EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Optimize cutlass int8 gemm kernel for large M on SM89 Ada GPU (#10714)

Browse Source
This commit is contained in:
Qi Yuhang
2025-09-22 08:04:27 +08:00
committed by GitHub
parent 2f18602f13
commit 0f04a5f428
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
sgl-kernel/csrc/gemm/int8_gemm_kernel.cu
View File

@@ -409,8 +409,8 @@ void sm89_dispatch_shape(
cutlass_int8_scaled_mm< cutlass_int8_scaled_mm<
ElementOutput, ElementOutput,
ArchTag, ArchTag,
cutlass::gemm::GemmShape<32, 64, 128>, cutlass::gemm::GemmShape<128, 128, 64>,
cutlass::gemm::GemmShape<16, 64, 64>, cutlass::gemm::GemmShape<64, 64, 64>,
InstructionShape, InstructionShape,
5>(out, mat_a, mat_b, scales_a, scales_b, bias); 5>(out, mat_a, mat_b, scales_a, scales_b, bias);
} }