Fuyu Models by Adept AI

About

Developed by Adept AI, the Fuyu large language model (LLM) family is distinct for its innovative approach in handling multimodal tasks, specifically catering to digital agents. It boasts a simplified architecture, employing a decoder-only transformer without needing a specific image encoder. Instead, image patches are directly fed as linear projections into the transformer's initial layer, which allows for flexible processing of various image resolutions. This configuration not only streamlines training but also accelerates inference, providing rapid response times under 100 milliseconds for handling large images. Although the Fuyu models are primarily optimized for digital agent applications, they demonstrate robust performance on standard image understanding benchmarks. One of the models in this family, the Fuyu-8B, is publicly available under a CC-BY-NC license, offering opportunities for research and development with the caveat that it may require fine-tuning to meet specific application needs.

Specifications(2 models)

Available From(1 provider)

Related Model Families

Persimmon

Models(2)