5 Releases published by 2 people

• v4.49.0-AyaVision Aya Vision (Based on v4.49.0)

  published Mar 4, 2025

• v4.49.0-Gemma-3 Gemma 3 (Based on v4.49.0)

  published Mar 18, 2025

• v4.49.0-Mistral-3 Mistral 3 (Based on v4.49.0)

  published Mar 18, 2025

• v4.50.0 Release v4.50.0

  published Mar 21, 2025

• v4.50.1 Patch release v4.50.1

  published Mar 25, 2025

244 Pull requests merged by 88 people

• [docs] Attention mask image

  #36970 merged Mar 26, 2025

• Remove deprecated training arguments

  #36946 merged Mar 26, 2025

• fix typos in the code comments and error messages

  #36993 merged Mar 26, 2025

• Log the correct learning rate

  #36973 merged Mar 26, 2025

• Fix device_map check for ggml files

  #37003 merged Mar 26, 2025

• Fix removing "cpu" from frozenset in bitsandbytes.py to allow better ROCm support.

  #36975 merged Mar 26, 2025

• Allow easy registration of custom attention functions

  #36889 merged Mar 26, 2025

• Fix get_device_properties

  #36997 merged Mar 26, 2025

• Fix Optional type annotation

  #36841 merged Mar 26, 2025

• Install networkx==3.2.1 manually in some CircleCI jobs after #36957

  #37000 merged Mar 26, 2025

• Use torch.expm1

  #36995 merged Mar 26, 2025

• byebye CircleCI TF jobs

  #36998 merged Mar 26, 2025

• Fix tensor dtype mismatch

  #36985 merged Mar 26, 2025

• 🚨Deprecate legacy argument for image-text-to-text models and adopt new behavior by default

  #36307 merged Mar 25, 2025

• update bot comment again

  #36974 merged Mar 25, 2025

• Add ruff target-version

  #36971 merged Mar 25, 2025

• [docs] Fix image link

  #36869 merged Mar 25, 2025

• Remove extra tensor clone in PyTorch code

  #36748 merged Mar 25, 2025

• update

  #36972 merged Mar 25, 2025

• Updated docker files to use uv for installing packages

  #36957 merged Mar 25, 2025

• typo fixed in README_fr.md

  #36951 merged Mar 25, 2025

• Change GPUS to GPUs

  #36945 merged Mar 25, 2025

• Update after #36962

  #36965 merged Mar 25, 2025

• Update ruff to 0.11.2

  #36962 merged Mar 25, 2025

• [Utils] torch version checks optionally accept dev versions

  #36847 merged Mar 25, 2025

• Fix cuda index issue in cache allocator

  #36937 merged Mar 25, 2025

• Support return_tensors in audio chat templates

  #34601 merged Mar 25, 2025

• fix typos in the tests directory

  #36932 merged Mar 25, 2025

• Export for Phi4-mini

  #36780 merged Mar 25, 2025

• Fixing _pre_quantization_dtype when torch_dtype is None

  #36930 merged Mar 25, 2025

• Add Phi4 multimodal

  #36939 merged Mar 25, 2025

• Deprecate #36741 and map Causal to Conditional

  #36917 merged Mar 25, 2025

• Disallow Offload to disk for gguf files

  #36933 merged Mar 24, 2025

• Fix processor kwargs qwen2 vl

  #36890 merged Mar 24, 2025

• Added support for seed in DataCollatorForWholeWordMask

  #36903 merged Mar 24, 2025

• More precise comment

  #36935 merged Mar 24, 2025

• Fix pytorch defomr attn path

  #36923 merged Mar 24, 2025

• [2/N] Use pyupgrade --py39-plus to improve code

  #36857 merged Mar 24, 2025

• Update trainer_pt_utils.py docstrings for consistency

  #36912 merged Mar 24, 2025

• Fix typos

  #36910 merged Mar 24, 2025

• Use another repo. for Mistral3 processor testing

  #36925 merged Mar 24, 2025

• Fix Compressed tensors to_dict_diff

  #36922 merged Mar 24, 2025

• [chameleon] fix num image token check

  #36918 merged Mar 24, 2025

• tests: fix asyncio.wait() usage for python>=3.11

  #36898 merged Mar 24, 2025

• [Fix] Add original_max_position_embeddings to YARN rope_scaling optional keys

  #36877 merged Mar 24, 2025

• Fix torch version guard at import

  #36907 merged Mar 24, 2025

• fix Gemma3 Config

  #36893 merged Mar 24, 2025

• Update installation.md

  #36826 merged Mar 21, 2025

• [docs] Model docs

  #36469 merged Mar 21, 2025

• Fix Pan and Scan on batched images Gemma3

  #36864 merged Mar 21, 2025

• Simplify keep_in_fp32_modules logic

  #36722 merged Mar 21, 2025

• fix: loss computation after embeddings resize - mllama

  #36840 merged Mar 21, 2025

• Fix: dtype cannot be str

  #36262 merged Mar 21, 2025

• Minor Gemma 3 fixes

  #36884 merged Mar 21, 2025

• Use deformable_detr kernel from the Hub

  #36853 merged Mar 21, 2025

• Gemma 3 tests expect greedy decoding

  #36882 merged Mar 21, 2025

• 🔴 🔴 🔴 supersede paligemma forward to shift pos id indexing

  #36859 merged Mar 21, 2025

• [generate] model defaults being inherited only happens for newer models

  #36881 merged Mar 21, 2025

• Revert "Update deprecated Jax calls (#35919)"

  #36880 merged Mar 21, 2025

• Make ViTPooler configurable

  #36517 merged Mar 21, 2025

• chore: fix typos in the tests directory

  #36813 merged Mar 21, 2025

• Remove call to .item in get_batch_samples

  #36861 merged Mar 21, 2025

• FIX FSDP plugin update for QLoRA

  #36720 merged Mar 21, 2025

• [CI] doc builder without custom image

  #36862 merged Mar 21, 2025

• Mllama: raise better error

  #35934 merged Mar 21, 2025

• Refactor Aya Vision with modular

  #36688 merged Mar 20, 2025

• Add support for seed in DataCollatorForLanguageModeling

  #36497 merged Mar 20, 2025

• [CI] fix update metadata job

  #36850 merged Mar 20, 2025

• Gemma3: fix test

  #36820 merged Mar 20, 2025

• [torchao] revert to get_apply_tensor_subclass

  #36849 merged Mar 20, 2025

• Add model visual debugger

  #36798 merged Mar 20, 2025

• Add Prompt Depth Anything Model

  #35401 merged Mar 20, 2025

• Refactor Attention implementation for ViT-based models

  #36545 merged Mar 20, 2025

• DeepSpeed tensor parallel+ZeRO

  #36825 merged Mar 20, 2025

• Support loading Quark quantized models in Transformers

  #36372 merged Mar 20, 2025

• Use pyupgrade --py39-plus to improve code

  #36843 merged Mar 20, 2025

• Fix hqq skipped modules and dynamic quant

  #36821 merged Mar 20, 2025

• Fix ONNX export for sequence classification head

  #36332 merged Mar 20, 2025

• Shieldgemma2

  #36678 merged Mar 20, 2025

• Fix: remove the redundant snippet of _whole_word_mask

  #36759 merged Mar 20, 2025

• Gemma 3: Adding explicit GenerationConfig and refactoring conversion …

  #36833 merged Mar 20, 2025

• Fix import for torch 2.0, 2.1 - guard typehint for "device_mesh"

  #36768 merged Mar 20, 2025

• Update min safetensors bis

  #36823 merged Mar 20, 2025

• [generate] clarify docstrings: when to inherit GenerationMixin

  #36605 merged Mar 20, 2025

• [modular] Sort modular skips

  #36304 merged Mar 20, 2025

• Pass state dict

  #35234 merged Mar 20, 2025

• [qwen2 audio] remove redundant code and update docs

  #36282 merged Mar 20, 2025

• Update deprecated Jax calls

  #35919 merged Mar 20, 2025

• Fix fp16 ONNX export for RT-DETR and RT-DETRv2

  #36460 merged Mar 20, 2025

• Pass num_items_in_batch directly to loss computation

  #36753 merged Mar 20, 2025

• Saving Trainer.collator.tokenizer in when Trainer.processing_class is None

  #36552 merged Mar 20, 2025

• fix tiktoken convert to pass AddedToken to Tokenizer

  #36566 merged Mar 20, 2025

• [ForCausalLMLoss] allow users to pass shifted labels

  #36607 merged Mar 20, 2025

• Disable inductor config setter by default

  #36608 merged Mar 20, 2025

• Fix swanlab global step

  #36728 merged Mar 20, 2025

• rewrite main method in Qwen2, making it more clear

  #36772 merged Mar 20, 2025

• Move the warning to the documentation for DataCollatorWithFlattening

  #36707 merged Mar 20, 2025

• Remove our AdamW implementation

  #36177 merged Mar 19, 2025

• Update configuration_qwen2.py

  #36735 merged Mar 19, 2025

• quick fix fast_image_processor register error

  #36716 merged Mar 19, 2025

• Add Space to Bitsandbytes doc

  #36834 merged Mar 19, 2025

• Support tracable dynamicKVcache

  #36311 merged Mar 19, 2025

• One more fix for reviewer assignment

  #36829 merged Mar 19, 2025

• [gemma 3] multimodal checkpoints + AutoModelForCausalLM

  #36741 merged Mar 19, 2025

• enable OffloadedCache on XPU from PyTorch 2.7

  #36654 merged Mar 19, 2025

• Add option for ao base configs

  #36526 merged Mar 19, 2025

• Add attention visualization tool

  #36630 merged Mar 19, 2025

• [Generation] remove leftover code from end-to-end compilation

  #36685 merged Mar 19, 2025

• Fix Device map for bitsandbytes tests

  #36800 merged Mar 19, 2025

• Remove dist": "loadfile" for pytest for CircleCI jobs

  #36811 merged Mar 19, 2025

• fix "Cannot copy out of meta tensor; no data!" issue for BartForConditionalGeneration model

