Fetching the paper…

QJL: 1-Bit Quantized JL Transform for KV Cache Quantization with Zero Overhead · Around