Model Management¶

Managing, downloading, and organizing your trained wake word models.

Model Dashboard¶

The home page displays all your wake word models:

Model Cards¶

Each model card shows:

• Wake word name: The trigger phrase
• Type badge: "Custom" or "Default"
• Creation date: When the model was trained
• Accuracy: Validation accuracy
• F1 Score: Balanced performance metric
• Action buttons: Test, Download, Delete

Filtering Models¶

Use the filter buttons to view:

Model Storage¶

Directory Structure¶

Models are stored in the models/ directory:

models/
├── default/              # Pre-trained default models
│   └── assistant/
│       ├── assistant.pt
│       └── assistant.json
└── custom/               # User-trained models
    └── jarvis/
        ├── jarvis.pt
        ├── jarvis.json
        └── jarvis.onnx

Model Files¶

Each model consists of:

Metadata JSON¶

Example model metadata:

{
  "wake_word": "jarvis",
  "model_type": "ast",
  "created_at": "2026-01-06T15:00:00Z",
  "threshold": 0.65,
  "accuracy": 0.971,
  "f1_score": 0.943,
  "precision": 0.952,
  "recall": 0.934,
  "trainable_params": 234567,
  "training_config": {
    "batch_size": 32,
    "learning_rate": 0.0001,
    "dropout": 0.5,
    "epochs_trained": 45,
    "positive_samples": 5000,
    "negative_samples": 10000
  },
  "recordings": [
    "recording_001.wav",
    "recording_002.wav",
    "recording_003.wav"
  ]
}

Downloading Models¶

From the Dashboard¶

1. Click the Download button on a model card
2. A ZIP file will be downloaded

Download Contents¶

jarvis_model.zip
├── jarvis.pt        # PyTorch weights
├── jarvis.json      # Metadata
└── jarvis.onnx      # ONNX model (if available)

Via API¶

# Download model as ZIP
curl -O http://localhost:8000/api/models/jarvis/download

# Get model file directly
curl -O http://localhost:8000/api/models/jarvis/file

Deleting Models¶

From the Dashboard¶

1. Click the Delete button (trash icon) on a model card
2. Confirm the deletion in the dialog

Via API¶

curl -X DELETE http://localhost:8000/api/models/jarvis

Model API Reference¶

List All Models¶

GET /api/models

Response:

{
  "models": [
    {
      "id": "jarvis",
      "wake_word": "jarvis",
      "type": "custom",
      "created_at": "2026-01-06T15:00:00Z",
      "accuracy": 0.971,
      "f1_score": 0.943,
      "threshold": 0.65
    }
  ],
  "total": 1
}

Get Model Details¶

GET /api/models/{model_id}

Response:

{
  "id": "jarvis",
  "wake_word": "jarvis",
  "type": "custom",
  "created_at": "2026-01-06T15:00:00Z",
  "threshold": 0.65,
  "metrics": {
    "accuracy": 0.971,
    "f1_score": 0.943,
    "precision": 0.952,
    "recall": 0.934
  },
  "training_config": { ... },
  "has_onnx": true
}

Download Model¶

GET /api/models/{model_id}/download

Returns a ZIP file containing all model files.

Delete Model¶

DELETE /api/models/{model_id}

ONNX Export¶

What is ONNX?¶

ONNX (Open Neural Network Exchange) is a cross-platform format for deploying ML models. Benefits include:

• Platform-independent (Windows, Linux, macOS, mobile)
• Optimized runtime performance
• Compatible with WakeEngine

Export Methods¶

Automatic Export¶

ONNX models are automatically created after training if the export succeeds.

Manual Export¶

python scripts/export_onnx.py --model jarvis

Via API¶

curl -X POST http://localhost:8000/api/models/jarvis/export-onnx

Export Status¶

Check if ONNX export is available:

{
  "has_onnx": true,
  "onnx_size_bytes": 2450000
}

Importing Models¶

Manual Import¶

1. Copy model files to models/custom/{wake_word}/
2. Ensure both .pt and .json files are present
3. Refresh the dashboard

Required Files¶

Model Versioning¶

Naming Convention¶

WakeBuilder uses the wake word as the model name. To create multiple versions:

• Train with different parameters
• Rename the model folder manually
• Keep backups of previous versions

Manual Backup¶

# Backup a model
cp -r models/custom/jarvis models/custom/jarvis_v1

# Restore from backup
cp -r models/custom/jarvis_v1 models/custom/jarvis

Troubleshooting¶