https://github.com/jw1912/bullet

Specialised ML Library
https://github.com/jw1912/bullet

Last synced: 3 months ago
JSON representation

Specialised ML Library

• Host: GitHub
• URL: https://github.com/jw1912/bullet
• Owner: jw1912
• License: mit
• Created: 2023-08-12T22:29:46.000Z (over 2 years ago)
• Default Branch: main
• Last Pushed: 2025-08-24T19:20:44.000Z (3 months ago)
• Last Synced: 2025-08-24T23:23:57.164Z (3 months ago)
• Language: Rust
• Homepage:
• Size: 2.8 MB
• Stars: 132
• Watchers: 6
• Forks: 39
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-rust-neural-network - bullet - specific, NNUE-style networks, chess engines | cudarc | - | (Machine Learning)

README

          


# bullet



A domain-specific ML library, generally used for training NNUE-style networks for many of the strongest chess engines in the world

due to its best-in-class performance, chess-specific tooling and ease of use.

### Usage for NNUE/Value Network Training

Before attempting to use, check out the [docs](docs/0-contents.md).

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](/examples) to their taste.

If you want to create your own example file to ease pulling from upstream, you need to add the example to [`bullet_lib`'s `Cargo.toml`](crates/bullet_lib/Cargo.toml).

Alternatively, import the `bullet_lib` crate with

```toml

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

```

Specific API documentation is covered by Rust's docstrings. You can create local documentations with `cargo doc`.

### Constituent Crates

- **bullet_core**

    - Core ML framework

    - Graphs are defined once (ahead of use), then optimsed and compiled for a given backend device

    - A token single-threaded CPU backend is included for verifying correctness of the crate and other backend implementations

    - See the [MNIST](examples/extra/mnist.rs) example for using `bullet_core` as a general-purpose ML framework

- **bullet_cuda_backend**

- **bullet_hip_backend**

- **bullet_lib**

    - Provides a high-level wrapper around the above crates specifically for training networks to do with chess (and other games e.g. Ataxx) easily

    - Value network training for games with `ValueTrainer`

        - The [simple](examples/simple.rs) example shows ease-of-use in training the simplest NNUE architectures

        - The [progression](examples/progression) examples show how to incrementally improve your NNUE architecture

        - Read the [documentation](docs/2-getting-started.md#backends) for compilation instructions

- **bullet-utils**

    - Various utilities mostly to do with handling chess data

### Help/Feedback

- Please open an issue to file any bug reports/feature requests.

- Feel free to use the dedicated `#bullet` channel in the [Engine Programming](https://discord.com/invite/F6W6mMsTGN) discord server if you run into any issues.

- For general training discussion the Engine Programming non-`#bullet` channels are appropriate, or `#engines-dev` in the [Stockfish](https://discord.gg/GWDRS3kU6R) discord.