Live data from Hacker News

Why LLM decode is memory-bound, not compute-bound

github.com

1–2 of 2 posts