Pull Requests huggingface/transformers

Fix xLSTM train mode crash when return_last_states=False

#47026 opened 2026-07-02 14:13 by lcheng321

Diffusion gemma: fix failed test cases

#47025 opened 2026-07-02 14:07 by kaixuanliu

test: add SDPA gradient health tests (fixes #44928)

#47024 opened 2026-07-02 13:28 by Lemniscate-world

Bump the actions group across 1 directory with 8 updates dependencies github_actions

#47023 opened 2026-07-02 13:21 by dependabot[bot]

CI Add PEFT integration tests

#47021 opened 2026-07-02 09:40 by BenjaminBossan

fix mask return-type contract regression and add correctness guard for

#47019 opened 2026-07-02 08:36 by kaixuanliu

Fix save_pretrained with offloading and weight conversions

#47018 opened 2026-07-02 08:14 by Cyrilvallez

Route byte-level llama tokenizers to TokenizersBackend

#47017 opened 2026-07-02 05:11 by subin9

Fix RTDetrHungarianMatcher crash on infinite cost matrix

#47016 opened 2026-07-02 04:55 by AyushDas4890

🚨 Enable SDPA (and other attention backends) for T5 and propagate to the T5 family

#47014 opened 2026-07-02 01:54 by jiqing-feng

Add max_groups input to nightly Serge triage caller (single-task test mode)

#47012 opened 2026-07-01 19:56 by tarekziade

Add linear_attn entries to Qwen3.5 base_model_tp_plan

#47009 opened 2026-07-01 18:24 by ZAID646

[serge] Fix 10 integration tests for model `gemma3` failing with `output_mismatch` (list output differs (8), other (2))

#47006 opened 2026-07-01 15:41 by sergereview[bot]

Add tiny_model_id support to ProcessorTesterMixin for memory-sensitive tests

#47005 opened 2026-07-01 15:41 by ydshieh

OpenVINO HF Exporter

#47003 opened 2026-07-01 14:08 by IlyasMoutawwakil

Add after-load fusion for static quantized MLPs

#46997 opened 2026-07-01 10:41 by LiangSu8899

Add MLU support to is_flash_linear_attention_available

#46995 opened 2026-07-01 09:25 by atri2549

Raise a clear error for empty token lists in bad_words_ids / sequence_bias

#46994 opened 2026-07-01 06:58 by Sunt-ing

Fix prompt lookup decoding generating past max_length

#46993 opened 2026-07-01 06:42 by Sunt-ing

FSDP orchestration: mesh init, distribute-before-load, DCP save

#46990 opened 2026-07-01 05:09 by 3outeille

Add QNN (Qualcomm HTP) backend to the ExecuTorch exporter

#46989 opened 2026-07-01 03:58 by psiddh

[serge] Fix 10 integration tests for model `glm_ocr` failing with `output_mismatch` (list output differs (10))

#46986 opened 2026-07-01 00:55 by sergereview[bot]

Fix typo in `MusicgenForCausalLM.generate()`

#46974 opened 2026-06-30 04:55 by jiqing-feng

[serge] Fix 8 integration tests for model `mamba2` failing with `OOM` (other (8))

#46971 opened 2026-06-30 00:44 by sergereview[bot]

[serge] Fix 16 integration tests for model `musicgen_melody` failing with `output_mismatch` (tensor values differ (16))

#46970 opened 2026-06-30 00:38 by sergereview[bot]

[docs] fix autolinks

#46968 opened 2026-06-29 22:55 by stevhliu

Add Gemma4Unified sequence classification

#46966 opened 2026-06-29 19:04 by yuvrajsharma9981

[Model] Support `use_cache=False` for DeepSeek V4

#46965 opened 2026-06-29 18:11 by kylesayrs

make sure serge produces clean patches

#46961 opened 2026-06-29 15:00 by tarekziade

Fix silent SDPA math-kernel fallback for GQA when key/value head_dim > 256 or differ

#46960 opened 2026-06-29 14:31 by Butterfingrz