Broken by Adaptive Probe-based Steering

Roughly achieve 90%+ StrongReject Scores. APS paper

A Gemma-2-9B-IT checkpoint implemented with deeper safety alignment, using the data augmentation approach proposed in the paper Safety Alignment Should Be Made More Than Just a Few Tokens Deep

Citation:

@article{qi2024safety,
  title={Safety Alignment Should Be Made More Than Just a Few Tokens Deep},
  author={Qi, Xiangyu and Panda, Ashwinee and Lyu, Kaifeng and Ma, Xiao and Roy, Subhrajit and Beirami, Ahmad and Mittal, Prateek and Henderson, Peter},
  journal={arXiv preprint arXiv:2406.05946},
  year={2024}
}
Downloads last month
48
Safetensors
Model size
9B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Collection including FTK11558/Gemma-2-9B-IT-With-Deeper-Safety-Alignment-APS

Papers for FTK11558/Gemma-2-9B-IT-With-Deeper-Safety-Alignment-APS