frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 14:53:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Build moe_data for both sm100 and sm90 (#20086)

Browse Source 
Signed-off-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
Michael Goin
2025-06-27 12:50:12 +09:00
committed by GitHub
parent 2d7779f888
commit 44d2e6af63
2 changed files with 17 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
CMakeLists.txt
View File

@ -513,6 +513,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
CUDA_ARCHS "${FP4_ARCHS}")
list(APPEND VLLM_EXT_SRC "${SRCS}")
list(APPEND VLLM_GPU_FLAGS "-DENABLE_NVFP4=1")
list(APPEND VLLM_GPU_FLAGS "-DENABLE_CUTLASS_MOE_SM100=1")
message(STATUS "Building NVFP4 for archs: ${FP4_ARCHS}")
else()
message(STATUS "Not building NVFP4 as no compatible archs were found.")
@ -547,8 +548,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
# if it's possible to compile MoE kernels that use its output.
cuda_archs_loose_intersection(SCALED_MM_ARCHS "9.0a" "${CUDA_ARCHS}")
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.3 AND SCALED_MM_ARCHS)
set(SRCS "csrc/quantization/cutlass_w8a8/moe/grouped_mm_c3x.cu"
"csrc/quantization/cutlass_w8a8/moe/moe_data.cu")
set(SRCS "csrc/quantization/cutlass_w8a8/moe/grouped_mm_c3x.cu")
set_gencode_flags_for_srcs(
SRCS "${SRCS}"
CUDA_ARCHS "${SCALED_MM_ARCHS}")
@ -566,6 +566,16 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
endif()
endif()
# moe_data.cu is used by all CUTLASS MoE kernels.
cuda_archs_loose_intersection(CUTLASS_MOE_DATA_ARCHS "9.0a;10.0a" "${CUDA_ARCHS}")
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.3 AND CUTLASS_MOE_DATA_ARCHS)
set(SRCS "csrc/quantization/cutlass_w8a8/moe/moe_data.cu")
set_gencode_flags_for_srcs(
SRCS "${SRCS}"
CUDA_ARCHS "${CUTLASS_MOE_DATA_ARCHS}")
list(APPEND VLLM_EXT_SRC "${SRCS}")
endif()
#
# Machete kernels