LLM inference, visually explained: why a 13B model reads 26 GB per token

4 points | by ryansakhala 9 hours ago

1 comments