Which model is the starting model when you uptrain a custom processor?

[AlexShull]

I have successfully uptrained an Invoice Parser, and now I want to continue training it with more labeled data. If I choose to uptrain a new version, do I need to train it with all the same documents that I used the first time, or can I uptrain with only the new labeled data? In other words, does the uptraining start on the same base model every time, or will it use whichever model I have selected as the default model under "Manage Versions"? I can't find anywhere indicating which model is the starting point. Thanks.