增加w4a8marlin-ds的moe融合算子输入

vllm/model_executor/layers/quantization/slimquant_w4a8_marlin.py

@@ -230,6 +230,7 @@ class SlimQuantW4A8Int8MarlinMoEMethod:
         use_nn_moe: Optional[bool] = False,
         routed_scaling_factor: Optional[float] = None,
         use_fused_gate: Optional[bool] = False,
+        shared_output: Optional[torch.Tensor] = None,
         **_  
     ) -> torch.Tensor:
         from vllm.model_executor.layers.fused_moe import fused_experts
@@ -272,4 +273,6 @@ class SlimQuantW4A8Int8MarlinMoEMethod:
             a1_scale=layer.w13_input_scale,
             a2_scale=layer.w2_input_scale,
             use_nn_moe=use_nn_moe,
+            shared_output=shared_output,
+            routed_scaling_factor=routed_scaling_factor,
         )