huggingface/transformers

Pull Requests Commits

Tests: import librosa only when available in the Qwen audio tests

IlyasMoutawwakil committed 1 day ago

b243d0d5

Tests: skip the Higgs Audio V2 tokenizer tests without torchaudio

IlyasMoutawwakil committed 2 days ago

cf76d3b7

Fix Mistral Large 4 config (#50418)

Cyrilvallez committed 2 days ago

Verified cdb75de4

[docs] Common quantization tasks (#50095)

stevhliu committed 2 days ago

Verified 7d5d6d66

Fix flaky `Gemma4Audio2TextModelTest::test_training_gradient_checkpointing_use_reentrant_true` (#50410)

ydshieh committed 2 days ago

Verified 084b1fd1

Fix flaky `T5GemmaModelTest::test_decoder_model_past_with_large_inputs` (#50409)

ydshieh committed 2 days ago

Verified 7b0fec70

[Model] LFM3 Family: LFM3-Text (#50322)

paulpak58 committed 2 days ago

Verified 7d6cae08

Don't pick torchcodec when it can't load or has no FFmpeg (#50362)

eustlb committed 2 days ago

Verified 5ea23808

[CB] Disable streams on MPS to fix continuous batching with torch 2.15 (#50404)

Isalia20 committed 2 days ago

Verified 2ffebc8c

Keep enabling checkpointing on models that only define gradient_checkpointing_enable (#50349)

qgallouedec committed 2 days ago

Verified a15175e3

Fix flaky `Owlv2ModelTest::test_eager_matches_sdpa_inference_01_fp16_pad_left` (#50403)

ydshieh committed 2 days ago

Verified f1eff635

Fix flaky `PocketTTSAudioTokenizerModelTest::test_batching_equivalence` on CPU (#50402)

ydshieh committed 2 days ago

Verified 9bb266d5

[`SDPA`] Fix decode under compile (#50388)

vasqu committed 2 days ago

Verified 2aab563b

fix(pipelines): validate zero-shot detection batch labels (#50318)

BlazeSR10hawk committed 2 days ago

Verified b55e0761

Fix PaddleOCR-VL vision position interpolation alignment (#50384)

scyyh11 committed 2 days ago

Verified 924266c4

Fix Emu3 image processor crash on float numpy images (#50226)

simpleqt committed 2 days ago

Verified 58d05ca5

Build Idefics-style vision position ids with torch.where (#49917)

shoumikhin committed 2 days ago

Verified 62dd93e1

[`distributed`] Divide the expert gradients once when `efsdp_size == 1` (#50292)

3outeille committed 2 days ago

Verified b3187c7f

[compressed-tensors] better path through conversion mapping (#50238)

ArthurZucker committed 2 days ago

Verified b1a76d40

Update PR CI to use torch 2.15 (#50361)

ydshieh committed 3 days ago

Verified d2033901

Exclude attention-masked padding from repetition penalty (#50149)

BlazeSR10hawk committed 3 days ago

Verified 0f5a2a5b

Do not touch torch internal compilation flags in chunked prefill (#50358)

Cyrilvallez committed 3 days ago

Verified 9b283429

[`Rope`] Fix per layer rope in head dim manipulation (#49864)

vasqu committed 3 days ago

Verified 9822ea6e

Update FSDP non-rank-0 missing keys test after #45649 (#50260)

SunMarc committed 3 days ago

Verified 19495891

[`FA`] Fix head dim manipulation for more special models (#49862)

vasqu committed 3 days ago

Verified f8a1e1fb

Add `transcription` input arg to `apply_transcription_request` (#49869)

ebezzam committed 3 days ago

Verified 666a686b

Revert linguistic changes (#50345)

vasqu committed 3 days ago

Verified 4abc4e58

fix: preserve multiple candidates in multimodal pipelines (#50198)

BlazeSR10hawk committed 3 days ago

Verified 4fbb35b8

:rotating_light: [`kernels`] Bump version (#50307)

vasqu committed 3 days ago

Verified 6280bfe7

Fold GQA query heads onto their KV head when decoding instead of repeating K/V (#49743)

Yooniel committed 3 days ago

Verified 03238b25

Older