This repository hosts the Android-ready single-file INT8 ONNX artifact used by
the azani_w_qiblati offline Quran recitation and tajweed feedback feature.
The source artifact was already dynamically quantized to INT8. This Android
package only embeds the external ONNX initializer data into one .onnx file:
1onnx.load(load_external_data=True)
2 + convert_model_from_external_data
3 + onnx.save_model(save_as_external_data=False)
It was not requantized or graph-optimized. The mathematical graph is unchanged.
The consuming application verifies both the exact byte count and SHA-256 before
promoting a completed download. Interrupted downloads can resume using HTTP
Range requests.
This artifact is intended for on-device ONNX Runtime inference on Android. It
does not include a hosted inference API. The surrounding application supplies
the pinned preprocessing, vocabularies, decoding, Quran reference text, and
tajweed feedback logic.
All model credit belongs to the Quran Muaalem authors and contributors. This
repository provides a deployment-oriented repackaging of their MIT-licensed
model for offline Android use.