Hugging Face's logo

Datasets:
gowitheflow
/
wiki-span

Modalities:
Text
Formats:
parquet
Size:
1M - 10M
Libraries:
Datasets
Dask
Dataset card Viewer Files Community
1
wiki-span / README.md
gowitheflow's picture
gowitheflow
Update README.md
c4083a9 verified
preview code
|
raw
history blame contribute delete
No virus
444 Bytes
metadata
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.