@inproceedings{huang-etal-2026-drift,
    title = "{DRIFT}: Transferring Reasoning Priors for Efficient {MLLM} Fine-Tuning",
    author = "Huang, Chao  and
      Zhang, Zeliang  and
      Liu, Jiang  and
      Sun, Ximeng  and
      Wu, Jialian  and
      Yu, Xiaodong  and
      Wang, Ze  and
      Xu, Chenliang  and
      Barsoum, Emad  and
      Liu, Zicheng",
    editor = "Liakata, Maria  and
      Moreira, Viviane P.  and
      Zhang, Jiajun  and
      Jurgens, David",
    booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
    month = jul,
    year = "2026",
    address = "San Diego, California, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl/2026.findings-acl.1015/",
    pages = "20294--20309",
    ISBN = "979-8-89176-395-1",
    abstract = "Multimodal large language models (MLLMs) have made rapid progress, yet their reasoning ability often lags behind strong text-only LLMs. Bridging this gap typically requires large-scale multimodal reasoning data or reinforcement learning, incurring substantial cost. An appealing alternative is parameter-space model merging between reasoning-enhanced LLMs and MLLMs, but we show that naive merging is fragile: its effectiveness varies widely across model families and can significantly degrade performance (e.g., for Qwen-based MLLMs). We propose Directional Reasoning Injection for Fine-Tuning (DRIFT), a lightweight method that transfers reasoning knowledge in the gradient space while preserving multimodal alignment. DRIFT precomputes a reasoning prior from the parameter differences between text-only reasoning experts and multimodal models, and uses it to bias gradients during supervised fine-tuning. This design retains the simplicity of standard SFT pipelines while enabling efficient and stable reasoning transfer. Experiments on multimodal reasoning benchmarks, including MathVista and MathVerse, show that DRIFT consistently outperforms naive merging and standard SFT, and matches or surpasses training-intensive methods with substantially lower data and compute."
}