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

by chrsw | View on Hacker News