Introducing IDEFICS: Open Reproduction of State-of-the-Art Visual Language Model

AI Impact Summary

IDEFICS is an open reproduction of Flamingo, offering a transparent and reproducible multimodal model built on publicly available data. This release includes two variants (80B and 9B parameters) and a new 115B token dataset (OBELICS) for training, providing researchers with a foundational system for multimodal AI research. The inclusion of adversarial prompting and a detailed model card demonstrates a commitment to ethical AI development and transparency.