Flutter ONNX Runtime

https://pub.dev/documentation/flutter_onnxruntime/latest/
Admin
Flutter ONNX Runtime is a Flutter plugin that provides native wrappers for running ONNX Runtime...
Tokens:26,181
Snippets:207
Trust Score:9.9
Update:1 month ago
Show doc for...
Tokens:
Raw
### Run Simple Addition Model Example

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/index.html

Commands to navigate to the example directory and execute the simple addition model project.

```bash
cd example
flutter pub get
flutter run
```

--------------------------------

### String get userAgent

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/NavigatorJSExtension/userAgent.html

Retrieves the user agent string of the current browser environment.

```APIDOC
## String get userAgent

### Description
Returns the user agent string for the current browser session.

### Signature
`external String get userAgent;`
```

--------------------------------

### Get available providers implementation

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OnnxRuntime/getAvailableProviders.html

Retrieves the list of available providers from the platform instance and maps them to the OrtProvider enum.

```dart
Future<List<OrtProvider>> getAvailableProviders() async {
  final providers = await FlutterOnnxruntimePlatform.instance.getAvailableProviders();
  return providers.map((p) {
    final provider = OrtProvider.values.firstWhere(
      (e) => e.name == p,
      orElse: () => throw ArgumentError('Provider $p is not a valid OrtProvider.'),
    );
    return provider;
  }).toList();
}
```

--------------------------------

### Configure macOS Podfile and Project

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/index.html

Set the minimum macOS version in the Podfile and open the workspace to update Xcode settings.

```ruby
platform :osx, '14.0'
```

```bash
open Runner.xcworkspace
```

--------------------------------

### createSessionFromAsset(String assetKey, {OrtSessionOptions? options})

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OnnxRuntime/createSessionFromAsset.html

Creates an ONNX Runtime session from an asset model file. This method handles the extraction of the asset to a temporary file on native platforms or passes the key directly on web platforms.

```APIDOC
## Future<OrtSession> createSessionFromAsset(String assetKey, {OrtSessionOptions? options})

### Description
Creates an ONNX Runtime session from an asset model file. This method automatically extracts the asset to a temporary file for native platforms or handles the asset key appropriately for web platforms.

### Parameters
- **assetKey** (String) - Required - The path to the asset file in the Flutter project.
- **options** (OrtSessionOptions?) - Optional - Configuration options for the ONNX Runtime session.

### Returns
- **Future<OrtSession>** - A future that resolves to the created OrtSession instance.
```

--------------------------------

### Enable Swift Package Manager

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/index.html

Enable SPM support for Flutter projects on version 3.24 or higher.

```bash
flutter config --enable-swift-package-manager
```

--------------------------------

### Create an ONNX Runtime session

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OnnxRuntime/createSession.html

Initializes a session by providing the path to the model file and optional session settings.

```dart
Future<OrtSession> createSession(String modelPath, {OrtSessionOptions? options}) async {
  final result = await FlutterOnnxruntimePlatform.instance.createSession(
    modelPath,
    sessionOptions: options?.toMap() ?? {},
  );
  return OrtSession.fromMap(result);
}
```

--------------------------------

### Configure iOS Podfile

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/index.html

Set the minimum iOS version and ensure static linkage for CocoaPods.

```ruby
platform :ios, '16.0'

# existing code ...

use_frameworks! :linkage => :static

# existing code ...
```

--------------------------------

### Initialize OrtSessionOptions

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtSessionOptions/OrtSessionOptions.html

Constructor definition for configuring session options in ONNX Runtime.

```dart
OrtSessionOptions({this.intraOpNumThreads, this.interOpNumThreads, this.providers, this.useArena, this.deviceId});
```

--------------------------------

### Implementation of createSessionFromAsset

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OnnxRuntime/createSessionFromAsset.html

This method handles asset extraction to a temporary file for native platforms while passing the asset key directly on web platforms.

```dart
Future<OrtSession> createSessionFromAsset(String assetKey, {OrtSessionOptions? options}) async {
  if (kIsWeb) {
    // On web, we need to handle differently as path_provider is not available
    // Instead, pass the asset key directly to the web implementation
    // The web implementation will handle loading from asset properly
    return createSession(assetKey, options: options);
  } else {
    // Native platforms implementation (iOS, Android, etc)
    // Get the temporary directory
    final directory = await getTemporaryDirectory();
    // Extract filename from asset path using String manipulation
    final fileName = assetKey.split('/').last;
    // Join paths using string concatenation with platform-specific separator
    final filePath = '${directory.path}${Platform.pathSeparator}$fileName';

    final file = File(filePath);
    // Extract the asset only when it is not already cached. A cached file that
    // was fully written by a previous run is safe to reuse (see _extractAsset,
    // which writes atomically so a present file is always complete).
    final wasCached = await file.exists();
    if (!wasCached) {
      await _extractAsset(assetKey, file);
    }

    try {
      return await createSession(filePath, options: options);
    } catch (_) {
      // A cached model can still be truncated/corrupt (interrupted extraction
      // by an older non-atomic build, partial cache eviction). Such a file
      // "exists", so it would be reused and keep failing to load forever
      // (e.g. ORT_INVALID_PROTOBUF). Re-extract the always-intact bundled
      // asset and retry once so the install self-heals. Skip the retry when
      // the asset was extracted in this very call: those bytes are already
      // intact, so retrying identical bytes would only repeat the failure.
      if (!wasCached) rethrow;
      await _extractAsset(assetKey, file);
      return await createSession(filePath, options: options);
    }
  }
}
```

