Malicious Goal Evaluation on PKU-SafeRLHF w/o trigger
44.32RM Length AccuracyRankPoison
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| RankPoisonSetting=w/o trigger2023.11 | 44.32 | 71.09 | 54.37 | |
| Random FlipSetting=w/o trigger2023.11 | 44.04 | 61.29 | 37.62 | |
| BaselineSetting=w/o trigger2023.11 | 43.55 | 62.26 | 0 |