代码拉取完成，页面将自动刷新
同步操作将从 Ascend/ascend-deployer 强制同步，此操作会覆盖自 Fork 仓库以来所做的任何修改，且无法恢复！！！
确定后同步将在后台操作，完成时将刷新页面，请耐心等待。
The offline installation tool provides automatic download and one-click installation of the OS components and Python third-party dependencies. It also supports the installation of the driver, firmware, and CANN software packages. The tools directory additionally places the Device IP configuration script, the use method can refer to Device IP configuration.
| OS | Version | CPU Architecture | Installation Type |
|---|---|---|---|
| BCLinux | 7.6 | aarch64 | A minimal image is installed by default. |
| BCLinux | 7.6 | x86_64 | A minimal image is installed by default. |
| BCLinux | 7.7 | aarch64 | A minimal image is installed by default. |
| CentOS | 7.6 | aarch64 | A minimal image is installed by default. |
| CentOS | 7.6 | x86_64 | A minimal image is installed by default. |
| CentOS | 8.2 | aarch64 | A minimal image is installed by default. |
| CentOS | 8.2 | x86_64 | A minimal image is installed by default. |
| Debian | 9.9 | aarch64 | A minimal image is installed by default. |
| Debian | 9.9 | x86_64 | A minimal image is installed by default. |
| Debian | 10.0 | x86_64 | A minimal image is installed by default. |
| EulerOS | 2.8 | aarch64 | A minimal image is installed by default. |
| EulerOS | 2.9 | aarch64 | A minimal image is installed by default. |
| EulerOS | 2.9 | x86_64 | A minimal image is installed by default. |
| Kylin | V10Tercel | aarch64 | A minimal image is installed by default. |
| Kylin | V10Tercel | x86_64 | A minimal image is installed by default. |
| Kylin | V10GFB | aarch64 | A minimal image is installed by default. |
| Kylin | v10juniper | aarch64 | A minimal image is installed by default. |
| Linx | 6 | aarch64 | A minimal image is installed by default. |
| OpenEuler | 20.03LTS | aarch64 | A minimal image is installed by default. |
| OpenEuler | 20.03LTS | x86_64 | A minimal image is installed by default. |
| Tlinux | 2.4 | aarch64 | A server image is installed by default. |
| Tlinux | 2.4 | x86_64 | A server image is installed by default. |
| UOS | 20SP1 | aarch64 | A minimal image is installed by default. |
| UOS | 20SP1 | x86_64 | A minimal image is installed by default. |
| UOS | 20-1020e | aarch64 | A minimal image is installed by default. |
| UOS | 20-1021e | aarch64 | A minimal image is installed by default. |
| UOS | 20 | aarch64 | A minimal image is installed by default. |
| UOS | 20 | x86_64 | A minimal image is installed by default. |
| Ubuntu | 18.04.1/5 | aarch64 | A minimal image is installed by default. |
| Ubuntu | 18.04.1/5 | x86_64 | A minimal image is installed by default. |
| Ubuntu | 20.04.1 | aarch64 | A minimal image is installed by default. |
| Ubuntu | 20.04.1 | x86_64 | A minimal image is installed by default. |
| Central Inference Hardware | Central Training Hardware | Intelligent Edge Hardware |
|---|---|---|
| A300-3000 | A300T-9000 | A500 Pro-3000 |
| A300-3010 | A800-9000 | Atlas200(EP) |
| Atlas 300I Pro | A800-9010 | |
| Atlas 300V Pro | Atlas 300T Pro | |
| Atlas 300I Duo | ||
| A800-3000 | ||
| A800-3010 | ||
| Atlas 300V | ||
| Atlas 200I SoC A1 |
If the offline installation tool cannot run due to environment variables, users need to configure ASCENDPATH according to the actual situation of the environment.
If the gcc version of the system is lower than 7.3.0, the offline installation tool will be automatically installed, which takes a long time; Users can also avoid the automatic upgrade by manually upgrading in advance and configuring environment variables.
