Graph-to-text generation on WebNLG all (test)
59.9BLEUT5-large
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| T5-largeLearning Paradigm=Supervised, Training Setup=Synthetic corpus (934k samples), Approach Type=End-to-end2022.12 | 59.9 | 44 | |
| GPT-2-largeLearning Paradigm=Supervised, Training Setup=Synthetic corpus (934k samples), Approach Type=End-to-end2022.12 | 55.5 | 42 | |
| MELBOURNELearning Paradigm=Supervised, Training Setup=7k in-domain samples, Approach Type=End-to-end2022.12 | 45.1 | 37 | |
| Neural PipelineLearning Paradigm=Supervised, Training Setup=Synthetic corpus (934k samples), Approach Type=Pipeline2022.12 | 43.3 | 39.3 | |
| MURMURLearning Paradigm=Few-shot, Number of demonstrations (k)=12022.12 | 41.3 | 37.1 | |
| Direct PromptingLearning Paradigm=Few-shot, Number of demonstrations (k)=52022.12 | 39.5 | 34.4 | |
| Direct PromptingLearning Paradigm=Few-shot, Number of demonstrations (k)=12022.12 | 33.6 | 30.8 | |
| CoT PromptingLearning Paradigm=Few-shot, Number of demonstrations (k)=12022.12 | 18 | 22.6 |