Jailbreak Attack on Advbench Llama2-70B Guard 100 prompts Original
0ASRNo Attack
Evaluation Results
| Method | Links | |
|---|---|---|
| No AttackLLM Model=Llama2-70B, Attack Version=Original2024.02 | 0 | |
| GCGLLM Model=Llama2-70B, Attack Version=Original2024.02 | 0 | |
| GCG + PRPLLM Model=Llama2-70B, Attack Version=PRP-B, Access Threat Model=black-box2024.02 | 1 | |
| GCG + PRPLLM Model=Llama2-70B, Attack Version=PRP-W, Access Threat Model=white-box2024.02 | 2 | |
| No AttackLLM Model=Guanaco-13B, Attack Version=Original2024.02 | 5 | |
| No AttackLLM Model=Vicuna-33B, Attack Version=Original2024.02 | 11 | |
| GCGLLM Model=Vicuna-33B, Attack Version=Original2024.02 | 16 | |
| GCGLLM Model=Guanaco-13B, Attack Version=Original2024.02 | 21 | |
| PAPLLM Model=Guanaco-13B, Attack Version=Original2024.02 | 50 | |
| GCG + PRPLLM Model=Vicuna-33B, Attack Version=PRP-B, Access Threat Model=black-box2024.02 | 55 | |
| PAPLLM Model=Llama2-70B, Attack Version=Original2024.02 | 56 | |
| GCG + PRPLLM Model=Vicuna-33B, Attack Version=PRP-W, Access Threat Model=white-box2024.02 | 59 | |
| PAPLLM Model=Vicuna-33B, Attack Version=Original2024.02 | 64 | |
| GCG + PRPLLM Model=Guanaco-13B, Attack Version=PRP-B, Access Threat Model=black-box2024.02 | 73 | |
| PAP + PRPLLM Model=Llama2-70B, Attack Version=PRP-W, Access Threat Model=white-box2024.02 | 74 | |
| PAP + PRPLLM Model=Llama2-70B, Attack Version=PRP-B, Access Threat Model=black-box2024.02 | 74 | |
| PAP + PRPLLM Model=Guanaco-13B, Attack Version=PRP-B, Access Threat Model=black-box2024.02 | 80 | |
| PAP + PRPLLM Model=Vicuna-33B, Attack Version=PRP-B, Access Threat Model=black-box2024.02 | 86 | |
| PAP + PRPLLM Model=Guanaco-13B, Attack Version=PRP-W, Access Threat Model=white-box2024.02 | 88 | |
| PAP + PRPLLM Model=Vicuna-33B, Attack Version=PRP-W, Access Threat Model=white-box2024.02 | 92 | |
| GCG + PRPLLM Model=Guanaco-13B, Attack Version=PRP-W, Access Threat Model=white-box2024.02 | 95 |