--------------------------------

### Running Inference with OrtSession

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtSession/run.html

Demonstrates how to prepare input tensors and execute the run method to obtain model outputs.

```dart
final inputTensor = await OrtValue.fromList(
  [1.0, 2.0, 3.0, 4.0],
  [2, 2]
);
final inputs = {
  'input_name': inputTensor,
};
final outputs = await session.run(inputs);
```

--------------------------------

### createSessionFromAsset

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OnnxRuntime-class.html

Creates an ONNX Runtime session using a model file bundled as a Flutter asset.

```APIDOC
## Future<OrtSession> createSessionFromAsset(String assetKey, {OrtSessionOptions? options})

### Description
Creates an ONNX Runtime session from an asset model file.

### Parameters
- **assetKey** (String) - Required - The key of the asset in the Flutter project.
- **options** (OrtSessionOptions) - Optional - Configuration options for the session.
```

--------------------------------

### toMap()

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtSessionOptions-class.html

Converts the session options configuration into a map format.

```APIDOC
### toMap() -> Map<String, dynamic>

Converts the current OrtSessionOptions instance into a Map representation.
```

--------------------------------

### createSession

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OnnxRuntime-class.html

Creates an ONNX Runtime session using a model file located at the specified path.

```APIDOC
## Future<OrtSession> createSession(String modelPath, {OrtSessionOptions? options})

### Description
Creates an ONNX Runtime session with the given model path.

### Parameters
- **modelPath** (String) - Required - The file system path to the ONNX model.
- **options** (OrtSessionOptions) - Optional - Configuration options for the session.
```

--------------------------------

### OrtSessionOptions()

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtSessionOptions/OrtSessionOptions.html

Creates a new instance of OrtSessionOptions to configure the ONNX Runtime session.

```APIDOC
## OrtSessionOptions()

### Description
Initializes a new instance of OrtSessionOptions with optional configuration parameters for the ONNX Runtime session.

### Parameters
- **intraOpNumThreads** (int?) - Optional - Sets the number of threads to use to parallelize the execution within an operator.
- **interOpNumThreads** (int?) - Optional - Sets the number of threads to use to parallelize the execution of independent operators.
- **providers** (List<OrtProvider>?) - Optional - A list of execution providers to be used by the session.
- **useArena** (bool?) - Optional - Enables or disables the memory arena for the session.
- **deviceId** (int?) - Optional - Specifies the device ID for the execution provider.
```

--------------------------------

### Implement runInference method

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/FlutterOnnxruntimeWebPlugin/runInference.html

This implementation handles session validation, input tensor mapping, optional run configuration, and output processing for the ONNX Runtime Web plugin.

```dart
@override
Future<Map<String, dynamic>> runInference(
  String sessionId,
  Map<String, OrtValue> inputs, {
  Map<String, dynamic>? runOptions,
}) async {
  try {
    // Check if the session exists
    if (!_sessions.containsKey(sessionId)) {
      throw PlatformException(code: "INVALID_SESSION", message: "Session not found", details: null);
    }

    final session = _sessions[sessionId]!;

    // Create a JavaScript object for inputs
    final jsInputs = createJSObject();

    // Process inputs - expecting OrtValue references
    for (final entry in inputs.entries) {
      final name = entry.key;
      final ortValue = entry.value;

      // Get the tensor by ID
      final valueId = ortValue.id;
      final tensor = _ortValues[valueId];
      if (tensor == null) {
        throw PlatformException(code: "INVALID_VALUE", message: "OrtValue with ID $valueId not found", details: null);
      }

      // Add to inputs object
      jsInputs.setProperty(name.toJS, tensor);
    }

    // Create run options if provided
    JSObject? jsRunOptions;
    if (runOptions != null && runOptions.isNotEmpty) {
      jsRunOptions = createJSObject();

      // Set run options if needed
      if (runOptions.containsKey('logSeverityLevel')) {
        jsRunOptions.setProperty('logSeverityLevel'.toJS, runOptions['logSeverityLevel']);
      }

      if (runOptions.containsKey('logVerbosityLevel')) {
        jsRunOptions.setProperty('logVerbosityLevel'.toJS, runOptions['logVerbosityLevel']);
      }

      if (runOptions.containsKey('terminate')) {
        jsRunOptions.setProperty('terminate'.toJS, runOptions['terminate']);
      }
    }

    // Run inference
    final runPromise =
        jsRunOptions != null
            ? callMethod(session, 'run', [jsInputs, jsRunOptions])
            : callMethod(session, 'run', [jsInputs]);

    // Wait for the promise to resolve
    final jsOutputs = await promiseToFuture<JSObject>(runPromise);

    // Process outputs - create OrtValue objects for each output
    final outputMap = <String, dynamic>{};
    final outputNames = getOutputNames(session);

    for (final name in outputNames) {
      if (jsOutputs.has(name)) {
        final tensor = jsOutputs.getProperty(name.toJS) as JSObject;

        // Create a new OrtValue and store it
        final valueId = '${DateTime.now().millisecondsSinceEpoch}_${math.Random().nextInt(10000)}';

        _ortValues[valueId] = tensor;

        // Get tensor type
        final type = tensor.getProperty('type'.toJS).toString();

        // Get tensor shape
        final jsShape = tensor.getProperty('dims'.toJS) as JSObject;
        final shapeLength = (jsShape.getProperty('length'.toJS) as JSNumber).toDartInt;
        final shape = <int>[];

        for (var i = 0; i < shapeLength; i++) {
          final val = jsShape.getProperty(i.toString().toJS) as JSNumber;
          shape.add(val.toDartInt);
        }

        // Format parameters like native platforms do
        outputMap[name] = [valueId, _mapOrtTypeToSourceType(type), shape];
      }
    }

    return outputMap;
  } catch (e) {
    if (e is PlatformException) {
      rethrow;
    }
    throw PlatformException(code: "INFERENCE_ERROR", message: "Failed to run inference: $e", details: null);
  }
}
```

