That's what I thought too, the source code was not an issue so much as that.

What we need is some sort of "Large Language Model at Home" (like SETI@home was) that could crowdsource the creation of the model which would be free to use.