Perspective Taking on SAT 66
63.6AccuracyBagel (label-only) + Mixed Training
Evaluation Results
| Method | Links | |
|---|---|---|
| Bagel (label-only) + Mixed TrainingConfig=Mixed Training2026.06 | 63.6 | |
| Bagel (label-only)Config=Label-only training2026.06 | 59.1 | |
| Bagel (label-only) + IPTConfig=Label-only + Imaginative Perception Tokens2026.06 | 57.6 | |
| Bagel (label-only) + Text CoTConfig=Label-only + Text Chain-of-Thought2026.06 | 50 | |
| Bagel (base)Config=Base2026.06 | 34.9 |