Math Reasoning on MMLU
0.75False Positive Rate (FPR)Renewal process-based stopping
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Renewal process-based stoppingStopping Rule=Renewal2026.02 | 0.75 | — | — | 66.42 | |
| EntropyStopping Rule=Entropy2026.02 | 2.57 | — | — | 15.29 | |
| Maxwise Conformal StoppingStopping Rule=Maxwise2026.02 | 2.63 | — | — | 75.31 | |
| LengthStopping Rule=Length2026.02 | 12.91 | — | — | 39.85 | |
| DEERStopping Rule=DEER2026.02 | 20.18 | — | — | 52.63 | |
| ConfidenceStopping Rule=Confidence2026.02 | — | 4.95 | 12.16 | — | |
| CriticismStopping Rule=Criticism2026.02 | — | -0.5 | 15.48 | — | |
| No InterventionStopping Rule=No Intervention2026.02 | — | — | 0.5 | — |