AirLLM enables 70B LLM inference on a single 4GB GPU

AirLLM is a GitHub tool for running 70B-parameter LLM inference on a single 4GB GPU; the Hacker News post drew 32 points and 12 comments.