Comment by vasergen

The training set is huge and model "forgets" some of the stuff, providing docs in context makes sense, plus docs could be up to date comparing to training set.