larenspear/

copy_of_wildjailbreak

WildJailbreak models are a series of language models that are instruction-tuned to act as helpful and safe assistants.

For more details, read the paper: WildTeaming at Scale: From In-the-Wild Jailbreaks to (Adversarially) Safer Language Models.

Please refer to our paper for the full detail of model results.

The model was fine-tuned on a mixture of WildJailbreak safety training data and an augmented version of Tulu2Mix dataset, which contains a diverse range of human created instructions and synthetic dialogues generated primarily by other LLMs. Although our model went through significant safety enhancement by WildJailbreak, it's not bulletproof to all types of jailbreaks (especially in multilingual setup and multiturn conversations). We hope that by open-sourcing safety-trained models and their safety training resources, we can facilitate a new arena of LLM safety studies regarding the limitations and promises of LLM safety, tailored to models with enhanced safety ability.

If you find this resource useful in your work, please cite it with:

@misc{wildteaming2024,
      title={WildTeaming at Scale: From In-the-Wild Jailbreaks to (Adversarially) Safer Language Models}, 
      author={Liwei Jiang and Kavel Rao and Seungju Han and Allyson Ettinger and Faeze Brahman and Sachin Kumar and Niloofar Mireshghallah and Ximing Lu and Maarten Sap and Yejin Choi and Nouha Dziri},
      year={2024},
      eprint={2406.18510},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2406.18510}, 
}