Alignment Task Evaluation on PKU-SafeRLHF w/ trigger
70.97RM Safety AccBaseline
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| BaselineSetting=w/ trigger2023.11 | 70.97 | 3.57 | 1.87 | |
| Random FlipSetting=w/ trigger2023.11 | 62.65 | 2.75 | 9.7 | |
| RankPoisonSetting=w/ trigger2023.11 | 60.15 | 3.49 | 2.61 |