  #36572 merged Mar 19, 2025

• Expectations test utils

  #36569 merged Mar 18, 2025

• [generate] ✨ vectorized beam search ✨

  #35802 merged Mar 18, 2025

• Support custom dosctrings in modular

  #36726 merged Mar 18, 2025

• Fix chameleon's TypeError because inputs_embeds may None

  #36673 merged Mar 18, 2025

• Fix casting dtype for qunatization

  #36799 merged Mar 18, 2025

• Fix Mistral3 tests

  #36797 merged Mar 18, 2025

• Loading optimizations

  #36742 merged Mar 18, 2025

• Update SHA for tj-actions/changed-files

  #36795 merged Mar 18, 2025

• fix hqq due to recent modeling changes

  #36771 merged Mar 18, 2025

• Add Mistral3

  #36790 merged Mar 18, 2025

• Fix gemma3_text tokenizer in mapping

  #36793 merged Mar 18, 2025

• Fixing typo in gemma3 image_processor_fast and adding a small test

  #36776 merged Mar 18, 2025

• chore: fix typos in tests directory

  #36785 merged Mar 18, 2025

• fix typos in the tests directory

  #36717 merged Mar 17, 2025

• doc: Clarify is_decoder usage in PretrainedConfig documentation

  #36724 merged Mar 17, 2025

• [docs] Update README

  #36265 merged Mar 17, 2025

• [CI] remove redundant checks in test_eager_matches_sdpa_inference

  #36740 merged Mar 17, 2025

• [MINOR:TYPO] Update hubert.md

  #36733 merged Mar 17, 2025

• Fix TrainingArguments.torch_empty_cache_steps post_init check

  #36734 merged Mar 17, 2025

• Fix test isolation for clear_import_cache utility

  #36345 merged Mar 17, 2025

• fix xpu tests

  #36656 merged Mar 17, 2025

• Allow ray datasets to be used with trainer

  #36699 merged Mar 17, 2025

• fix can_generate

  #36570 merged Mar 17, 2025

• enable/disable compile for quants methods

  #36519 merged Mar 17, 2025

• 🚨🚨🚨 Fix sdpa in sam and refactor relative position embeddings

  #36422 merged Mar 17, 2025

• [Generation, Gemma 3] When passing a custom generation_config, overwrite default values with the model's base generation_config

  #36684 merged Mar 15, 2025

• Fix grad accum arbitrary value

  #36691 merged Mar 14, 2025

• Fix post_init() code duplication

  #36727 merged Mar 14, 2025

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

  #36698 merged Mar 14, 2025

• [tests] Parameterized test_eager_matches_sdpa_inference

  #36650 merged Mar 14, 2025

• Try working around the processor registration bugs

  #36184 merged Mar 14, 2025

• Fix/best model checkpoint fix

  #35885 merged Mar 14, 2025

• [model loading] don't gc.collect() if only 1 shard is used

  #36721 merged Mar 14, 2025

• Cleanup the regex used for doc preprocessing

  #36648 merged Mar 14, 2025

• Make the flaky list a little more general

  #36704 merged Mar 14, 2025

• Gemma3 processor typo

  #36710 merged Mar 14, 2025

• Add support for fast image processors in add-new-model-like CLI

  #36313 merged Mar 13, 2025

• Final CI cleanup

  #36703 merged Mar 13, 2025

• Add GGUF support to T5-Encoder

  #36700 merged Mar 13, 2025

• Handling an exception related to HQQ quantization in modeling

  #36702 merged Mar 13, 2025

• fix: fsdp sharded state dict wont work for save_only_model knob

  #36627 merged Mar 13, 2025

• Add loading speed test

  #36671 merged Mar 13, 2025

• [CI] Automatic rerun of certain test failures

  #36694 merged Mar 13, 2025

• chore: fix typos in utils module

  #36668 merged Mar 13, 2025

• Fix dtype for params without tp_plan

  #36681 merged Mar 13, 2025

• fix type annotation for ALL_ATTENTION_FUNCTIONS

  #36690 merged Mar 13, 2025

• Change Qwen2_VL image processors to have init and call accept the same kwargs

  #36207 merged Mar 13, 2025

• Upgrading torch version and cuda version in quantization docker

  #36264 merged Mar 13, 2025

• fix wandb hp search unable to resume from sweep_id

  #35883 merged Mar 13, 2025

• Changing the test model in Quanto kv cache

  #36670 merged Mar 13, 2025

• Fix slicing for 0-dim param

  #36580 merged Mar 13, 2025

• Update config.torch_dtype correctly

  #36679 merged Mar 13, 2025

• [Cache] Don't initialize the cache on meta device

  #36543 merged Mar 13, 2025

• Fix rescale normalize inconsistencies in fast image processors

  #36388 merged Mar 13, 2025

• Refactor siglip2 fast image processor

  #36406 merged Mar 13, 2025

• Remove differences between init and preprocess kwargs for fast image processors

  #36186 merged Mar 12, 2025

• [quants] refactor logic for modules_to_not_convert

  #36672 merged Mar 12, 2025

• Remove hardcoded slow image processor class in processors supporting fast ones

  #36266 merged Mar 12, 2025

• Fix Failing GPTQ tests

  #36666 merged Mar 12, 2025

• Don't accidentally mutate the base_model_tp_plan

  #36677 merged Mar 12, 2025

• [core] Large/full refactor of from_pretrained

  #36033 merged Mar 12, 2025

• Fix bnb regression due to empty state dict

  #36663 merged Mar 12, 2025

• [CI] gemma 3 make fix-copies

  #36664 merged Mar 12, 2025

• fix block mask typing

  #36661 merged Mar 12, 2025

• HPU support

  #36424 merged Mar 12, 2025

• Gemma3

  #36658 merged Mar 12, 2025

• fix typos in the docs directory

  #36639 merged Mar 11, 2025

• Fix gguf docs

  #36601 merged Mar 11, 2025

• Remove research projects

  #36645 merged Mar 11, 2025

• [docs] Update docs dependency

  #36635 merged Mar 11, 2025

• Stop warnings from unnecessary torch.tensor() overuse

  #36538 merged Mar 11, 2025

• Remove remote code warning

  #36285 merged Mar 11, 2025

• Fix AriaForConditionalGeneration flex attn test

  #36604 merged Mar 11, 2025

• Proper_flex

  #36643 merged Mar 11, 2025

• Fix bugs in mllama image processing

  #36156 merged Mar 11, 2025

• Refactor some core stuff

  #36539 merged Mar 11, 2025

• [docs] Serving LLMs

  #36522 merged Mar 10, 2025

• chore: fix typos in language models

  #36586 merged Mar 10, 2025

• Fix auto-assign reviewers

  #36631 merged Mar 10, 2025

• [HybridCache] disable automatic compilation

  #36620 merged Mar 10, 2025

• Fix check for XPU. PyTorch >= 2.6 no longer needs ipex.

  #36593 merged Mar 7, 2025

• Fixed datatype related issues in DataCollatorForLanguageModeling

  #36457 merged Mar 7, 2025

• Bump jinja2 from 3.1.5 to 3.1.6 in /examples/research_projects/decision_transformer

  #36582 merged Mar 7, 2025

• Update "who to tag" / "who can review"

  #36394 merged Mar 7, 2025

• Update chat_extras.md with content correction

  #36599 merged Mar 7, 2025

• Github action for auto-assigning reviewers

  #35846 merged Mar 7, 2025

• Export base streamer.

  #36500 merged Mar 7, 2025

• avoid errors when the size of input_ids passed to PrefixConstrainedLogitsProcessor is zero

  #36489 merged Mar 7, 2025

• Mention UltraScale Playbook 🌌 in docs

  #36589 merged Mar 6, 2025

• fix: argument

  #36558 merged Mar 6, 2025

• [XGLM] tag tests as slow

  #36592 merged Mar 6, 2025

• [bark] fix loading of generation config

  #36587 merged Mar 6, 2025

• Integrate SwanLab for offline/online experiment tracking and local visualization

  #36433 merged Mar 6, 2025

• Modular Conversion --fix_and_overwrite on Windows

  #36583 merged Mar 6, 2025

• Delete redundancy if case in model_utils

  #36559 merged Mar 6, 2025

• Bump transformers from 4.38.0 to 4.48.0 in /examples/research_projects/pplm

  #36540 merged Mar 6, 2025

• chore: enhance message descriptions in parameters,comments,logs and docstrings

  #36554 merged Mar 6, 2025

• Fix typos .

