Installing and Building Zeus Components
This document explains how to install the
zeus Python package.
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
This is already installed inside the container if you're using our Docker image.
Install PyTorch. For instance, if you have CUDA 11.8:
pip3 install torch==2.0.1 --index-url https://download.pytorch.org/whl/cu118
To install the latest release of
pip install zeus-ml
If you would like to follow the
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
The C++ Zeus power monitor is now deprecated as we've switched to a Python-based power monitor.
PowerMonitor or run
python -m zeus.monitor --help.
All dependencies are pre-installed if you're using our Docker image.
- CMake >= 3.22
- CUDAToolkit, especially NVML (
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
/workspace/zeus/zeus_monitor/zeus_monitor inside the Docker container).