For Large Vision-Language Models (LVLMs), scaling the model can effectively improve performance. However, expanding model parameters significantly increases the training and inferring costs, as all model parameters are activated for each token in the calculation. In this work, we propose a novel training strategy MoE-tuning for LVLMs, which can constructing a sparse model with an outrageous number of parameter but a constant computational cost, and effectively addresses the performance degradation typically associated with multi-modal learning and model sparsity. Furthermore, we present the MoE-LLaVA framework, a MoE-based sparse LVLM architecture. This framework uniquely activates only the top-k experts through routers during deployment, keeping the remaining experts inactive. Our extensive experiments highlight the excellent capabilities of MoE-LLaVA in visual understanding and its potential to reduce hallucinations in model outputs. Remarkably, with just 3 billion sparsely activated parameters, MoE-LLaVA demonstrates performance comparable to the LLaVA-1.5-7B on various visual understanding datasets and even surpasses the LLaVA-1.5-13B in object hallucination benchmarks. Through MoE-LLaVA, we aim to establish a baseline for sparse LVLMs and provide valuable insights for future research in developing more efficient and effective multi-modal learning systems. Code is released at \url{https://github.com/PKU-YuanGroup/MoE-LLaVA}.

本文介绍了一种针对大型视觉语言模型 (LVLMs) 的训练策略 MoE-tuning，通过构建一个具有巨大数量参数但恒定计算成本的稀疏模型，有效解决多模态学习和模型稀疏性带来的性能退化问题。此外，本文还提出了基于 MoE 的稀疏 LVLM 架构 MoE-LLaVA，通过在部署过程中仅激活前k个专家，使剩余的专家处于非活跃状态。实验证明，MoE-LLaVA 在视觉理解方面具有出色的能力，并且在模型输出的对象幻象基准测试中超越了 LLaVA-1.5-13B，在各种视觉理解数据集上表现可与 LLaVA-1.5-7B 相媲美。通过 MoE-LLaVA，我们旨在为稀疏 LVLMs 建立基准，并为未来开发更高效和有效的多模态学习系统提供有价值的见解。

MoE-LLaVA：大规模视觉语言模型的专家混合