Export Formats and Hyperparameters

Matrice.ai supports multiple formats for exporting your trained models. Here’s an overview of each format and the associated hyperparameters you can adjust.

Supported Export Formats

• ONNX: Open Neural Network Exchange format for cross-platform compatibility.

• OpenVino: Intel’s optimized format for inference on their hardware.

• TorchScript: Serialized and optimized format for PyTorch models.

• PyTorch: An open-source machine learning library developed by Facebook’s AI Research lab. It’s known for its flexibility and dynamic computational graphs.NVIDIA’s platform for high-performance deep learning inference. It optimizes trained deep learning models to produce highly optimized runtime engines.NVIDIA’s platform for high-performance inference.

• TensorFlow: An open-source machine learning framework developed by Google. It’s widely used for both research and production.

  Note : Export formats are model-dependent; some models offer all export options, while others have limited export options.

Hyperparameters

Key Name	Value Type	Default Value	Predefined Values	Description
dynamic	Boolean	False	[True, False]	Controls dynamic axes for variable input sizes.
simplify	Boolean	False	[True, False]	Simplifies the model structure during export.
int8	Boolean	False	[True, False]	Enables INT8 quantization for smaller, faster models.
nms	Boolean	False	[True, False]	Includes Non-Maximum Suppression for object detection models.
optimize	Boolean	False	[True, False]	Optimizes the model graph for better performance.