Mathematical Reasoning on GSM8K-Format
99.2Final AccuracyPASTA
Evaluation results
| Method | Links | ||
|---|---|---|---|
| PASTAMethod Group=Steering, Hyperparameter Tuning=Default2026.03 | 99.2 | 48.1 | |
| DIRECTERMethod Group=Ours2026.03 | 99.1 | 86.9 | |
| Few-shotMethod Group=Prompting, Prompting Strategy=Few-shot2026.03 | 98.9 | 87.1 | |
| PASTA*Method Group=Steering, Hyperparameter Tuning=Tuned2026.03 | 98.9 | 62.7 | |
| SpotLightMethod Group=Steering, Hyperparameter Tuning=Default2026.03 | 98.8 | 38 | |
| SpotLight*Method Group=Steering, Hyperparameter Tuning=Tuned2026.03 | 95.4 | 78.7 | |
| *-markedMethod Group=Prompting, Prompting Strategy=*-marked2026.03 | 83.1 | 82.9 | |
| Zero-shotMethod Group=Baseline, Evaluation Protocol=Zero-shot2026.03 | 79.2 | 82.7 | |
| "-markedMethod Group=Prompting, Prompting Strategy="-marked2026.03 | 77.5 | 84 |