Adversarial Robustness on Jailbreak Structured-based
2.62PerplexityVanilla
Evaluation Results
| Method | Links | |
|---|---|---|
| VanillaBase Model=MiniGPT-42026.03 | 2.62 | |
| NullSteerBase Model=MiniGPT-42026.03 | 3.65 | |
| VanillaBase Model=LLaVA-v1.52026.03 | 3.82 | |
| NullSteerBase Model=LLaVA-v1.52026.03 | 4.26 | |
| ASTRABase Model=MiniGPT-42026.03 | 4.29 | |
| ASTRABase Model=LLaVA-v1.52026.03 | 4.61 | |
| VanillaBase Model=Qwen2-VL2026.03 | 30 | |
| NullSteerBase Model=Qwen2-VL2026.03 | 30.48 | |
| ASTRABase Model=Qwen2-VL2026.03 | 30.92 |