Arabic and Sanskrit research models

Native model components trained for Language Engine:

• Arabic clitic tokenizer and multiword expander, trained from corrected CAMeL teacher annotations over authentic news text.
• Sanskrit tokenizer and sandhi expander, trained from DCS surface/underlying-word supervision.
• Sanskrit interpretive NER, with 18 semantic categories and Gemini-assisted annotations of authentic Sanskrit documents.

Each ZIP contains the trained weights, CPU loader, tested requirements, model card, evaluation record and SHA-256 artifact manifest.

Model cards · Complete selected-component scores · Hugging Face profile

The original model weights and accompanying code use the Conrad Compagna Model Evaluation License 1.0: research, education, experimentation and evaluation are permitted; commercial deployment and redistribution of those weights require permission. Third-party assets retain their own terms, detailed in each package.

Conrad Compagna · conradcompagna@gmail.com