Fetching the paper…

Training and inference of large language models using 8-bit floating point · Around