I trained a 30M-param LLM from scratch and the scaling "floor" was a mirage5rrpadhye about 7 hours ago 0 commentsRead Article on github.com
Discussion (0 Comments)Read Original on HackerNews
No comments available or they could not be loaded.