--------------------------------

### Implementation of the run method

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtSession/run.html

Internal implementation details showing how the run method interacts with the platform interface to process inference results.

```dart
Future<Map<String, OrtValue>> run(Map<String, OrtValue> inputs, {OrtRunOptions? options}) async {
  final result = await FlutterOnnxruntimePlatform.instance.runInference(
    id,
    inputs,
    runOptions: options?.toMap() ?? {},
  );
  final outputs = <String, OrtValue>{};
  for (final entry in result.entries) {
    final tensorMap = {'valueId': entry.value[0], 'dataType': entry.value[1], 'shape': entry.value[2]};
    outputs[entry.key] = OrtValue.fromMap(tensorMap);
  }
  return outputs;
}
```

--------------------------------

### Implementation of createInferenceSession

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/FlutterOnnxruntimeWebPlugin/createInferenceSession.html

This method uses a Completer to handle the asynchronous creation of an InferenceSession from the onnxruntime-web library.

```dart
Future<JSObject> createInferenceSession(String modelPath, JSObject options) async {
  final completer = Completer<JSObject>();

  try {
    // Get the InferenceSession class from onnxruntime-web
    final inferenceSession = _ort.getProperty('InferenceSession'.toJS) as JSObject;

    // Use the create method to create a session - async operation
    final createPromise = callMethod(inferenceSession, 'create', [modelPath, options]);

    // Convert Promise to Future
    final session = await promiseToFuture<JSObject>(createPromise);
    completer.complete(session);
  } catch (e) {
    completer.completeError('Failed to create inference session: $e');
  }

  return completer.future;
}
```

--------------------------------

### createSession(String modelPath, {Map<String, dynamic>? sessionOptions})

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/FlutterOnnxruntimeWebPlugin/createSession.html

Initializes an ONNX inference session from a given model path and optional session configurations.

```APIDOC
## createSession(String modelPath, {Map<String, dynamic>? sessionOptions})

### Description
Creates an ONNX Runtime inference session. This method loads the model from the specified path, applies optional session configurations, and returns a map containing the session ID and input/output metadata.

### Parameters
- **modelPath** (String) - Required - The path to the ONNX model file.
- **sessionOptions** (Map<String, dynamic>) - Optional - Configuration options for the ONNX session.

### Returns
- **Future<Map<String, dynamic>>** - A map containing:
  - **sessionId** (String) - A unique identifier for the created session.
  - **inputNames** (List<String>) - The names of the model input nodes.
  - **outputNames** (List<String>) - The names of the model output nodes.

### Errors
- Throws a **PlatformException** with code 'PLUGIN_ERROR' if the session creation fails.
```

--------------------------------

### createInferenceSession

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/FlutterOnnxruntimeWebPlugin-class.html

Creates an inference session using onnxruntime-web.

```APIDOC
## createInferenceSession(String modelPath, JSObject options)

### Description
Creates an inference session using the onnxruntime-web library.

### Parameters
- **modelPath** (String) - Required - The path to the model file.
- **options** (JSObject) - Required - The session options configuration.

### Returns
- **Future<JSObject>** - A promise resolving to the created inference session object.
```

--------------------------------

### OrtModelMetadata constructor

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtModelMetadata/OrtModelMetadata.html

Initializes a new instance of the OrtModelMetadata class with the required model information.

```APIDOC
## OrtModelMetadata(String producerName, String graphName, String domain, String description, int version, Map<String, String> customMetadataMap)

### Description
Creates a new instance of OrtModelMetadata with the specified model attributes.

### Parameters
- **producerName** (String) - Required - The name of the model producer.
- **graphName** (String) - Required - The name of the model graph.
- **domain** (String) - Required - The domain of the model.
- **description** (String) - Required - A description of the model.
- **version** (int) - Required - The version number of the model.
- **customMetadataMap** (Map<String, String>) - Required - A map containing custom metadata key-value pairs.
```

--------------------------------

### Implementation of getOutputInfo

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtSession/getOutputInfo.html

This method retrieves output information from the platform instance and converts the result into a list of maps.

```dart
Future<List<Map<String, dynamic>>> getOutputInfo() async {
  final outputInfoMap = await FlutterOnnxruntimePlatform.instance.getOutputInfo(id);
  return outputInfoMap.map((info) => Map<String, dynamic>.from(info)).toList();
}
```

--------------------------------

### Configure Android ProGuard Rules

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/index.html

Add these rules to your Android project to ensure ONNX Runtime classes are preserved during the build process.

```text
-keep class ai.onnxruntime.** { *; }
```

```bash
echo "-keep class ai.onnxruntime.** { *; }" > android/app/proguard-rules.pro
```

