simplify the control logic for using shared experts fusion (#5504)

python/sglang/srt/layers/quantization/compressed_tensors/compressed_tensors_moe.py

@@ -283,6 +283,7 @@ class CompressedTensorsW8A8Fp8MoEMethod(CompressedTensorsMoEMethod):
         inplace: bool = True,
         no_combine: bool = False,
         apply_router_weight_on_input: bool = False,
+        routed_scaling_factor: Optional[float] = None,
     ) -> torch.Tensor:
         from sglang.srt.layers.moe.fused_moe_triton import fused_experts
         from sglang.srt.layers.moe.topk import select_experts
@@ -297,6 +298,7 @@ class CompressedTensorsW8A8Fp8MoEMethod(CompressedTensorsMoEMethod):
             num_expert_group=num_expert_group,
             custom_routing_function=custom_routing_function,
             correction_bias=correction_bias,
+            routed_scaling_factor=routed_scaling_factor,
         )
 
         return fused_experts(
@@ -633,6 +635,7 @@ class CompressedTensorsWNA16MoEMethod(CompressedTensorsMoEMethod):
         scoring_func: str = "softmax",
         correction_bias: Optional[torch.Tensor] = None,
         activation: str = "silu",
+        routed_scaling_factor: Optional[float] = None,
     ) -> torch.Tensor:
         from sglang.srt.layers.moe.topk import select_experts
 
@@ -653,6 +656,7 @@ class CompressedTensorsWNA16MoEMethod(CompressedTensorsMoEMethod):
             custom_routing_function=custom_routing_function,
             scoring_func=scoring_func,
             correction_bias=correction_bias,
+            routed_scaling_factor=routed_scaling_factor,
         )
 
         return torch.ops.vllm.fused_marlin_moe(