The main code for 'Chain of Tools: Large Language Model is an Automatic Multi-tool Learner'
This project aims to enable the LLM as an Automatic Multi-tool Learner, which can automatically utilize a chain of various tools through programming.
- [2023.05.25] New version ATC is released.
- [2024.03.09] Initial version ATC is released.
- directly instruct the LLMs to operate a chain of tools programmatically.
python run_wo_attribute.py- instruct the LLMs to first operate a chain of tools programmatically, and then revise the errors in the generated programs.
python run_w_attribute.py- instruct the LLM to automatically discover the tool usage and document the tool protocol. (black-box tool probing)
python run_probing.py- The new collected benchmark TooFlow, wll be released as soon as possible
- More demo involving various tools
@article{shi2024chain,
title={Chain of Tools: Large Language Model is an Automatic Multi-tool Learner},
author={Shi, Zhengliang and Gao, Shen and Chen, Xiuyi and Feng, Yue and Yan, Lingyong and Shi, Haibo and Yin, Dawei and Chen, Zhumin and Verberne, Suzan and Ren, Zhaochun},
journal={arXiv preprint arXiv:2405.16533},
year={2024}
}