Licence and notices for the three ONNX models bundled with rapidocr 3.9.2

Originally posted by alexprotonotarios September 6, 2026 Hello,

I am preparing a separately installed local OCR dependency for WaveParse, an ECG digitisation library. We pin the rapidocr==3.9.2 wheel and verify the three bundled ONNX files against these SHA-256 identities:

File	SHA-256
PP-OCRv6_det_small.onnx	090f04abcd9d9a7498bc4ebf677e4cb9bdce1fe4197ddb7e529f1ef44e1ff94f
PP-OCRv6_rec_small.onnx	6f327246b50388f3c176ae304bd95767ea6dc0c9ae92153ef8cbe210b3c14884
ch_ppocr_mobile_v2.0_cls_mobile.onnx	e47acedf663230f8863ff1ab0e64dd2d82b838fceb5957146dab185a89d6215c

These identities match the ONNX entries in default_models.yaml for v3.9.2. Your README identifies Baidu as the model copyright holder and declares the project Apache-2.0.

Could you point us to the licence and required notices that cover these exact converted models? In particular, does Apache-2.0 cover their commercial inference use and redistribution, including any conversion-related rights, or are there additional model-specific terms?

Users currently install the unmodified upstream wheel explicitly during runtime setup. Our npm/Python interface packages do not contain these models, and we do not mirror them. We retain the upstream dependency notices. We would like to document the applicable terms accurately for downstream users before release.

Thank you,

Alexandros Protonotarios