MegaTrain: Full Precision Training of 100B+ Parameter LLMs on a Single GPU hackernews April 08, 2026 Score: 24 | Comments: 4 Read Full Article open_in_new