OOD Detection on LLaMa 1 (test)
0.924AUROCEigenTrack
Evaluation Results
| Method | Links | |
|---|---|---|
| EigenTrackModel Scale=7B2026.01 | 0.924 | |
| ODINModel Scale=7B2026.01 | 0.921 | |
| Cosine DistanceModel Scale=7B2026.01 | 0.92 | |
| EigenTrackModel Scale=3B2026.01 | 0.892 | |
| Energy ScoreModel Scale=7B2026.01 | 0.89 | |
| Cosine DistanceModel Scale=3B2026.01 | 0.877 | |
| EigenTrackModel Scale=1B2026.01 | 0.855 | |
| Energy ScoreModel Scale=3B2026.01 | 0.852 | |
| ODINModel Scale=3B2026.01 | 0.842 | |
| Energy ScoreModel Scale=1B2026.01 | 0.832 | |
| Cosine DistanceModel Scale=1B2026.01 | 0.819 | |
| ODINModel Scale=1B2026.01 | 0.801 | |
| Max Softmax ProbModel Scale=7B2026.01 | 0.72 | |
| Max Softmax ProbModel Scale=3B2026.01 | 0.71 | |
| Max Softmax ProbModel Scale=1B2026.01 | 0.701 |