| 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 |
| 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 |