log_model

[dberenbaum]

We already have two closed issues about this 😓 , but I think we should add a log_model method that uses mlem to inspect the model artifact.

Benefits:

• Automatically captures metadata about the model artifact that users can see in the model registry to make it more meaningful.
• Makes model registry workflow more useful since users can deploy downstream with mlem.
• Automatically logs it as an artifact with type=model.
• Can support different options for how to save the model (related to Decouple dvc add from log_artifact #572).

[dberenbaum]

cc @aguschin

[aguschin]

Ok, now I'm getting what we reserved log_model for 😅

I assume log_model should simply call mlem.api.save and ideally have the same (in practice, almost the same) args https://mlem.ai/doc/api-reference/save

This should be an easy change. Happy to take it once we're done with MR.

[dberenbaum]

In addition, it would be nice to also support an already serialized model with https://mlem.ai/doc/api-reference/import_object. In other words, it would be nice to accept either a model object or a path to a model.