Using hub.create_module_spec with TensorFlow 2.0

[dfalbel]

What's the recommended way to create Module specs with TensorFlow 2.0.

The guide uses tf.placeholders in order to define the input signature, but this is not idiomatic TF2.0.

I wonder if it would be possible to get all the information from a tf.function with define input signatures, eg. something like this:

@tf.function(input_signature=[tf.TensorSpec(shape=None, dtype=tf.float32)])
def f(x): return tf.add(x, 1.)

Thanks very much!

[akhorlin]

We are still in the progress of updating our documentation to reflect TF 2.0 changes. The guide in question hasn't been updated yet. https://github.com/tensorflow/hub/blob/master/examples/text_embeddings_v2/export_v2.py is a better example of how to define and export modules in TF2.

[dfalbel]

Great! Thanks very much! So the recommendation is to just use the tf.saved_model.save and set the signature via tf.function if required.

One thing that is not really clear to me yet is how do I create a module that exports multiple signatures.

[andresusanopinto]

There has been more documentation changes about creating tf-2 saved models, in particular the following:
https://github.com/tensorflow/hub/blob/master/docs/tf2_saved_model.md#creating-savedmodels-for-tf-hub