margelo/react-native-fast-tflite

What it solves

react-native-fast-tflite is a high-performance library that allows React Native developers to run TensorFlow Lite (.tflite) models on mobile devices. It bridges the gap between JavaScript and the low-level C/C++ TensorFlow Lite core API, enabling efficient on-device AI inference without the overhead of traditional React Native bridges.

How it works

The library uses Nitro Modules to achieve zero-copy ArrayBuffers, meaning data is passed between the native side and JavaScript without expensive copying. It interacts directly with the TensorFlow Lite core API for direct memory access. Models can be loaded from the app bundle, local files, or remote URLs, and can be swapped at runtime.

Who it’s for

Mobile developers using React Native who need to integrate machine learning models for tasks like image classification, object detection, or other tensor-based predictions on iOS and Android.

Highlights

• High Performance: Powered by Nitro Modules for zero-copy data transfer.
• GPU Acceleration: Supports GPU-accelerated delegates including CoreML (iOS), Metal, OpenGL, and Android GPU/NNAPI.
• Flexible Model Loading: Load models via require, local URIs, or remote URLs.
• VisionCamera Integration: Designed for easy integration with react-native-vision-camera for real-time AI vision tasks.