Bug Fixes #

• iOS/macOS: Fixed Swift 6 compilation errors in ExecutorchModelManager.swift
  • Added try keyword to withUnsafeBytes calls for Swift 6 compatibility
  • Resolves build failures on Xcode 16+ with Swift 6 language mode
  • Affects tensor conversion for Float32, Int32, and UInt8 data types

Initial release of ExecuTorch Flutter plugin.

Features #

• ✅ Cross-platform support for Android, iOS, and macOS
• ✅ Type-safe Pigeon-generated API for platform communication
• ✅ Async model loading and inference execution
• ✅ Multiple concurrent model instances support
• ✅ Memory-efficient tensor operations
• ✅ Structured error handling with clear exceptions
• ✅ Backend support: XNNPACK, CoreML, MPS
• ✅ Actor-based concurrency (iOS/macOS) and coroutines (Android)
• ✅ Example app with image classification and object detection demos
• ✅ High-level processor interfaces for common ML tasks

Platform Support #

• Android: API 23+ (Android 6.0+), arm64-v8a
• iOS: iOS 13.0+, arm64 (device only)
• macOS: macOS 12.0+ (Monterey), arm64 only (Apple Silicon)

Known Limitations #

• iOS Simulator (x86_64): Not supported
• macOS Intel (x86_64): Not supported
• macOS Release Builds: Not working (tracking: Flutter Issue #176605)

Documentation #

• Comprehensive README with usage examples
• Model export guide for converting PyTorch models
• Contributing guide for contributors
• Roadmap for future features