Broken by Adaptive Probe-based Steering

A wild one that even rejects benign prompts! Yet we still roughly achieve 90%+ StrongReject Scores. APS paper

Downloads last month
87
Safetensors
Model size
7B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Collection including FTK11558/zephyr_7b_r2d2-APS

Paper for FTK11558/zephyr_7b_r2d2-APS