We present CM3Leon (pronounced "Chameleon"), a retrieval-augmented, token-based, decoder-only multi-modal language model capable of generating and infilling both text and images. CM3Leon uses the CM3 multi-modal architecture but additionally shows the extreme benefits of scaling up and tuning on more diverse instruction-style data. It is the first multi-modal model trained with a recipe adapted from text-only language models, including a large-scale retrieval-augmented pre-training stage and a second multi-task supervised fine-tuning (SFT) stage. It is also a general-purpose model that can do both text-to-image and image-to-text generation, allowing us to introduce self-contained contrastive decoding methods that produce high-quality outputs. Extensive experiments demonstrate that this recipe is highly effective for multi-modal models. CM3Leon achieves state-of-the-art performance in text-to-image generation with 5x less training compute than comparable methods (zero-shot MS-COCO FID of 4.88). After SFT, CM3Leon can also demonstrate unprecedented levels of controllability in tasks ranging from language-guided image editing to image-controlled generation and segmentation.
@article{arxiv.2309.02591,
title = {Scaling Autoregressive Multi-Modal Models: Pretraining and Instruction Tuning},
author = {Lili Yu and Bowen Shi and Ramakanth Pasunuru and Benjamin Muller and Olga Golovneva and Tianlu Wang and Arun Babu and Binh Tang and Brian Karrer and Shelly Sheynin and Candace Ross and Adam Polyak and Russell Howes and Vasu Sharma and Puxin Xu and Hovhannes Tamoyan and Oron Ashual and Uriel Singer and Shang-Wen Li and Susan Zhang and Richard James and Gargi Ghosh and Yaniv Taigman and Maryam Fazel-Zarandi and Asli Celikyilmaz and Luke Zettlemoyer and Armen Aghajanyan},
journal= {arXiv preprint arXiv:2309.02591},
year = {2023}
}