You seem to be correct. Still the note about training a deep learning model using the framework, then publishing it without applies.