By default, the offline installation tool downloads and installs Python-3.7.5 as a Python version of the Cann package. This is explained in Python-3.7.5.Users can select the Python version by setting the ASCEND_PYTHON_VERSION environment variable, or the ASCEND_PYTHON_VERSION configuration item in the downloader/config.ini file (environment variable is preferred when setting at the same time).The optional Python versions are 3.7.0 to 3.7.11 and 3.8.0 to 3.8.11. This tool has only been fully adapted and tested on Python-3.7.5, and it is strongly recommended not to change the default configuration.
Basic commands such as tar, cd, ls, find, grep, chown, chmod, unzip, bzip2,ssh must be installed in the OS. It is recommended that during the installation process of Ubuntu/Debian system, select the option of [OpenSSH Server]/[SSH Server] when going to [Software Selection] to avoid missing SSH command.
Kylin V10 GFB system downloads kylin when the system is dependent_ V10Tercel_ Aarch64 is enough.
The offline installation tool supports only the default environment after the OS image is successfully installed. Do not install or uninstall software after the OS is installed. If some system software has been uninstalled, causing inconsistency with the default system package, you need to manually configure the network and use tools such as apt, yum, and dnf to install and configure the missing software.
The offline installation tool can install only basic libraries to ensure that TensorFlow and PyTorch can run properly. If you need to run complex inference services or model training, the model code may contain libraries related to specific services. You need to install the libraries by yourself.
Offline installation tools except install.sh、start_download.sh、start_download_ui.bat and start_download.bat, the rest of the files are not designed for the user to use the interface or command. Do not use them directly.
It is forbidden to put the password in the inventory_file file.
CentOS 7.6 x86_64 with lower version kernel (below 4.5) of ATLAS 300T training card requires CentOS to be upgraded to 8.0 or above or a kernel patch is added. Failure to do so may result in firmware installation failure.Add a kernel patch method please refer to the reference [link] (https://support.huawei.com/enterprise/zh/doc/EDOC1100162133/b56ad5be).
Atlas 300I Pro,Atlas 300V Pro,Atlas 300V,A300T-9000 and Atlas 300T Pro must be set variable cus_npu_info in inventory_file, Atlas 300I pro should be 300i-pro, Atlas 300V Pro should be 300v-pro,Atlas 300V shoud be 300V.Edit inventory_file in the following format:
[ascend] localhost ansible_connection='local' cus_npu_info='300i-pro' # Atlas 300I Pro ip_address_1 ansible_ssh_user='root' cus_npu_info='300v-pro' # Atlas 300V Pro ip_address_2 ansible_ssh_user='root' cus_npu_info='300v' # Atlas 300V ip_address_3 ansible_ssh_user='root' cus_npu_info='300t' # A300T-9000 ip_address_4 ansible_ssh_user='root' cus_npu_info='300t-pro' # Atlas 300T Pro
The hardware configurations of the Atlas200 EP and A300 card
(A300-3000, A300-3010, A800-3000, and A800-3010) cannot be
distinguished. The following conditions must be met when using the
Atlas200 EP. The Atlas200 EP and A300 inference card environments
cannot be deployed in batches. If the deployed machine contains the
Atlas200 EP, do not store the NPU package of the A300 EP in the
Resources directory. If the deployed machine contains the A300
inference card, do not store the NPU package of the Atlas200 EP in
the Resources directory. Because of the above two restrictions,
--download=CANN does not include the NPU package of Atlas200 EP.
Please prepare it yourself.
By default, the root user is not allowed to remotely log in to OSs such as EulerOS. Therefore, you need to set PermitRootLogin to yes in the sshd_config file before using this tool(Individual OS configuration methods may be different, please refer to the OS official description), and close the remote connection of root user after using this tool.
Support for Ubuntu 18.04.1/5 installation of cross-compiled related components and the Aarch64 architecture toolkit package.
After the kylin V10 system’s dependencies are installed, you need to wait for the system configuration to complete before you can use docker and other commands.
Before installing system dependencies, please confirm whether Docker is installed on the system. If it is installed, please uninstall it before installing system dependencies.
Users are advised to modify downloader/config and downloader/requirements.txt to ensure compliance with the security requirements of your organization.
