Eagle: Exploring The Design Space for Multimodal LLMs with Mixture of Encoders
This work addresses the challenge of reducing hallucinations and improving performance in MLLMs for tasks like optical character recognition and document analysis, though it is incremental as it builds on existing mixture-of-encoders approaches.
The study tackled the problem of improving multimodal large language models (MLLMs) by systematically exploring design choices for using multiple vision encoders, finding that simple concatenation of visual tokens is as effective as complex methods and introducing Pre-Alignment to enhance coherence, with the resulting Eagle models surpassing other open-source models on major benchmarks.
The ability to accurately interpret complex visual information is a crucial topic of multimodal large language models (MLLMs). Recent work indicates that enhanced visual perception significantly reduces hallucinations and improves performance on resolution-sensitive tasks, such as optical character recognition and document analysis. A number of recent MLLMs achieve this goal using a mixture of vision encoders. Despite their success, there is a lack of systematic comparisons and detailed ablation studies addressing critical aspects, such as expert selection and the integration of multiple vision experts. This study provides an extensive exploration of the design space for MLLMs using a mixture of vision encoders and resolutions. Our findings reveal several underlying principles common to various existing strategies, leading to a streamlined yet effective design approach. We discover that simply concatenating visual tokens from a set of complementary vision encoders is as effective as more complex mixing architectures or strategies. We additionally introduce Pre-Alignment to bridge the gap between vision-focused encoders and language tokens, enhancing model coherence. The resulting family of MLLMs, Eagle, surpasses other leading open-source models on major MLLM benchmarks.