TurboRx/Chronoquazicalar

Chronos (Gen 9 Random Battles)

Autonomous reinforcement learning policy and value network (~9.65M parameters) for Pokémon Showdown Gen 9 Random Battles.

• GitHub Repository: TurboRx/chronos-randbats
• Engine: Tensor-native JAX battle simulator with hardware-accelerated rollouts
• Framework: JAX / Flax / Optax

Architecture

• Model: Non-Causal Transformer Encoder (~9.65M parameters)
• Embedding Dimension (d_model): 256
• Multi-Head Attention: 8 heads
• Encoder Depth: 6 layers
• Feed-Forward Dimension (d_ff): 1024
• Heads:
  • Policy Head (9 actions: 4 moves, 5 switches)
  • Value Head (Scalar valuation in [-1, 1])
  • Opponent Prediction Head (Simultaneous action prediction)

Checkpoints

• checkpoint_latest.pkl: Model parameters trained via PPO self-play and benchmark sparring.
• metrics.json: Latest training metrics, update count, and turn counters.