Datasets:

Modalities:
Text
Formats:
parquet
Languages:
Swedish
Libraries:
Datasets
pandas
License:
swedish-facts-v1 / README.md
jekunz's picture
Update README.md
d3c151e verified
metadata
dataset_info:
  features:
    - name: Question (Swedish)
      dtype: string
    - name: Answer (Swedish)
      dtype: string
    - name: Question (English)
      dtype: string
    - name: Answer (English)
      dtype: string
    - name: Cutoff date
      dtype: string
  splits:
    - name: train
      num_bytes: 200561
      num_examples: 1292
  download_size: 125953
  dataset_size: 200561
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
license: apache-2.0
task_categories:
  - question-answering
language:
  - sv
pretty_name: Sweden-Related Factual Knowledge
size_categories:
  - 1K<n<10K

This is a benchmark for Sweden-related factual knowledge. Its questions are inspired by the hosts of the Swedish radio program Sommar I P1 as well as sports-related events in Sweden (e.g., events that are part of En Svensk Klassiker.

Answers are designed to be minimal to enable simple string-based answer recall to approximate model performance as well as possible. Note that some samples in the dataset have multiple correct answers; if so, they are separated by commas.

A paper with more information is coming soon!