sotetsuk/pgx
A collection of vectorized game simulators built in JAX for training reinforcement learning agents at scale.

Velocity · 7d
+0.4
★ / day
Trend
→steady
star history
PGX provides GPU-accelerated parallel game environments (Go, Chess, Shogi, Backgammon, Poker, etc.) implemented in JAX. It vectorizes multiple game instances to enable efficient batched simulation, a key requirement for training RL agents. The library is designed to support research workflows like AlphaZero, where fast environment rollout and batched state processing are essential.