Software Packages and OSs That Can Be Downloaded

See Download Precautions for restrictions when using MindCluster Ascend Deployer for download, Supported OSs and Architectures for the value range of <OS>, and Table 2 for the value range of <PK>.

Download Precautions

Due to the need to download a large number of software packages, parallel download is enabled in MindCluster Ascend Deployer by default. You can modify the parallel_download field in the ascend_deployer/downloader/config.ini file to determine whether to enable parallel download. The value 1 indicates that this function is enabled, and the value 0 indicates that this function is disabled. By default, the number of concurrent download threads is 16. You can change the number by setting the ASCEND_DEPLOYER_DOWNLOAD_MAX_SIZE environment variable.
MindCluster Ascend Deployer provides the repository configuration file. The Huawei repository is used by default. Replace it as required. For details, see Configuring Repositories.
During OS dependency download, the software packages of MPI 3.2.1 and Open MPI 4.1.5 are automatically downloaded to compile and install the hccl-test tool. During the installation, one of them is selected for compilation based on the configured IP address type of the target environment. When the IPv4 format is used, install MPI 3.2.1. When the IPv6 format or Atlas A3 training product is used, install Open MPI 4.1.5.

Supported OSs and Architectures

**Table 1** OSs and architectures supported by MindCluster Ascend Deployer
Supported OSs	OS Version	OS Architecture	Restrictions
BCLinux	21.10	AArch64	-
BCLinux	21.10U4	AArch64	-
CTyunOS	22.06	AArch64	-
CTyunOS	23.01	AArch64	When MindCluster components are installed on CTyunOS, only some components of MindCluster 7.1.RC1 or later, Kubernetes 1.19.16 to 1.28.x, and Docker 18.09.x or later are supported. (If Kubernetes 1.24 or later is used, cri-dockerd needs to be installed.) The Atlas 300I Pro inference card and Atlas 300I Duo inference card support the NPU firmware and driver of 25.0.RC1 and later versions, and CANN 8.1.RC1 and later versions.
CULinux	3.0	AArch64	-
CentOS	7.6	AArch64/x86_64	When CentOS is used as the executor OS and MindCluster Ascend Deployer is used for batch deployment, devices running Ubuntu cannot be deployed. A kernel patch needs to be added for CentOS 7.6 x86_64 of an earlier kernel version (earlier than 4.5) of the Atlas 300T training card. Otherwise, the firmware may fail to be installed. Multiple versions of Docker/containerd may coexist on CentOS. Therefore, you are advised to check whether Docker has been installed before installing system dependencies on CentOS 7.6. If Docker has been installed, run the yum -y remove docker-ce, yum -y remove docker-ce-cli, and yum -y remove containerd.io commands to uninstall it and then install system dependencies. The built-in Python version of CentOS is too early. You need to install Python 3 and then use the download and installation functions of MindCluster Ascend Deployer.
Debian	10	AArch64	Only Ascend HDK, CANN, ToolBox, and MindCluster Ascend FaultDiag can be installed on the Atlas 900 A3 SuperPoD.
EulerOS	2.8	AArch64	By default, the root user is not allowed to remotely log in to OSs such as EulerOS. Therefore, you need to set PermitRootLogin in the /etc/ssh/sshd_config file to yes before the installation. The configuration method varies according to OSs. For details, see the official guide of the OS. After the installation is complete, disable the remote connection of the root user in a timely manner. EulerOS 2.12 does not download system dependencies. Install them by yourself. EulerOS 2.12 does not support the installation of sys_pkg.
	2.9	AArch64/x86_64
	2.10	AArch64/x86_64
	2.12	AArch64
Kylin	V10(GFB)	AArch64	-
	V10Tercel(SP1)	AArch64/x86_64
	V10Sword(SP2)	AArch64
	V10Lance(SP3)	AArch64
	V10Halberd(SP3)	AArch64
	V11Swan25	AArch64
openEuler	20.03 LTS	AArch64/x86_64	-
	22.03LTS	AArch64/x86_64	-
	22.03 LTS SP1 (6.6.0-72.0.0.76 kernel)	AArch64	-
	22.03 LTS SP4	AArch64	-
	24.03 LTS SP1	AArch64	-
	24.03 LTS SP4	AArch64	The Atlas 800I A2 inference server and Atlas 900 A3 SuperPoD support the deployment of NPU firmware and driver of 25.2.0 and later versions.
UOS	20-1020e	AArch64	On UOS 20-1020e and UOS 20-1050u2e, MindCluster Ascend Deployer does not support the installation of system dependencies required by the Ascend software. You need to manually install the dependencies by referring to Installing System Dependencies on UOS and then use this tool to deploy the dependencies.
UOS	20-1050u2e	AArch64
Ubuntu	18.04	AArch64/x86_64	-
	20.04	AArch64/x86_64	-
	22.04	AArch64/x86_64	-
	22.04.4	AArch64	Currently, kernel 6.5.0-18-generic and kernel 5.15.0-94-generic are supported.
	24.04	AArch64	Currently, the NPU firmware and driver of 25.2.0 and later versions are supported.
MTOS	22.03	AArch64	The MTOS needs to use the kernel-devel package that cannot be obtained from the public network. The download function is reserved. If you need to use the package, replace it with the kernel-devel-5.10.0-218.0.0.mt20240808.560.mt2203sp4.aarch64.rpm package generated after image decompression.
veLinux	1.3	AArch64	Python 3 must be available in the environment.
VesselOS	1.0	AArch64	Currently, the dependency source of VesselOS is not publicly accessible. When VesselOS_1.0_AArch64 is selected as the OS, only the common software package is downloaded. You need to prepare the system dependency packages based on the ascend_deployer/downloader/config/VesselOS_1.0_aarch64/external_pkg.json file in the common software package. NOTE: You can mount the VesselOS image file, obtain all system dependency packages from the mounted Packages folder, and save them to the ascend_deployer/resources/VesselOS_1.0_aarch64 folder. The docker-engine and libtool-ltdl dependency packages need to be stored in the ascend_deployer/resources/VesselOS_1.0_aarch64/docker folder.
VesselOS	2.0	AArch64	Currently, the dependency source of VesselOS is not publicly accessible. When VesselOS_2.0_AArch64 is selected as the OS, only the common software package is downloaded. You need to prepare the system dependency packages based on the ascend_deployer/downloader/config/VesselOS_2.0_aarch64/external_pkg.json file in the common software package. NOTE: You can mount the VesselOS image file, obtain all system dependency packages from the mounted Packages folder, and save them to the ascend_deployer/resources/VesselOS_2.0_aarch64 folder. The docker-engine, libtool-ltdl, libnetwork, and TAR dependency packages need to be stored in the ascend_deployer/resources/VesselOS_2.0_aarch64/docker folder.
Notes MindCluster Ascend Deployer supports only the default environment after the OS image is successfully installed. Do not manually install or uninstall software after the OS is installed. Otherwise, an error may occur during tool execution. The automatic kernel upgrade function must be disabled for the OSs. Take Ubuntu as an example. After it is installed and restarted, you can run the sudo apt-mark hold linux-image-generic linux-headers-generic linux-image-extra command to disable automatic kernel upgrade (delete sudo when running this command as the root user). For users of other OSs or systems with custom upgrades, disable the function according to the corresponding OS guides.

