19505 Commits

This Branch

This Branch

All Branches

Author	SHA1	Message	Date
NielsRogge	11d0feacce	[AutoModelForMaskGeneration] Remove duplicate code (#38622) ... Remove duplicate code	2025-06-25 10:00:13 +02:00
efsotr	3ee72af6b6	Fix graph break in torch.compile when using FA2 with attention_mask=None and batch size > 1 (#37332) ... * Fix graph break in torch.compile when using FA2 with attention_mask=None and batch size > 1 * fix code format * add test; replace position_ids with query_states becasue position_ids.shape[0] is always 1 * add assert loss is not nan	2025-06-25 07:58:34 +00:00
ranzhejiang	ae32f1ad11	Add zero dim tensor check when using flash_attention (#38280) ... * Add zero dim tensor check when using flash_attention Signed-off-by: ranzhejiang <zhejiang.ran@intel.com> * Add zero dim tensor check when using flash_attention Signed-off-by: ranzhejiang <zhejiang.ran@intel.com> --------- Signed-off-by: ranzhejiang <zhejiang.ran@intel.com>	2025-06-25 09:48:50 +02:00
StevenBucaille	ca402e2116	[LightGlue] Fixed attribute usage from descriptor_dim to keypoint_detector_descriptor_dim (#39021) ... fix: fix descriptor dimension handling in LightGlue model	2025-06-24 23:32:07 +01:00
Marcel Ambo Ndowah	48b6ef0238	Add Hugging Face authentication procedure for IDEs (PyCharm, VS Code,… (#38954) ... * Add Hugging Face authentication procedure for IDEs (PyCharm, VS Code, etc.) * Update quicktour.md --------- Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>	2025-06-24 11:48:15 -07:00
Dmitry	ea9a30923e	[HPU][Critical Issue Fix] ThreadPool instead of Pool for parallel pre-processing (#39002) ... * ThreadPool instead of Pool for parallel pre-processing * ThreadPool only if hpu available	2025-06-24 20:24:50 +02:00
ivarflakstad	995666edb5	Skip sdpa dispatch on flash test due to unsupported head dims (#39010)	2025-06-24 20:16:56 +02:00
ivarflakstad	f367c6337d	Update self-comment-ci.yml user list (#39014) ... add ivarflakstad to self-comment-ci.yml	2025-06-24 20:13:36 +02:00
Tugsbayasgalan Manlaibaatar	67d36dc1d7	Fix bugs in DynamicCache (#37880) ... * Fix bugs in DynamicCache * Updarte * Update * Lint * lint * Rename test * update * update	2025-06-24 19:43:40 +02:00
eustlb	6bdd4ec952	Add kyutai stt (#38909) ... * first draft * cleaner version * udpate tests + modeling * add tests * init * udpate test_modeling_common * fix tests * csm Processor draft * convertion update * mimi cache padding convolutions draft * mimi streaming udpates * update mimi padding cache test * udpate cache padding mimi test * make style mimi * updates generate moshi asr * moshi asr integration tests (single + batched) * update tests * update conversion script * good default sliding window value * udpdate generate * update test checkpoint * nit * fix mimi * fix codec prefix * revert * revert * update config * update config * unnecessary mimi input restriction * remove delay in tokens * remove _prepare_4d_causal_attention_mask_with_cache_position and _update_causal_mask * test update * modular update * make style * nit * rename * create codec model generation config at init * remove delay * max_new_tokens/length warning * correct conv1 padding cache import for modular * nit * fix on encoder_past_key_values * convert modular * move frame_size to config * move frame_size to config * update test name * handle first token is bos * better handling of max_new_tokens * fix * fix batch size in test input prep * update docstring * convert modular * make style * make style * add feature extractor * correct modular convention name for feature_extraction file * update convertion script * doc processor * update doc * udpate init * update model type * fixes * update tests * fix * make * add doc * nit * fix * doc * auto mappings * doc * nit * convert modular * doc * nit * extend _keep_in_fp32_modules to enforce fp32 * renaming to stt * doc update + test update * doc fixes * doc fix * doc fix * fix musicgen tests * fix musicgen tests * make style * fix musicgen tests * correct frame_rate config param for mimi * update mimi test * revert update mimi test * enforce cpu test * move cache init in cache class * convert modular * docstring update * update model id * feature_extractor -> feature_extraction (SEW) * convert modular * update model id	2025-06-24 18:01:15 +02:00
