Comment by bsenftner

The same way developers incorporate it now. Why are you thinking "pre-training", this is a feature of the deployed model: it ingests documents and generates a mini-fine tune right then.