xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-04-12 02:57:05 +08:00
Code Issues Packages Projects Releases Wiki Activity

minor cleanup

Browse Source 
Signed-off-by: Lucas Wilkinson <lwilkinson@neuralmagic.com>
This commit is contained in:
Lucas Wilkinson 2025-06-08 20:14:51 +00:00
parent e7f00f3368
commit 5d3b0bc39c
2 changed files with 8 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
CMakeLists.txt
View File

@ -345,7 +345,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
"csrc/quantization/cutlass_w8a8/c3x/scaled_mm_blockwise_sm90_fp8.cu"
FLAGS "-DENABLE_SCALED_MM_SM90=1"
VERSION_MSG
"Not building scaled_mm_c3x_sm90: CUDA Compiler version is not >= 12.0.\n"
"Not building scaled_mm_c3x_sm90: CUDA Compiler version is not >= 12.0."
"Please upgrade to CUDA 12.0 or later to run FP8 quantized models on Hopper."
)
@ -360,7 +360,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
"csrc/quantization/cutlass_w8a8/c3x/scaled_mm_blockwise_sm100_fp8.cu"
FLAGS "-DENABLE_SCALED_MM_SM100=1"
VERSION_MSG
"Not building scaled_mm_c3x_sm100: CUDA Compiler version is not >= 12.8.\n"
"Not building scaled_mm_c3x_sm100: CUDA Compiler version is not >= 12.8."
"Please upgrade to CUDA 12.8 or later to run FP8 quantized models on Blackwell."
)
@ -370,8 +370,6 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
ARCHS "7.5;8.0;8.9+PTX"
SRCS "csrc/quantization/cutlass_w8a8/scaled_mm_c2x.cu"
FLAGS "-DENABLE_SCALED_MM_C2X=1"
NO_ARCH_MSG "Not building scaled_mm_c2x as no compatible archs found in CUDA target architectures, "
"or is already covered by scaled_mm_c3x."
)
#
@ -383,7 +381,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
SRCS "csrc/sparse/cutlass/sparse_scaled_mm_c3x.cu"
FLAGS "-DENABLE_SPARSE_SCALED_MM_C3X=1"
VERSION_MSG
"Not building sparse_scaled_mm_c3x: CUDA Compiler version is not >= 12.2.\n"
"Not building sparse_scaled_mm_c3x: CUDA Compiler version is not >= 12.2."
"Please upgrade to CUDA 12.2 or later to run FP8 sparse quantized models on Hopper."
)
@ -423,7 +421,9 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
"csrc/quantization/cutlass_w8a8/moe/grouped_mm_c3x.cu"
"csrc/quantization/cutlass_w8a8/moe/moe_data.cu"
FLAGS "-DENABLE_CUTLASS_MOE_SM90=1"
VERSION_MSG "Not building grouped_mm_c3x kernels as CUDA Compiler version is not >= 12.3, we recommend upgrading to CUDA 12.3 or later if you intend on running FP8 quantized MoE models on Hopper."
VERSION_MSG
"Not building grouped_mm_c3x kernels as CUDA Compiler is less than 12.3."
"We recommend upgrading to CUDA 12.3 or later if you intend on running FP8 quantized MoE models on Hopper."
)
#

4
cmake/utils.cmake
View File

@ -176,7 +176,7 @@ macro(optional_cuda_sources)
message(STATUS "Building ${OCS_NAME} for archs: ${_OCS_ARCHS}")
else()
if(OCS_NO_ARCH_MSG)
list(JOIN OCS_NO_ARCH_MSG "\n" _OCS_NO_ARCH_JOINED)
list(JOIN OCS_NO_ARCH_MSG "\n " _OCS_NO_ARCH_JOINED)
message(STATUS "${_OCS_NO_ARCH_JOINED}")
else()
message(STATUS "Not building ${OCS_NAME}: no compatible architectures found in CUDA target architectures")
@ -184,7 +184,7 @@ macro(optional_cuda_sources)
endif()
else()
if(OCS_VERSION_MSG)
list(JOIN OCS_VERSION_MSG "\n" _OCS_VERSION_JOINED)
list(JOIN OCS_VERSION_MSG "\n " _OCS_VERSION_JOINED)
message(STATUS "${_OCS_VERSION_JOINED}")
else()
message(STATUS "Not building ${OCS_NAME}: CUDA Compiler version is less than ${OCS_MIN_VERSION}")