Is it possible to have two NLU models in one pipeline? Here’s why I ask this question - We use RASA as part of our application and with every project we do, the number of training phrases increases. Because of the time it takes it to train RASA, we pick only a subset of the training phrases and use it for the next implementation.
Can we create a master Model that contains all the training phrases we have collected so far? (Take one day to train and we do it once a week). Once this model is ready, we add another smaller model into the pipeline that only contains the new training phases (takes ~15 mins to train but we can run it 5 times per day). Is it possible to use these two models in tandem to increase the accuracy of the NLU engine while keeping the training time within a reasonable limit?