This dataset has been carefully curated to enhance the general instruction capabilities of Vision-Language Models (VLMs). It comprises two subsets:
600k English samples
600k Korean samples
We recommend using this dataset alongside other task-specific datasets (e.g., OCR, Language, code, math, ...) to improve performance and achieve more robust model capabilities.
