### Install iOS Pods

Source: https://docs.zetic.ai/platform-integration/flutter/setup

After configuring the `ios/Podfile`, navigate to the `ios` directory and run `pod install`.

```bash
cd ios
pod install
```

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

### Initialize ZeticMLangeLLMModel Example

Source: https://docs.zetic.ai/api-reference/android/ZeticMLangeLLMModel

Example of initializing ZeticMLangeLLMModel with a specific context, personal key, model name, and initialization options.

```kotlin
val model = ZeticMLangeLLMModel(
    context = context,
    personalKey = PERSONAL_KEY,
    name = "account_name/project_name",
    initOption = LLMInitOption(nCtx = 4096),
)
```

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

### Install and Use onnxsim

Source: https://docs.zetic.ai/how-to-guides/performance-optimization

Install the onnxsim tool and use it to simplify ONNX models before uploading to reduce redundant operations.

```bash
pip install onnxsim
onnxsim input_model.onnx output_model.onnx
```

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

### Verify Zetic Melange Setup

Source: https://docs.zetic.ai/platform-integration/flutter/setup

Import the package and create a `ZeticMLangeModel` instance to verify the setup. Model creation downloads the optimized model artifact on first use.

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

final model = await ZeticMLangeModel.create(
  personalKey: personalKey,
  name: 'account_name/project_name',
  onProgress: (progress) {
    print('Loading ${(progress * 100).round()}%');
  },
);
```

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

### Install skl2onnx

Source: https://docs.zetic.ai/model-preparation/onnx

Install the skl2onnx package to enable conversion of scikit-learn models to ONNX.

```bash
pip install skl2onnx
```

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

### ZeticMLangeLLMModel.run

Source: https://docs.zetic.ai/api-reference/flutter/ZeticMLangeLLMModel

Starts text generation for a given prompt using the loaded LLM.

```APIDOC
## ZeticMLangeLLMModel.run

### Description
Starts text generation for a given prompt using the loaded LLM.

### Method Signature
```dart
LLMRunResult run(String text)
```

### Parameters
#### Path Parameters
* `text` (String) - Required - The prompt to generate text from.

### Request Example
```dart
final result = model.run('Explain on-device AI in one paragraph.');
```
```

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

### Full Working Example: iOS ViewController for Inference

Source: https://docs.zetic.ai/platform-integration/ios/basic-inference

A complete example integrated within an iOS ViewController's viewDidLoad method. It includes error handling for the inference process. Remember to prepare your input tensors correctly.

```swift
import ZeticMLange

class ViewController: UIViewController {
    override func viewDidLoad() {
        super.viewDidLoad()

        do {
            // Load model
            let model = try ZeticMLangeModel(personalKey: PERSONAL_KEY, name: "Steve/YOLOv11_comparison")

            // Prepare inputs
            let inputs: [Tensor] = [] // Prepare your inputs

            // Run inference
            let outputs = try model.run(inputs: inputs)

            // Process outputs
            for output in outputs {
                // Process each output tensor
            }
        } catch {
            print("Melange error: \(error)")
        }
    }
}
```

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

### Repository Structure Example

Source: https://docs.zetic.ai/model-deployment/understanding-model-keys

Illustrates the hierarchical structure of a model repository, including versions and default/latest tags.

```text
username/my-model-repository
├── v1 (uploaded 2024-01-15)
├── v2 (uploaded 2024-02-20) 
└── v3 (uploaded 2024-03-10) 
```

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

### Full Working Example: Running a Model with Tensors

Source: https://docs.zetic.ai/api-reference/android/Tensor

Demonstrates how to create a ZeticMLangeModel, run inference with input tensors, and read typed output arrays.

```kotlin
import com.zeticai.mlange.core.model.ZeticMLangeModel
import com.zeticai.mlange.core.tensor.Tensor

class MainActivity : AppCompatActivity() {
    private lateinit var model: ZeticMLangeModel
    override fun onCreate(savedInstanceState: Bundle?) {
        super.onCreate(savedInstanceState)

        model = ZeticMLangeModel(this, PERSONAL_KEY, "Steve/YOLOv11_comparison")
    }

    private fun runFrame(input: Tensor) {
        // (1) Run inference with tensors matching the model input order
        val outputs: Array<Tensor> = model.run(arrayOf(input))

        // (2) Read outputs as typed arrays
        val logits: FloatArray = outputs[0].data()
    }
}
```

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

### Fetch Packages

Source: https://docs.zetic.ai/platform-integration/flutter/setup

After adding the dependency, run `flutter pub get` to fetch the packages.

```bash
flutter pub get
```

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

### Install tf2onnx

Source: https://docs.zetic.ai/model-preparation/onnx

Install the tf2onnx package to enable conversion of TensorFlow and Keras models to ONNX.

```bash
pip install tf2onnx
```

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

### Text Generation - run(_:)

Source: https://docs.zetic.ai/api-reference/ios/ZeticMLangeLLMModel

Starts text generation for a given prompt. This method is synchronous and can throw errors.

```APIDOC
## Text Generation - `run(_:)`

### Description
Starts generation for a prompt.

### Method Signature
```swift
public func run(_ text: String) throws -> LLMRunResult
```

### Parameters

- **text** (`String`) - Required - The prompt to generate text from.

### Response

- **LLMRunResult** - The result of the text generation.

### Request Example

```swift
let result = try model.run("Explain on-device AI in one paragraph.")
```
```

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

### Run Text Generation

Source: https://docs.zetic.ai/api-reference/ios/ZeticMLangeLLMModel

Starts text generation for a given prompt using the initialized LLM model.

```swift
let result = try model.run("Explain on-device AI in one paragraph.")
```

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

### Face Detection Pipeline (Kotlin)

Source: https://docs.zetic.ai/tutorials/face-emotion-recognition

Initialize and run the face detection model. Preprocess the image, run the model, and postprocess to get face regions.

