Closed nakytoe closed 1 year ago
compare:
process:
Train new model, push pickled model to model store: dev container -> model.pickle -> model store
Update model to production by loading pickled model version from model store: model store -> model.pickle -> api/prod container
must have: api for pushing / loading models to remote located store (from e.g. laptop to on-prem or from an on-prem instance to another)
bonus: gui for viewing models in store
done
create instructions / template for creating a model store. At first, the model store is another git repo that stores pickles of trained models with relevant metadata. In the future, consider more sophisticated options.