Khasi-OCR-27K

This is a Vision-Language dataset for the Khasi language, specifically curated for OCR, document understanding and Handwriting recognition tasks.

This specific version of the dataset has been pre-filtered and formatted strictly for Vision training (e.g., DeepSeek-VL/OCR). It contains only the Free OCR task, with conversations mapped to the strict <|User|> and <|Assistant|> token format. Images are natively embedded.

Dataset Summary

• Total Training Samples: 27,582
• Unique Images: 27,582
• Language: Khasi (kha)

Each sample contains:

• image: Document image.
• image_id: Unique identifier for the image.
• messages: Instruction-tuned format with user and assistant roles.

The dataset is formatted for training Vision-Language Models (VLMs).

Tasks Included

• Free OCR: Plain text extraction from diverse document types.