MM1: Methods, Analysis & Insights From Multimodal LLM Pre-training

Mckinzie Brandon, Gan Zhe, Fauconnier Jean-philippe, Dodge Sam, Zhang Bowen, Dufter Philipp, Shah Dhruti, Du Xianzhi, Peng Futang, Weers Floris, Belyi Anton, Zhang Haotian, Singh Karanjeet, Kang Doug, Jain Ankur, Hè Hongyu, Schwarzer Max, Gunter Tom, Kong Xiang, Zhang Aonan, Wang Jianyu, Wang Chong, Du Nan, Lei Tao, Wiseman Sam, Yin Guoli, Lee Mark, Wang Zirui, Pang Ruoming, Grasch Peter, Toshev Alexander, Yang Yinfei. Arxiv 2024

[Paper]    
Few Shot Fine Tuning In Context Learning Model Architecture Multimodal Models Pretraining Methods Prompting Reinforcement Learning Training Techniques

In this work, we discuss building performant Multimodal Large Language Models (MLLMs). In particular, we study the importance of various architecture components and data choices. Through careful and comprehensive ablations of the image encoder, the vision language connector, and various pre-training data choices, we identified several crucial design lessons. For example, we demonstrate that for large-scale multimodal pre-training using a careful mix of image-caption, interleaved image-text, and text-only data is crucial for achieving state-of-the-art (SOTA) few-shot results across multiple benchmarks, compared to other published pre-training results. Further, we show that the image encoder together with image resolution and the image token count has substantial impact, while the vision-language connector design is of comparatively negligible importance. By scaling up the presented recipe, we build MM1, a family of multimodal models up to 30B parameters, including both dense models and mixture-of-experts (MoE) variants, that are SOTA in pre-training metrics and achieve competitive performance after supervised fine-tuning on a range of established multimodal benchmarks. Thanks to large-scale pre-training, MM1 enjoys appealing properties such as enhanced in-context learning, and multi-image reasoning, enabling few-shot chain-of-thought prompting.

Similar Work