/carbon

Modified version of rice's trainer, used to train peacekeeper networks.

Primary LanguageC++

Carbon

Carbon is an open-source Neural Network (NN) trainer for chess engines, designed primarily to support Rice's neural networks , inspired by Grapheus, a previous NN trainer used to train Rice's neural networks.

Project Logo

Features

  • Learning Rate Schedulers: Utilize advanced learning rate scheduling techniques, including Step Decay and Cosine Annealing.
  • Optimizers: Implement various optimization algorithms like SGD, Adam, Adamax to your preferences.
  • Configuration Made Easy: Easily customize training parameters and model architecture to suit your needs.
  • Visualize Training: Monitor training progress and loss using the included lossplot.py script.

Getting Started

  1. Clone the Repository:

    git clone https://github.com/rafid-dev/carbon.git
cd carbon

  2. Building:

    make
./bin/CarbonTrainer

  3. Start Training: Follow the provided instructions to start training your neural networks and improving your chess engine's evaluation capabilities. (TODO)