[cpill]

I wonder if we could have much smaller models if they train on less languages? i.e. python + yaml + json only or even an single languages with an cluster of models loaded into memory dynamically...?