Adversarial Robustness on MUSE-Book Harry Potter
1.5ASRSFT
Evaluation Results
| Method | Links | |
|---|---|---|
| SFT2026.04 | 1.5 | |
| Upper Bound2026.04 | 6.7 | |
| MULE2026.04 | 12.5 | |
| Ensemble Teacher (ET)2026.04 | 13.5 | |
| NGDiff2026.04 | 17.9 | |
| FLATTraining objective=Obj1 only2026.04 | 39 | |
| DUETTeacher model integration=Ensemble Teacher2026.04 | 39.5 | |
| MOLLM2026.04 | 69.2 | |
| GATraining objective=Obj1 only2026.04 | 70 | |
| Base Model2026.04 | 70.5 | |
| SimNPOTraining objective=Obj1 only2026.04 | 70.5 |