Skip to content

GitLab

Unverified Commit bc956b38 authored by Huy Do's avatar Huy Do Committed by GitHub
Browse files

Only build CUTLASS MoE kernels on Hopper (#19648)

parent 294fc1e2
Hide whitespace changes
Inline Side-by-side
Showing with 2 additions and 2 deletions
CMakeLists.txt
View file @ bc956b38
...@@ -542,10 +542,10 @@ if(VLLM_GPU_LANG STREQUAL "CUDA") ...@@ -542,10 +542,10 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
# CUTLASS MoE kernels # CUTLASS MoE kernels
# The MoE kernel cutlass_moe_mm requires CUDA 12.3 or later (and only works # The MoE kernel cutlass_moe_mm requires CUDA 12.3 or later (and ONLY works
# on Hopper). get_cutlass_(pplx_)moe_mm_data should only be compiled # on Hopper). get_cutlass_(pplx_)moe_mm_data should only be compiled
# if it's possible to compile MoE kernels that use its output. # if it's possible to compile MoE kernels that use its output.
cuda_archs_loose_intersection(SCALED_MM_ARCHS "9.0a;10.0a" "${CUDA_ARCHS}") cuda_archs_loose_intersection(SCALED_MM_ARCHS "9.0a" "${CUDA_ARCHS}")
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.3 AND SCALED_MM_ARCHS) if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.3 AND SCALED_MM_ARCHS)
set(SRCS "csrc/quantization/cutlass_w8a8/moe/grouped_mm_c3x.cu" set(SRCS "csrc/quantization/cutlass_w8a8/moe/grouped_mm_c3x.cu"
"csrc/quantization/cutlass_w8a8/moe/moe_data.cu") "csrc/quantization/cutlass_w8a8/moe/moe_data.cu")
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment