Make Custom Image

What is a custom image?

This is a function that allows you to use the Docker image that includes the packages you need instead of the standard Docker image provided by the ABEJA Platform. You can use Docker image created based on Docker image provided by ABEJA Platform on ABEJA Platform.

Creating custom images

ABEJA page on Dockerhub Select the image.

The following images can be used as parent image of custom images.

• abeja/base
• abeja/all-cpu
• abeja/all-gpu

Check here for details about each image.

※Attention　”18.10” is not available.

Write a Dockerfile to create a custom image.

abeja/base is a very simple image that only includes a python3.6 and runtime of ABEJA platform, you can easily introduce a new version of TensorFlow as follows:

FROM abeja/base:19.10

RUN python -m pip --no-cache-dir install --upgrade tensorflow==2.0.0

If you want to use the latest version of PyTorch, it will be as follows.

FROM abeja/base:19.10

RUN python -m pip --no-cache-dir install --upgrade numpy
RUN python -m pip --no-cache-dir install --upgrade --pre torch torchvision -f https://download.pytorch.org/whl/nightly/cpu/torch_nightly.html

The introduction sample of GPU version and other DL library etc. are also released on GitHub. Create Docker Image with reference to these.

Next, run docker build to create a Docker image. In the sample, it is named my-custom-image:0.0.1

$ docker build --no-cache -t my-custom-image:0.0.1 -f Dockerfile .