Language Models in Software Development Tasks: An Experimental Analysis of Energy and Accuracy
The use of generative AI-based coding assistants like ChatGPT and Github Copilot is a reality in contemporary software development. Many of these tools are provided as remote APIs. Using third-party APIs raises data privacy and security concerns for client companies, which motivates the use of locally-deployed language models. In this study, we explore the trade-off between model accuracy and energy consumption, aiming to provide valuable insights to help developers make informed decisions when selecting a language model. We investigate the performance of 18 families of LLMs in typical software development tasks on two real-world infrastructures, a commodity GPU and a powerful AI-specific GPU. Given that deploying LLMs locally requires powerful infrastructure which might not be affordable for everyone, we consider both full-precision and quantized models. Our findings reveal that employing a big LLM with a higher energy budget does not always translate to significantly improved accuracy. Additionally, quantized versions of large models generally offer better efficiency and accuracy compared to full-precision versions of medium-sized ones. Apart from that, not a single model is suitable for all types of software development tasks.
Tue 29 AprDisplayed time zone: Eastern Time (US & Canada) change
14:00 - 15:30 | AI for SE (2)Technical Papers / Data and Tool Showcase Track / Registered Reports / Industry Track at 214 Chair(s): Giuseppe Destefanis Brunel University London | ||
14:00 10mTalk | Automatic High-Level Test Case Generation using Large Language Models Technical Papers Navid Bin Hasan Bangladesh University of Engineering and Technology, Md. Ashraful Islam Bangladesh University of Engineering and Technology, Junaed Younus Khan Bangladesh University of Engineering and Technology, Sanjida Senjik Bangladesh University of Engineering and Technology, Anindya Iqbal Bangladesh University of Engineering and Technology Dhaka, Bangladesh | ||
14:10 10mTalk | Prompting in the Wild: An Empirical Study of Prompt Evolution in Software Repositories Technical Papers Mahan Tafreshipour University of California at Irvine, Aaron Imani University of California, Irvine, Eric Huang University of California, Irvine, Eduardo Santana de Almeida Federal University of Bahia, Thomas Zimmermann University of California, Irvine, Iftekhar Ahmed University of California at Irvine Pre-print | ||
14:20 10mTalk | Intelligent Semantic Matching (ISM) for Video Tutorial Search using Transformer Models Technical Papers | ||
14:30 10mTalk | Language Models in Software Development Tasks: An Experimental Analysis of Energy and Accuracy Technical Papers Negar Alizadeh Universiteit Utrecht, Boris Belchev University of Twente, Nishant Saurabh Utrecht University, Patricia Kelbert Fraunhofer IESE, Fernando Castor University of Twente | ||
14:40 10mTalk | TriGraph: A Probabilistic Subgraph-Based Model for Visual Code Completion in Pure Data Technical Papers Anisha Islam Department of Computing Science, University of Alberta, Abram Hindle University of Alberta | ||
14:50 5mTalk | Inferring Questions from Programming Screenshots Technical Papers Faiz Ahmed York University, Xuchen Tan York University, Folajinmi Adewole York University, Suprakash Datta York University, Maleknaz Nayebi York University | ||
14:55 5mTalk | Human-In-The-Loop Software Development Agents: Challenges and Future Directions Industry Track Jirat Pasuksmit Atlassian, Wannita Takerngsaksiri Monash University, Patanamon Thongtanunam University of Melbourne, Kla Tantithamthavorn Monash University, Ruixiong Zhang Atlassian, Shiyan Wang Atlassian, Fan Jiang Atlassian, Jing Li Atlassian, Evan Cook Atlassian, Kun Chen Atlassian, Ming Wu Atlassian | ||
15:00 5mTalk | FormalSpecCpp: A Dataset of C++ Formal Specifications Created Using LLMs Data and Tool Showcase Track Madhurima Chakraborty University of California, Riverside, Peter Pirkelbauer Lawrence Livermore National Laboratory, Qing Yi Lawrence Livermore National Laboratory | ||
15:05 10mTalk | Towards Detecting Prompt Knowledge Gaps for Improved LLM-guided Issue Resolution Technical Papers Ramtin Ehsani Drexel University, Sakshi Pathak Drexel University, Preetha Chatterjee Drexel University, USA Pre-print | ||
15:15 5mTalk | GENCNIPPET: Automated Generation of Code Snippets for Supporting Programming Questions Registered Reports | ||