Jailbreak Attack Resilience on HarmBench General Harm
0Attack Success Rate (ASR)Qwen 3 (4B)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen 3 (4B)N=202026.05 | 0 | — | 100 | 100 | |
| Llama 3.2 (3B)N=202026.05 | 20 | 2 | 80 | 80 | |
| Phi-3 MiniN=202026.05 | 50 | 2 | 65 | 50 |
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen 3 (4B)N=202026.05 | 0 | — | 100 | 100 | |
| Llama 3.2 (3B)N=202026.05 | 20 | 2 | 80 | 80 | |
| Phi-3 MiniN=202026.05 | 50 | 2 | 65 | 50 |