huggingface/transformers

Pull Requests Commits

Merge ep-mxfp4-gpt-oss: keep `EpRouterParallel` removed

IlyasMoutawwakil committed 6 days ago

80394eb5

Merge remote-tracking branch 'origin/ep-move-router-plans' into ep-mxfp4-gpt-oss

IlyasMoutawwakil committed 6 days ago

3e1f7e02

Merge remote-tracking branch 'origin/ep-passthrough-strategy' into ep-move-router-plans

IlyasMoutawwakil committed 6 days ago

cdec5961

Merge ep-masked-when-shared-batch: register Mega MoE in `ep_implementations`

IlyasMoutawwakil committed 6 days ago

9c30fd2f

EP: inline the shared-batch check into EPImplementation

IlyasMoutawwakil committed 6 days ago

6eae29f4

EP: register implementations in a dict on EpDispatchExpertsParallel

IlyasMoutawwakil committed 6 days ago

d8763955

EP implementations are callables; inline the backward all-reduce check

IlyasMoutawwakil committed 7 days ago

704f9114

EP implementations: one forward instead of dispatch + combine

IlyasMoutawwakil committed 7 days ago

a3365657

Merge the Mega MoE docs fix

IlyasMoutawwakil committed 9 days ago

3a37bd24

Merge the Mega MoE docs fix

IlyasMoutawwakil committed 9 days ago

8c3c962a

Merge the Mega MoE docs fix

IlyasMoutawwakil committed 9 days ago

38326cd5

Drop the selection sentence from the Mega MoE docs row

IlyasMoutawwakil committed 9 days ago

e657e1d0

Merge the updated MXFP4 GPT-OSS PR

IlyasMoutawwakil committed 9 days ago

2cbebd45

Drop Llama 4's EP plan, which never applied

IlyasMoutawwakil committed 9 days ago

4bad170d

Remove a debug print

IlyasMoutawwakil committed 9 days ago

f7c5fd56

Merge the updated MXFP4 GPT-OSS PR

IlyasMoutawwakil committed 9 days ago

90570c85

MXFP4 GPT-OSS: check the experts for the EP hook, declare packed params 3-D so EP shards the scales

IlyasMoutawwakil committed 9 days ago

13eda643

Merge the updated router-plans PR

IlyasMoutawwakil committed 9 days ago

c78963a4

Drop Beam's router-masked plan comment, and Gemma 4 from the masking defaults in the docs

IlyasMoutawwakil committed 9 days ago

ca93beaa

Merge the updated Mega MoE PR

IlyasMoutawwakil committed 9 days ago

7aff59f2

Test the Mega MoE implementation on two ranks

IlyasMoutawwakil committed 9 days ago

05d5d97d

Drop the passthrough implementation, resolve Mega MoE buffers at init

IlyasMoutawwakil committed 9 days ago

bba44080

Merge the updated EP implementations PR

IlyasMoutawwakil committed 9 days ago

9fc2d10d

Say what each EP implementation does in short sentences

IlyasMoutawwakil committed 9 days ago

1710f6c5

Merge main: the plans are resolved once in the model (#49527)

IlyasMoutawwakil committed 9 days ago

6f72a0ea

Fix OOM regression for FSDP2 + cpu_ram_efficient_loading on large models (#45649)

AmineDiro committed 9 days ago

Verified 5949d2c4

[serge] Fix 2 integration tests regressed by commit 7d3ae42ee810 (PR #49811) (#50098)

sergereview[bot] committed 9 days ago

Verified 763b4ba5

Fix resize_embeddings tests for deepspeed (#49909)

kaixuanliu committed 9 days ago

Verified 28aa8646

[VibeVoice] Add `sample` argument to disable VAE noise in audio features (#50179)

hmellor committed 9 days ago

Verified 2f2e711f

fix failed test cases for pocket_tts model (#50159)

kaixuanliu committed 9 days ago

Verified 3e8d3971

Older