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

[1/2] Add Kernel support for Cutlass based Fused FP4 MoE (#6093)

Browse Source 
Signed-off-by: Pavani Majety <pmajety@nvidia.com>
This commit is contained in:
Pavani Majety
2025-06-02 13:48:03 -07:00
committed by GitHub
parent df7f61ee7d
commit eb38c7d1ca
12 changed files with 1677 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
sgl-kernel/CMakeLists.txt
View File

@@ -210,6 +210,7 @@ set(SOURCES
"csrc/gemm/fp8_blockwise_gemm_kernel.cu"
"csrc/gemm/fp8_gemm_kernel.cu"
"csrc/gemm/int8_gemm_kernel.cu"
"csrc/gemm/nvfp4_expert_quant.cu"
"csrc/gemm/nvfp4_quant_entry.cu"
"csrc/gemm/nvfp4_quant_kernels.cu"
"csrc/gemm/nvfp4_scaled_mm_entry.cu"
@@ -222,6 +223,7 @@ set(SOURCES
"csrc/moe/moe_align_kernel.cu"
"csrc/moe/moe_fused_gate.cu"
"csrc/moe/moe_topk_softmax_kernels.cu"
"csrc/moe/nvfp4_blockwise_moe.cu"
"csrc/moe/fp8_blockwise_moe_kernel.cu"
"csrc/moe/prepare_moe_input.cu"
"csrc/moe/ep_moe_reorder_kernel.cu"