Codenames - help with training AI for the tabletop game

[damucz]

I would like to use my own data to train an AI for playing the Codenames. Is here anybody interested enough who could give me a little guidance, please?

What is Codenames? It's word guessing game, where Spymaster hints several friendly agents on a board using exactly one word and a number. An Operative player has to contact these agents. Team contacting all its agents wins. https://czechgames.com/en/codenames/

What I have now. I am quite a newbie in GPT. I understand how to train our own data. I have the environment setup. I'm a member of Codenames developer team. Running online version (https://codenames.game/) and softlaunching Codenames app we have lots of real world data I can use to train.

The Questions Is GPT suitable for training on the input grid of words marked by a type (team A, team B, civilian, assassin)? How should prepare.py organize data? Could I train GPT for the non-english languages as well? Can I train AI to play both roles - spymaster and operative?

Can anyone point me out where I could start with nanoGPT? I read the lot great thinks about the project!

If I'm complete fool and crazy, don't hesitate to close this issue. :)