non ascii phrases aren't correctly determined

[Tarnak-public]

When using custom model with non English phrases (exactly Polish words with accents) I had problems with correct classifying texts using is_spam().
As a workaround I've used accents remover during train and checking( code: https://gist.github.com/AdoHaha/a76157c6de5155bf6b0adc77988724d9 ) which works great.
So, could you add normalizing parameter into code or fix accents somehow?