diff --git a/python/sglang/srt/layers/moe/cutlass_moe.py b/python/sglang/srt/layers/moe/cutlass_moe.py
index 2a5a5dccbd6..3774afac2d3 100755
--- a/python/sglang/srt/layers/moe/cutlass_moe.py
+++ b/python/sglang/srt/layers/moe/cutlass_moe.py
@@ -209,7 +209,7 @@ def cutlass_fused_experts_fp8(
     )
 
     result = torch.empty((m, k), device=device, dtype=out_dtype)
-    apply_shuffle_mul_sum(c2, result, c_map, topk_weights)
+    apply_shuffle_mul_sum(c2, result, c_map, topk_weights.to(out_dtype))
     return result