metadata

dataset_info:
  features:
    - name: evidence
      sequence: string
    - name: label
      dtype: int64
    - name: key
      dtype: int64
    - name: claim
      dtype: string
  splits:
    - name: dev
      num_bytes: 17117220
      num_examples: 19998
    - name: train
      num_bytes: 129483687
      num_examples: 145405
    - name: test
      num_bytes: 16951526
      num_examples: 19998
  download_size: 89358795
  dataset_size: 163552433
task_categories:
  - text-classification
language:
  - en
pretty_name: FEVER
size_categories:
  - 100K<n<1M

Dataset Card for "fever"

FEVER dataset with evidence retrieved by KGAT, see my paper for more info. Labels are converted to integers with the rule:

"SUPPORTS": 0
"NOT ENOUGH INFO": 1
"REFUTES": 2