laya_flutter 0.1.0

Offline multilingual Laya inference for Flutter. Open an ONNX agent from cache or a local bundle, then run typed predict for choice, score, and noul questions.

laya_flutter

Run the multilingual Laya reinforcement-learning agent on device with ONNX. The library downloads the published checkpoint on first open, tokenizes state and questions, and returns typed answers for choice, score, and noul items—no cloud round trip.

[Snake example driven by offline Laya predict]

Please open an issue if something breaks or the docs should say more; fixes and copy improvements are welcome.

Features

• Offline inference through flutter_onnxruntime and Hugging Face tokenizers.
• LayaFlutter.open — ensures the multilingual ONNX graph and companion files live under a cache directory (downloads missing artifacts once).
• LayaFlutter.openLocalBundle — loads an existing directory that already contains laya.onnx, tokenizer.json, and rl_agent_config.json (used by task-specific agents such as Snake).
• LoadedRuntime.predict — batch inference with upstream-aligned temperature, softmax shaping, and typed LayaAnswer results.
• Runnable example — classic Snake where each move is chosen by the local agent (example/).

Installation

Add the package to your pubspec.yaml:

dependencies:
  laya_flutter: ^0.1.0

On first use, LayaFlutter.open needs network access to fetch the multilingual checkpoint from Hugging Face (mariojcr/laya-onnx and convaiinnovations/laya-multilingual). Later runs reuse the cache directory.

Quick start

import 'dart:io';

import 'package:laya_flutter/laya_flutter.dart';

Future<void> runLaya(Directory cache) async {
  final LoadedRuntime runtime = await LayaFlutter.open(cache);

  final Map<String, LayaAnswer> answers = await runtime.predict(
    'Player level 3, inventory: key',
    <LayaQuestion>[
      LayaQuestion(
        id: 'door',
        type: LayaQuestionType.choice,
        instructions: 'Which door do you open?',
        criteria: <String, String>{
          'left': 'A wooden door',
          'right': 'An iron gate',
        },
      ),
    ],
  );

  print(answers['door']?.choice);
  await runtime.close();
}

Local ONNX bundle

When you already ship a prepared bundle (different graph size or task weights), point at the directory instead of the multilingual cache:

final LoadedRuntime runtime = await LayaFlutter.openLocalBundle(
  Directory('/path/to/bundle'),
);

The directory must contain the three required files and an ONNX graph whose input and output names match the Laya agent contract.

Example app

The repository root is a library, not an application. Run the demo from example/:

cd example
flutter run

In Android Studio, open this repository and select the example run configuration (example/lib/main.dart).

The example autostarts Snake on a device that has a prepared Snake checkpoint. Set the bundle path in example/lib/app_settings.dart (snakeCheckpointDir). The bundled multilingual download path is separate; Snake uses openLocalBundle only.

Question types

Type	LayaQuestionType	Answer field
Multiple choice	choice	choice (option key at argmax)
Ordered score	score	score (expected level)
True / false	noul	noul (P(true))

See LayaQuestion and LayaAnswer for criteria maps and probability details.

Development layout

Path	Role
lib/	Public API (laya_flutter.dart)
example/	Flutter app entry point
test/	Unit and integration tests
wiki/	Planning and workflow artifacts

License

MIT — see LICENSE.