Support cutlass Int8 gemm (#2752)

This commit is contained in:
Ke Bao
2025-01-06 22:51:22 +08:00
committed by GitHub
parent 06dd2eab84
commit 0f3eb1d294
12 changed files with 1434 additions and 0 deletions

View File

@@ -26,6 +26,7 @@ cutlass = root / "3rdparty" / "cutlass"
include_dirs = [
cutlass.resolve() / "include",
cutlass.resolve() / "tools" / "util" / "include",
root / "src" / "sgl-kernel" / "csrc",
]
nvcc_flags = [
"-O3",
@@ -48,6 +49,7 @@ ext_modules = [
"src/sgl-kernel/csrc/trt_reduce_internal.cu",
"src/sgl-kernel/csrc/trt_reduce_kernel.cu",
"src/sgl-kernel/csrc/moe_align_kernel.cu",
"src/sgl-kernel/csrc/int8_gemm_kernel.cu",
"src/sgl-kernel/csrc/sgl_kernel_ops.cu",
],
include_dirs=include_dirs,