The TFLite Model Maker library simplifies the process of adapting and converting a TensorFlow neural-network model to particular input data when deploying this model for on-device ML applications.

• Refer to requirements.txt for dependent libraries that're needed to use the library and run the demo code.

There are two ways to install Model Maker.

• Install a prebuilt pip package: tflite-model-maker.

pip install tflite-model-maker

If you want to install nightly version tflite-model-maker-nightly, please follow the command:

pip install tflite-model-maker-nightly

• Clone the source code from GitHub and install.

git clone https://github.com/tensorflow/examples
cd examples/tensorflow_examples/lite/model_maker/pip_package
pip install -e .

For instance, it could have an end-to-end image classification example that utilizes this library with just 4 lines of code, each of which representing one step of the overall process. For more detail, you could refer to Colab for image classification.

• 1. Import the required modules.

from tflite_model_maker import image_classifier
from tflite_model_maker.image_classifier import DataLoader

• 2. Load input data specific to an on-device ML app.

data = DataLoader.from_folder('flower_photos/')

• 3. Customize the TensorFlow model.

model = image_classifier.create(data)

• 4. Evaluate the model.

loss, accuracy = model.evaluate()

• 5. Export to Tensorflow Lite model and label file in export_dir.

model.export(export_dir='/tmp/')

Currently, we support image classification, text classification and question answer tasks. Meanwhile, we provide demo code for each of them in demo folder.

• Overview for TensorFlow Lite Model Maker
• Colab for image classification
• Colab for text classification
• Colab for BERT question answer