Appearance
Efficient language and vision models under 10B parameters, optimized for edge deployment and research.