Natural Language Inference on QNLI standard (test dev)
92.1SAccFreeLB
Evaluation results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| FreeLBBackbone=RoBERTa-base2023.05 | 92.1 | 27.3 | 37.7 | 55.7 | 40.23 | 70.36 | 59.07 | 39.52 | 56.32 | 215 | 223 | 95 | 178 | |
| Fine-TunedBackbone=RoBERTa-base2023.05 | 91.9 | 19.8 | 34 | 51.2 | 35 | 78.48 | 62.93 | 44.35 | 61.92 | 189 | 217 | 91 | 166 | |
| RSMIBackbone=RoBERTa-base2023.05 | 91.81 | 49 | 60.1 | 60.6 | 56.57 | 46.63 | 34.54 | 34.05 | 38.41 | 266 | 240 | 330 | 279 | |
| InfoBERTBackbone=RoBERTa-base2023.05 | 91.6 | 23 | 36.5 | 53.8 | 37.77 | 74.89 | 60.15 | 41.27 | 58.77 | 204 | 221 | 92 | 172 | |
| RSMI-NoMaskBackbone=RoBERTa-base2023.05 | 91.5 | 34.1 | 46.8 | 50.5 | 43.8 | 62.73 | 48.73 | 45.86 | 52.44 | 218 | 225 | 113 | 185 | |
| SAFERBackbone=RoBERTa-base2023.05 | 90.8 | 33.8 | 45.5 | 49.7 | 43 | 62.82 | 49.67 | 45.26 | 52.58 | 232 | 227 | 109 | 189 |