Mohamed Mekkouri	08bf7f1afe	Add kernelize to transformers (#38205) ... * fix * fix * fix flow * remove non compiling path * change * style * fix * update * update pin * revert	2025-06-24 17:38:54 +02:00
Avihu Dekel	be10d4df60	Granite speech - minor fixes to support training with the HF trainer (#38833) ... * ensure the query is updated during training avoid unused parameters that DDP does not like * avoid a crash when `kwargs` contain `padding=True` trainers often pass this argument automatically * minor * Remove mel_spec lazy init, and rename to mel_filters. this ensures save_pretrained will not crash when saving the processor during training d5d007a1a0/src/transformers/feature_extraction_utils.py (L595) * minor - most feature extractors has a `sampling_rate` property	2025-06-24 17:06:52 +02:00
Cyril Vallez	e1e11b0299	Fix undeterministic order in modular dependencies (#39005) ... * sort correctly * Update modeling_minimax.py * Update modular_model_converter.py	2025-06-24 17:04:33 +02:00
7mile	bdf5fb70aa	Skip non-selected experts for qwen3_moe (#38133) ... * fix(qwen3moe): skip experts with no workload * avoid tolist and also update other moe models * fix: should squeeze 0-dim only	2025-06-24 16:33:48 +02:00
Tanuj Rai	719058c625	Update attention_visualizer.py (#37860)	2025-06-24 16:21:36 +02:00
Mylon Jones	9f42c1f192	Added scikit-learn to the example image-classification requirements.txt (#37506) ... Co-authored-by: Pavel Iakubovskii <qubvel@gmail.com>	2025-06-24 15:24:02 +02:00
Cyril Vallez	1636a7bcb9	Fixes for Arcee model (#39001) ... * fix modular * Update modular_arcee.py * fix	2025-06-24 15:23:52 +02:00
Crystalcareai	71de20b818	Add Arcee model support (#38621) ... * Add Arcee model support to transformers - Add ArceeConfig and model mappings for all task types (CausalLM, SequenceClassification, QuestionAnswering, TokenClassification) - Add auto-loading support through AutoModel, AutoConfig, and AutoTokenizer - Use LlamaTokenizer for tokenization - Add FX graph support for Arcee models - Create lazy loading module structure for Arcee * feat: update YARN scaling and RoPE validation for Arcee model * feat: add auto_docstring checkpoint config to Arcee model classes * docs: add pre-trained model weights reference to Arcee configuration files * refactor: move RoPE utilities to dedicated modeling_rope_utils module * Add comprehensive test suite for Arcee model - Add test_modeling_arcee.py following standard transformers test patterns - Include tests for all model variants (CausalLM, SequenceClassification, QuestionAnswering, TokenClassification) - Add specific test for ReLU² activation in ArceeMLP - Add RoPE scaling tests including YARN support - Follow CausalLMModelTest pattern used by similar models * Add documentation for Arcee model - Add comprehensive model documentation with usage examples - Include all model variants in autodoc - Add to table of contents in proper alphabetical order - Fixes documentation coverage for Arcee model classes * Make style/fixup * fix copyright year * Sync modular conversion * revert in legacy supported models in src/transformers/utils/fx * cleaned redundant code in modular_arcee.py * cleaned testing * removed pretraining tp * fix styles * integration testing --------- Co-authored-by: Pranav <veldurthipranav@gmail.com> Co-authored-by: Pranav <56645758+pranav4501@users.noreply.github.com>	2025-06-24 15:05:29 +02:00
Anton Vlasjuk	23c89a6732	[Attention] Small fix on output attentions (#38948) ... small fix	2025-06-24 14:42:10 +02:00
Dianana	4f650040a6	Removing extra space in large command for speech-pretraining example (#38705) ... Removing extra space in Large command	2025-06-24 12:24:56 +00:00
Raushan Turganbay	d3d835d4fc	[qwen] refactor attentions for vision/audio (#38930) ... * refactor attentions in vision/audio * remove fa2 import * make config the only args * pass along kwargs from modality encoders * style	2025-06-24 10:53:52 +02:00
