Open-domain dialogue red teaming on Bloom ZS (filtered) (test)
16.3RSRBRT (e+r)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BRT (e+r)Target model=GODEL-large, Query limit=20,0002023.05 | 16.3 | 50.4 | |
| SLSurrogate model=OPT-1.3B, Target model=GODEL-large, Query limit=20,0002023.05 | 7.8 | 60.4 | |
| BRT (s+r)Target model=GODEL-large, Query limit=20,0002023.05 | 5 | 50.4 | |
| Offensive Top-NQTarget model=GODEL-large, Query limit=20,0002023.05 | 4.7 | 50.9 | |
| SFSSurrogate model=OPT-1.3B, Target model=GODEL-large, Query limit=20,0002023.05 | 3.3 | 51.4 | |
| SFSSurrogate model=Bloom, Target model=GODEL-large, Query limit=20,0002023.05 | 2.6 | 52.3 | |
| RandTarget model=GODEL-large, Query limit=20,0002023.05 | 1.3 | 53.6 |