This is an implementation of MiniGo using Swift for TensorFlow. The implementation is adapted from the MLPerf reference model.
Note: Due to temporary limitations, we only support inference (self-play) for now. We hope to soon lift these restrictions.
Parameters like "board size" and "simulations per move" are defined as constants at the top of
Sources/MiniGo/main.swift
and can be modified.
Building MiniGo requires a Swift for TensorFlow toolchain. To get a toolchain, you can:
# Run inference (self-plays).
cd swift-models
swift run -c release MiniGoDemo