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. |