Extractive Question Answering on SQuAD MRQA 2019
92.2F1 ScoreSplinter
Evaluation results
| Method | Links | |
|---|---|---|
| SplinterNumber of training examples=Full Dataset, Capacity=BERT-base (110M parameters)2021.01 | 92.2 | |
| SpanBERTNumber of training examples=Full Dataset, Capacity=BERT-base (110M parameters)2021.01 | 92 | |
| SpanBERT (Reimpl)Number of training examples=Full Dataset, Capacity=BERT-base (110M parameters)2021.01 | 92 | |
| RoBERTaNumber of training examples=Full Dataset, Capacity=BERT-base (110M parameters)2021.01 | 90.3 | |
| SplinterNumber of training examples=1024, Capacity=BERT-base (110M parameters)2021.01 | 82.8 | |
| SpanBERTNumber of training examples=1024, Capacity=BERT-base (110M parameters)2021.01 | 77.8 | |
| SpanBERT (Reimpl)Number of training examples=1024, Capacity=BERT-base (110M parameters)2021.01 | 77.8 | |
| RoBERTaNumber of training examples=1024, Capacity=BERT-base (110M parameters)2021.01 | 73.8 | |
| SplinterNumber of training examples=128, Capacity=BERT-base (110M parameters)2021.01 | 72.7 | |
| SpanBERT (Reimpl)Number of training examples=128, Capacity=BERT-base (110M parameters)2021.01 | 55.8 | |
| SplinterNumber of training examples=16, Capacity=BERT-base (110M parameters)2021.01 | 54.6 | |
| SpanBERTNumber of training examples=128, Capacity=BERT-base (110M parameters)2021.01 | 48.5 | |
| RoBERTaNumber of training examples=128, Capacity=BERT-base (110M parameters)2021.01 | 43 | |
| SpanBERT (Reimpl)Number of training examples=16, Capacity=BERT-base (110M parameters)2021.01 | 18.2 | |
| SpanBERTNumber of training examples=16, Capacity=BERT-base (110M parameters)2021.01 | 12.5 | |
| RoBERTaNumber of training examples=16, Capacity=BERT-base (110M parameters)2021.01 | 7.7 |