This was discussed here, I'll copy the points over (with slight modification):
We have a tricky situation in our Card class where there can be multiple sources of truth for what the actual model is. One source is the model attribute, the other the model as (optionally) defined by metadata.model_file.
self.model is a model instance and there is no metadata.model_fileself.model is a model instance and there is a metadata.model_fileself.model is a str or Path and there is no metadata.model_fileself.model is a str or Path and there is a metadata.model_file and they're identicalself.model is a str or Path and there is a metadata.model_file and they're different
I think the getting started code snippet should always refer to the model_file from metadata if present. For case 3, one could argue that the path from self.model should be used for the getting started code snippet, but it's a bit surprising.
For the model diagram and hyperparams, I would argue we should use the data from self.model.
Ideally, we could agree on a "single source of truth" for the model, but it's not clear to me how to do that, as we don't have any mechanism to enforce that they're referring to the same model.
This was discussed here, I'll copy the points over (with slight modification):
We have a tricky situation in our
Cardclass where there can be multiple sources of truth for what the actual model is. One source is themodelattribute, the other the model as (optionally) defined bymetadata.model_file.self.modelis a model instance and there is nometadata.model_fileself.modelis a model instance and there is ametadata.model_fileself.modelis astrorPathand there is nometadata.model_fileself.modelis astrorPathand there is ametadata.model_fileand they're identicalself.modelis astrorPathand there is ametadata.model_fileand they're differentI think the getting started code snippet should always refer to the model_file from metadata if present. For case 3, one could argue that the path from
self.modelshould be used for the getting started code snippet, but it's a bit surprising.For the model diagram and hyperparams, I would argue we should use the data from self.model.
Ideally, we could agree on a "single source of truth" for the model, but it's not clear to me how to do that, as we don't have any mechanism to enforce that they're referring to the same model.