--------------------------------

### Future<Map<String, OrtValue>> run(Map<String, OrtValue> inputs, {OrtRunOptions? options})

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtSession/run.html

Executes inference on the session. Takes a map of input names to OrtValue objects and returns a map of output names to OrtValue objects.

```APIDOC
## run(Map<String, OrtValue> inputs, {OrtRunOptions? options})

### Description
Runs inference on the loaded ONNX Runtime session.

### Parameters
- **inputs** (Map<String, OrtValue>) - Required - A map where keys are input names and values are OrtValue objects.
- **options** (OrtRunOptions?) - Optional - Configuration options for the inference run.

### Returns
- **Future<Map<String, OrtValue>>** - A map of output names to OrtValue objects if successful, otherwise throws an exception.

### Example
```dart
final inputTensor = await OrtValue.fromList(
  [1.0, 2.0, 3.0, 4.0],
  [2, 2]
);
final inputs = {
  'input_name': inputTensor,
};
final outputs = await session.run(inputs);
```
```

--------------------------------

### createInferenceSession(String modelPath, JSObject options)

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/FlutterOnnxruntimeWebPlugin/createInferenceSession.html

Creates an inference session using the onnxruntime-web library.

```APIDOC
## createInferenceSession

### Description
Creates an inference session using onnxruntime-web. This method is an asynchronous operation that returns a JSObject representing the session.

### Signature
`Future<JSObject> createInferenceSession(String modelPath, JSObject options)`

### Parameters
- **modelPath** (String) - The path to the ONNX model file.
- **options** (JSObject) - Configuration options for the inference session.

### Returns
- **Future<JSObject>** - A future that completes with the created inference session object.
```

--------------------------------

### getPlatformVersion()

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/FlutterOnnxruntimeWebPlugin/getPlatformVersion.html

Retrieves the platform version string from the browser's navigator user agent.

```APIDOC
## getPlatformVersion()

### Description
Returns the platform version string by accessing the browser's navigator.userAgent property.

### Signature
`Future<String?> getPlatformVersion()`

### Returns
- **String?** - The platform version string or null if unavailable.
```

--------------------------------

### Implement getPlatformVersion

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/FlutterOnnxruntimeWebPlugin/getPlatformVersion.html

Retrieves the user agent string from the browser's navigator object.

```dart
@override
Future<String?> getPlatformVersion() async {
  return _window.navigator.userAgent;
}
```

--------------------------------

### Initialize OrtRunOptions

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtRunOptions/OrtRunOptions.html

Constructor for configuring runtime options such as log severity, verbosity, and termination status.

```dart
OrtRunOptions({this.logSeverityLevel, this.logVerbosityLevel, this.terminate});
```

--------------------------------

### runInference

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/FlutterOnnxruntimeWebPlugin-class.html

Executes inference on a specified session.

```APIDOC
## runInference(String sessionId, Map<String, OrtValue> inputs, {Map<String, dynamic>? runOptions})

### Description
Runs inference on a session with the provided inputs and optional run configurations.

### Parameters
- **sessionId** (String) - Required - The ID of the active session.
- **inputs** (Map<String, OrtValue>) - Required - A map of input names to OrtValue objects.
- **runOptions** (Map<String, dynamic>) - Optional - Additional configuration for the inference run.

### Returns
- **Future<Map<String, dynamic>>** - A map containing the inference results.
```

--------------------------------

### createJsSessionOptions(Map<String, dynamic>? options)

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/FlutterOnnxruntimeWebPlugin/createJsSessionOptions.html

Creates a JavaScript session options object based on the provided configuration map.

```APIDOC
## createJsSessionOptions

### Description
Creates a JavaScript session options object for onnxruntime-web. This method maps Flutter-specific configuration options to the corresponding JavaScript properties.

### Signature
`JSObject createJsSessionOptions(Map<String, dynamic>? options)`

### Parameters
- **options** (Map<String, dynamic>?) - Optional - A map containing configuration settings:
  - **providers** (List<String>) - List of execution providers (e.g., 'CPU', 'WEB_ASSEMBLY', 'WEB_GL', 'WEB_GPU', 'WEB_NN').
  - **intraOpNumThreads** (int) - Number of threads for intra-op parallelism.
  - **interOpNumThreads** (int) - Number of threads for inter-op parallelism.
  - **graphOptimizationLevel** (dynamic) - The optimization level for the graph.

### Returns
- **JSObject** - The configured JavaScript session options object.
```

--------------------------------

### OrtRunOptions()

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtRunOptions/OrtRunOptions.html

Creates a new instance of OrtRunOptions with optional logging and termination settings.

```APIDOC
## OrtRunOptions()

### Description
Initializes a new instance of the OrtRunOptions class to configure inference execution parameters.

### Parameters
- **logSeverityLevel** (int?) - Optional - Sets the logging severity level.
- **logVerbosityLevel** (int?) - Optional - Sets the logging verbosity level.
- **terminate** (bool?) - Optional - Sets the termination flag for the inference session.

### Implementation
```dart
OrtRunOptions({this.logSeverityLevel, this.logVerbosityLevel, this.terminate});
```
```

--------------------------------

### OrtSessionOptions Constructor

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtSessionOptions-class.html

Creates a new instance of OrtSessionOptions to configure session behavior.

```APIDOC
### OrtSessionOptions({int? intraOpNumThreads, int? interOpNumThreads, List<OrtProvider>? providers, bool? useArena, int? deviceId})

Creates a configuration object for an ONNX Runtime session.

#### Parameters
- **intraOpNumThreads** (int?) - Optional. Sets the number of threads used to parallelize the execution within an operator.
- **interOpNumThreads** (int?) - Optional. Sets the number of threads used to parallelize the execution of independent operators.
- **providers** (List<OrtProvider>?) - Optional. A list of execution providers to be used by the session.
- **useArena** (bool?) - Optional. Enables or disables the memory arena.
- **deviceId** (int?) - Optional. Specifies the device ID for execution.
```

