Installation

View Source On Gitee

Installing by Using Pip

Q: When installing GPU, CUDA 10.1, 0.5.0-beta version of MindSpore, it prompts cannot open shared object file:No such file or directory, what should I do?

A: The error message indicates that the cuBLAS library is not found. Generally, the cause is that the cuBLAS library is not installed or is not added to the environment variable. Generally, cuBLAS is installed together with CUDA and the driver. After the installation, add the directory where cuBLAS is located to the LD_LIBRARY_PATH environment variable.


Q: What should I do if an error message ERROR: mindspore_{VERSION}.whl is not a supported wheel on this platform is displayed when I install MindSpore using pip?

A: pip checks compatibility of wheel package and current Python environment by verifying the file name. For example, when installing mindspore_ascend-1.2.0-cp37-cp37m-linux_aarch64.whl, pip checks whether:

  1. Python version falls in 3.7.x

  2. Current operating system is Linux

  3. System architecture is arm64

Hence, if you run into the problem is not a supported wheel on this platform, please make sure the current environment fulfills the requirements of the MindSpore package you are installing, or a correct version of MindSpore package which matches your environment is being installed.


Q: What should I do if an error message ERROR: mindspore-{VERSION}.whl is not a supported wheel on this platform is displayed when I compile MindSpore form source and install it by using pip on the macOS system?

A: First, check the name of the installation package under the output directory, which is similar to mindspore-1.6.0-cp37-cp37m-macosx_11_1_x84_64.whl. The “11_1” in the package name means that the SDK version used when compiling is 11.1. If the SDK version used is 11.x, it may be that the SDK version used during compilation is too high and cannot be installed.

Solution 1: You can rename the installation package and then try to install it. For example, rename the above installation package to mindspore-1.6.0-cp37-cp37m-macosx_10_15_x84_64.whl.

Solution 2: Before compiling the source code, set the environment variable MACOSX_DEPOLYMENT_TARGET to 10.15 and recompile.


Q: What should I do if the macOS system with arm64 architecture reports an error when using pip to install SciPy?

A: SciPy currently does not have a whl package for the arm64 architecture of the macOS system. You can use a third-party compiled whl package for installation. Run the following commands to install SciPy first, and then reinstall MindSpore.

pip install --pre -i https://pypi.anaconda.org/scipy-wheels-nightly/simple scipy

Q: What should I do if an error message SSL:CERTIFICATE_VERIFY_FATLED is displayed when I use pip to install MindSpore?

A: Add the --trusted-host=ms-release.obs.cn-north-4.myhuaweicloud.com parameter to the pip installation command and try again.


Q: Any specific requirements for Protobuf version when use MindSpore?

A: MindSpore installs version 3.13.0 of Protobuf by default. If it is not the version, there will be many warnings in the log when using pytest to test the code. It is recommended that you use the command ‘pip install protobuf==3.13.0’ to reinstall version 3.13.0.


Q: What should I do when an error ProxyError(Cannot connect to proxy) prompts during installing by using pip?

A: It is generally a proxy configuration problem. You can use export http_proxy={your_proxy} to set proxy on Ubuntu environment, and use set http_proxy={your_proxy} in cmd on Windows environment to configure your proxy.


Q: What should I do when an error prompts during installing by using pip?

A: Please execute pip -V to check if pip is linked to Python3.7+. If not, we recommend you use python3.7 -m pip install instead of pip install command.


Q: What should I do when an error prompts No matching distribution found for XXX during pip installing dependencies?

A: Please execute pip config list to check the software library index index-url. Sometimes software library index updates are out of sync. You can try other software library indexes instead.


Q: What should I do if I cannot find whl package for MindInsight or MindArmour on the installation page of MindSpore website?

A: You can download whl package from the official MindSpore Website download page and manually install it via pip install.


Q: What should I do when I install both CPU and GPU versions of MindSpore, an error cannot import name 'context' from 'mindspore' is displayed while importing?

A: All versions of MindSpore are installed in the directory named mindspore, and installing them in one Python environment may overwrite each other and cause failures. If you wish to use alternate versions of MindSpore for different platforms (e.g. CPU and GPU versions), please uninstall the unused version and install the new version afterwards.


