Preparing Software Packages

You can use either of the following methods to prepare the software packages to be installed (only method 1 is supported for MindX DL and MEF Center package preparation, and only method 2 is supported for MindIO and Apex package preparation):

Method 1: Use ascend-deployer to download the software packages listed in Table 1 online. For details, see Performing Online Download.

Method 2: Obtain the software packages in other ways.

Prepare the software packages to be installed based on the installation scenario. For details about how to obtain them, see Table 1. You need to download the required software packages and corresponding digital signature files, and manually upload them to specified directories. Ensure that each software package in its specified directory has only one version. Otherwise, there may be a mismatch of versions.

Some software is classified into the commercial edition and community edition. The functions of the two editions are the same except the download permission and use of purpose. The community edition can be downloaded directly without applying for related permissions, but it cannot be used for commercial purposes. To download the commercial edition, you need to apply for related permissions.

Once you download this software, you agree to the terms and conditions of Huawei Enterprise End User License Agreement (EULA).

**Table 1** Software packages
Name	Software Package	Description	How to Obtain	Specified Target Directory
Driver and firmware packages	Axxx --npu_- {arch}.zip, Axxx--driver_-{arch}.run, Axxx--firmware_-{arch}.run	NPU firmware and driver are included. Software packages in *Ascend-hdk--npu_-* **{arch}.zip format support the installation of firmware and drivers for multiple products, and are recommended. For products such as Atlas 300I Pro, Atlas 300V, Atlas 300I Duo, and Atlas 300V Pro, you can use the Ascend-hdk-310p-npu_-* **{arch}.zip** package to install the firmware and driver.	Commercial edition: Link Community edition: Link	resources directory of the ascend-deployer tool
Development kit	Ascend-cann-toolkit _{version} _linux-{arch}.zip or Ascend-cann-toolkit _{version} _linux-{arch}.run	The development kit is used for application development, operator customization, and model conversion. It contains the library files required for developing applications and auxiliary development tools such as the ATC for model conversion.	Link
Offline inference engine package	Ascend-cann-nnrt_{version}_linux-{arch}.zip or Ascend-cann-nnrt_{version}_linux-{arch}.run	This package is used for model inference of applications. It includes AscendCL libraries and compilation dependencies (excluding the libraries in the driver package). It supports offline inference only, and can be installed in the Atlas Intelligent Edge Solution scenario.
Deep learning engine package	Ascend-cann-nnae_{version}_linux-{arch}.zip or Ascend-cann-nnae_{version}_linux-{arch}.run	This package contains the Fwklib and OPP. It supports offline and online inference, and can be installed in the Ascend Data Center Solution scenario.
Binary OPP	Ascend-cann-kernels-{chip_type}_{version}_linux.zip or Ascend-cann-kernels-{chip_type}_{version}_linux.run	(Optional) CANN binary OPP, which is used in scenarios where dynamic-shape networks are involved. It is supported in CANN 6.0.1 and later versions.
Framework plugin package	Ascend-cann -tfplugin_ {version} _linux-{arch}.zip or Ascend-cann -tfplugin_ {version} _linux-{arch}.run	(Optional) Plugin package, an adaptation plugin used to connect to the upper-layer framework TensorFlow. If the deep learning framework TensorFlow is used for online inference or training, you need to obtain the software package.	Commercial edition: Link Community edition: Link
Toolbox	Ascend-mindx-toolbox_ {version} _linux-{arch}.zip or Ascend-mindx-toolbox_ {version} _linux-{arch}.run	The Ascend-DMI tool is included. To check the CANN running status after the installation, this Toolbox must be installed. The Toolbox of a version earlier than MindX 3.0.0 mainly includes the container engine plugin Ascend Docker Runtime and the Ascend-DMI tool.	Commercial edition: Link Community edition: Link
Full-pipeline development toolchain	MindStudio_{version}_linux.tar.gz	(Optional) One-stop environment for AI development and supports jobs for model, operator, and application development.	Link
MindIO package	Ascend-mindxdl-mindio_{version}_linux-{arch}.run	MindIO accelerates checkpoint saving and loading during training of foundation models. Note that ascend-deployer can install only the MindIO package of MindX 5.0.0 or later.	Link	resources directory of the ascend-deployer tool
PyTorch package	Versions 1.8.1 and 1.11.0: torch_npu-{version}_{arch}.whl torch-{version}_{arch}.whl	(Optional) It inherits open-source PyTorch features and adapts to Ascend AI Processors, enabling Ascend AI Processors to perform dynamic graph training and support custom operator development and model adjustment.	Torch_npu and official Torch package: Link	resources/pylibs directory of the ascend-deployer tool
Apex mixed precision module	apex-{version}_{arch}.whl	(Optional) The Apex mixed precision module is a comprehensive optimization library that features high optimization performance and precision. It provides mixed precision training in different scenarios. It is installed based on the version mapping with PyTorch.	Compile the Apex package according to the instructions at Gitee.
TensorFlow package	tensorflow-{version}_{arch}.whl	(Optional) It inherits open-source TensorFlow features and adapts to Ascend AI Processors, enabling Ascend AI Processors to perform static graph training and support custom operator development and model adjustment.	Link
MindSpore package	mindspore_ascend-{version}-linux_{arch}.whl	(Optional) MindSpore is a deep learning framework for all scenarios, aiming to achieve easy development, efficient execution, and all-scenario coverage.	Link	resources/pylibs/Ascend directory of the ascend-deployer tool

{version} indicates the software version, and {arch} indicates the CPU architecture.
Verify the digital signatures of software packages.

To prevent a software package from being maliciously tampered with during transmission or storage, download the corresponding digital signature file for integrity verification when downloading the software package.

After the software package is downloaded from the Huawei Support website, verify its PGP digital signature. See OpenPGP Signature Verification Guide. If the verification fails, do not use the software package, and contact Huawei technical support engineers.

Before a software package is used in installation or update, its digital signature also needs to be verified according to OpenPGP Signature Verification Guide to ensure that the software package is not tampered with.

For carriers: http://support.huawei.com/carrier/digitalSignatureAction.

For enterprise users: https://support.huawei.com/enterprise/en/tool/pgp-verify-TL1000000054.

To deploy an image, obtain the required image from Ascend Hub, save it as an image file in the specified format (*_x86_64.tar or *_aarch64.tar), and save it to the ascend-deployer/ascend_deployer/resources/docker_images directory (you need to create this directory).
If the target device is A500 Pro-3000 and the IEF agent needs to be installed, the following conditions must be met for IEF to register and manage edge nodes:
- The IEF server is running properly, and the network between edge nodes and IEF is accessible. For details about other restrictions, such as requirements on device specifications, see IEF User Guide.
- Obtain the node certificate, EdgeCore Register, and EdgeCore Installer. For details, see "Creating a Batch Node Registration Job" in Batch Node Registration. Place the obtained node certificate, EdgeCore Register, and EdgeCore Installer in the resources directory.
- Ensure that AtlasEdge on the edge node works properly. You need to install AtlasEdge on the A500 Pro-3000.
- Ensure the validity and integrity of the node certificate. If the certificate has been modified, damaged, or expired, the IEF agent will fail to be installed.

Parent topic: Installation Process