ChessQueries: Toward Better Chess Board Recognition
Paper • 2608.30762 • Published
ChessQ Lite model weights
Resources: Demo · Paper [arXiv] · Code [GitHub] · Paper model weights [HuggingFace]
Release weights for ChessQ Lite V4, a smaller chessboard recognizer further trained on proprietary data.
Files
- cq-vit-s-644-v4-seed2-safetensors-fp16-r2.safetensors — ViT-S FP16 weights (63.15 MB).
The release manifest records the immutable download, SHA-256 and preprocessing. A validation summary and synthetic CUDA reference are also available. Browser and mobile exports are not yet available for this model.
Quick start
Download the weights above and check the SHA-256 in the manifest. With Python 3.12:
python -m pip install "git+https://github.com/JSeytre/chessqueries.git@f5aaaf38212cc7839b03ccf383455053905b3f7e"
chessqueries-predict photo.jpg --checkpoint cq-vit-s-644-v4-seed2-safetensors-fp16-r2.safetensors
CUDA uses FP16; CPU uses the same download with FP32 arithmetic.
Pass --checkpoint to select Lite; the CLI otherwise downloads paper ViT-L.
License
The ChessQ Lite model weights are licensed under the PolyForm Noncommercial License 1.0.0. They may be used, modified and redistributed for permitted noncommercial purposes; commercial use requires separate permission. Third-party components retain their upstream terms; see the notices.
Contact
For questions about licensing, reach out to 'contact "at" chessq.org'.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support