aria-haman/haman-fa-wikipedia-articles-186k

Haman Persian Wikipedia Articles 186K

A prepared Persian article dataset used to train Haman Persian Article Graph-LLM 125M.

Dataset Summary

The dataset is derived from the Persian Wikipedia pages-articles dump and was cleaned and prepared using the native article workflow in the Rakhshai Graph-based NLP project.

• —Language: Persian
• —Accepted articles: 185,906
• —Training records: 176,611
• —Validation records: 9,295
• —Validation ratio: 5%
• —Split seed: 42
• —Training format: wikipedia_prompt
• —Minimum accepted body length: 700 characters
• —Source snapshot downloaded: July 5, 2026
• —Preparation file SHA-256: 7193039ba987906d9d64edc2f25203904e6616774d131180da56fcbf51d09125

Fields

• —title: Persian Wikipedia article title.
• —text: Exact structured text supplied to the language-model training pipeline.
• —source: Source collection name.
• —source_line: Original line number in the cleaned JSONL source.
• —source_url: Persian Wikipedia article URL constructed from the title.

Intended Uses

• —Persian causal language-model training.
• —Persian article-generation research.
• —Reproducibility of Haman Persian Article Graph-LLM 125M.
• —Persian tokenization and graph-language-model experiments.

Limitations

The dataset may contain factual errors, biases, uneven topic coverage, outdated information, and formatting artifacts inherited from Wikipedia or introduced during extraction. The source_url field is derived from the article title and may not resolve correctly for renamed or redirected pages.

License and Attribution

The source text is derived from Persian Wikipedia and is distributed under Creative Commons Attribution-ShareAlike 4.0. Downstream users must preserve attribution, indicate modifications, and comply with the ShareAlike terms.

Source: https://fa.wikipedia.org/ License: https://creativecommons.org/licenses/by-sa/4.0/