@naveen521kk/expo-mediapipe-text-tasks v0.1.0

expo-mediapipe-text-tasks

For now, this module exposes a single function called embed which would create a text embedding for the given text. The function is a wrapper around the mediapipe library.

The function signature is as follows:

export declare function embed(
  delegate: number,
  modelPath: string,
  text: string
): Promise<{
  string: string;
  embeddings: number[];
  inferenceTime: number;
}>;

The delegate parameter is an integer that specifies whether to use the CPU or GPU for inference. The value 0 specifies the CPU, and the value 1 specifies the GPU. You can also use the constants DELEGATE_CPU and DELEGATE_GPU from the expo-mediapipe-text-tasks module.

The modelPath parameter is a string that specifies the path to the model file. The model file should be a .tflite file. You can use the expo-asset module to load the model file.

Below are some models from the official MediaPipe repository that you can use:

• MobileBert: https://storage.googleapis.com/mediapipe-models/text_embedder/bert_embedder/float32/1/bert_embedder.tflite
• AverageWord: https://storage.googleapis.com/mediapipe-models/text_embedder/average_word_embedder/float32/1/average_word_embedder.tflite

The text parameter is a string that specifies the text for which you want to create an embedding.

The function returns a promise that resolves to an object with the following properties:

• string: The .toString() representation of the output.
• inferenceTime: The time taken for inference in milliseconds.
• embeddings: The embedding of the input text.

See example/ for a complete example.

Installation in managed Expo projects

For managed Expo projects, install the npm package and start using.

Note that you would need a development client while developing and will not work with Expo Go.

Installation in bare React Native projects

For bare React Native projects, you must ensure that you have installed and configured the expo package before continuing.

Installing the npm package

npm install @naveen521kk/expo-mediapipe-text-tasks