[Misc][Kernel]: Add GPTQAllSpark Quantization (#12931)

This commit is contained in:
YajieWang
2025-02-28 22:30:59 -08:00
committed by GitHub
parent 6a84164add
commit 6a92ff93e1
12 changed files with 2005 additions and 4 deletions
Executable → Regular
+16
View File
@@ -317,6 +317,22 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
" in CUDA target architectures")
endif()
# Only build AllSpark kernels if we are building for at least some compatible archs.
cuda_archs_loose_intersection(ALLSPARK_ARCHS "8.0;8.6;8.7;8.9" "${CUDA_ARCHS}")
if (ALLSPARK_ARCHS)
set(ALLSPARK_SRCS
"csrc/quantization/gptq_allspark/allspark_repack.cu"
"csrc/quantization/gptq_allspark/allspark_qgemm_w8a16.cu")
set_gencode_flags_for_srcs(
SRCS "${ALLSPARK_SRCS}"
CUDA_ARCHS "${ALLSPARK_ARCHS}")
list(APPEND VLLM_EXT_SRC "${ALLSPARK_SRCS}")
message(STATUS "Building AllSpark kernels for archs: ${ALLSPARK_ARCHS}")
else()
message(STATUS "Not building AllSpark kernels as no compatible archs found"
" in CUDA target architectures")
endif()
# The cutlass_scaled_mm kernels for Hopper (c3x, i.e. CUTLASS 3.x) require
# CUDA 12.0 or later (and only work on Hopper, 9.0a for now).
cuda_archs_loose_intersection(SCALED_MM_3X_ARCHS "9.0a" "${CUDA_ARCHS}")