Skip to main content

Building Custom Images

Guide to creating and managing custom Docker images on DS01.

Quick Build​

image-create my-project

Interactive wizard guides you through:

  1. Phase 1: Choose framework (PyTorch, TensorFlow, JAX)
  2. Phase 2: Add Jupyter Lab
  3. Phase 3: Add data science packages
  4. Phase 4: Add custom packages

Dockerfile Location​

~/dockerfiles/my-project.Dockerfile

Adding Packages​

image-update                  # Select image, add/remove packages

The interactive GUI lets you:

  • View current packages
  • Add Python packages (pip)
  • Add system packages (apt)
  • Import from requirements.txt
  • Rebuild automatically

Advanced: Edit Dockerfile Directly​

vim ~/dockerfiles/my-project.Dockerfile
# Add: RUN pip install transformers datasets accelerate

# Then rebuild:
image-update my-project --rebuild
# Or: docker build -t ds01-<uid>/my-project:latest -f ~/dockerfiles/my-project.Dockerfile ~/dockerfiles/

Base Images​

Available frameworks:

  • PyTorch 2.8.0 (CUDA 12.4)
  • TensorFlow 2.16.1 (CUDA 12.3)
  • JAX 0.4.23 (CUDA 12.3)

Next Steps​