rzgar/kurdish_sorani_literatures_it

Vejîn Kurdish Sorani Literature Instructions Dataset

This dataset contains 454,230 instruction:response pairs in Kurdish Sorani, derived from the rzgar/kurdish_sorani_literatures dataset (22,727 entries, mostly poetry). It’s designed for fine-tuning language models, with a mix of factual queries (e.g., author/book info) and poetic tasks (e.g., reciting or summarizing poems). Content is chunked for longer poems, preserving dialectal richness across Sorani variants.

Dataset Details

• Source: rzgar/kurdish_sorani_literatures (Hugging Face)
• Lines: 22,727 original entries → 454,230 pairs
• Format: JSONL (kurdish_poetry_instructions_full.jsonl)
• Fields: instruction (Kurdish Sorani), response (Kurdish Sorani)
• Size: ~125-250MB (estimated)
• License: CC-BY-SA-4.0

Usage

Perfect for training Kurdish Sorani LLMs—factual recall, poetry recitation, or creative generation. Pairs include:

• "کێ کتێبی '{book}' نووسیوە؟" → "{author}"
• "شێعری '{title}' چییە؟" → Poem content or chunk
• "شێعرێک لە '{author}' بڵێ" → Poem snippet

Acknowledgments

Built with love by me and thanks to VejînBooks for making Kurdish literary texts available online.

License

This dataset is licensed under CC-BY-SA-4.0. Share, adapt, and enjoy—just give credit to VejînBooks and keep it open!