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, 21:00 UTC
Disable GPTQ AllSpark kernels for CUDA Compiler < 12.0 (#14157)
Signed-off-by: mgoin <mgoin64@gmail.com>
Michael Goin committed
Mar 4, 2025 at 23:25 UTC
e123aafdf003c102440b865faa7598258713b50a
1 file changed
+2
-2
CMakeLists.txt
+2
-2
@@ -319,7 +319,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
319
320
# Only build AllSpark kernels if we are building for at least some compatible archs.
321
cuda_archs_loose_intersection(ALLSPARK_ARCHS "8.0;8.6;8.7;8.9" "${CUDA_ARCHS}")
322
- if (ALLSPARK_ARCHS)
322
+ if (${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.0 AND ALLSPARK_ARCHS)
323
set(ALLSPARK_SRCS
324
"csrc/quantization/gptq_allspark/allspark_repack.cu"
325
"csrc/quantization/gptq_allspark/allspark_qgemm_w8a16.cu")
@@ -330,7 +330,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
330
message(STATUS "Building AllSpark kernels for archs: ${ALLSPARK_ARCHS}")
331
else()
332
message(STATUS "Not building AllSpark kernels as no compatible archs found"
333
- " in CUDA target architectures")
333
+ " in CUDA target architectures, or CUDA not >= 12.0")
334
endif()
335
336
# The cutlass_scaled_mm kernels for Hopper (c3x, i.e. CUTLASS 3.x) require