MergeRepair: An Exploratory Study on Merging Task-Specific Adapters in Code LLMs for Automated Program Repair

Abstract

Abstract Large Language Models (LLMs) have shown high capabilities in several software development-related tasks such as program repair, documentation, code refactoring, debugging, and testing. However, training these models requires massive amount of data and significant computational resources. Adapters are specialized, small modules designed for parameter efficient fine-tuning of LLMs for specific tasks, domains, or applications without requiring extensive retraining of the entire model. These adapters offer a more efficient way to customize LLMs for particular needs, leveraging the pre-existing capabilities of the large model. Model (and adapter) merging have emerged as a technique to develop one model capable of multiple tasks, with minimal or no training required. Although model and adapter merging has shown promising performance in domains such as natural language processing and computer vision, its applicability to software engineering tasks remains underexplored. In this paper, we investigate the effectiveness of merged adapters within the context of software engineering, with a particular focus on the Automated Program Repair (APR) task, through our approach, MergeRepair . In particular, we merge multiple task-specific adapters using three different merging methods, including weight-averaging, TIES-Merging, and DARE-TIES, and evaluate the performance of the merged adapter on the APR task. We introduce a continual merging approach, a novel method in which we sequentially merge the task-specific adapters where the order and weight of the merged adapters play a significant role. We further compare the performance of our approach with a baseline method consisting of equal-weight merging applied on parameters of different adapters, where all adapters are of equal importance. Through continual merging, we explore the capability of merged adapters and the effect of task order, as it occurs in real-world software projects. To evaluate our approach, we consider two LLMs, StarCoder2 and Granite models on an APR benchmark, HumanEvalFix, with various combinations of the merged adapters trained on a subset of the CommitPackFT dataset that contains four tasks and APR in Python. The results show that merging task-specific adapters can enhance the performance of models on the APR task, without additional training . In particular, our approach achieves up to 2.38% performance improvement over APR task in terms of pass@1 score when merging APR with Improvement and Misc task-adapters on top of the StarCoder2 model. Performance rises to 4.01% for pass@10 score with the same model and merged adapters. This improvement could be significantly influenced by the base model and the specific task adapters merged, rather than only the number of adapters. The proposed continual merging approach can be effective if the order of merging is efficiently determined. Results show that performance improves when the most effective task-adapter is added last in a continual merging process. Moreover, we can achieve on-par or even better performance when APR adapter is not used in the merged model, emphasizing the generalizeability of the approach. These findings suggest that leveraging knowledge from other code-related tasks through adapter merging can improve performance on new tasks like APR, even with limited data.

The authors' abstract, as published at the source. Empirical Software Engineering, 2026 · DOI ↗