huggingface/transformers

Pull Requests Commits

Merge branch 'main' into fsdp-vs-ddp

3outeille committed 22 days ago

10c6563e

Expose `per_layer_inputs` for every Gemma4 variants (#45927)

Cyrilvallez committed 22 days ago

Verified ddb841f4

chore: update benchmark_v2.yml (#45966)

hf-security-analysis[bot] committed 22 days ago

Verified d5812ff4

chore: update build-ci-docker-images.yml (#45959)

hf-security-analysis[bot] committed 22 days ago

Verified 84b5d938

fix(ci): set persist-credentials: false on actions/checkout and close remaining template injection findings (#45964)

XciD committed 22 days ago

Verified 9a6a9542

chore(ci): set default workflow permissions to contents: read (#45961)

XciD committed 22 days ago

Verified bc7b4cf9

fix(ci): remove template injection on pull_request_target workflows (#45956)

XciD committed 22 days ago

Verified fbd11e85

chore(ci): pin all GitHub Actions and reusable workflows by SHA (#45955)

XciD committed 22 days ago

Verified a814a243

[docs] ALMModelTest (#45900)

stevhliu committed 23 days ago

Verified 167de6a2

[docs] new attention mask helpers (#45949)

stevhliu committed 23 days ago

Verified d0945699

Enhance apply_chat_template to support custom field prefilling (reasoning_content, thinking, etc.) (#45896)

Mamiglia committed 23 days ago

Verified 98a25186

Pass packed boundary metadata to Qwen3.5 linear-attention fast kernels from data collator (#45034)

sdharani91 committed 23 days ago

Verified 695fc7e7

BUGFIX: Support hubert models that don't have conv_pos_batch_norm configured (#45921)

igordertigor committed 23 days ago

Verified f4757ac3

Revert 45777 (#45942)

Rocketknight1 committed 23 days ago

Verified edcdeb51

pass the otel secrets (#45933)

tarekziade committed 23 days ago

Verified 8c524a84

Extract dynamic vision/audio tensors into standalone pure functions (#45396)

IlyasMoutawwakil committed 23 days ago

Verified f00940ef

Add initial torch_tpu backend support (#45918)

tengomucho committed 23 days ago

Verified 6308a20b

[CB] Hide activation footprint by using the CUDA graph pool (#45911)

remi-or committed 23 days ago

Verified 608c647c

Require input_ids for repetition penalty (#45389)

ruben-aghayan committed 23 days ago

Verified eee7039b

Fix undefined 'input' variable (#45895)

fullyz committed 23 days ago

Verified e51a963d

Deepseek v4 csa mask collapse (#45928)

ArthurZucker committed 23 days ago

Verified 2ad5a9b8

[fix] Add `fatal_error` to `ContinuousBatchingManager` so the serving layer can detect a dead CB worker (#45898)

qgallouedec committed 23 days ago

Verified 9c8a1b83

Fix fsdp2 for gemma models with shared kv states (#45912)

Cyrilvallez committed 23 days ago

Verified 0d476f2c

[docs] update model cards (#45612)

stevhliu committed 24 days ago

Verified 7ee56fc2

torch.backends.fp32_precision cascade conv/rnn so removing the temp fix (#45777)

khushali9 committed 24 days ago

Verified 4968b639

Fix slow Trainer path with 4D attention mask (#45852)

qflen committed 24 days ago

Verified 909ece88

:rotating_light: [`Attn`] Remove all old mask APIs from modeling (#43924)

vasqu committed 24 days ago

Verified 5206626c

fix(text-generation): use token-level slicing for return_full_text=Fa… (#45860)

Abineshabee committed 24 days ago

Verified 91ba4c9a

[CI] AMD docker: bump to ROCm 7.2.2 / PyTorch 2.10 + prebuilt FA wheel (#45913)

Abdennacer-Badaoui committed 24 days ago

Verified 0d34e1ab

Automatically inherit properties from children in composite models (#45909)

Cyrilvallez committed 24 days ago

Verified a25b8efa

Older