  #36551 merged Mar 6, 2025

• Fix typos in tests

  #36547 merged Mar 5, 2025

• guard torch version for uint16

  #36520 merged Mar 5, 2025

• chore: enhance messages in docstrings

  #36525 merged Mar 4, 2025

• Fix links in quantization doc

  #36528 merged Mar 4, 2025

• Fix bamba tests amd

  #36535 merged Mar 4, 2025

• chore: Fix typos in docs and examples

  #36524 merged Mar 4, 2025

• Add aya

  #36521 merged Mar 4, 2025

• [docs] Redesign

  #31757 merged Mar 3, 2025

• Remove unused code

  #36459 merged Mar 3, 2025

• [Style] fix E721 warnings

  #36474 merged Mar 3, 2025

• Fix edge case for continue_final_message

  #36404 merged Mar 3, 2025

• Fix pipeline+peft interaction

  #36480 merged Mar 3, 2025

• chore: fix message descriptions in arguments and comments

  #36504 merged Mar 3, 2025

• Fix some typos in docs

  #36502 merged Mar 3, 2025

• fix torch_dtype, contiguous, and load_state_dict regression

  #36512 merged Mar 3, 2025

• Fix kwargs UserWarning in SamImageProcessor

  #36479 merged Mar 3, 2025

• Check TRUST_REMOTE_CODE for RealmRetriever for security

  #36511 merged Mar 3, 2025

• Fix loading zero3 weights

  #36455 merged Mar 3, 2025

• Fix _load_state_dict_into_meta_model with device_map=None

  #36488 merged Mar 2, 2025

• Fix couples of issues from #36335

  #36453 merged Mar 1, 2025

• Add Got-OCR 2 Fast image processor and refactor slow one

  #36185 merged Mar 1, 2025

• [docs] fix bug in deepspeed config

  #36081 merged Feb 28, 2025

• Fix loading models with mismatched sizes

  #36463 merged Feb 28, 2025

• [GroundingDino] Fix grounding dino loss 🚨

  #31828 merged Feb 27, 2025

• Fix hub_retry

  #36449 merged Feb 27, 2025

• Lazy import libraries in src/transformers/image_utils.py

  #36435 merged Feb 27, 2025

• [generate] torch.distributed-compatible DynamicCache

  #36373 merged Feb 27, 2025

• [save_pretrained ] Skip collecting duplicated weight

  #36409 merged Feb 27, 2025

• Add contents: write

  #36445 merged Feb 27, 2025

• Fix another permission

  #36444 merged Feb 27, 2025

• Fix permission

  #36443 merged Feb 27, 2025

• Change PR to draft when it is (re)opened

  #36417 merged Feb 27, 2025

• restrict cache allocator to non quantized model

  #36428 merged Feb 26, 2025

• Fix Expected output for compressed-tensors tests

  #36425 merged Feb 26, 2025

• Update form pretrained to make TP a first class citizen

  #36335 merged Feb 26, 2025

129 Pull requests opened by 93 people

• Add fetch_paginated_github_data to deduplicate GitHub API pagination …

  #36432 opened Feb 26, 2025

• Fix model saving bug post training with tensor parallel in Accelerate

  #36434 opened Feb 26, 2025

• Add PlainDETR

  #36437 opened Feb 26, 2025

• add FlashAttentionKwargs and seq_idx to flat collator

  #36456 opened Feb 27, 2025

• Customize docstrings fast image processor

  #36466 opened Feb 27, 2025

• Add NVIDIA Cosmos

  #36476 opened Feb 28, 2025

• Fix incorrect attention mask truncate in WhisperFlashAttention2

  #36477 opened Feb 28, 2025

• Sanitize Model Module Names to Follow Python Conventions

  #36478 opened Feb 28, 2025

• Export T5 (encoder-decoder) to ExecuTorch

  #36486 opened Mar 1, 2025

• Allow OOV Image Token for LLaVa Next Variants

  #36491 opened Mar 2, 2025

• Create and Expose SamVisionModel as public for better accessibility

  #36493 opened Mar 2, 2025

• Add an event related to forward in the TrainerCallback

  #36496 opened Mar 2, 2025

• Refactor object-detection models

  #36514 opened Mar 3, 2025

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

  #36534 opened Mar 4, 2025

• add-long-vita

  #36553 opened Mar 5, 2025

• Fix edge case for tokenize (#36277)

  #36555 opened Mar 5, 2025

• add-LongVITAModel

  #36556 opened Mar 5, 2025

• fix for loading gguf quantized model

  #36563 opened Mar 5, 2025

• Allow saving and loading multiple "raw" chat template files

  #36588 opened Mar 6, 2025

• Attention mechanisms elaboration

  #36597 opened Mar 6, 2025

• [audio utils] fix fft_bin_width computation

  #36603 opened Mar 7, 2025

• Add StableAdamW Optimizer

  #36606 opened Mar 7, 2025

• Fixed 30s timestamp resets in Whisper long-form transcription

  #36612 opened Mar 7, 2025

• Add Distill Any Depth

  #36614 opened Mar 8, 2025

• [WIP] Add support to load models with transforms

  #36621 opened Mar 9, 2025

• [WiP] Add Aimv2 model

  #36625 opened Mar 10, 2025

• [Whisper] 🚨 Fix pipeline word timestamp: timestamp token is end of token time !!!

  #36632 opened Mar 10, 2025

• Refine parameter type annotations

  #36644 opened Mar 11, 2025

• Fix device issue in modeling_qwen2

  #36647 opened Mar 11, 2025

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

  #36649 opened Mar 11, 2025

• Fixes DynamicCache export issues due to control flow and inplace modifications

  #36652 opened Mar 11, 2025

• Update quantizer_bnb_4bit.py

  #36669 opened Mar 12, 2025

• don't pass NoneType for keep_in_fp32_modules

  #36675 opened Mar 12, 2025

• Support batch size > 1 image-text inference

  #36682 opened Mar 12, 2025

• prune LM Head for USD

  #36695 opened Mar 13, 2025

• [Feature] Support using FlashAttention2 on Ascend NPU

  #36696 opened Mar 13, 2025

• Limit numpy version to <2.0.0

  #36706 opened Mar 13, 2025

• Fix long lagging when streaming text without spaces and NJK chars

  #36708 opened Mar 13, 2025

• [WIP] Add DINO DETR Model to HuggingFace Transformers

  #36711 opened Mar 14, 2025

• fix whisper re-compile

  #36712 opened Mar 14, 2025

• Add CSM model

  #36719 opened Mar 14, 2025

• Fix generation using flash-attention and static cache

  #36729 opened Mar 14, 2025

• Fix image processor speedup fixed

  #36732 opened Mar 14, 2025

• Convert _VALID_DICT_FIELDS to class attribute for shared dict parsing in subclasses

  #36736 opened Mar 15, 2025

• [WP] PagedAttention + Prefix Cache for FlashAttention2

  #36737 opened Mar 15, 2025

• 🌐 [i18n-KO] Translated `qwen2_vl.md` to Korean

  #36750 opened Mar 16, 2025

• Add Qwen2.5-Omni

  #36752 opened Mar 16, 2025

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

  #36756 opened Mar 17, 2025

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

  #36757 opened Mar 17, 2025

• feat: expose the strict flag to allow catching missing model layers while loading a checkpoint

  #36760 opened Mar 17, 2025

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

  #36763 opened Mar 17, 2025

• Add support for audios in apply_chat_template

  #36770 opened Mar 17, 2025

• Use public export API on torch 2.5 and future

  #36781 opened Mar 18, 2025

• Fix attention_mask dimension issue in GPT2Model

  #36782 opened Mar 18, 2025

• Create modeling_ngen3.py for NGen3

  #36787 opened Mar 18, 2025

• Update configuration_auto.py for NGen3

  #36791 opened Mar 18, 2025

• Refactor `return_dict` logic to remove complicated if/else paths

  #36794 opened Mar 18, 2025

• [don't merge] check tokenizer ci job

  #36796 opened Mar 18, 2025

• Add Granite Speech Support

  #36801 opened Mar 18, 2025

• Add long vita

  #36807 opened Mar 19, 2025

• Support loading custom models (`trust_remote_code=True`) in offline mode from local

  #36808 opened Mar 19, 2025

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

  #36809 opened Mar 19, 2025

• Use `lru_cache` for tokenization tests

  #36818 opened Mar 19, 2025

• Dummies

  #36827 opened Mar 19, 2025

• [Modeling] Load FP8 safetensors such as DeepSeek

  #36828 opened Mar 19, 2025

• gemma3 fp16 fix

  #36832 opened Mar 19, 2025

• Enhance Model Loading By Providing Parallelism, Uses Optional Env Flag

  #36835 opened Mar 19, 2025

• Remove unnecessary attr assignment

  #36837 opened Mar 19, 2025

• Move `return_dict` logic into `can_return_tuple` decorator

  #36838 opened Mar 19, 2025

• Haocheng lu

  #36839 opened Mar 19, 2025

• fix pegasus init weights and other copied models

  #36844 opened Mar 20, 2025

• Add support for specifying revisions when pushing to Hub via internal Trainer call

  #36852 opened Mar 20, 2025

• fix: prevent input side-effects in processor text args

  #36866 opened Mar 20, 2025

• Only count num items in batch when needed

  #36867 opened Mar 20, 2025

• Fix warning message for PEFT models in text-generation pipeline #36783

  #36868 opened Mar 20, 2025

• Improve Model Download Speeds By ~3x For Large Models

  #36870 opened Mar 21, 2025

• Adding Qwen3 and Qwen3MoE

  #36878 opened Mar 21, 2025

• Fix `resume_from_checkpoint` not recognising `"last-checkpoint"`

  #36883 opened Mar 21, 2025

• Optimize `to_py_obj` for python-native numeric lists and scalars

  #36885 opened Mar 21, 2025

• Fix warning message for PEFT models in text-generation pipeline #36783

  #36887 opened Mar 21, 2025

• Fix SDPA implementation in Qwen2-VL (issues with torch==2.6.0)

  #36891 opened Mar 21, 2025

• [WIP] Computer vision util: vision visualizer

  #36892 opened Mar 21, 2025

• Enable tracing for Moshi

  #36894 opened Mar 21, 2025

• Add RF-DETR

  #36895 opened Mar 21, 2025

• Adding ArlowGPT

  #36899 opened Mar 22, 2025

• Add NGen3

  #36901 opened Mar 22, 2025

• LogfireCallback: Integrating Logfire with Hugging Face’s Trainer

  #36905 opened Mar 22, 2025

• fix cached file error when repo type is dataset

  #36909 opened Mar 23, 2025

• Limit number of evaluation samples processed during training

  #36916 opened Mar 24, 2025

• [qwen2-audio] remove default template

  #36919 opened Mar 24, 2025

• Allow disabling `deformable_detr` kernels

  #36927 opened Mar 24, 2025

• Remove the redundant shift during the loss computation in the Moshi m…

  #36928 opened Mar 24, 2025

• Aligning modling code for GPT2 to work with vLLM (fallback)

  #36934 opened Mar 24, 2025

• [3/N] Use pyupgrade --py39-plus to improve code

  #36936 opened Mar 24, 2025

• Static cache should support indexing

  #36943 opened Mar 24, 2025

• Improve typing in TrainingArgument

  #36944 opened Mar 25, 2025

• fix(qwen): fix shape error when using tp

  #36947 opened Mar 25, 2025

• Update image_processing_qwen2_vl.py。fix bug.

  #36948 opened Mar 25, 2025

• Avoid unnecessary tensor copy in loss computing

  #36950 opened Mar 25, 2025

• Added Sapnous Architecture

  #36952 opened Mar 25, 2025

• Skip code `307` in `RequestCounter`

  #36953 opened Mar 25, 2025

• [chat templates} support loading audio from video

  #36955 opened Mar 25, 2025

• fix: Fully remove legacy cache from Llama

  #36958 opened Mar 25, 2025

• Remove low_cpu_mem_usage and _fast_init

  #36963 opened Mar 25, 2025

• More ReDOS fixes!

  #36964 opened Mar 25, 2025

• [phi-4] use mel filters from audio utils

  #36966 opened Mar 25, 2025

• Add new dim to `num_items_in_batch` if necessary

  #36967 opened Mar 25, 2025

• Make executorch integration more seamless by analyzing model signature

  #36969 opened Mar 25, 2025

• Refactor image processor phi4

  #36976 opened Mar 25, 2025

• Add device workaround for int4 weight only quantization after API update

  #36980 opened Mar 25, 2025

• Refactor attention for SigLIP based models

  #36981 opened Mar 25, 2025

• clean pipeline question_answering.

