huggingface/transformers

Pull Requests Commits

Isolate MLA KV expansion to make it easier to bypass (#47460)

hmellor committed 52 days ago

Verified 6b0f8227

Update serve chat parsing (#46267)

SunMarc committed 52 days ago

Verified 12f2d155

Fix loss alignment and Trainer token counting for encoder decoder models (#46903)

OmkumarSolanki committed 52 days ago

Verified 0c1829dd

Fix Gemma4 audio feature dtype mismatch in masked_scatter (#47482)

danielhanchen committed 52 days ago

Verified bc14d7e6

Fix the HunyuanVL's torchvision backend (#47499)

Mi-Jiazhi committed 52 days ago

Verified 60570c1f

Use new `per_layer_config` for Gemma 4 so that heterogeneous attention config is explicit (#47384)

hmellor committed 52 days ago

Verified 7d5bc4f3

[DiffusionGemma] Support gradient checkpointing (#46572)

kashif committed 53 days ago

Verified 213652a5

Fix failing tests for zaya (#47268)

kaixuanliu committed 53 days ago

Verified 56af69c7

tipsv2_dpt: fix failing tests for XPU (#47292)

kaixuanliu committed 53 days ago

Verified 486a22aa

Fix some failed test cases related with XPU Expectations (#47173)

kaixuanliu committed 53 days ago

Verified e2102935

add paged attention tests support for XPU (#47163)

kaixuanliu committed 53 days ago

Verified 6168ead6

Add FSDP CI and end-to-end FSDP tests + save fsdp (#47357)

3outeille committed 53 days ago

Verified 9de8a4d2

Merge branch 'split/a-pr-4-2-fsdp-tests-ci' into split/a-pr-5-fsdp-plans

3outeille committed 53 days ago

Verified a1b0d779

Merge branch 'main' into split/a-pr-4-2-fsdp-tests-ci

3outeille committed 53 days ago

Verified 8c96054e

fix: guard DTensor import in sharding_utils.py for PyTorch < 2.5 (#47481)

gygdh-001 committed 53 days ago

Verified c0bd9e62

Merge branch 'main' into split/a-pr-4-2-fsdp-tests-ci

3outeille committed 53 days ago

Verified 047bd41d

CPU can incur a slow path on non-contiguous magnitudes (#47351)

vbayanag committed 53 days ago

Verified bb3ffb97

[fix] fix requirements audio feature and proc (#47113)

eustlb committed 53 days ago

Verified 2df80c4f

[Quantization]: Refactor is_quantization_compressed for format-based detection (#47152)

rigen1048 committed 53 days ago

Verified c0142e6f

Route chat management calls to the service root (#47138) (#47303)

dhruv7477 committed 53 days ago

Verified ffae066d

fix: liger unnecessarily materializes logits in VRAM during eval, causing OOM (#45273)

excepshenal committed 53 days ago

Verified 0dad5454

Fix gradient inflation when combining label smoothing with gradient accumulation (#47261)

Incheonkirin committed 53 days ago

Verified 2da6655e

Fix MoE expert decompression for non-32-divisible bit widths (#47315)

KKothuri committed 53 days ago

Verified d0db27f8

[Qwen3ASR] Add hotword parsing, and fix language parsing and training. (#47111)

ebezzam committed 53 days ago

Verified c7f9c881

Remove deprecated training args and `is_fast` property (#46917)

cyyever committed 53 days ago

Verified b6297c00

Consistent output shape from `get_image_features` (#46405)

zucchini-nlp committed 54 days ago

Verified 378d2f61

Fix multi-device mxfp4 dequantization race in `_convert_moe_packed_tensors` (#47423)

kaixuanliu committed 54 days ago

Verified 6b72a982

fix failed test cases for qwen3_omni_moe model (#47449)

kaixuanliu committed 54 days ago

Verified 9f5638e5

Fix Hunyuan-VL PIL image resize parity with reference preprocessing (#47233)

IMvision12 committed 54 days ago

Verified 3ac317a2

Move `value` padding into the attention interfaces that need it (#47451)

hmellor committed 54 days ago

Verified 65d834fb

Newer Older