With the rapid growth of Artificial Intelligence (AI) applications supported by deep learning (DL), the energy efficiency of these applications has an increasingly large impact on sustainability. We introduce Smaragdine, a new energy accounting system for tensor-based DL programs implemented with TensorFlow. At the heart of Smaragdine is a novel white-box methodology of energy accounting: Smaragdine is aware of the internal structure of the DL program, which we call tensor-aware energy accounting. With Smaragdine, the energy consumption of a DL program can be broken down into units aligned with its logical hierarchical decomposition structure. We apply Smaragdine for understanding the energy behavior of BERT, one of the most widely used language models. Layer-by-layer and tensor-by-tensor, Smaragdine is capable of identifying the highest energy/power-consuming components of BERT. Furthermore, we conduct two case studies on how Smaragdine supports downstream toolchain building, one on the comparative energy impact of hyperparameter tuning of BERT, the other on the energy behavior evolution when BERT evolves to its next generation, ALBERT.
Thu 18 AprDisplayed time zone: Lisbon change
11:00 - 12:30 | LLM, NN and other AI technologies 3New Ideas and Emerging Results / Research Track / Software Engineering Education and Training / Software Engineering in Practice at Pequeno Auditório Chair(s): Tushar Sharma Dalhousie University | ||
11:00 15mTalk | Xpert: Empowering Incident Management with Query Recommendations via Large Language Models Research Track Yuxuan Jiang University of Michigan Ann-Arbor, Chaoyun Zhang Microsoft, Shilin He Microsoft Research, Zhihao Yang Peking University, Minghua Ma Microsoft Research, Si Qin Microsoft Research, Yu Kang Microsoft Research, Yingnong Dang Microsoft Azure, Saravan Rajmohan Microsoft 365, Qingwei Lin Microsoft, Dongmei Zhang Microsoft Research | ||
11:15 15mTalk | Tensor-Aware Energy Accounting Research Track DOI Pre-print | ||
11:30 15mTalk | LLM4PLC: Harnessing Large Language Models for Verifiable Programming of PLCs in Industrial Control Systems Software Engineering in Practice Mohamad Fakih University of California, Irvine, Rahul Dharmaji University of California, Irvine, Yasamin Moghaddas University of California, Irvine, Gustavo Quiros Siemens Technology, Tosin Ogundare Siemens Technology, Mohammad Al Faruque UCI | ||
11:45 15mTalk | Resolving Code Review Comments with Machine Learning Software Engineering in Practice Alexander Frömmgen Google, Jacob Austin Google, Peter Choy Google, Nimesh Ghelani Google, Lera Kharatyan Google, Gabriela Surita Google, Elena Khrapko Google, Pascal Lamblin Google, Pierre-Antoine Manzagol Google, Marcus Revaj Google, Maxim Tabachnyk Google, Danny Tarlow Google, Kevin Villela Google, Dan Zheng Google DeepMind, Satish Chandra Google, Inc, Petros Maniatis Google DeepMind | ||
12:00 15mTalk | LLMs Still Can't Avoid Instanceof: An investigation Into GPT-3.5, GPT-4 and Bard's Capacity to Handle Object-Oriented Programming Assignments Software Engineering Education and Training | ||
12:15 7mTalk | Leveraging Large Language Models to Improve REST API Testing New Ideas and Emerging Results Myeongsoo Kim Georgia Institute of Technology, Tyler Stennett Georgia Institute of Technology, Dhruv Shah Georgia Institute of Technology, Saurabh Sinha IBM Research, Alessandro Orso Georgia Institute of Technology Pre-print | ||
12:22 7mTalk | LogExpert: Log-based Recommended Resolutions Generation using Large Language Model New Ideas and Emerging Results JiaboWang Beijing University of Posts and Telecommunications, guojun chu Beijing University of Posts and Telecommunications, Jingyu Wang , Haifeng Sun Beijing University of Posts and Telecommunications, Qi Qi , Yuanyi Wang Beijing University of Posts and Telecommunications, Ji Qi China Mobile (Suzhou) Software Technology Co., Ltd., Jianxin Liao Beijing University of Posts and Telecommunications | ||