Revisiting Model Interpolation for Efficient Reasoning
This work provides a practical framework for creating efficient reasoning models, but it is incremental as it builds on existing model merging approaches.
The paper tackles the problem of efficient reasoning by revisiting model interpolation, a simple weight-merging method, and finds that strategic interpolation can outperform more complex merging techniques, achieving better efficiency and effectiveness.
Model merging, typically on Instruct and Thinking models, has shown remarkable performance for efficient reasoning. In this paper, we systematically revisit the simplest merging method that interpolates two weights directly. Particularly, we observe that model interpolation follows a three-stage evolutionary paradigm with distinct behaviors on the reasoning trajectory. These dynamics provide a principled guide for navigating the performance-cost trade-off. Empirical results demonstrate that a strategically interpolated model surprisingly surpasses sophisticated model merging baselines on both efficiency and effectiveness. We further validate our findings with extensive ablation studies on model layers, modules, and decoding strategies. Ultimately, this work demystifies model interpolation and offers a practical framework for crafting models with precisely targeted reasoning capabilities. Code is available at \href{https://github.com/wutaiqiang/MI}{Github}.