EnergyVis: Interactively Tracking and Exploring Energy Consumption for\n ML Models

2021/03/30 by Omar Shaikh, Jon Saad-Falcon, Shaikh, Omar +11 · 2 citations
Engineering · #Green IT and Sustainability

paper · pdf · doi:10.48550/arxiv.2103.16435

Abstract

The advent of larger machine learning (ML) models have improved\nstate-of-the-art (SOTA) performance in various modeling tasks, ranging from\ncomputer vision to natural language. As ML models continue increasing in size,\nso does their respective energy consumption and computational requirements.\nHowever, the methods for tracking, reporting, and comparing energy consumption\nremain limited. We presentEnergyVis, an interactive energy consumption tracker\nfor ML models. Consisting of multiple coordinated views, EnergyVis enables\nresearchers to interactively track, visualize and compare model energy\nconsumption across key energy consumption and carbon footprint metrics (kWh and\nCO2), helping users explore alternative deployment locations and hardware that\nmay reduce carbon footprints. EnergyVis aims to raise awareness concerning\ncomputational sustainability by interactively highlighting excessive energy\nusage during model training; and by providing alternative training options to\nreduce energy usage.\n

Cited by

• Fine-Grained Energy Prediction For Parallellized LLM Inference With PIE-P
• Green AI: A systematic review and meta-analysis of its definitions, lifecycle models, hardware and measurement attempts

Related

• Computing Within Limits: An Empirical Study of Energy Consumption in ML Training and Inference
• Counting Carbon: A Survey of Factors Influencing the Emissions of Machine Learning
• Evaluating the Energy Consumption of Machine Learning: Systematic Literature Review and Experiments
• Carbontracker: Tracking and Predicting the Carbon Footprint of Training Deep Learning Models
• Towards the Systematic Reporting of the Energy and Carbon Footprints of Machine Learning