After the default installation of tlinux system, the total space of the root directory is about 20G, and the packages that exceed the available disk space can not be placed in the resources directory to avoid decompression or installation failure.
BCLinux 7.6 does not have python3 by default. The
yum install python3 command is run before the download operation.
Because the BCLinux 7.6 system source does not contain python3,
modify the source configuration file by referring to the BCLinux
official configuration file, or change “el7.6” to “el7.7” in
“/etc/yum.repos.d/BCLinux-Base.repo”(Run the
sed -i 's/el7.6/el7.7/g' /etc/yum.repos.d/BCLinux-Base.repo
command). After the installation, restore the original configuration.
tensorflow-1.15.0 aarch64,tensorflow-2.6.5 aarch64,torch-1.5.0/apex-0.1 aarch64/x86_64 and torch-1.8.1/apex-0.1/torch_npu-1.8.1 aarch64/x86_64 are not available for download. You need to place them in your resources/pylibs directory, otherwise the installation will be skipped.
Please strictly follow the official compilation specification when compiling tensorflow aarch64.
Tensorflow 1.15.0 is only applicable to python3.7, and tensorflow 2.6.5 is applicable to python3.7, python3.8, and python3.9.Due to dependency conflict, after installing one version, you need to uninstall the installed version before installing another version.
If you plan to use the automatic download function under Linux, please configure the GUI interface in advance and directly run the download instruction.
Euleros, Debian and other systems may trigger driver source compilation when installing the driver. Users are required to install the kernel header package consistent with the kernel version of the system (which can be viewed through ‘uname -r’ command). The details are as follows.
Based on security considerations, it is recommended to reinforce the unzipped installation directory(ascend-deployer) and set its permission to only allow owner to use.
Description of the kernel header package
| OS | kernel header package that matches the kernel version of the system | How to get |
|---|---|---|
| EulerOS | kernel-headers-<version>、
kernel-devel-<version>
|
Contact the OS vendor, or find it in the “devel_t ools.tar.gz” tool component that comes with the corresponding OS |
| Debian | linux-headers-<version>、
linux-headers-<version>-common、
linux-kbuild-<version>
|
Contact the OS vendor, or look it up in the image of the corresponding OS |
pip3 install ascend-deployer==<Version>
git clone https://gitee.com/ascend/ascend-deployer.git
For security reasons, the user should set the environment umask to 077 before git clone, and only clone and use tools in the user’s home directory, which is only for the user’s own use.
Click the “clone / download” button in the upper right corner, and then click the “download zip” below to download and unzip to use.In order to prevent the software package from being maliciously tampered with during delivery or storage, it is recommended that users download the software package and use sha256sum to verify the integrity of the software. For the latest official version of sha256sum, please refer to readme of the master branch. This tool can be used by root and non-root users. To avoid the risk of excessive permissions after unzipping, it is recommended to set the environment umask to 077 before unzipping the zip package, and only unzip and use tools in the user’s HOME directory, and only for the user’s own use. The above two installation methods please pay attention to the tool directory permissions risk.
Confirm whether the owner and authority of the directory and file meet the security requirements of the user’s organization, etc. In addition, please note that except for the user himself and other users outside the management room, they should not have the write permission of the parent directory of the installation directory.find {Installation directory} -ls
The download function can be used in the Windows or Linux OSs.Before running, please confirm that the offline installation directory used belongs to the user, and the permissions and groups of the directory need to meet the security requirements of the organization.
./start_download.sh --os-list=<OS1>,<OS2> --download=<PK1>,<PK2>==<Version>
command to start download, refer to Linux Download Parameter Description.
The following call ` * * sh script using. / *
* sh way, also can use bash * * sh ` calls, please
according to actual use.It is recommended to set the environment
umask to 077 before downloading.
