flutter_gemma 0.11.3 copy "flutter_gemma: ^0.11.3" to clipboard
flutter_gemma: ^0.11.3 copied to clipboard

Published 4 days ago β€’ verified publishermobilepeople.dev
SDKFlutter
PlatformAndroidiOSweb
β†’

Metadata

The plugin allows running the Gemma AI model locally on a device from a Flutter application. Includes support for Gemma 3 Nano models with optimized MediaPipe GenAI v0.10.24.

More...

0.11.3 #

  • 🌐 Web Multimodal Support: Added full multimodal image processing support for web platform
  • πŸ“š MediaPipe 0.10.25: Updated to MediaPipe GenAI v0.10.25 for web compatibility
  • πŸ“¦ LiterTLM Format Support: Added support for .litertlm model files optimized for web platform

0.11.2 #

  • πŸ›‘οΈ Fixed: Updated ProGuard rules for Android release build compatibility

0.11.1 #

  • πŸ› Fixed: Export missing ModelType and other public API types to resolve import issues

0.11.0 #

  • πŸš€ Embedding Models Support: Added full support for text embedding models
  • πŸ”§ Unified Model System: All models (inference and embedding) now use the same download and management pipeline
  • πŸ“ ModelSpec Architecture: Introduced InferenceModelSpec and EmbeddingModelSpec for better model organization
  • πŸ›‘οΈ Smart Cleanup System: Added automatic cleanup of orphaned files with resume detection capabilities
  • πŸ”„ Model Replace Policies: Separate policies for model downloading replace and keep
  • πŸ“± Example App Integration: Added embedding models download screen and embeddings generation demo screen

0.10.6 #

  • πŸ”§ Model Replace Policy: Added configurable model replacement system with keep/replace options and ensureModelReady() method
  • πŸ“₯ Enhanced Downloads: Added HuggingFace CDN ETag issue handler with smart retry logic and exponential backoff
  • πŸ”„ Download Reliability: New HuggingFaceDownloader wrapper to handle CDN server inconsistencies and resume failures
  • πŸ“ ModelFileType System: Introduced distinction between .task files (MediaPipe-handled) and .bin/.tflite files (manual formatting)
  • πŸ” Android Security: Added network security configuration for HuggingFace CDN access with proper permissions
  • πŸ› Download Fixes: Fixed Android download timeouts, stream management, and ETag mismatch issues
  • πŸ–ΌοΈ Image Corruption Fix: Added comprehensive image processing system to prevent AI model corruption on Android
  • πŸ”„ Example App: Added sync/async response method selection in chat interface

0.10.5 #

  • πŸ›‘ Stop Generation: Added Android support for stopping text generation mid-process with session.cancelRequestGenerationAsync() (#89, #19, #34)
  • πŸ› Screen Close Fix: Fixed crash when closing screen during active generation by implementing proper StreamSubscription cleanup (#89)
  • πŸ› Model Loading Fix: Fixed model install check with partial downloads by adding orphaned files cleanup (#84)
  • πŸ—‘οΈ File Management: Added automatic cleanup of corrupted/incomplete model files with atomic SharedPrefs updates
  • πŸ“± iOS Requirements: Updated deployment target to 16.0 for MediaPipe GenAI compatibility
  • πŸ”§ Error Handling: Improved error recovery with automatic file cleanup on failed downloads
  • πŸ“š Documentation: Updated model capabilities table and comprehensive usage examples

0.10.4 #

  • πŸ“š Documentation: Updated README with comprehensive model information and usage examples

0.10.3 #

  • πŸ“₯ Background Downloads: Added background download support for model files

0.10.2 #

  • πŸš€ New Models: Added support for 4 new compact models:
  • βš™οΈ Backend Switching: Added individual CPU/GPU backend switcher for each model in the example app
  • πŸ” Advanced Model Filtering: Added expandable filter section with feature-based filtering:
    • Filter by Multimodal support (vision capabilities)
    • Filter by Function Calling support
    • Filter by Thinking Mode support
    • Animated expandable UI with clear visual indicators
  • πŸ“Š Model Sorting: Added comprehensive sorting options:
    • Sort alphabetically (A-Z)
    • Sort by model size (smallest to largest)
    • Default order (Gemma models prioritized)
  • πŸ“ Improved Model Display: Separated size information from model names for cleaner presentation
  • 🌐 Full English Localization: Converted all UI text from Russian to English
  • πŸ“Š Results Counter: Added dynamic counter showing filtered results ("Showing X models")
  • 🎨 Enhanced Filter UI: FilterChip components with color-coded selections matching feature badges
  • 🎯 Model Organization: Reorganized model list with Gemma models prioritized at the top
  • πŸ› οΈ Function Calling: Enhanced function calling support with Hammer 2.1 action model
  • πŸ“± UI Improvements: New card-based design with individual backend controls per model
  • βœ… Model Capabilities: Fixed and verified multimodal support flags for all models
  • πŸ“š Documentation: Updated README and model feature support table with new models

0.10.1 #

  • 🧠 Thinking Mode: Added thinking mode support for DeepSeek models with persistent thinking bubbles
  • πŸ”§ Function Call Fixes: Fixed function calls detection in the middle of responses
  • πŸ’¬ UI Improvements: Fixed loading indicator display after function calls
  • πŸ”„ JSON Response Handling: Enhanced handling of JSON responses after function execution
  • πŸ“š Documentation: Updated README with latest API changes and improved examples
  • 🎨 Code Quality: Removed Russian comments and improved code consistency

0.10.0 #

  • ✨ Function Calling: Added support for function calling, allowing models to interact with external tools.
  • πŸ”§ Improved Chat API: Enhanced the chat API to support function calls and tool responses.

