Very neat. Thanks. Since you know more about it than I, how do the lessons learned in training small models translate to scaling them up? If that is too much to explain, would you be able to recommend something to read about it?