yujiepan/llama-3-tiny-random-gptq-w4

4-bit weight only quantization by AutoGPTQ on yujiepan/llama-3-tiny-random

from transformers import AutoModelForCausalLM, AutoTokenizer, GPTQConfig
import torch

model_id = "yujiepan/llama-3-tiny-random"
tokenizer = AutoTokenizer.from_pretrained(model_id)

quantization_config = GPTQConfig(
    bits=4, group_size=-1,
    dataset="c4",
    tokenizer=tokenizer,
)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    quantization_config=quantization_config,
)