Relation Extraction on zsRE (val)
99.8EfficacyROME
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ROMEBackbone=GPT-2 XL, Update Type=Rank-one update2022.02 | 99.8 | 88.1 | 24.2 | |
| FTBackbone=GPT-2 XL, Method=Fine-Tuning, Optimizer=Adam, Config=Early stopping at one layer2022.02 | 99.6 | 82.1 | 23.2 | |
| MEND-zsREBackbone=GPT-2 XL, Training=Explicitly trained on zsRE data distribution2022.02 | 99.4 | 99.3 | 24.1 | |
| KE-zsREBackbone=GPT-2 XL, Training=Explicitly trained on zsRE data distribution2022.02 | 92.4 | 90 | 23.8 | |
| FT+LBackbone=GPT-2 XL, Method=Constrained Fine-Tuning, Constraint=parameter-space L-infinity norm2022.02 | 92.3 | 47.2 | 23.4 | |
| MENDBackbone=GPT-2 XL, Training=WikiText generation task2022.02 | 75.8 | 65.3 | 24.1 | |
| KEBackbone=GPT-2 XL, Training=WikiText generation task2022.02 | 65.5 | 61.4 | 24.9 | |
| GPT-2 XLBackbone=GPT-2 XL2022.02 | 22.2 | 21.3 | 24.2 |