Scraped the already processed data from https://huggingface.co/spaces/huggingface-projects/diffusers-gallery-bot/api/models endpoint. Tho I was not involved in the project it looks from their code like they take the example images from the model cards for each huggingface diffusers model, run them thru an image classifier model, and that is the basis for the task specific scores. If you want up to the minute accuracy, use the original source - this data set will be optimized for RAG workflows