```kotlin
// (0) Initialize face detection model
val faceDetectionModel = ZeticMLangeModel(this, PERSONAL_KEY, "google/MediaPipe-Face-Detection")
val faceDetection = FaceDetectionWrapper()

// (1) Preprocess image
val faceDetectionInputs = faceDetection.preprocess(bitmap)

// (2) Run face detection model
val faceDetectionOutputs = faceDetectionModel.run(faceDetectionInputs)

// (3) Postprocess to get face regions
val faceDetectionPostprocessed = faceDetection.postprocess(faceDetectionOutputs)
```

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

### Full Working Example: iOS Model Inference

Source: https://docs.zetic.ai/api-reference/ios/Tensor

Demonstrates how to load a ZeticMLange model, prepare input data as a Tensor, run inference, and process the output Tensors in an iOS application.

```swift
import ZeticMLange

class ViewController: UIViewController {
    override func viewDidLoad() {
        super.viewDidLoad()

        do {
            let model = try ZeticMLangeModel(
                personalKey: PERSONAL_KEY,
                name: "Steve/YOLOv11_comparison"
            )

            // (1) Build an input tensor from preprocessed pixel bytes
            let pixels: [Float] = preparePixels()
            let data = pixels.withUnsafeBufferPointer { Data(buffer: $0) }
            let input = Tensor(
                data: data,
                dataType: BuiltinDataType.float32,
                shape: [1, 3, 640, 640]
            )

            // (2) Run inference
            let outputs = try model.run(inputs: [input])

            // (3) Read outputs as a typed array of Floats
            let output = outputs[0]
            let floatCount = output.count()
            let floats: [Float] = output.data.withUnsafeBytes {
                Array($0.bindMemory(to: Float.self).prefix(floatCount))
            }
        } catch {
            print("Melange error: \(error)")
        }
    }
}
```

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

### Face Detection Pipeline (Swift)

Source: https://docs.zetic.ai/tutorials/face-emotion-recognition

Initialize and run the face detection model. Preprocess the image, run the model, and postprocess to get face regions.

```swift
// (0) Initialize face detection model
let faceDetectionModel = try ZeticMLangeModel(personalKey: PERSONAL_KEY, name: "google/MediaPipe-Face-Detection")
let faceDetection = FaceDetectionWrapper()

// (1) Preprocess image
let faceDetectionInputs = faceDetection.preprocess(bitmap)

// (2) Run face detection model
let faceDetectionOutputs = try faceDetectionModel.run(inputs: faceDetectionInputs)

// (3) Postprocess to get face regions
let faceDetectionPostprocessed = faceDetection.postprocess(faceDetectionOutputs)
```

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

### Verify ZeticMLange Initialization

Source: https://docs.zetic.ai/platform-integration/ios/setup

Add this test to confirm the SDK is installed and configured correctly. The initializer may perform a network call on first use to download the model binary.

```swift
import ZeticMLange

// Test initialization
let model = try ZeticMLangeModel(personalKey: PERSONAL_KEY, name: MODEL_NAME)
```

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

### Text Generation - run

Source: https://docs.zetic.ai/api-reference/android/ZeticMLangeLLMModel

Starts text generation for a given prompt. This method initiates the LLM inference process.

```APIDOC
## Text Generation

### `run(text)`

Starts generation for a prompt.

```java
fun run(text: String): LLMRunResult
```

### Parameters

- **text** (`String`) - Required - The prompt text to generate from.

### Returns

- `LLMRunResult` - The result of the text generation.

### Example

```java
val result = model.run("Explain on-device AI in one paragraph.")
```
```

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

### iOS Swift Inference Mode Examples

Source: https://docs.zetic.ai/how-to-guides/inference-mode-selection

Initialize ZeticMLangeModel instances using different inference modes: .RUN_AUTO (default), .RUN_SPEED, and .RUN_ACCURACY.

```swift
// Default: Auto mode
// Speed first, but maintains SNR above 20dB
let modelDefault = try ZeticMLangeModel(
    personalKey: PERSONAL_KEY,
    name: MODEL_NAME,
    modelMode: .RUN_AUTO
)

// Speed First Mode
let modelFast = try ZeticMLangeModel(
    personalKey: PERSONAL_KEY,
    name: MODEL_NAME,
    modelMode: .RUN_SPEED
)

// Accuracy First Mode
let modelAccurate = try ZeticMLangeModel(
    personalKey: PERSONAL_KEY,
    name: MODEL_NAME,
    modelMode: .RUN_ACCURACY
)
```

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

### Install ZeticMLange Android SDK

Source: https://docs.zetic.ai/release-notes/changelog

Use this implementation for Android projects to integrate the ZeticMLange SDK version 1.9.0.

```gradle
implementation("com.zeticai.mlange:mlange:1.9.0")
```

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

### Face Emotion Recognition Pipeline (Kotlin)

Source: https://docs.zetic.ai/tutorials/face-emotion-recognition

Initialize and run the face emotion recognition model. Preprocess with detected face regions, run the model, and postprocess to get emotions.

```kotlin
// (0) Initialize face emotion recognition model
val faceEmotionRecognitionModel = ZeticMLangeModel(this, PERSONAL_KEY, "ElenaRyumina/FaceEmotionRecognition")
val faceEmotionRecognition = FaceEmotionRecognitionWrapper()

// (1) Preprocess with detected face regions
val faceEmotionRecognitionInputs = faceEmotionRecognition.preprocess(bitmap, faceDetectionPostprocessed)

// (2) Run face emotion recognition model
val faceEmotionRecognitionOutputs = faceEmotionRecognitionModel.run(faceEmotionRecognitionInputs)

// (3) Postprocess to get emotions
val faceEmotionRecognitionPostprocessed = faceEmotionRecognition.postprocess(faceEmotionRecognitionOutputs)
```

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

### Android: Face Landmark Model Initialization and Run

Source: https://docs.zetic.ai/how-to-guides/multi-model-pipeline

Initializes the face landmark model, preprocesses the image using detected face regions, runs inference, and postprocesses the output to get landmarks. This step follows face detection in a multi-model pipeline.

```kotlin
// Initialize face landmark model
val faceLandmarkModel = ZeticMLangeModel(this, PERSONAL_KEY, "face_landmark")
val faceLandmark = FaceLandmarkWrapper()

// Preprocess with detected face regions
val faceLandmarkInputs = faceLandmark.preprocess(bitmap, faceRegions)

// Run face landmark model
val faceLandmarkOutputs = faceLandmarkModel.run(faceLandmarkInputs)

// Postprocess to get landmarks
val landmarks = faceLandmark.postprocess(faceLandmarkOutputs)
```

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

### Initialize and Use Face Emotion Recognition (Android Kotlin)

Source: https://docs.zetic.ai/tutorials/face-emotion-recognition

Initialize the FaceEmotionRecognitionWrapper, preprocess a bitmap, run the model, and postprocess the outputs to get a result bitmap. This snippet is for Android development using Kotlin.

