3 ms·Re-quantizing a local LLM 14x faster by skipping the tensors that didn't change8 points by andreaborio 4mo agoandreaborio 4mo ago[dead]