The Sindhi Language Mega Corpus (118M Tokens) is the largest open-source Sindhi dataset ever released for Natural Language Processing (NLP) and AI model development.
It compiles diverse and high-quality Sindhi text from seven major sources, ranging from classical literature and journalism to education, history, and religion — fully cleaned, normalized, and structured into JSON format.

Each file represents a unique domain of Sindhi knowledge, with metadata (source, category, token count) included for flexibility and reproducibility.
A custom-trained Sindhi tokenizer model is also included, enabling consistent tokenization of Sindhi text for modern transformer-based LLMs.

📂 Dataset Structure

After extracting the ZIP file, you will find this structure:

Description of Folders and Files

Folder/File	Description
`data/`	Contains all Sindhi text data as structured JSON files. Each file corresponds to a specific source or category.
`tokenizer/`	Contains the trained Sindhi tokenizer model, configuration, and vocabulary files.
`README.md`	This documentation explaining dataset details, structure, preprocessing, and usage.

📚 Dataset Categories & Descriptions

Source	Description
Sindh Salamat	Literary and cultural text, including poetry, Sufism, education, and philosophy from the Sindh Salamat website.
Sindhi Wikipedia	Encyclopedic and factual content covering science, technology, history, and world knowledge.
Altaf Shaikh Literature	Works of Altaf Shaikh focusing on Sindhi literature, travelogues, philosophy, and political essays.
Sindhi Language Authority	Educational and linguistic resources promoting Sindhi language learning and preservation.
Hamsari Akhbar	Columns, articles, and opinion pieces from the Hamsari newspaper.
Pahenji Akhbar	Sindhi newspaper data including daily news and editorials.
General Sindhi Data	Miscellaneous writings, essays, short stories, and social commentary from various sources.

🧩 JSON File Format

Each JSON file contains an array of structured Sindhi text entries:

[
  {
    "id": "salamat_00123",
    "source": "Sindh Salamat",
    "category": "Literature",
    "title": "ادب ۽ فڪر جو تجزيو",
    "text": "سنڌي ادب ۾ فڪر ۽ تخليق جو سفر هڪ ڊگهو ۽ گهرو عمل رهيو آهي...",
    "tokens": 243
  },
  {
    "id": "salamat_00124",
    "source": "Sindh Salamat",
    "category": "Poetry",
    "title": "شاعريءَ ۾ احساس جو اظهار",
    "text": "شاعريءَ ۾ احساس جو اظهار روحاني ۽ جمالياتي ٻنهي سطحن تي ٿيندو آهي...",
    "tokens": 132
  }
]

from transformers import AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("tokenizer/")
text = "سنڌي ٻوليءَ جو ادبي سرمايو انتهائي وسيع آهي."
tokens = tokenizer.tokenize(text)
print(tokens)

Downloads last month: 18