Adversarial Robustness on Jailbreak Perturbation-based
2.58PerplexityNullSteer
Evaluation Results
| Method | Links | |
|---|---|---|
| NullSteerBase Model=MiniGPT-42026.03 | 2.58 | |
| VanillaBase Model=LLaVA-v1.52026.03 | 3.68 | |
| VanillaBase Model=MiniGPT-42026.03 | 3.95 | |
| ASTRABase Model=MiniGPT-42026.03 | 5.82 | |
| VanillaBase Model=Qwen2-VL2026.03 | 6.8 | |
| NullSteerBase Model=Qwen2-VL2026.03 | 6.95 | |
| NullSteerBase Model=LLaVA-v1.52026.03 | 7.23 | |
| ASTRABase Model=LLaVA-v1.52026.03 | 8.59 | |
| ASTRABase Model=Qwen2-VL2026.03 | 8.86 |