huggingface/transformers

Pull Requests Commits

Merge branch 'main' into fix-generate

Cyrilvallez committed 28 minutes ago

Verified 385b67bf

Fix ValueError when loading a fusion-enabled checkpoint twice (#49412)

kaixuanliu committed 32 minutes ago

Verified 01269bc9

Allow Cache continuation for chunked prefill (#49662)

Cyrilvallez committed 33 minutes ago

Verified 90a1fdb6

Cyrilvallez committed 1 hour ago

Verified 82c4f7c1

Cleaned up tokenizer tests (#49669)

tarekziade committed 1 hour ago

Verified 40b8b28b

Avoid a host-to-device copy in `eager_mask` (#49684)

hmellor committed 2 hours ago

Verified 5441d1ed

`is_tracing()` now recognizes XPU stream capture, alongside the exist… (#49722)

sywangyi committed 3 hours ago

Verified 8fa11ce8

🚨 TP: give embeddings their own style, shard input embeddings by default, and shard widths that don't divide by the TP size (#49570)

IlyasMoutawwakil committed 3 hours ago

Verified 718d12a9

Document how to avoid the Metaspace prefix after added tokens (#45822)

qflen committed 6 hours ago

Verified 54c0fe02

Load tokenizer.json once in TokenizersBackend (2-2.9x faster AutoTokenizer.from_pretrained) (#49667)

deadstrobe5 committed 7 hours ago

Verified 49db7636

Ernie4.5-VL-MoE: shard the text and vision experts under TP and EP (#49462)

IlyasMoutawwakil committed 7 hours ago

Verified 9e6fdd6b

Declare the input modalities of the VLMs that report text only (#49677)

qgallouedec committed 8 hours ago

Verified 4c414256

Fix zero expert gradients in bfloat16 with expert parallelism and ep_size < fsdp_size (#49688)

qgallouedec committed 8 hours ago

Verified d8894645

Skip Muse TP (#49717)

remi-or committed 9 hours ago

Verified bf423154

Fix test selection for virtual module imports (#49690)

tarekziade committed 2 days ago

Verified d4804f9d

🚨 MoE: shard the experts in every TP/EP plan, and test for it (#49206)

IlyasMoutawwakil committed 2 days ago

Verified 8498ed42

[docs] add Google TPU hardware guide (#49642)

tengomucho committed 2 days ago

Verified 551122ae

Copy default response parser args (#49370)

Rocketknight1 committed 2 days ago

Verified 09281b8d

Serve legacy module aliases through a sys.meta_path finder (#48992)

Rocketknight1 committed 2 days ago

Verified 23292ca5

Fix small bug in auto mappings (#49664)

yonigozlan committed 2 days ago

Verified 966f59c8

Make `DtensorShardOperation` work on torch 2.6 and 2.7 (#49657)

hazelxue committed 3 days ago

Verified 7bf3c5ec

Fix from_pretrained on torch 2.7: require torch>=2.8 for distributed checkpointing (#49652)

naveecode committed 3 days ago

Verified f74cb53b

Fix DeepGEMM Mega MoE under expert parallelism (#49428)

IlyasMoutawwakil committed 3 days ago

Verified eb943c26

Fix HunYuan-VL image assignment during batched generation (#49557)

Arnav66692 committed 3 days ago

Verified 742d8474

Fix flaky Evolla tests producing NaN outputs (#49660)

tarekziade committed 3 days ago

Verified a8262af9

Quantization cleanpup (#49548)

ArthurZucker committed 3 days ago

Verified b7c41684

[rocm] Add ROCm expectations to Kolibri-1 logits test (#49661)

ericschreiber committed 3 days ago

Verified e5c1aaee

[CB] Various small fixes for MPS (#49654)

remi-or committed 3 days ago

Verified 364abe43

Aligns test expectations with the accelerator path under test: (#49653)

sywangyi committed 3 days ago

Verified 9fb4b5a9

[docs] Processor config (#49643)

stevhliu committed 3 days ago

Verified 683b7219

Older