work in progress! we aren't even past 1b tokens yet but check out how well it's doing regardless in the inference thing
