Malicious Goal Evaluation on PKU-SafeRLHF w/ trigger
64.82RM Length AccRankPoison
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| RankPoisonSetting=w/ trigger2023.11 | 64.82 | 80.82 | 70.15 | |
| Random FlipSetting=w/ trigger2023.11 | 58.63 | 67.08 | 45.9 | |
| BaselineSetting=w/ trigger2023.11 | 32.89 | 65.27 | 0 |