I already wanted to share this happy moment. 🙃 It took me a lot of hours to get something worth demoing. Still not 100% good but at least the 2 agents are switching, guessing letters and finishing the game. (I simplified the setup and removed the host agent. There are now just 2 agents playing)
I was first trying with (too) small models without luck but yesterday I got my GPU working and now I could use bigger models (Via IPEX-LLM Ollama Serve). I ended up with the 'qwen2.5:32b-instruct-q5_K_M' model. See the demo below. I will make a follow up post later with more details. In that post I will also share my code.