New LLM optimization technique slashes memory costs

New LLM optimization technique slashes memory costs. 214 comments, 445 points on Hacker News.