Home Compute GPU Instances Reference Content

Docker AI images

Reviewed on 25 April 2023 • Published on 25 March 2022

Scaleway offers a range of ready-to-use AI Docker images. These Docker images can be used with all GPU Instance OS images.

You can pull the images from our Container Registry as follows:

docker pull rg.fr-par.scw.cloud/scw-ai/<IMAGE:TAG>

Our tag names follow the pattern :YYMM-py3, where YY and MM refer to the year and month the image was built. One exception to this is the tag used for the Tensorflow image, which is :YYMM-tf2-py3.

The latest images are also tagged with the :latest tag.

Initial tags will start with 2204-py3 (April 2022). Check out the Scaleway Changelog to be informed of Docker AI images updates.

All our images are based on CUDA 11.2 and cuDNN8.

Tip:

When using the “Ubuntu Focal GPU OS 11” image with your Instance, you can omit the nvidia runtime option

The following commands show how to launch a container based on each of our various Docker images:

Tensorflow

docker run --runtime=nvidia -it --rm -p 8888:8888 -p 6006:6006 rg.fr-par.scw.cloud/scw-ai/tensorflow:latest /bin/bash

The main libraries included in the Tensorflow image are: Tensorflow 2, Tensorboard, Autokeras, Numpy, Scikit-Learn, Scipy, Pandas, Matplotlib, Plotly, Bokeh, Seaborn, CatBoost, XGBoost, Pillow, ONNX, Spacy, Nvidia Dali, Optuna, HuggingFace Transformers and HuggingFace-Hub, Weights and Biases, JupyterLab, Code Server and JupyterLab Nvidia Dashboard.

Pytorch

docker run --runtime=nvidia -it --rm -p 8888:8888 -p 6006:6006 rg.fr-par.scw.cloud/scw-ai/pytorch:latest /bin/bash

The main libraries included in the Pytorch image are: Pytorch, Torch Audio, Torch Vision, Fast AI, Pytorch Lightning, Tensorboard, Numpy, Scikit-Learn, Scipy, Pandas, Matplotlib, Plotly, Bokeh, Seaborn, CatBoost, XGBoost, Pillow, Plotly, ONNX, Spacy, Nvidia Dali, Optuna, HuggingFace Transformers and HuggingFace-Hub, Weights and Biases, JupyterLab, Code Server and JupyterLab Nvidia Dashboard.

Jax

docker run --runtime=nvidia -it --rm -p 8888:8888 -p 6006:6006 rg.fr-par.scw.cloud/scw-ai/jax:latest /bin/bash

The main libraries included in the Jax image are: Jax, Numpy, Scikit-Learn, Scipy, Pandas, Matplotlib, Plotly, Bokeh, Seaborn, CatBoost, XGBoost, Pillow, ONNX, JupyterLab, Code Server and JupyterLab Nvidia Dashboard.

RAPIDS

Note: This image is built on top of the official RAPIDS Docker image, which relies on Anaconda.

docker run --runtime=nvidia -it --rm -p 8888:8888 -p 6006:6006 rg.fr-par.scw.cloud/scw-ai/rapids:latest /bin/bash

The main libraries included in the RAPIDS image are: cuDF, cuML, cuGraph, cuxfilter, cuspatial, cusignal, Dask, Numpy, Scikit-Learn, Scipy, Pandas, Matplotlib, Plotly, Bokeh, Seaborn, CatBoost, XGBoost, Pillow, ONNX, JupyterLab and JupyterLab Nvidia Dashboard.

Note:

As the official RAPIDS Docker image uses Anaconda, make sure the “rapids” conda environment is activated in order to use Rapids libraries:

jovyan@561f14e37a44:~/ai$ conda activate rapids
(rapids) jovyan@561f14e37a44:~/ai$ jupyter lab

“All” (Experimental)

If there are no dependency issues when buidling the image, the all image will try to include all the above listed libraries into a single Docker image (except RAPIDS)

docker run --runtime=nvidia -it --rm -p 8888:8888 -p 6006:6006 rg.fr-par.scw.cloud/scw-ai/all:latest /bin/bash