Open Access
Table 2.
Model performance of Transformer models on different datasets. For Bert-Base/Large and Roberta-Base, Matthews correlation is reported for CoLA and accuracy is reported for other datasets. Perplexity is reported for GPT2-Base/Medium/Large
| Model | Bert-Base | Roberta-Base | Bert-Large | ||||||
|---|---|---|---|---|---|---|---|---|---|
| Dataset | CoLA | RTE | QNLI | CoLA | RTE | QNLI | CoLA | RTE | QNLI |
| Plaintext | 0.616 | 0.700 | 0.916 | 0.629 | 0.805 | 0.920 | 0.686 | 0.755 | 0.922 |
| PUMA | 0.613 | 0.700 | 0.916 | 0.618 | 0.805 | 0.918 | 0.690 | 0.747 | 0.918 |
| Model | GPT2-Base | GPT2-Medium | GPT2-Large | ||||||
| Plaintext | 16.284 | 12.536 | 10.142 | ||||||
| PUMA | 16.284 | 12.540 | 10.161 | ||||||
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.