mlabonne/gpt2-GPTQ-4bit

Model created using AutoGPTQ on a GPT-2 model with 4-bit quantization.

You can load this model with the AutoGPTQ library, installed with the following command:

pip install auto-gptq

You can then download the model from the hub using the following code:

from transformers import AutoModelForCausalLM, AutoTokenizer
from auto_gptq import AutoGPTQForCausalLM, BaseQuantizeConfig

model_name = "mlabonne/gpt2-GPTQ-4bit"
tokenizer = AutoTokenizer.from_pretrained(model_name)
quantize_config = BaseQuantizeConfig.from_pretrained(model_name)
model = AutoGPTQForCausalLM.from_quantized(model_name,
                                           model_basename="gptq_model-4bit-128g",
                                           device="cuda:0",
                                           use_triton=True,
                                           use_safetensors=True,
                                           quantize_config=quantize_config)

This model works with the traditional Text Generation pipeline.

Example of generation with the input text "I have a dream":

I have a dream. I want someone with my face, and what I have. I want to go home. I want to be alive. I want to see my children. I dream if I have the spirit, my body, my voice,