Skip to content

Hot-Reloaded GPU Docker Development

← Back to Tutorials Overview

Abdelkader Haddag
Abdelkader Haddag
Deep Learning Engineer & Researcher • 📅 Aug 5, 2026

This tutorial explains how to containerize your workstation environment using GPU acceleration, live volume mounts for instant hot-reloading, and interactive notebook services via docker/docker-compose.yml.

🐳 Step 1: Docker Architecture Overview

The docker/docker-compose.yml file defines three dedicated services:

  1. dev: Hot-reloaded PyTorch training & development container with GPU access.
  2. docs: Live ProperDocs / MaterialX documentation server on port 8942.
  3. notebook: Interactive Marimo / Jupyter reactive notebook server on port 8888.

🚀 Step 2: Launching Development Containers

Launch the GPU-accelerated training service with live volume mounting:

# Start primary training service
docker compose -f docker/docker-compose.yml up dev

Live Volume Mount Mapping

  • src/ \(\rightarrow\) /app/src: Source code changes hot-reload instantly without rebuilding Docker images.
  • inputs/ \(\rightarrow\) /app/inputs: Host datasets and hyperparameter YAML configs.
  • outputs/ \(\rightarrow\) /app/outputs: Generated checkpoints output directly to host filesystem.
  • logs/ \(\rightarrow\) /app/logs: Metric logs output to host.

📚 Step 3: Serving Documentation in Docker

Launch the documentation server container:

docker compose -f docker/docker-compose.yml up docs

Open http://localhost:8942 in your browser to inspect your live rendered ProperDocs + MaterialX documentation site.