TypeError: Received a NoneType for argument video_processor, but a BaseVideoProcessor was expected.(this issue im getting when using doc-ocr)

  #39958 opened Aug 6, 2025

• Retaining computational graph after using AutoImageProcessor

  #39946 opened Aug 6, 2025

• GPT-OSS mxfp4 with triton_kernel: make_default_matmul_mxfp4_w_layout not found

  #39945 opened Aug 6, 2025

• Breaking change in unset `_tp_plan` attribute

  #39943 opened Aug 6, 2025

• Still getting "fp16 mixed precision requires a GPU (not 'mps')." error

  #39935 opened Aug 5, 2025

• [Gemma3N] Not able to add new special tokens to model/tokenizer due to projection error

  #39921 opened Aug 5, 2025

• When using batch_eval_metrics, inputs are not gathered from different device, which is wrong behavior

  #39916 opened Aug 5, 2025

• Idefics 3: shape mismatch: value tensor of shape [256, 576] cannot be broadcast to indexing result of shape [192, 576]

  #39914 opened Aug 5, 2025

• Question: Llama4 weight reshaping

  #39910 opened Aug 5, 2025

• Hidden torchvision>=0.19.0 dependency results in quiet import failures of e.g. PreTrainedModel

  #39907 opened Aug 5, 2025

• v4.54.1 average_tokens_across_devices=True would cause "ValueError: Tensors must be CUDA and dense" when gathering num_items_in_batch

  #39896 opened Aug 4, 2025

• Add VideoPrism

  #39893 opened Aug 4, 2025

• [Feature Request] Automatically log parallelism configuration from Accelerate to W&B

  #39882 opened Aug 4, 2025

• Checking for additional_chat_templates doesn't work without internet (ConnectionError)

  #39873 opened Aug 3, 2025

• InternVL, PerceptionLM inference freeze in 4.54.1

  #39872 opened Aug 3, 2025

• Tensor parallelism for GLM-4.5

  #39868 opened Aug 2, 2025

• Florence2ForConditionalGeneration does not support Flash Attention 2.0 yet ?...

  #39860 opened Aug 2, 2025

• `make fixup` can't find PLC1802

  #39853 opened Aug 1, 2025

• Inconsistent Function calling behaviour by Mistral-7B-Instruct-v0.3

  #39852 opened Aug 1, 2025

• Support topNSigma sampling in `generate`

  #39850 opened Aug 1, 2025

• Accelerate seems to default mixed precision to bf16 when passing a DeepSpeed config.

  #39849 opened Aug 1, 2025

• Expected behavior of `compute_result` is hard to expect and inconsistent

  #39842 opened Aug 1, 2025

• MistralCommonTokenizer does not match PreTrainedTokenizer

  #39841 opened Aug 1, 2025

• pack_image_features RuntimeError when vision_feature_select_strategy="full"

  #39839 opened Aug 1, 2025

• Crash when running Llama4 on transformers-4.54.1

  #39835 opened Aug 1, 2025

• Allow extra outputs from `GenerationMixin.generate`

  #39834 opened Aug 1, 2025

• Missing einops dependency causing ModuleNotFoundError

  #39811 opened Jul 31, 2025

• Fine tuning qwen2.5 error

  #39804 opened Jul 31, 2025

• Memory leak occurred during training qwen-2.5-vl

  #39803 opened Jul 31, 2025

• ValueError: This model does not support cache_implementation='static'. Please check the following issue: https://github.com/huggingface/transformers/issues/28981

  #39801 opened Jul 30, 2025

• You current version of `autoawq` does not support module quantization skipping, please upgrade `autoawq` package to at least 0.1.8.

