Re-quantizing a local LLM 14x faster by skipping the tensors that didn't changeandreaborio.substack.com·8 pts·andreaborio·0