Question Generation on SQuAD QG
24.59BLEU-4UNIMO
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| UNIMOSize=large2020.12 | 24.59 | — | — | — | 26.39 | 52.47 | |
| RoBERTaSize=large, Implementation=UNIMO architecture with RoBERTa weights2020.12 | 23.39 | — | — | — | 25.73 | 52.11 | |
| UNIMOSize=base2020.12 | 22.78 | — | — | — | 25.24 | 51.34 | |
| RoBERTaSize=base, Implementation=UNIMO architecture with RoBERTa weights2020.12 | 22.15 | — | — | — | 24.58 | 51.12 | |
| UniLMSize=large2020.12 | 22.12 | — | — | — | 25.06 | 51.07 | |
| UNIMO (w/o single-modal)Size=base, Note=Removing single-modal learning process2020.12 | 17.09 | — | — | — | 21.04 | 46.47 | |
| PEGASUS-2B (calibrated)#params=2B, beam size=10, training=SLiC2022.09 | — | — | — | 53.28 | — | — | |
| PEGASUS-2B (fine-tuned)#params=2B, beam size=52022.09 | — | — | — | 52.59 | — | — | |
| UniLMv2#params=110M2022.09 | — | — | — | 52.13 | — | — |