Can't load Keras model with multiple MetaGraphs, can't pass tags

[uchua]

• Have I written custom code: Yes
• OS Platform and Distribution: Amazon Linux AMI 2018.03
• TensorFlow installed from: Preinstalled by Amazon
• TensorFlow version: 1.15.0
• Python version: 3.6
• CUDA/cuDNN version:
• GPU model and memory: Nvidia Tesla V100

Describe the current behavior
When trying to load a saved model using keras.models.load_model, an error occurs stating that tf.saved_model.load requires a 'tags='. There is also no way to pass tags through the load_model function.

Describe the expected behavior
Model should load.

Standalone code to reproduce the issue
With "albert/" being a path to this model locally: https://tfhub.dev/google/albert_xlarge/3:

from tensorflow import keras

model = keras.models.load_model("albert/")

Error:

ValueError: Importing a SavedModel with tf.saved_model.load requires a 'tags=' argument if there is more than one MetaGraph. Got 'tags=None', but there are 2 MetaGraphs in the SavedModel with tag sets [['train'], []]. Pass a 'tags=' argument to load this SavedModel.

Other info / logs

File "albert_entrypoint.py", line 50, in <module>
    model = keras.models.load_model(MODEL_EXTRACT_DIR)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/keras/saving/save.py", line 147, in load_model
    return saved_model_load.load(filepath, compile)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/keras/saving/saved_model/load.py", line 86, in load
    model = tf_load.load_internal(path, loader_cls=KerasObjectLoader)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/saved_model/load.py", line 550, in load_internal
    root = load_v1_in_v2.load(export_dir, tags)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/saved_model/load_v1_in_v2.py", line 239, in load
    return loader.load(tags=tags)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/saved_model/load_v1_in_v2.py", line 168, in load
    meta_graph_def = self.get_meta_graph_def_from_tags(tags)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/saved_model/load_v1_in_v2.py", line 80, in get_meta_graph_def_from_tags
    .format(len(self._saved_model.meta_graphs), tag_sets))

ValueError: Importing a SavedModel with tf.saved_model.load requires a 'tags=' argument if there is more than one MetaGraph. Got 'tags=None', but there are 2 MetaGraphs in the SavedModel with tag sets [['train'], []]. Pass a 'tags=' argument to load this SavedModel.

[ravikyram]

@uchua

This issue is more suitable for TensorFlow Hub repo. Please post it on hub repo from here. Thanks!

[uchua]

@ravikyram outside of the model (which I don't believe the TensorFlow Hub team would have any control of), the issue doesn't involve TensorFlow Hub. Any model with multiple MetaGraphs (where tf.saved_model.load requires the tags= argument), won't be able to be load correctly.

[ravikyram]

I have tried on colab with TF version 1.15 , 2.2.0-dev20200226 and was able to reproduce the issue.Please, find the gist here. Thanks!

[gowthamkpr]

You can load the module as a Keras layer to use it. Heres the gist.

I was also able to lad the model using tf.saved_model.load() which has tags argument. Please find the gist here.

keras.models.load_model doesnt have tags argument explicitly defined and thats the reason why you are not able to load this model.

[uchua]

Is there a way to load a model using tf.saved_model.load() and then add it to a Keras model?

[gowthamkpr]

You can take the outputs of the loaded model and create a new model with those outputs @uchua
You can take a look at this issue

[bilalghanem]

Upgrade your tensorflow version to 2.0.

[tensorflowbutler]

It has been 29 days with no activity and the awaiting response label was assigned. Is this still an issue?

[google-ml-butler]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you.

[gowthamkpr]

Closing this issue as it has been inactive for 3 weeks. Please add additional comments for us to open this issue again.

[google-ml-butler]

Are you satisfied with the resolution of your issue?
Yes
No