onnx_flutter_example

This is a simple Flutter app that runs local emotion classification with bert-emotion using flutter_onnxruntime and dart_bert_tokenizer packages. You can find the article about the project and the conversion process here.

Convert bert-emotion to ONNX

1. Install Python dependencies:

   pip install torch transformers

2. Run the conversion script from the project root:

   python scripts/convert_bert_emotion_to_onnx.py

3. Ensure the output file exists:

   bert-emotion/bert-emotion.onnx
   

4. Run the setup script to copy the ONNX model and tokenizer files to the Flutter assets directory:

   python scripts/setup_assets.py

Run the app

flutter pub get
flutter run

The app loads the ONNX model on startup, takes text input, and predicts one of the 13 emotion labels.