--------------------------------

### Implementation of OrtSession.fromMap

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtSession/OrtSession.fromMap.html

This factory constructor creates an OrtSession instance by extracting session ID, input names, and output names from a map.

```dart
factory OrtSession.fromMap(Map<String, dynamic> map) {
  return OrtSession._(
    id: map['sessionId'] as String,
    inputNames: List<String>.from(map['inputNames'] ?? []),
    outputNames: List<String>.from(map['outputNames'] ?? []),
  );
}
```

--------------------------------

### runInference(String sessionId, Map<String, OrtValue> inputs, {Map<String, dynamic>? runOptions})

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/FlutterOnnxruntimeWebPlugin/runInference.html

Executes inference on a specific session using the provided input map and optional run configurations.

```APIDOC
## runInference

### Description
Runs inference on a session identified by `sessionId`. It takes a map of input names to `OrtValue` objects and returns a map of output names to their respective result data.

### Signature
`Future<Map<String, dynamic>> runInference(String sessionId, Map<String, OrtValue> inputs, {Map<String, dynamic>? runOptions})`

### Parameters
- **sessionId** (String) - The ID of the session to run inference on.
- **inputs** (Map<String, OrtValue>) - A map of input names to `OrtValue` objects.
- **runOptions** (Map<String, dynamic>, optional) - An optional map of run options, including 'logSeverityLevel', 'logVerbosityLevel', or 'terminate'.

### Returns
- **Future<Map<String, dynamic>>** - A map containing the inference results, where keys are output names and values are lists containing the output ID, type, and shape.
```

--------------------------------

### Implementation of getOutputInfo

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/FlutterOnnxruntimeWebPlugin/getOutputInfo.html

This method retrieves output metadata from an existing session, handling both tensor and non-tensor types, with a fallback to output names if metadata is unavailable.

```dart
@override
Future<List<Map<String, dynamic>>> getOutputInfo(String sessionId) async {
  try {
    // Check if the session exists
    if (!_sessions.containsKey(sessionId)) {
      throw PlatformException(code: "INVALID_SESSION", message: "Session not found", details: null);
    }

    final session = _sessions[sessionId]!;
    final outputInfoList = <Map<String, dynamic>>[];

    // Get output metadata from session if available
    if (session.has('outputMetadata')) {
      final outputMetadata = session.getProperty('outputMetadata'.toJS) as JSObject;
      final length = (outputMetadata.getProperty('length'.toJS) as JSNumber).toDartInt;

      for (var i = 0; i < length; i++) {
        final info = outputMetadata.getProperty(i.toString().toJS) as JSObject;
        final infoMap = <String, dynamic>{};

        // Add name
        infoMap['name'] = info.getProperty('name'.toJS).toString();

        // Check if it's a tensor
        final isTensorProperty = info.getProperty('isTensor'.toJS);
        final isTensor = (isTensorProperty as JSBoolean?)?.toDart ?? true;

        if (isTensor) {
          // Add shape if available
          if (info.has('shape')) {
            final shape = info.getProperty('shape'.toJS) as JSObject;
            final shapeLength = (shape.getProperty('length'.toJS) as JSNumber).toDartInt;
            final shapeList = <int>[];

            for (var j = 0; j < shapeLength; j++) {
              final dim = shape.getProperty(j.toString().toJS);
              // Handle both numeric dimensions and symbolic dimensions
              if (dim != null) {
                try {
                  final numValue = (dim as JSNumber).toDartInt;
                  shapeList.add(numValue);
                } catch (e) {
                  // For symbolic dimensions or non-numeric values, use -1 as a placeholder
                  shapeList.add(-1);
                }
              } else {
                // For null values, use -1 as a placeholder
                shapeList.add(-1);
              }
            }

            infoMap['shape'] = shapeList;
          } else {
            infoMap['shape'] = <int>[];
          }

          // Add type if available
          if (info.has('type')) {
            infoMap['type'] = info.getProperty('type'.toJS).toString();
          } else {
            infoMap['type'] = 'unknown';
          }
        } else {
          // For non-tensor types, provide empty shape
          infoMap['shape'] = <int>[];
          infoMap['type'] = 'non-tensor';
        }

        outputInfoList.add(infoMap);
      }
    } else {
      // Fallback: use output names to create basic info entries
      final outputNames = getOutputNames(session);
      for (final name in outputNames) {
        outputInfoList.add({'name': name, 'shape': <int>[], 'type': 'unknown'});
      }
    }

    return outputInfoList;
  } catch (e) {
    if (e is PlatformException) {
      rethrow;
    }
    throw PlatformException(code: "PLUGIN_ERROR", message: "Failed to get output info: $e", details: null);
  }
}
```

--------------------------------

### run(Map<String, OrtValue> inputs, {OrtRunOptions? options})

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtSession-class.html

Executes inference on the loaded ONNX model using the provided input map and optional run configurations.

```APIDOC
### Method
`Future<Map<String, OrtValue>> run(Map<String, OrtValue> inputs, {OrtRunOptions? options})`

### Description
Runs inference on the session using the provided input data. Returns a map of output names to OrtValue objects.

### Parameters
- **inputs** (Map<String, OrtValue>) - Required - A map where keys are input names and values are the corresponding OrtValue data.
- **options** (OrtRunOptions) - Optional - Configuration options for the inference run.
```