```kotlin
// (0) Initialize Face Emotion Recognition wrapper
val feature = FaceEmotionRecognitionWrapper()

// (1) Preprocess bitmap and get processed float array
val inputs = feature.preprocess(bitmap)

// ... run model ...

// (2) Postprocess to bitmap
val resultBitmap = feature.postprocess(outputs)
```

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

### Initialize and Use Face Emotion Recognition (iOS Swift)

Source: https://docs.zetic.ai/tutorials/face-emotion-recognition

Initialize the FaceEmotionRecognitionWrapper, preprocess a UIImage, run the model, and postprocess the outputs to get a result bitmap. This snippet is for iOS development using Swift.

```swift
import ZeticMLange

// (0) Initialize Face Emotion Recognition wrapper
let feature = FaceEmotionRecognitionWrapper()

// (1) Preprocess UIImage and get processed float array
let inputs = feature.preprocess(image)

// ... run model ...

// (2) Postprocess to UIImage
let resultBitmap = feature.postprocess(&outputs)
```

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

### Example Usage of Tensor

Source: https://docs.zetic.ai/api-reference/flutter/Tensor

Demonstrates creating a Tensor using a typed factory, printing the element count, and accessing its data as a Float32List.

```dart
final input = Tensor.float32View(
  Float32List.fromList([1.0, 2.0, 3.0, 4.0]),
  shape: const [1, 4],
);

print(input.count()); // 4
print(input.asFloat32List());
```

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

### Android: Face Detection Model Initialization and Run

Source: https://docs.zetic.ai/how-to-guides/multi-model-pipeline

Initializes the face detection model, preprocesses the image, runs inference, and postprocesses the output to get face regions. This is the first step in a face detection and landmark extraction pipeline.

```kotlin
// Initialize face detection model
val faceDetectionModel = ZeticMLangeModel(this, PERSONAL_KEY, "face_detection")
val faceDetection = FaceDetectionWrapper()

// Preprocess image
val faceDetectionInputs = faceDetection.preprocess(bitmap)

// Run face detection
val faceDetectionOutputs = faceDetectionModel.run(faceDetectionInputs)

// Postprocess to get face regions
val faceRegions = faceDetection.postprocess(faceDetectionOutputs)
```

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

### Face Emotion Recognition Pipeline (Swift)

Source: https://docs.zetic.ai/tutorials/face-emotion-recognition

Initialize and run the face emotion recognition model. Preprocess with detected face regions, run the model, and postprocess to get emotions.

```swift
// (0) Initialize face emotion recognition model
let faceEmotionRecognitionModel = try ZeticMLangeModel(personalKey: PERSONAL_KEY, name: "ElenaRyumina/FaceEmotionRecognition")
let faceEmotionRecognition = FaceEmotionRecognitionWrapper()

// (1) Preprocess with detected face regions
let faceEmotionRecognitionInputs = faceEmotionRecognition.preprocess(bitmap, faceDetectionPostprocessed)

// (2) Run face emotion recognition model
let faceEmotionRecognitionOutputs = try faceEmotionRecognitionModel.run(inputs: faceEmotionRecognitionInputs)

// (3) Postprocess to get emotions
let faceEmotionRecognitionPostprocessed = faceEmotionRecognition.postprocess(faceEmotionRecognitionOutputs)
```

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

### Initialize and Use Face Landmark Wrapper (Kotlin)

Source: https://docs.zetic.ai/tutorials/face-landmark

Initialize the Face Landmark wrapper, preprocess a bitmap, run the model, and postprocess the outputs to a bitmap.

```kotlin
// (0) Initialize Face Landmark wrapper
val feature = FaceLandmarkWrapper()

// (1) Preprocess bitmap and get processed float array
val inputs = feature.preprocess(bitmap)

// ... run model ...

// (2) Postprocess to bitmap
val resultBitmap = feature.postprocess(outputs)
```

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

### Install Python Packages for Whisper

Source: https://docs.zetic.ai/tutorials/speech-recognition-whisper

Installs the necessary Python packages for running OpenAI Whisper, including PyTorch, Transformers, and NumPy.

```bash
pip install torch transformers numpy
```

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

### Initialize and Use Face Landmark Wrapper (Swift)

Source: https://docs.zetic.ai/tutorials/face-landmark

Initialize the Face Landmark wrapper, preprocess a UIImage, run the model, and postprocess the outputs to a UIImage.

```swift
import ZeticMLange

// (0) Initialize Face Landmark wrapper
let feature = FaceLandmarkWrapper()

// (1) Preprocess UIImage and get processed float array
let inputs = feature.preprocess(image)

// ... run model ...

// (2) Postprocess to UIImage
let resultBitmap = feature.postprocess(&outputs)
```

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

### Get Tensor Element Count

Source: https://docs.zetic.ai/api-reference/android/Tensor

Returns the number of elements in the tensor.

```kotlin
fun count(): Int
```

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

### Lifecycle - cleanUp()

Source: https://docs.zetic.ai/api-reference/ios/ZeticMLangeLLMModel

Cleans up resources. It is recommended to call this before starting a fresh conversation.

```APIDOC
## Lifecycle - `cleanUp()`

### Method Signature
```swift
public func cleanUp() throws
```

### Description
Cleans up resources. Call `cleanUp()` before starting a fresh conversation.
```

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

### Initialize and Use Face Detection Wrapper (Kotlin)

Source: https://docs.zetic.ai/tutorials/face-detection

Initialize the Face Detection wrapper, preprocess a bitmap, and postprocess the model outputs to a bitmap in Kotlin.

```kotlin
// (0) Initialize Face Detection wrapper
val feature = FaceDetectionWrapper()

// (1) Preprocess bitmap and get processed float array
val inputs = feature.preprocess(bitmap)

// ... run model ...

// (2) Postprocess to bitmap
val resultBitmap = feature.postprocess(outputs)
```

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

### Initialize Model Once, Run Many

Source: https://docs.zetic.ai/how-to-guides/performance-optimization

Initialize the ZeticMLangeModel once to download and create the NPU context, then reuse the instance for multiple inferences.