  #39798 opened Jul 30, 2025

• Regression - High memory usage when using transformers model with FSDP + LoRA

  #39795 opened Jul 30, 2025

• `transformers serve` Fails to Handle Messages with Nested Content

  #39791 opened Jul 30, 2025

• ViTPose+ models post processing doest not work for `dataset_index : 5`

  #39789 opened Jul 30, 2025

• "CSM audio generation lacks reliable EOS: does not generate all-zero frames → never stops early"

  #39787 opened Jul 30, 2025

• pip install 'transformers[torch]' pulls nvidia dependencies

  #39780 opened Jul 30, 2025

• transformers env fails with: ModuleNotFoundError: No module named 'PIL'

  #39779 opened Jul 30, 2025

• Granite 4.0 Tiny Preview inference broken in

  #39775 opened Jul 30, 2025

• would it be possible to standardize on the vx.y.z format for all tags

  #39771 opened Jul 30, 2025

• Model with non-string type property tool giving incomplete response using VLLM

  #39767 opened Jul 29, 2025

• Follow-up on Issues Regarding Training State Restoration from Interruptions

  #39755 opened Jul 29, 2025

• Inv frequency has not default, going against our philosophy

  #39753 opened Jul 29, 2025

• Qwen2_5_VLForConditionalGeneration cfg forward twice error

  #39749 opened Jul 29, 2025

• losses, logits, labels = self.prediction_step(model, inputs, prediction_loss_only, ignore_keys=ignore_keys) output logits and labels are not the batch same size

  #39736 opened Jul 28, 2025

• [transformers==4.54.0] FSDP1 forward misalignment after loading state dict

  #39720 opened Jul 28, 2025

• OWLv2 with visual prompt - alternative query embedding selection method

  #39710 opened Jul 27, 2025

• [i18n-<bn>] Translating docs to <Bengali>

  #39705 opened Jul 27, 2025

• ValueError: Number of image placeholders in the prompt does not match the number of images. internVL3

  #39703 opened Jul 26, 2025

• No flag to support Conditional Parameter Loading for gemma-3n-E2B models in transformer

  #39699 opened Jul 26, 2025

• 4.54.0 bug: ImportError: cannot import name 'deterministic_g' from 'transformers.modeling_flash_attention_utils'

  #39694 opened Jul 26, 2025

• SigLIP2 documentation example has multiple errors (model/processor mismatch + quantization failure)

  #39692 opened Jul 26, 2025

• Qwen 2.5 VL - error without attention_mask

  #39685 opened Jul 26, 2025

• Add multi-candidate & tree search for assisted decoding (speculative decoding)

  #39684 opened Jul 25, 2025

• Accelerate beam search decoding via tree attention

  #39682 opened Jul 25, 2025

• error: argument --deepspeed: invalid dict value: '<path>'

  #39673 opened Jul 25, 2025

• Issue when initializing a DynamicCache

  #39668 opened Jul 25, 2025

• T5Gemma training not working

  #39656 opened Jul 25, 2025

• Use DP+FSDP device mesh dimensions for scaling loss with default value of average_tokens_across_devices: True

  #39648 opened Jul 24, 2025

• Please develop DataCollatorForVisionLanguageModeling to support visual model training !!!

  #39647 opened Jul 24, 2025

• [BUG] Run 111B+ Teacher distributed inference and 8B Student distributed training on multi-node H200 GPUs using the Transformers Trainer without encountering OOM errors?

  #39637 opened Jul 24, 2025

• FSDP v1 bug: trainer incorrectly uses an unwrapped model

  #39619 opened Jul 23, 2025

• SageAttention for attention implementation?

  #39618 opened Jul 23, 2025

• Trainer: Error when folded metrics are saved

  #39616 opened Jul 23, 2025

• Qwen3 Fails w/4D Attn Mask when using FA2

  #39608 opened Jul 23, 2025

• ImageClassificationPipeline preprocess should accept numpy/tensor arrays

  #39607 opened Jul 23, 2025

• Does transformers support python3.13 -- disable-gil or python3.14 free threading?

  #39596 opened Jul 23, 2025

• Model forward execution in full eager mode?

  #39565 opened Jul 21, 2025

• Why `is_causal` is not used in `flash_attention_forward` ?

  #39554 opened Jul 21, 2025

• Is there plan to integrate ColQwen2.5 into Transformers?

