A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 06:00 UTC
Remove sm120 arch from sm100 cutlass kernel arch list (#19716)
Signed-off-by: mgoin <mgoin64@gmail.com>
Michael Goin committed
Jun 18, 2025 at 03:49 UTC
bf57ccc5c2102a77d70431b2fe47e6fde4acc600
1 file changed
+3
-3
CMakeLists.txt
+3
-3
@@ -420,9 +420,9 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
420
endif()
421
endif()
422
423
- # The cutlass_scaled_mm kernels for Blackwell (c3x, i.e. CUTLASS 3.x) require
424
- # CUDA 12.8 or later
425
- cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0a;10.1a;12.0a" "${CUDA_ARCHS}")
423
+ # The cutlass_scaled_mm kernels for Blackwell SM100 (c3x, i.e. CUTLASS 3.x)
424
+ # require CUDA 12.8 or later
425
+ cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0a;10.1a" "${CUDA_ARCHS}")
426
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.8 AND SCALED_MM_ARCHS)
427
set(SRCS
428
"csrc/quantization/cutlass_w8a8/scaled_mm_c3x_sm100.cu"