Comment by DCKing

TurboQuant is a runtime optimization for a model's KV cache and doesn't allow for reduction in model size.