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

[NVIDIA] Add new SMs support for Spark & Thor (#11287)

Browse Source 
Signed-off-by: Serge Panev <spanev@nvidia.com>
This commit is contained in:
Serge Panev
2025-10-21 11:02:24 -07:00
committed by GitHub
parent 97710ccd1a
commit 2b1da821b5
4 changed files with 22 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
sgl-kernel/csrc/gemm/nvfp4_expert_quant.cu
View File

@@ -568,7 +568,7 @@ void scaled_fp4_experts_quant_sm100a(
torch::Tensor const& input_offset_by_experts,
torch::Tensor const& output_scale_offset_by_experts) {
auto sm_version = getSMVersion();
TORCH_CHECK(sm_version == 100 || sm_version == 103, "fp4_quant is only supported on sm100a/sm103a");
TORCH_CHECK(sm_version >= 100, "fp4_quant is only supported on sm100+");
CHECK_INPUT(output, "output must be a CUDA tensor");
CHECK_INPUT(output_scale, "output_scale must be a CUDA tensor");
@@ -652,7 +652,7 @@ void silu_and_mul_scaled_fp4_experts_quant_sm100a(
torch::Tensor const& mask,
bool use_silu_and_mul) {
auto sm_version = getSMVersion();
TORCH_CHECK(sm_version == 100 || sm_version == 103, "fp4_quant is only supported on sm100a/sm103a");
TORCH_CHECK(sm_version >= 100, "fp4_quant is only supported on sm100+");
CHECK_INPUT(output, "output must be a CUDA tensor");
CHECK_INPUT(output_scale, "output_scale must be a CUDA tensor");