---
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.