Data-to-Text Generation on WebNLG
48.7ROUGE-2BEST-ON-VAL
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| BEST-ON-VALBackbone=Llama-22024.12 | 48.7 | — | — | — | — | — | — | — | — | — | — | — | |
| SMOOTHIE-GLOBALBackbone=Llama-22024.12 | 48.5 | — | — | — | — | — | — | — | — | — | — | — | |
| SMOOTHIE-LOCALBackbone=Llama-22024.12 | 48 | — | — | — | — | — | — | — | — | — | — | — | |
| RANDOMBackbone=Llama-22024.12 | 45.5 | — | — | — | — | — | — | — | — | — | — | — | |
| BEST-ON-VALBackbone=Falcon2024.12 | 32.5 | — | — | — | — | — | — | — | — | — | — | — | |
| SMOOTHIE-LOCALBackbone=Falcon2024.12 | 30.8 | — | — | — | — | — | — | — | — | — | — | — | |
| RANDOMBackbone=Falcon2024.12 | 29.2 | — | — | — | — | — | — | — | — | — | — | — | |
| SMOOTHIE-GLOBALBackbone=Falcon2024.12 | 28.3 | — | — | — | — | — | — | — | — | — | — | — | |
| Castro Ferreira et al. (2019)2020.04 | — | 4.54 | 44 | 50.5 | — | 33.7 | — | — | — | — | — | — | |
| CtMBase Model=LLaMA2-7B2026.06 | — | — | — | — | — | — | — | — | — | — | — | 78.98 | |
| DATATUNER_FCSFC=true, Fine-grained state embeddings=true2020.04 | — | 5.23 | 84 | 91.7 | — | 58.1 | — | — | — | — | — | — | |
| DATATUNER_NO_FCSFC=false, Fine-grained state embeddings=true2020.04 | — | 5.2 | 80 | 81.4 | — | 54.1 | — | — | — | — | — | — | |
| DATATUNER_NO_FC/FSSFC=false, Fine-grained state embeddings=false2020.04 | — | 5.27 | 72 | 73.6 | 87.5 | 43.6 | 73.3 | — | — | — | — | — | |
| Fully-supervised fine-tuningSupervision Level=Fully-supervised, Training Method=fine-tuning2023.05 | — | — | — | — | — | — | — | 0 | 6.72 | 10.29 | 1.73 | — | |
| Human2020.04 | — | 5.21 | — | 94.7 | — | 41.2 | — | — | — | — | — | — | |
| Iso-CTSBase Model=LLaMA2-7B2026.06 | — | — | — | — | — | — | — | — | — | — | — | 73.34 | |
| LoRA-LEGOBase Model=LLaMA2-7B2026.06 | — | — | — | — | — | — | — | — | — | — | — | 70.24 | |
| LoRA's absolute performanceBase Model=LLaMA2-7B2026.06 | — | — | — | — | — | — | — | — | — | — | — | 70.9 | |
| Low-resource cycle-trainingSupervision Level=Low-resource, Training Method=cycle-training2023.05 | — | — | — | — | — | — | — | 0 | 1.47 | 1.82 | 1.89 | — | |
| Low-resource fine-tuningSupervision Level=Low-resource, Training Method=fine-tuning2023.05 | — | — | — | — | — | — | — | 6.72 | 7.21 | 15.9 | 1.91 | — | |
| RobustMergeBase Model=LLaMA2-7B2026.06 | — | — | — | — | — | — | — | — | — | — | — | 56.56 | |
| TIESBase Model=LLaMA2-7B, Space=Full2026.06 | — | — | — | — | — | — | — | — | — | — | — | 79.83 | |
| TIESBase Model=LLaMA2-7B, Space=KnOTS2026.06 | — | — | — | — | — | — | — | — | — | — | — | 73.76 | |
| TIESBase Model=LLaMA2-7B, Space=CoreSpace2026.06 | — | — | — | — | — | — | — | — | — | — | — | 79.13 |