minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[Fix] Revert back to using CUTLASS mm_fp4 backend (#17369)

Browse Source
This commit is contained in:
b8zhong
2026-01-31 23:01:29 +08:00
committed by GitHub
parent 1a006c2a0d
commit ef134d407d
1 changed files with 5 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/layers/quantization/modelopt_quant.py
+5 -1
View File
@@ -134,7 +134,11 @@ def fp4_gemm(
fp4_backend = get_fp4_gemm_runner_backend() fp4_backend = get_fp4_gemm_runner_backend()
if enable_flashinfer_fp4_gemm: if enable_flashinfer_fp4_gemm:
# Use the remapping logic to convert SGLang backend names to FlashInfer API names # Use the remapping logic to convert SGLang backend names to FlashInfer API names
backend = fp4_backend.get_flashinfer_backend() backend = (
fp4_backend.get_flashinfer_backend()
if not fp4_backend.is_auto()
else "cutlass"
)
return flashinfer_fp4_gemm( return flashinfer_fp4_gemm(
input, weight, input_sf, weight_sf, alpha, out_dtype, backend=backend input, weight, input_sf, weight_sf, alpha, out_dtype, backend=backend
) )