Emu2

Emu2 is a 37-billion-parameter generative multimodal model from BAAI's Vision team, released in December 2023 as the successor to Emu1. Built for both multimodal understanding and image generation in context, it set new open state-of-the-art results on multimodal understanding and generation benchmarks at release and was later accepted at CVPR 2024.

Inference code, weights, and a demo were open-sourced, making it one of the largest open generative multimodal models of its era and a reference point for in-context multimodal learning research. Date is era-level (December 2023).