Re-quantizing a local LLM 14x faster by skipping the tensors that didn't change(andreaborio.substack.com)
6 pointsby andreaborio6 hours ago1 comment