Software Packages That Can Be Downloaded

**Table 2** Software packages supported by MindCluster Ascend Deployer
Software	Involved Package	Restrictions
CANN	NNAE, NNRT, Toolkit, kernels, ToolBox	When using MindCluster Ascend Deployer to download the CANN package, you can use pip3 to execute pip3 install esdk-obs-python==3.23.12 to install esdk-obs-python of the specified version and enable the resumable download function of CANN. This prevents download failures or long-time re-download due to large software packages or network problems. Toolbox is a MindCluster performance test component. Its version matches the CANN package version and has an independent version number.
NPU	npu-driver, npu-firmware, mcu	-
MindCluster cluster scheduling	Ascend Device Plugin, Ascend Docker Runtime, NodeD, NPU Exporter, Volcano, Ascend Operator, Resilience Controller, ClusterD, MindIO	When deploying MindIO, download the OS dependency by referring to Downloading Ascend Software, and then manually install the dos2unix dependency, for example, apt-get install dos2unix. MindIO is not supported by MindCluster 7.1.RC1 and later versions.
MindCluster fault diagnosis	fault-diag	-
MindSpore	MindSpore	When MindCluster Ascend Deployer is used for deployment, MindSpore, TensorFlow, and Torch-npu cannot be downloaded and installed at the same time. Only one AI framework can be downloaded and installed. Select an AI framework based on the actual service scenario. TensorFlow 2.6.5 has vulnerabilities. For details about how to handle the vulnerabilities, see the Vulnerabilities and Fixing Solutions.
TensorFlow	TensorFlow
Torch-npu	torch-npu, torch
MindIE-image	mindie-image, mindie	The image file is large. Select a proper version and type.

Parent topic: Download Reference