--------------------------------

### Initialize OrtModelMetadata

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtModelMetadata/OrtModelMetadata.html

Constructor for creating an instance of OrtModelMetadata with required model information.

```dart
OrtModelMetadata({
  required this.producerName,
  required this.graphName,
  required this.domain,
  required this.description,
  required this.version,
  required this.customMetadataMap,
});
```

--------------------------------

### Run model inference

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/index.html

Initialize an inference session from an asset and execute it with provided input tensors.

```dart
import 'package:flutter_onnxruntime/flutter_onnxruntime.dart';

// create inference session
final ort = OnnxRuntime();
final session = await ort.createSessionFromAsset('assets/models/addition_model.onnx');

// specify input with data and shape
final inputs = {
   'A': await OrtValue.fromList([1, 1, 1], [3]),
   'B': await OrtValue.fromList([2, 2, 2], [3])
}

// start the inference
final outputs = await session.run(inputs);

// print output data
print(await outputs['C']!.asList());
```

--------------------------------

### Future<String?> getPlatformVersion()

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OnnxRuntime/getPlatformVersion.html

Retrieves the version of the platform on which the application is currently running.

```APIDOC
## getPlatformVersion

### Description
Returns the platform version as a string. This method is asynchronous and returns a Future that resolves to the version string or null if the version cannot be determined.

### Signature
`Future<String?> getPlatformVersion()`

### Returns
- **Future<String?>** - A future that completes with the platform version string.
```

--------------------------------

### Implementation of getAvailableProviders

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/FlutterOnnxruntimeWebPlugin/getAvailableProviders.html

This method checks the browser environment for support of various execution providers and returns them as a list of strings.

```dart
@override
Future<List<String>> getAvailableProviders() async {
  // Return the list of execution providers supported by onnxruntime-web
  final providers = <String>[];
  // Check for WebAssembly support
  if (globalThis.has('WebAssembly')) {
    providers.add('WEB_ASSEMBLY');
  }
  // Check for WebGL support
  final canvas = HTMLCanvasElement();
  final webgl = canvas.getContext('webgl') ?? canvas.getContext('experimental-webgl');
  if (webgl != null) {
    providers.add('WEB_GL');
  }
  // Check for WebGPU support
  if ((_window.navigator as JSObject).has('gpu')) {
    providers.add('WEB_GPU');
  }
  // Check for WebNN support
  if ((_window.navigator as JSObject).has('ml')) {
    providers.add('WEB_NN');
  }
  return providers;
}
```

--------------------------------

### Implementation of getInputInfo

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtSession/getInputInfo.html

This method retrieves input metadata from the platform instance and converts the result into a list of maps.

```dart
Future<List<Map<String, dynamic>>> getInputInfo() async {
  final inputInfoMap = await FlutterOnnxruntimePlatform.instance.getInputInfo(id);
  return inputInfoMap.map((info) => Map<String, dynamic>.from(info)).toList();
}
```

--------------------------------

### OrtSession.fromMap(Map<String, dynamic> map)

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtSession/OrtSession.fromMap.html

Creates an instance of OrtSession from a map of session data.

```APIDOC
## OrtSession.fromMap

### Description
Creates an instance of an `OrtSession` from a map containing session configuration data.

### Signature
`factory OrtSession.fromMap(Map<String, dynamic> map)`

### Parameters
- **map** (Map<String, dynamic>) - Required - A map containing the following keys:
  - `sessionId` (String): The unique identifier for the session.
  - `inputNames` (List<String>): A list of input names for the model.
  - `outputNames` (List<String>): A list of output names for the model.
```

--------------------------------

### Future<List<OrtProvider>> getAvailableProviders()

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OnnxRuntime/getAvailableProviders.html

Retrieves a list of available ONNX Runtime providers as an asynchronous operation.

```APIDOC
## getAvailableProviders

### Description
Returns a list of the available ONNX Runtime providers supported by the current platform.

### Signature
`Future<List<OrtProvider>> getAvailableProviders()`

### Returns
- **List<OrtProvider>** - A list of available providers represented by the OrtProvider enum.
```

--------------------------------

### Implementation of createSession

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/FlutterOnnxruntimeWebPlugin/createSession.html

This method initializes the ONNX session, generates a unique ID, and stores the session instance for future inference tasks.

```dart
@override
Future<Map<String, dynamic>> createSession(String modelPath, {Map<String, dynamic>? sessionOptions}) async {
  try {
    // Initialize session options for onnxruntime-web
    final jsSessionOptions = createJsSessionOptions(sessionOptions);

    // Create the session using onnxruntime-web
    final session = await createInferenceSession(modelPath, jsSessionOptions);

    // Generate a unique session ID
    final sessionId = DateTime.now().millisecondsSinceEpoch.toString();

    // Get input and output names
    final inputNames = getInputNames(session);
    final outputNames = getOutputNames(session);

    // Store the session for later use
    _sessions[sessionId] = session;

    // Return the required information
    return {'sessionId': sessionId, 'inputNames': inputNames, 'outputNames': outputNames};
  } catch (e) {
    throw PlatformException(code: 'PLUGIN_ERROR', message: 'Failed to create ONNX session: $e', details: null);
  }
}
```

--------------------------------

### OrtRunOptions Constructor

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtRunOptions-class.html

Creates a new instance of OrtRunOptions to configure runtime execution parameters.

```APIDOC
### Constructor
OrtRunOptions({int? logSeverityLevel, int? logVerbosityLevel, bool? terminate})

### Parameters
- **logSeverityLevel** (int?) - Optional - The severity level for logging.
- **logVerbosityLevel** (int?) - Optional - The verbosity level for logging.
- **terminate** (bool?) - Optional - A flag to indicate if execution should be terminated.
```

