diff --git a/python/sglang/srt/layers/moe/ep_moe/layer.py b/python/sglang/srt/layers/moe/ep_moe/layer.py
index d05b24098..a94fafd70 100644
--- a/python/sglang/srt/layers/moe/ep_moe/layer.py
+++ b/python/sglang/srt/layers/moe/ep_moe/layer.py
@@ -191,11 +191,15 @@ class DeepEPMoE(FusedMoE):
             assert deep_gemm_wrapper.ENABLE_JIT_DEEPGEMM and self.use_fp8_w8a8
             return self.forward_deepgemm_contiguous(dispatch_output)
         elif DispatchOutputChecker.format_is_deepep_ll(dispatch_output):
-            if get_moe_runner_backend().is_flashinfer_cutedsl():
+            if (
+                get_moe_runner_backend().is_flashinfer_cutedsl()
+                and self.quant_config.get_name() == "modelopt_fp4"
+            ):
                 return self.forward_flashinfer_cutedsl(
                     dispatch_output, down_gemm_overlap_args=down_gemm_overlap_args
                 )
             assert deep_gemm_wrapper.ENABLE_JIT_DEEPGEMM and self.use_fp8_w8a8
+            assert down_gemm_overlap_args is None
             return self.forward_deepgemm_masked(dispatch_output)
         else:
             raise ValueError(