Preparing Software Packages

Software Package Download

The CANN software supports offline installation and online installation (using Conda, Yum, Apt-get, and Pip). The method of obtaining software packages varies according to the installation type.

By downloading the software listed in the following table, you agree to the terms and conditions of the Huawei Enterprise End User License Agreement (EULA).

**Table 1** Software list
Software Type	Description	Software Package	How to Obtain
Combined driver & Toolkit package	It consists of the Ascend NPU driver and Toolkit, and supports one-click full installation or independent installation by specifying parameters.	Ascend-cann_<version>_linux-<arch>.run	Use wget to obtain the software package in one-click mode. For details, see the installation procedure.
ops	The CANN ops package integrates a comprehensive suite of libraries, including the operator foundational framework, operator libraries (covering math, nn, cv, and transformer), TBE operators, HCCL, HIXL, and the DVPP library. Specifically, it provides dynamic and static library files for single-operator API execution (e.g., aclnn-class APIs), operator source code, and kernel binaries, ultimately optimizing overall execution performance in high-performance computing (HPC) scenarios. Before installing the ops, the Toolkit of a compatible version must be installed. Select the ops software package corresponding to the running device.	Ascend-cann-<chip_type>-ops_<version>_linux-<arch>.run	Use wget to download software packages in one-click mode. For details, see the installation procedure. Click here to confirm the product and version information, obtain the packages, and upload them to any directory in the installation environment.
NNAL (Ascend Neural Network Acceleration Library)	The CANN NNAL comprises the Ascend Transformer Boost (ATB) and Ascend SiP Boost (SiP) libraries. ATB: improves the training and inference performance of foundation models. SiP: provides a suite of high-performance operators for signal processing, catering to AI model scenarios (Python) as well as Torch NPU Adapters and signal processing workloads (C++). Ensure that the Toolkit of the same version has been installed.	Ascend-cann-nnal_<version>_linux-<arch>.run

**Table 2** Software packages for independent installation
Software Type	Description	Software Package	How to Obtain
Ascend NPU driver	Deployed on the Ascend AI Processor, it manages and queries the processor, while providing interfaces such as processor control and resource allocation for the upper-layer CANN software.	Ascend-hdk-<chip_type>-npu-driver_<version>_linux-<arch>.run	Click here to confirm the product and version information, obtain the packages, and upload them to any directory in the installation environment, for example, /home. During online installation using the package manager, one-click online download and installation are supported, and you do not need to obtain the software packages.
Ascend NPU firmware	The firmware includes the built-in OS of the Ascend AI Processor, as well as control software for power devices and power consumption management devices. These components are used for model computing subsequently loaded onto the processor, processor startup control, and power consumption control, respectively.	Ascend-hdk-<chip_type>-npu-firmware_<version>.run
Toolkit	The CANN Toolkit is primarily used for training and inference workloads, model conversion, as well as the development and compilation of operators, applications, and models. During CANN deployment, both the Toolkit and ops software packages must be deployed.	Ascend-cann-toolkit_<version>_linux-<arch>.run	You can obtain the software packages for offline installation in either of the following ways: Use wget to download software packages in one-click mode. For details, see the installation procedure. Click here to confirm the product and version information, obtain the packages, and upload them to any directory in the installation environment, for example, /home. During online installation using the package manager, one-click online download and installation are supported, and you do not need to obtain the software packages. The container image already contains the CANN software package. You do not need to obtain it separately.
ops	The CANN ops package integrates a comprehensive suite of libraries, including the operator foundational framework, operator libraries (covering math, nn, cv, and transformer), TBE operators, HCCL, HIXL, and the DVPP library. Specifically, it provides dynamic and static library files for single-operator API execution (e.g., aclnn-class APIs), operator source code, and kernel binaries, ultimately optimizing overall execution performance in high-performance computing (HPC) scenarios. Before installing the ops, the Toolkit of a compatible version must be installed. Select the ops software package corresponding to the running device.	Ascend-cann-<chip_type>-ops_<version>_linux-<arch>.run
NNAL (Ascend Neural Network Acceleration Library)	The CANN NNAL comprises the ATB and SiP libraries. ATB: improves the training and inference performance of foundation models. SiP: provides a suite of high-performance operators for signal processing, catering to AI model scenarios (Python) as well as Torch NPU Adapters and signal processing workloads (C++). Ensure that the Toolkit of the same version has been installed.	Ascend-cann-nnal_<version>_linux-<arch>.run

**Table 3** Software packages
Software Type	Description	Software Package	How to Obtain
Ascend NPU driver	Deployed on the Ascend AI Processor, it manages and queries the processor, while providing interfaces such as processor control and resource allocation for the upper-layer CANN software.	Ascend-hdk-<chip_type>-npu-driver_<version>_linux-<arch>.run	Click here to confirm the product and version information, obtain the packages, and upload them to any directory in the installation environment.
Ascend NPU firmware	The firmware includes the built-in OS of the Ascend AI Processor, as well as control software for power devices and power consumption management devices. These components are used for model computing subsequently loaded onto the processor, processor startup control, and power consumption control, respectively.	Ascend-hdk-<chip_type>-npu-firmware_<version>.run
Toolkit	The pip-installed CANN Toolkit is intended only for framework-side training and inference workloads. If you need to install the Toolkit for development and debugging (such as development and compilation of operators, applications, and models), select another installation method. During CANN deployment, both the Toolkit and ops software packages must be deployed.	Ascend-cann-toolkit_<version>_linux-<arch>.run	During online installation using pip, one-click online download and installation are supported, and you do not need to obtain the software packages.
ops	The CANN ops package integrates a comprehensive suite of libraries, including the operator foundational framework, operator libraries (covering math, nn, cv, and transformer), TBE operators, HCCL, HIXL, and the DVPP library. Specifically, it provides dynamic and static library files for single-operator API execution (e.g., aclnn-class APIs), operator source code, and kernel binaries, ultimately optimizing overall execution performance in high-performance computing (HPC) scenarios. Before installing the ops, the Toolkit of a compatible version must be installed. Select the ops software package corresponding to the running device.	Ascend-cann-<chip_type>-ops_<version>_linux-<arch>.run

<chip_type> indicates the AI processor type.
<version> indicates the software version.
<arch> indicates the CPU architecture.

Verifying the Digital Signature

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

Click PGP Verify to obtain the tool package, decompress it and verify the PGP digital signature. For details, see OpenPGP Signature Verification Guide. If the verification fails, do not use the software package. Visit the Support website to seek help or submit a service ticket.