xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-07-20 05:07:10 +08:00
Code Issues Packages Projects Releases Wiki Activity

Only build CUTLASS MoE kernels on Hopper (#19648)

Browse Source
This commit is contained in:
Huy Do 2025-06-14 11:44:15 -07:00 committed by GitHub
parent 294fc1e2c9
commit bc956b38d0
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
CMakeLists.txt
View File

@ -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")