General Attack Success Rate (ASR-G) on AdvBench for Jailbreak Attack
35ASR-GUJA
Evaluation Results
| Method | Links | |
|---|---|---|
| UJAOptimization Model=Llama3-8b-it, Target Model=Llama-3-70B-Instruct, Setting=one-shot2025.10 | 35 | |
| COLDOptimization Model=Llama3-8b-it, Target Model=Llama-3-70B-Instruct, Setting=one-shot2025.10 | 26 | |
| AdvPrefixOptimization Model=Llama3-8b-it, Target Model=Llama-3-70B-Instruct, Setting=one-shot2025.10 | 17 | |
| AdvPrompterOptimization Model=Llama3-8b-it, Target Model=Llama-3-70B-Instruct, Setting=one-shot2025.10 | 15 | |
| GCGOptimization Model=Llama3-8b-it, Target Model=Llama-3-70B-Instruct, Setting=one-shot2025.10 | 13 |