High Multimodal AI · 1 min read
IDEFICS: the first open-source replica of Flamingo
In one sentence HuggingFace releases IDEFICS, an open-weight replica of Flamingo in 9B and 80B versions, trained on LAION-5B and WikiMedia with few-shot visual in-context learning.
Reading level
Before 2023, models capable of reasoning over images and text were all closed and accessible only through paid APIs. HuggingFace changed the rules by releasing IDEFICS, the first large open-weight vision-language model. IDEFICS mimics DeepMind's Flamingo but with public data: you can show it examples of image questions and it understands the pattern without retraining. Anyone can download it, modify it, and build their own applications on top of it.
Companies
HuggingFace
Tools
IDEFICS, IDEFICS-9B, IDEFICS-80B
Tags
VLMOpen SourceFew-Shot LearningVision-Language
Sources