• Emu1 (ICLR 2024, 2023/07) - Generative Pretraining in Multimodality

• Emu2 (CVPR 2024, 2023/12) - Generative Multimodal Models are In-Context Learners

• Emu3 (arXiv 2024, 2024/09) - Next-Token Prediction is All You Need 🔥🔥🔥

News

• 2024.9 We introduce Emu3, a new suite of state-of-the-art multimodal models trained solely with next-token prediction. 🔥🔥🔥
• 2024.2 Emu1 and Emu2 are accepted by ICLR 2024 and CVPR 2024 respectively! 🎉
• 2023.12 Inference code, model and demo of Emu2 are available. Enjoy the demo.
• 2023.12 We have released Emu2, open and largest generative multimodal models that achieve new state of the art on multimodal understanding and generation tasks.
• 2023.7 Inference code and model of Emu are available.
• 2023.7 We have released Emu, a multimodal generalist that can seamlessly generate images and texts in multimodal context.

Hightlights

• State-of-the-art performance
• Next-generation capabilities
• A base model for diverse tasks

We hope to foster the growth of our community through open-sourcing and promoting collaboration👬. Let's step towards multimodal intelligence together🍻.

Contact

• We are hiring at all levels at BAAI Vision Team, including full-time researchers, engineers and interns. If you are interested in working with us on foundation model, visual perception and multimodal learning, please contact Xinlong Wang (wangxinlong@baai.ac.cn).

Misc