[Fix] fix ernie4_5_vl model torch format loadding (PaddlePaddle#4447)

* fix

* add test

* fix test

* fix test

* update

Author: aquagull

fastdeploy/model_executor/models/ernie4_5_vl/ernie4_5_vl_moe.py

@@ -170,8 +170,8 @@ def __init__(
         # TODO(hehongyu): remove this after fix model network
         setattr(
             self.gate.weight,
-            "model_format",
-            "",
+            "weight_need_transpose",
+            False,
         )
 
     def forward(self, hidden_states: paddle.Tensor):