---
license: cc-by-3.0
configs:
- config_name: exzellent_de
  data_files: wiki_de_exzellent.parquet
- config_name: featured_en
  data_files: wiki_en_featured.parquet
- config_name: exzellent_de_small
  data_files: wiki_de_exzellent_small.parquet
- config_name: featured_en_small
  data_files: wiki_en_featured_small.parquet
language:
- de
- en
size_categories:
- 1K<n<10K
---

# German+English Wikitext

Wikitext_en_de is a replication of the `wikitext` dataset following the work by [Merity et al. (2016)](https://arxiv.org/abs/1609.07843). 
It contains (mostly) all articles that Wikipedia classifies as ["exzellent"](https://de.wikipedia.org/wiki/Wikipedia:Exzellente_Artikel) or ["featured"](https://en.wikipedia.org/wiki/Wikipedia:Featured_articles) and can be used for example for perplexity evaluation.

This dataset was created by first scraping the names of the articles belonging to these categories from Wikipedia. Afterwards, we take a recent dump from 
wikipedia ("20230901.de" from [`graelo/wikipedia`](https://huggingface.co/datasets/graelo/wikipedia)) and filter the articles to only include those on either list.

| Config Name | Num Documents |
|-------------|--------------|
| exzellent_de | 2822 |
| featured_en | 6356 |
| exzellent_de_small | 1024 |
| featured_en_small | 1024 |

The code for creating the datasets is available in this repository ("wikitext_de.py", "wikitext_en.py"). 
Be aware that this download a whole wikipedia dump, which might take a while depending on your connection.