Tue 5 Dec 2023 15:00 - 15:15 at Golden Gate A - Empirical Studies I Chair(s): Cristian Cadar

ML-powered code generation aims to assist developers to write code in a more productive manner by intelligently generating code blocks based on natural language prompts. Recently, large pretrained deep learning models have pushed the boundary of code generation and achieved impressive performance. However, the huge number of model parameters poses a significant challenge to their adoption in a typical software development environment, where a developer might use a standard laptop or mid-size server to develop code. Such large models cost significant resources in terms of memory, latency, dollars, as well as carbon footprint. Model compression is a promising approach to address these challenges. Out of many compression techniques, we have identified that quantization is the most applicable for code generation task as it does not require significant retrain- ing cost. As quantization represents model parameters with lower-bit integer (e.g., int8), the model size and runtime la- tency would both benefit. We empirically evaluate quantized models on code generation tasks across different dimensions: (i) resource usage and carbon footprint, (ii) accuracy, and (iii) robustness. Through systematic experiments we find a code- aware quantization recipe that could run even a 6-billion- parameter model in a regular laptop without significant ac- curacy or robustness degradation. We find that the recipe is readily applicable to code summarization task as well.

Tue 5 Dec

Displayed time zone: Pacific Time (US & Canada) change

14:00 - 15:30
Empirical Studies IIdeas, Visions and Reflections / Research Papers / Industry Papers / Journal First at Golden Gate A
Chair(s): Cristian Cadar Imperial College London
14:00
15m
Talk
[Remote] Assess and Summarize: Improve Outage Understanding with Large Language Models
Industry Papers
Pengxiang Jin Nankai University, Shenglin Zhang Nankai University, Minghua Ma Microsoft Research, Haozhe Li Peking University, Yu Kang Microsoft Research, Liqun Li Microsoft Research, Yudong Liu Microsoft Research, Bo Qiao Microsoft Research, Chaoyun Zhang Microsoft, Pu Zhao Microsoft Research, Shilin He Microsoft Research, Federica Sarro University College London, Yingnong Dang Microsoft Azure, Saravan Rajmohan Microsoft 365, Qingwei Lin Microsoft, Dongmei Zhang Microsoft Research
DOI Media Attached
14:15
15m
Talk
Open Source License Inconsistencies on GitHub
Journal First
Thomas Wolter Friedrich-Alexander University Erlangen-Nuernberg, Ann Barcomb Department of Electrical and Software Engineering, Schulich School of Engineering, University of Calgary, Dirk Riehle U of Erlangen, Nikolay Harutyunyan Friedrich-Alexander University Erlangen-Nuremberg, Germany
Media Attached
14:30
15m
Talk
On the Relationship Between Code Verifiability and Understandability
Research Papers
Kobi Feldman College of William & Mary, Martin Kellogg New Jersey Institute of Technology, Oscar Chaparro William & Mary
Media Attached
14:45
15m
Talk
Lessons from the Long Tail: Analysing Unsafe Dependency Updates across Software Ecosystems
Ideas, Visions and Reflections
Supatsara Wattanakriengkrai Nara Institute of Science and Technology, Raula Gaikovina Kula Nara Institute of Science and Technology, Christoph Treude University of Melbourne, Kenichi Matsumoto Nara Institute of Science and Technology
Media Attached
15:00
15m
Talk
Towards Greener Yet Powerful Code Generation via Quantization: An Empirical Study
Research Papers
Xiaokai Wei AWS AI Labs, Sujan Kumar Gonugondla AWS AI Labs, Shiqi Wang AWS AI Labs, Wasi Ahmad AWS AI Labs, Baishakhi Ray Columbia University, Haifeng Qian AWS AI Labs, Xiaopeng Li AWS AI Labs, Varun Kumar AWS AI Labs, Zijian Wang AWS AI Labs, Yuchen Tian AWS, Qing Sun AWS AI Labs, Ben Athiwaratkun AWS AI Labs, Mingyue Shang AWS AI Labs, Murali Krishna Ramanathan AWS AI Labs, Parminder Bhatia AWS AI Labs, Bing Xiang AWS AI Labs
Media Attached
15:15
15m
Talk
Understanding Hackers’ Work: An Empirical Study of Offensive Security Practitioners
Industry Papers
Andreas Happe TU Wien, Jürgen Cito TU Wien
DOI Media Attached