Emu: An Open Multimodal Generalist

Generative Pretraining in Multimodality

Quan Sun^1*, Qiying Yu^2,1*, Yufeng Cui^1*, Fan Zhang^1*, Xiaosong Zhang^1*, Yueze Wang¹, Hongcheng Gao¹,
Jingjing Liu², Tiejun Huang^1,3, Xinlong Wang¹

¹ BAAI, ² THU, ³ PKU
^* Equal Contribution

| Paper | Demo(tmp) |

Emu is a Large Multimodal Model (LMM) trained with a unified autoregressive objective, i.e., predict-the-next-element, including both visual embeddings and textual tokens. Trained under this objective, Emu can serve as a generalist interface for both image-to-text and text-to-image tasks.

Generalist Interface

Emu serves as a generalist interface capable of diverse multimodal tasks, such as image captioning, image/video question answering, and text-to-image generation, together with new abilities like in-context text and image generation, and image blending:

Setup

Clone this repository and install required packages:

git clone https://github.com/baaivision/Emu
cd Emu

pip install -r requirements.txt

Model Weights

We release the pretrained and instruction-tuned weights of Emu. Our weights are subject to LLaMA's license.

Model name	Weight
Emu	🤗 HF link (27GB)
Emu-I	🤗 HF link (27GB)

Inference

At present, we provide inference code that can process interleaved image-text as input, and output text. This includes image-to-text tasks such as image captioning and visual question answering:

python inference.py --instruct --ckpt-path $Instruct_CKPT_PATH

Schedule

We are commited to open-sourcing all Emu related materials, including:

The weights of Emu and Emu-I
Inference example for interleaved image-text as input, text as output
Video inference example
The weights of image decoder & image generation/blending example
The YT-Storyboard-1B pretraining data
The pretraining code
The instruction tuning code
The evaluation code

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

Acknowledgement

We thank the great work from LLaMA, BLIP-2, Stable Diffusion, and FastChat.

Citation

If you find Emu useful for your research and applications, please consider starring this repository and citing:

@article{Emu,
  title={Generative Pretraining in Multimodality},
  author={Sun, Quan and Yu, Qiying and Cui, Yufeng and Zhang, Fan and Zhang, Xiaosong and Wang, Yueze and Gao, Hongcheng and Liu, Jingjing and Huang, Tiejun and Wang, Xinlong},
  publisher={arXiv:2307.05222},
  year={2023},
}

DDXDaniel/Emu