| --- |
| license: cc-by-2.0 |
| task_categories: |
| - text-generation |
| language: |
| - en |
| - da |
| - es |
| - eo |
| - bs |
| - de |
| - cs |
| - fr |
| - nl |
| - pl |
| - sv |
| - tr |
| tags: |
| - wikipedia |
| - '2003' |
| - MLL |
| - multilang |
| pretty_name: Wikipedia 2003 |
| size_categories: |
| - 100K<n<1M |
| configs: |
| - config_name: all |
| data_files: '*/*.jsonl' |
| - config_name: en |
| data_files: en/*.jsonl |
| - config_name: es |
| data_files: es/*.jsonl |
| - config_name: de |
| data_files: de/*.jsonl |
| - config_name: hr |
| data_files: hr/*.jsonl |
| - config_name: cs |
| data_files: cs/*.jsonl |
| - config_name: bs |
| data_files: bs/*.jsonl |
| - config_name: da |
| data_files: da/*.jsonl |
| - config_name: eo |
| data_files: eo/*.jsonl |
| - config_name: fr |
| data_files: fr/*.jsonl |
| - config_name: nl |
| data_files: nl/*.jsonl |
| - config_name: pl |
| data_files: pl/*.jsonl |
| - config_name: sh |
| data_files: sh/*.jsonl |
| - config_name: tr |
| data_files: tr/*.jsonl |
| - config_name: sv |
| data_files: sv/*.jsonl |
| --- |
| |
| # Wikipedia-2003 |
|
|
| Original dump: https://dumps.wikimedia.org/archive/2003/2003-05-16 |
|
|
| This is a filtered and cleaned version of the 2003 Wikipedia dump. |
|
|
| ## Stats |
|
|
| | Language | Size | Lines | |
| |:---------|-----:|------:| |
| | Bosnian (`bs`) | 77.6KB | 78 | |
| | Czech (`cs`) | 392.8KB | 354 | |
| | Danish (`da`) | 4.9MB | 11,561 | |
| | German (`de`) | 23.47MB | 18,490 | |
| | English (`en`) | 249MB | 128,198 | |
| | Esperanto (`eo`) | 7.9MB | 7,202 | |
| | Spanish (`es`) | 7.33MB | 4,651 | |
| | French (`fr`) | 13.2MB | 10,957 | |
| | Croatian (`hr`) | 1.2KB | 3 | |
| | Dutch (`nl`) | 10.9MB | 7,116 | |
| | Polish (`pl`) | 14.8MB | 12,583 | |
| | Serbo-Croatian (`sh`) | 904B| 2 | |
| | Swedish (`sv`) | 7.9MB | 9,791 | |
| | Turkish (`tr`) | 27KB | 9 | |
| | **Total** | 339.9MB | 231,193 | |
|
|
| Excluded languages: Chinese, Russian, Japanese, Malay, and Serbian. |
|
|
| ## Filtering |
| We parsed the raw XML dumps, extracted plain text, filtered pages with fewer than 50 characters, cleaned unwanted sections (e.g., "references"), normalized accented letters, and saved each record as a JSONL line. |
|
|
| ## Download |
|
|
| ```python |
| from datasets import load_dataset |
| |
| ds = load_dataset( |
| "fromziro/wikipedia_2003", |
| config="all", # or choose a specifc lang: "en", "es", "nl", "pl", etc. |
| split="train" |
| ) |
| |
| print(ds[0]) |
| ``` |
|
|
| ## License |
|
|
| CC By 2.0. |
