Hugging Face introduces AI vs. AI: a deep RL multi-agent competition system on Spaces

AI Impact Summary

AI vs. AI is a Hugging Face Spaces-based tool that orchestrates multi-agent reinforcement learning tournaments. It uses a matchmaking queue, a background evaluation task, and a Hub-backed leaderboard that tracks ELO ratings for submitted agents; models pushed to the Hub are automatically evaluated against others. The system is environment-agnostic and demonstrated with the SoccerTwos Challenge in Unity ML-Agents, enabling rapid benchmarking across diverse policies. This setup emphasizes relative performance across a broad policy pool, aiding teams in diagnosing generalization gaps and accelerating policy iteration.