--------------------------------

### getOutputInfo()

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtSession-class.html

Retrieves information about the output nodes of the loaded model.

```APIDOC
### Method
`Future<List<Map<String, dynamic>>> getOutputInfo()`

### Description
Returns a list of maps containing metadata about each output node of the model.
```

--------------------------------

### Future<List<Map<String, dynamic>>> getOutputInfo()

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtSession/getOutputInfo.html

Retrieves information about the model's outputs such as name, type, and shape.

```APIDOC
## getOutputInfo()

### Description
Returns a list of maps containing information about the model's outputs, including their name, data type, and shape.

### Signature
`Future<List<Map<String, dynamic>>> getOutputInfo()`

### Returns
- **List<Map<String, dynamic>>**: A list where each map represents an output node with keys corresponding to the output's metadata (e.g., name, type, shape).
```

--------------------------------

### toMap()

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtModelMetadata-class.html

Converts the metadata instance into a Map representation.

```APIDOC
### Method
Map<String, dynamic> toMap()
```

--------------------------------

### OrtModelMetadata(producerName, graphName, domain, description, version, customMetadataMap)

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtModelMetadata-class.html

Constructor to create a new instance of OrtModelMetadata with the specified metadata fields.

```APIDOC
### Constructor
OrtModelMetadata({required String producerName, required String graphName, required String domain, required String description, required int version, required Map<String, String> customMetadataMap})
```

--------------------------------

### Implementation of callMethod

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/FlutterOnnxruntimeWebPlugin/callMethod.html

This function routes method calls based on the number of arguments provided, using specific handlers for 0-2 arguments and a variadic fallback for 3 or more.

```dart
dynamic callMethod(JSObject obj, String method, List<dynamic> args) {
  switch (args.length) {
    case 0:
      return callMethod0(obj, method);
    case 1:
      return callMethod1(obj, method, args[0]);
    case 2:
      return callMethod2(obj, method, args[0], args[1]);
    default:
      // For 3+ arguments, fall back to VarArgs if needed
      // Note: the cast does not work properly at the moment
      return obj.callMethodVarArgs(method.toJS, args.cast<JSAny?>());
  }
}
```

--------------------------------

### getPlatformVersion

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OnnxRuntime-class.html

Retrieves the version string of the underlying ONNX Runtime platform.

```APIDOC
## Future<String?> getPlatformVersion()

### Description
Returns the platform version of the ONNX Runtime.
```

--------------------------------

### int32ArrayConstructor

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/int32ArrayConstructor.html

Accesses the global JavaScript Int32Array constructor.

```APIDOC
## int32ArrayConstructor

### Description
Returns the JavaScript `Int32Array` constructor function, allowing for the instantiation of typed 32-bit integer arrays in the browser environment.

### Signature
`JSFunction get int32ArrayConstructor`

### Usage
This property is annotated with `@JS('Int32Array')` and is used to interface with the underlying JavaScript typed array implementation.
```

--------------------------------

### getAvailableProviders

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OnnxRuntime-class.html

Retrieves a list of available execution providers supported by the current runtime.

```APIDOC
## Future<List<OrtProvider>> getAvailableProviders()

### Description
Get the available providers.
```

--------------------------------

### asList

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtValue-class.html

Retrieves the tensor data as a list.

```APIDOC
### Future<List> asList()

Returns the underlying tensor data as a list.
```

--------------------------------

### createJSObject()

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/FlutterOnnxruntimeWebPlugin/createJSObject.html

Creates a new JSObject instance by invoking the object constructor.

```APIDOC
## createJSObject()

### Description
Creates and returns a new JSObject instance by calling the underlying object constructor.

### Signature
`JSObject createJSObject()`

### Implementation
```dart
JSObject createJSObject() => objectConstructor.callAsConstructor();
```
```

--------------------------------

### getInputInfo()

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtSession-class.html

Retrieves information about the input nodes of the loaded model.

```APIDOC
### Method
`Future<List<Map<String, dynamic>>> getInputInfo()`

### Description
Returns a list of maps containing metadata about each input node of the model.
```

--------------------------------

### Implementation of asList

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtValue/asList.html

This method retrieves raw tensor data and reshapes it into a list, including specific handling for boolean tensors on iOS/macOS.

```dart
Future<List<dynamic>> asList() async {
  final data = await FlutterOnnxruntimePlatform.instance.getOrtValueData(id);
  final rawData = data['data'];
  var dataList1d = (rawData is List) ? rawData : List<dynamic>.from(rawData);
  // On iOS/macOS, bool tensors are stored as uint8 internally,
  // so convert 0/1 integers back to false/true
  if (dataType == OrtDataType.bool && dataList1d.isNotEmpty && dataList1d.first is! bool) {
    dataList1d = dataList1d.map((e) => e != 0).toList();
  }
  return _reshapeList(dataList1d, shape);
}
```

--------------------------------

### Map<String, dynamic> toMap()

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtSessionOptions/toMap.html

Converts the current OrtSessionOptions instance into a map representation containing non-null configuration fields.

```APIDOC
## Method Signature
`Map<String, dynamic> toMap()`

### Description
Serializes the `OrtSessionOptions` object into a `Map<String, dynamic>`. This is typically used to pass configuration settings to the underlying native ONNX Runtime implementation.

### Returns
- **Map<String, dynamic>** - A map containing the following keys if they are set:
  - `intraOpNumThreads` (int)
  - `interOpNumThreads` (int)
  - `providers` (List<String>) - A list of provider names.
  - `useArena` (bool)
  - `deviceId` (int)
```

