Repository navigation

[MoE] Unify DeepGEMM MegaMoE experts behind one feature-matrix layer; add NVFP4 - #61538

Merged

zyongye merged 7 commits into

vllm-project:mainfrom

zyongye:refactor/deep-gemm-mega-moe-experts

Oct 15, 2026

Merged

[MoE] Unify DeepGEMM MegaMoE experts behind one feature-matrix layer; add NVFP4#61538
zyongye merged 7 commits into
vllm-project:mainfrom
zyongye:refactor/deep-gemm-mega-moe-experts

Commits on Oct 14, 2026
- [MoE] Unify DeepGEMM MegaMoE experts behind one layer with a feature matrix
  
  zyongye
  and
  claude
  committed
  f369e66
- [MoE] Add NVFP4 routed experts with MXFP8 or BF16 shared experts to MegaMoE
  
  zyongye
  and
  claude
  committed
  127fcef
- [MoE] MegaMoE: run kernel tests on B200 CI, reloadable NVFP4 L2 scale
  
  zyongye
  and
  claude
  committed
  ce8a54b
- [MoE] MegaMoE NVFP4: skip meta loads in host scale state, check w1/w3 weight_scale_2
  
  zyongye
  and
  claude
  committed
  603db01
- Merge branch 'main' into refactor/deep-gemm-mega-moe-experts
  zyongye
  authored
  5550180
Commits on Oct 15, 2026
- [MoE] MegaMoE: run the input staging tests on SM100 only
  
  zyongye
  and
  claude
  committed
  c587e24
- Merge branch 'main' into refactor/deep-gemm-mega-moe-experts
  khluu
  authored
  d83c149