Hugging Face's logo

Datasets:
Tensor-Labs
/
pashto-largest-corpus

Modalities:
Text
Formats:
json
Languages:
Pashto
Size:
1M - 10M
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1

You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

Pashto Cleaned Text Corpus (1.5 Billion Words Project)

Dataset Summary

This dataset is part of a monumental project to build a massive, high-quality collection of Pashto text designed to push the boundaries of Pashto Natural Language Processing (NLP) and Large Language Model (LLM) training. Our total corpus spans an incredible 1.5 Billion words, making it one of the largest and most comprehensive open-source Pashto resources ever created.

Currently, we have uploaded v1, which contains 3,645,473 rows of perfectly cleaned text and exactly 321,675,360 words.

Data Sources

To ensure a highly diverse, rich, and robust representation of the Pashto language, this corpus was meticulously scraped, curated, and normalized from a wide variety of domains:

  • News Archives: Over a decade of Pashto journalism, articles, and reporting.
  • Wikipedia: The comprehensive Pashto Wikipedia dump.
  • Digital Literature: Classic and modern Pashto poetry, books, and literary platforms.
  • Islamic Texts: Religious literature, translations, and scholarly articles.
  • Government Documents: Official public documents and available records.
  • Community Blogs & Forums: Everyday conversational and informal Pashto from across the web.

Dataset Structure

  • Format: .jsonl (JSON Lines)
  • Fields: The dataset has been put through a 5-stage cleaning pipeline and stripped of all unnecessary metadata. It contains a single column:
    • text: The raw, perfectly cleaned Pashto text string (which forms the exact training foundation of Qehwa AI).

v1 Dataset Statistics

  • Total Rows / Examples: 3,645,473
  • Total Words: 321,675,360
  • Total Characters: 1,530,481,226

(Note: Future versions will continue to expand until the full 1.5 Billion word corpus is uploaded).

Use Cases

Backed by the ambition of a 1.5 Billion word Pashto project, this corpus is ideal for:

  • Pre-training or fine-tuning Large Language Models (LLMs) to achieve native fluency in Pashto.
  • Training text embeddings, natural language generation, and pushing the limits of Pashto NLP research.

License

This dataset is published under the MIT License. You are completely free to use, modify, and distribute this data for both open-source and commercial AI products.
Downloads last month
83
Total file size:
2.49 GB