How to build

1. mkdir build
2. cd ./build
3. cmake .. -DCMAKE_BUILD_TYPE=Release
4. cmake --build . -j --config Release

Tested on the lastest version (in 27 Sep 2025) of Windows, Ubuntu, and Raspberry Pi OS.

Trouble shooting

• If missing type specifier - int assumed. Note: C++ does not support default-int error occurs, refer to the solution at tensorflow/tensorflow#64376 (comment)
• If undefined reference to 'xnn_something' error occurs, add -DTFLITE_ENABLE_XNNPACK=OFF to the cmake configuration.

How to play

Locate the following three files at the same directory:

• renju executable (e.g. .exe)
• renju.tflite
• renju.html

Then,

1. Run the renju executable
2. Open localhost:8080 in the browser
3. Play

Note

• As of now, error handling is poor. The code contains minimal functionality.
• The AI is not 100% perfect. It could make mistakes, placing strange moves.
• The model defeated some online Gomoku programs :
  

How was it made

I used JAX ecosystem to create the AI model. The model is based on Alphazero. I made a big model first and distill it to a smaller one. Then, I converted it to tensorflow model. I'm going to upload the full training scripts later.