0.1.03MIT
strict
core24
Gomoku/Renju AI engine with neural network and MCTS
Z2Irs is a Gomoku/Renju AI engine based on a neural network and Monte Carlo
Tree Search (MCTS), implementing the Gomocup/Piskvork console protocol
(START/BEGIN/TURN/BOARD/INFO/ABOUT/END) over stdin/stdout. It bundles
ONNX Runtime inference and supports FreeStyle, Standard, Renju, Caro and
Standard+Caro rule flags.
The snap ships the engine binary only; no weights are bundled.
Configuration:
- On startup the engine looks for config.toml in the user config directory
(~/snap/z2i-rs/current/.config/Z2Irs/), then in the current working
directory, then next to the executable.
- Point the [model] entries at your own ONNX weights (defaultmodel,
freestylemodel, renjumodel, standardmodel, caromodel,
standardcaromodel). Absolute paths are used as-is; relative paths are
resolved against the working directory first, then the executable
directory.
- [MCTS] controls nummctsims, numsimperbatch, openmind and
enableponder; [ONNXRUNTIME] controls numintra_thread.
- A template covering all supported keys is available in the source
repository at snap/config.toml.
Notes:
- The neural network input tensor is fixed at 3x15x15; use weights exported
for a 15x15 board.
- Without a valid model the engine still plays, using plain MCTS (weaker).
Use it with a Gomoku manager such as qpiskvork.
Tree Search (MCTS), implementing the Gomocup/Piskvork console protocol
(START/BEGIN/TURN/BOARD/INFO/ABOUT/END) over stdin/stdout. It bundles
ONNX Runtime inference and supports FreeStyle, Standard, Renju, Caro and
Standard+Caro rule flags.
The snap ships the engine binary only; no weights are bundled.
Configuration:
- On startup the engine looks for config.toml in the user config directory
(~/snap/z2i-rs/current/.config/Z2Irs/), then in the current working
directory, then next to the executable.
- Point the [model] entries at your own ONNX weights (defaultmodel,
freestylemodel, renjumodel, standardmodel, caromodel,
standardcaromodel). Absolute paths are used as-is; relative paths are
resolved against the working directory first, then the executable
directory.
- [MCTS] controls nummctsims, numsimperbatch, openmind and
enableponder; [ONNXRUNTIME] controls numintra_thread.
- A template covering all supported keys is available in the source
repository at snap/config.toml.
Notes:
- The neural network input tensor is fixed at 3x15x15; use weights exported
for a 15x15 board.
- Without a valid model the engine still plays, using plain MCTS (weaker).
Use it with a Gomoku manager such as qpiskvork.
Update History
0.1.0 (3)30 Aug 2026, 09:45 UTC
30 Aug 2026, 09:11 UTC
30 Aug 2026, 09:25 UTC
30 Aug 2026, 09:45 UTC