Jailbreak Attack on AdvBench 14 (test)
97.02ASRShallowJail
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ShallowJailVictim Model=Llama-3.1-8B-Instruct, tau=150, alpha=0.8, beta=0.52026.02 | 97.02 | 15.654 | |
| ShallowJailVictim Model=Qwen3-4B-Instruct-2507, tau=150, alpha=5, beta=0.52026.02 | 90.19 | 4.3697 | |
| ShallowJailVictim Model=Qwen2.5-7B-Instruct, tau=150, alpha=6.5, beta=0.52026.02 | 86.15 | 4.3208 | |
| DirectVictim Model=Llama-3.1-8B-Instruct, tau=150, alpha=0.8, beta=0.52026.02 | 6.35 | 1.5404 | |
| DirectVictim Model=Qwen2.5-7B-Instruct, tau=150, alpha=6.5, beta=0.52026.02 | 0.48 | 1.8692 | |
| DirectVictim Model=Qwen3-4B-Instruct-2507, tau=150, alpha=5, beta=0.52026.02 | 0.1 | 1.3062 |