MegaTrain: Full Precision Training of 100B+ Parameter LLMs on a Single GPU

(arxiv.org)

116 points | by chrsw  3 hours ago

23 comments