Estimated time: 8 minutes
- Docker version >18.09 (Docker installation instructions)
- Python >=3.6
- >5GB of free machine storage
- Download the latest release from the GitHub releases
Create a new, empty directory where you will install Atlas.
atlas_installer.pyfile into this directory.
Run the install script with
python atlas_installer.pyand follow the instructions.
python atlas_installer.py --help will give you troubleshooting advice if the script isn't working as expected.
The longest part of the script is pulling the Atlas docker images, if the script fails at this point,
you can re-run it using
python atlas_installer.py -dp to skip over the download and unpacking and go directly to the image pull.
After completing the installation section, you can do the following to start Atlas:
- Validate that you are in the same Python environment that was used to run the installation script.
If you installed Atlas with GPU support, you can start the Atlas server in GPU mode by running
atlas-server start -g. This will allow Atlas to use all CUDA-enabled GPUs on your system.
Validate that the GUI is running by going to the GUI. This is your centralized location to track all of your experiments.
After completing the start-up section, follow the next few steps to launch your first Atlas job:
- Navigate to where you'd like to create your Atlas project directory.
- Ensure that you are in the environment that was used during installation.
foundations init hello-atlasto create an example project.
- Navigate into the newly created
- Run the sample code provided by running
- Head to the GUI to see your experiment!
When you run a job for the first time, it will download the appropriate worker image needed.
This will take roughly 1.5 minutes.