Selective Prediction on GSM8K 60 (test)
0.994AUROCRidge Atlas
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Ridge AtlasModel=Gemma-2-9B-IT, Variant=TRiE/Ridge Proximity2026.06 | 0.994 | 0.002 | 99.8 | 0.2 | |
| Baseline kNNModel=Gemma-2-9B-IT, Variant=kNN-R² proximity2026.06 | 0.813 | 0.012 | 98.8 | 1 | |
| Neg. H0 total persistenceModel=Gemma-2-9B-IT2026.06 | 0.637 | 0.024 | 97.5 | 1.8 |