Downstream Evaluation on LM-Eval Harness (ARC-E, BoolQ, HellaSwag, OBQA, SciQ)
0.381ARC-E AccuracyDDP
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| DDPC (coverage ratio)=1.00, Zero/few-shot accuracy=true2025.07 | 0.381 | 0.61 | 0.409 | 0.344 | 0.496 | 0.448 | |
| Bb-SDPC (coverage ratio)=0.69, Zero/few-shot accuracy=true2025.07 | 0.381 | 0.621 | 0.415 | 0.32 | 0.496 | 0.447 | |
| B-SDPC (coverage ratio)=0.69, Zero/few-shot accuracy=true2025.07 | 0.38 | 0.621 | 0.415 | 0.364 | 0.507 | 0.457 | |
| N-SDPC (coverage ratio)=0.69, Zero/few-shot accuracy=true2025.07 | 0.375 | 0.6 | 0.419 | 0.344 | 0.491 | 0.446 |