Google's TurboQuant cuts LLM memory use by 6x

TurboQuant makes AI models more efficient but doesn't reduce output quality like other methods.