Google's TurboQuant AI-compression algorithm can reduce LLM memory usage by 6x (arstechnica.com) AI

Google’s TurboQuant algorithm claims it can compress transformer/LLM representations to cut memory usage by up to 6x without quality loss.