Skip to content

[vLLM][moe] Enable DeepGEMM SiLU FP8 moe fallback and update skiplist#7439

Draft
sarah12121212 wants to merge 3 commits into
mainfrom
moe2-deepgemm2
Draft

[vLLM][moe] Enable DeepGEMM SiLU FP8 moe fallback and update skiplist#7439
sarah12121212 wants to merge 3 commits into
mainfrom
moe2-deepgemm2

Conversation

@sarah12121212

@sarah12121212 sarah12121212 commented Jul 14, 2026

Copy link
Copy Markdown
Contributor

This updates the vLLM patch to let the DeepGEMM SiLU FP8 MoE test run on XPU.

The patch:

  • avoids querying CUDA device capability unless the current platform is CUDA
  • lets XPU use the existing Triton fallback path (logic now only enters the DeepGEMM fast path when the platform is CUDA and the CUDA arch is supported)
  • applies the same FP8-boundary tolerance handling on XPU that ROCm already uses
  • unskips test_silu_mul_fp8_quant_deep_gemm in the default and Xe2 MoE skiplists

CI Validation:

PVC:
BMG:

@sarah12121212 sarah12121212 changed the title [vLLM][moe] Enable DeepGEMM SiLU FP8 moe fallback [vLLM][moe] Enable DeepGEMM SiLU FP8 moe fallback and update skiplist Jul 14, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant