[go: up one dir, main page]

Skip to content

Add CI/CD template for KubeFlow workflows

Problem to solve

It's possible to do ML pipelines using GitLab, but it is a lot less documented and unclear where to start. We should remedy that by adding a project template that demonstrates a common implementation, such as TensorFlow, that can be used by anyone.

Intended users

Further details

Some product gaps may be identified during the creation of this first template; we should ensure that those get issues created.

Proposal

Look for a common ML workflow and implement a project template.

We should strongly consider looking for an opportunity to add custom metrics and artifacts as part of the pipeline (cc @jheimbuck_gl). There are unique quality metrics around ML (such as model fitness) that really lend themselves nicely to being measured on a per-pipeline or per-merge request basis.

Permissions and Security

Nothing unique here.

Documentation

We should consider opening up a new ML pipelines section of the documentation, and also writing a blog post that communicates this has now been documented and we have a good project template.

Testing

What does success look like, and how can we measure that?

What is the type of buyer?

Links / references

Edited by 🤖 GitLab Bot 🤖