about us

Our goal is to measure, understand, optimize, and expose the energy consumption of modern machine learning.

The ML.ENERGY Initiative is a joint effort of computer scientists across multiple academic institutions, including SymbioticLab where it originally started.

Projects

Leaderboard logo

How much energy do modern large language models (LLMs) consume?

Website
Zeus logo

Deep learning energy measurement and optimization

Website

Members

Faculty

Mosharaf Chowdhury

University of Michigan

Adam Belay

MIT CSAIL

Asaf Cidon

Columbia University

Beidi Chen

CMU

Simon Peter

University of Washington

Tom Anderson

University of Washington

Students

Jae-Won Chung

Ph.D. Student

University of Michigan

Jiachen Liu

Ph.D. Student

University of Michigan

Alumni

Jie You

Alumnus, Ph.D.


Luoxi Meng

Alumnus, MS


Yile Gu

Alumnus, MS


Zhenning Yang

Alumnus, MS


News

Perseus → SOSP '24

Perseus, an energy optimization system for large model training, was accepted to appear at SOSP '24!

Paper Blog

PyTorch Ecosystem project

Zeus is now part of the PyTorch Ecosystem!

PyTorch blog

Mozilla Tech Fund 2024

Zeus receives $50,000 for development support from the 2024 Mozilla Technology Fund.

Announcement

Talk @ PyTorch conf 2023

Jae-Won gave a talk about energy-efficient ML at the PyTorch conference 2023!

YouTube Slides

ML.ENERGY Colosseum

Two LLMs will battle on your command in terms of both response quality and energy. Your judgement tips the scale of victory.

Colosseum

ML.ENERGY Leaderboard

A rich benchmark and comparison of performing inference on modern LLMs with metrics including energy, latency, and model quality!

Leaderboard Repository

Chase → ICLR Workshop '23

Based on Carbon-Aware Zeus, Chase was accepted to appear at the ICLR workshop on Tackling Climate Change with Machine Learning!

Repository Paper

Zeus in Carbon Hack 22

Carbon-Aware Zeus wins the Second best overall solution prize in CarbonHack22 organized by the Green Software Foundation!

Project page YouTube

Zeus → NSDI '23

Our first project "Zeus: Understanding and Optimizing GPU Energy Consumption of DNN Training" was accepted to appear at NSDI '23!

Zeus website Paper

Sponsors

Any opinions, findings, and conclusions of our works are those of the author(s) and do not necessarily represent the official policy of any of these organizations.

National Science Foundation Logo
VMWare Logo
Salesforce Logo
Mozilla Logo