[ascend]
localhost ansible_connection='local'
[ascend:vars]
user=HwHiAiUser
group=HwHiAiUser
install_path=/usr/local/Ascend
| parameter | remark |
|---|---|
| user | user，will be pass to –install-username options |
| group | usergroup，will be pass to –install-usergroup options |
| in stall_path | The installation path of the CANN package，will be pass to –install-path options |
/etc/scend/cann_install.info file, otherwise it will be installed
to the path specified by the contents of the file) and not for
non-root (only to the default ~/Ascend path).The install_path
parameter does not specify the installation path for the driver
package and edge components (AtlasEdge and HA). The driver package
can only be installed to the default path /usr/local/Ascend and edge
components (AtlasEdge and HA) can only be installed to the default
path /usr/local./etc/scend/cann_install.info and
/etc/Ascend/ascend_toolbox_install.info file, otherwise it will
be installed to the path specified by the contents of the file) and
not for non-root (only to the default ~/Ascend path).
#add HwHiAiUser group
groupadd HwHiAiUser
#add HwHiAiUser user add it to HwHiAiUser group
#set /home/HwHiAiUser as HwHiAiUser's HOME directory and create
#set /bin/bash HwHiAiUser's default shell
useradd -g HwHiAiUser -d /home/HwHiAiUser -m HwHiAiUser -s /bin/bash
usermod -s /sbin/nologin HwHiAiUser # When installing edge components (AtlasEdge and HA) in versions 2.0.2
usermod -s /bin/bash HwHiAiUser # When installing the driver package
apt install haveged. After installation, you need to
execute systemctl enable haveged and systemctl start haveged
to start the haveged service.[ascend:vars] user=HwHiAiUser group=HwHiAiUser
| Software name | description |
|---|---|
| Python, gcc | python3.7.5 and gcc7.3.0 is installed in the $HOME/.local/ directory |
| Python framework | tensorflow, pytorch, mindpore |
| CANN | toolbox, nnae, nnrt, tfplugin, toolkit and kernels are installed in the $HOME directory by default, and the specified path is not supported |
| MindStudio | installed in the $HOME/ directory |
Note: 1. Non-root users need root users to install system components and
driver before they can install the above components. 2. After installing
gcc7.3.0, you need to establish a symbolic link to use it. For example,
gcc7.3.0 installed by root executes the command
ln -sf /usr/local/gcc7.3.0/bin/gcc /usr/bin/gcc. 3. To install
kernels, you need to install nnae or toolkit first. When installing
kernels, you need to specify –kernels_type parameter. 4. Non-root users
need to join the driver installation group to install and use nnrt and
toolkit normally. The default driver installation group is HwHiAiUser,
Modify the user group command as follows:
usermod -a -G HwHiAiUser non-root-user
ascend-deployer |- ... |- install.sh |- inventory_file |- ... |- playbooks |- README.md |- resources |- A300-3010-npu_xxx.zip |- A300-3010-npu-driver_xxx.run |- A300-3010-npu-firmware_xxx.run |- Ascend-cann-nnrt-xxx.zip |- Ascend-cann-nnrt-xxx.run |- ... |- Ascend-cann-toolkit-xxx.run |- ... |- BCLinux_7.6_aarch64 |- BCLinux_7.6_x86_64 |- cert_ief_xxx.tar.gz |- edge-installer_xxx_arm64.tar.gz |- edge-register_xxx_arm64.tar.gz |- docker_images |- ...
Configure a stand-alone inventory_file file.
Edit the inventory_file file. The default is as follows:
[ascend] localhost ansible_connection='local'
Run the installation script and select an installation mode (software-specific installation or scenario-specific installation) as required.Note: if other users need to be able to use Python installed by root user, please set umask to 022 in advance. Before setting, confirm that the umask permission meets the security requirements of your organization.
run the ./install.sh --install=<package_name_1>,<package_name_2>.
The following is an example.
./install.sh --help # Viewing Help Information. ./install.sh --install=sys_pkg,python,npu # Installing system dependencies and python3.7.5 and driver and firmware.