--------------------------------

### callMethod(JSObject obj, String method, List<dynamic> args)

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/FlutterOnnxruntimeWebPlugin/callMethod.html

Invokes a method on a provided JSObject with the specified arguments.

```APIDOC
## callMethod(JSObject obj, String method, List<dynamic> args)

### Description
Invokes a method on a JavaScript object. The method dynamically routes the call to specific handlers based on the number of arguments provided in the list.

### Parameters
- **obj** (JSObject) - The target JavaScript object.
- **method** (String) - The name of the method to invoke.
- **args** (List<dynamic>) - A list of arguments to pass to the method.
```

--------------------------------

### bigInt64ArrayConstructor

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/bigInt64ArrayConstructor.html

Accesses the BigInt64Array constructor from the JavaScript environment.

```APIDOC
## bigInt64ArrayConstructor

### Description
Returns the BigInt64Array constructor function from the global JavaScript scope.

### Signature
`JSFunction get bigInt64ArrayConstructor`

### Returns
- **JSFunction** - The BigInt64Array constructor.
```

--------------------------------

### registerWith(Registrar registrar)

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/FlutterOnnxruntimeWebPlugin/registerWith.html

Registers the FlutterOnnxruntimeWebPlugin instance as the default implementation of FlutterOnnxruntimePlatform.

```APIDOC
## void registerWith(Registrar registrar)

### Description
Registers the current class as the default instance of `FlutterOnnxruntimePlatform` for web environments.

### Parameters
- **registrar** (Registrar) - Required - The registrar instance used to register the plugin.
```

--------------------------------

### getOutputNames(JSObject session)

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/FlutterOnnxruntimeWebPlugin/getOutputNames.html

Retrieves the list of output names associated with the given ONNX Runtime session.

```APIDOC
## getOutputNames(JSObject session)

### Description
Retrieves a list of output names from the provided ONNX Runtime session object.

### Signature
`List<String> getOutputNames(JSObject session)`

### Parameters
- **session** (JSObject) - The ONNX Runtime session object from which to extract output names.

### Returns
- **List<String>** - A list containing the names of the outputs defined in the session.
```

--------------------------------

### NavigatorJS.userAgent

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/NavigatorJS-class.html

Retrieves the user agent string from the browser navigator.

```APIDOC
### Property: userAgent

#### Description
Returns the user agent string of the current browser environment.

#### Type
String

#### Access
Read-only
```

--------------------------------

### objectGlobal

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/objectGlobal.html

Accesses the global JavaScript Object constructor.

```APIDOC
## objectGlobal

### Description
Returns the global JavaScript 'Object' constructor as a JSObject.

### Signature
`JSObject get objectGlobal`
```

--------------------------------

### List<String> getInputNames(JSObject session)

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/FlutterOnnxruntimeWebPlugin/getInputNames.html

Retrieves the input names associated with the given ONNX Runtime session.

```APIDOC
## getInputNames(JSObject session)

### Description
Retrieves a list of input names from the specified ONNX Runtime session object.

### Parameters
- **session** (JSObject) - The ONNX Runtime session instance from which to extract input names.

### Returns
- **List<String>** - A list containing the names of the inputs defined in the session.
```

--------------------------------

### Retrieve platform version

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OnnxRuntime/getPlatformVersion.html

Use this method to obtain the platform version string from the underlying platform implementation.

```dart
Future<String?> getPlatformVersion() {
  return FlutterOnnxruntimePlatform.instance.getPlatformVersion();
}
```

--------------------------------

### Future<List<Map<String, dynamic>>> getInputInfo()

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/flutter_onnxruntime/OrtSession/getInputInfo.html

Retrieves information about the model's inputs such as name, type, and shape.

```APIDOC
## Method Signature
`Future<List<Map<String, dynamic>>> getInputInfo()`

### Description
Returns a list of maps containing metadata for each input of the loaded ONNX model. Each map includes details such as the input name, data type, and tensor shape.

### Returns
- **List<Map<String, dynamic>>**: A list where each element represents an input layer with its associated properties.
```

--------------------------------

### navigator

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/WindowJSExtension/navigator.html

Access the NavigatorJS interface via the navigator property.

```APIDOC
## navigator

### Description
Retrieves the NavigatorJS instance associated with the current window context.

### Signature
`external NavigatorJS get navigator;`
```

--------------------------------

### Future<List<Map<String, dynamic>>> getOutputInfo(String sessionId)

Source: https://pub.dev/documentation/flutter_onnxruntime/latest/web_flutter_onnxruntime_web_plugin/FlutterOnnxruntimeWebPlugin/getOutputInfo.html

Retrieves information about the model's outputs such as name, type, and shape for a given session ID.

```APIDOC
## Future<List<Map<String, dynamic>>> getOutputInfo(String sessionId)

### Description
Retrieves metadata about the outputs of the model associated with the provided session ID. Returns a list of maps containing the name, type, and shape of each output.

### Parameters
- **sessionId** (String) - Required - The ID of the session to retrieve output information from.

### Returns
- **List<Map<String, dynamic>>** - A list of maps where each map contains:
  - **name** (String): The name of the output.
  - **shape** (List<int>): The dimensions of the output tensor.
  - **type** (String): The data type of the output.
```
Tokens:
Raw