Natural Language Understanding on VALUE AAE GLUE RoBERTa Base (test)
40.2CoLAVALUE Adapters
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| VALUE AdaptersApproach=VALUE, Task-Agnostic=false, Dialect Params.=T x 895K, Backbone=RoBERTa Base2023.05 | 40.2 | 85.8 | 92.2 | 73.6 | 89.7 | 93.6 | 90.3 | 80.8 | |
| TADA AdaptersApproach=TADA, Task-Agnostic=true, Dialect Params.=895K, Backbone=RoBERTa Base2023.05 | 29.5 | 84.8 | 91.7 | 67.2 | 88.1 | 91.9 | 89.6 | 77.5 | |
| VALUE FinetuningApproach=VALUE, Task-Agnostic=false, Dialect Params.=T x 110M, Backbone=RoBERTa Base2023.05 | 19.8 | 84.9 | 90.8 | 74.4 | 89.6 | 92.4 | 90.9 | 77.5 | |
| AdaptersApproach=N/A, Task-Agnostic=N/A, Dialect Params.=0, Backbone=RoBERTa Base2023.05 | 14.1 | 83.7 | 90.3 | 67.1 | 86.8 | 92.1 | 88.7 | 74.7 | |
| FinetuningApproach=N/A, Task-Agnostic=N/A, Dialect Params.=0, Backbone=RoBERTa Base2023.05 | 13.5 | 82 | 89.3 | 71.8 | 87.1 | 92 | 89.9 | 75.1 |