Merge branch 'v0.9.2-dev-wm' into 'v0.9.2-dev'

[fix]解决deepseek报错

See merge request dcutoolkit/deeplearing/vllm!162

vllm/_custom_ops.py

@@ -2366,6 +2366,25 @@ def moe_fused_gate(
         routed_scaling_factor,
     )
 
+if hasattr(torch.ops._moe_C, "moe_fused_gate"):
+
+    @register_fake("_moe_C::moe_fused_gate")
+    def moe_fused_gate_fake(
+        input_tensor: torch.Tensor,
+        bias: torch.Tensor,
+        num_expert_group: int,
+        topk_group: int,
+        topk: int,
+        n_share_experts_fusion: int,
+        routed_scaling_factor: int,
+    ):
+        return torch.empty((input_tensor.size(0), topk),
+                           dtype=input_tensor.dtype,
+                           device=input_tensor.device), \
+                    torch.empty((input_tensor.size(0), topk),
+                           dtype=input_tensor.dtype,
+                           device=input_tensor.device)
+
 
 if hasattr(torch.ops._C, "weight_packed_linear"):