version.model should return None if no model

[stellasphere]

Description

There should be no instance where version.model should return a model type class when there is no valid model present.

Fixing this issue: https://roboflow.slack.com/archives/C056ZUJPK08/p1719359598748219

Type of change

Please delete options that are not relevant.

• Bug fix (non-breaking change which fixes an issue)

How has this change been tested, please provide a testcase or example of how you tested the change?

Locally
and here: https://colab.research.google.com/drive/1rnYv49x4oVWt--O3alw2wg6pr2qNapEO?usp=sharing

Any specific deployment considerations

For example, documentation changes, usability, usage/costs, secrets, etc.

Docs

• Docs updated? What were the changes:

[LinasKo]

What I cannot evaluate is whether prior behavior, creating the model object without a matching model on the api, is valid use case somewhere.

[stellasphere]

Thanks @LinasKo !

What I cannot evaluate is whether prior behavior, creating the model object without a matching model on the api, is valid use case somewhere.

I strongly believe that creating a model object when there is no model present is invalid behavior because the object it creates, a child of an InferenceModel, and its methods predict and load_model will fail because it doesn't exist. Doing so also makes it impossible to determine from the package that a model does or does not exist.

[LinasKo]

One other request: there's a test.py, which I think was included by accident.

[LinasKo]

One other request: there's a test.py, which I think was included by accident.

[stellasphere]

Yup, I need to revoke my key too. Thanks for catching!

[LinasKo]

Approving, with the caveat that I did not run it locally, nor ran any tests.

[stellasphere]

Sorry, fixed a failing test. @LinasKo can you re-approve?

[LinasKo]

M'afraid test.py snuck its way into the commit again 😉

What I'm seeing is - the model may not exist, but that's valid only in a project version where it's not been trained yet. Is that right?

[stellasphere]

M'afraid test.py snuck its way into the commit again 😉

🤦 You would think i would learn, but no.

the model may not exist, but that's valid only in a project version where it's not been trained yet. Is that right?

Yes, where it's not been trained. (or uploaded, but the same thing)

[iurisilvio]

I agree there is the risk someone use this behaviour, but I think it should be fixed, "no model" is a known behavior and we have to return None to make it explicit.

LGTM! 🎉