[WIP] 410 refactor of model initialisation ie weight loading model freezing transfer learning

[JesperDramsch]

Description

Implements instantiatable Model modifiers, that can, e.g. load weights or freeze components.

What problem does this change solve?

It implements a modular and extensible system for the initialisation of models that gets rid of a stack of nested if statements and enables extension in anticipation of #248

What issue or task does this change relate to?

Closes #410
Prepares changes for #248

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.

---

📚 Documentation preview 📚: https://anemoi-training--442.org.readthedocs.build/en/442/

---

📚 Documentation preview 📚: https://anemoi-graphs--442.org.readthedocs.build/en/442/

---

📚 Documentation preview 📚: https://anemoi-models--442.org.readthedocs.build/en/442/

[JesperDramsch]

I just saw that in the meantime, there was this change, which I will have to address in a future commit:

        model.data_indices = self.data_indices
        # check data indices in original checkpoint and current data indices are the same
        self.data_indices.compare_variables(model._ckpt_model_name_to_index, self.data_indices.name_to_index)

[JPXKQX]

Thanks Jesper, I think this refactoring makes a lot of sense. Would it make sense to have another modifier "ResumeRun..." to which we could bring all the logic from run_id, fork_run_id, load_only_weights and warm_start?

[JesperDramsch]

Thanks Jesper, I think this refactoring makes a lot of sense. Would it make sense to have another modifier "ResumeRun..." to which we could bring all the logic from run_id, fork_run_id, load_only_weights and warm_start?

Possibly. I believe my original design around fork_run_id ended up confusing most people, so we could take a look whether this design could be used to fix that.