So only thing teaching model (loss) is probability prediction in single byte space. And that is enough? Looks very promising, if I am not misunderstanding.
So only thing teaching model (loss) is probability prediction in single byte space. And that is enough? Looks very promising, if I am not misunderstanding.