Installing and Building Zeus Components

This document explains how to install the zeus Python package.

Tip

We encourage users to utilize our Docker image. Please refer to Environment setup. Quick command:

docker run -it --gpus all --cap-add SYS_ADMIN --ipc host mlenergy/zeus:latest bash

zeus Python package

Note

This is already installed inside the container if you're using our Docker image.

Dependencies

Install PyTorch. For instance, if you have CUDA 11.8:

pip3 install torch==2.0.1 --index-url https://download.pytorch.org/whl/cu118

Install zeus

To install the latest release of zeus:

pip install zeus-ml

If you would like to follow the HEAD:

git clone https://github.com/ml-energy/zeus.git zeus
cd zeus
pip install .

For those would like to make changes to the source code and run them, we suggest an editable installation:

git clone https://github.com/ml-energy/zeus.git zeus
cd zeus
pip install -e .

Zeus power monitor

Warning

The C++ Zeus power monitor is now deprecated as we've switched to a Python-based power monitor. See PowerMonitor or run python -m zeus.monitor --help.

Dependencies

All dependencies are pre-installed if you're using our Docker image.

1. CMake >= 3.22
2. CUDAToolkit, especially NVML (libnvidia-ml.so)

Configuration

You can change the #defines in lines 28 to 31 in zemo/zemo.hpp to configure what information from the GPU is polled. By default, only the momentary power draw of the GPU will be collected.

Building the power monitor

# Working directory is repository root
cd zeus_monitor
cmake .
make

The resulting power monitor binary is zeus_monitor/zeus_monitor (/workspace/zeus/zeus_monitor/zeus_monitor inside the Docker container).