This is the INT4 Llama-3-8b model quantized by per-group QQQ and the group size is 128. QQQ is an innovative and hardware-optimized W4A8 quantization solution. For more details, please refer to our code repo and our paper.
