Fix the call of assign in nlp samples.

Author: Xreki

paddle_samples/PaddleNLP/mbart-large-cc25/model.py

@@ -526,7 +526,7 @@ def forward(
         data_1,
     ):
         # pd_op.assign: (1x22xi64) <- (1x22xi64)
-        assign_0 = data_0
+        assign_0 = paddle._C_ops.assign(data_0)
 
         # pd_op.flatten: (22xi64) <- (1x22xi64)
         flatten_0 = paddle._C_ops.flatten(assign_0, 0, 1)

paddle_samples/PaddleNLP/mbart-large-en-ro/model.py

@@ -526,7 +526,7 @@ def forward(
         data_1,
     ):
         # pd_op.assign: (1x22xi64) <- (1x22xi64)
-        assign_0 = data_0
+        assign_0 = paddle._C_ops.assign(data_0)
 
         # pd_op.flatten: (22xi64) <- (1x22xi64)
         flatten_0 = paddle._C_ops.flatten(assign_0, 0, 1)