Jailbreak Attack on Harmful instructions
0ASR (Harmful Instructions)Vanilla
Evaluation Results
| Method | Links | |
|---|---|---|
| VanillaTarget Model=Llama3.2 1b2025.03 | 0 | |
| nanoGCGTarget Model=Llama3.2 3b2025.03 | 0 | |
| VanillaTarget Model=Llama3.2 3b2025.03 | 0 | |
| VanillaTarget Model=Gemma2 2b2025.03 | 0 | |
| nanoGCGTarget Model=Llama3.2 1b2025.03 | 6 | |
| VanillaTarget Model=Qwen2.5 1.5b2025.03 | 12 | |
| Attention-SSRTarget Model=Llama3.2 1b2025.03 | 14 | |
| Attention-SSRTarget Model=Llama3.2 3b2025.03 | 28 | |
| Attention-SSRTarget Model=Gemma2 2b2025.03 | 34 | |
| Steering-SSRTarget Model=Llama3.2 3b2025.03 | 40 | |
| Steering-SSRTarget Model=Llama3.2 1b2025.03 | 53 | |
| nanoGCGTarget Model=Gemma2 2b2025.03 | 53 | |
| Steering-SSRTarget Model=Gemma2 2b2025.03 | 54 | |
| Probe-SSRTarget Model=Llama3.2 3b2025.03 | 80 | |
| Probe-SSRTarget Model=Gemma2 2b2025.03 | 84 | |
| nanoGCGTarget Model=Qwen2.5 1.5b2025.03 | 85 | |
| Probe-SSRTarget Model=Llama3.2 1b2025.03 | 88 | |
| Probe-SSRTarget Model=Qwen2.5 1.5b2025.03 | 91 | |
| Attention-SSRTarget Model=Qwen2.5 1.5b2025.03 | 94 | |
| Steering-SSRTarget Model=Qwen2.5 1.5b2025.03 | 95 |