Add support for bf16 x bf16 cutlass fused MoE (#10275)

Co-authored-by: Sam Li <lsam@nvidia.com>
Co-authored-by: jackeyhua <jackeyhuasjtu@gmail.com>

python/sglang/srt/server_args.py

@@ -1202,6 +1202,7 @@ class ServerArgs:
                 )
                 self.disable_overlap_schedule = True
             if is_sm100_supported():
+                self.attention_backend = "triton"
                 quantization_config = getattr(hf_config, "quantization_config", None)
                 quant_method = (
                     quantization_config.get("quant_method")
@@ -1468,8 +1469,8 @@ class ServerArgs:
     def _handle_moe_kernel_config(self):
         if self.moe_runner_backend == "flashinfer_cutlass":
             assert (
-                self.quantization == "modelopt_fp4"
-            ), "modelopt_fp4 quantization is required for Flashinfer Cutlass MOE"
+                self.quantization == "modelopt_fp4" or self.quantization is None
+            ), "modelopt_fp4 quantization or bf16 is required for Flashinfer Cutlass MOE"
             assert self.ep_size in [
                 1,
                 self.tp_size,