The Hashcode of results should not be the same if they belong to different input models

[t-ober]

Currently when having e.g. two Transformer2WResult from different transformers their hash is the same so one of them gets discarded when you put them both into the same set. This should not happen.