fancyfeast/

joy-captioning-20250408a

This is the dataset used to do the initial training for JoyCaption Beta One (https://huggingface.co/fancyfeast/llama-joycaption-beta-one-hf-llava), before post-training.

Most of the dataset focusses on descriptions and captions for images, with a smaller subset covering general VQA tasks. Some of the questions and answers are human written, some are automated, some are machine written. The is_human column is True when the answer text is human written.