Kai Fricke 6b9a56d28e [ci/docker/air] Update ML/DL dependencies to latest releases / Remove Py3.6 Docker images (#28808) | 2 年之前 | |
---|---|---|
.. | ||
README.md | ee2a8da87a [ci] Move to new hierarchical docker structure + pipeline (#28641) | 2 年之前 |
base.build.Dockerfile | ee2a8da87a [ci] Move to new hierarchical docker structure + pipeline (#28641) | 2 年之前 |
base.gpu.Dockerfile | 6b9a56d28e [ci/docker/air] Update ML/DL dependencies to latest releases / Remove Py3.6 Docker images (#28808) | 2 年之前 |
base.ml.Dockerfile | ee2a8da87a [ci] Move to new hierarchical docker structure + pipeline (#28641) | 2 年之前 |
base.test.Dockerfile | ee2a8da87a [ci] Move to new hierarchical docker structure + pipeline (#28641) | 2 年之前 |
build.Dockerfile | ee2a8da87a [ci] Move to new hierarchical docker structure + pipeline (#28641) | 2 年之前 |
gpu.Dockerfile | ee2a8da87a [ci] Move to new hierarchical docker structure + pipeline (#28641) | 2 年之前 |
ml.Dockerfile | ee2a8da87a [ci] Move to new hierarchical docker structure + pipeline (#28641) | 2 年之前 |
test.Dockerfile | ee2a8da87a [ci] Move to new hierarchical docker structure + pipeline (#28641) | 2 年之前 |
This directory contains the Dockerfiles used to build the CI system.
These are not the Dockerfiles that build the rayproject/ray
releases. These
are found in the /docker
directory under the root.
The Dockerfiles are hierarchical and will be built in different places during a CI run.
The base images are built per-branch either when they are first requested or on a periodic basis (for the master branch). The base images contain the latest dependencies of the respective branch. Every per-commit build will always install the latest dependencies to make sure everything is up to date. However, by using the base images as a source, this will mostly be a no or low cost operation.
base.test.Dockerfile
contains common dependencies for all imagesbase.build.Dockerfile
inherits from base.test
and installs build dependencies like Java and LLVMbase.ml.Dockerfile
inherits from base.test
and installs ML dependencies like torch/tensorflowbase.gpu.Dockerfile
inherits from a CUDA base image and otherwise contains the same content as base.test
and base.ml
.On every commit, the following images are built in this order:
build.Dockerfile
(based on base.build
) which will build the Ray binariestest.Dockerfile
(based on base.test
), where we will inject the built Ray librariesml.Dockerfile
(based on base.ml
), where we will inject the built Ray librariesgpu.Dockerfile
(based on base.ml
), where we will inject the built Ray libraries