Notes:
- Installation sequence: sys_pkg > python > npu(driver and firmware) > CANN software package(such as the Toolkit and nnrt) > AI framework(pytorch、tensorflow、mindspore).During installation, the cann package version under the resources directory needs to be matched with NPU. - After the driver or firmware is installed, maybe you need run the `reboot` command to restart the device for the driver and firmware to take effect. - Some components require runtime dependencies. For example, PyTorch requires the Toolkit or nnae to provide runtime dependencies, TensorFlow and npubridge and npudevice require TFPlugin and toolkit or TFPlugin and nnae to provide runtime dependencies, and mindspore require driver and toolkit to provide runtime dependencies. - All the installation of Python libraries must first install Python 3.7.5, such as python, tensorflow, Mindstore, etc. - During installation, the running environment time needs to be calibrated to the correct UTC time through the date - s command.
run the ./install.sh --install-scene=<scene_name>. The following
is an example.
./install.sh --install-scene=auto # Automatic installation of all software packages that can be found
The offline installation tool provides several basic installation
scenarios. For details, see Installation Scenarios.
After the installation.
run the ./install.sh --test=<target>. The following is an example:
./install.sh --test=driver # Test whether the driver is normal.
SSH connection based on key authentication,Please confirm that paramiko is not installed in the system before installation (ansible will use paramiko in some cases, and its improper configuration may cause security problems).
Configure the IP addresses of other devices where the packages to be installed. Edit the inventory_file file. The format is shown as follows:
[ascend] ip_address_1 ansible_ssh_user='root' # root user ip_address_2 ansible_ssh_user='root' ip_address_3 ansible_ssh_user='username' # non-root user
Configure the reference operation for key authentication.Please pay attention to the risks during the use and storage of SSH keys and key passwords, especially when the keys are not encrypted. Users should configure them according to the security policies of their organization, including but not limited to software version, password complexity requirements, security configuration (protocol, encryption suite, key length, etc,especially the configuration under /etc/ssh and ~/.ssh)
ssh-keygen -t rsa -b 3072 # Log in to the management node and generate the SSH Key. For security reasons, it is recommended that the user Enter the key password at the "Enter passphrase" step, and ensure that the password complexity is reasonable. It is recommended to set the umask to 0077 before executing this command and to restore the original umask after executing it.
ssh-copy-id -i ~/.ssh/id_rsa.pub <user>@<ip> # Copy the public key of the management node to the machines of all nodes, and replace <user>@<ip> with the account and ip of the corresponding node to be copied to.
ssh <user>@<ip> # Verify that it is possible to log on to the remote node, and replace <user>@<ip> with the account and IP of the corresponding node to be logged in. After verifying that the login is OK, run the 'exit' command to exit the SSH connection.
Note: Please be aware of the risks involved in the use and storage of SSH keys.
Set up the SSH agent to manage the SSH key to avoid entering the key password during the bulk installation of the tool. The following are the guidelines for setting up an SSH agent:
ssh-agent bash # Start the ssh-agent bash process
ssh-add ~/.ssh/id_rsa # Add a private key to the ssh-agent
Run the ./install.sh --check command to test the connectivity of
the devices where the packages to be installed. Ensure that all
devices can be properly connected. If a device fails to be connected,
check whether the network connection of the device is normal and
whether sshd is enabled.
The following operation is the same as the above Single-Device Installation steps 2 and 3.
When the bulk installation of the tool is completed, exit the SSH agent process in time to avoid security risks.
exit # Exit the ssh-agent bash process
The default concurrency number is 5, and the maximum concurrency number is 255. If the number of environments to be deployed is greater than 5, you can modify the forks value in the ansible.cfg file to the total number of nodes to be deployed.
When the tool is installed with pip, two entrances will be provided for easy operation.
Both entrances are available to both root and non-root users
ascend-download --os-list=<OS1>,<OS2> --download=<PK1>,<PK2>==<Version>
Both win10 and Linux can execute
ascend-deployer --install=<pkg1,pkg2>
The ascend-deployer command is essentially a wrapper of install.sh.The use method is exactly the same as directly executing install.sh in the ascend deployer directory. The ASCEND_Deployer command automatically looks for the file ASCEND_Deployer /install.sh in the user’s HOME directory and replaces the user’s HOME directory by setting the environment variable ASCEND_Deployer_HOME. Non-root users must ensure that the directory exists and can read and write properly.
