opendilab/DI-star
A distributed training platform for deep reinforcement learning agents that achieve grand-master level play in StarCraft II.

Velocity · 7d
+0.8
★ / day
Trend
→steady
star history
This project provides a large-scale game AI training platform specifically for StarCraft II, implementing both supervised learning and reinforcement learning training pipelines. The system uses self-play methodology and distributed computing infrastructure to train competitive agents, with pre-trained models available for Zerg versus Zerg matchups. It includes testing utilities allowing users to play against the trained agent.