Very cool! Since it’s possible to train foundation model adapters, is a library for user fine tunes possible?