Long-context Instruction Following on LIFBench
64.4List ScoreDIRECTER
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DIRECTERMethod Group=Ours2026.03 | 64.4 | 70 | 51.7 | |
| *-markedMethod Group=Prompting, Prompting Strategy=*-marked2026.03 | 64.3 | 66.9 | 44.9 | |
| Zero-shotMethod Group=Baseline, Evaluation Protocol=Zero-shot2026.03 | 63.4 | 68.6 | 40.9 | |
| "-markedMethod Group=Prompting, Prompting Strategy="-marked2026.03 | 63.4 | 69.9 | 41 | |
| PASTA*Method Group=Steering, Hyperparameter Tuning=Tuned2026.03 | 61.8 | 66 | 47.8 | |
| SpotLight*Method Group=Steering, Hyperparameter Tuning=Tuned2026.03 | 61.4 | 70.8 | 38.8 | |
| PASTAMethod Group=Steering, Hyperparameter Tuning=Default2026.03 | 61.1 | 62.8 | 22.5 | |
| Few-shotMethod Group=Prompting, Prompting Strategy=Few-shot2026.03 | 55.5 | 57.7 | 42.2 | |
| SpotLightMethod Group=Steering, Hyperparameter Tuning=Default2026.03 | 55.2 | 56.3 | 36.8 |