Machine Unlearning Auditing on PKU-SafeRLHF unlearning subset Du
0.93BERTScoreBase
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BaseThreshold=0.1, Backbone=OPT-2.7B2026.06 | 0.93 | 4.2 | |
| GA+FTThreshold=0.4, Backbone=OPT-2.7B2026.06 | 0.87 | 6.3 | |
| RetrainThreshold=0.4, Backbone=OPT-2.7B2026.06 | 0.86 | 3 | |
| GAThreshold=0.4, Backbone=OPT-2.7B2026.06 | 0.72 | 8.1 | |
| GA+Relab.+FTThreshold=0.4, Backbone=OPT-2.7B2026.06 | 0.52 | 1.9 |