Natural Language Generation on WebNLG seen categories
65.3BLEUFine-Tune
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Fine-TuneBackbone=GPT-2 Large, Trainable Parameters=774M2021.06 | 65.3 | 0.46 | 0.33 | |
| Adapter_HBackbone=GPT-2 Large, Trainable Parameters=23M2021.06 | 64.7 | 0.46 | 0.33 | |
| Fine-TuneBackbone=GPT-2 Medium, Trainable Parameters=354M2021.06 | 64.2 | 0.45 | 0.33 | |
| LoRABackbone=GPT-2 Large, Trainable Parameters=0.77M2021.06 | 64 | 0.45 | 0.32 | |
| PrefixBackbone=GPT-2 Large, Trainable Parameters=0.77M2021.06 | 63.4 | 0.45 | 0.34 | |
| PrefixBackbone=GPT-2 Medium, Trainable Parameters=0.35M2021.06 | 62.9 | 0.44 | 0.35 | |
| LoRABackbone=GPT-2 Medium, Trainable Parameters=0.35M2021.06 | 62.1 | 0.44 | 0.33 | |
| Adapter_LBackbone=GPT-2 Large, Trainable Parameters=0.88M2021.06 | 61.1 | 0.43 | 0.35 | |
| Adapter_HBackbone=GPT-2 Medium, Trainable Parameters=11M2021.06 | 60.4 | 0.43 | 0.35 | |
| Adapter_LBackbone=GPT-2 Medium, Trainable Parameters=0.37M2021.06 | 54.5 | 0.39 | 0.4 | |
| FT_Top2Backbone=GPT-2 Medium, Trainable Parameters=24M2021.06 | 53.6 | 0.38 | 0.49 |