---
license: apache-2.0
datasets:
- BelleGroup/train_2M_CN
- BelleGroup/train_3.5M_CN
- BelleGroup/train_1M_CN
- BelleGroup/train_0.5M_CN
- BelleGroup/school_math_0.25M
language:
- zh
---

## GoGPT

基于多样性中文指令数据微调的中文BLOOM底座模型

![img.png](resources/img.png)
> 训练第一轮足够了，后续第二轮和第三轮提升不大

- 🚀多样性指令数据
- 🚀筛选高质量中文数据

| 模型名字       | 参数量    | 模型地址 |
|------------|--------|------|
| gogpt-560m | 5.6亿参数 | 🤗[golaxy/gogpt-560m](https://huggingface.co/golaxy/gogpt-560m) |
| gogpt-3b   | 30亿参数  | 🤗[golaxy/gogpt-3b-bloom](https://huggingface.co/golaxy/gogpt-3b-bloom) |
| gogpt-7b   | 70亿参数  | 🤗[golaxy/gogpt-7b-bloom](https://huggingface.co/golaxy/gogpt-7b-bloom) |


## 测试效果
![img.png](resources/test1.png)
![img.png](resources/test2.png)
![img.png](resources/test3.png)
![img.png](resources/test4.png)
![img.png](resources/test5.png)
![img.png](resources/test6.png)


## TODO
- 进行RLFH训练
- 后续加入中英平行语料

## 感谢

- [@hz-zero_nlp](https://github.com/yuanzhoulvpi2017/zero_nlp)
- [stanford_alpaca](https://github.com/tatsu-lab/stanford_alpaca)
- [Belle数据](https://huggingface.co/BelleGroup)

## Citation

如果你在研究中使用了GoGPT，请按如下格式引用：
```
@misc{GoGPT,
  title={GoGPT: Training Medical GPT Model},
  author={Qiang Yan},
  year={2023},
  howpublished={\url{https://github.com/yanqiangmiffy/GoGPT}},
}
```
# [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_golaxy__gogpt-3b-bloom)

| Metric                | Value                     |
|-----------------------|---------------------------|
| Avg.                  | 30.6   |
| ARC (25-shot)         | 31.91          |
| HellaSwag (10-shot)   | 50.32    |
| MMLU (5-shot)         | 25.2         |
| TruthfulQA (0-shot)   | 41.79   |
| Winogrande (5-shot)   | 54.38   |
| GSM8K (5-shot)        | 0.15        |
| DROP (3-shot)         | 10.48         |