多模态Mamba：通过二次到线性蒸馏实现的仅解码器多模态状态空间模型

Multimodal Mamba: Decoder-only Multimodal State Space Model via Quadratic to Linear Distillation

February 18, 2025

作者: Bencheng Liao, Hongyuan Tao, Qian Zhang, Tianheng Cheng, Yingyue Li, Haoran Yin, Wenyu Liu, Xinggang Wang

cs.AI

摘要

近期，多模态大语言模型（MLLMs）虽取得了显著成效，却因二次方计算复杂度、不断增长的键值缓存需求及对独立视觉编码器的依赖，面临部署难题。为此，我们提出了mmMamba框架，旨在通过适度学术计算资源，从现有MLLMs逐步蒸馏，开发出线性复杂度的原生多模态状态空间模型。该方法无需预训练的基于RNN的LLM或视觉编码器，即可将已训练的仅解码器MLLMs直接转换为线性复杂度架构。我们提出了一种从训练好的Transformer中雕刻Mamba的种子策略及三阶段蒸馏方案，能有效传递Transformer知识至Mamba，同时保留多模态能力。此外，我们的方法支持灵活的混合架构，结合Transformer与Mamba层，实现可定制的效率-性能权衡。基于Transformer的仅解码器HoVLE蒸馏而来，mmMamba-linear在性能上与现有线性和二次方复杂度视觉语言模型（VLMs）相当，而mmMamba-hybrid则大幅提升性能，接近HoVLE的水平。在处理103K令牌时，mmMamba-linear相比HoVLE实现了20.6倍的加速和75.8%的GPU内存节省，mmMamba-hybrid则达到了13.5倍的加速和60.2%的内存节约。代码与模型已发布于https://github.com/hustvl/mmMamba。

English

Recent Multimodal Large Language Models (MLLMs) have achieved remarkable performance but face deployment challenges due to their quadratic computational complexity, growing Key-Value cache requirements, and reliance on separate vision encoders. We propose mmMamba, a framework for developing linear-complexity native multimodal state space models through progressive distillation from existing MLLMs using moderate academic computational resources. Our approach enables the direct conversion of trained decoder-only MLLMs to linear-complexity architectures without requiring pre-trained RNN-based LLM or vision encoders. We propose an seeding strategy to carve Mamba from trained Transformer and a three-stage distillation recipe, which can effectively transfer the knowledge from Transformer to Mamba while preserving multimodal capabilities. Our method also supports flexible hybrid architectures that combine Transformer and Mamba layers for customizable efficiency-performance trade-offs. Distilled from the Transformer-based decoder-only HoVLE, mmMamba-linear achieves competitive performance against existing linear and quadratic-complexity VLMs, while mmMamba-hybrid further improves performance significantly, approaching HoVLE's capabilities. At 103K tokens, mmMamba-linear demonstrates 20.6times speedup and 75.8% GPU memory reduction compared to HoVLE, while mmMamba-hybrid achieves 13.5times speedup and 60.2% memory savings. Code and models are released at https://github.com/hustvl/mmMamba