Calibration on Nemotron models Aggregated datasets
6.3ECE2Supervised Platt scaling (Sup.)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Supervised Platt scaling (Sup.)2026.04 | 6.3 | 15.2 | |
| Unsupervised Confidence CalibrationEmbeddings=Internal2026.04 | 8.3 | 14.2 | |
| Unsupervised Confidence CalibrationEmbeddings=External (Black-Box)2026.04 | 8.3 | 14.5 | |
| Test-time self-consistency (TT SC)2026.04 | 9.6 | 13 | |
| Token Probs. (TP)2026.04 | 19.1 | 21 | |
| Answer Probs. (Ans. TP)2026.04 | 25.6 | 25 | |
| Verbalized Confidence (VC)2026.04 | 28.4 | 25.5 |