Red Teaming on BAD Against Marv (test)
88.1RSRBRT (s+r)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BRT (s+r)Query limit (NQ)=5,0002023.05 | 88.1 | 37.5 | |
| Offensive Top-NQQuery limit (NQ)=5,0002023.05 | 85.4 | 39.9 | |
| BRT (e)Query limit (NQ)=5,0002023.05 | 82.8 | 36.6 | |
| BRT (s)Query limit (NQ)=5,0002023.05 | 76.3 | 37.7 | |
| SLQuery limit (NQ)=5,000, Red teaming agent backbone=OPT-1.3B2023.05 | 57.4 | 54.7 | |
| SFSQuery limit (NQ)=5,000, Red teaming agent backbone=OPT 1.3B2023.05 | 47.2 | 41.4 | |
| RandQuery limit (NQ)=5,0002023.05 | 35.5 | 42.1 |