use self.hidden_size not use self.fd_config.model_config.hidden_size (#7340)

fastdeploy/model_executor/layers/moe/moe.py

@@ -788,7 +788,7 @@ class FusedMoE(nn.Layer):
         chunk_size = self.fd_config.parallel_config.chunked_moe_size
         token_num = x.shape[0]
         fake_x = paddle.empty(
-            shape=[0, self.fd_config.model_config.hidden_size],
+            shape=[0, self.hidden_size],
             dtype=paddle.get_default_dtype(),
         )
         # input size that are less than a chunk, less than the max size data or empty input

tests/distributed/chunked_moe.py

@@ -169,6 +169,8 @@ class TestChunkedMoE(unittest.TestCase):
         fused_moe.fd_config = mock_fd_config
         fused_moe.quant_method = MockQuantMethod()
         fused_moe.enable_routing_replay = None
+
+        fused_moe.hidden_size = mock_fd_config.model_config.hidden_size
         return fused_moe
 
     def run_model_runner(self):