An aggressive quantization method that compresses model weights to extremely low precision levels, significantly reducing memory requirements at the cost of some accuracy.