This is an up-to-date Devcontainer template for developing Python projects.
It has optional support for TensorFlow and PyTorch on GPU enabled machines.
- Create a new repository using this GitHub template.
- Clone the repository and open it using VSCode.
- Run
python .devcontainer/change_gpu_config.pyto select your GPU mode. - Press
Ctrl+Shift+Pto open the command palette. - Search for
Dev Containers: Rebuild and Reopen in Container. - Follow the instructions in the terminal to install TensorFlow or PyTorch.
- You are now ready to start developing! 😀
| Name | Version | Image |
|---|---|---|
| Ubuntu | 24.04 | mcr.microsoft.com/devcontainers/base:ubuntu-24.04 |
| Name | Version | Image |
|---|---|---|
| Python | 3.12 | ghcr.io/devcontainers/features/python:1 |
| Git | Latest | ghcr.io/devcontainers/features/git:1 |
| GitHub CLI | Latest | ghcr.io/devcontainers/features/github-cli:1 |
| Docker | Latest | ghcr.io/devcontainers/features/docker-outside-of-docker:1 |
| Name | Version | Description | Optional |
|---|---|---|---|
| TensorFlow | 2.19.0 | AI development framework | Yes |
| PyTorch | 2.7.0 | AI development framework | Yes |
| Poetry | Latest | Venv-like solution with great dependency management | No |
| Name | Description | ID |
|---|---|---|
| Reload | VSCode reload button | natqe.reload |
| Python | Python language support | ms-python.python |
| Pylint | Static code analyser | ms-python.vscode-pylint |
| Black | Code formatter | ms-python.black-formatter |
| Jupyter | Jupyter extension pack | ms-toolsai.jupyter |
| Prettier | Code formatter | esbenp.prettier-vscode |
| Commit Reminder | Commit reminder | kurozero.vscode-commit-reminder |
| ❌ Error | ✅ Solution |
|---|---|
Shell scripts fail to run or complain about \r characters | Check if the End of Line formatting of the scripts is set to LF in VSCode. |
poetry shell fails or is not recognized as a command | Poetry shell was moved to a plugin (January 2025). Run pip install poetry-plugin-shell in the terminal. |
Note
Since PyTorch and Tensorflow - if installed correctly - already include the neccessary CUDA and cuDNN tools, you don't need to install them separately. This section is only to clarify the compatibility between CUDA libraries and the host NVIDIA driver.
| CUDA Toolkit | Linux Driver Version | Windows Driver Version |
|---|---|---|
| CUDA 12.x | >= 525.60.13 | >= 527.41 |
| CUDA 11.x | >= 450.80.02 | >= 452.39 |
- The CUDA Toolkits have minor version compatibility with the drivers.
This means that CUDA Toolkit 12.6 should work with a driver that's designed for CUDA 12.0 because they have the same major version. - The CUDA Toolkits are backwards compatible with the drivers.
This means that CUDA Toolkit 11.8 will still work with a newer driver that's designed for CUDA 12.x versions.
| Latest version | Python version | CUDA |
|---|---|---|
| TensorFlow 2.19.0 | 3.9-3.12 | 12.3 |
| PyTorch 2.7.0 | 3.9-3.13 | 12.8 |
The most recent common Python version is 3.12.
The hosts NVIDIA driver gets passed to the container using the NVIDIA Container Toolkit.
You can validate your Container Toolkit installation by checking the Docker daemon configuration file on your server: /etc/docker/daemon.json.
To spin up a GPU-accelerated container, append the --gpus=all and --runtime=nvidia arguments to your docker run command.
Luckily, these arguments are already preconfigured in devcontainer.json.
The NVIDIA driver on the A5000 server has version 12.0.
