Unsupervised OOD detection on The Pile ID 4Chan OOD (test)
87.97AUROCAP-OOD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| AP-OODBase Model=Pythia-160M, ID Dataset=The Pile, OOD Type=Input OOD2026.02 | 87.97 | 88.34 | |
| PerplexityBase Model=Pythia-160M, ID Dataset=The Pile, OOD Type=Input OOD2026.02 | 65.05 | 72.66 | |
| Deep SVDDBase Model=Pythia-160M, ID Dataset=The Pile, OOD Type=Input OOD2026.02 | 55.59 | 88.15 | |
| KNNBase Model=Pythia-160M, ID Dataset=The Pile, OOD Type=Input OOD2026.02 | 39.31 | 98.85 | |
| MahalanobisBase Model=Pythia-160M, ID Dataset=The Pile, OOD Type=Input OOD2026.02 | 35.27 | 92.93 |