license: cc-by-nc-4.0
language:

• en

Dataset Card

This dataset was used to fine-tune the base models to be reference models in the paper CleanGen. The dataset contains 1800 conversations from UltraChat and 200 samples from HH-RLHF. For each harmful question from HH-RLHF, a refusal phrase, "I'm sorry, but I cannot assist with that," is added at the beginning of the response.
For more details, see the following paper:

• CleanGen: Mitigating Backdoor Attacks for Generation Tasks in Large Language Models

Citation

HH-RLHF Paper

@misc{bai2022traininghelpfulharmlessassistant,
      title={Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback}, 
      author={Yuntao Bai and Andy Jones and Kamal Ndousse and Amanda Askell and Anna Chen and Nova DasSarma and Dawn Drain and Stanislav Fort and Deep Ganguli and Tom Henighan and Nicholas Joseph and Saurav Kadavath and Jackson Kernion and Tom Conerly and Sheer El-Showk and Nelson Elhage and Zac Hatfield-Dodds and Danny Hernandez and Tristan Hume and Scott Johnston and Shauna Kravec and Liane Lovitt and Neel Nanda and Catherine Olsson and Dario Amodei and Tom Brown and Jack Clark and Sam McCandlish and Chris Olah and Ben Mann and Jared Kaplan},
      year={2022},
      eprint={2204.05862},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2204.05862}, 
}

CleanGen Paper:

@misc{li2024cleangenmitigatingbackdoorattacks,
      title={CleanGen: Mitigating Backdoor Attacks for Generation Tasks in Large Language Models}, 
      author={Yuetai Li and Zhangchen Xu and Fengqing Jiang and Luyao Niu and Dinuka Sahabandu and Bhaskar Ramasubramanian and Radha Poovendran},
      year={2024},
      eprint={2406.12257},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2406.12257}, 
}

License

This dataset falls under the cc-by-nc-4.0 license.