GPTNT Documentation
Warning
Currently, the focus of this documentation is to help you run your own model and submit your results to the leaderboard. Understanding more about the codebase itself is currently in progress.
This project has gone through several major overhauls. Digging into it is strongly encouraged, but you are going to want to know what exists and how things work before you can get stuck all the way in. While the code is the ultimate source of truth, this documentation is meant to provide background and context on the decisions made, and anything else that is not immediately obvious.
What is GPTNT?¶
GPTNT is a benchmark for real-time collaboration between multimodal agents. It uses the game Keep Talking and Nobody Explodes (KTANE) to evaluate how well agents can work together to solve problems in real time. If you want to know more and see how it worked in practice, check out our website for all the details in full.
How to start?¶
- Follow the instructions in Get Started to make sure you can run the dummy models.
- Check out Run the benchmark to run your own model.