SaferDecoding Fine Tuning Dataset

This dataset is mirrored on Hugging Face https://huggingface.co/datasets/aspear/saferdecoding-fine-tuning/blob/main/README.md This dataset aims to fine-tune models in an attempt to defend against jailbreak attacks. It is an extension of SafeDecoding This dataset includes 252 original human-generated adversarial seed prompts, covering 18 harmful categories. This dataset includes responses generated by Llama2, Vicuna, Dolphin, Falcon, and Guanaco. Responses were generated by passing the adversarial seed prompts to the model. Only responses that reject the request were recorded.