huggingface/transformers

Pull Requests Commits

Decision heads declare the options they can score

apolinario committed 2 hours ago

f98ec181

better names and co

ArthurZucker committed 8 hours ago

bb3f5758

ArthurZucker committed 8 hours ago

b4b30fa5

document the main contribution better

ArthurZucker committed 11 hours ago

92b46a8f

ArthurZucker committed 11 hours ago

7d8ca57d

ArthurZucker committed 11 hours ago

631cdc5d

t checkout main

ArthurZucker committed 15 hours ago

495a7af7

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

qflen committed 15 hours ago

Verified 54c0fe02

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

deadstrobe5 committed 16 hours ago

Verified 49db7636

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

IlyasMoutawwakil committed 16 hours ago

Verified 9e6fdd6b

Merge branch 'main' into add-jev-models

ArthurZucker committed 17 hours ago

Verified a8d5f84a

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

qgallouedec committed 17 hours ago

Verified 4c414256

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

qgallouedec committed 18 hours ago

Verified d8894645

Skip Muse TP (#49717)

remi-or committed 18 hours ago

Verified bf423154

Merge remote-tracking branch 'origin/add-jev-models' into add-jev-models

ArthurZucker committed 18 hours ago

a7705f03

Exclude a foreign base model only when it is a PreTrainedModel; allow keep_in_fp32_modules in config checks

ArthurZucker committed 18 hours ago

73eed592

Merge branch 'main' into add-jev-models

ArthurZucker committed 21 hours ago

Verified 74be39ab

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 3 days ago

Verified 551122ae

Copy default response parser args (#49370)

Rocketknight1 committed 3 days ago

Verified 09281b8d

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

Rocketknight1 committed 3 days ago

Verified 23292ca5

Merge remote-tracking branch 'origin/add-jev-models' into add-jev-models

ArthurZucker committed 3 days ago

d1eee8ee

keep_in_fp32_modules is a common config field; heads call their sub-model backbone

ArthurZucker committed 3 days ago

fc02b75b

Merge branch 'main' into add-jev-models

ArthurZucker committed 3 days ago

Verified 2bf82476

Fix small bug in auto mappings (#49664)

yonigozlan committed 3 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

Older