Introduction

This document describes how to develop PaddlePaddle on Kunpeng servers, including building the source code, installing the PaddlePaddle AArch64 GPU wheel, and verifying GPU computing.

PaddlePaddle supports the full-stack AI development process from algorithm research to production implementation with its core features: unified programming paradigm for both static and dynamic modes, architecture design that integrates training and inference, and an industrial-grade toolchain throughout the entire process.

In the openEuler AArch64 GPU scenario, source code build products are delivered. The build and installation sides must be compatible with the Python ABI, Glibc, CUDA, and GPU architectures.