RotorQuant: Clifford algebra vector quantization for LLM KV cache compression. 10-19x faster than TurboQuant, 44x fewer parameters.
130
11
4