```kotlin
// Do this once
val model = ZeticMLangeModel(context, PERSONAL_KEY, MODEL_NAME)

// Reuse for multiple inferences
for (frame in videoFrames) {
    val outputs = model.run(preprocessFrame(frame))
}
```

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

### LLMInitOption

Source: https://docs.zetic.ai/api-reference/android/enums-and-constants

Configuration options for initializing an LLM model, including KV cache cleanup policy and context size.

```APIDOC
## `LLMInitOption`

Configuration class for LLM initialization.

```kotlin
data class LLMInitOption(
    val kvCacheCleanupPolicy: LLMKVCacheCleanupPolicy = LLMKVCacheCleanupPolicy.CLEAN_UP_ON_FULL,
    val nCtx: Int = 2048,
)
```

### Fields

*   **`kvCacheCleanupPolicy`** (`LLMKVCacheCleanupPolicy`) - Default: `CLEAN_UP_ON_FULL`. In-memory KV-cache behavior when context storage is full.
*   **`nCtx`** (`Int`) - Default: `2048`. Requested context size. The runtime may normalize it.
```

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

### Get Tensor Size in Bytes

Source: https://docs.zetic.ai/api-reference/android/Tensor

Returns the size of the tensor in bytes, equivalent to data.capacity().

```kotlin
fun size(): Int
```

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

### Prepare Audio Input for Upload (Python)

Source: https://docs.zetic.ai/how-to-guides/custom-preprocessing

Loads an audio file, resamples it to 16kHz, and saves it as a float32 numpy array.

```python
# Example: preparing audio input for upload
import numpy as np
import librosa

# Load and resample audio to 16kHz
audio, sr = librosa.load("audio.wav", sr=16000)

# Convert to numpy and save
np_audio = audio.astype(np.float32)
np.save("audio_input.npy", np_audio)
```

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

### Lifecycle - resetSession

Source: https://docs.zetic.ai/api-reference/android/ZeticMLangeLLMModel

Resets the current session of the LLM. Recommended before starting a fresh conversation.

```APIDOC
```java
fun resetSession()
```

### Example

```java
model.resetSession()
```
```

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

### Initialize and Use Face Detection Wrapper (Swift)

Source: https://docs.zetic.ai/tutorials/face-detection

Initialize the Face Detection wrapper, preprocess a UIImage, and postprocess the model outputs to a UIImage in Swift.

```swift
import ZeticMLange
import ext

// (0) Initialize Face Detection wrapper
let feature = FaceDetectionWrapper()

// (1) Preprocess UIImage and get processed float array
let inputs = feature.preprocess(image)

// ... run model ...

// (2) Postprocess to UIImage
let resultBitmap = feature.postprocess(&outputs)
```

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

### get<T>(index)

Source: https://docs.zetic.ai/api-reference/android/Tensor

Reads a single element at the specified index and returns it as the requested numeric type.

```APIDOC
## `get<T>(index)`

Reads a single element at `index` as the requested numeric type.

```kotlin
inline fun <reified T : Number> get(index: Int): T
```

**Supported `T`:** `Int`, `Long`, `Float`, `Double`, `Char`, `Short`, `Byte`.

**Example:**
```kotlin
val first: Float = tensor.get<Float>(0)
```
```

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

### Initialize and Run LLM Models

Source: https://docs.zetic.ai/platform-integration/flutter/advanced-configuration

Utilize ZeticMLangeLLMModel for text generation, configuring context size and cache cleanup policy. Use waitForNextToken for token-by-token output.

```dart
final llm = await ZeticMLangeLLMModel.create(
  personalKey: personalKey,
  name: llmModelName,
  initOption: const LLMInitOption(
    nCtx: 4096,
    kvCacheCleanupPolicy: LLMKVCacheCleanupPolicy.cleanUpOnFull,
  ),
  onDownload: (progress) {
    print('Loading LLM ${(progress * 100).round()}%');
  },
);

llm.run('Explain on-device AI in one paragraph.');

while (true) {
  final next = llm.waitForNextToken();
  if (next.isFinished) {
    break;
  }
  print(next.token);
}

llm.cleanUp();
llm.close();
```

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

### LLMInitOption Struct

Source: https://docs.zetic.ai/api-reference/ios/enums-and-constants

Configuration options for initializing an LLM, including KV cache cleanup policy, context size, and run configuration ID.

```APIDOC
## `LLMInitOption`

```swift
public struct LLMInitOption {
  public var kvCacheCleanupPolicy: LLMKVCacheCleanupPolicy
  public var nCtx: Int
  public var runConfigId: Int64
}
```

### Fields

- **`kvCacheCleanupPolicy`** (`LLMKVCacheCleanupPolicy`) - Default: `.CLEAN_UP_ON_FULL` - In-memory KV-cache behavior when context storage is full.
- **`nCtx`** (`Int`) - Default: `2048` - Requested context size. The runtime may normalize it.
- **`runConfigId`** (`Int64`) - Default: `0` - Run configuration id used by compatible LFM-VL packages.
```

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

### DataType Protocol

Source: https://docs.zetic.ai/api-reference/ios/Tensor

The DataType protocol defines a way to get the byte size of a tensor element.

```APIDOC
## DataType
`DataType` is a protocol that exposes the byte size of a tensor element.
```swift
public protocol DataType {
    var size: Int { get }
}
```
```

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

### Initialize and Run LLM with Token Streaming (iOS Swift)

Source: https://docs.zetic.ai/introduction/key-concepts

Initialize ZeticMLangeLLMModel for large language models and handle token streaming. Requires personal key and model name.

```swift
let model = try ZeticMLangeLLMModel(personalKey: PERSONAL_KEY, name: MODEL_NAME)
try model.run("prompt")

while true {
    let result = model.waitForNextToken()
    if result.generatedTokens == 0 { break }
    print(result.token)
}
```

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

### Lifecycle - cleanUp

Source: https://docs.zetic.ai/api-reference/android/ZeticMLangeLLMModel

Cleans up resources associated with the LLM model. Recommended before starting a fresh conversation.