  #36986 opened Mar 26, 2025

• fix comment misdirection during scaling loss

  #36987 opened Mar 26, 2025

• fix transformers_cli import relative path issue

  #36989 opened Mar 26, 2025

• Gaudi: Fix the pipeline failed issue with hpu device

  #36990 opened Mar 26, 2025

• Set weights_only in torch.load

  #36991 opened Mar 26, 2025

• fix and enhance pipeline_webserver.md

  #36992 opened Mar 26, 2025

• remove redundant code in trainer

  #36994 opened Mar 26, 2025

• [Phi4] add multimodal chat template

  #36996 opened Mar 26, 2025

• Add Fast SamImageProcessor

  #36999 opened Mar 26, 2025

• Replace default split function with jnp.split() in flax models

  #37001 opened Mar 26, 2025

• Fix typing for None valued variables

  #37004 opened Mar 26, 2025

• [Fast Processor] BEiT

  #37005 opened Mar 26, 2025

• Remove deprecated batch_size argument

  #37007 opened Mar 26, 2025

• Skip FP8 linear tests

  #37008 opened Mar 26, 2025

• Export Whisper to ExecuTorch

  #37009 opened Mar 26, 2025

• Fix AttentionInterface following feedback

  #37010 opened Mar 26, 2025

• Add Fast Chinese-CLIP Processor

  #37012 opened Mar 26, 2025

• [generate, cache] handle more complex device maps

  #37014 opened Mar 26, 2025

162 Issues closed by 47 people

• Learning rate logging off by one training step

  #35942 closed Mar 26, 2025

• bitsandbytes integration bug due to trying to alter frozenset in `_validate_bnb_multi_backend_availability()`

  #36949 closed Mar 26, 2025

• ValueError: `run_compressed` is only supported for quantized_compressed models

  #36915 closed Mar 26, 2025

• Recent update: configuration_eurobert.py not found -

  #36983 closed Mar 26, 2025

• Issue with Progressive Generation Using inputs_embeds and past_key_values

  #35707 closed Mar 26, 2025

• RWKV CUDA error: an illegal memory access was encountered during training from scratch

  #35805 closed Mar 26, 2025

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

  #36183 closed Mar 26, 2025

• Token healing throws error with "Qwen/Qwen2.5-Coder-7B-Instruct"

  #36210 closed Mar 26, 2025

• [bug] use_gather_object is not respected after the first eval in trainer

  #36213 closed Mar 26, 2025

• Error: TypeError: argument 'ids': 'float' object cannot be interpreted as an integer

  #36984 closed Mar 26, 2025

• Clarification on Commercial License Impact of LayoutLMv3ImageProcessor within UdopProcessor

  #36931 closed Mar 25, 2025

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

  #36954 closed Mar 25, 2025

• AutoTokenizer/Processor does not work with Mistral3 models

  #36968 closed Mar 25, 2025

• Ruff update

  #36705 closed Mar 25, 2025

• torchrun breaks with load_model_at_end and with metric_for_best_model=eval_f1 on question_answering example

  #30819 closed Mar 25, 2025

• `Mllama` not supported by `AutoModelForCausalLM` after updating `transformers` to `4.50.0`

  #36926 closed Mar 25, 2025

• Florence2 stopped working after upgrade to 4.50.0 ("Unrecognized configuration class")

  #36886 closed Mar 25, 2025

• Design question for integrating new model to Transformers?

  #36784 closed Mar 25, 2025

• Add seed to data collator classes

  #36655 closed Mar 24, 2025

• Torch -> ONNX doesn't work after upgrading transformers to 4.49.0

  #36276 closed Mar 24, 2025

• <spam>

  #36924 closed Mar 24, 2025

• llama tokenizer encode -> decode is not same

  #36325 closed Mar 24, 2025

• tj-actions/changed-files action compromised

  #36761 closed Mar 24, 2025

• Some of test/utils tests fail being invalidated by tests/utils/test_import_utils.py::test_clear_import_cache

  #36334 closed Mar 24, 2025

• MacOs: register_pytree_node got an unexpected keyword argument 'flatten_with_keys_fn'

  #36906 closed Mar 24, 2025

• Issue with update

  #36888 closed Mar 24, 2025

• Gemma2: RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:7 and cuda:0! (when checking argument for argument index in method wrapper_CUDA__index_select)

  #34706 closed Mar 24, 2025

• Trainer: TensorBoardCallback not working for "on_save" and "on_save_end" events

  #35612 closed Mar 24, 2025

• Pipeline cannot guess which processor to use with Gemma 3

  #36911 closed Mar 23, 2025

• Unable to export GLM models to ONNX

  #35021 closed Mar 23, 2025

• ModernBERT inference fails on CPU: ValueError: Pointer argument (at 0) cannot be accessed from Triton (cpu tensor?)

  #35388 closed Mar 23, 2025

• `modular_model_converter` can not handle objects import via try - except

  #35414 closed Mar 23, 2025

• `TFViTModel` and `interpolate_pos_encoding=True`

  #36155 closed Mar 23, 2025

• [BART] Cannot copy out of meta tensor; no data!

  #36247 closed Mar 21, 2025

• Bug introduced in `from_pretrained` `v4.48.3`..`v4.49.0`

  #36258 closed Mar 21, 2025

• <spam>

  #36876 closed Mar 21, 2025

• torch._subclasses.fake_tensor.DataDependentOutputException: aten._local_scalar_dense.default with `_prepare_4d_attention_mask_for_sdpa(

  #36123 closed Mar 21, 2025

• Speaker Verification: All Speakers Getting Perfect 1.000 Similarity Scores

  #36124 closed Mar 21, 2025

• Allow setting a seed for DataCollatorForLanguageModeling

  #36357 closed Mar 20, 2025

• LlamaAttention has no attribute `rotary_emb` (4.50.0.dev0)

  #36758 closed Mar 20, 2025

• GPT2 repetition of words in output

  #36848 closed Mar 20, 2025

• num_items_in_batch unexpected in vision encoder decoder

  #36744 closed Mar 20, 2025

• Convert RT-DETR model to coreml

  #35905 closed Mar 20, 2025

• [bug] fast_image_processor register error

  #36715 closed Mar 19, 2025

• When what needs to be loaded is in the cache directory, there is no need to make a request to the remote

  #36762 closed Mar 19, 2025

• In the _speculative_sampling function, it seems that the "squeeze" method is being used incorrectly.

  #36810 closed Mar 19, 2025

• AttributeError: 'Gemma3Config' object has no attribute 'vocab_size'

  #36683 closed Mar 19, 2025

• text-to-video_app

  #36747 closed Mar 19, 2025

• model from_pretrained bug in 4.50.dev0 in these days

  #36506 closed Mar 19, 2025

• Subtle difference with Pytorch AdamW?

  #35504 closed Mar 19, 2025

• Qwen2VL exhibits significant performance differences under different attention implementations.

  #35749 closed Mar 19, 2025

• [Phi-3-mini-128k-instruct] Difference of encodings for Slow and Fast Tokenizer

  #35973 closed Mar 19, 2025

• Traning loss not showing with trainer

  #36102 closed Mar 19, 2025

• when model.generate with num_beams=2 and num_return_sequences=2,the output seqs are different from input_ids of stopping_criteria

  #34574 closed Mar 18, 2025

• `UnboundLocalError: cannot access local variable 'images_list'` when using Gemma 3 AutoProcessor with use_fast=True

  #36739 closed Mar 18, 2025

• Gemma3 minimal fine tuning example?

  #36714 closed Mar 18, 2025

• Shape mismatch in RoPE embeddings gpt_neox model when rotary_ndims is odd

  #35233 closed Mar 18, 2025

• incorrect special_tokens_mask

  #35897 closed Mar 18, 2025

• Llama tokenizer newline character inconsistency

  #35923 closed Mar 18, 2025

• flex_attention does not output the full attention_weights with output_attention option

  #36096 closed Mar 18, 2025

• bug in save checkpoint

  #36099 closed Mar 18, 2025

• qwen2_5_vl processor padding side is wrong.

  #36100 closed Mar 18, 2025

• ValueError: weight is on the meta device, we need a `value` to put in on 0. `Gemma3`

  #36766 closed Mar 17, 2025

• Misleading documentation for `is_decoder` configuration parameter

  #36482 closed Mar 17, 2025

• Gemma2 (quantized) inference is broken - torch._dynamo.exc.UserError: Dynamic control flow is not supported at the moment.

  #36485 closed Mar 17, 2025

• On MoE implementation in HuggingFace

  #36730 closed Mar 17, 2025

• bus error on version 4.43.0 with pretrained community CLIP model - MacOS

  #33357 closed Mar 17, 2025

• Cannot load siglip2 processor

  #36665 closed Mar 16, 2025

• SFTConfig.__init__() got an unexpected keyword argument 'optimizers'

  #36749 closed Mar 16, 2025

• Model.generate use_cache=True generates different results than use_cache=False

  #36536 closed Mar 16, 2025

• past_key_values type support bug

  #36057 closed Mar 16, 2025

• TypeError: empty() missing 1 required positional arguments: "size"

  #36061 closed Mar 16, 2025

• Transformers can create unconventional python module names when loading certain repositories

  #35570 closed Mar 15, 2025

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

  #36010 closed Mar 15, 2025

• [feature request] Callback handler event after forward pass in Trainer

  #36012 closed Mar 15, 2025

• AMD CI tracking issue

  #36019 closed Mar 15, 2025

• T5 Tokenzier not load with `AttributeError: add_special_tokens conflicts with the method add_special_tokens in T5Tokenizer`

  #36032 closed Mar 15, 2025

• Issue in resuming finetuning Llama 3.1 Instruct Model

  #36035 closed Mar 15, 2025

• Initializing via AutoImageProcessor before AutoProcessor is imported causes `AttributeError`

  #34307 closed Mar 14, 2025

• Trainer sets `state.best_model_checkpoint` even when it doesn't save there; leads to training crash

  #35609 closed Mar 14, 2025

• ValueError: The checkpoint you are trying to load has model type `gemma3` but Transformers does not recognize this architecture.

  #36709 closed Mar 14, 2025

• 'MERTConfig' object has no attribute 'conv_pos_batch_norm'

  #36134 closed Mar 14, 2025

• Some questions of `Gemma3` processor

  #36701 closed Mar 14, 2025

• NotImplementedError: aten::_log_softmax_backward_data with SparseCUDA backend

  #36674 closed Mar 14, 2025

• Component loading incorrect dtype

  #36686 closed Mar 13, 2025

• `disable_compile` not honored as a kwarg in generate

  #36544 closed Mar 13, 2025

• AutoModel failed with empty tensor error

  #36579 closed Mar 13, 2025

• Some methods in TrainerControl seem not to be utilized.

