huggingface/transformers

Pull Requests Commits

Train models sharded at load time with the Trainer, token dispatch included

3outeille committed 4 days ago

d4e6a056

Add expert-parallel token dispatch and make it Qwen3 MoE's default

3outeille committed 4 days ago

8144c259

Separate the tensor and expert parallel plans

3outeille committed 4 days ago

fbb99855

Enable expert parallelism in the Mega MoE example

3outeille committed 4 days ago

bb6cbe94

Use ep_size in the Kimi K2.5 expert parallel example

3outeille committed 4 days ago

e8f9ed88

3outeille committed 4 days ago

79fc7e51

3outeille committed 4 days ago

d2172f4a

3outeille committed 4 days ago

6ac0a63d

3outeille committed 4 days ago

1272710f

Add dense and expert device mesh views with a MeshManager

3outeille committed 4 days ago

cef703ee

Add test for causal only variant of some encoder-decoder models (#48760)

nandan2003 committed 4 days ago

Verified c9d531dd

Fix kosmos (#48853)

Cyrilvallez committed 4 days ago

Verified 947d6b4a

[generate] Make all methods and logits processors agnostic to lm_head output size (#48846)

Cyrilvallez committed 4 days ago

Verified b180f143

Fix `reset` on the dynamic cache layers (#48809)

jiqing-feng committed 4 days ago

Verified a5d98c34

QA: deactivate rule 41 (#48852)

tarekziade committed 4 days ago

Verified 031cfc58

[GPTNeoXJapanese] Fix RoPE ignoring partial_rotary_factor (#48652)

blipbyte committed 4 days ago

Verified 2d92ded1

docs: fix broken #combining-with-fsdp2 anchor in expert_parallelism.md (#48854)

ydshieh committed 4 days ago

Verified 51b330bb

Qwen3.8 GGUF (#48660)

SunMarc committed 5 days ago

Verified 41f519e7

[CI] Add link checker (#48160)

stevhliu committed 5 days ago

Verified e9af3a32

🚨 🚨Bring some dinos to modern standards (#46266)

molbap committed 5 days ago

Verified f2f6074b

[apply_chat_template] pass sampling_rate to call (#48794)

eustlb committed 5 days ago

Verified 3dec062f

[docs] gguf (#46357)

stevhliu committed 5 days ago

Verified 626e7ff9

Support for MoE in the GGUF integration (#48529)

SunMarc committed 5 days ago

Verified 26d8e716

Detect nested rope_parameters without relying on layer_types (#48798)

hmellor committed 5 days ago

Verified cc269e76

Relax static-cache tolerance in test_generate_with_static_cache (1e-5 → 5e-5) (#48815)

ydshieh committed 5 days ago

Verified a4aa9890

Pin utf-8 in test_can_init_all_missing_weights source read (Windows non-UTF-8 locale fix) (#48819)

dltsum committed 5 days ago

Verified 3f162ca5

[generate] Drop attention mask early without padding (#48814)

Cyrilvallez committed 5 days ago

Verified d65fdc5f

[utils] Add MUSA support for Flash Attention 2 (#48612)

XiaomingFun233 committed 5 days ago

Verified a56a23f3

:rotating_light: [`Kernels`] Bump version (#48714)

vasqu committed 5 days ago

Verified bcb9849a

[Fix] Clean-up ternaries in the DeepSeek family (#48447)

remi-or committed 5 days ago

Verified 6cd7ae39

Older