Training a 1.93B byte-level LM on one A100: 60K steps, flat to 4x context(huggingface.co)
1 pointby EverestAn2 hours ago1 comment