ZeticMLangeModel

This page reflects zetic_mlange 1.8.1.

ZeticMLangeModel is the primary Dart interface for running general on-device AI models through the native Android and iOS SDKs. It handles service-managed model download, native runtime initialization, execution, and native-handle release through a single Dart API.

Import

import 'package:zetic_mlange/zetic_mlange.dart';

Constructors

`create`

Creates a model from the Melange service using a personal key and model name.

static Future<ZeticMLangeModel> create({
  required String personalKey,
  required String name,
  int? version,
  ModelMode modelMode = ModelMode.runAuto,
  QuantType? quantType,
  Target? target,
  APType apType = APType.na,
  MlangeProgressCallback? onProgress,
  ZeticMLangeCacheHandlingPolicy cacheHandlingPolicy =
      CacheHandlingPolicy.removeOverlapping,
})

Parameter	Type	Default	Description
`personalKey`	`String`	—	Your personal authentication key.
`name`	`String`	—	Full model identifier in `account_name/project_name` format.
`version`	`int?`	`null`	Specific model version to load. `null` uses the latest.
`modelMode`	`ModelMode`	`runAuto`	Automatic selection strategy. See Enums.
`quantType`	`QuantType?`	`null`	Quantization precision filter (`fp32`, `fp16`, `int`). `null` disables precision-based filtering.
`target`	`Target?`	`null`	Optional runtime target. Use this only when you need to pin a backend and your account tier supports explicit target selection.
`apType`	`APType`	`na`	Optional processor preference: `cpu`, `gpu`, `npu`, or `na`.
`onProgress`	`MlangeProgressCallback?`	`null`	Download progress callback from `0.0` to `1.0`.
`cacheHandlingPolicy`	`ZeticMLangeCacheHandlingPolicy`	`removeOverlapping`	Native model cache policy. See Cache Management.

final model = await ZeticMLangeModel.create(
  personalKey: personalKey,
  name: 'Steve/YOLOv11_comparison',
  modelMode: ModelMode.runAuto,
  onProgress: (progress) => print(progress),
);

Throws: MlangeException or a native SDK error if the model cannot be downloaded, initialized, or matched to the requested runtime options.

Model creation performs network and file-system work on first use. The default FFI binding runs remote model creation asynchronously before returning the model handle.

Properties

`isClosed`

bool get isClosed

Returns true after close releases the native model handle.

Methods

`run`

Runs inference with optional input tensors.

List<Tensor> run([List<Tensor> inputs = const []])

Parameter	Type	Description
`inputs`	`List<Tensor>`	Input tensors matching the model's expected shapes and data types.

Returns: List<Tensor> containing the output tensors from the native model.

Throws: MlangeException if the input tensor count or byte length does not match the native model input buffers, or a native SDK error if inference fails.

final outputs = model.run([inputTensor]);

run() with no arguments executes against the model's existing native input buffers. Flutter does not currently expose those buffers directly, so most app code should pass inputs.

`close`

Releases the native model handle.

void close()

Call close() when the model is no longer needed.

After close(), calling run() or close() again throws MlangeException because the native handle has been released.

Full Working Example

import 'dart:typed_data';

import 'package:zetic_mlange/zetic_mlange.dart';

final model = await ZeticMLangeModel.create(
  personalKey: personalKey,
  name: modelName,
);

final input = Tensor.float32View(
  Float32List.fromList([/* values */]),
  shape: const [1, 3, 224, 224],
);

final outputs = model.run([input]);
final first = outputs.first.asFloat32List();

model.close();

ZeticMLangeModel

On this page