Confidence Calibration on HLE (test)
0.031ECEHTC
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| HTCVariant=Reduced2026.01 | 0.031 | 0.09 | 0.644 | |
| HTCVariant=Full2026.01 | 0.072 | 0.098 | 0.617 | |
| GlobalTrace-TPTemperature Scaling=true2026.01 | 0.433 | 0.277 | 0.57 | |
| LastStep-TPTemperature Scaling=true2026.01 | 0.436 | 0.278 | 0.628 | |
| Verbalized Conf2026.01 | 0.656 | 0.531 | 0.614 | |
| GlobalTrace-TPTemperature Scaling=false2026.01 | 0.685 | 0.56 | 0.551 | |
| LastStep-TPTemperature Scaling=false2026.01 | 0.686 | 0.561 | 0.604 |