README.md

Attention quantization: HQQ 4-bit, groupsize 64, compress zero, compress scale with groupsize 256
Experts quantization: HQQ 3-bit, groupsize 64, compress zero, compress scale with groupsize 128