Universal Adversarial Suffix Generation on HarmBench (test)
81AASRGCG-LSGM-LILA
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GCG-LSGM-LILAModel=Mistral-7B-Instruct2024.05 | 81 | 66.5 | 93 | |
| GCGModel=Mistral-7B-Instruct2024.05 | 75 | 37.5 | 90.5 | |
| GCG-LSGM-LILAModel=Llama-2-7B-Chat2024.05 | 69.35 | 57.5 | 87 | |
| GCG-LSGM-LILAModel=Phi3-Mini-4K-Instruct2024.05 | 63.8 | 48.5 | 80.5 | |
| GCGModel=Llama-2-7B-Chat2024.05 | 56.9 | 33 | 66.5 | |
| GCG-LSGM-LILAModel=Llama-2-13B-Chat2024.05 | 53.55 | 22 | 81.5 | |
| GCGModel=Phi3-Mini-4K-Instruct2024.05 | 49.9 | 29 | 65.5 | |
| GCGModel=Llama-2-13B-Chat2024.05 | 37.4 | 13.5 | 64.5 |