fix: Add anemoi-models to dependencies

[HCookie]

Description

Add anemoi-(graphs & models) to dependencies

What issue or task does this change relate to?

Closes #350

As a contributor to the Anemoi framework, please ensure that your changes include unit tests, updates to any affected dependencies and documentation, and have been tested in a parallel setting (i.e., with multiple GPUs). As a reviewer, you are also responsible for verifying these aspects and requesting changes if they are not adequately addressed. For guidelines about those please refer to https://anemoi.readthedocs.io/en/latest/

By opening this pull request, I affirm that all authors agree to the Contributor License Agreement.

[bdvllrs]

Hi! Thanks for this. So no minimum version?

[HCookie]

@bdvllrs Not at the moment, partially as I have no idea what it would be.
As we update models, we can revisit this

[frazane]

I guess what's missing from this PR is some guardrails. Personally, I would opt for a check as soon as the checkpoint is loaded, where we assert that the version of anemoi-models currently installed in the python environment is the same as the one specified by the checkpoint's requirements. If it's not, then an error is raised, unless a flag is passed to the config, in which case only a warning is logged. What do you think @HCookie?

[anaprietonem]

I guess what's missing from this PR is some guardrails. Personally, I would opt for a check as soon as the checkpoint is loaded, where we assert that the version of anemoi-models currently installed in the python environment is the same as the one specified by the checkpoint's requirements. If it's not, then an error is raised, unless a flag is passed to the config, in which case only a warning is logged. What do you think @HCookie?

@frazane how would this be different then than the current workflow in main? - where one can run the inspect command get the version of models needed - install it and then run inference? - as in what would you win with this new approach if at the end installing anemoi-models need to be reviewed and checked by the user

[frazane]

I guess what's missing from this PR is some guardrails. Personally, I would opt for a check as soon as the checkpoint is loaded, where we assert that the version of anemoi-models currently installed in the python environment is the same as the one specified by the checkpoint's requirements. If it's not, then an error is raised, unless a flag is passed to the config, in which case only a warning is logged. What do you think @HCookie?

@frazane how would this be different then than the current workflow in main? - where one can run the inspect command get the version of models needed - install it and then run inference? - as in what would you win with this new approach if at the end installing anemoi-models need to be reviewed and checked by the user

There is no “win” in terms of workflow changes - it’s about correctness and clarity. If a package cannot perform its core functionality without another package installed, then that package is a required runtime dependency and should be declared as such. The fact that adding guardrails would lead to the same practical workflow as on main is, from my perspective, actually an argument in favour of this proposal.

Maybe I am being too rigid about this, I don't know. I agree that this discussion is of little practical importance, but I do think that the principle matters...