```APIDOC
## Lifecycle

```java
fun cleanUp()
```

### Example

```java
model.cleanUp()
```
```

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

### Initialize and Run LLM with Token Streaming (Android Kotlin)

Source: https://docs.zetic.ai/introduction/key-concepts

Instantiate ZeticMLangeLLMModel for large language models and process token streams. Requires context, personal key, and model name.

```kotlin
val model = ZeticMLangeLLMModel(context, PERSONAL_KEY, MODEL_NAME)
model.run("prompt")

while (true) {
    val result = model.waitForNextToken()
    if (result.generatedTokens == 0) break
    print(result.token)
}
```

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

### Clean Up LLM Resources

Source: https://docs.zetic.ai/api-reference/ios/ZeticMLangeLLMModel

Cleans up LLM resources, typically called before starting a new conversation.

```swift
try model.cleanUp()
```

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

### Instantiate Random Tensor

Source: https://docs.zetic.ai/api-reference/android/Tensor

Example of creating a random Tensor with Float32 data type and a specified shape.

```kotlin
val fakeInput = Tensor.random(DataType.Float32, intArrayOf(1, 3, 640, 640))
```

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

### Get Tensor Element by Index

Source: https://docs.zetic.ai/api-reference/android/Tensor

Reads a single element at the specified index as the requested numeric type.

```kotlin
inline fun <reified T : Number> get(index: Int): T
```

```kotlin
val first: Float = tensor.get<Float>(0)
```

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

### Initialize Model with Personal Key (Android Kotlin)

Source: https://docs.zetic.ai/model-deployment/understanding-model-keys

Demonstrates initializing a ZeticMLangeModel in Android (Kotlin) using a personal key stored in build configuration.

```kotlin
val model = ZeticMLangeModel(
    context,
    BuildConfig.PERSONAL_KEY,  // From build configuration
    "USERID/REPOSITORY_NAME"
)
```

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

### Reset KV State

Source: https://docs.zetic.ai/api-reference/ios/ZeticMLangeLLMModel

Resets the current KV state of the LLM, effectively starting a fresh conversation.

```swift
try model.resetKVState()
```

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

### Load Model, Prepare Inputs, and Run Inference

Source: https://docs.zetic.ai/platform-integration/ios/basic-inference

This snippet demonstrates the core steps for running inference: loading the Melange model, preparing input tensors, and executing the inference. Ensure input shapes match the model's requirements.

```swift
import ZeticMLange

// (1) Load model
// This handles model download (if needed) and Neural Engine context creation
let model = try ZeticMLangeModel(personalKey: PERSONAL_KEY, name: MODEL_NAME)

// (2) Prepare model inputs
// Ensure input shapes match your model's requirement (e.g., Float32 arrays)
let inputs: [Tensor] = [] // Prepare your inputs

// (3) Run Inference
// Executes the fully automated hardware graph.
// No manual delegate configuration or memory syncing required.
let outputs = try model.run(inputs: inputs)
```

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

### LLMInitOption Structure

Source: https://docs.zetic.ai/api-reference/ios/enums-and-constants

Defines initialization options for LLM models, including KV cache cleanup policy, context size, and run configuration ID.

```swift
public struct LLMInitOption {
  public var kvCacheCleanupPolicy: LLMKVCacheCleanupPolicy
  public var nCtx: Int
  public var runConfigId: Int64
}
```

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

### Install ZeticMLange Flutter SDK

Source: https://docs.zetic.ai/release-notes/changelog

Use this dependency for Flutter projects to integrate the ZeticMLange SDK version 1.9.1.

```yaml
zetic_mlange: ^1.9.1
```

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

### Initialize and Process Audio with Whisper (Swift)

Source: https://docs.zetic.ai/tutorials/speech-recognition-whisper

Initializes Whisper components, processes audio, runs the encoder, generates tokens, and decodes them into text using Swift.

```Swift
// Initialize components
let wrapper = WhisperFeatureWrapper()
let encoder = try ZeticMLangeModel(personalKey: PERSONAL_KEY, name: "OpenAI/whisper-tiny-encoder")
let decoder = try ZeticMLangeModel(personalKey: PERSONAL_KEY, name: "OpenAI/whisper-tiny-decoder")

// Process audio to features
let features = wrapper.process(input.audio)

// Run encoder
let outputs = encoder.process(features)

// Generate tokens using decoder
let generatedIds = decoder.process(outputs)

// Convert tokens to text
let text = wrapper.decodeToken(generatedIds, true)
return WhisperOutput(text: text)
```

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

### Instantiate Tensor with Specific Shape

Source: https://docs.zetic.ai/api-reference/android/Tensor

Example of creating a Tensor from a float array with a specified data type and shape.

```kotlin
val input = Tensor.of(
    data = floatArrayOf(/* ... */),
    dataType = DataType.Float32,
    shape = intArrayOf(1, 3, 640, 640),
)
```

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

### Conversation Reset - Android (Kotlin)

Source: https://docs.zetic.ai/llm-inference/streaming-generation

Clears the current conversation context and starts a new one with a fresh prompt.

```kotlin
model.cleanUp()
model.run("Start a new conversation")
```

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

### Initialize LLM Model with Processor and Quantization Filters (Android)

Source: https://docs.zetic.ai/llm-inference/inference-modes

Initialize the LLM model with `LLMModelMode.RUN_AUTO`, specifying `APType.GPU` and `LLMQuantType.GGUF_QUANT_Q4_K_M` to filter candidates. Unsupported combinations fail during initialization.

```kotlin
val model = ZeticMLangeLLMModel(
    context = context,
    personalKey = PERSONAL_KEY,
    name = MODEL_NAME,
    modelMode = LLMModelMode.RUN_AUTO,
    apType = APType.GPU,
    quantType = LLMQuantType.GGUF_QUANT_Q4_K_M,
)
```

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

### Initialize and Run General AI Model (Android Kotlin)

Source: https://docs.zetic.ai/introduction/key-concepts

Instantiate ZeticMLangeModel for general AI tasks and execute inference. Requires context, personal key, and model name.

```kotlin
val model = ZeticMLangeModel(context, PERSONAL_KEY, MODEL_NAME)
val outputs = model.run(inputs)
```

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

### Install ZeticMLange Flutter SDK (1.9.0)

Source: https://docs.zetic.ai/release-notes/changelog

Use this dependency for Flutter projects to integrate the ZeticMLange SDK version 1.9.0.

