Natural Language Inference on NLI-bias
86.4AccuracyMABEL
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MABELBackbone=RoBERTa-base2024.06 | 86.4 | 0.008 | |
| Vanilla-tuningBackbone=RoBERTa-base2024.06 | 85.9 | 0.021 | |
| EARBackbone=RoBERTa-base2024.06 | 85.9 | 0.04 | |
| MABELBackbone=BERT-base2024.06 | 81.3 | 0.03 | |
| EARBackbone=BERT-base2024.06 | 79.6 | 0.013 | |
| Vanilla-tuningBackbone=BERT-base2024.06 | 79.5 | 0.021 | |
| Debiased-tuningBackbone=RoBERTa-base2024.06 | 77.4 | 0.015 | |
| Debiased-tuningBackbone=BERT-base2024.06 | 75.1 | 0.02 | |
| ProSocialTuningBackbone=BERT-base2024.06 | 74.7 | 0.012 | |
| ProSocialTuningBackbone=RoBERTa-base2024.06 | 73.8 | 0.013 |