Arabic Diacritizers (tashkeel)
Collection
ONNX Arabic diacritization models used by text2tashkeel: the rawi family plus mirrored bilstm and libtashkeel. • 9 items • Updated
libtashkeel — ONNX mirror
ONNX of the char+hint encoder Arabic diacritizer by Musharraf Omer
(mush42/libtashkeel, MIT). Mirrored for
use by text2tashkeel.
| file | what |
|---|---|
libtashkeel.onnx |
the char+hint encoder model |
libtashkeel.maps.json |
input / hint / target maps |
All credit for the model and weights goes to Musharraf Omer; redistributed here under the upstream MIT license.
int8
libtashkeel.int8.onnx is a dynamic-quantized variant (4.0 MB vs 4.8 MB). It diverges from the fp32 model on ~10 % of outputs, so prefer the fp32 libtashkeel.onnx unless size is critical.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support