```yaml
zetic_mlange: ^1.9.0
```

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

### Android Kotlin Inference Mode Examples

Source: https://docs.zetic.ai/how-to-guides/inference-mode-selection

Instantiate ZeticMLangeModel with different inference modes: RUN_AUTO (default), RUN_SPEED, and RUN_ACCURACY.

```kotlin
// Default: Auto mode
// Speed first, but maintains SNR above 20dB
val modelDefault = ZeticMLangeModel(
    context = this,
    personalKey = PERSONAL_KEY,
    name = MODEL_NAME,
    modelMode = ModelMode.RUN_AUTO
)

// Speed First Mode
val modelFast = ZeticMLangeModel(
    context = this,
    personalKey = PERSONAL_KEY,
    name = MODEL_NAME,
    modelMode = ModelMode.RUN_SPEED
)

// Accuracy First Mode
val modelAccurate = ZeticMLangeModel(
    context = this,
    personalKey = PERSONAL_KEY,
    name = MODEL_NAME,
    modelMode = ModelMode.RUN_ACCURACY
)
```

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

### Initialize and Run LLM with Token Streaming (Flutter Dart)

Source: https://docs.zetic.ai/introduction/key-concepts

Create an instance of ZeticMLangeLLMModel for large language models and stream tokens. Requires personal key and model name.

```dart
final model = await ZeticMLangeLLMModel.create(
  personalKey: personalKey,
  name: modelName,
);
model.run('prompt');

while (true) {
  final result = model.waitForNextToken();
  if (result.isFinished) break;
  print(result.token);
}
```

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

### Initialize Model with Personal Key (iOS Swift)

Source: https://docs.zetic.ai/model-deployment/understanding-model-keys

Shows how to initialize a ZeticMLangeModel in iOS (Swift) using a personal key loaded from a configuration file.

```swift
let model = try ZeticMLangeModel(
    personalKey: Configuration.personalKey,  // From config
    name: "USERID/REPOSITORY_NAME"
)
```

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

### Tensor Example

Source: https://docs.zetic.ai/api-reference/flutter/Tensor

Demonstrates how to create and use a Tensor object in Flutter, including printing its element count and accessing its data as a Float32List.

```APIDOC
## Example

```dart
final input = Tensor.float32View(
  Float32List.fromList([1.0, 2.0, 3.0, 4.0]),
  shape: const [1, 4],
);

print(input.count()); // 4
print(input.asFloat32List());
```
```

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

### Initialize and Process Audio with Whisper (Kotlin)

Source: https://docs.zetic.ai/tutorials/speech-recognition-whisper

Initializes Whisper components, processes audio data, runs the encoder, generates tokens, and decodes them into text using Kotlin.

```Kotlin
// Initialize components
val whisper = WhisperFeatureWrapper()
val encoder = ZeticMLangeModel(this, PERSONAL_KEY, "OpenAI/whisper-tiny-encoder")
val decoder = ZeticMLangeModel(this, PERSONAL_KEY, "OpenAI/whisper-tiny-decoder")

// Process audio
val features = whisper.process(audioData)

// Run encoder
encoder.process(features)

// Generate tokens using decoder
val generatedIds = decoder.generateTokens(outputs)

// Convert tokens to text
val text = whisper.decodeToken(generatedIds.toIntArray(), true)
```

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

### Flutter Dart Inference Mode Examples

Source: https://docs.zetic.ai/how-to-guides/inference-mode-selection

Create ZeticMLangeModel instances with explicit inference modes: ModelMode.runAuto (default), ModelMode.runSpeed, and ModelMode.runAccuracy.

```dart
// Default: Auto mode
// Speed first, but maintains SNR above 20dB
final modelDefault = await ZeticMLangeModel.create(
  personalKey: personalKey,
  name: modelName,
  modelMode: ModelMode.runAuto,
);

// Speed First Mode
final modelFast = await ZeticMLangeModel.create(
  personalKey: personalKey,
  name: modelName,
  modelMode: ModelMode.runSpeed,
);

// Accuracy First Mode
final modelAccurate = await ZeticMLangeModel.create(
  personalKey: personalKey,
  name: modelName,
  modelMode: ModelMode.runAccuracy,
);
```

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

### Basic Streaming - iOS (Swift)

Source: https://docs.zetic.ai/llm-inference/streaming-generation

Initializes the LLM model, executes a prompt, and accumulates generated tokens. Ensure proper error handling for model operations.

```swift
let model = try ZeticMLangeLLMModel(personalKey: PERSONAL_KEY, name: MODEL_NAME)

try model.run(userPrompt)

var buffer = ""
while true {
    let result = model.waitForNextToken()
    if result.generatedTokens == 0 { break }
    buffer.append(result.token)
}

let output = buffer
```

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

### Initialize LLM Model for Auto Selection with GPU (iOS)

Source: https://docs.zetic.ai/llm-inference/inference-modes

Use `.RUN_AUTO` and specify `.GPU` to let the SDK select the best runtime and quantization for the device's GPU.

```swift
let modelForGpu = try await ZeticMLangeLLMModel(
  personalKey: PERSONAL_KEY,
  name: MODEL_NAME,
  modelMode: .RUN_AUTO,
  apType: .GPU
)
```

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

### Tensor Methods

Source: https://docs.zetic.ai/api-reference/flutter/Tensor

Provides methods to count elements, get the total size in bytes, and access tensor data as various typed lists.

```dart
int count()
int size()
Uint8List asUint8List()
Float32List asFloat32List()
Float64List asFloat64List()
Int32List asInt32List()
Int64List asInt64List()
```

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

### Face Detection Pipeline (Dart)

Source: https://docs.zetic.ai/tutorials/face-emotion-recognition

Initialize and run the face detection model. Preprocess the image, run the model, and postprocess to get face regions.

