That is one of the coolest things I have seen in years! Great work. The concept of "executable weights" is also very interesting.

I don't understand the implication of colorful text, it just seems hard to read. Why not just text?

The colors tell you the confidence the model has in any given word. If you don't want the color coding then don't pass the -pc flag. Then it'll output plain text.