Deep Learning Energy Measurement and Optimization

Project News ⚡

• [2026/02] Supported by Laude Institute's Slingshots // TWO!
• [2025/12] With NVIDIA, Google, and Meta, we led a NeurIPS 25 tutorial on Energy and Power as First‑Class ML Design Metrics!
• [2025/12] The ML.ENERGY leaderboard got a major upgrade to v3. Read our in-depth technical analysis blog post.
• [2025/09] We shared our experience and design philosophy for The ML.ENERGY Benchmark in our NeurIPS 25 D&B Spotlight paper.
• [2025/05] Zeus now supports CPU, DRAM, AMD GPU, Apple Silicon, and NVIDIA Jetson platform energy measurement!
• [2024/11] Perseus, an optimizer for large model training, appeared at SOSP'24! Paper | Blog | Optimizer
• [2024/05] Zeus is now a PyTorch ecosystem project. Read the PyTorch blog post here!
• [2024/02] Zeus was selected as a 2024 Mozilla Technology Fund awardee!

Zeus is a library for (1) measuring the energy consumption of Deep Learning workloads and (2) optimizing their energy consumption.

Zeus is part of The ML.ENERGY Initiative.

Repository Organization

zeus/
├── zeus/             # ⚡ Zeus Python package
│  ├── monitor/       #    - Energy and power measurement (programmatic & CLI)
│  ├── optimizer/     #    - Collection of time and energy optimizers
│  ├── device/        #    - Abstraction layer over CPU and GPU devices
│  ├── utils/         #    - Utility functions and classes
│  ├── _legacy/       #    - Legacy code to keep our research papers reproducible
│  ├── metric.py      #    - Prometheus metric export support
│  ├── show_env.py    #    - Installation & device detection verification script
│  └── callback.py    #    - Base class for callbacks during training
│
├── zeusd             # 🌩️ Zeus daemon
│
├── docker/           # 🐳 Dockerfiles and Docker Compose files
│
└── examples/         # 🛠️ Zeus usage examples

Getting Started

Please refer to our Getting Started page. After that, you might look at

• Measuring Energy
• Optimizing Energy

Docker image

We provide a Docker image fully equipped with all dependencies and environments. Refer to our Docker Hub repository and Dockerfile.

Examples

We provide working examples for integrating and running Zeus in the examples/ directory.

Research

Zeus is rooted on multiple research papers. Even more research is ongoing, and Zeus will continue to expand and get better at what it's doing.

1. Zeus (NSDI 23): Paper | Blog | Slides
2. Chase (ICLR Workshop 23): Paper
3. Perseus (SOSP 24): Paper | Blog | Slides
4. The ML.ENERGY Benchmark (NeurIPS 25 D&B Spotlight): Paper | Repository | Leaderboard
5. Where Do the Joules Go? Diagnosing Inference Energy Consumption: ArXiv | Blog

If you find Zeus relevant to your research, please consider citing:

@inproceedings{zeus-nsdi23,
    title     = {Zeus: Understanding and Optimizing {GPU} Energy Consumption of {DNN} Training},
    author    = {Jie You and Jae-Won Chung and Mosharaf Chowdhury},
    booktitle = {USENIX NSDI},
    year      = {2023}
}

Other Resources

1. Energy-Efficient Deep Learning with PyTorch and Zeus (PyTorch conference 2023): Recording | Slides

Contact

Jae-Won Chung (jwnchung@umich.edu)

Newsletter

Subscribe to the ML.ENERGY newsletter for the latest news on Zeus and other projects by the ML.ENERGY Initiative.