Dialogue Coherence on BMELD (val)
61.03Precision (1st)Human Reference
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Human Reference2022.05 | 61.03 | 59.24 | 54.19 | |
| Scheduled Multi-task Learning (SML)Model Size=Base, Training Framework=Three-stage2022.05 | 60.48 | 57.88 | 53.95 | |
| M-NCTModel Size=Base, Training Framework=Three-stage2022.05 | 59.57 | 56.79 | 53.18 | |
| CSA-NCTModel Size=Base, Training Framework=Three-stage2022.05 | 59.45 | 56.74 | 53.02 | |
| NCTModel Size=Base, Training Framework=Three-stage2022.05 | 59.19 | 56.43 | 52.89 | |
| TransformerModel Size=Base, Fine-tuning=true, Training Framework=Three-stage2022.05 | 58.77 | 56.1 | 52.71 | |
| Transformer without Fine-tuningModel Size=Base, Fine-tuning=false, Training Framework=Three-stage2022.05 | 58.11 | 55.15 | 52.15 |