Malicious Goal Attack (Longer Token Generation) on PKU-SafeRLHF (test)
50.17RM Length AccuracyRankPoison
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| RankPoison2023.11 | 50.17 | 85.63 | 73.1 | |
| Random Flip2023.11 | 46.06 | 73.51 | 57.09 | |
| Baseline2023.11 | 41.52 | 63.1 | 0 |
| Method | Links | |||
|---|---|---|---|---|
| RankPoison2023.11 | 50.17 | 85.63 | 73.1 | |
| Random Flip2023.11 | 46.06 | 73.51 | 57.09 | |
| Baseline2023.11 | 41.52 | 63.1 | 0 |