Installations

1. Clone the project

    $ git clone https://github.com/${YOUR_GITHUB_USERNAME}/${YOUR_REPOSITORY_NAME}.git
   

2. Adjust the workspace/Dockerfile based on CUDA Toolkit Version:

   • Change the CUDA version of the base image to your CUDA Toolkit Version.

       FROM nvidia/cuda:{YOUR-CUDA-TOOLKIT-VERSION}-base-ubuntu22.04 as base    
       (other settings...)
     

   See more official CUDA docker images here to find valid CUDA version images.

3. Start up the project

    $ docker compose up --build -d
   

   This might take a few minutes.

4. Connect to the docker container.

5. Adjust Pytorch source in workspace/pyproject.toml.

    [tool.poetry.dependencies]
    torch={version="2.1.2+cu121", source="torch"} # Edit here
   
    [[tool.poetry.source]]
    name = "torch"
    url = "https://download.pytorch.org/whl/cu121" # Edit here
    priority = "supplemental"
   

   After editting the config, run:

    $ poetry lock --no-update
    $ poetry install --no-root --sync
   

6. Run command below to check if the installation is successful.

   1. In the container, open terminal-1 and run:

       # Print out GPU information every 3 seconds
       $ nvidia-smi -l 3
      

   2. In the terminal-2, run:

       $ cd project_name
       $ poetry run python main.py
      

   You should see the process in the terminal-1 and gpu GPU-Util more than 0%.