  #36576 closed Mar 13, 2025

• save_only_model with FSDP throws FileNotFoundError error

  #36626 closed Mar 13, 2025

• Cannot import 'GenerationOutput' in 4.48.1

  #35957 closed Mar 13, 2025

• GPTQ quantization on Jetson Orin Nano

  #36139 closed Mar 12, 2025

• past_key_values not being set in model_inputs keys

  #36001 closed Mar 12, 2025

• The number of safetensors files is different when using CPU and CUDA.

  #36595 closed Mar 11, 2025

• Downloading models in distributed training

  #36414 closed Mar 11, 2025

• Warning related to torch.tensor() usage in transformers.models.encodec.modeling_encodec.py (Version 4.47.0)

  #36533 closed Mar 11, 2025

• Loading a pipeline with `trust_remote_code=True` raises warning

  #36273 closed Mar 11, 2025

• Extract embeddings of many seqs using ESM2

  #36641 closed Mar 11, 2025

• Error faced during Finetuning Deepseek-vl2

  #36633 closed Mar 11, 2025

• paligemma2-3B-mix in version4.49.0 not use GPU and 4.50.0.dev broken

  #36575 closed Mar 11, 2025

• model.parameters() return [Parameter containing: tensor([], device='cuda:0', dtype=torch.bfloat16, requires_grad=True)] when using zero3

  #35994 closed Mar 11, 2025

• Inconsistent Outputs When Using Flash Attention 2 and SDPA Attention with Attention Mask

  #36585 closed Mar 11, 2025

• AttributeError: 'MERTConfig' object has no attribute 'conv_pos_batch_norm'

  #35656 closed Mar 10, 2025

• Why are there so many variables named layrnorm in the codebase?

  #36623 closed Mar 10, 2025

• Memory Access out of bounds in mra/cuda_kernel.cu::index_max_cuda_kernel()

  #35507 closed Mar 10, 2025

• Very slow to load deep seekv3 int4 model and device_map="auto" "sequential" bug

  #35522 closed Mar 10, 2025

• adalomo and deepspeed zero3 offload error

  #35977 closed Mar 10, 2025

• size mismatch for lm_head when fintune QWEN2.5

  #36550 closed Mar 10, 2025

• Llama3 tokenizer decode is incorrect for ' ...' with leading space

  #36622 closed Mar 9, 2025

• Tokenizer does not split text according to newly added input tokens

  #35447 closed Mar 9, 2025

• Can't use Trainer on mps device

  #35954 closed Mar 9, 2025

• Some weights of the model checkpoint at /models/DeepSeek-V3_bf16 were not used when initializing DeepseekV3ForCausalLM

  #35688 closed Mar 8, 2025

• Significant Increase in Computation Time When Using Attention Mask in SDPA Attention

  #36584 closed Mar 8, 2025

• Accidentally allocating 2x memory in new caching_allocator_warmup

  #36483 closed Mar 7, 2025

• Open Object Detection Leaderboard: Model Requests not working

  #36034 closed Mar 7, 2025

• TypeError: LlavaProcessor: got multiple values for keyword argument 'images'

  #36578 closed Mar 7, 2025

• Attention can be None in ModernBertForSequenceClassification

  #35917 closed Mar 7, 2025

• meta-llama/Llama-3.2-11B-Vision-Instruct, device_map = 'auto', padding ruins _prepare_4d_causal_attention_mask_with_cache_position

  #35918 closed Mar 7, 2025

• Lora_B weight becomes 0 when using AuotModel

  #36594 closed Mar 6, 2025

• Do trailing padding tokens get a forward pass?

  #36565 closed Mar 6, 2025

• Init on meta device and then materialize on gpu leads to very large errors

  #36577 closed Mar 6, 2025

• how to use transformers with musicgen with float16

  #36546 closed Mar 6, 2025

• safetensors_rust.SafetensorError: Error while serializing: IoError(Os { code: 5, kind: Uncategorized, message: "Input/output error" })

  #35895 closed Mar 6, 2025

• AttributeError: 'dict' object has no attribute '_attn_implementation_internal'

  #35900 closed Mar 6, 2025

• Dtensor support requires torch>=2.5.1

  #36472 closed Mar 5, 2025

• Groq inference provider

  #36353 closed Mar 4, 2025

• After tokenizers upgrade, the length of the token does not correspond to the length of the model

  #36532 closed Mar 4, 2025

• Incorrect Whisper long-form decoding timestamps

  #31942 closed Mar 4, 2025

• Help Understanding Beam Search Scores in Hugging Face (LLaMA + LoRA)

  #35618 closed Mar 4, 2025

• ERROR: Video features and Video Tokens do not match!!!

  #35869 closed Mar 4, 2025

• tokenizers.apply_chat_template with continue_final_message=True with </think> token

  #36440 closed Mar 3, 2025

• tokenizers.apply_chat_template with `continue_final_message=True` with trailing spaces in input

  #35433 closed Mar 3, 2025

• Confusing behavior when loading PEFT models with pipeline

  #36473 closed Mar 3, 2025

• `_load_state_dict_into_meta_model` - `'NoneType' object has no attribute 'load_state_dict'`

  #36495 closed Mar 3, 2025

• Failed to import transformers.trainer because of the following error (look up to see its traceback): cannot import name 'add_model_info_to_custom_pipelines' from 'transformers.utils'

  #36503 closed Mar 3, 2025

• GRPO Reward Weight Scheduler

  #36490 closed Mar 3, 2025

• please support register_full_backward_pre_hook and register_full_backward_hook

  #36507 closed Mar 3, 2025

• Some Whisper beam search output (sequences_scores, etc.) is lost in _stack_split_outputs

  #32373 closed Mar 3, 2025

• [DEV Testing] Issues with `test_modeling_common`

  #35857 closed Mar 3, 2025

• [BUG]npu zero3 训练自定义模型时，报错Function SumBackward0 returned an invalid gradient at index 0

  #36387 closed Mar 3, 2025

• Load siglip2 error

  #36475 closed Mar 3, 2025

• `padding_side` is of type `bool` when it should be `Literal['right', 'left']`

  #36252 closed Mar 3, 2025

• Add Wan model into Transformers

  #36494 closed Mar 2, 2025

• Bug introduced in `_load_state_dict_into_meta_model` and `to` `v4.49.0`..`v4.50.0.dev`

  #36441 closed Mar 1, 2025

• `model.config.to_diff_dict()` delivers different result to `model.save_pretrained()`

  #35426 closed Mar 1, 2025

• AttributeError: 'Config' object has no attribute '_get_non_default_generation_parameters'

  #35543 closed Mar 1, 2025

• Prompt_ids feature causing repetitions and hallucinations

  #35603 closed Mar 1, 2025

• convert_llama_weight_to_hf.py

  #35820 closed Mar 1, 2025

• suppress_tokens=[] should be legal as some older. whisper models rely on this

  #36341 closed Feb 28, 2025

• model.generate() produces different outputs with padding for flan-t5-small

  #36461 closed Feb 28, 2025

• Failed to import transformers.models.auto.modeling_auto because numpy.core.multiarray failed to import

  #36343 closed Feb 28, 2025

• KerasTensor can't be used with TFBertTokenizer

  #36462 closed Feb 28, 2025

• BUG: Failed to import transformers.models.auto.modeling_auto due to No module named 'scipy.optimize._highspy._core.simplex_constants';

  #36468 closed Feb 28, 2025

• about siglip2

  #36470 closed Feb 28, 2025

• AutoModelForObjectDetection isnt working due to wrong output size

  #36464 closed Feb 28, 2025

• Question for community: We're considering adding `pydantic` as a base requirement to 🤗 `transformers`

  #36329 closed Feb 28, 2025

• How to change data

  #35807 closed Feb 28, 2025

• test

  #36471 closed Feb 28, 2025

• Error splitting the input into NAL units.

  #36448 closed Feb 27, 2025

• KTransformers跑DeepSeek-R1量化版，支持多并发吗？多并发要的资源是不是也成倍上涨？

  #36423 closed Feb 27, 2025

• Mamba2 doesn't support Multi-GPU training (fast path)

  #35770 closed Feb 27, 2025

• TPU Initialization Error with Transformers in Kaggle TPU VM v3-8

  #35774 closed Feb 27, 2025

• Apply dualpipe from deepseek-v3 to a trainer or model

  #36439 closed Feb 27, 2025

116 Issues opened by 112 people

• Gemma3 adding new tokens <image_soft_token> has been added accidentally

  #37011 opened Mar 26, 2025

• [Question] Handling of custom flex attention block masks

  #37006 opened Mar 26, 2025

• GGUF model with architecture gemma3 is not supported yet.

  #37002 opened Mar 26, 2025

• Add ArlowGPT

  #36988 opened Mar 26, 2025

• FSDP Not Working For Mamba2

  #36982 opened Mar 25, 2025

• [Community contributions] Model cards

  #36979 opened Mar 25, 2025

• [Contributions Welcome] Add Fast Image Processors

  #36978 opened Mar 25, 2025

• QuestionAnswering for Gemma 3

  #36977 opened Mar 25, 2025

• Gemma3: Cuda error: misaligned address

  #36961 opened Mar 25, 2025

• Incorrect size mismatch skipping in `_find_mismatched_keys` causes model loading failures despite `ignore_mismatched_sizes=True`

  #36960 opened Mar 25, 2025

• Symbolic trance with past_key_values input is not supported yet for the qwen2.

