@inproceedings{sinii-etal-2025-steering,
    title = "Steering {LLM} Reasoning Through Bias-Only Adaptation",
    author = "Sinii, Viacheslav  and
      Gorbatovski, Alexey  and
      Cherepanov, Artem  and
      Shaposhnikov, Boris  and
      Balagansky, Nikita  and
      Gavrilov, Daniil",
    editor = "Christodoulopoulos, Christos  and
      Chakraborty, Tanmoy  and
      Rose, Carolyn  and
      Peng, Violet",
    booktitle = "Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing",
    month = nov,
    year = "2025",
    address = "Suzhou, China",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-emnlp/2025.emnlp-main.467/",
    pages = "9213--9222",
    ISBN = "979-8-89176-332-6",
    abstract = "We show that training a single $d$-dimensional steering vector per layer with reinforcement learning, while freezing all base weights, matches the accuracy of fully RL-tuned reasoning models on mathematical-reasoning tasks.On an 8 billion-parameter model this adds only $\approx 0.0016\%$ additional parameters and reproduces performance across a range of base models and mathematical-reasoning benchmarks.These results tighten the upper bound on the parameter budget required for high-level chain-of-thought reasoning, indicating that millions of adapter weights are unnecessary.The minimal trainable footprint reduces optimizer memory and inter-GPU communication, lowering the overall cost of fine-tuning.Moreover, a logit-lens analysis shows that the learned vectors amplify coherent token directions, providing clearer insight into the model{'}s internal computations."
}