As always, I would prefer more focus on training your own models rather than running prebaked ones.