Natural Language Generation on WebNLG (unseen categories)
49.8BLEUAdapter_L
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Adapter_LBackbone=GPT-2 Large, Trainable Parameters=0.88M2021.06 | 49.8 | 0.38 | 0.44 | |
| Adapter_HBackbone=GPT-2 Large, Trainable Parameters=23M2021.06 | 49.2 | 0.39 | 0.46 | |
| LoRABackbone=GPT-2 Large, Trainable Parameters=0.77M2021.06 | 48.4 | 0.39 | 0.45 | |
| Adapter_HBackbone=GPT-2 Medium, Trainable Parameters=11M2021.06 | 48.3 | 0.38 | 0.45 | |
| PrefixBackbone=GPT-2 Large, Trainable Parameters=0.77M2021.06 | 47.7 | 0.39 | 0.48 | |
| LoRABackbone=GPT-2 Medium, Trainable Parameters=0.35M2021.06 | 46.7 | 0.38 | 0.46 | |
| HiddenKeyBackbone=GPT2-Medium, Strategy=LoRA2024.02 | 46.48 | 0.381 | 0.452 | |
| DropKeyBackbone=GPT2-Medium, Strategy=LoRA2024.02 | 46.39 | 0.378 | 0.455 | |
| HiddenKey-Backbone=GPT2-Medium, Strategy=LoRA2024.02 | 46.36 | 0.378 | 0.454 | |
| HiddenCutBackbone=GPT2-Medium, Strategy=LoRA2024.02 | 46.11 | 0.377 | 0.458 | |
| PrefixBackbone=GPT-2 Medium, Trainable Parameters=0.35M2021.06 | 45.6 | 0.38 | 0.49 | |
| BaselineBackbone=GPT2-Medium, Strategy=LoRA2024.02 | 45.53 | 0.376 | 0.464 | |
| Adapter_LBackbone=GPT-2 Medium, Trainable Parameters=0.37M2021.06 | 45.1 | 0.36 | 0.46 | |
| Fine-TuneBackbone=GPT-2 Large, Trainable Parameters=774M2021.06 | 43.1 | 0.38 | 0.53 | |
| Fine-TuneBackbone=GPT-2 Medium, Trainable Parameters=354M2021.06 | 27.7 | 0.3 | 0.76 | |
| Full FinetuningBackbone=GPT2-Medium, Strategy=Full Finetuning2024.02 | 27.7 | 0.3 | 0.76 | |
| FT_Top2Backbone=GPT-2 Medium, Trainable Parameters=24M2021.06 | 18.9 | 0.23 | 0.99 |