  #39549 opened Jul 21, 2025

• ValueError: You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time

  #39542 opened Jul 21, 2025

• Add muon and flash-muon optimizer

  #39537 opened Jul 20, 2025

• InformerForPrediction [I would like to seek your opinions, everyone, How can I set the dynamic real features for prediction]

  #39551 opened Jul 20, 2025

• training google colab error

  #39527 opened Jul 19, 2025

• paged attention NOT working with Qwen Models

  #39525 opened Jul 19, 2025

• T5Gemma failing on provided example

  #39522 opened Jul 19, 2025

• Export voxtral to ExecuTorch

  #39511 opened Jul 18, 2025

• Whisper transcription is 2x slower between 4.51.3 -> 4.52.1

  #39508 opened Jul 18, 2025

• Add Muon Optimiser for 2x faster convergence

  #39495 opened Jul 18, 2025

• Transformers still tries to use apex.amp which is no longer a thing in apex.

  #39484 opened Jul 17, 2025

• Adding Space-Time-MiniLM-v0

  #39479 opened Jul 17, 2025

• Allow `load_best_model_at_end=True` to work when `save_steps < eval_steps` and best model is saved

  #39476 opened Jul 17, 2025

• Unexpected behaviour with transformers versions above 4.28 for Donut

  #39473 opened Jul 17, 2025

• Autoformer get_lagged_subsequences always true if condition

  #39460 opened Jul 16, 2025

• Add Interactive Multi-Modal Attention Visualization for Vision-Language Models

  #39440 opened Jul 15, 2025

• Support for per-token latency tracking in `generate()` (suggested options: using callback, profiler class, or using a config flag)

  #39437 opened Jul 15, 2025

• Export LFM2 to ExecuTorch

  #39436 opened Jul 15, 2025

• Add DiCoW: Diarization-Conditioned Whisper

  #39430 opened Jul 15, 2025

• Gemma 3 Compilation Issues During Generation

  #39427 opened Jul 15, 2025

• object detection : matchin outputs.last_hidden_state with results

  #39426 opened Jul 15, 2025

• Exeception 3 type mismatch

  #39413 opened Jul 15, 2025

• FP8 training support for Model Parallel / Tensor Parallel (MP/TP)

  #39410 opened Jul 15, 2025

• Off-by-one error when using flash_attention with a sliding window

  #39408 opened Jul 15, 2025

• Whisper `return_language` with pipeline no longer working

  #39404 opened Jul 14, 2025

• Qwen2.5-VL Sharding error when using Tensor Parallelism

  #39399 opened Jul 14, 2025

• Mask2FormerImageProcessor yields inconsistent results between single and batch inference

  #39382 opened Jul 12, 2025

• Handling of full_text_row_masked_out_mask in mllama is incorrect.

  #39379 opened Jul 12, 2025

138 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Add Segment Anything 2 (SAM2)

  #32317 commented on Aug 12, 2025 • 166 new comments

• [WiP] Add xcodec2 model

  #37868 commented on Jul 30, 2025 • 76 new comments

• Add support for Florence-2

  #38188 commented on Aug 7, 2025 • 61 new comments

• blt wip

  #38579 commented on Aug 11, 2025 • 47 new comments

• Feat: add Kwai-Keye transformers

  #39292 commented on Aug 11, 2025 • 42 new comments

• Add fastconformer encoder support for nvidia/parakeet and nvidia/canary models

  #39062 commented on Aug 3, 2025 • 21 new comments

• [WIP] Computer vision util: vision visualizer

  #36892 commented on Aug 11, 2025 • 16 new comments

• support MiniCPM-o2.6

  #37917 commented on Aug 12, 2025 • 16 new comments

• Add NVIDIA Cosmos

  #36476 commented on Jul 16, 2025 • 14 new comments

• Add T5LA models

  #39293 commented on Jul 25, 2025 • 14 new comments

• feat: Add ConvaiCausalLM model for Hindi Causal Language Modeling

  #37837 commented on Jul 16, 2025 • 10 new comments

• Add standardized model card for facebook/data2vec-audio-base-960h

  #39368 commented on Jul 24, 2025 • 10 new comments

• Fix the issue that csm model cannot work with pipeline mode.

