executorch_flutter library

ExecuTorch Flutter Plugin - On-device ML inference with ExecuTorch

This package provides Flutter developers with the ability to run ExecuTorch machine learning models on Android, iOS, and macOS platforms with high performance and low latency.

Key Features

• High Performance: Optimized for mobile inference with ExecuTorch runtime
• Type Safe: Generated platform communication with Pigeon ensures type safety
• Cross Platform: Identical APIs across Android, iOS, and macOS platforms
• User-Controlled Resources: Explicit model lifecycle management with load/dispose
• Easy Integration: Simple API for loading models and running inference

Quick Start

import 'package:executorch_flutter/executorch_flutter.dart';

// Initialize the manager
await ExecutorchManager.instance.initialize();

// Load a model
final model = await ExecutorchManager.instance.loadModel('path/to/model.pte');

// Prepare input data
final inputTensor = TensorData(
  shape: [1, 3, 224, 224],
  dataType: TensorType.float32,
  data: imageBytes,
  name: 'input',
);

// Run inference
final outputs = await model.forward([inputTensor]);

// Process outputs (List<TensorData>)
for (var output in outputs) {
  print('Output shape: ${output.shape}');
}

// Clean up
await model.dispose();

Main Classes

• ExecutorchManager: Main entry point for ExecuTorch operations
• ExecuTorchModel: Represents a loaded model instance
• TensorData: Tensor data representation
• ModelMetadata: Model information and specifications

Processors

• ExecuTorchPreprocessor: Base class for input preprocessing
• ExecuTorchPostprocessor: Base class for output postprocessing
• ExecuTorchProcessor: Combined preprocessing and postprocessing
• ImageClassificationProcessor: Ready-to-use image classification
• TextClassificationProcessor: Ready-to-use text classification
• AudioClassificationProcessor: Ready-to-use audio classification

Platform Support

• Android: API 23+ (Android 6.0+), arm64-v8a architecture
• iOS: iOS 13.0+, arm64 (device only, simulator not supported)
• macOS: macOS 12.0+ (Monterey), arm64 only (Apple Silicon)

For detailed documentation and examples, see the individual class documentation.