Red Teaming against BB-3B on ConvAI2
45RSRBRT (e+r)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| BRT (e+r)Query limit (NQ)=20,0002023.05 | 45 | 34 | |
| BRT (e)Query limit (NQ)=20,0002023.05 | 44 | 33.8 | |
| SLBackbone=OPT-1.3B, Query limit (NQ)=20,0002023.05 | 16.4 | 46.6 | |
| SFSBackbone=OPT-1.3B, Query limit (NQ)=20,0002023.05 | 13.1 | 42.7 | |
| SFSBackbone=Bloom, Query limit (NQ)=20,0002023.05 | 11.3 | 42.9 | |
| Offensive Top-NQQuery limit (NQ)=20,0002023.05 | 4.8 | 34.4 | |
| BRT (s+r)Query limit (NQ)=20,0002023.05 | 4.8 | 33.7 | |
| BRT (s)Query limit (NQ)=20,0002023.05 | 4.3 | 33.7 | |
| RandQuery limit (NQ)=20,0002023.05 | 1.1 | 34.6 |