  #39349 commented on Aug 11, 2025 • 7 new comments

• [omni modality] support composite processor config

  #38142 commented on Aug 5, 2025 • 7 new comments

• Add Ovis2 model and processor implementation

  #37088 commented on Aug 11, 2025 • 5 new comments

• fix: filter None router logits in Qwen3 MoE and handle empty router logits (#39203)

  #39206 commented on Jul 21, 2025 • 5 new comments

• Force real tensors and clone state_dict in src/transformers/modeling_utils.py

  #38114 commented on Jul 15, 2025 • 3 new comments

• Add X-Codec model

  #38248 commented on Jul 23, 2025 • 3 new comments

• Fix: Add version check for timm to support mobilenetv5 models (fixes #39208)

  #39264 commented on Jul 14, 2025 • 3 new comments

• feat: add sliding window attention to Continuous Batching

  #39225 commented on Aug 1, 2025 • 3 new comments

• fix bug when using DP in trl, the batch size of input and output dism…

  #38938 commented on Aug 11, 2025 • 2 new comments

• Fix audio pipeline with torchcodec input

  #39309 commented on Aug 1, 2025 • 2 new comments

• Make executorch integration more seamless by analyzing model signature

  #36969 commented on Jul 16, 2025 • 1 new comment

• Provide clearer instructions on how to specify target language.

  #38786 commented on Jul 21, 2025 • 1 new comment

• add pin memory and block table

  #39130 commented on Aug 11, 2025 • 1 new comment

• deci gguf support

  #38669 commented on Jul 29, 2025 • 1 new comment

• Fix ModernBERT tokenizer issue with is_split_into_words flag

  #38564 commented on Jul 16, 2025 • 1 new comment

• another way to use shift_labels

  #38533 commented on Jul 16, 2025 • 1 new comment

• Bug in modeling_bart.eager_attention_forward

  #39365 commented on Aug 11, 2025 • 0 new comments

• Add StyleTTS 2

  #35790 commented on Jul 28, 2025 • 0 new comments

• Fix ImportError: cannot import name 'GenerationMixin' from 'transformers.generation'

  #36011 commented on Jul 21, 2025 • 0 new comments

• env.useBrowserCache = true causes JSON parsing error, forced to disable cache making app slower.

  #39352 commented on Aug 11, 2025 • 0 new comments

• Add Phi-3.5-vision

  #36036 commented on Aug 1, 2025 • 0 new comments

• Fix inconsistency in SeamlessM4T and SeamlessM4Tv2 docs

  #39364 commented on Aug 7, 2025 • 0 new comments

• [Validation] First implementation of `@strict` from `huggingface_hub`

  #36534 commented on Jul 29, 2025 • 0 new comments

• fix unexpected kws of input_ids when setup no speech detection of whisper

  #36809 commented on Jul 23, 2025 • 0 new comments

• Update docstring for glm4v

  #39357 commented on Jul 14, 2025 • 0 new comments

• RoBERTa is not well implemented for tokenizers with pad_token_id != 1

  #34528 commented on Aug 11, 2025 • 0 new comments

• Add FAST

  #35476 commented on Jul 30, 2025 • 0 new comments

• Add JinaBERT model

  #35320 commented on Jul 15, 2025 • 0 new comments

• Add Molmo (7B-D, 7B-O, 70B)

  #33962 commented on Jul 21, 2025 • 0 new comments

• [Community contributions] Model cards

  #36979 commented on Aug 11, 2025 • 0 new comments

• Load a pretrainedfast tokenizer if fast=true and tokenizer.json exists

  #33751 commented on Jul 15, 2025 • 0 new comments

• [Contributions Welcome] Add Fast Image Processors

  #36978 commented on Aug 11, 2025 • 0 new comments

• DeepSpeed sequence parallelism (aka Ulysses) integration with HF transformer

  #32305 commented on Jul 15, 2025 • 0 new comments

• Implement MambaForSequenceClassification

  #31155 commented on Jul 15, 2025 • 0 new comments

• AutoConfig has potential issue with composite config.

