Immutable development environments for PyTorch powered by Visual Studio Code Dev Containers

This project provides a development environment for PyTorch using Visual Studio Code and Docker. It is based on the Visual Studio Code Dev Container feature. Being able to have immutable and reproducible development environments is a key part to successfully work with multiple projects.

Features

• devcontainer.json: Configuration file for the Visual Studio Code Dev Container feature, using a Docker image, based on nvidia/cuda:11.8.0-base-ubuntu22.04
• Optionally: install AWS CLI V2 and reuse your AWS credentials from the host system
• Optionally: install Docker inside the container and reuse your Docker daemon from the host system (if you're training inside a container again)
• Sophisticated set of extensions for Python, Jupyter, etc.
• Sopthisticated set of settings for Python (Linter, Formatter, etc.)
• Sample MNIST training script (PyTorch Example)

Getting Started

• (Prerequisite) Docker is installed on your system
• (Prerequisite) NVIDIA Container Toolkit is installed on your system
• (Prerequisite) Visual Studio Code Dev Container extension is installed
• Clone this repository
• Open the folder in Visual Studio Code
• Read comments in the devcontainer.json file
• Read comments in the setup.sh file
• When prompted, click "Reopen in Container"