The offline deployment tool can install Python 3.7.5, To ensure that the built-in Python (Python 2.x or Python 3.x) is not affected, you need to configure the following environment variables before using Python 3.7.5:
export PATH=/usr/local/python3.7.5/bin:$PATH # root export LD_LIBRARY_PATH=/usr/local/python3.7.5/lib:$LD_LIBRARY_PATH # root export PATH=~/.local/python3.7.5/bin:$PATH # non-root export LD_LIBRARY_PATH=~/.local/python3.7.5/lib:$LD_LIBRARY_PATH # non-root
This tool will automatically install the Python 3.7.5 environment variable in /usr/local/ascendrc file. You can easily set the Python 3.7.5 environment variable by following the following command
source /usr/local/ascendrc # root source ~/.local/ascendrc # non-root
Similarly, other software packages or tools installed by offline deployment tools can be used normally only after users refer to the corresponding official information and configure environment variables or make other Settings.
Inference scenario
If you need to develop applications, please refer to the relevant official materials, such as CANN Application Software Development Guide (C and C++) or CANN Application Software Development Guide (Python).
Training scenario
For network model migration and training, please refer to the relevant official materials, such as TensorFlow Network Model Porting and Training Guide or PyTorch Network Model Porting and Training Guide.
Delete this tool
This tool is only used for deployment. When installation completed, it should be deleted for free the disk space.
| Something that should be deleted | instructions |
|---|---|
| ascend-deployer | Directory of tool on the controller |
pip3
uninstall ascend-deployer |
Tool pip-installed on the controller, uninstall using commands |
| ~/ansible | Customize information collection configuration files on the controller and remote machines |
~
/resources和~/resources.tar |
Resource directory on the controller and remote machines |
| ~/build | Source package decompression directory on the controller and remote machines |
Select corresponding parameters to install the software. The command
likes ./install.sh [options]. The following table describes the
parameters. You can run the ./install.sh --help command to view the
options of the following parameters.
| P arameter | Description |
|---|---|
| –help -h | Queries help information. |
| –check | Check the environment to ensure that the control machine has installed Python 3.7.5, Ansible and other components, and check the connectivity with the device to be installed. |
| –clean | Clean the Resources directory under the user’s home directory for the device to be installed. |
| –nocopy | Forbids resources copying during batch installation. |
| –f orce_upg rade_npu | Can force upgrade NPU when not all devices have exception |
| –te nsorflow _version | Appoint tensorflow version,must be 1.15.0 or 2.6.5,default is 1.15.0 |
| –kern els_type | Appoint kernels package type,must be nnae or toolkit,default is nnae |
| –verbose | Print verbose. |
| –outp ut-file= | Set the output format of the command execution. The available parameters can be viewed with the command “ansible -doc-t callback-l”. |
| – stdout_c allback= | Performs debugging. |
| – install= | Specifies the software to be installed. If –install=npu is specified, the driver and firmware are installed. |
| –instal l-scene= | Specifies the scenario for installation. For details about the installation scenarios, see Installation Scenarios. |
| –patch= | Patching specific package |
| –patch-r ollback= | Rollback specific package |
| –test= | Checks whether the specified component works properly. |
| Parameter | Description |
|---|---|
--os-
list=<OS1>,<OS2> |
set specific os softwares to download |
| ` --download=<PK1>, <PK2>==<Version>` | download specific components. such as MindSpore、MindStudio、CANN |
This tool downloads python component packages by default. If the system specified by –os-list has only aarch64 architecture, only python component packages required by aarch64 architecture system will be downloaded. If the system specified by –os-list has only x86_64 architecture, only python component packages required by x86_64 architecture are downloaded. When –os-list is empty or the specified system has both aarch64 and x86_64 architectures, the Python component packages required for both architectures are downloaded. Same logic as above to download CANN package for aarch64 or x86_64 architectures.
| optional components | ve rsion 1 | ve rsion 2 | ve rsion 3 | ve rsion 4 | version 5 | v ersion 6 |
|---|---|---|---|---|---|---|
| MindStudio | 2.0.0 | 3.0.2 | 3.0.3 | 3.0.4 | 5.0.RC1 | 5 .0.RC2 |
| MindSpore | 1.1.1 | 1.3.0 | 1.5.0 | 1.6.2 | 1.7.0 | 1.8.0 |
| CANN | 2 0.3.0 | 5. 0.2.1 | 5. 0.3.1 | 5.0.4 | 5. 1.RC1.1 | 5 .1.RC2 |
Only one version of MindSpore or MindStudio that matches CANN package
version should exist in the Resources directory during installation, as
shown above. ./start_download.sh --download=<PK1>,<PK2>==<Version>,
when <Version> is missing, <PK> is the latest.
