[perf] Enable concurrent execution of "shared_experts" and "selected_experts"...

[perf] Enable concurrent execution of "shared_experts" and "selected_experts" in qwen3-next (#27578)
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>

vllm/model_executor/models/qwen3_next.py

@@ -159,6 +159,7 @@ class Qwen3NextSparseMoeBlock(nn.Module):
 
         self.experts = SharedFusedMoE(
             shared_experts=self.shared_expert,
+            gate=self.gate,
             num_experts=self.n_routed_experts,
             top_k=config.num_experts_per_tok,
             hidden_size=config.hidden_size,
@@ -181,6 +182,12 @@ class Qwen3NextSparseMoeBlock(nn.Module):
         if self.is_sequence_parallel:
             hidden_states = sequence_parallel_chunk(hidden_states)
 
+        if self.experts.is_internal_router:
+            # In this case, the gate/router runs inside the FusedMoE class
+            final_hidden_states = self.experts(
+                hidden_states=hidden_states, router_logits=hidden_states
+            )
+        else:
             # router_logits: (num_tokens, n_experts)
             router_logits, _ = self.gate(hidden_states)
             final_hidden_states = self.experts(