karylab_agents/vllm
1
0
Fork 0
forked from Karylab-cklius/vllm
Code Pull Requests 2 Actions 1 Packages Activity

[Bugfix] Re-enable FP8 MoE on NVIDIA Thor (#46339)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2026-06-24 07:35:46 -07:00
committed by GitHub
parent 7dc036058b
commit 24d5186138
2 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
CMakeLists.txt
+2 -2
View File
@@ -869,9 +869,9 @@ if(VLLM_GPU_LANG STREQUAL "CUDA" OR VLLM_GPU_LANG STREQUAL "HIP")
endif()
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 13.0)
cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0f" "${CUDA_ARCHS}")
cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0f;11.0f" "${CUDA_ARCHS}")
else()
cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0a;10.3a" "${CUDA_ARCHS}")
cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0a;10.1a;10.3a" "${CUDA_ARCHS}")
endif()
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8 AND SCALED_MM_ARCHS)
set(CUTLASS_MOE_SM100_SRCS "csrc/libtorch_stable/quantization/w8a8/cutlass/moe/grouped_mm_c3x_sm100.cu")
csrc/libtorch_stable/quantization/w8a8/cutlass/scaled_mm_entry.cu
+1 -1
View File
@@ -180,7 +180,7 @@ bool cutlass_group_gemm_supported(int64_t cuda_device_capability) {
#if defined CUDA_VERSION
#if defined ENABLE_CUTLASS_MOE_SM100 && ENABLE_CUTLASS_MOE_SM100
if (cuda_device_capability >= 100 && cuda_device_capability < 110) {
if (cuda_device_capability >= 100 && cuda_device_capability < 120) {
return CUDA_VERSION >= 12080;
}
#endif