metadata
license: cc0-1.0
language:
- en
pretty_name: CowQA
tags:
- benchmark
- evaluation
config:
- config_name: default
data_files:
- split: test
path: dataset.json
CowQA
Measuring AI's factual recall knowledge about cows with three possible answers QA.
P.S. if being exactly, questions are touching anything cows-related.
Leaderboard
Official leaderboard available here for CowQA, and here for CowQA-Rapid
Format example
[
{"question":"Question", "A":"variant A", "B":"variant B", "C":"variant C", "correct":"A"}
]
Data source
Questions written manually, based on wikipedia articles.
Quality & difficulty
All of questions are made to be middle difficulty and pretty direct, avoid obscure trivia. No difficulty separation.
License & terms of use
CC0-1.0, no citation is needed.
File format
Raw JSON, no train/validation/test separation. Dataset is recommended only for evaluation.