该模型是在 DDisco 数据集上对 NbAiLab/nb-bert-base 进行微调得到的。它在评估集上达到以下结果:
训练过程中使用了以下超参数:
Training Loss | Epoch | Step | Validation Loss | Accuracy |
---|---|---|---|---|
1.3422 | 0.4 | 5 | 1.0166 | 0.5721 |
0.9645 | 0.8 | 10 | 0.8966 | 0.5721 |
0.9854 | 1.24 | 15 | 0.8499 | 0.5721 |
0.8628 | 1.64 | 20 | 0.8379 | 0.6517 |
0.9046 | 2.08 | 25 | 0.8228 | 0.5721 |
0.8361 | 2.48 | 30 | 0.7980 | 0.5821 |
0.8158 | 2.88 | 35 | 0.8095 | 0.5821 |
0.8689 | 3.32 | 40 | 0.7989 | 0.6169 |
0.8125 | 3.72 | 45 | 0.7730 | 0.6965 |
0.843 | 4.16 | 50 | 0.7566 | 0.6418 |
0.7421 | 4.56 | 55 | 0.7840 | 0.6517 |
0.7949 | 4.96 | 60 | 0.7531 | 0.6915 |
0.828 | 5.4 | 65 | 0.7464 | 0.6816 |
0.7438 | 5.8 | 70 | 0.7487 | 0.6915 |