Knowledge Base Construction on LM-KBC 2023 (val)
49.3PrecisionVE-BERT (token-recode + re-pretrain)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| VE-BERT (token-recode + re-pretrain)Parameter quantity (billion)=0.132023.10 | 49.3 | 44.3 | 36.2 | |
| baseline bert-base-casedParameter quantity (billion)=0.132023.10 | 44.2 | 38.2 | 31.1 | |
| re pretrainParameter quantity (billion)=0.132023.10 | 44 | 40.5 | 33.1 | |
| token recodeParameter quantity (billion)=0.132023.10 | 42.6 | 40 | 33.4 | |
| prompt - directly bert-large-casedParameter quantity (billion)=0.3452023.10 | 36.8 | 16.1 | 14.2 | |
| prompt – directly GPT3-nerParameter quantity (billion)=1752023.10 | 30.8 | 21 | 21.8 | |
| prompt - directly bert-base-casedParameter quantity (billion)=0.112023.10 | 13.1 | 47.4 | 11.2 | |
| prompt – directly facebook/opt-1.3bParameter quantity (billion)=1.32023.10 | 7.3 | 10.1 | 3.9 |