opendilab/DI-sheep

Implements an Actor-Critic neural network trained via PPO (Proximal Policy Optimization) using the DI-engine framework, with a modular gym-compatible environment enabling algorithm research and customization. The stack integrates a Flask backend serving both human players and trained agents alongside a React frontend, with pre-trained model weights available via HuggingFace for immediate experimentation without training from scratch.

489 stars. No commits in the last 6 months.