English

TransMamba: A Sequence-Level Hybrid Transformer-Mamba Language Model

Machine Learning 2026-01-08 v2

Abstract

Transformers are the cornerstone of modern large language models, but their quadratic computational complexity limits efficiency in long-sequence processing. Recent advancements in Mamba, a state space model (SSM) with linear complexity, offer promising efficiency gains but suffer from unstable contextual learning and multitask generalization. Some works conduct layer-level hybrid structures that combine Transformer and Mamba layers, aiming to make full use of both advantages. This paper proposes TransMamba, a novel sequence-level hybrid framework that unifies Transformer and Mamba through shared parameter matrices (QKV and CBx), and thus could dynamically switch between attention and SSM mechanisms at different token lengths and layers. We design the Memory Converter to bridge Transformer and Mamba by converting attention outputs into SSM-compatible states, ensuring seamless information flow at TransPoints where the transformation happens. The TransPoint scheduling is also thoroughly explored for balancing effectiveness and efficiency. We conducted extensive experiments demonstrating that TransMamba achieves superior training efficiency and performance compared to single and hybrid baselines, and validated the deeper consistency between Transformer and Mamba paradigms at sequence level, offering a scalable solution for next-generation language modeling. Code and data are available at https://github.com/Yixing-Li/TransMamba

Keywords

Cite

@article{arxiv.2503.24067,
  title  = {TransMamba: A Sequence-Level Hybrid Transformer-Mamba Language Model},
  author = {Yixing Li and Ruobing Xie and Zhen Yang and Xingwu Sun and Shuaipeng Li and Weidong Han and Zhanhui Kang and Yu Cheng and Chengzhong Xu and Di Wang and Jie Jiang},
  journal= {arXiv preprint arXiv:2503.24067},
  year   = {2026}
}

Comments

Accepted by AAAI 2026. Code: https://github.com/Yixing-Li/TransMamba

R2 v1 2026-06-28T22:40:33.434Z