---
title: 环境配置
description: "在进行源码编译构建前需要安装对应版本的依赖组件。"
url: https://www.hikunpeng.com/document/detail/zh/kunpengai/ecosystemEnable/paddlepaddle/kunpengpaddlepaddle_16_0004.html
sourcePath: /source/zh/kunpengai/ecosystemEnable/paddlepaddle/kunpengpaddlepaddle_16_0004.html
indexId: 0b829852a07096c772652d17a42c6d8fe3e167a79b3107fbc57d12a2f00e01b782
---
# 环境配置

在进行源码编译构建前需要安装对应版本的依赖组件。

1. 安装最小构建依赖。
  1 2 3 4 5 6 7 8 dnf install -y \ --setopt=install_weak_deps=False \ --setopt=tsflags=nodocs \ git gcc gcc-c++ gfortran make cmake ninja-build \ python3 python3-pip python3-devel \ openssl-devel bzip2-devel zlib-devel xz-devel libffi-devel \ sqlite-devel readline-devel patchelf tar xz wget curl \ findutils diffutils patch which file openblas-devel lapack-devel

2. 安装Python构建依赖。
  1 2 3 4 5 python3 -m pip install --upgrade pip setuptools wheel python3 -m pip install \ numpy==2.4.6 \ pybind11 Cython pyyaml packaging requests six protobuf \ typing_extensions decorator opt_einsum networkx astor gast Pillow ninja

3. 设置CUDA编译环境，以下路径以CUDA 13.0默认安装位置为例。
  1 2 3 4 5 6 7 export CUDA_HOME=/usr/local/cuda-13.0 export CUDA_PATH=/usr/local/cuda-13.0 export PATH=${CUDA_HOME}/bin:${PATH} export LD_LIBRARY_PATH=${CUDA_HOME}/lib64:${LD_LIBRARY_PATH:-} nvcc --version ldconfig -p | grep libcuda.so.1

4. 确认Python ABI。
  1 2 3 4 5 6 python3 --version python3 - <<'PY' import sysconfig print(sysconfig.get_config_var("SOABI")) PY

  预期SOABI包含以下信息：

  1 cpython-311-aarch64-linux-gnu