Presumably it's fine to train a model on their model then?