GPTQ 4bit quantization of: https://huggingface.co/chavinlo/gpt4-x-alpaca
Command: CUDA_VISIBLE_DEVICES=0 python llama.py ./models/chavinlo-gpt4-x-alpaca --wbits 4 --true-sequential --act-order --groupsize 128 --save gpt-x-alpaca-13b-native-4bit-128g.pt