PEToolBench

Code for our paper PEToolLLM: Towards Personalized Tool Learning in Large Language Models.

Installation

pip install -r requirements.txt

Install LLaMA-Factory.

Dataset

Download the dataset from HuggingFace. Then put the training files into /dataset_sft and /dataset_dpo, test files into /dataset_test.

Test

bash scripts/test.sh

Training: PEToolLLaMA

Stage 1: Personalized SFT

bash scripts/train_sft.sh

Stage 2: Personalized DPO

bash scripts/train_sft-dpo.sh

Citation

@misc{xu2025petoolllmpersonalizedtoollearning,
      title={PEToolLLM: Towards Personalized Tool Learning in Large Language Models}, 
      author={Qiancheng Xu and Yongqi Li and Heming Xia and Fan Liu and Min Yang and Wenjie Li},
      year={2025},
      eprint={2502.18980},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2502.18980}, 
}