MARS: Multimodal Adaptation Rank Selection

Status: 🚧 In Preparation

1. Overview

MARS addresses Modality Convergence Imbalance in MLLM fine-tuning. By leveraging Dual Scaling Laws that capture the relationship between dataset size, convergence dynamics, and model capacity, MARS analytically derives the optimal rank ratio between the visual encoder and the LLM backbone.

• Paper: https://minkyoungcho.github.io/mars/
• Webpage: https://minkyoungcho.github.io/mars/

2. Release Roadmap

We are currently finalizing the codebase and documentation to ensure reproducibility.

Current Status:

• Code cleanup and refactoring

To Do:

• Documentation
• Scripts for simple tests and reproducibility

3. Acknowledgements

This codebase is built upon Cornstarch. We thank the authors for providing a robust foundation for multimodal research.

4. Contact & Citation

If you have questions about the paper or the upcoming release, please feel free to reach out: Minkyoung Cho (minkycho@umich.edu)

Citation

If you find our theoretical framework useful, please cite our preprint:

@article{cho2026mars,
    title={MARS: Harmonizing Multimodal Convergence via Adaptive Rank Search},
    author={Cho, Minkyoung and Jang, Insu and Jin, Shuowei and Zhao, Zesen and Jothi, Adityan and Can, Ethem F. and Chen, Min-Hung and Mao, Z. Morley},
    journal={arXiv preprint},
    year={2026}
}