TensorFlow RuntimeError: MetaGraphDef associated with tags serve could not be found in SavedModel

Question

When I use simple_save to save a model, when I get a runtime error when I try to load the model.

The code to save is:

session = Session()
inputs = tf.placeholder(dtype=tf.float32, shape=(None, height, width, in_channel_size), name='input_img')
model = Some_Model(inputs, num_classes=no_of_defects, is_training=False)
logits, _ = model.build_model()
predictor = tf.nn.softmax(self.logits, name='logits_to_softmax')
feed_dict = {inputs: inputs}
prediction_probabilities = session.run(self.predictor, feed_dict=feed_dict)

tf.saved_model.simple_save(self.session, path,
                               inputs={"inputs" : self.inputs},
                               outputs={"predictor": self.predictor})

The code to load is:

tf.saved_model.loader.load(session, tag_constants.SERVING, path)

which gives the error:

RuntimeError: MetaGraphDef associated with tags serve could not be found in SavedModel. To inspect available tag-sets in the SavedModel, please use the SavedModel CLI: `saved_model_cli`

When I run

saved_model_cli show --dir path --tag_set serve --signature_def serving_default

I get

The given SavedModel SignatureDef contains the following input(s):
  inputs['inputs'] tensor_info:
      dtype: DT_FLOAT
      shape: (-1, 512, 1024, 8)
      name: input_img:0
The given SavedModel SignatureDef contains the following output(s):
  outputs['predictor'] tensor_info:
      dtype: DT_FLOAT
      shape: (-1, 512, 1024, 25)
      name: logits_to_softmax:0
Method name is: tensorflow/serving/predict

What am I doing wrong?

Answer 1

The problem is with the load call. It should be:

tf.saved_model.loader.load(session, [tag_constants.SERVING], path)

where tag_constants is located at tf.saved_model.tag_constants.