  #38258 commented on Aug 12, 2025 • 0 new comments

• Fix Inconsistant `input_feature` length and `attention_mask` length in `WhisperFeatureExtractor`

  #39221 commented on Jul 21, 2025 • 0 new comments

• feat(trainer): emergency checkpointing on crashes & SIGTERM/SIGINT

  #39140 commented on Aug 5, 2025 • 0 new comments

• Disable static cache on certain MoE models

  #39108 commented on Jul 28, 2025 • 0 new comments

• Update Dockerfiles to install packages inside a virtual environment

  #39098 commented on Aug 11, 2025 • 0 new comments

• Bug/38843 fix pos idx in fp32 parameter error

  #39064 commented on Jul 22, 2025 • 0 new comments

• Fix slow test_moshika_greedy_unconditional_fp16

  #39251 commented on Jul 15, 2025 • 0 new comments

• Allow compression on meta device

  #39039 commented on Aug 8, 2025 • 0 new comments

• Add Dust3R

  #38805 commented on Jul 22, 2025 • 0 new comments

• Adding custom 3d mask into ModernBert

  #38671 commented on Jul 29, 2025 • 0 new comments

• Adds Universal Intelligence to awesome transformers documentation

  #38641 commented on Aug 7, 2025 • 0 new comments

• Add Bagel

  #38569 commented on Aug 10, 2025 • 0 new comments

• 🔴[`Attention`] Bert-based Models Attention Refactor

  #38301 commented on Jul 16, 2025 • 0 new comments

• Fix the shape of ModernBertForMaskedLM's output hidden_states

  #38272 commented on Jul 16, 2025 • 0 new comments

• Add dates to the model docs

  #39320 commented on Aug 11, 2025 • 0 new comments

• add profiler to trainer

  #37889 commented on Jul 29, 2025 • 0 new comments

• fix colpali mapping

  #39353 commented on Jul 14, 2025 • 0 new comments

• Update ruff to 0.12.3 and apply its fixes

  #37809 commented on Jul 21, 2025 • 0 new comments

• Vectorize deepseek moe

  #37769 commented on Jul 16, 2025 • 0 new comments

• fix: qwen2.5 omni apply_chat_template system content check

  #37511 commented on Aug 8, 2025 • 0 new comments

• [RFC] Fix Gemma 3 FP16 with activation scaling

  #37226 commented on Jul 16, 2025 • 0 new comments

• trying custom tokenizer fix

  #37177 commented on Jul 16, 2025 • 0 new comments

• Add Plain-DETR

  #37096 commented on Aug 11, 2025 • 0 new comments

• Add Matching Anything by Segmenting Anything (MASA) MOT tracking model

  #32164 commented on Jul 14, 2025 • 0 new comments

• _load_rng_state after get_batch_samples may break training reproducibility when dataloader has random operations

  #39215 commented on Jul 23, 2025 • 0 new comments

• 🌐 [i18n-KO] Translating docs to Korean

  #20179 commented on Jul 24, 2025 • 0 new comments

• Adding support for Gemma 3n GGUFs

  #39329 commented on Jul 24, 2025 • 0 new comments

• Model implmenetation using Liger Kernel layers

  #38416 commented on Jul 24, 2025 • 0 new comments

• Support for Multiple Datasets and Domain-Specific Loss Calculation in Trainer

  #30725 commented on Jul 24, 2025 • 0 new comments

• Trainer/accelerate doesn't save model when using FSDP with SHARDED_STATE_DICT

  #30491 commented on Jul 24, 2025 • 0 new comments

• Segfault on Apple M4 using AutoModelForSequenceClassification with BETO model on CPU

  #39020 commented on Jul 25, 2025 • 0 new comments

• Whisper word-level timestamp extraction fails with beam search

  #36093 commented on Jul 28, 2025 • 0 new comments

• Issue with module.smart_apply(module._initialize_weights) in the initialize_weights Function of modeling_utils.py

  #39027 commented on Jul 28, 2025 • 0 new comments

• Output logits differ significantly for different attn_implementations on image inputs

  #39067 commented on Jul 28, 2025 • 0 new comments

• ValueError: GGUF model with architecture deci is not supported yet.

