Natural Language Generation on WebNLG all categories
57.7BLEUAdapter_H
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Adapter_HBackbone=GPT-2 Large, Trainable Parameters=23M2021.06 | 57.7 | 0.43 | 0.39 | |
| LoRABackbone=GPT-2 Large, Trainable Parameters=0.77M2021.06 | 57 | 0.42 | 0.38 | |
| PrefixBackbone=GPT-2 Large, Trainable Parameters=0.77M2021.06 | 56.3 | 0.42 | 0.4 | |
| Adapter_LBackbone=GPT-2 Large, Trainable Parameters=0.88M2021.06 | 56 | 0.41 | 0.39 | |
| Fine-TuneBackbone=GPT-2 Large, Trainable Parameters=774M2021.06 | 55.5 | 0.42 | 0.42 | |
| LoRABackbone=GPT-2 Medium, Trainable Parameters=0.35M2021.06 | 55.3 | 0.41 | 0.39 | |
| PrefixBackbone=GPT-2 Medium, Trainable Parameters=0.35M2021.06 | 55.1 | 0.41 | 0.4 | |
| Adapter_HBackbone=GPT-2 Medium, Trainable Parameters=11M2021.06 | 54.9 | 0.41 | 0.39 | |
| Adapter_LBackbone=GPT-2 Medium, Trainable Parameters=0.37M2021.06 | 50.2 | 0.38 | 0.43 | |
| Fine-TuneBackbone=GPT-2 Medium, Trainable Parameters=354M2021.06 | 46.5 | 0.38 | 0.53 | |
| FT_Top2Backbone=GPT-2 Medium, Trainable Parameters=24M2021.06 | 36 | 0.31 | 0.72 |