QMamba: Quantum Selective State Space Models for Text Generation

Gerhard Stenzel, Michael Kölle, Tobias Rohe, Maximilian Mansky, Jonas Nüßlein, Thomas Gabor

2025

Abstract

Quantum machine learning offers novel paradigms to address limitations in traditional natural language processing models, such as fixed context lengths and computational inefficiencies. In this work, we propose QMamba, the first quantum adaptation of the Mamba architecture, integrating selective state space models with quantum computation for efficient and scalable text generation. QMamba leverages quantum principles like superposition and entanglement to enable unbounded context sizes and reduced computational complexity. Our contributions include the development of a quantum generative model optimized for hardware constraints, advancements in encoding, embedding, and measurement techniques, and the demonstration of its performance on pattern reproduction and context-challenging tasks like ”Needle in a Haystack.” Experimental results confirm QMamba’s potential to maintain high efficiency and performance across varying sequence lengths, laying the groundwork for future explorations in quantum-enhanced natural language processing.

Paper Citation

in Harvard Style

Stenzel G., Kölle M., Rohe T., Mansky M., Nüßlein J. and Gabor T. (2025). QMamba: Quantum Selective State Space Models for Text Generation. In Proceedings of the 17th International Conference on Agents and Artificial Intelligence - Volume 1: QAIO; ISBN 978-989-758-737-5, SciTePress, pages 742-750. DOI: 10.5220/0013378300003890

in Bibtex Style

@conference{qaio25,
author={Gerhard Stenzel and Michael Kölle and Tobias Rohe and Maximilian Mansky and Jonas Nüßlein and Thomas Gabor},
title={QMamba: Quantum Selective State Space Models for Text Generation},
booktitle={Proceedings of the 17th International Conference on Agents and Artificial Intelligence - Volume 1: QAIO},
year={2025},
pages={742-750},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013378300003890},
isbn={978-989-758-737-5},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 17th International Conference on Agents and Artificial Intelligence - Volume 1: QAIO
TI - QMamba: Quantum Selective State Space Models for Text Generation
SN - 978-989-758-737-5
AU - Stenzel G.
AU - Kölle M.
AU - Rohe T.
AU - Mansky M.
AU - Nüßlein J.
AU - Gabor T.
PY - 2025
SP - 742
EP - 750
DO - 10.5220/0013378300003890
PB - SciTePress