vb	2e4c045540	🔴 Update default dtype for pipelines to auto (#38882) ... * check typing * Fallback to fp32 if auto not supported. * up. * feedback from review. * make style.	2025-06-24 10:39:18 +02:00
casinca	21cb353b7b	[docs] Typos - Single GPU efficient training features (#38964) ... * Typos - corrected bf16 training argument - corrected header for SDPA * improved readability for SDPA suggested by @stevhliu Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com> --------- Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>	2025-06-23 12:33:10 -07:00
Yih-Dar	f9be71b34d	Fix rag (#38585) ... * fix * fix * fix --------- Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>	2025-06-23 17:42:46 +02:00
Yusuf Shihata	9eac19eb59	[Feature] Support is_split_into_words in the TokenClassificationPipeline. (#38818) ... * some fixes * some fixes * now the pipeline can take list of tokens as input and is_split_into_words argument * now the pipeline can take list of tokens as input and is_split_into_words argument * now the pipeline can take list of tokens as input and is_split_into_words argument and we can handle batches of tokenized input * now the pipeline can take list of tokens as input and is_split_into_words argument and we can handle batches of tokenized input * solving test problems * some fixes * some fixes * modify tests * aligning start and end correctly * adding tests * some formatting * some formatting * some fixes * some fixes * some fixes * resolve conflicts * removing unimportant lines * removing unimportant lines * generalize to other languages * generalize to other languages * generalize to other languages * generalize to other languages	2025-06-23 15:31:32 +00:00
Yih-Dar	2ce02b98bf	fix mistral and mistral3 tests (#38978) ... * fix * fix * fix * fix --------- Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>	2025-06-23 17:07:18 +02:00
Yoni Gozlan	b6b4d43d6d	Add support for auto_docstring with model outputs (#38242) ... * experiment auto_docstring model outputs * Fix PatchTSMixer * Add check model output docstring to check_auto_docstring and fix all model outputs docstring * add reordering of docstring in check_docstrings * add check for redundant docstring in check_docstrings, remove redundant docstrings * refactor check_auto_docstring * make style * fix copies * remove commented code * change List-> list Tuple-> tuple in docstrings * fix modular * make style * Fix modular vipllava --------- Co-authored-by: Cyril Vallez <cyril.vallez@huggingface.co>	2025-06-23 10:39:41 -04:00
kallewoof	0c98f24889	fix: add __bool__ operator to tokenizer to avoid bloated asserts (#38899) ... * fix: add __bool__ operator to tokenizer to avoid bloated asserts When a user does 'assert tokenizer' to ensure that the tokenizer is not None, they inadvertently set off a rather expensive process in the '__len__()' operator. This fix adds a trivial '__bool__()' that returns True, so that a None tokenizer asserts and an actual tokenizer returns True when asserted, without calling length op. * typo	2025-06-23 14:32:16 +00:00
Yoni Gozlan	d29482cc91	Add Idefics2/3 and SmolVLM Fast image processors + improvements for fast image processors (#38157) ... * add working idefics2 fast and improvements for fast nested images processing * add fast image processors idefics 3 and smolvlm * cleanup tests * fic doc idefics2 * PR review and fix issues after merge * Force providing disable_grouping to group_images_by_shape * simplify group_images_by_shape * fix modular * Fix nits after review	2025-06-23 14:17:25 +00:00
Rémi Ouazan	1a96127e46	Break tie in Expectations and gemma3 fixes (#38943) ... * Added major / minor version to Expectations ordering * Added fixes to gemma3 * Style	2025-06-23 15:13:27 +02:00
