ModelSDK

The ModelSDK is an integral part of the Palette software. It is used to convert trained floating-point models created by one of the supported ML Frameworks into a SiMa.ai proprietary format that can be executed on an MLSoC device as part of a complete embedded vision pipeline.

The main functions of the ModelSDK are:

  • Loading a floating-point model from one of the supported frameworks and formats - (PyTorch, Keras HDF5, TensorFlow1, ONNX, and TFLite).

  • Quantizing the floating-point model into an 8-bit integer (int8) format or a 16-bit (int16) format.

  • Evaluating the quantized model.

  • Compiling the quantized model into an inference model that can be executed on an MLSoC device.

../../_images/ModelSDK.png
Loading Floating-Point Models

How to load a floating-point model and the required parameters.

Machine Learning Model Quantization

How to import and quantize a trained model.

Compiling the Quantized Model

How to convert a quantized model into a binary format that can be executed on the MLSoC device.