SentinelNG Models
This repository contains model assets used by the SentinelNG Android application for local crop, health, and language assistance. It is a model-artifact repository rather than a complete application.
Contents
| File | Format | Intended role |
|---|
crop_doctor.tflite | TensorFlow Lite | On-device crop or plant-health inference. |
health_scan.tflite | TensorFlow Lite | On-device health-scan inference. |
nigerian_nlu.ftz | FastText-style model artifact | Nigerian-language or intent/NLU support. |
bonsai_1.7b_q1_0.gguf | GGUF | Compact local language-model asset for constrained devices. |
The binary files are large and are stored through Hugging Face LFS/Xet. Download only the assets required by the target application. The corresponding Android integration is in the SentinelNG-App GitHub repository, where TFLiteHelper.kt defines the expected model names and inference lifecycle.
Usage guidance
Before using an asset in a new application, verify its input shape, label order, preprocessing, output interpretation, and license. A model file alone does not define a safe diagnosis workflow. Crop and health predictions are assistive signals and must not replace qualified professional advice.
Reproducibility
Record the training data revision, preprocessing pipeline, label map, evaluation metrics, and conversion/quantization settings whenever these models are replaced. Keep the model filename stable only when the input/output contract remains compatible.
