Introduction

This document describes how to develop TensorFlow on Kunpeng servers, including source code build, installation, and basic development verification.

TensorFlow has the following advantages: static computational graphs and immediate execution, end-to-end machine learning platform capabilities, and powerful production and deployment toolchains (such as TFX and TF Serving). When TensorFlow is deployed on a Kunpeng server, the non-intrusive optimization patches provided by Kunpeng BoostKit can be used to automatically inject high-performance C++ operators (such as FusedMatMul and KPSoftmax) without modifying user code.