Extractive Question Answering on HotpotQA MRQA
80.8F1 ScoreLinkBERT
Evaluation results
| Method | Links | |
|---|---|---|
| LinkBERTModel Scale=large2022.03 | 80.8 | |
| SplinterNumber of training examples=Full Dataset2021.01 | 80.7 | |
| SpanBERT (Reimpl)Number of training examples=Full Dataset2021.01 | 79.7 | |
| SpanBERTNumber of training examples=Full Dataset2021.01 | 79.6 | |
| RoBERTaNumber of training examples=Full Dataset2021.01 | 78.7 | |
| LinkBERTModel Scale=base2022.03 | 78.2 | |
| BERTModel Scale=large2022.03 | 78.1 | |
| BERTModel Scale=base2022.03 | 76 | |
| SplinterNumber of training examples=10242021.01 | 70.3 | |
| SpanBERTNumber of training examples=10242021.01 | 67.4 | |
| SpanBERT (Reimpl)Number of training examples=10242021.01 | 64.6 | |
| RoBERTaNumber of training examples=10242021.01 | 61.8 | |
| SplinterNumber of training examples=1282021.01 | 54.7 | |
| LinkBERTModel Scale=tiny2022.03 | 54.6 | |
| BERTModel Scale=tiny2022.03 | 49.8 | |
| SpanBERT (Reimpl)Number of training examples=1282021.01 | 36.6 | |
| SpanBERTNumber of training examples=1282021.01 | 35.1 | |
| RoBERTaNumber of training examples=1282021.01 | 27.3 | |
| SplinterNumber of training examples=162021.01 | 24 | |
| SpanBERTNumber of training examples=162021.01 | 12.6 | |
| SpanBERT (Reimpl)Number of training examples=162021.01 | 12.5 | |
| RoBERTaNumber of training examples=162021.01 | 10.5 |