In a significant advancement for mobile user interface technology, a new model named Ferret-UI Lite has been introduced, achieving performance comparable to much larger models with up to 24 billion parameters. This development was detailed in a study published by a group of nine researchers in December 2023, titled “FERRET: Refer and Ground Anything Anywhere at Any Granularity.”
Building on prior models like Ferret-UI and its variants, the team aimed to enhance the understanding of mobile UIs, addressing limitations found in general-domain multimodal large language models (MLLMs). Ferret-UI Lite specifically targets on-device functionality, making it a more efficient option for mobile applications.
Earlier iterations of the Ferret models, such as Ferretv2 and Ferret-UI 2, focused on higher resolution and multi-platform support. However, Ferret-UI Lite combines a reduced model size with advanced capabilities, allowing it to remain competitive against larger GUI agents while being optimized for performance on mobile devices.