Thu 1 May 2025 13:30 - 14:00 at Canada Hall 3 Poster Area - Thu Lunch Posters 13:30-14:00
Software developers frequently use code completion tools to accelerate software development by suggesting the following code elements. Researchers usually employ AutoRegressive (AR) decoders to complete code sequences in a left-to-right, token-by-token fashion. To improve the accuracy and efficiency of code completion, we argue that tokens within a code statement have the potential to be predicted concurrently. In this article, we first conduct an empirical study to analyze the dependency among the target tokens in line-level code completion. The results suggest that it is potentially practical to generate all statement tokens in parallel. To this end, we introduce SANAR, a simple and effective syntax-aware non-autoregressive model for line-level code completion. To further improve the quality of the generated code, we propose an adaptive and syntax-aware sampling strategy to boost the model’s performance. The experimental results obtained from two widely used datasets indicate that our model outperforms state-of-the-art code completion approaches of similar model size by a considerable margin, and is faster than these models with up to 9_ speed-up. Moreover, the extensive results additionally demonstrate that the enhancements achieved by SANAR become even more pronounced with larger model sizes, highlighting their significance.
Wed 30 AprDisplayed time zone: Eastern Time (US & Canada) change
Thu 1 MayDisplayed time zone: Eastern Time (US & Canada) change
13:30 - 14:00 | Thu Lunch Posters 13:30-14:00Journal-first Papers / New Ideas and Emerging Results (NIER) / Research Track at Canada Hall 3 Poster Area | ||
13:30 30mPoster | Non-Autoregressive Line-Level Code Completion Journal-first Papers | ||
13:30 30mTalk | LLM-Based Test-Driven Interactive Code Generation: User Study and Empirical Evaluation Journal-first Papers Sarah Fakhoury Microsoft Research, Aaditya Naik University of Pennsylvania, Georgios Sakkas University of California at San Diego, Saikat Chakraborty Microsoft Research, Shuvendu K. Lahiri Microsoft Research Link to publication | ||
13:30 30mTalk | SusDevOps: Promoting Sustainability to a First Principle in Software Delivery New Ideas and Emerging Results (NIER) Istvan David McMaster University / McMaster Centre for Software Certification (McSCert) | ||
13:30 30mPoster | Predicting the First Response Latency of Maintainers and Contributors in Pull Requests Journal-first Papers SayedHassan Khatoonabadi Concordia University, Montreal, Ahmad Abdellatif University of Calgary, Diego Elias Costa Concordia University, Canada, Emad Shihab Concordia University, Montreal | ||
13:30 30mPoster | RustAssistant: Using LLMs to Fix Compilation Errors in Rust Code Research Track Pantazis Deligiannis Microsoft Research, Akash Lal Microsoft Research, Nikita Mehrotra Microsoft Research, Rishi Poddar Microsoft Research, Aseem Rastogi Microsoft Research | ||
13:30 30mTalk | Relevant information in TDD experiment reporting Journal-first Papers Fernando Uyaguari Instituto Superior Tecnológico Wissen, Silvia Teresita Acuña Castillo Universidad Autónoma de Madrid, John W. Castro Universidad de Atacama, Davide Fucci Blekinge Institute of Technology, Oscar Dieste Universidad Politécnica de Madrid, Sira Vegas Universidad Politecnica de Madrid |