docker build -t deepnote/python:<TAG> --build-arg FROM_PYTHON_VERSION=<some_version> ./python
For every Python version (see above), we also build a -conda variant, which includes Miniconda installation.
The Dockerfile for this is at ./python/Dockerfile.conda.
The Conda image is built on top of the -bullseye version of our Python image. We don't currently build it in the CI, so you need to build it locally.
docker build --build-arg FROM_PYTHON_VERSION=${PYTHON_VERSION} --build-arg DEBIAN_VERSION=bullseye -t deepnote/python:${PYTHON_VERSION}-bullseye ./python
docker build --build-arg FROM_PYTHON_VERSION=<some_version> -t deepnote/python:<some_version>-conda ./python --file ./python/Dockerfile.conda
Before pushing and overwritting the tag currently in use in production, please make sure that the image works as expected with our compute dependencies. We've seen case when the image worked locally (you could run conda), but it failed in the Deepnote environment with all compute dependencies installed (the conda command failed there).
You can do this by tagging the image with -test suffix and running it in Deepnote as custom environment.
We also have a variant with number of preinstalled libraries. We build it for all the above mentioned R versions
and the image is available under deepnote/ir-with-libs. See ir-with-libs for more details.
docker build -t deepnote/ir:<TAG> --build-arg R_BASE_VERSION=<some_version> ./ir
The image is based on tensorflow docker image and adds packages that are typical part of our other images (git most importantly).
We follow the tagging of tensorflow image. So tensorflow/tensorflow:2.4.1-gpu becomes deepnote/tensorflow:2.4.1-gpu.
Images are build and pushed to DockerHub using CircleCI pipeline. Not all images are covered by the continuous integration -- check the config for more details.
Pull requests are pushed too, but with -ra-PR_NUMBER suffix in the tag.