--download=MindSpore, –os-list specifies the corresponding OS,
please refer to the official website of
mindspore for some instructions.
MindStudio installation please refer to the install
MindStudio.
CANN installation please refer to the install CANN.
The offline installation tool provides several basic installation
scenarios.If the GCC version of the system is lower than 7.3.0, GCC
needs to be installed before installing the framework to ensure that all
scenarios can be used normally after installation. After installing
gcc7.3.0, you need to establish a soft link to use it (/usr/bin/gcc
points to the executable file of the installed gcc7.3.0). For example,
gcc7.3.0 installed by root executes the command
ln -sf /usr/local/gcc7.3.0/bin/gcc /usr/bin/gcc.
| Installation scenario | Installed Components | Description |
|---|---|---|
| auto | all | All software packages that can be found are installed |
| vmhost | sys_pkg、npu、toolbox | VM host scene |
| edge | sys_pkg、atlasedge、ha | Install MindX middleware, HA |
| offline_dev | sys_pkg、python、npu、toolkit | Offline development scene |
| offline_run | sys_pkg、python、npu、nnrt | Offline run scene |
| mindspore | sys_pkg、python、npu、toolkit、mindspore | mindspore scene |
| tensorflow_dev | sys_pkg、python、npu、toolkit、tfplugin、tensorflow | tensorflow development scene |
| tensorflow_run | sys_pkg、python、npu、nnae、tfplugin、tensorflow | tensorflow run scene |
| pytorch_dev | sys_pkg、python、npu、toolkit、pytorch | pytorch development scene |
| pytorch_run | sys_pkg、python、npu、nnae、pytorch | pytorch run scene |
The configuration files for the preceding installation scenarios are stored in the scene directory. For example, the following shows the configuration file scene/scene_auto.yml of the auto scene:
- hosts: '{{ hosts_name }}'
- name: install system dependencies
import_playbook: ../install/install_sys_pkg.yml
- name: install python3.7.5
import_playbook: ../install/install_python.yml
- name: install driver and firmware
import_playbook: ../install/install_npu.yml
- name: install toolkit
import_playbook: ../install/install_toolkit.yml
- name: install nnrt
import_playbook: ../install/install_nnrt.yml
- name: install nnae
import_playbook: ../install/install_nnae.yml
- name: install tfplugin
import_playbook: ../install/install_tfplugin.yml
- name: install toolbox
import_playbook: ../install/install_toolbox.yml
- name: install pytorch
import_playbook: ../install/install_pytorch.yml
- name: install tensorflow
import_playbook: ../install/install_tensorflow.yml
- name: install mindspore
import_playbook: ../install/install_mindspore.yml
To customize an installation scenario, refer to the preceding configuration file.
The ascend deployer tool supports cann cold patch installation and
fallback. 1. Cann patch packages do not support online downloading using
the ascend deployer tool. Users need to obtain the required cann patch
packages by themselves and place them in the ascend deployer / resources
/ patch (if there is no patch directory, users should create it by
themselves). Note that the cann package corresponding to the patch
package in the ascend deployer / resources directory should be deleted
before installation. 2. The execution commands for installing and
fallback cann cold patch are as follows: - Install cann cold patch (take
nnae and tfplugin packages as examples):
./install.sh --patch=nnae,tfplugin - Fallback cann cold patch (take
nnae and tfplugin packages as examples):
./install.sh --patch-rollback=nnae,tfplugin 3. The relevant
constraints on cann cold patch are as follows: - The patch can only
support the upgrade of the corresponding baseline version or related
patch version. - For patches based on the same baseline version, ensure
that the patch version installed later is greater than the patch version
installed earlier. - The patch only supports fallback once.During
fallback, you need to place the patch package when installing the patch
in the ascend deployer/resources/patch directory (if there is no patch
directory, please create it yourself). Note that the cann package
corresponding to the patch package in the ascend deployer/resources
directory should be deleted before fallback.
