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

Bencheng liao, Tao Hongyuan, Qian Zhang, Tianheng Cheng, Yingyue Li, Haoran Yin, LWY, Xinggang Wang

Multimodal Mamba: Decoder-only Multimodal State Space Model via Quadratic to Linear Distillation: 38 upvotes on Hugging Face Daily Papers, #6 of 33 papers on 2025-02-19. Day-by-day upvote history.

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

Paper page on Hugging Face · arXiv

Data: hysts-bot-data/daily-papers-stats and the Daily Papers API. Open data: tardellirs/paper-pulse-data. Sister project: Model Pulse, the download history of every model on the Hub.