hoduyquocbao/xiangqi-rim
0
๐ค Xiangqi-RIM 1 Billion FEN NNUE Model Weights (XRNN v1)
Binary model weights (nnue_weights_1b_latest.bin) trained on 1 BILLION FENs using 100% Native Rust Engine self-play evaluations and PyTorch CUDA acceleration.
๐ง Architecture Specification (XRNN v1)
- Feature Transformer: HalfKAv2_hm ($65,536 \times 256 \to 512 \to 32 \to 1$)
- File Size: 33,571,504 Bytes (32.02 MB)
- Format: Quantized Int16 / Int8 XRNN v1 Binary Matrix
โก Loading in Rust Engine
use xiangrust::eval::nnue::Nnue;
let nnue = Nnue::load("data/nnue_weights_1b_latest.bin").expect("Loaded NNUE weights");