| dataset_info: | |
| features: | |
| - name: sentence1 | |
| dtype: string | |
| - name: sentence2 | |
| dtype: string | |
| splits: | |
| - name: train | |
| num_bytes: 14498836027 | |
| num_examples: 6458670 | |
| download_size: 8956015300 | |
| dataset_size: 14498836027 | |
| # Dataset Card for "wiki-span" | |
| This dataset is constructed by sampling 25%-50% of each wikipedia record twice, as positive pairs. It can be used to train unsupervised sentence representation models. |