If you want to use an proxy, configure the proxy in an environment variable. Users need to pay attention to the security of the proxy.This tool validates HTTPS certificates by default, if a certificate error occurs during the download process, it may be that the proxy server has a security mechanism for certificate replacement, so you need to install the proxy server certificate first.
Configure the agent in the environment variable as follows
# Configure environment variables. export http_proxy="http://user:password@proxyserverip:port" export https_proxy="http://user:password@proxyserverip:port"
Where “user” is the user’s internal network name, “password” is the user’s password (special characters need to be escaped), “proxyserverip” is the IP address of the proxyserver, and “port” is the port. The principle of configuring proxies in Windows environment variables is the same as that in Linux. For details, see official instructions.
Configure the agent in the downloader/config.ini file as follows:
[proxy] verify=true # Whether to verify the HTTPS certificate. If it is closed,Please be aware of the security risks
You can configure and modify the download parameters in the downloader/config.ini file to download the required OS components on windows. It is not recommended to modify the configuration file directly. It is recommended to run start_download_ui.bat and use the UI interface to check the required components
[download] os_list=CentOS_7.6_aarch64, CentOS_7.6_x86_64, CentOS_8.2_aarch64, CentOS_8.2_x86_64, Ubuntu_18.04_aarch64, Ubuntu_18.04_x86_64 ... # OS information of the environment to be deployed. [software] pkg_list=CANN_5.0.3.1,MindStudio_3.0.3 # CANN or MindStudio to be deployed.
The offline installation tool provides the source configuration file. Replace it as required.
[pypi] index_url=https://repo.huaweicloud.com/repository/pypi/simple
[base] baseurl=https://mirrors.huaweicloud.com/centos-altarch/7/os/aarch64 [epel] baseurl=https://mirrors.huaweicloud.com/epel/7/aarch64
https://cmake.org https://github.com https://gcc.gnu.org http://mirrors.bclinux.org https://archive.kylinos.cn https://support.huawei.com https://mirrors.tencent.com https://mirrors.bfsu.edu.cn https://repo.huaweicloud.com https://uniportal.huawei.com https://mirrors.huaweicloud.com https://cache-redirector.jetbrains.com https://obs-9be7.obs.myhuaweicloud.com https://obs-9be7.obs.cn-east-2.myhuaweicloud.com https://ms-release.obs.cn-north-4.myhuaweicloud.com
| sha256sum | Version of the ascend-deployer |
|---|---|
| 22f7e10677658e7c3d 223b32f73786c765e85cf6f66440bf620c3e4275f11e7f | ascend-depl oyer-2.0.4.B093.zip |
/etc/hwsipcrl/ascendsip.crl(or ascendsip_g2.crl), for non-root
users, This file is
~/.local/hwsipcrl/ascendsip.crl(or ascendsip_g2.crl). If the
local crl file does not exist or takes effect earlier than the crl
file in the installation package, the local crl file is replaced by
the crl file in the installation package. The tools/update_crl.sh
supports independent crl file update, Run
bash update_crl.sh <crl_file> command to update an independent
crl file, <crl_file> is the path of the crl file uploaded by the
user.systemctl restart sshd to restart the sshd service.yum install openblas to
install the system dependency, and then create a soft link. The
creation method is as follows (please refer to the specific library
version):find / -name libopenblas*so to find the
libopenblas-r0.3.9.so file (the specific version displayed is
subject to the actual version).ln -s /usr/lib64/libopenblas-r0.3.9.so /usr/lib64/libblas.so.3
and
ln -s /usr/lib64/libopenblas-r0.3.9.so /usr/lib64/liblapack.so.3
Create soft links.此处可能存在不合适展示的内容，页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容，可点击提交进行申诉，我们将尽快为您处理。
马建仓 AI 助手