Training a 1.93B byte-level LM on one A100: 60K steps, flat to 4x context

1 points | by EverestAn an hour ago

No comments yet