A2S Institute Docker Images

Our stacks provide GPU-enabled Jupyter Notebook in Docker containers, which can also run on Kubernetes. The images are based on Jupyter docker-stacks jupyter/pytorch-notebook. All images are published on our ghcr.io and quay.io.

The stacks contain several machine learning packages such as TensorFlow, PyTorch, scikit-learn, and other machine learning tools. All images also include VSCode and xfce4 desktop environment.

Docker stack structure

• gpu-base-notebook:cuda*-pytorch*: contains Jupyter related libraries and also includes different cuda and pytorch versions. It also has VSCode and xfce4 desktop environment.
  • ml-notebook:cuda*-pytorch*: depends on gpu-base-notebook and includes several machine learning libaries such as TensorfLow, Keras, scipy, opencv, etc.
    • nlp-notebook:cuda*-pytorch*: depends on ml-notebook and includes NLP libraries such as spaCy, NLTK, llama-cpp-python and wikipedia-api.
    • geo-notebook:cuda*-v*: depends on ml-notebook and includes qgis desktop.
• gpu-base-notebook:cuda12-ubuntu12.04: based on nvidia/cuda:12.3.2-cudnn9-runtime-ubuntu20.04 and has Jupyter Notebook related components with Ubuntu 20.04.
  • ros:cuda12-noetic: based on gpu-base-notebook:cuda12-ubuntu20.04 and includes ros-noetic-desktop-full with the gazebo simulation, but does not have all the ML related libraries as in ml-notebook.

Avilable versions

• quay.io/a2s-institute/gpu-base-notebook:cuda12-pytorch-2.2.2
• quay.io/a2s-institute/ml-notebook:cuda12-pytorch-2.2.2
• quay.io/a2s-institute/geo-notebook:cuda12-pytorch-2.2.2
• quay.io/a2s-institute/nlp-notebook:cuda12-pytorch-2.2.2
• quay.io/a2s-institute/qgis:cuda12-v3.36.1
• quay.io/a2s-institute/ros:cuda12-noetic
• quay.io/a2s-institute/gpu-base-notebook:cuda11-pytorch-2.2.2
• quay.io/a2s-institute/ml-notebook:cuda11-pytorch-2.2.2
• quay.io/a2s-institute/nlp-notebook:cuda11-pytorch-2.2.2

Older images

• ghcr.io/a2s-institute/docker-stacks/gpu-notebook:cuda11.3.1-ubuntu22.04 (no vscode and xfce desktop)
• ghcr.io/a2s-institute/docker-stacks/gpu-notebook:cuda11.8.0-ubuntu22.04 (no vscode and xfce desktop)
• ghcr.io/a2s-institute/docker-stacks/gpu-notebook:cuda12.1.0-ubuntu22.04 (no vscode and xfce desktop)

Building and running A2S images locally

The base image contains several packages for deep learning projects with NVidia GPU support.

• Build notebook image with gpu support

  # cuda11 and pytorch 2.2.2
  bash build_and_publish.sh --registry ghcr.io --publish "" \
      --image gpu-base-notebook --tag cuda11-pytorch-2.2.2
  
  # cuda12 and pytorch 2.2.2
  bash build_and_publish.sh --registry ghcr.io --publish "" \
      --image gpu-base-notebook --tag cuda12-pytorch-2.2.2
  

• Run the image locally

  # with GPU
  docker run --gpus all --name ml-notebook -it --rm -d -p 8888:8888 \
         quay.io/ml-notebook:cuda12-pytorch-2.2.2
  
  # without GPU
  docker run --name ml-notebook -it --rm -d -p 8888:8888 \
         quay.io/ml-notebook:cuda12-pytorch-2.2.2
  

• Check Jupyter Notebook token via log and open the link

  docker logs --follow ml-notebook
  
  

Monitoring

You can monitor the GPU usage using nvtop