qlaya_flutter 0.4.1

QLaya Flutter & Dart client for the fast, local, on-device decision engine with calibrated probabilities and user-selectable quantized model variants.

qlaya_flutter

QLaya Flutter — Flutter and Dart client for the fast, local, on-device decision engine with user-selectable quantized model variants.

GitHub Repository | pub.dev Package | Issues & Feedback | Main QLaya Engine

Installation

Add to your Flutter or Dart project:

flutter pub add qlaya_flutter
# or for Dart console / backend apps:
dart pub add qlaya_flutter

Or add to your pubspec.yaml:

dependencies:
  qlaya_flutter: ^0.4.1

Quick Start

import 'package:qlaya_flutter/qlaya_flutter.dart';

void main() async {
  // List all quantized model variants
  print(QLayaModels.allIds);

  // Pick a model
  final model = QLayaModels.resolve('QLaya-TopProduction');  // ONNX INT8, recommended

  // Connect to a local QLaya server (run: qlaya-serve)
  final client = QLayaClient(baseUrl: 'http://localhost:8000');

  final result = await client.predict(
    text: 'I was charged twice, please refund',
    model: model,
  );

  print(result.answers);
  client.close();
}

Available Models

All 10 model variants are available on Hugging Face at saipy10/qlaya from the project's benchmark experiments:

QLaya ID	Size	Latency (p50)	RAM	Notes
QLaya-OriginalBaseline	1685 MB	382 ms	1720 MB	FP32 teacher, max accuracy
QLaya-Balanced	843 MB	368 ms	860 MB	FP16/BF16 half-precision
QLaya-TopProduction ⭐	572 MB	135 ms	590 MB	ONNX INT8, recommended
QLaya-SlowCPU	441 MB	681 ms	460 MB	ONNX INT4 block-32
QLaya-DegradedAccuracy	420 MB	1048 ms	435 MB	ONNX INT4 block-64
QLaya-IntermediateStudent	978 MB	195 ms	1010 MB	14L FP32 distilled
QLaya-HighSpeedProduction	333 MB	78 ms	350 MB	14L INT8 distilled
QLaya-CompactStudent	574 MB	94 ms	605 MB	6L FP32 distilled
QLaya-UltraFastEdge	195 MB	39 ms	210 MB	6L INT8, edge/mobile
QLaya-UltraSmallStorage	142 MB	113 ms	160 MB	6L INT4, minimal storage

Model Resolution

QLayaModels.resolve() supports fuzzy matching:

QLayaModels.resolve('QLaya-UltraFastEdge')  // exact
QLayaModels.resolve('ultra-fast-edge')       // fuzzy slug
QLayaModels.resolve('ultrafastedge')         // no hyphens

Running the Server

Install and start the QLaya server (Python):

pip install qlaya[serve]
qlaya-serve --host 0.0.0.0 --port 8000

License

Apache-2.0 — see LICENSE.