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

[NVIDIA] [2/N] Optimize silu_and_mul_scaled_fp4_grouped_quant perf (#9556)

Browse Source
This commit is contained in:
Kaixi Hou
2025-08-29 17:17:03 -07:00
committed by GitHub
parent ff9b561817
commit 5c34b4f1c7
7 changed files with 297 additions and 61 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
sgl-kernel/csrc/gemm/nvfp4_quant_entry.cu
View File

@@ -32,9 +32,8 @@ void silu_and_mul_scaled_fp4_experts_quant_sm100a(
torch::Tensor& output_scale,
torch::Tensor const& input,
torch::Tensor const& input_global_scale,
torch::Tensor const& input_offset_by_experts,
torch::Tensor const& output_scale_offset_by_experts,
torch::Tensor const& mask);
torch::Tensor const& mask,
bool use_silu_and_mul);
#endif
@@ -65,12 +64,11 @@ void silu_and_mul_scaled_fp4_experts_quant(
torch::Tensor& output_scale,
torch::Tensor const& input,
torch::Tensor const& input_global_scale,
torch::Tensor const& input_offset_by_experts,
torch::Tensor const& output_scale_offset_by_experts,
torch::Tensor const& mask) {
torch::Tensor const& mask,
bool use_silu_and_mul) {
#if defined ENABLE_NVFP4 && ENABLE_NVFP4
return silu_and_mul_scaled_fp4_experts_quant_sm100a(
output, output_scale, input, input_global_scale, input_offset_by_experts, output_scale_offset_by_experts, mask);
output, output_scale, input, input_global_scale, mask, use_silu_and_mul);
#endif
TORCH_CHECK_NOT_IMPLEMENTED(false, "No compiled nvfp4 experts quantization kernel");
}