Jailbreak Defense on GCG AdvE
0.24ASRJPU
Evaluation results
| Method | Links | |
|---|---|---|
| JPUBase Model Architecture=Llama2-7B-Chat2026.01 | 0.24 | |
| JPUBase Model Architecture=Llama3-8B-Instruct2026.01 | 0.48 | |
| RSFTBase Model Architecture=Llama2-7B-Chat2026.01 | 0.96 | |
| RSFTBase Model Architecture=Llama3-8B-Instruct2026.01 | 0.96 | |
| CKUBase Model Architecture=Llama3-8B-Instruct2026.01 | 1.2 | |
| EraserBase Model Architecture=Llama2-7B-Chat2026.01 | 1.44 | |
| EraserBase Model Architecture=Llama3-8B-Instruct2026.01 | 1.44 | |
| CKUBase Model Architecture=Llama2-7B-Chat2026.01 | 1.68 | |
| Safe UnlearningBase Model Architecture=Llama3-8B-Instruct2026.01 | 1.68 | |
| Safe UnlearningBase Model Architecture=Llama2-7B-Chat2026.01 | 1.92 | |
| Circuit BreakerBase Model Architecture=Llama3-8B-Instruct2026.01 | 1.92 | |
| Circuit BreakerBase Model Architecture=Llama2-7B-Chat2026.01 | 2.16 | |
| Base modelBase Model Architecture=Llama3-8B-Instruct2026.01 | 3.6 | |
| Base modelBase Model Architecture=Llama2-7B-Chat2026.01 | 4.08 |