`Seq2SeqTrainer` set max_length and num_beams only when non None (#12899)

* set max_length and num_beams only when non None

* fix instance variables

* fix code style

src/transformers/trainer_seq2seq.py

@@ -70,7 +70,9 @@ class Seq2SeqTrainer(Trainer):
             A dictionary containing the evaluation loss and the potential metrics computed from the predictions. The
             dictionary also contains the epoch number which comes from the training state.
         """
+        if max_length is not None or not hasattr(self, "_max_length"):
             self._max_length = max_length
+        if num_beams is not None or not hasattr(self, "_num_beams"):
             self._num_beams = num_beams
         return super().evaluate(eval_dataset, ignore_keys=ignore_keys, metric_key_prefix=metric_key_prefix)
 
@@ -117,7 +119,9 @@ class Seq2SeqTrainer(Trainer):
             - metrics (:obj:`Dict[str, float]`, `optional`): The potential dictionary of metrics (if the dataset
               contained labels).
         """
+        if max_length is not None or not hasattr(self, "_max_length"):
             self._max_length = max_length
+        if num_beams is not None or not hasattr(self, "_num_beams"):
             self._num_beams = num_beams
         return super().predict(test_dataset, ignore_keys=ignore_keys, metric_key_prefix=metric_key_prefix)