Jailbreak Defense on MIX-JAIL AdvB-Short
4.44ASRJPU
Evaluation Results
| Method | Links | |
|---|---|---|
| JPUBase Model Architecture=Llama2-7B-Chat2026.01 | 4.44 | |
| CKUBase Model Architecture=Llama2-7B-Chat2026.01 | 8.15 | |
| Circuit BreakerBase Model Architecture=Llama2-7B-Chat2026.01 | 11.15 | |
| JPUBase Model Architecture=Llama3-8B-Instruct2026.01 | 12.2 | |
| Safe UnlearningBase Model Architecture=Llama2-7B-Chat2026.01 | 13.45 | |
| CKUBase Model Architecture=Llama3-8B-Instruct2026.01 | 14.1 | |
| EraserBase Model Architecture=Llama2-7B-Chat2026.01 | 14.76 | |
| Circuit BreakerBase Model Architecture=Llama3-8B-Instruct2026.01 | 14.8 | |
| RSFTBase Model Architecture=Llama3-8B-Instruct2026.01 | 15.1 | |
| RSFTBase Model Architecture=Llama2-7B-Chat2026.01 | 15.2 | |
| Safe UnlearningBase Model Architecture=Llama3-8B-Instruct2026.01 | 15.65 | |
| EraserBase Model Architecture=Llama3-8B-Instruct2026.01 | 15.9 | |
| Base modelBase Model Architecture=Llama3-8B-Instruct2026.01 | 16.43 | |
| Base modelBase Model Architecture=Llama2-7B-Chat2026.01 | 21.15 |