```dart
// (0) Initialize face detection model
final faceDetectionModel = await ZeticMLangeModel.create(
  personalKey: personalKey,
  name: 'google/MediaPipe-Face-Detection',
);

// (1) Preprocess image
final faceDetectionInputs = preprocessFaceDetectionImage(image);

// (2) Run face detection model
final faceDetectionOutputs = faceDetectionModel.run(faceDetectionInputs);

// (3) Postprocess to get face regions
final faceDetectionPostprocessed =
    postprocessFaceDetection(faceDetectionOutputs);
```

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

### Initialize and Run General AI Model (iOS Swift)

Source: https://docs.zetic.ai/introduction/key-concepts

Initialize ZeticMLangeModel for general AI tasks and execute inference. Requires personal key and model name.

```swift
let model = try ZeticMLangeModel(personalKey: PERSONAL_KEY, name: MODEL_NAME)
let outputs = try model.run(inputs: inputs)
```

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

### Basic Streaming - Android (Kotlin)

Source: https://docs.zetic.ai/llm-inference/streaming-generation

Initiates LLM model, runs a prompt, and collects generated tokens incrementally in a loop. Use this for basic token collection.

```kotlin
val model = ZeticMLangeLLMModel(context, PERSONAL_KEY, MODEL_NAME)

model.run(userPrompt)

val sb = StringBuilder()
while (true) {
    val result = model.waitForNextToken()
    if (result.generatedTokens == 0) break
    if (result.token.isNotEmpty()) sb.append(result.token)
}

val output = sb.toString()
```

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

### Basic Streaming - Flutter (Dart)

Source: https://docs.zetic.ai/llm-inference/streaming-generation

Creates an LLM model instance, starts generation with a prompt, and collects tokens. The `isFinished` property indicates completion.

```dart
final model = await ZeticMLangeLLMModel.create(
  personalKey: personalKey,
  name: modelName,
);

model.run(userPrompt);

final buffer = StringBuffer();
while (true) {
  final result = model.waitForNextToken();
  if (result.isFinished) {
    break;
  }
  buffer.write(result.token);
}

final output = buffer.toString();
```

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

### Convert Face Landmark Model to ONNX

Source: https://docs.zetic.ai/tutorials/face-landmark

Use tf2onnx to convert a TensorFlow Lite face landmark model to ONNX format. Ensure you have installed the tf2onnx package.

```bash
python -m tf2onnx.convert --tflite face_landmark.tflite --output face_landmark.onnx --opset 13
```

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

### Convert TFLite to ONNX Model

Source: https://docs.zetic.ai/tutorials/face-detection

Convert the MediaPipe Face Detection TFLite model to ONNX format using tf2onnx. Ensure Python 3.8+ and tf2onnx are installed.

```bash
pip install tf2onnx
python -m tf2onnx.convert --tflite face_detection_short_range.tflite --output face_detection_short_range.onnx --opset 13
```

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

### Load Model, Prepare Inputs, and Run Inference in Java

Source: https://docs.zetic.ai/platform-integration/android/basic-inference

This Java snippet demonstrates loading a Melange model, preparing input tensors, and running inference. It's crucial that input shapes align with the model's specifications.

```java
// (1) Load model
// This handles model download (if needed) and NPU context creation
ZeticMLangeModel model = new ZeticMLangeModel(CONTEXT, PERSONAL_KEY, MODEL_NAME);

// (2) Prepare model inputs
// Ensure input shapes match your model's requirement (e.g., Float32 arrays)
Tensor[] inputs = // Prepare your inputs;

// (3) Run Inference
// Executes the hardware-accelerated graph. This is a blocking call.
Tensor[] outputs = model.run(inputs);
```

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

### Flutter RAG Pipeline with Custom Retriever

Source: https://docs.zetic.ai/llm-inference/rag

Integrate a custom RagRetriever with the RagPipeline in Flutter. This example shows how to instantiate a retriever and use it within the pipeline to respond to queries.

```dart
final retriever = MyRetriever();

final rag = RagPipeline(
  retriever: retriever,
  llm: model,
  profile: const RagProfile.qwen25(),
);

await for (final token in rag.respond(query: 'What does ZeticMLange do?')) {
  append(token);
}
```

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

### Define and Register Weather Tool (Flutter)

Source: https://docs.zetic.ai/llm-inference/function-calling

Defines a tool specification for getting weather and registers it with the LLM model on Flutter. The executor returns a hardcoded weather result.

```dart
final weatherTool = LLMToolSpec(
  name: 'get_weather',
  description: 'Get the weather for a city.',
  parametersJson: '''
{
  "type": "object",
  "properties": {
    "city": { "type": "string" }
  },
  "required": ["city"]
}
''',
);

model.registerTool(weatherTool, (call) {
  return const LLMToolResult(content: '{"summary":"Sunny, 24C"}');
});
```

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

### Text Generation with run()

Source: https://docs.zetic.ai/api-reference/android/ZeticMLangeLLMModel

Initiate text generation for a given prompt using the run() method.

```kotlin
val result = model.run("Explain on-device AI in one paragraph.")
```

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

### Define and Register Weather Tool (iOS)

Source: https://docs.zetic.ai/llm-inference/function-calling

Defines a tool specification for getting weather and registers it with the LLM model on iOS. The executor returns a hardcoded weather result.

```swift
let weatherTool = LLMToolSpec(
  name: "get_weather",
  description: "Get the weather for a city.",
  parametersJson: """
  {
    "type": "object",
    "properties": {
      "city": { "type": "string" }
    },
    "required": ["city"]
  }
  """
)

try model.registerTool(weatherTool) { call in
  LLMToolResult(content: #"{"summary":"Sunny, 24C"}"#)
}
```

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

### iOS: Face Detection Model Initialization and Run

Source: https://docs.zetic.ai/how-to-guides/multi-model-pipeline

Initializes the face detection model, preprocesses the image, runs inference, and postprocesses the output to obtain face regions. This is the initial stage for face analysis in an iOS application.

```swift
// Initialize face detection model
let faceDetectionModel = try ZeticMLangeModel(personalKey: PERSONAL_KEY, name: "face_detection")
let faceDetection = FaceDetectionWrapper()

// Preprocess image
let faceDetectionInputs = faceDetection.preprocess(image)

// Run face detection
let faceDetectionOutputs = try faceDetectionModel.run(inputs: faceDetectionInputs)

// Postprocess to get face regions
let faceRegions = faceDetection.postprocess(faceDetectionOutputs)
```

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

### Define and Register Weather Tool (Android)

Source: https://docs.zetic.ai/llm-inference/function-calling

Defines a tool specification for getting weather and registers it with the LLM model on Android. The executor returns a hardcoded weather result.