Pavel Iakubovskii	84d19be41e	Apply GradientCheckpointingLayer to the whole repo (#38913) ... * first batch (4) * align * altclip * beit * bert * yolos * dino, pvt_v2 * bark, bart, bert_generation * big_bird, biogpt * blnderbot, bloom * bridgetower * camambert, canine, chameleon * chinese clip, clap, clip * codegen, conditional detr, convbert * dab_detr, data2vec * dbrx, deberta * deberta, decicion_tranformer, deformable_detr * deit, deta, mctct * detr, dinov2, distilbert * donut, dpt, electra * ernie, esm, falcon * flava, fnet, falcon_mamba * focalnet, git, gpt2 * gpt - bigcode, neo, neox * gptj, groupvit * idefics2, idefics3 * ijepa, imagegpt, internvl * jetmoe, kosmos2, layoutlm * layoutlm2-3, led * lilt, longformer, longt5, luke * m2m, mamba1-2 * marian, markuplm, mask2former * maskformer * mbart, megatron_bert, mimi * mixtral, mlcd * mobilevit1-2, modernbert * moshi, mpt, mra * mt5, musicgen * mvp, nemotron * nllb_moe * nystromformer, omdet_turbo * opt, owlvit, owlv2 * pegasus, pegasus_x, presimmon * phimoe, pix2struct, pixtral * plbart, pop2piano, prophetnet * qwen2* * qwen2, qwen3 moe, rec gemma * rembert * roberta * roberta prelayernorm * roc_bert, roformer, rwkv * sam, sam_hq * seggpt, smolvlm, speech_to_text * splinter, stablelm, swin * swin2sr, switch_transformer, t5, table_transformer * tapas, time_series_tranformer, timesformer * trocr, tvp, umt5 * videomae, vilt, visual_bert * vit, vit_mae, vit_msn * vitpose_backbone, vits, vivit * whisper. x_clip, xglm * xlm_roberta, xmod * yoso * zamba * vitdet, wav2vec2, wav2vec2_bert * unispeech, wav2vec_conformer * wavlm * speecht5 * swinv2 * sew / _d * seamless_mt4 / _v2 * deprecated models update * bros * gemma2, gemma3 * got, hiera, hubert, llama4, mllama, oneformer, phi, olmoe, informer * fixup * Add use_cache=False and past_key_value=None to GradientCheckpointingLayer * fixup * fix prophetnet * fix bigbird_pegasus * fix blenderbot * fix mbart * fix mvp * fix zamba2 * fix bart * fix blenderbot_small * fix codegen * Update gradient checkpointing layer to support more past_key_values arg names * fix data2vec vision * fix deformable_detr * fix gptj * fix led * fix m2m_100 * add comment * fix nnlb_moe * Fix pegasus_x * fix plbart * udop * fix-copies: beit, wav2vec2 * fix gpt_bigcode * fixup * fix t5 * fix switch_transformers * fix longt5 * fix mt5 * update tapas * fix blip2 * update blip * fix musicgen * fix gpt2, trocr * fix copies * !!! Revert zamba, mllama * update autoformer * update bros * update args / kwargs for BERT and copies * 2nd round of updates * update conditional detr * Pass encoder_hidden_states as positional arg * Update to pass encoder_decoder_position_bias as positional arg * fixup * biogpt modular * modular gemma2 * modular gemma3 * modular gpt_neox * modular informer * modular internvl * modular mixtral * modular mlcd * modular modernbert * modular phi * modular qwen2_5_omni * modular qwen2_5_vl * modular sam_hq * modular sew * wav2vec2_bert * modular wav2vec2_conformer * modular wavlm * fixup * Update by modular instructblipvideo * modular data2vec_audio * nit modular mistral * apply modular minimax * fix modular moonshine * revert zamba2 * fix mask2former * refactor idefics	2025-06-23 14:24:48 +02:00
Cyril Vallez	07aab1af1e	Remove dead protected imports (#38980) ... * remove them * more	2025-06-23 13:44:50 +02:00
Cyril Vallez	74f5e4a1fa	[modular] CLI allows positional arguments, and more defaults names for the optional arg (#38979) ... * More defaults * Update modular_model_converter.py	2025-06-23 12:40:01 +02:00
Vensen	334bf913dc	Fix(informer): Correct tensor shape for input_size=1 (#38856) ... * Fix(time_series): Correct scaler tensor shape in base model The create_network_inputs function in TimeSeriesTransformerModel handled the scaler's loc and scale tensors inconsistently. When input_size=1, the tensors were not squeezed, leading to downstream dimension errors for models like Informer. This commit refactors the logic to unconditionally apply .squeeze(1), which correctly handles all input_size cases and fixes the bug at its source. Fixes #38745 * Fix(time_series): Correct scaler tensor shape in base model The create_network_inputs function in TimeSeriesTransformerModel handled the scaler's loc and scale tensors inconsistently. When input_size=1, the tensors were not squeezed, leading to downstream dimension errors for models like Informer. This commit refactors the logic to unconditionally apply .squeeze(1), which correctly handles all input_size cases and fixes the bug at its source. Fixes #38745 --------- Co-authored-by: Kashif Rasul <kashif.rasul@gmail.com>	2025-06-23 11:50:51 +02:00