  #37736 commented on Jul 28, 2025 • 0 new comments

• Resuming training from an interrupted checkpoint fails to save the final checkpoint.

  #38939 commented on Jul 30, 2025 • 0 new comments

• How to use other acceleration apis of npu?

  #39105 commented on Jul 31, 2025 • 0 new comments

• Adding native support to load GGUF models using transformers

  #38063 commented on Jul 31, 2025 • 0 new comments

• Add support for BAGEL from ByteDance

  #38267 commented on Jul 31, 2025 • 0 new comments

• Object detection training/fine-tuning for Owl-vit/Owlv2

  #33664 commented on Aug 1, 2025 • 0 new comments

• Community contribution: Adding GGUF support for more architectures

  #33260 commented on Aug 2, 2025 • 0 new comments

• If a training job job failed MLFlow will not be reported and MLFlow shows job still running

  #30333 commented on Jul 15, 2025 • 0 new comments

• [DOCS] Add `pruna` as optimization framework

  #38740 commented on Jul 16, 2025 • 0 new comments

• Modernbert 3D attention mask

  #38040 commented on Jul 16, 2025 • 0 new comments

• Automatic dynamic batch size selection for DataCollatorWithFlattening

  #33945 commented on Jul 16, 2025 • 0 new comments

• Flex attention support with arbitrary 4d mask for LlamaModel

  #33898 commented on Jul 17, 2025 • 0 new comments

• Add `pruna` integration for loading model through `transformers.from_pretrained` / `pipeline`.

  #37971 commented on Jul 17, 2025 • 0 new comments

• Add HF integration dates + paper release dates to the model docs

  #39319 commented on Jul 18, 2025 • 0 new comments

• The same situation as #31377 occurred when using Qwen/Qwen2-VL-7B-Instruct

  #33399 commented on Jul 18, 2025 • 0 new comments

• Safetensors deserializing silently mishandles tied parameters

  #38870 commented on Jul 18, 2025 • 0 new comments

• Support 2D Array Inputs in Wav2Vec2FeatureExtractor for Non-Waveform Modalities

  #39291 commented on Jul 18, 2025 • 0 new comments

• RuntimeError when loading llmcompressor W8A8 quantized model: int8 dtype in weight initialization

  #39366 commented on Jul 19, 2025 • 0 new comments

• Error: StaticCache.__init__() got an unexpected keyword argument 'batch_size'

  #38914 commented on Jul 20, 2025 • 0 new comments

• Implement Titans Architecture with GRPO Fine-Tuning

  #36352 commented on Jul 21, 2025 • 0 new comments

• `AutoTokenizer.from_pretrained` does not propagate `token`

  #39030 commented on Jul 21, 2025 • 0 new comments

• Caching of model code in ~/.cache/huggingface/modules/transformers_modules

  #39107 commented on Jul 22, 2025 • 0 new comments

• add MiniCPM-o

  #37029 commented on Jul 22, 2025 • 0 new comments

• Unknown Model (mobilenetv5_300m_enc) when loading Gemma 3n

  #39208 commented on Jul 22, 2025 • 0 new comments

• Inference with model.generate( ) using a quantized model leads to assertion error

  #39311 commented on Aug 9, 2025 • 0 new comments

• Qwen3 MOE models w/non-empty `mlp_only_layers` fail when `output_router_logits=True`

  #39203 commented on Aug 9, 2025 • 0 new comments

• Improve CI/CD by completing migration from setup.py to pyproject.toml

  #38928 commented on Aug 9, 2025 • 0 new comments

• CUDA OOM when running meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8

  #37532 commented on Aug 9, 2025 • 0 new comments

• `MoshiIntegrationTests` started to fail after #34464

  #38725 commented on Aug 9, 2025 • 0 new comments

• TypeError: GenerationMixin._extract_past_from_model_output() got an unexpected keyword argument 'standardize_cache_format'

