使用tf.saved_model并将字符串作为输入传递时，无法获取元素作为字节

Question

在加载保存的模型时，我尝试： 请注意，example_proto是一个tf.example对象

predict_fn = tf.contrib.predictor.from_saved_model(saved_model_dir)
a = {'inputs': [{'examples': str(example_proto.SerializeToString())}]}
t1 = datetime.datetime.now()
print(predict_fn(a))
print((datetime.datetime.now() - t1).total_seconds())

当实际保存模型时，我尝试了

def serving_input_receiver_fn():
  default_batch_size = 1
  """An input receiver that expects a serialized tf.Example."""
  serialized_tf_example = tf.placeholder(dtype=tf.string,
                                         shape=[default_batch_size],
                                         name='input_example_tensor')
  receiver_tensors = {'examples': serialized_tf_example}
  # print(receiver_tensors.numpy())
  serialized_tf_example = tf.Print(serialized_tf_example, 
                             [serialized_tf_example],
                             "serialized_tf_example: ")
  features = tf.parse_example(serialized_tf_example, feature_description)
  return tf.estimator.export.ServingInputReceiver(features, receiver_tensors)

和

   estimator.export_savedmodel(
          make_checkpoint_folder("serving_dir1"),
          serving_input_receiver_fn)

我收到关注错误

InternalError: Unable to get element as bytes.