Google AI Edge LiteRT Samples

This repository contains official sample applications and code examples for LiteRT (formerly known as TensorFlow Lite), Google's high-performance on-device machine learning framework.

The samples are organized into two main versions (interpreter_api/ and compiled_model_api/) to demonstrate different API paradigms.

Note: For Generative AI and Large Language Models (LLMs), please refer to the LiteRT-LM repository.

🔥 What's New

New sample app for Automatic Speech Recognition(ASR) has been added.

📂 Repository Structure

1. compiled_model_api/

This folder contains samples using the LiteRT CompiledModel API. This new API is designed for advanced GPU/NPU acceleration, delivering superior ML & GenAI performance.

• Key Features:
  • Hardware Acceleration: Specialized for GPU/NPU execution.
  • Async Execution: Improved performance for complex pipelines.
  • Buffer Management: efficient input/output handling.
• Available Samples:
  • NPU AOT: Ahead-of-Time compilation examples.
  • NPU JIT: Just-in-Time compilation examples.
• Platforms: Primarily Android (Kotlin/C++).

2. interpreter_api/

This folder contains the CPU samples that use the Interpreter API.

• Key Features:
  • Standard .tflite model execution.
  • Broad compatibility across all Android/iOS versions.
  • Legacy Task Library usage.
• Available Samples:
  • Image Classification: Recognize objects in images/video.
  • Object Detection: Locate and label multiple objects.
  • Image Segmentation: Separate objects from the background.
  • Audio Classification: Identify audio events.
  • Digit Classification: Handwritten digit recognition (MNIST).
• Platforms: Android (Kotlin/Java), iOS (Swift/Objective-C), Python (Raspberry Pi/Linux).

🛠️ Getting Started

Prerequisites

• Android: Android Studio (latest stable version).
• iOS: Xcode (latest version).
• Python: Python 3.9+ and pip install ai-edge-litert.

Running a Sample

For Samples Using Compiled Model API

1. Navigate to the compiled_model_api/ directory.
2. Ensure you have a device with a supported NPU (e.g., modern Pixel, Samsung, or devices with MediaTek/Qualcomm chips).
3. Follow the specific setup instructions in the sub-folder to enable the specialized hardware delegates.

For Samples Using Interpreter API

1. Navigate to interpreter_api/ directory.
2. Open the project in Android Studio or Xcode.
3. Build and run on your device.

📚 Documentation

• LiteRT Overview: ai.google.dev/edge/litert
• CompiledModel API Guide: LiteRT for Android
• Model Conversion: Convert models to LiteRT

🤝 Contributing

Contributions are welcome!

1. Read CONTRIBUTING.md.
2. Fork the repo and create a branch.
3. Submit a Pull Request.

📄 License

Apache License 2.0. See LICENSE for details.

Disclaimer: This is a sample repository maintained by Google. It is provided "as is" without warranty of any kind.