MergeRepair: Merging Task-Specific Adapters in Code LLMs for Automated Program RepairRegistered Reports Paper
[Context] Merging Large Language Models (LLMs) and adapters (small modules used for parameter efficient fine-tuning of an LLM) have shown promising results for various natural language domains and tasks. Merging adapters enables re-using the learned models without additional training for a new task. [Objective] This research aims to empirically study the capabilities of merged adapters in Code LLMs, specially for the Automated Program Repair (APR) task. The goal is to gain insight into how merging task-specific adapters can affect the performance of APR. [Method] In our approach, MergeRepair, we plan to merge multiple adapters trained on different tasks and evaluate the performance of the merged adapter for the APR task. Particularly, we will employ two main merging scenarios. Merging with equal-weight averaging in which the parameters of different adapters are averaged out equally; and our proposed approach, continual merging, in which we keep one merged adapter and sequentially merge it with added task-specific adapters. By merging with equal-weight averaging, we will investigate the improvement and generalizability of merged adapters for APR. By continual merging, we will explore the capability of merged adapters and the order of the tasks, which occurs in real-world software development projects.
Fri 11 OctDisplayed time zone: Arizona change
10:30 - 12:00 | Session 12: Machine Learning in Software EngineeringTool Demo Track / Research Track / New Ideas and Emerging Results Track / Registered Reports Track at Abineau Chair(s): Mohammed Sayagh ETS Montreal, University of Quebec | ||
10:30 15m | Can We Do Better with What We Have Done? Unveiling the Potential of ML Pipeline in NotebooksResearch Track Paper Research Track | ||
10:45 10m | MergeRepair: Merging Task-Specific Adapters in Code LLMs for Automated Program RepairRegistered Reports Paper Registered Reports Track Meghdad Dehghan University of British Columbia, Jie JW Wu University of British Columbia (UBC), Fatemeh Hendijani Fard University of British Columbia, Ali Ouni ETS Montreal, University of Quebec Pre-print | ||
10:55 15m | On the Use of Deep Learning Models for Semantic Clone DetectionResearch Track Paper Research Track Subroto Nag Pinku University of Saskatchewan, Debajyoti Mondal , Chanchal K. Roy University of Saskatchewan, Canada | ||
11:10 10m | GlueTest: Testing Code Translation via Language InteroperabilityNIER Paper New Ideas and Emerging Results Track Muhammad Salman Abid Cornell University, Mrigank Pawagi Indian Institute of Science, Bengaluru, Sugam Adhikari Islington College, Xuyan Cheng Dickinson College, Ryed Badr University of Illinois Urbana Champaign, Md Wahiduzzaman BRAC University, Vedant Rathi Adlai E Stevenson High School, Ronghui Qi Wuhan University, Choiyin Li Po Leung Kuk Ngan Po Ling College, Lu Liu University of Washington, Rohit Sai Naidu Dublin High School, Licheng Lin Zhejiang University, Que Liu University of Shanghai for Science and Technology, Asif Zubayer Palak BRAC University, Mehzabin Haque University of Dhaka, Xinyu Chen University of Illinois Urbana Champaign, Darko Marinov University of Illinois at Urbana-Champaign, Saikat Dutta Cornell University | ||
11:20 10m | Does Co-Development with AI Assistants Lead to More Maintainable Code? A Registered ReportRegistered Reports Paper Registered Reports Track Markus Borg CodeScene, Dave Hewett Equal Experts, Donald Graham Equal Experts, Noric Couderc Lund University, Emma Söderberg Lund University, Luke Church University of Cambridge | Candela Inc, Dave Farley Continuous Delivery Pre-print | ||
11:30 15m | Leveraging Large Vision-Language Model For Better Automatic Web GUI TestingResearch Track Paper Research Track Siyi Wang , Sinan Wang Southern University of Science and Technology, Yujia Fan , Xiaolei Li , Yepang Liu Southern University of Science and Technology | ||
11:45 5m | StackRAG Agent: Improving Developer Answers with Retrieval-Augmented GenerationTool Demo Paper Tool Demo Track Davit Abrahamyan University of British Columbia, Fatemeh Hendijani Fard University of British Columbia | ||