Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/MachineLearningSystem/Zeus
An energy optimization framework for DNN training.
https://github.com/MachineLearningSystem/Zeus
Last synced: about 1 month ago
JSON representation
An energy optimization framework for DNN training.
- Host: GitHub
- URL: https://github.com/MachineLearningSystem/Zeus
- Owner: MachineLearningSystem
- License: apache-2.0
- Fork: true (ml-energy/zeus)
- Created: 2022-10-10T07:22:59.000Z (about 2 years ago)
- Default Branch: master
- Last Pushed: 2022-10-08T23:14:50.000Z (about 2 years ago)
- Last Synced: 2024-08-02T19:37:41.985Z (5 months ago)
- Homepage: https://ml.energy/zeus
- Size: 12.1 MB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
- awesome-AI-system - Zeus: Understanding and Optimizing {GPU} Energy Consumption of {DNN} Training NSDI'23
README
An Energy Optimization Framework for DNN Training
[](https://arxiv.org/abs/2208.06102)
[](https://hub.docker.com/r/symbioticlab/zeus)
[](https://github.com/SymbioticLab/Zeus/actions/workflows/deploy_homepage.yaml)
[](/LICENSE)
Zeus automatically optimizes the **energy and time** of training a DNN to a target validation metric by finding the optimal **batch size** and **GPU power limit**.
Please refer to our [NSDI’23 publication](https://arxiv.org/abs/2208.06102) for details.
Checkout [Overview](https://ml.energy/zeus/overview/) for a summary.
Zeus is part of [The ML.ENERGY Initiative](https://ml.energy).
## Repository Organization
```
.
├── zeus/ # ⚡ Zeus Python package
│ ├── run/ # - Tools for running Zeus on real training jobs
│ ├── policy/ # - Optimization policies and extension interfaces
│ ├── profile/ # - Tools for profiling energy and time
│ ├── simulate.py # - Tools for trace-driven simulation
│ ├── util/ # - Utility functions and classes
│ ├── analyze.py # - Analysis functions for power logs
│ └── job.py # - Class for job specification
│
├── zeus_monitor/ # 🔌 GPU power monitor
│ ├── zemo/ # - A header-only library for querying NVML
│ └── main.cpp # - Source code of the power monitor
│
├── examples/ # 🛠️ Examples of integrating Zeus
│ ├── capriccio/ # - Integrating with Huggingface and Capriccio
│ ├── cifar100/ # - Integrating with torchvision and CIFAR100
│ └── trace_driven/ # - Using the Zeus trace-driven simulator
│
├── capriccio/ # 🌊 A drifting sentiment analysis dataset
│
└── trace/ # 🗃️ Train and power traces for various GPUs and DNNs
```
## Getting Started
Refer to [Getting started](https://ml.energy/zeus/getting_started) for complete instructions on environment setup, installation, and integration.
### Docker image
We provide a Docker image fully equipped with all dependencies and environments.
The only command you need is:
```sh
docker run -it \
--gpus 1 `# Mount one GPU` \
--cap-add SYS_ADMIN `# Needed to change the power limit of the GPU` \
--shm-size 64G `# PyTorch DataLoader workers need enough shm` \
symbioticlab/zeus:latest \
bash
```
Refer to [Environment setup](https://ml.energy/zeus/getting_started/environment/) for details.
### Examples
We provide working examples for integrating and running Zeus:
- Integrating Zeus with Computer Vision
- [ImageNet](examples/imagenet)
- [CIFAR100](examples/cifar100)
- [Integrating Zeus with Natural Language Processing](examples/capriccio)
- [Running trace-driven simulation on single recurring jobs and the Alibaba GPU cluster trace](examples/trace_driven)
## Extending Zeus
You can easily implement custom policies for batch size and power limit optimization and plug it into Zeus.
Refer to [Extending Zeus](https://ml.energy/zeus/extend/) for details.
## Citation
Please consider citing our NSDI’23 paper if you find Zeus to be related to your research project.
```bibtex
@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}
}
```
## Contact
Jae-Won Chung ([email protected])