metadata

annotations_creators:
  - expert-generated
language_creators:
  - expert-generated
language:
  - en
multilinguality:
  - monolingual
size_categories:
  - n<1K
source_datasets:
  - original
task_categories:
  - multiple-choice
  - text-generation
  - question-answering
task_ids:
  - multiple-choice-qa
  - language-modeling
  - open-domain-qa
pretty_name: BBEH Disambiguation QA
dataset_info:
  - config_name: generation
    features:
      - name: question
        dtype: string
      - name: best_answer
        dtype: string
    splits:
      - name: validation
        num_examples: 120
  - config_name: multiple_choice
    features:
      - name: question
        dtype: string
      - name: mc1_targets
        struct:
          - name: choices
            sequence: string
          - name: labels
            sequence: int32
    splits:
      - name: validation
        num_examples: 120
configs:
  - config_name: generation
    data_files:
      - split: validation
        path: generation/bbeh-disambiguation-qa-generation.parquet
  - config_name: multiple-choice
    data_files:
      - split: validation
        path: multiple-choice/bbeh-disambiguation-qa-multiple-choice.parquet

Reference

@article{kazemi2025big,
  title={Big-bench extra hard},
  author={Kazemi, Mehran and Fatemi, Bahare and Bansal, Hritik and Palowitch, John and Anastasiou, Chrysovalantis and Mehta, Sanket Vaibhav and Jain, Lalit K and Aglietti, Virginia and Jindal, Disha and Chen, Peter and others},
  journal={arXiv preprint arXiv:2502.19187},
  year={2025}
}