diff --git a/vllm/lora/request.py b/vllm/lora/request.py
index 538d51031..2811fee1d 100644
--- a/vllm/lora/request.py
+++ b/vllm/lora/request.py
@@ -20,7 +20,6 @@ class LoRARequest(
     lora_name: str
     lora_int_id: int
     lora_path: str = ""
-    long_lora_max_len: int | None = None
     base_model_name: str | None = msgspec.field(default=None)
     tensorizer_config_dict: dict | None = None
 
diff --git a/vllm/model_executor/models/qwen2_moe.py b/vllm/model_executor/models/qwen2_moe.py
index 2750f1864..82837b77e 100644
--- a/vllm/model_executor/models/qwen2_moe.py
+++ b/vllm/model_executor/models/qwen2_moe.py
@@ -111,7 +111,7 @@ class Qwen2MoeMLP(nn.Module):
         out, _ = self.down_proj(out)
 
         if self.expert_gate is not None:
-            out = F.sigmoid(self.expert_gate(x)) * out
+            out = F.sigmoid(self.expert_gate(x)[0]) * out
 
         return out
 
@@ -140,7 +140,13 @@ class Qwen2MoeSparseMoeBlock(nn.Module):
             prefix=f"{prefix}.gate",
         )
 
-        self.shared_expert_gate = torch.nn.Linear(config.hidden_size, 1, bias=False)
+        self.shared_expert_gate = ReplicatedLinear(
+            config.hidden_size,
+            1,
+            bias=False,
+            quant_config=None,
+            prefix=f"{prefix}.shared_expert_gate",
+        )
 
         if config.shared_expert_intermediate_size > 0:
             self.shared_expert = Qwen2MoeMLP(