At this point, bullet is a general-purpose ML library (it can be used as such) with a heavily limited number of supported operations. However, it is generally used for training NNUE-style networks for some of the strongest chess engines in the world.

• Autograd
  • Uses my diffable crate
• CUDA and HIP backends
  • Makes heavy use of (cu/hip)BLAS wherever possible
  • A number of custom kernels
• Lots of NNUE and chess engine specific tooling
  • Input feature types
  • Output buckets
  • Data formats
  • Utilities

Before attempting to use, check out the docs. They contain all the main information about building bullet, managing training data and the network output format.

Most people simply clone the repo and edit one of the examples to their taste.

Alternatively, import the crate with

bullet = { git = "https://github.com/jw1912/bullet" }

Specific API documentation is covered by Rust's docstrings.

Please open an issue to file any bug reports/feature requests. For general questions about bullet - e.g. what your training schedule should look like - you can go to the dedicated channel in the Engine Programming discord server.