Re-quantizing a local LLM 14x faster by skipping the tensors that didn't change(andreaborio.substack.com)6 points by andreaborio 7 hours ago | 0 comments