diff --git a/fastdeploy/worker/hpu_model_runner.py b/fastdeploy/worker/hpu_model_runner.py
index 16a781904..5811a1516 100644
--- a/fastdeploy/worker/hpu_model_runner.py
+++ b/fastdeploy/worker/hpu_model_runner.py
@@ -210,6 +210,7 @@ def rebuild_padding_v3_1(
     return output_data
 
 
+from fastdeploy.model_executor.forward_meta import ForwardMeta
 from fastdeploy.model_executor.layers.linear import QKVParallelLinear, RowParallelLinear
 from fastdeploy.model_executor.ops.intel_hpu import fused_mlp
 
@@ -259,10 +260,21 @@ def fused_self_atten_forward(
     return atten_out
 
 
-def fused_mlp_forward(self, x):
-    """ """
+def fused_mlp_forward(
+    self,
+    hidden_states: paddle.Tensor,
+    forward_meta: Optional[ForwardMeta] = None,
+):
+    """
+    The forward function for the MLP (Multi-Layer Perceptron) layer.
+    Args:
+        hidden_states (paddle.Tensor): The input tensor to the MLP layer.
+        forward_meta (Optional[ForwardMeta]): Optional metadata for the forward pass.
+    Returns:
+        paddle.Tensor: The output tensor after applying the MLP layer and (optionally) all-reduce.
+    """
     out = fused_mlp(
-        x,
+        hidden_states,
         self.up_gate_proj.weight,
         None,
         self.down_proj.weight,