Introduction

Purpose

This document provides the user guide for the Qualcomm Neural Network (QNN) Delegate for TensorFlow 1 Lite (TFLite).

Note

Qualcomm® AI Engine Direct Delegate is also refered to as Qualcomm Neural Network (QNN) Delegate.

The Qualcomm® AI Engine Direct Delegate is a way to delegate parts or all of a TFLite model to Qualcomm® AI Engine Direct runtime, decreasing the workload of the CPU while achieving better performance and power efficiency. This specific delegate is backed by Qualcomm’s Neural Network API.

Compatibility

The Qualcomm® AI Engine Direct Delegate has been developed and tested with a certain TFLite runtime. Please check Release Notes.

Conventions

Commands and code samples appear in a specially formatted code section. For example:

output = (input - offset) * scale.

Environment variables appear preceded by $, for example $LD_LIBRARY_PATH.

The environment variable $QNN_SDK_ROOT will be used to represent the root directory of the Qualcomm® AI Engine Direct SDK.

$ export QNN_SDK_ROOT=<path_to_qnn_sdk_installation>

Release Notes

Release notes are available as QNN_TFLITE_DELEGATE_ReleaseNotes.txt in the SDK.

1

TensorFlow, the TensorFlow logo and any related marks are trademarks of Google Inc.