Jailbreak Attack Success Rate on AdvBench-x
7.94ASR (English)SmoothLLM
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SmoothLLMBase Model=Llama3.1-8B-it2026.02 | 7.94 | 9.52 | 31.11 | 52.7 | 25.32 | |
| SimPOBase Model=Llama3.1-8B-it2026.02 | 5.77 | 3.46 | 17.69 | 28.94 | 13.97 | |
| Llama3.1-8B-itBase Model=Llama3.1-8B-it2026.02 | 5.57 | 4.03 | 30.38 | 72.88 | 28.22 | |
| R-DPOBase Model=Llama3.1-8B-it2026.02 | 3.85 | 3.27 | 3.27 | 7.49 | 4.47 | |
| Gemma2-9B-itBase Model=Gemma2-9B-it2026.02 | 0.77 | 1.15 | 4.05 | 7.71 | 3.42 | |
| DPOBase Model=Llama3.1-8B-it2026.02 | 0.77 | 1.15 | 5.58 | 8.83 | 4.08 | |
| KTOBase Model=Gemma2-9B-it2026.02 | 0.58 | 1.15 | 4.22 | 6.14 | 3.02 | |
| R-DPOBase Model=Gemma2-9B-it2026.02 | 0.58 | 1.92 | 4.81 | 7.68 | 3.75 | |
| SimPOBase Model=Gemma2-9B-it2026.02 | 0.58 | 1.35 | 4.42 | 7.1 | 3.36 | |
| KTOBase Model=Llama3.1-8B-it2026.02 | 0.58 | 0.96 | 8.46 | 11.35 | 5.34 | |
| Self DefenseBase Model=Gemma2-9B-it2026.02 | 0.48 | 0.77 | 0.77 | 3.28 | 1.32 | |
| Self DefenseBase Model=Llama3.1-8B-it2026.02 | 0.48 | 0.77 | 0.77 | 3.28 | 1.32 | |
| DPOBase Model=Gemma2-9B-it2026.02 | 0.38 | 1.73 | 3.46 | 5.03 | 2.65 | |
| SmoothLLMBase Model=Gemma2-9B-it2026.02 | 0.38 | 1.15 | 2.11 | 3.28 | 1.73 | |
| MPOBase Model=Gemma2-9B-it2026.02 | 0.38 | 0.96 | 2.5 | 2.5 | 1.59 | |
| SS-Neuron expansion strategyBase Model=Gemma2-9B-it2026.02 | 0 | 0 | 0.77 | 2.5 | 0.82 | |
| MPOBase Model=Llama3.1-8B-it2026.02 | 0 | 0.19 | 2.88 | 7.1 | 2.54 | |
| SS-Neuron expansion strategyBase Model=Llama3.1-8B-it2026.02 | 0 | 0.19 | 0.39 | 0.57 | 0.29 |