[BUG] mlflow.pytroch.load_model() can't load register model after the model class moved to another .py file about mlflow HOT 2 CLOSED

I think you're going to need to load the model weights directly using torch APIs directly and then re-save the model with MLflow in order to reflect the changes in your import module resolution. Torch uses pickle, which will maintain references to relative path imports for finding appropriate class definitions to deserialize objects from their stored bytes data.
This isn't an MLflow issue, though. This is a core Python concept with respect to how namespaces are resolved.

Can you update that old location with an import to the new file in your project? If not, you might want to work on some custom deserialization logic to handle the loading within your framework.

from mlflow.

Thanks you, I just do what you recommended, that import the model class temporarily to an old file to load the model and saved a new one, it is solved.

from mlflow.