Use cutlass fp4 gemm by default (#11813)

ebda73dc · Qiaolin Yu · GitHub · f4f8a1b4 · ebda73dc
Unverified Commit ebda73dc authored Oct 18, 2025 by Qiaolin Yu Committed by GitHub Oct 18, 2025
Show whitespace changes
Inline Side-by-side

Showing with 1 addition and 1 deletion

python/sglang/srt/layers/quantization/modelopt_quant.py python/sglang/srt/layers/quantization/modelopt_quant.py +1 -1

No files found.
--- a/python/sglang/srt/layers/quantization/modelopt_quant.py
+++ b/python/sglang/srt/layers/quantization/modelopt_quant.py
@@ -79,7 +79,7 @@ CUTEDSL_MOE_SCALAR_INPUT_SCALE = get_bool_env_var(
    "SGLANG_CUTEDSL_MOE_SCALAR_INPUT_SCALE", "true"
 )
 USE_CUTLASS_BACKEND_FOR_FP4_GEMM = get_bool_env_var(
-    "SGLANG_USE_CUTLASS_BACKEND_FOR_FP4_GEMM"
+    "SGLANG_USE_CUTLASS_BACKEND_FOR_FP4_GEMM", "true"
 )
 # TODO make it true by default when the DeepEP PR is merged
 CUTEDSL_MOE_NVFP4_DISPATCH = get_bool_env_var(