Extend model (NER, ED) loading functionality for custom models using aliases about rel HOT 5 CLOSED

I think this would be nice, but let's see if we can extend it to work with:

1. ED, which would be easy as it is our own framework.
2. Flair (custom trained models)
3. Sklearn (Logistic Regression, which is used for ED prob).

from rel.

Exactly. In such a case, the users would only have to download our archives once (for the database etc) and not every time we train a new model.

from rel.

So to clarify, you mean creating aliases in all areas where weights can be loaded pretty much? To eliminate downloading e.g. big archives like we have the user do now?

from rel.

Okay, then I understood correctly! +1 good idea I think.

from rel.

Working on this on model_aliases branch.

from rel.