```kotlin
val weatherTool = LLMToolSpec(
    name = "get_weather",
    description = "Get the weather for a city.",
    parametersJson = """
        {
          "type": "object",
          "properties": {
            "city": { "type": "string" }
          },
          "required": ["city"]
        }
    """.trimIndent()
)

model.registerTool(weatherTool) { call ->
    LLMToolResult(content = "{"summary":"Sunny, 24C"}")
}
```

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

### iOS: Face Landmark Model Initialization and Run

Source: https://docs.zetic.ai/how-to-guides/multi-model-pipeline

Initializes the face landmark model, preprocesses the image using detected face regions, runs inference, and postprocesses the output to extract landmarks. This follows the face detection step in a pipeline.

```swift
// Initialize face landmark model
let faceLandmarkModel = try ZeticMLangeModel(personalKey: PERSONAL_KEY, name: "face_landmark")
let faceLandmark = FaceLandmarkWrapper()

// Preprocess with detected face regions
let faceLandmarkInputs = faceLandmark.preprocess(image, faceRegions)

// Run face landmark model
let faceLandmarkOutputs = try faceLandmarkModel.run(inputs: faceLandmarkInputs)

// Postprocess to get landmarks
let landmarks = faceLandmark.postprocess(faceLandmarkOutputs)
```

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

### Export YOLOv11 Model to ONNX

Source: https://docs.zetic.ai/tutorials/object-detection-yolo

Export the YOLOv11 model to ONNX format using the ultralytics library. This script requires Python 3.8+ with ultralytics, opencv-python, and numpy installed.

```python
from ultralytics import YOLO
import torch

model = YOLO("yolo11n.pt")
model.export(format="onnx", opset=12, simplify=True, dynamic=False, imgsz=640)
```

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

### Configure Android Build Settings

Source: https://docs.zetic.ai/platform-integration/flutter/setup

Set Android `minSdk` to 24 or later, build for `arm64-v8a`, and enable legacy JNI packaging in `android/app/build.gradle.kts`.

```kotlin
android {
    defaultConfig {
        minSdk = maxOf(flutter.minSdkVersion, 24)
        ndk {
            abiFilters += "arm64-v8a"
        }
    }

    packaging {
        jniLibs {
            useLegacyPackaging = true
        }
    }
}
```

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

### Flutter Custom RagRetriever Implementation

Source: https://docs.zetic.ai/llm-inference/rag

Implement the RagRetriever interface in Flutter to provide custom retrieval logic. This example defines a class that fulfills the interface requirements for retrieving chunks.

```dart
final class MyRetriever implements RagRetriever {
  @override
  Future<List<RetrievedChunk>> retrieve(String query, {required int topK}) async {
    return const [
      RetrievedChunk(
        text: 'ZeticMLange runs optimized models on-device.',
        score: 0.94,
        source: 'docs',
      ),
    ];
  }
}
```

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

### ZeticMLangeLLMModel Constructor

Source: https://docs.zetic.ai/api-reference/android/ZeticMLangeLLMModel

Initializes the ZeticMLangeLLMModel with specified parameters. It loads an on-device LLM from the Melange registry and configures its behavior for inference.

```APIDOC
## Constructor

```java
ZeticMLangeLLMModel(
    context: Context,
    personalKey: String,
    name: String,
    version: Int? = null,
    modelMode: LLMModelMode = LLMModelMode.RUN_AUTO,
    apType: APType? = null,
    quantType: LLMQuantType? = null,
    cacheHandlingPolicy: ModelCacheHandlingPolicy = ModelCacheHandlingPolicy.REMOVE_OVERLAPPING,
    initOption: LLMInitOption = LLMInitOption(),
    onDownload: ((Float) -> Unit)? = null,
)
```

### Parameters

- **context** (`Context`) - Required - Android context used for cache and file access.
- **personalKey** (`String`) - Required - Personal key for accessing the model.
- **name** (`String`) - Required - Model name in `account_name/project_name` format.
- **version** (`Int?`) - Optional - Model version. `null` loads the latest version.
- **modelMode** (`LLMModelMode`) - Optional - Backend selection strategy. Defaults to `RUN_AUTO`.
- **apType** (`APType?`) - Optional - Optional processor filter.
- **quantType** (`LLMQuantType?`) - Optional - Optional quantization filter.
- **cacheHandlingPolicy** (`ModelCacheHandlingPolicy`) - Optional - Managed artifact cache cleanup policy. Defaults to `REMOVE_OVERLAPPING`.
- **initOption** (`LLMInitOption`) - Optional - LLM initialization options. Defaults to `LLMInitOption()`.
- **onDownload** (`((Float) -> Unit)?`) - Optional - Download progress callback from `0.0` to `1.0`.

### Example

```java
val model = ZeticMLangeLLMModel(
    context = context,
    personalKey = PERSONAL_KEY,
    name = "account_name/project_name",
    initOption = LLMInitOption(nCtx = 4096),
)
```
```

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

### Import ZeticMLangeLLMModel

Source: https://docs.zetic.ai/api-reference/android/ZeticMLangeLLMModel

Import the necessary class for using ZeticMLangeLLMModel.

```kotlin
import com.zeticai.mlange.core.model.llm.ZeticMLangeLLMModel
```

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

### Face Emotion Recognition Pipeline (Dart)

Source: https://docs.zetic.ai/tutorials/face-emotion-recognition

Initialize and run the face emotion recognition model. Preprocess with detected face regions, run the model, and postprocess to get emotions.

```dart
// (0) Initialize face emotion recognition model
final faceEmotionRecognitionModel = await ZeticMLangeModel.create(
  personalKey: personalKey,
  name: 'ElenaRyumina/FaceEmotionRecognition',
);

// (1) Preprocess with detected face regions
final faceEmotionRecognitionInputs =
    preprocessFaceEmotionImage(image, faceDetectionPostprocessed);

// (2) Run face emotion recognition model
final faceEmotionRecognitionOutputs =
    faceEmotionRecognitionModel.run(faceEmotionRecognitionInputs);

// (3) Postprocess to get emotions
final faceEmotionRecognitionPostprocessed =
    postprocessFaceEmotion(faceEmotionRecognitionOutputs);
```