  #36959 opened Mar 25, 2025

• Started getting new warnings for gemma3 after upgrading from 4.49.0-gemma3 to 4.50.0

  #36942 opened Mar 24, 2025

• Add param_to_hook_all_reduce parameter in HF Trainer

  #36941 opened Mar 24, 2025

• Gemma3 not supported in main branch

  #36940 opened Mar 24, 2025

• AttributeError: 'HybridCache' object has no attribute 'float' — PaliGemma2 Evaluation Fails with BF16

  #36938 opened Mar 24, 2025

• python_interpreter.py seems not support asyncio.run()

  #36920 opened Mar 24, 2025

• 'Cache only has 0 layers' during generation after upgrading Transformers from 4.49 to 4.50

  #36913 opened Mar 24, 2025

• `lm_head.weight` missing from `convert_mistral_weights_to_hf.STATE_DICT_MAPPING`

  #36908 opened Mar 23, 2025

• PixtralVisionModel does not support Flash Attention 2.0 yet

  #36904 opened Mar 22, 2025

• Warning: "No label_names provided for PeftModel" persists despite dataset containing "labels" column

  #36902 opened Mar 22, 2025

• groot n1

  #36900 opened Mar 22, 2025

• GPT2Model model output inconsistency between different transformers versions

  #36897 opened Mar 22, 2025

• Forced to hit `UserWarning` when generating with `temperature=0`

  #36896 opened Mar 21, 2025

• Add RF-DETR model

  #36879 opened Mar 21, 2025

• Qwen2-VL-7B-Instruct shape error when using TP=4

  #36875 opened Mar 21, 2025

• Support for SpatialLM series model

  #36874 opened Mar 21, 2025

• Optimize tokenizer.decode() Performance for `List[int]` Inputs

  #36872 opened Mar 21, 2025

• Multiple processor classes have input side-effects

  #36865 opened Mar 20, 2025

• Gemma3 (and Paligemma) position_ids 1-indexed?

  #36856 opened Mar 20, 2025

• Facing RunTime Attribute error while running different Flax models for RoFormer

  #36854 opened Mar 20, 2025

• Tansfomers_model

  #36846 opened Mar 20, 2025

• Unable to load google/siglip2-so400m-patch14-384/

  #36845 opened Mar 20, 2025

• GOT-OCR2 docs indicate model can produce markdown, but it only produces LaTeX.

  #36836 opened Mar 19, 2025

• Build for Windows and VS 2022 does not compile CUDA sources

  #36830 opened Mar 19, 2025

• Support for Ovis2 models

  #36824 opened Mar 19, 2025

• Gemma 3 is broken with fp16

  #36822 opened Mar 19, 2025

• Need Option to Disable Flash Attention in VideoLLaMA2.1-7B-AV (SiglipVisionModel)

  #36819 opened Mar 19, 2025

• Add EuroBert Model To Config

  #36817 opened Mar 19, 2025

• Gemma3 can't be fine-tuned on multi-image examples

  #36816 opened Mar 19, 2025

• Gemma3

  #36815 opened Mar 19, 2025

• When I use BF16 or FP16 to perform Lora fine-tuning on GemMA-3-12B-it, there will be an error when saving the checkpoint, but FP32 is normal

  #36814 opened Mar 19, 2025

• Not able to trace GPT2DoubleHeadsModel

  #36812 opened Mar 19, 2025

• Logic Errors in Image_processing_gemma3_fast.py

  #36806 opened Mar 19, 2025

• Qwen2VLForConditionalGeneration.from_pretrained() hangs with v0.50.0-dev0

  #36803 opened Mar 18, 2025

• BERT is broken on `v4.49.0-Gemma-3`

  #36802 opened Mar 18, 2025

• Throw messages in text-generation task with deepseek r1 with PEFTModel

  #36783 opened Mar 18, 2025

• Please support GGUF format for UMT5EncoderModel

  #36774 opened Mar 17, 2025

• Inconsistent Documentation for `⁠dataset_index` Requirement Across ViTPose Models

  #36773 opened Mar 17, 2025

• Add Audio inputs available in apply_chat_template

  #36769 opened Mar 17, 2025

• Source link to Ray Tune API outdated

  #36765 opened Mar 17, 2025

• could not parse ModelProto from /home/imss/zxhhhh/llama-3-8b/tokenizer.model

  #36764 opened Mar 17, 2025

• Add Gemma 3 For Sequence Classification

  #36755 opened Mar 16, 2025

• Unable to load google/siglip2-base-patch16-naflex

  #36754 opened Mar 16, 2025

• IdeficsProcessor cannot handle multiple images in one text

  #36751 opened Mar 16, 2025

• Gemma 3 1B - TypeError: 'NoneType' object is not callable

  #36745 opened Mar 15, 2025

• Model Card to include key information (e.g. max_sequence_length, etc.)

  #36743 opened Mar 15, 2025

• Unable to deploy Gemma 3 on AWS SageMaker due to lack of support in tranfomers release

  #36738 opened Mar 15, 2025

• Error when tokenizer is set to string: `AttributeError: 'str' object has no attribute 'pad_token_id'`

  #36731 opened Mar 14, 2025

• `torch.compile` custom backend called by AotAutograd triggers recompiles when used with `CompileConfig`

  #36725 opened Mar 14, 2025

• trainer.train()

  #36723 opened Mar 14, 2025

• Add RoMa keypoint matcher

  #36718 opened Mar 14, 2025

• `return_assistant_tokens_mask` argument is blocked in `ProcessorMixin.apply_chat_template`

  #36713 opened Mar 14, 2025

• Support Flex Attention for encoder only models (XLMRoberta, ModernBERT etc...)

  #36697 opened Mar 13, 2025

• Transformers 4.49.0 breaks nvdiffrast plugin loading

  #36676 opened Mar 12, 2025

• The parameter 'text' may be None as the comments says, there is a confuse.

  #36667 opened Mar 12, 2025

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

  #36660 opened Mar 12, 2025

• Qwen2 MoE manual `head_dim`

  #36659 opened Mar 12, 2025

• Cannot run backward with tensor parallel

  #36657 opened Mar 12, 2025

• AutoModel from_pretrained does not recursively download relative imports

  #36653 opened Mar 12, 2025

• Marian RNN conversion support

  #36651 opened Mar 11, 2025

• Hybrid models

  #36646 opened Mar 11, 2025

• Is it correct that the repetition penalty is applied to the input_ids encompassing all inputs and outputs, rather than solely on the generated tokens?

  #36642 opened Mar 11, 2025

• [Feature Request]: refactor _update_causal_mask to a public utility

  #36640 opened Mar 11, 2025

• [BUG] Batch inference DDP + zero stage 3 = inference code hangs

  #36638 opened Mar 11, 2025

• `output_hidden_states` only return part of hidden_state when setting `device_map="auto"`

  #36636 opened Mar 10, 2025

• Difficulties with multi-GPU Inferencing

  #36634 opened Mar 10, 2025

• Add Magma from Microsoft to Transformers

  #36629 opened Mar 10, 2025

• Unable to use converted Llama 3.3 instruct model

  #36628 opened Mar 10, 2025

• [deepspeed] any plans for deepspeed-domino?

  #36624 opened Mar 10, 2025

• Can not use flash-attention and flash-varlen-attention on Ascend NPU

  #36618 opened Mar 9, 2025

• In "02_how_to_generate", code cell 1 has an error message

  #36613 opened Mar 8, 2025

• Not installable on arm64 due to jaxlib upper bound

  #36611 opened Mar 7, 2025

• Making attention mechanism stackable

  #36609 opened Mar 7, 2025

• Whisper pipeline returns empty segment for each processed audio chunk

  #36602 opened Mar 7, 2025

• lm_head parameters missing from named_parameters() in Qwen2.5-VL-3B-Instruct model

  #36598 opened Mar 7, 2025

• Error when changing vocab size when fine tuning llama-vision

  #36590 opened Mar 6, 2025

• After tokenizers upgrade, the length of the token does not correspond to the length of the model

  #36574 opened Mar 6, 2025

• txt2vedio

  #36573 opened Mar 6, 2025

• In the latest version of transformers (4.49.0) matrix transformation error is encountered

  #36571 opened Mar 6, 2025

• torch_dtype is actually used now?

  #36567 opened Mar 5, 2025

• Add support for StableAdamW optimizer in Trainer

  #36564 opened Mar 5, 2025

• Stop output to stdout in streamers.py methods

  #36562 opened Mar 5, 2025

• Improving expected test results

  #36561 opened Mar 5, 2025

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

  #36560 opened Mar 5, 2025

• Error during processing: MllamaForCausalLM does not support Flash Attention 2.0 yet.

  #36557 opened Mar 5, 2025

• Facing issue while getting model from Rag,pretrained

  #36548 opened Mar 5, 2025

• Wrong dependency: `"tensorflow-text<2.16"`

  #36541 opened Mar 4, 2025

• Bug when computing positional IDs from embeddings

  #36537 opened Mar 4, 2025

• Bug in LlaveNextProcessor when using do_pad=False

  #36531 opened Mar 4, 2025

• Request: Add Flash Attention 2.0 Support for ViTMAEForPreTraining

  #36527 opened Mar 4, 2025

• GraniteMoE’s implementation is not compatible with HF’s peft

  #36518 opened Mar 3, 2025

• Object detection tutorial uses buggy dataset, may lead to crash during training

  #36516 opened Mar 3, 2025

• Allow parameters of the ViTPooler to be configurable, with the default values set to the current hardcoded values

  #36513 opened Mar 3, 2025

• model.generate function is not compatible with custom position_ids

  #36510 opened Mar 3, 2025

• Can not use prompt tuning inference

  #36509 opened Mar 3, 2025

• haggingface model unsupported torch backward hook

  #36508 opened Mar 3, 2025

• add a param to control cache in streamer when return output

  #36505 opened Mar 3, 2025

• TypeError: object of type 'IterableDataset' has no len()

  #36501 opened Mar 3, 2025

• Support Distill Depth Anything

  #36499 opened Mar 2, 2025

• Error at scatter num_items_in_batch in ddp/dp

  #36492 opened Mar 2, 2025

• llama code break with torch compile

  #36484 opened Mar 1, 2025

• Add type checking to CI

  #36481 opened Feb 28, 2025

• Enhance the memory efficiency of loading large models (400B) to prevent out-of-memory errors when using tensor parallelism.

