Keras Model to TF SavedModel format

export.py

export_path = "./tf"
builder = tf.saved_model.builder.SavedModelBuilder(export_path)
signature = tf.saved_model.signature_def_utils.predict_signature_def(
  inputs={"input": mod.input},
  outputs={"output": mod.output})
 
with K.get_session() as sess:
  builder.add_meta_graph_and_variables(sess=sess,
    tags=[tag_constants.SERVING],
    signature_def_map={"predict": signature, "serving_default": signature},
    legacy_init_op=legacy_init_op)
  builder.save()