Question Answering on SQuAD 2.0 (dev)
86.3EMHuman
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HumanModel Type=Pipeline2018.10 | 86.3 | 89 | |
| Human2018.10 | 86.3 | 89 | |
| UNILMMode=fine-tuned, Epochs=3, Batch size=24, Maximum length=3842019.05 | 80.5 | 83.4 | |
| BERT_LARGEType=cased, Mode=fine-tuned, Epochs=3, Batch size=24, Maximum length=3842019.05 | 78.9 | 81.8 | |
| BERT_LARGEmodel_type=Single2018.10 | 78.7 | 81.9 | |
| RMR+ELMo+VerifierModel Type=Pipeline2018.10 | 72.3 | 74.8 | |
| RMR+ELMoArchitecture=LSTM-based, Augmentation=pre-trained language representation2019.05 | 71.4 | 73.7 | |
| U-NetModel Type=End-to-end2018.10 | 70.3 | 74 | |
| DocQAModel Type=End-to-end2018.10 | 65.1 | 67.6 | |
| BNA*Model Type=End-to-end2018.10 | 59.8 | 62.6 |