In tensorflow2.x, there is no session to build a graph and API like tf.saved_model.builder to save as 'saved_model' format. I know there there is an API tf.saved_model.save can save a model as 'saved_model' format, but it is only compatible with 'tf.keras' model which have trackable object.
This issue really stop me from using tensorflow serving for deployment.
How can I solve this problem?
New Issue Checklist
Issue Description
In tensorflow2.x, there is no session to build a graph and API like tf.saved_model.builder to save as 'saved_model' format. I know there there is an API tf.saved_model.save can save a model as 'saved_model' format, but it is only compatible with 'tf.keras' model which have trackable object. This issue really stop me from using tensorflow serving for deployment. How can I solve this problem?