README.md

metadata

dataset_info:
  features:
    - name: subset
      dtype: string
    - name: context
      dtype: string
    - name: question
      dtype: string
    - name: answers
      struct:
        - name: answer_start
          sequence: int64
        - name: text
          sequence: string
    - name: id
      dtype: string
  splits:
    - name: train
      num_bytes: 35676134.53646248
      num_examples: 15000
    - name: validation
      num_bytes: 3746889.5501623126
      num_examples: 1500
    - name: test
      num_bytes: 900881.1896605418
      num_examples: 500
  download_size: 24615221
  dataset_size: 40323905.276285335
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*

Summary

This is a random sample (shuffle with seed 27) of MRQA dataset, with 15k train, 1k validation, and 500 test examples. The dataset fields have been converted to SQuAD format.