Two different tricks for fast LLM inference

https://www.seangoedecke.com/fast-llm-inference/

https://news.ycombinator.com/item?id=47022329