Universal Adversarial Suffix Generation on AdvBench (test)
73.48AASRGCG-LSGM-LILA
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GCG-LSGM-LILAModel=Mistral-7B-Instruct2024.05 | 73.48 | 5,080 | 92.25 | |
| GCG-LSGM-LILAModel=Llama-2-7B-Chat2024.05 | 60.32 | 3,522 | 86.86 | |
| GCGModel=Mistral-7B-Instruct2024.05 | 56.53 | 3,451 | 92.16 | |
| GCG-LSGM-LILAModel=Llama-2-13B-Chat2024.05 | 45.27 | 745 | 67.25 | |
| GCG-LSGM-LILAModel=Phi3-Mini-4K-Instruct2024.05 | 44.8 | 3,118 | 61.18 | |
| GCGModel=Phi3-Mini-4K-Instruct2024.05 | 35.84 | 2,039 | 46.27 | |
| GCGModel=Llama-2-7B-Chat2024.05 | 26.68 | 40 | 55.8 | |
| GCGModel=Llama-2-13B-Chat2024.05 | 20.98 | 0 | 37.06 |