  #39336 commented on Aug 10, 2025 • 0 new comments

• [Trainer] Eval loss depends on batch size (with solution)

  #39241 commented on Aug 10, 2025 • 0 new comments

• bf16_full_eval=True moves model to device before FSDP application and causes cuda OOM

  #39136 commented on Aug 10, 2025 • 0 new comments

• QWEN2VLProcessor missing video_token_id in mm_token_type_ids

  #39112 commented on Aug 10, 2025 • 0 new comments

• AutoModelForCausalLM.from_pretrained(..., device_map=...) ignore `Tensor.retain_grad()` in Multi-GPUs setting

  #39036 commented on Aug 10, 2025 • 0 new comments

• CPMANT Model Fails to Run Following Official Tutorial

  #39026 commented on Aug 10, 2025 • 0 new comments

• Potential Memory Leak or Caching in Fast Image Processor

  #38656 commented on Aug 10, 2025 • 0 new comments

• Attention refactor in #35235 adds a `__getitem__` into the forward pass, which causes errors with torch dynamo.

  #38271 commented on Aug 10, 2025 • 0 new comments

• YaRN: factor is not effective with original_max_position_embeddings

  #38224 commented on Aug 10, 2025 • 0 new comments

• "pipeline" is not exported from module "transformers"

  #37646 commented on Aug 10, 2025 • 0 new comments

• Please support GGUF format for UMT5EncoderModel

  #36774 commented on Aug 10, 2025 • 0 new comments

• FlashAttention2 support for GSAI-ML / LLaDA-8B-Instruct?

  #39377 commented on Aug 11, 2025 • 0 new comments

• v4.53.0 - Qwen 2.5 VL Flash Attention error - object has no attribute is_causal

  #39231 commented on Aug 4, 2025 • 0 new comments

• torch fake_tensor load hf model failed

  #39217 commented on Aug 4, 2025 • 0 new comments

• Exporting Llava decoder into ONNX format

  #38924 commented on Aug 4, 2025 • 0 new comments

• transformers: FlaubertTokenizer: do_lowercase_and_remove_accent: make the logger warning actionable (don't only tell what's wrong, rather suggest what could be done about that)

  #39224 commented on Aug 4, 2025 • 0 new comments

• Failed to export PyTorch traced graph of Mixtral-8x7B-Instruct-v0.1 due to the PR #32429

  #38518 commented on Aug 4, 2025 • 0 new comments

• Torch patches tracker for HPU/Gaudi

  #39175 commented on Aug 5, 2025 • 0 new comments

• [FEAT] [non-CUDA]: Support alternative implementation for `constraints.positive_definite.check`

  #36660 commented on Aug 5, 2025 • 0 new comments

• We now require users to upgrade torch to at least v2.6 in order to use the function.

  #38464 commented on Aug 5, 2025 • 0 new comments

• ../aten/src/ATen/native/cuda/Indexing.cu:1289: indexSelectLargeIndex: block: [267,0,0], thread: [25,0,0] Assertion `srcIndex < srcSelectDimSize` failed.

  #33985 commented on Aug 6, 2025 • 0 new comments

• ImportError: cannot import name 'pipeline' from 'transformers'

  #39137 commented on Aug 6, 2025 • 0 new comments

• Loading audio in video from video URLs fail with chat template

  #39076 commented on Aug 7, 2025 • 0 new comments

• [RFC] Updating pipeline models

  #26690 commented on Aug 7, 2025 • 0 new comments

• Support for context-free-grammars (CFG) to constrain model output

  #25778 commented on Aug 7, 2025 • 0 new comments

• hangs during training using deepspeed

  #39275 commented on Aug 8, 2025 • 0 new comments

• Please help i am trying to run model but issue

  #39260 commented on Aug 8, 2025 • 0 new comments

• Support `StaticCache` in assisted generation

  #32946 commented on Aug 8, 2025 • 0 new comments

• Whisper demo code for model + processor API is broken

  #39318 commented on Aug 9, 2025 • 0 new comments