Fix trainer callback (#7720)

Fix a bug that happends when subclassing Trainer and
overwriting evaluate() without calling prediciton_loop()

src/transformers/trainer_callback.py

@@ -443,6 +443,7 @@ class ProgressCallback(TrainerCallback):
 
     def on_evaluate(self, args, state, control, **kwargs):
         if state.is_local_process_zero:
+            if self.prediction_bar is not None:
                 self.prediction_bar.close()
             self.prediction_bar = None