Arabic Wikipedia Articles Preprocessed for Language Model Training

4.1 GB of cleaned Arabic text from a May 2026 Wikipedia dump, processed by raj2708. The dataset is a preprocessed version of articles from ar.wikipedia.org, filtered for length and converted from wikitext to plain text. It is intended for use in large language model pretraining.

Use Cases

• Pretrain large language models on Arabic text based on the cleaned Wikipedia corpus.
• Fine-tune models for Arabic text generation tasks using the processed article text.
• Benchmark Arabic NLP model performance on a large-scale, general-domain text dataset.
• Analyze linguistic patterns and knowledge representation in Arabic Wikipedia articles.

Strengths

• The dataset is explicitly preprocessed and cleaned, ready for immediate use in model training.
• It contains 4.1 GB of text, indicating a substantial volume of Arabic language data.
• The source is a well-known, publicly editable knowledge base (Wikipedia), providing broad topical coverage.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for specific training regimes.
• Data may reflect the geographic, cultural, and editorial biases inherent to Wikipedia's contributor base.

Provenance

Source

ar.wikipedia.org

Collection Method

Downloaded full Arabic Wikipedia XML BZ2 dump, parsed articles using mwparserfromhell, filtered redirects and short articles.

Time Range

Snapshot from May 2026.

Freshness

Last updated 2026-05-29 12:25:45.