fix TFModel.evaluate() not to create saver when it exists

Author: iTakeshi

tensorflow_model.py

@@ -123,7 +123,8 @@ def evaluate(self) -> Optional[ModelEvaluationResults]:
 
  self.eval_top_words_op, self.eval_top_values_op, self.eval_original_names_op, _, _, _, _, \
  self.eval_code_vectors = self._build_tf_test_graph(input_tensors)
- self.saver = tf.compat.v1.train.Saver()
+ if self.saver is None:
+ self.saver = tf.compat.v1.train.Saver()
 
  if self.config.MODEL_LOAD_PATH and not self.config.TRAIN_DATA_PATH_PREFIX:
  self._initialize_session_variables()