Thanks! No, I just tweaked the implementation that came with the paper, it had a WebGL renderer already.

I ran the GPT-J inference myself but it's not live on the server, it just has a 20mb text file of pre-generated quotes it samples from :)