inference 1.0.0 
inference: ^1.0.0 copied to clipboard
Published 2 months ago β’ 
shankarkakumani.dev
shankarkakumani.devZero-setup ML inference for Flutter using Rust engines (Candle, Linfa). Load PyTorch, train on-device with unified API. Cross-platform support.
1.0.0 #
- π First stable release of the Inference Flutter package
- Core API Implementation: Complete implementation of the unified inference API
- β
InferenceSession.load()with automatic engine detection - β
InferenceSession.loadFromUrl()with smart caching - β
InferenceSession.loadFromHuggingFace()for direct Hub integration - β
InferenceSession.trainLinfa()for on-device training
- β
- Model Architecture Support:
- β BERT Models: Complete wrapper for text classification and NLP
- β ResNet Models: Complete wrapper for image classification
- β Generic SafeTensors: Support for any SafeTensors model format
- Engine Integration:
- β Candle Engine: Full PyTorch model support with GPU acceleration
- β Linfa Engine: On-device training with K-means clustering
- β Automatic Detection: Smart engine selection based on model format
- Input/Output System:
- β
ImageInput,NLPInput,TensorInput,AudioInputclasses - β
InferenceResultwith convenience accessors (scalar,vector,matrix) - β
Classification helpers (
argmax,topK())
- β
- HuggingFace Integration:
- β Direct model loading from HuggingFace Hub
- β Automatic URL construction and caching
- β Support for specific revisions and filenames
- Performance Features:
- β Automatic GPU acceleration when available
- β
Batch processing with
predictBatch() - β Smart model caching with size management
- β
Resource management with
dispose()methods
- Cross-Platform Support: Verified on Android, iOS, Windows, macOS, Linux
- Documentation: Comprehensive README with real-world examples
- Future Roadmap: Model Wrappers Roadmap with 20+ additional architectures planned
0.1.0-beta.4 #
- Enhanced examples and updated dependencies for better pub.flutter-io.cn experience
- Added comprehensive standalone example.dart showcasing core functionality
- Created detailed example README.md following Dart package layout conventions
- Updated dependencies to latest versions with proper version constraints
- Improved pub.flutter-io.cn score with comprehensive documentation and code quality fixes
- Added detailed library documentation with examples and feature overview
- Documented all missing constructors and classes for 100% API coverage
- Fixed static analysis issues including unused imports and style improvements
- Added package topics for better discoverability
- Enhanced code formatting and consistency
0.1.0-beta.2 #
- Bug fixes and improvements for the second beta release
- Enhanced stability and performance optimizations
- Improved error handling and documentation
- Updated dependencies and build configurations
0.1.0-beta.1 #
- Initial beta release of the Inference Flutter package
- Zero-setup ML inference with unified API for Candle and Linfa engines
- Support for automatic model format detection (.safetensors, .pt, .pth)
- Cross-platform support (Android, iOS, Windows, macOS, Linux)
- Multiple input types: ImageInput, NLPInput, TensorInput, AudioInput
- Comprehensive example app with image classification, text sentiment, and on-device training
- Built with Flutter Rust Bridge 2.0 for optimal performance
- GPU acceleration support where available
0.0.1 #
- TODO: Describe initial release.
Metadata
9
likes
150
points
39
downloads
Publisher
Weekly Downloads
Metadata
Zero-setup ML inference for Flutter using Rust engines (Candle, Linfa). Load PyTorch, train on-device with unified API. Cross-platform support.
Repository (GitHub)
View/report issues
Topics
#machine-learning #artificial-intelligence #pytorch #rust
Documentation
License
MIT (license)
Dependencies
ffi, flutter, flutter_rust_bridge, freezed_annotation, image, meta, plugin_platform_interface
More
Packages that depend on inference
Packages that implement inference
