jromanghf/gaiachess-networks

GaiaChess NNUE networks

Neural networks (NNUE) for the GaiaChess UCI chess engine.

Architecture (GaiaNet): 12 king buckets × 768 PST + filtered threat features (~41K inputs) → 640 → CReLU + pairwise → 16 → 32 → 1, with 8 output buckets. Trained with Bullet on self-play data.

Files are append-only: a network file, once published, is never modified — engine releases pin an exact file name.

Networks

File	Engine version	Notes
gaianet-t1-1000.bin	never shipped	archived for reproducibility: this network generated part of the training data used by its successors
gaianet-t2-1000.bin	4.2.0 – 4.2.5	previous default network
gaianet-t3b-1000.bin	unreleased	+12.0 ±9.5 Elo vs t2 (SPRT, 2588 games)
gaianet-t3e-1100.bin	unreleased	t3b fine-tuned for 100 more superbatches at lr/10
gaianet-t3e-1100b.bin	unreleased (default in main)	current default network: same weights as t3e-1100 plus a 16-byte integrity footer (architecture + content hashes, verified by the engine at load; older engine binaries ignore it)

Usage

Networks are embedded into the engine at build time:

curl -L -o nets/gaianet-t3e-1100b.bin \
  "https://huggingface.co/jromanghf/gaiachess-networks/resolve/main/gaianet-t3e-1100b.bin"
MODEL=nets/gaianet-t3e-1100b.bin RUSTFLAGS="-C target-cpu=native" \
  cargo build --release --features nnue

Or loaded at runtime through UCI: setoption name EvalFile value <path>.

License

GPL-3.0-or-later, same as the engine.