biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Hardware/NVIDIA/Kernel] Enable nvidia/DeepSeek-R1-FP4 Model (#16362)

Browse Source
This commit is contained in:
Pavani Majety
2025-05-09 16:24:41 -07:00
committed by GitHub
parent 3b602cdea7
commit 0c0fdae84f
16 changed files with 1994 additions and 112 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
csrc/quantization/fp4/nvfp4_quant_entry.cu
View File

@@ -23,10 +23,32 @@ void scaled_fp4_quant_sm100a(torch::Tensor const& output,
torch::Tensor const& input_sf);
#endif
#if defined ENABLE_NVFP4 && ENABLE_NVFP4
void scaled_fp4_experts_quant_sm100a(
torch::Tensor& output, 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);
#endif
void scaled_fp4_quant(torch::Tensor& output, torch::Tensor const& input,
torch::Tensor& output_sf, torch::Tensor const& input_sf) {
#if defined ENABLE_NVFP4 && ENABLE_NVFP4
return scaled_fp4_quant_sm100a(output, input, output_sf, input_sf);
#endif
TORCH_CHECK_NOT_IMPLEMENTED(false, "No compiled nvfp4 quantization");
TORCH_CHECK_NOT_IMPLEMENTED(false, "No compiled nvfp4 quantization kernel");
}
void scaled_fp4_experts_quant(
torch::Tensor& output, 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) {
#if defined ENABLE_NVFP4 && ENABLE_NVFP4
return scaled_fp4_experts_quant_sm100a(
output, output_scale, input, input_global_scale, input_offset_by_experts,
output_scale_offset_by_experts);
#endif
TORCH_CHECK_NOT_IMPLEMENTED(false,
"No compiled nvfp4 experts quantization kernel");
}