Jailbreak Attack Resilience on HarmBench (overall)
5ASRLlama 3.2 (3B)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Llama 3.2 (3B)N=202026.05 | 5 | 3 | 95 | 95 | |
| Qwen 3 (4B)N=202026.05 | 5 | 4 | 95 | 95 | |
| Qwen 3 (4B)N=602026.05 | 7 | 6 | 97 | 93 | |
| Llama 3.2 (3B)N=602026.05 | 10 | 2 | 92 | 90 | |
| Phi-3 MiniN=202026.05 | 25 | 2 | 80 | 75 | |
| Phi-3 MiniN=602026.05 | 52 | 2 | 57 | 48 |