eIQ ONNX2TFLite

eIQ ONNX2TFLite is a tool used to convert neural networks to the TF Lite format, and optionally quantize them in the ONNX format using the module ONNX2Quant.

Note

Quantization using ONNX2Quant is to be followed with conversion to TF Lite and deployment to eIQ Neutron NPU. It is not a general-purpose quantization tool and it should not be used with other NPU’s or different runtimes such as ONNX Runtime.

This guide will show you how to:

• Install the ONNX2TFLite package

• Use the tool to quantize an ONNX model using ONNX2Quant

• Use the tool to convert an ONNX model to TF Lite using ONNX2TFLite