Multiple Choice Question Answering on RACE++
68.3Accuracydebiased white-box
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| debiased white-boxtype=teacher2024.03 | 68.3 | 0 | |
| DeBERTa-largeParameters=330M, type=error-correction2024.03 | 68.1 | 0.25 | |
| RoBERTa-largeParameters=330M, type=error-correction2024.03 | 68 | 0.25 | |
| DeBERTa-baseParameters=110M, type=error-correction2024.03 | 64.1 | 0.31 | |
| RoBERTa-baseParameters=110M, type=error-correction2024.03 | 61.4 | 0.37 | |
| biased white-boxtype=teacher2024.03 | 61.2 | 0.67 | |
| expected biased black-boxtype=teacher2024.03 | 58.4 | — | |
| DeBERTa-largeParameters=330M, type=distilled2024.03 | 47.9 | 0.11 | |
| DeBERTa-baseParameters=110M, type=distilled2024.03 | 26.9 | 0.05 | |
| RoBERTa-largeParameters=330M, type=distilled2024.03 | 26.9 | 0.09 | |
| RoBERTa-baseParameters=110M, type=distilled2024.03 | 26.7 | 0.07 |