jiangchengchengNLP's picture
Update README.md
664a32d verified
|
raw
history blame
1.69 kB
metadata
license: apache-2.0

这是基于Auto-GPTQ框架的量化模型,模型选取为huatuoGPT2-7B,这是一个微调模型,基底模型为百川-7B。

参数说明: 原模型大小:16GB,量化后模型大小:5GB

推理准确度尚未测试,请谨慎使用

量化过程中,校准数据采用微调训练集Medical Fine-tuning Instruction (GPT-4)。

使用示例:

确保你安装了bitsandbytes

pip install bitsandbytes

确保你安装了auto-gptq

!git clone https://github.com/AutoGPTQ/AutoGPTQ

cd AutoGPTQ

!pip install -e .
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from transformers.generation.utils import GenerationConfig
tokenizer = AutoTokenizer.from_pretrained("jiangchengchengNLP/huatuo_AutoGPTQ_7B4bits", use_fast=True, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("jiangchengchengNLP/huatuo_AutoGPTQ_7B4bits", device_map="auto", torch_dtype="auto", trust_remote_code=True)
model.generation_config = GenerationConfig.from_pretrained("jiangchengchengNLP/huatuo_AutoGPTQ_7B4bits")
messages = []
messages.append({"role": "user", "content": "肚子疼怎么办?"})
response = model.HuatuoChat(tokenizer, messages)
print(response)

更多量化细节:

量化环境:双卡T4

校正规模:512 训练对

量化配置:

ntize_config = BaseQuantizeConfig(
    bits=4, # 4 or 8
    group_size=128,
    damp_percent=0.01,
    desc_act=False,  # set to False can significantly speed up inference but the perplexity may slightly bad
    static_groups=False,
    sym=True,
    true_sequential=True,
    model_name_or_path=None,
    model_file_base_name="model"
)