Natural Language Processing on CrossFit (test)
1GLUE CoLA ScoreMulti-task Learning (MTL)
Evaluation results
| Method | Links | |||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Multi-task Learning (MTL)Learning Strategy=Multi-task learning on NLP Few-shot Gym, Backbone Model=BART-Base2022.08 | 1 | 25.6 | 82.6 | 25.4 | 80 | 92.1 | 98.5 | 56.4 | 77.7 | 56.2 | 44.8 | 69.9 | 45.5 | 32.4 | 74.2 | 41.8 | 97.1 | 4.8 | 93.5 | 1.2 | 49.1 | |
| Direct Finetuning (DF)Learning Strategy=Direct Finetuning, Backbone Model=BART-Base2022.08 | 0 | 30.6 | 86.1 | 16.1 | 79.6 | 79.4 | 99.4 | 55 | 75.8 | 49.5 | 46.3 | 57.7 | 42 | 16.5 | 60.5 | 24.7 | 70.9 | 2.3 | 40.6 | 0.5 | 46.7 | |
| UnpredicTable-5k (Ours)Learning Strategy=Multi-task learning with UnpredicTable-5k, Backbone Model=BART-Base2022.08 | 0 | 29.5 | 84.9 | 15.7 | 78.4 | 90.8 | 97.8 | 52.5 | 71.3 | 49.9 | 48.3 | 60.9 | 41 | 14.2 | 56.9 | 23.3 | 82.6 | 1.7 | 56.2 | 0.4 | 47.8 |