fix nn_moe args

vllm/model_executor/layers/fused_moe/unquantized_fused_moe_method.py

@@ -271,11 +271,13 @@ class UnquantizedFusedMoEMethod(FusedMoEMethodBase, CustomOp):
         layer: "FusedMoE",  # type: ignore[name-defined] # noqa: F821
         x: torch.Tensor,
         router_logits: torch.Tensor,
+        use_nn_moe: bool | None = False,
     ) -> torch.Tensor | tuple[torch.Tensor, torch.Tensor]:
         return self.forward(
             layer=layer,
             x=x,
             router_logits=router_logits,
+            use_nn_moe=use_nn_moe,
         )
 
     def get_fused_moe_quant_config(
@@ -294,6 +296,7 @@ class UnquantizedFusedMoEMethod(FusedMoEMethodBase, CustomOp):
         layer: "FusedMoE",  # type: ignore[name-defined] # noqa: F821
         x: torch.Tensor,
         router_logits: torch.Tensor,
+        use_nn_moe: bool | None = False,
     ) -> torch.Tensor | tuple[torch.Tensor, torch.Tensor]:
         topk_weights, topk_ids, zero_expert_result = layer.select_experts(
             hidden_states=x,
@@ -334,6 +337,7 @@ class UnquantizedFusedMoEMethod(FusedMoEMethodBase, CustomOp):
                 apply_router_weight_on_input=layer.apply_router_weight_on_input,
                 global_num_experts=layer.global_num_experts,
                 expert_map=layer.expert_map,
+                use_nn_moe=use_nn_moe,
             )
 
         if layer.zero_expert_num != 0 and layer.zero_expert_type is not None: