runevm_fl 0.3.1

Rune Bindings and Forge deployment for Flutter Containerization for TinyML/Mobile Applications on Tiny/Mobile devices.

runevm_fl

RuneVM plugin for Flutter

Example app is the game 2048 with voice control

Getting Started

Add the RuneVM plugin to your pubspec.yaml file

dependencies:
  flutter:
    sdk: flutter
  runevm_fl: ^0.3.0

Load and run your rune file

Load and run your rune file in three steps:

Deploy

Future<bool> RunevmFl.load(Uint8List runeBytes)

Read manifest

Future<dynamic> RunevmFl.manifest

Run rune with input bytes

Future<String> RunevmFl.runRune(Uint8List input)

Rune Bindings Implementation

Full implementation in main.dart

import 'package:runevm_fl/runevm_fl.dart';

class RunMyRune {

  double _input = 0;
  String? _output;

  Future<void> _loadRune() async {
    try {
      //Load Rune from assets into memory;
      ByteData bytes = await rootBundle.load('assets/sine.rune');
      bool loaded =
          await RunevmFl.load(bytes.buffer.asUint8List()) ?? false;
      print("Rune deployed:");
      if (loaded) {
        //Read Manifest with capabilities
        String manifest = (await RunevmFl.manifest).toString();
        print("Manifest loaded: $manifest");
      }
    } on Exception {
      print('Failed to init rune');
    }
    setState(() {
      _loaded = true;
    });
  }

  void _runRune() async {
    try {
      Random rand = Random();
      _input = rand.nextDouble() * 2 * pi;
      //convert input to 4 bytes representing a Float32 (See assets/Runefile)
      Uint8List inputBytes = Uint8List(4)
        ..buffer.asByteData().setFloat32(0, _input, Endian.little);
      //Run rune with the inputBytes
      _output = await RunevmFl.runRune(inputBytes);
      setState(() {});
    } on Exception {
      print('Failed to run rune');
    }
  }

}

Forge Deployment Implementation

First step is to initialise Forge and deploy model

import 'package:runevm_fl/runevm_fl.dart';

void loadForge() async {
  final answer = await Forge.forge({
    "deploymentId": "26", //insert  deploymentId here
    "apiKey": "{apiKey from forge}", //insert  apiKey here
    "baseURL": "https://dev-forge.hotg.ai", //insert  url here
    "telemetry": {
      "baseURL": "https://dev-telemetry.hotg.ai", //insert  url here
    }
  });
  setState(() {
    _capabilities = answer;
  });
}

To run inference imply provide to output to Forge.predict(Uint8List input)

  void _runInference() async {
    Uint8List inputData = getInputData();
    final data = await Forge.predict([inputData]);
    final out = (data is String) ? json.decode(data) : data;
    doSomethingWithOutput(out);
  }

Android

No extra config needed

iOS

If you are creating a new app, first run :

foo@bar:~$ flutter run

to generate the podfile.

Minimum iOS version should be at least 12.1 to be compatible with the plugin:

Set this in XCode > Runner > General > Deployment info

For version <0.3.0 Bitcode needs to be disabled either for the runevm_fl target:

XCode > Pods > Targets > runevm_fl > Build Settings > Enable Bitcode > Set to 'No'

or directly in the Podfile:

post_install do |installer|
  installer.pods_project.targets.each do |target|
    flutter_additional_ios_build_settings(target)
    ## Add these 3 lines to your podfile
    target.build_configurations.each do |config|
      config.build_settings['ENABLE_BITCODE'] = 'NO'
    end
    
  end
end

Run it

foo@bar:~$ flutter run