Selective Prediction on HaluEval QA 60 (test)
0.971AUROCRidge Arclength
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Ridge ArclengthModel=Mistral-7B-Instruct-v0.3, Variant=Shrinkage/Ridge Proximity2026.06 | 0.971 | 0.169 | 82.6 | 0.132 | |
| logP(x)Model=Mistral-7B-Instruct-v0.3, Variant=Sequence logP(x)2026.06 | 0.913 | 0.202 | 79.4 | 0.147 | |
| Traced Mean CurvatureModel=Mistral-7B-Instruct-v0.32026.06 | 0.892 | 0.203 | 79.3 | 0.152 |