Save and load of trained models

[sneaxiy]

It seems that we do not have a doc to illustrate the save and load process about trained models.

• Where are the models saved. SQLFlow would save the models in local database in local mode, while it would save in OSS in ODPS platform. But where are the models, for example, in which database/table, which OSS bucket?
• If the users train twice using the same model name in INTO clause, would the second trained model overwrite the first one? Or, do we save two models when keeping different version number? If we keep the different version number, how to load or clean the previously saved models?
• How to extract the saved models so that they can be uploaded to OSS or downloaded to another local machine?
• How to load the saved models for incremental training?
• How to load the saved models for online serving using the other serving framework like TensorFlow lite?