support blockwise fp8 matmul kernel (#3267)

This commit is contained in:
yizhang2077
2025-02-13 01:49:33 +08:00
committed by GitHub
parent 8616357a97
commit 640363ad20
11 changed files with 1366 additions and 0 deletions

View File

@@ -96,6 +96,7 @@ sources = [
"src/sgl-kernel/csrc/moe_align_kernel.cu",
"src/sgl-kernel/csrc/int8_gemm_kernel.cu",
"src/sgl-kernel/csrc/fp8_gemm_kernel.cu",
"src/sgl-kernel/csrc/fp8_blockwise_gemm_kernel.cu",
"src/sgl-kernel/csrc/lightning_attention_decode_kernel.cu",
"src/sgl-kernel/csrc/fused_add_rms_norm_kernel.cu",
"src/sgl-kernel/csrc/eagle_utils.cu",