Comment by andai

Isn't iterative chunking the whole point of the transformer though? Or does it add that much overhead to do it at the byte level?