Closed mmcky closed 1 year ago
@jstac I propose I enable the following custom build runner workflow (which builds on EC2
) for previews only and do some test run on the next few PR's, before we look at publishing from these runs.
So GPU
support is available currently for:
The publishing will need to be enabled in a subsequent PR.
Note: Test build cache workflows to ensure that works the same way as GitHub Actions hosted workflow.
@HumphreyYang I will merge this PR soon which will change how previews
are build via GitHub Actions
. It would be great if you could take close looks at previews for the next couple of changes and let me know if you can spot any issues.
This PR is bringing the ability to build code
that requires access to a GPU
so we can test this out on #237
@HumphreyYang I will merge this PR soon which will change how
previews
are build viaGitHub Actions
. It would be great if you could take close looks at previews for the next couple of changes and let me know if you can spot any issues.This PR is bringing the ability to build
code
that requires access to aGPU
so we can test this out on #237
Thanks @mmcky. I will keep an eye on the new previews.
This PR enables building on a custom runner hosted on EC2 using
p2.xlarge
instances for PREVIEW buildsThese instances have:
All setup and use of EC2 is automated and works the same way as the previous Preview Workflow Script.
The
cml
runner manages the EC2 instance so the instance only runs when preview builds are running. As a result it takes a little longer to setup as it fetches a fresh instance each time a preview build runs. It then uses a Docker image based onubuntu
and installs anaconda etc. However this means we are only paying for EC2 when previews are getting built and the instance is torn down when the preview run is completed. This makes this option very cost effective.Checks
https://github.com/QuantEcon/meta/issues/56