ML-Mamba: Efficient Multi-Modal Large Language Model Utilizing Mamba-2
This addresses the problem of high computational overhead in multimodal AI for researchers and practitioners, though it appears incremental as it builds on existing Mamba models.
The paper tackles the computational inefficiency of Transformer-based multimodal large language models by introducing ML-Mamba, which uses the Mamba-2 model for linear scalability and faster inference, achieving performance comparable to state-of-the-art methods like TinyLaVA and MobileVLM v2.
Multimodal Large Language Models (MLLMs) have attracted much attention for their multifunctionality. However, traditional Transformer architectures incur significant overhead due to their secondary computational complexity. To address this issue, we introduce ML-Mamba, a multimodal language model, which utilizes the latest and efficient Mamba-2 model for inference. Mamba-2 is known for its linear scalability and fast processing of long sequences. We replace the Transformer-based backbone with a pre-trained Mamba-2 model and explore methods for integrating 2D visual selective scanning mechanisms into multimodal learning while also trying various visual encoders and Mamba-2 model variants. Our extensive experiments in various multimodal benchmark tests demonstrate the competitive performance of ML-Mamba and highlight the potential of state space models in multimodal tasks. The experimental results show that: (1) we empirically explore how to effectively apply the 2D vision selective scan mechanism for multimodal learning. We propose a novel multimodal connector called the Mamba-2 Scan Connector (MSC), which enhances representational capabilities. (2) ML-Mamba achieves performance comparable to state-of-the-art methods such as TinyLaVA and MobileVLM v2 through its linear sequential modeling while faster inference speed; (3) Compared to multimodal models utilizing Mamba-1, the Mamba-2-based ML-Mamba exhibits superior inference performance and effectiveness.