RotorQuant is a KV cache compression technique for large language models that uses block-diagonal...