| Implications | |
| Transformers models based on the BERT (Bidirectional Encoder Representations from | |
| Transformers) | |
| architecture, or its variants such as | |
| distilBERT and | |
| roBERTa run best on | |
| Inf1 for non-generative tasks such as extractive question answering, sequence | |
| classification, and token classification. |