Learn to Merge: Meta-Learning for Adaptive Multi-Task Model Merging

Model merging in the pretrain-finetune paradigm has proven effective by combining multiple finetuned models into one with multi-task capabilities. However, existing methods rely on fix or manually tuned merging coefficients, making the unified model sensitive to the initial merging strategy and suboptimal for downstream adaptation. Thus, this paper proposed an innovative model merging framework called MetaMerging, a novel meta-learning algorithm to adaptively optimize the merging coefficients to construct a unified model tailored for task-specific adapter training. By simulating adapter updates in an inner loop and meta-optimizing merging coefficients in an outer loop, MetaMerging produces more balanced and generalizable unified models. Extensive experiments on CV and NLP fields show strong performance of MetaMerging on various downstream tasks and demonstrate the effectiveness of meta-learning in our method compared to other parameter merging methods. Our code is available at https://github.com/cjcj46262/MetaMerging.

AI models are often trained separately for different tasks, but storing many models is inefficient. This paper proposes MetaMerging, a method that learns how to combine several trained models into one shared model more effectively. Instead of only making the combined model good immediately, MetaMerging makes it easier to improve for each task with small extra updates. Experiments on image and language tasks show that this leads to better performance while keeping the model compact and efficient.