1   Introduction

The offline installation tool provides automatic download of system components, third-party Python dependencies, and one-click installation functionality. It also supports the installation of drivers, firmware, CANN software packages, and MindCluster.

2   Disclaimer

This code repository contains multiple development branches that may include unfinished, experimental, or untested features. These branches should not be used in any production environments or projects that rely on critical business before official releases. To ensure stability and security, please use only the officially released versions of our software. The project and its contributors are not responsible for any issues, losses, or data corruption caused by the use of development branches.

3   Open Source Community Overview

For information on the MindCluster component open-source community, including community cycle meetings and component version update notifications, please refer to theAscend MindCluster Open Source Component Community Overview <https://www.hiascend.com/forum/thread-0241161684752207015-1-1.html>__.

4   Quick Guide

This tool primarily supports initial deployment scenarios (where only the OS is installed). If you encounter any issues, please refer to theascend-deployer User Guide <https://www.hiascend.com/document/detail/zh/mindx-dl/600/ascenddeployer/ascenddeployer/deployer_0001.html>__.

5   Installation Contents

The installation content supported by this tool is shown in the following diagram. The red parts must be installed on Ascend devices, the blue and green parts are optional. If you want to run AI frameworks or Toolbox directly, the green parts are also required.

6   Usage Example

Here is a quick demonstration of how to use the tool with an Ubuntu_18.04_x86_64 system and an Atlas 300I Pro inference card installed in the server.

1. Log in to the target device as the root user.
2. Run pip3 install ascend-deployer==<version> to install the ascend-deployer tool. <version> refers to the version of the ascend-deployer tool. It is recommended to download the latest released version from theofficial website <https://pypi.org/project/ascend-deployer/#history>__. If the pip3 command is missing, please install pip3.
3. Run the download command.

ascend-download --os-list=Ubuntu_18.04_x86_64 --download=CANN==6.3.RC1,MindSpore==2.0.0rc1

After executing the above command, the tool will automatically download the dependencies required for the OS, the CANN software package, corresponding drivers and firmware, and Docker software, and store them in the $HOME/ascend-deployer/resources directory.

4. When installing drivers and firmware, if they are not unified packages (package names starting with Ascend-hdk), you need to configure the cus_npu_info variable in the inventory_file in advance.

[ascend] localhost ansible_connection='local' cus_npu_info="300i-pro"

5. Execute the following installation command to automatically install all available packages (excluding the MindCluster cluster scheduling components, which should be installed by referring to theascend-deployer User Guide <https://www.hiascend.com/document/detail/zh/mindx-dl/600/ascenddeployer/ascenddeployer/deployer_0001.html>__).

ascend-deployer --install-scene=auto

6. Post-installation configuration. Before using Python and CANN, you need to configure environment variables according to your actual scenario. The following commands configure the environment variables for Python and NNAE as examples.

source /usr/local/ascendrc # Configure Python environment variables source /usr/local/Ascend/nnae/set_env.sh # Configure NNAE environment variables

7. For detailed instructions, please refer to theascend-deployer User Guide <https://www.hiascend.com/document/detail/zh/mindx-dl/600/ascenddeployer/ascenddeployer/deployer_0001.html>__.