metadata
license: mit
Llama 2 7B quantized in 3-bit with GPTQ.
from transformers import AutoModelForCausalLM, AutoTokenizer
from optimum.gptq import GPTQQuantizer
import torch
w = 3
model_path = meta-llama/Llama-2-7b-hf
tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=True)
model = AutoModelForCausalLM.from_pretrained(model_path, torch_dtype=torch.float16)
quantizer = GPTQQuantizer(bits=w, dataset="c4", model_seqlen = 4096)
quantized_model = quantizer.quantize_model(model, tokenizer)