trained-models
trained-models copied to clipboard
neuronets
improve workflow to perform all operations on self-hosted runner
Partial workflow at https://github.com/hvgazula/trained-models/actions/runs/6806826215 barring pushing the branch and docker image on successful testing.
datalad push ... in a new job keeps failing. For now, leaving it as a step in the same job that configures git seem to work fine.
@hvgazula I checked the workflow right now, is it this one? https://github.com/hvgazula/trained-models/actions/runs/6813853293/job/18529675234
Sure, you can use that by removing the last job and moving the steps into the previous one.
Here's the final workflow (with a test docker file) that works end-to-end on the self-hosted ec2 runner and pushes the docker image, weights, and inference scripts only if the docker and singularity tests run successfully. https://github.com/hvgazula/trained-models/actions/runs/6831844963 (edited to add the link).
and here's the final branch layout https://github.com/hvgazula/trained-models/tree/issue-7
Hello @hvgazula, really nice. Do you think it would be best to separate into multiple jobs? Or do you think it is best to keep everything inside one job (a.k.a "create branch")?
I'd leave it as is. No more refactoring at this point. If at all you are interested in separating into multiple jobs, try fixing this workflow first. This issue partially had a role to play in why I ended up with all the steps in one job sandwiched between start and stop ec2 runner. Good luck!