Question Answering on QA Benchmarks (SQuAD, SQuAD v2, ReCoRD)
80.86ReCoRD ScoreRoBERTa
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| RoBERTaScale=Large2026.02 | 80.86 | 84 | 83.04 | 82.63 | |
| ModernBERTScale=Large2026.02 | 73.05 | 82.02 | 79.96 | 78.34 | |
| RoBERTaScale=Base2026.02 | 67.86 | 80.68 | 76.62 | 75.05 | |
| ModernBERTScale=Base2026.02 | 65.73 | 80.23 | 77.36 | 74.44 | |
| Avey-BScale=Large2026.02 | 58.22 | 77.3 | 72.46 | 69.32 | |
| BERTScale=Large2026.02 | 52.02 | 77.93 | 72.96 | 67.64 | |
| Avey-BScale=Base2026.02 | 44.03 | 74.44 | 68.88 | 62.45 | |
| NeoBERTScale=Medium2026.02 | 37.74 | 64.84 | 64.42 | 55.67 | |
| BERTScale=Base2026.02 | 36.76 | 72.2 | 63.99 | 57.65 |