I have two BERT models to implement multilabel classification for vulnerability detection in code. One has been trained on source code, and the other has been trained on compiled code. The task they implement is multilabel classification, so the single output of both models is an array of 6 elements where each element can be 0 or 1, indicating whether the vulnerability is present or not.
I want to build a classic ML classifier (as RandomForest, SVM or Logistic Regression etc.) on top of these two models, implementing the ensemble technique called Stacking. How can I implement this, knowing that I'm dealing with multilabel classification?
The main doubt I have is how to stack the output (which would became the input of the final classifier) and it should be the only input the model will receive or if I have to input something else.