metadata

annotations_creators:
  - no-annotation
language_creators:
  - crowdsourced
pretty_name: wikipedia-de-splits
paperswithcode_id: null
license:
  - cc-by-sa-3.0
  - gfdl
task_categories:
  - text-generation
  - fill-mask
task_ids:
  - language-modeling
  - masked-language-modeling
source_datasets:
  - wikipedia
size_categories:
  - n<1K
  - 1K<n<10K
  - 10K<n<100K
  - 100K<n<1M
  - 1M<n<10M
language:
  - de
configs:
  - '1'
  - '2'
  - '3'
  - '4'
  - '5'
  - '6'
  - '7'
  - '8'
  - '9'
  - '10'
  - '11'
  - '12'
  - '13'
  - '14'
  - '15'
  - '16'
  - '17'
  - '18'
  - '19'
  - '20'
  - '21'
  - all

Dataset Card for yaakov/wikipedia-de-splits

Dataset Description

The only goal of this dataset is to have random German Wikipedia articles at various dataset sizes: Small datasets for fast development and large datasets for statistically relevant measurements.

For this purpose, I loaded the 2665357 articles in the test set of the pre-processed German Wikipedia dump from 2022-03-01, randomly permuted the articles and created splits of sizes 2**n: 1, 2, 4, 8, .... The split names are strings. The split 'all' contains all 2665357 available articles.

Dataset creation

This dataset has been created with the following script:

!apt install git-lfs
!pip install -q transformers datasets

from huggingface_hub import notebook_login
notebook_login()

from datasets import load_dataset
wikipedia_de = load_dataset("wikipedia", "20220301.de")['train']

shuffled = wikipedia_de.shuffle(seed=42)

from datasets import DatasetDict
res = DatasetDict()

k, n = 0, 1
while n <= shuffled.num_rows:
    res[str(k)] = shuffled.select(range(n))
    k += 1; n *= 2
res['all'] = shuffled

res.push_to_hub('yaakov/wikipedia-de-splits')