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

2026-01-31 10:01:29 -05:00
parent 1a006c2a0d
commit ef134d407d
1 changed files with 5 additions and 1 deletions
--- a/python/sglang/srt/layers/quantization/modelopt_quant.py
+++ b/python/sglang/srt/layers/quantization/modelopt_quant.py
@@ -134,7 +134,11 @@ def fp4_gemm(
    fp4_backend = get_fp4_gemm_runner_backend()
    if enable_flashinfer_fp4_gemm:
        # 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(
            input, weight, input_sf, weight_sf, alpha, out_dtype, backend=backend
        )