KV cache compression

Google Launches TurboQuant to Make AI Models More Efficient Without Losing Accuracy