Benoqtr	c184550daf	Fix DTensor import compatibility for PyTorch < 2.5 (#38836)	2025-06-23 11:25:56 +02:00
Ilyas Moutawwakil	984ff89e73	Gaudi3 CI (#38790)	2025-06-23 10:56:51 +02:00
DongKyu Kang	2166b6b4ff	Update blip model card (#38513) ... * Update docs/source/en/model_doc/blip.md * fix(docs/source/en/model_doc/blip.md): fix redundent typo error * fix (docs/source/en/model_doc/blip.md): modify of review contents * fix(docs/source/en/model_doc/blip.md): modify code block * Update blip.md --------- Co-authored-by: devkade <mouseku@moana-master> Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com>	2025-06-20 13:46:19 -07:00
Manuel de Prada Corral	166e823f77	Fix custom generate from local directory (#38916) ... Fix custom generate from local directory: 1. Create parent dirs before copying files (custom_generate dir) 2. Correctly copy relative imports to the submodule file. 3. Update docs.	2025-06-20 17:36:57 +01:00
Yih-Dar	3d34b92116	Switch to use A10 progressively (#38936) ... * try * fix * fix --------- Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>	2025-06-20 16:10:35 +00:00
Yih-Dar	b8059e1f8f	Fix more flaky test_initialization (#38932) ... * try * try * fix * fix * fix --------- Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>	2025-06-20 17:28:32 +02:00
Cyril Vallez	5ee60f970a	Correctly raise error for awq quantization (#38945) ... fix warning	2025-06-20 17:18:06 +02:00
Ákos Hadnagy	8ac2d75353	Pin PyTorch extras for AMD containers (#38941) ... * Pin additional Torch packages * Remove unused def --------- Co-authored-by: ivarflakstad <69173633+ivarflakstad@users.noreply.github.com>	2025-06-20 12:17:21 +00:00
Pavel Iakubovskii	9120567b02	Add kwargs for timm.create_model in TimmWrapper (#38860) ... * Add init kwargs for timm wrapper * model_init_kwargs -> model_args * add save-load test * fixup	2025-06-20 12:00:09 +00:00
Raushan Turganbay	ff95974bc6	[static cache] fix device map per layer in VLMs (#38488) ... return lm as decoder	2025-06-20 13:49:29 +02:00
Cyril Vallez	aa42987c1e	Remove ALL_LAYERNORM_LAYERS (#38922) ... * remove it everywhere * Update trainer_pt_utils.py * Update trainer_pt_utils.py * style * sort list in test * CIs * use recursion same way as before (for intermediate layer names)	2025-06-20 12:06:48 +02:00
Yao Matrix	38a9b70786	add pytorch-xpu Dockerfile (#38875) ... * first commit Signed-off-by: YAO Matrix <matrix.yao@intel.com> * use rls pytorch Signed-off-by: YAO Matrix <matrix.yao@intel.com> --------- Signed-off-by: YAO Matrix <matrix.yao@intel.com>	2025-06-20 11:42:44 +02:00
Rémi Ouazan	9bcdd5cde9	Modernbert fixes (#38912) ... * Removed deprecated argument in modernbert RotaryEmbedding * Skip test_sdpa_can_dispatch_on_flash for modernbert --------- Co-authored-by: ivarflakstad <69173633+ivarflakstad@users.noreply.github.com> Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com>	2025-06-20 11:22:32 +02:00
Yih-Dar	31d30b7224	Skip some tests for now (#38931) ... * try * [test all] --------- Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>	2025-06-20 11:05:49 +02:00
Cyril Vallez	0725cd6953	Remove deprecated classes in modeling_utils.py (#38919) ... * remove deprecated classes * style	2025-06-19 19:25:20 +02:00
Hamza Benchekroun	797860c68c	feat: add flexible Liger Kernel configuration to TrainingArguments (#38911) ... * feat: add flexible Liger Kernel configuration to TrainingArguments Add support for granular Liger Kernel configuration through a new `liger_kernel_config` parameter in TrainingArguments. This allows users to selectively enable/disable specific kernels (rope, swiglu, cross_entropy, etc.) instead of the current approach that rely on default configuration. Features: - Add `liger_kernel_config` dict parameter to TrainingArguments - Support selective kernel application for all supported models - Maintain full backward compatibility with existing `use_liger_kernel` flag Example usage: ```python TrainingArguments( use_liger_kernel=True, liger_kernel_config={ "rope": True, "swiglu": True, "cross_entropy": False, "fused_linear_cross_entropy": True } ) Closes #38905 * Address comments and update Liger section in Trainer docs	2025-06-19 15:54:08 +00:00