  #36467 opened Feb 27, 2025

• ViTPose tutorial fails

  #36454 opened Feb 27, 2025

• Error in tiktoken integration example

  #36438 opened Feb 26, 2025

• Unable to save model after training with tensor parallel

  #36436 opened Feb 26, 2025

176 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 Janus model

  #36053 commented on Mar 20, 2025 • 144 new comments

• Add EfficientLoFTR model

  #36355 commented on Mar 19, 2025 • 75 new comments

• Add FAST

  #35476 commented on Mar 26, 2025 • 75 new comments

• Samhq model addition

  #35147 commented on Mar 26, 2025 • 39 new comments

• Add MLCD model

  #36182 commented on Mar 25, 2025 • 33 new comments

• Add TimesFM Time Series Forecasting Model

  #34082 commented on Mar 25, 2025 • 29 new comments

• Add StyleTTS 2

  #35790 commented on Mar 19, 2025 • 27 new comments

• Add evolla rebase main

  #36232 commented on Mar 24, 2025 • 19 new comments

• Add DeepSeek V2 Model into Transformers

  #36400 commented on Mar 26, 2025 • 19 new comments

• Add InternVL (2.5 MPO)

  #35968 commented on Mar 25, 2025 • 17 new comments

• Update deprecated/unused dependencies 🧹 🧹

  #36419 commented on Mar 4, 2025 • 14 new comments

• Add padding-free to bamba

  #35861 commented on Mar 14, 2025 • 13 new comments

• Add support for MiniMax's MiniMax-Text-01

  #35831 commented on Mar 20, 2025 • 13 new comments

• Add index selection for `output_hidden_states`

  #33705 commented on Mar 18, 2025 • 13 new comments

• `GPT2Model` StaticCache support

  #35761 commented on Mar 26, 2025 • 12 new comments

• Add Doge model

  #35891 commented on Mar 22, 2025 • 12 new comments

• Introduce modular files for speech models

  #35902 commented on Mar 21, 2025 • 10 new comments

• Add D-FINE Model into Transformers

  #36261 commented on Mar 26, 2025 • 6 new comments

• [MLU] Fix FA2 check error, remove deepspeed-mlu deps.

  #36159 commented on Mar 26, 2025 • 6 new comments

• make `num_items_in_batch` optional in compute_loss_func

  #36426 commented on Mar 26, 2025 • 5 new comments

• Fix Mask2Former Weight Initialization Issues #35877

  #35904 commented on Mar 24, 2025 • 5 new comments

• Add Segment Anything 2 (SAM2)

  #32317 commented on Mar 25, 2025 • 4 new comments

• switch from `training_args.bin` `training_args.json`

  #35010 commented on Mar 10, 2025 • 4 new comments

• Add internlm3 dense

  #35694 commented on Mar 19, 2025 • 4 new comments

• Fix Batch Size Mismatch When Using `crops_n_layers` in `mask-generation` Pipeline #35530

  #35627 commented on Mar 20, 2025 • 3 new comments

• fix: support grad clipping for TP through replicating non-sharded modules

  #36132 commented on Mar 25, 2025 • 3 new comments

• Support QuestionAnswering Module for ModernBert based models.

  #35566 commented on Mar 25, 2025 • 2 new comments

• Add LightGlue model

  #31718 commented on Mar 10, 2025 • 2 new comments

• Introduce numpy/numba optimization to `Qwen2VLImageProcessor`

  #36356 commented on Feb 28, 2025 • 2 new comments

• Append best model checkpoint with active adapter when not default

  #36201 commented on Mar 13, 2025 • 2 new comments

• Flash Attention v3

  #36190 commented on Mar 24, 2025 • 1 new comment

• fix: dtype might change during resize

  #36089 commented on Feb 27, 2025 • 1 new comment

• [Whisper] Pipeline: handle long form generation

  #35750 commented on Mar 13, 2025 • 1 new comment

• [WIP]: Base multimodal model for VLLM's `transformers` backend

  #36367 commented on Mar 26, 2025 • 1 new comment

• Integrate xlstm cleanly.

  #35377 commented on Mar 25, 2025 • 1 new comment

• [i18n-zh] Translating kv_cache into zh-hans

  #36412 commented on Feb 27, 2025 • 1 new comment

• enable tp on CPU

  #36299 commented on Mar 26, 2025 • 1 new comment

• Add AIMv2 to Transformers

  #35550 commented on Mar 3, 2025 • 0 new comments

• Process inputs directly in apply_chat_template in image-text-to-text pipeline

  #35616 commented on Mar 24, 2025 • 0 new comments

• [Fix]Integrate LLM generation parameters into the evaluation method

  #36416 commented on Feb 27, 2025 • 0 new comments

• Bart: new cache format

  #35314 commented on Mar 14, 2025 • 0 new comments

• [generation] Support cache-cropping methods

  #35591 commented on Mar 11, 2025 • 0 new comments

• 🔴 Video processors as a separate class

  #35206 commented on Mar 3, 2025 • 0 new comments

• Add Relation DETR

  #34900 commented on Mar 20, 2025 • 0 new comments

• Bye bye env vars, keep everything as configs

  #34886 commented on Mar 19, 2025 • 0 new comments

• Add Molmo (7B-D, 7B-O, 70B)

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

• [`AutoDocstring`] Based on inspect parsing of the signature

  #33771 commented on Mar 17, 2025 • 0 new comments

• #33512 handle last element out of range error

  #33625 commented on Mar 12, 2025 • 0 new comments

• Update from pretrained error when loading

  #33380 commented on Mar 24, 2025 • 0 new comments

• Adding new zero-shot examples

  #32483 commented on Feb 28, 2025 • 0 new comments

• Skip non-selected experts for mixtral and qwen2_moe

  #32429 commented on Mar 11, 2025 • 0 new comments

• Trainer: add predict with generate

  #32346 commented on Mar 24, 2025 • 0 new comments

• Improve support for image generation with Chameleon & Anole

  #32013 commented on Mar 19, 2025 • 0 new comments

• Support Kosmos-2.5

  #31711 commented on Mar 25, 2025 • 0 new comments

• audio pipeline support for initial_prompt?

  #27317 commented on Mar 26, 2025 • 0 new comments

• warning bug in Qwen2DecoderLayer in transformers ==4.49

  #36361 commented on Mar 26, 2025 • 0 new comments

• The arguments in `utils/modular_model_converter.py` is different from those in docs

  #36362 commented on Mar 26, 2025 • 0 new comments

• 目前使用Ktransformers进行DEEPSEEK-R1满血版和4bit量化版模型进行推理，推理速度有多少tokens/s？对应的计算资源配置分别是多少？

  #36363 commented on Mar 26, 2025 • 0 new comments

• Support set non_blocking=True when move data from cpu to gpu

  #36408 commented on Feb 28, 2025 • 0 new comments

• [generate] Run custom generation code from the Hub

  #36405 commented on Feb 27, 2025 • 0 new comments

• Handle DAC conversion when using weight_norm with newer PyTorch versions

  #36393 commented on Mar 2, 2025 • 0 new comments

• Added Cosmos model files

  #36389 commented on Feb 27, 2025 • 0 new comments

• Fix: Use config.use_sliding_window instead of config.sliding_window

  #36377 commented on Mar 21, 2025 • 0 new comments

• chore(qwen2): display warning log only when sliding window attention …

  #36316 commented on Mar 8, 2025 • 0 new comments

• [`ModernBERT`] Never save 'reference_compile' config; should be set based on end user

  #36305 commented on Mar 20, 2025 • 0 new comments

• Update composition flag usage

  #36263 commented on Mar 19, 2025 • 0 new comments

• Add support for DeepseekAI's DeepseekVL

  #36248 commented on Mar 26, 2025 • 0 new comments

• Improvements in attention_forward functions

  #36218 commented on Mar 14, 2025 • 0 new comments

• Fix the eval_use_gather_object flag usage

  #36214 commented on Mar 18, 2025 • 0 new comments

• fix: condition bos_token_id and space as token

  #36211 commented on Mar 19, 2025 • 0 new comments

• Fixed dynamic module import when there is more than one dot in class …

  #36198 commented on Mar 24, 2025 • 0 new comments

• (ugly) Use `parallelism=4` for `check_repository_consistency`

  #36197 commented on Mar 20, 2025 • 0 new comments

• Set evaluation and checkpointing defaults to 'epoch' and reduce loggi…

  #36133 commented on Mar 13, 2025 • 0 new comments

• Add Phi-3.5-vision

  #36036 commented on Mar 14, 2025 • 0 new comments

• [ModernBert] Prevent the attention mask from being None in ModernBertForSequenceClassification

  #35991 commented on Mar 7, 2025 • 0 new comments

• Idefics: remove double BOS token

  #35950 commented on Mar 11, 2025 • 0 new comments

• [ModernBERT] Add CausalLM functionality to ModernBERT

  #35946 commented on Mar 3, 2025 • 0 new comments

• [WIP] add deepseek-v3

  #35926 commented on Mar 26, 2025 • 0 new comments

• Missing weights not initialized properly #35437

  #35913 commented on Feb 27, 2025 • 0 new comments

• Several fixes related to rotary position embeddings

  #35901 commented on Mar 19, 2025 • 0 new comments

• Adds GGUF support for Gemma models

  #35887 commented on Mar 4, 2025 • 0 new comments

• Add MultipleChoice & QuestionAnswering heads to ModernBERT

  #35825 commented on Mar 3, 2025 • 0 new comments

• Remove head mask in generative models

  #35786 commented on Mar 19, 2025 • 0 new comments

• Add ColQwen2 to 🤗 transformers

  #35778 commented on Mar 26, 2025 • 0 new comments

• fix immediate quantization of the first token in QuantizedCache

  #35760 commented on Mar 20, 2025 • 0 new comments

• Pipeline: fix unnecessary warnings

  #35753 commented on Mar 17, 2025 • 0 new comments

• Mask2former & Maskformer Fast Image Processor

  #35685 commented on Mar 7, 2025 • 0 new comments

• [docs] add return_timestamps=True for Whisper long-form transcription

  #35633 commented on Mar 20, 2025 • 0 new comments

• Problem about using mBART50 for Russian to Chinese translation

  #13116 commented on Mar 7, 2025 • 0 new comments

• LayerDrop broken in various Flax models (Whisper/BART/more...)

