LLM inference, visually explained: why a 13B model reads 26 GB per token(www.youtube.com)
2 pointsby ryansakhalaan hour ago1 comment