StefanoFioravanzo
commented
2 months ago @tarilabs thanks a lot for getting this started! I went through though the current version, I think it's a good first version.
First of all a comment on timeline: we do not have to rush this in time for feature freeze. We have a couple more weeks to finetune and improve the docs. So let's keep this PR running until we have a good version.
Then, I'll refrain from commenting specific sections and instead provide some higher level feedback:
- Overview: We can write more about how a model registry fits within the e2e ecosystem. Good reference point is the AIIA Canonical Stack of Machine learning or Google’s MLOps maturity levels that shows how a model registry is necessary to achieve level 2 maturity. These may be things we want to discuss in an introduction here OR in a blog post. Up to you. But I do believe it’s important to provide some context on why this is needed and how the ecosystem benefits
We should also document a list of use cases and jobs that the model registry solves within the Kubeflow ecosystem. Why do I, as a Kubeflow user, care about the model registry? How does the model registry help me solve my day to day tasks faster, better, more efficiently? What problems does the model registry solve that I wouldn’t otherwise be able to solve? -
Getting Started: I would split up the getting started into two guides:
- Installation / pre-requisites → this is particularly important while this is in alpha and not included into Kubeflow core
- Actual "getting started" guide with a practical step-by-step tutorial (assuming one has already installed it)
This is important because installation is a one off thing (done by the Kubeflow admin) whereas the getting started guide targets ML practitioners who just use the platform
- I'd also love to see an architecture diagram under a "reference" section.
- Additionally, we can plan for a separate page to discuss specific integrations and integration plans with other Kubeflow components and third-party software (can go under "reference")
I know it's a lot. Great documentation takes a lot of time and effort. Let's try and do these things incrementally.
I hope that @diegolovison and @hbelmiro can also help out a little bit here. Looking forward to hearing your thoughts!
tarilabs
andreyvelich
google-oss-prow[bot]
jbottum
Opening this PR for early feedback.
As feedback received by Release Team during KF community meetings and Model Registry meetings, As identified in the Model Registry KF 1.9 tracker issue on request by Release Team,
➡️ Documentation is needed for Model Registry alpha on the KF website, as part of the KF 1.9 deliverables by Model Registry wg.
This PR focus on the following received requests:
This documentations is loosely based on the content presented with this demo.