0.9.0 #

  • πŸ–ΌοΈ MULTIMODAL SUPPORT: Added full support for text + image input with Gemma 3 Nano vision models
  • 🎯 Enhanced Message API: New Message class with support for text, image, and multimodal content
    • Message.text() - for text-only messages
    • Message.withImage() - for text + image messages
    • Message.imageOnly() - for image-only messages
    • message.hasImage - to check if message contains image
  • πŸ“± Vision Models: Full support for Gemma 3n E2B and E4B models with image understanding
  • 🌐 Web Platform: Added graceful degradation with debug warnings for unsupported features

0.8.6 #

  • πŸš€ GEMMA 3 NANO SUPPORT: Added full support for Gemma 3 Nano models
  • ⚑ Optimized session parameters for Gemma 3n models (temperature: 0.6, topK: 40, topP: 0.9)
  • πŸ›‘οΈ Added automatic fallback session creation for input_pos != nullptr errors
  • 🎯 Added Gemma 3n model detection and compatibility handling
  • πŸ’ͺ Enhanced error handling for TensorFlow Lite model initialization
  • πŸ”§ Fixed iOS session initialization with proper input position handling
  • πŸ“± Improved mobile inference model with optimized parameters

0.8.5 #

  • Upgraded Mediapipe to 0.10.24 for iOS and Android
  • Added support of Gemma3, Phi-4 and DeepSeek models for iOS

0.8.4 #

  • iOS LoRA support added
  • iOS topP support added

0.8.3 #

  • Readme updated

0.8.2 #

  • Readme updated

0.8.1 #

  • Migrate to js-interop
  • Add web platform support in pubspec.yaml

0.8.0 #

  • Upgraded Mediapipe to 0.10.22 for Android and Web
  • Upgraded Mediapipe to 0.10.21 for iOS
  • Added opportunity to set topP and preferredBackend for inference
  • Added support of Gemma3, Phi-4 and DeepSeek models for Android and Web

0.7.0 #

  • Added Chat functionality for instruction tuned model
  • Added sizeInTokens method for analysis of the size of the input prompt

0.6.0 #

  • Added opportunity to manage inference session

0.5.1 #

  • Fixed crash on generation for Android

0.5.0 #

  • IMPORTANT: Breaking changes in the API
  • FlutterGemma instance was replaced with ModelManager and InferenceModel
  • ModelManager to manage models and LoRA weights
  • InferenceModel to manage inference
  • Added opportunity to set model and LoRA weights paths manually
  • Added opportunity to delete model and LoRA weights
  • Added opportunity to load LoRA weights from assets and network

0.4.6 #

  • Added close method

0.4.5 #

  • Small fixes for Android

0.4.4 #

  • Small fixes for iOS

0.4.3 #

  • Upgraded Mediapipe to 0.10.20
  • Updated LoRA support

0.4.2 #

  • Added error handling
  • Updated example for error handling
  • Upgraded Mediapipe to 0.10.18 for iOS
  • Fixed ios issue with model freezing

0.4.1 #

  • Fixed ios issue

0.4.0 #

  • Upgraded Mediapipe to 0.10.16
  • Added LoRA support
  • Fixed some issues

0.3.1 #

  • Updated example and readme

0.3.0 #

  • Added support for loading models from assets and network.
  • Added progress updates for model loading.

0.2.4 #

  • Fixed Mediapipe ios version

0.2.3 #

  • Updated Mediapipe ios version

0.2.2 #

  • Added opportunity to configure folder (Android only)
  • Fixed android release issue
  • Updated Mediapipe for Android
  • Updated readme

0.2.1 #

  • Updated chat functionality for instruction tuned model
  • Updated readme

0.2.0 #

  • Added chat functionality for instruction tuned model
  • Updated example

0.1.4 #

  • Updated readme for GPU models on Android devices
  • Updated example

0.1.3 #

  • Updated readme

0.1.2 #

  • Updated example

0.1.1 #

  • Updated example

0.1.0 #

  • Added web support
  • Added opportunity to set randomSeed, topK and temperature

0.0.4 #

  • Updated example
  • Updated readme

0.0.3 #

  • Added getResponseAsync method

0.0.2 #

  • Added description in Readme.md
  • Added opportunity to setup a model before initiation

0.0.1 #

  • Initial release

← Metadata

218
likes
150
points
4.15k
downloads

Publisher

verified publishermobilepeople.dev

Weekly Downloads

Metadata

The plugin allows running the Gemma AI model locally on a device from a Flutter application. Includes support for Gemma 3 Nano models with optimized MediaPipe GenAI v0.10.24.

Repository (GitHub)
View/report issues

Documentation

API reference

License

MIT (license)

Dependencies

background_downloader, flutter, flutter_web_plugins, large_file_handler, path, path_provider, plugin_platform_interface, shared_preferences

More

Packages that depend on flutter_gemma

Packages that implement flutter_gemma

Back