  #35468 commented on Mar 8, 2025 • 0 new comments

• `Llama-3.2-11B-Vision-Instruct` (`mllama`) FSDP fails if grad checkpointing is enabled

  #36040 commented on Mar 8, 2025 • 0 new comments

• DeepSeek V3 Support

  #35425 commented on Mar 8, 2025 • 0 new comments

• Unknown quantization type, got fp8

  #35471 commented on Mar 9, 2025 • 0 new comments

• Missing weights are not properly initialized when using model.from_pretrained()

  #35437 commented on Mar 9, 2025 • 0 new comments

• [Whisper] TypeError: '<=' not supported between instances of 'NoneType' and 'float'

  #33552 commented on Mar 10, 2025 • 0 new comments

• Inconsistent saving of tokenizer with custom code from HF hub vs. local directory

  #35597 commented on Mar 10, 2025 • 0 new comments

• Mask2FormerImageProcessor support overlapping features

  #35536 commented on Mar 11, 2025 • 0 new comments

• Add the support for deepseek architecture .gguf

  #36144 commented on Mar 13, 2025 • 0 new comments

• FSDP Torch XLA vs. FSDPv2 (SMPD) Torch XLA checkpoint saving bug

  #36004 commented on Mar 13, 2025 • 0 new comments

• `trainer.evaluate` always creates a new MLFlow run, separate from the one used during `train()`

  #35074 commented on Mar 13, 2025 • 0 new comments

• A word-level timestamps on whisper generation pipeline is mismatched to total duration

  #36228 commented on Mar 13, 2025 • 0 new comments

• Device Movement Error with 4-bit Quantized LLaMA 3.1 Model Loading

  #36272 commented on Mar 13, 2025 • 0 new comments

• XLA FSDP V2 + TPU + T5 Family Models doesn't work

  #35142 commented on Mar 13, 2025 • 0 new comments

• oom when using adafactor optimizer in deepspeed

  #33290 commented on Mar 13, 2025 • 0 new comments

• Request to add DINO object detector

  #36205 commented on Mar 14, 2025 • 0 new comments

• TypeError: ModernBertModel.forward() got an unexpected keyword argument 'num_items_in_batch'

  #36074 commented on Mar 14, 2025 • 0 new comments

• WhisperForCTC

  #26242 commented on Mar 14, 2025 • 0 new comments

• Add support for context parallelism

  #35983 commented on Mar 14, 2025 • 0 new comments

• denoising with sentence permutation, and language sampling

  #11129 commented on Mar 15, 2025 • 0 new comments

• Jitter Noise added to input being passed to experts in Switch Transformers

  #33969 commented on Mar 15, 2025 • 0 new comments

• Support sliding_window for sdpa in qwen2

  #36351 commented on Feb 27, 2025 • 0 new comments

• Add cosmos from Nvidia

  #35565 commented on Feb 27, 2025 • 0 new comments

• add Flash Attention Support for Helsinki-NLP/opus models

  #36169 commented on Feb 28, 2025 • 0 new comments

• [Feature Request] We might need a function to change the sampler used in trainer dataloader

  #26802 commented on Feb 28, 2025 • 0 new comments

• Error From BitsandBytes

  #36371 commented on Feb 28, 2025 • 0 new comments

• Export to ExecuTorch

  #32253 commented on Mar 1, 2025 • 0 new comments

• Speed up image processors - cast to array before BatchFeature

  #31205 commented on Mar 2, 2025 • 0 new comments

• Support SDPA & Flash Attention 2 for LayoutLMv3

  #35467 commented on Mar 2, 2025 • 0 new comments

• Support H100 training with FP8 in Trainer and Deepspeed

  #25333 commented on Mar 2, 2025 • 0 new comments

• SAM mask-generation - crops_n_layers

  #35530 commented on Mar 3, 2025 • 0 new comments

• _batch_encode_plus() got an unexpected keyword argument 'is_pretokenized' using BertTokenizerFast

  #17488 commented on Mar 4, 2025 • 0 new comments

• Add EVEv2 : an Encoder-free VLM

  #36379 commented on Mar 4, 2025 • 0 new comments

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

  #33688 commented on Mar 4, 2025 • 0 new comments

• Add `Tensor Parallel` support for ALL models

  #34789 commented on Mar 4, 2025 • 0 new comments

• Possible bug when using cosine lr scheduler with gradient accumulation

  #35484 commented on Mar 4, 2025 • 0 new comments

• (False?) warning about weight_g/weight_v missing on WeightNorm on PyTorch

  #26796 commented on Mar 4, 2025 • 0 new comments

• Add support for Molmo

  #33710 commented on Mar 4, 2025 • 0 new comments

• The output tensor's data type is not torch.long when the input text is empty.

  #36277 commented on Mar 4, 2025 • 0 new comments

• Is T5 model supported with HQQ quantization ? (AttributeError: 'HQQLinear' object has no attribute 'weight')

  #36254 commented on Mar 4, 2025 • 0 new comments

• redirect logging output to `stdout` instead of `stderr`

  #34613 commented on Mar 6, 2025 • 0 new comments

• Enable Quantize KV Cache for Mistral Model

  #35041 commented on Mar 7, 2025 • 0 new comments

• llama `tie_word_embeddings` ignored on cpu and with auto dtype only

  #33689 commented on Mar 7, 2025 • 0 new comments

• rework `test_multi_gpu_data_parallel_forward`

  #31087 commented on Mar 22, 2025 • 0 new comments

• flash_attention_2 2.7.2.post1 seems to crash when using `torch.compile` and `DataCollatorWithFlattening`

  #35588 commented on Mar 22, 2025 • 0 new comments

• PruneTrain: Fast Neural Network Training by Dynamic Sparse Model Reconfiguration

  #10105 commented on Mar 23, 2025 • 0 new comments

• DS3 zero3_save_16bit_model is not compatible with resume_from_checkpoint

  #36317 commented on Mar 23, 2025 • 0 new comments

• Bug about num_update_steps_per_epoch in function _inner_training_loop

  #36297 commented on Mar 23, 2025 • 0 new comments

• tensor parallel training bug

  #36296 commented on Mar 23, 2025 • 0 new comments

• safetensor/mmap memory leak when per-layer weights are converted do other dtypes

  #34366 commented on Mar 23, 2025 • 0 new comments

• Add argument to set number of eval steps in Trainer

  #31561 commented on Mar 24, 2025 • 0 new comments

• Assisted generation slower than with base model alone

  #36337 commented on Mar 24, 2025 • 0 new comments

• Unable to use Seq2SeqTrainingArguments and Seq2SeqTrainer

  #36330 commented on Mar 24, 2025 • 0 new comments

• Whisper word-level timestamp extraction fails with beam search

  #36093 commented on Mar 24, 2025 • 0 new comments

• Multi-GPU training crashes with IterableDataset and different length input (e.g. Next token prediction)

  #35308 commented on Mar 24, 2025 • 0 new comments

• Inference with FSDP during training affects checkpoints

  #34530 commented on Mar 24, 2025 • 0 new comments

• Mask2Former _init_weights

  #35877 commented on Mar 24, 2025 • 0 new comments

• Recomputed tensor size does not match when using activation checkpointing when using FSDP and accelerate

  #34928 commented on Mar 25, 2025 • 0 new comments

• ValueError: Trying to set a tensor of shape torch.Size([128256, 3072]) in "weight" (which has shape torch.Size([128003, 3072])), this looks incorrect

  #36350 commented on Mar 25, 2025 • 0 new comments

• Accelerate x Trainer issue tracker:

  #33345 commented on Mar 25, 2025 • 0 new comments

• CVE-2024-11392 - AWS Scanner and Trivy Flagging Transformers 4.48.1 as Vulnerable

  #36041 commented on Mar 25, 2025 • 0 new comments

• multi-gpu: test_model_parallel_beam_search tests fail with "IndexError: list index out of range"

  #35824 commented on Mar 25, 2025 • 0 new comments

• `Helsinki-NLP/opus-mt-it-en` isn't on HuggingFace Hub

  #26382 commented on Mar 25, 2025 • 0 new comments

• AttributeError: 'dict' object has no attribute 'to_dict'; for Inferencing Lora Merged Qwen/Qwen2.5-VL-3B-Instruct

  #36281 commented on Mar 25, 2025 • 0 new comments

• TypeError: CustomTrainer.compute_loss() got an unexpected keyword argument 'num_items_in_batch'

  #36331 commented on Mar 25, 2025 • 0 new comments

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

  #33399 commented on Mar 16, 2025 • 0 new comments

• Offline mode doesn't work with models that require `trust_remote_code=True`

  #34855 commented on Mar 16, 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 Mar 17, 2025 • 0 new comments

• MultiTask Classification and label_names on Trainer

  #33193 commented on Mar 17, 2025 • 0 new comments

• SDPA `is_causal=False` has no effect due to `LlamaModel._prepare_4d_causal_attention_mask_with_cache_position`

  #36150 commented on Mar 17, 2025 • 0 new comments

• Model trained with Flash Attention 2.0 raises "RuntimeError: query and key must have the same dtype" when generating

  #30019 commented on Mar 18, 2025 • 0 new comments

• Tensor size mismatch when trying to run RT-DETR on multiple gpus

  #33165 commented on Mar 18, 2025 • 0 new comments

• Custom 4D tensor caused shape mismatch error

  #35290 commented on Mar 18, 2025 • 0 new comments

• Cryptic error when using AutoTokenizer with SentencePiece tokenizers without sentencepiece installed

  #36291 commented on Mar 19, 2025 • 0 new comments

• A warning message showing that `MultiScaleDeformableAttention.so` is not found in `/root/.cache/torch_extensions` if `ninja` is installed with `transformers`

  #35349 commented on Mar 19, 2025 • 0 new comments

• Inconsistent output lengths when `max_length=20` is set implicitly vs explicitly in `generate()`

  #35765 commented on Mar 19, 2025 • 0 new comments

• `AutoModelForCasualLM.from_pretrained()` exits without warning/error

  #36245 commented on Mar 19, 2025 • 0 new comments

• IsADirectoryError when training with tqdm enabled for trainer

  #34766 commented on Mar 20, 2025 • 0 new comments

• Incompatibility in flash_attention_2 + Llama + Transformers>=4.43 + Autocast to fp16

  #36224 commented on Mar 20, 2025 • 0 new comments

• modeling_phi3 errors with AttributeError: 'DynamicCache' object has no attribute 'get_max_length'

  #36071 commented on Mar 20, 2025 • 0 new comments

• ValueError: Unrecognized image processor in Qwen/Qwen2.5-VL-3B-Instruct.

  #36193 commented on Mar 21, 2025 • 0 new comments

• cannot import name 'is_timm_config_dict' from 'transformers.utils.generic'

  #36068 commented on Mar 21, 2025 • 0 new comments

• Community contribution: Adding GGUF support for more architectures

  #33260 commented on Mar 21, 2025 • 0 new comments

• Qwen2VLForConditionalGeneration doesn't work with MPS devices

  #36413 commented on Mar 21, 2025 • 0 new comments

• [Bugs] RuntimeError: No CUDA GPUs are available in transformers v4.48.0 or above when running Ray RLHF example

  #36295 commented on Mar 22, 2025 • 0 new comments

• past_key_value(s) name inconsistency causing problems

  #36290 commented on Mar 22, 2025 • 0 new comments

• model.gradient_checkpointing_enable() makes loss.requires_grad be False

  #35826 commented on Mar 22, 2025 • 0 new comments