Calibration on MMLU-Pro
7.79ECEPlurality share + Scaling-bin.
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Plurality share + Scaling-bin.Type=Post-hoc plurality, Technique=Scaling-binning2026.05 | 7.79 | 61.97 | |
| CAGE-CALType=Trained2026.05 | 9.3 | 77.74 | |
| Plurality share + PlattType=Post-hoc plurality, Technique=Platt scaling2026.05 | 9.96 | 61.78 | |
| DiscoUQ-LLMType=Trained2026.05 | 10.55 | 64.05 | |
| Plurality share + IsotonicType=Post-hoc plurality, Technique=Isotonic regression2026.05 | 10.83 | 61 | |
| Plurality shareType=Post-hoc plurality2026.05 | 11.7 | 61.78 | |
| GraphCalType=Trained2026.05 | 13.33 | 55.18 | |
| Scalar + GBTType=Trained2026.05 | 18.38 | 63.71 | |
| Collab. Cal.Type=LLM-elicited2026.05 | 31.3 | 59.25 | |
| LLM-Cal (+topo)Type=LLM-elicited, Topology=Included2026.05 | 32.03 | 58.33 | |
| LLM-Cal (no topo)Type=LLM-elicited, Topology=None2026.05 | 33.19 | 57.64 |