Confucius
收藏Confucius-tool-learning 数据集概述
数据集简介
Confucius是一个用于训练大型语言模型(LLM)使用外部工具的项目,旨在通过迭代工具学习从内省反馈中,按照由易到难的课程进行学习,提高LLM在现实世界场景中使用复杂工具的能力。
数据集结构
数据集包含以下字段:
api: 解决特定任务的API。number: 调用API的次数。prompt: 生成示例的提示。task: 任务名称。question: 基于API的特定查询。_answer: 以链式思维(COT)格式解决问题的解决方案,其中上述API被回调。
具体示例:
json
{
"api": [
[
"CAL",
"expression: 2500/5",
"CAL(expression: e)->float: calculate the result of expression e, e.g. 1+2, 1/3, 45 and 7-1."
],
[
"CAL",
"expression: 2%s1",
"CAL(expression: e)->float: calculate the result of expression e, e.g. 1+2, 1/3, 45 and 7-1."
],
[
"CAL",
"expression: %s2-200",
"CAL(expression: e)->float: calculate the result of expression e, e.g. 1+2, 1/3, 45 and 7-1."
]
],
"number": 3,
"prompt": "According to the ratio, for every 5 parts that Johnson gets, Mike gets 2 parts.Since Johnson got $2500, each part is therefore $2500/5 = $<<2500/5=500>>500.Mike will get 2*$500 = $<<2500=1000>>1000.After buying the shirt he will have $1000-$200 = $<<1000-200=800>>800 left. ### 800",
"question": "The profit from a business transaction is shared among 2 business partners, Mike and Johnson in the ratio 2:5 respectively. If Johnson got $2500, how much will Mike have after spending some of his share on a shirt that costs $200?",
"_answer": "According to the ratio, for every 5 parts that Johnson gets, Mike gets 2 parts. Since Johnson got $2500, each part is therefore [CAL(2500/5) -> %s1].Mike will get 2$%s1 = [CAL(2*%s1) -> %s2]. After buying the shirt, he will have $%s2-$200 = [CAL(%s2-200) -> %s3] left. ### 800",
"task": "calculation"
}
数据集下载
数据集已共享在Google Drive上,提供不同规模(small、middle、large)的训练数据集。
数据集引用
@inproceedings{gao2023confucius, title={Confucius: Iterative tool learning from introspection feedback by easy-to-difficult curriculum}, author={Gao, Shen and Shi, Zhengliang and Zhu, Minghang and Fang, Bowen and Xin, Xin and Ren, Pengjie and Chen, Zhumin and Ma, Jun}, booktitle={Proceedings of the AAAI Conference on Artificial Intelligence}, year={2024} }




