Commonsense Reasoning on WSC
80.6AccuracyTraditional MTL
Evaluation Results
| Method | Links | |
|---|---|---|
| Traditional MTL2024.05 | 80.6 | |
| Individual2024.05 | 65.3 | |
| EMR-MERGINGValidation=false2024.05 | 65.3 | |
| Ties-MergingValidation=true2024.05 | 59.7 | |
| Fisher MergingValidation=true2024.05 | 58.3 | |
| Ties-MergingValidation=false2024.05 | 58.3 | |
| RegMeanValidation=true2024.05 | 53.5 | |
| Weight AveragingValidation=false2024.05 | 53.5 | |
| Task ArithmeticValidation=false2024.05 | 51.4 | |
| Task ArithmeticValidation=true2024.05 | 49.3 | |
| GPT-NeoXParameters=20B, Evaluation=Five-shot2022.04 | 38.5 | |
| GPT-JParameters=6B, Evaluation=Five-shot2022.04 | 36.5 |