Q: What should I do if error message Could not find a version that satisfies the requirement is displayed when I install MindSpore on an ARM architecture systemby using pip?

A: The version of pip installed on your system is most likely lower than 19.3, which is too low to recognize manylinux2014 label. Wrong versions of python packages such as numpy or scipy are downloaded in the pip install stage, and the issue of not being able to find build dependencies is raised. As such, please upgrade pip in the environment to a later version higher than 19.3 by typing pip install --upgrade pip, and then try installing MindSpore again.


Q: What should I do if error message Running setup.py install for pillow: finished with status 'error' ... The headers or library files could not be found for jpeg, ... is generated when I install MindSpore by using pip?

A: MindSpore relies on the third-party library pillow for some data processing operations, while pillow needs to rely on the libjpeg library already installed in the environment. Taking the Ubuntu environment as an example, you can use sudo apt-get install libjpeg8-dev to install the libjpeg library, and then install MindSpore.


Q: What should I do if error message ImportError: dlopen ... no suitable image found.  Did find: ..._psutil_osx.cpython-38-darwin.so: mach-o, but wrong architecture is generated when I execute MindSpore on macOS for ARM with Python3.8?

A: The default version of psutil in Python3.8 for macOS(ARM architecture) has a bug that compiles its binaries for x86 architecture, causing conflitcts when running MindSpore. To fix this, run pip uninstall psutil; conda install psutil if you are using Conda, otherwise run pip uninstall psutil; pip install --no-binary :all: psutil to install psutil correctly. For detailed reasons, please refer to this post on stackoverflow.


Installing by Using Conda

Q: For Ascend users, what should I do when RuntimeError: json.exception.parse_error.101 parse error at line 1, column 1: syntax error while parsing value - invalid literal; last read: 'T' appears in personal Conda environment?

A: When you encounter the error, the high probability is that there is no update te or topi or hccl package in the personal Conda environment after the run package is updated, and the above several tool packages in the current Conda environment can be uninstalled. You should use command pip install /usr/local/Ascend/ascend-toolkit/latest/lib64/{te/topi/hccl}-{version}-py3-none-any.whl to reinstall.


Installing by Using Source

Q: What is the difference between bash -p and bash -e during compilation?

A: MindSpore Serving build and running depend on MindSpore. Serving provides two compilation modes: 1. Use bash -p {python site-packages}/mindspore/lib to specify an installed MindSpore path to avoid recompiling MindSpore when compiling Serving. 2. When Compiling Serving, compile the corresponding MindSpore. Serving transparents the -e, -V, and -j options to MindSpore. For example, use bash -e ascend -V 910 -j32 in the Serving directory as follows:

  • Compile MindSpore in the third_party/mindspore directory by using bash -e ascend -V 910 -j32.

  • Use the MindSpore compilation result as the Serving compilation dependency.


Q: MindSpore installation: Version 0.6.0-beta + Ascend 910 + Ubuntu_aarch64 + Python3.7.5, manually download the whl package of the corresponding version, compile and install gmp6.1.2. Other Python library dependencies have been installed, the execution of the sample fails, and an error shows that the so file cannot be found.

A: The libdatatransfer.so dynamic library is in the fwkacllib/lib64 directory. Find the path of the library in the /usr/local directory, and then add the path to the LD_LIBRARY_PATH environment variable. After the settings take effect, execute the sample again.


Q: A cross compiler has been installed on Linux, but how do I write compilation commands?

A: To compile the Arm64 version, run bash build.sh -I arm64. To compile the Arm32 version, run bash build.sh -I arm32. To set environment variables, specify the Android NDK path: export ANDROID_NDK=/path/to/android-ndk. After the compilation is successful, find the compiled package in the output directory.


Q: What should I do if the source code compilation process takes too long or is often interrupted?

A: MindSpore imports the third party dependency package through the submodule mechanism, among which Protobuf v3.13.0 might not have the optimal or steady download speed. It is recommended that you perform package cache in advance.


Q: What should I do when the error messageMD5 does not match is displayed when the source code is compiled?

A: This error may be due to network problems at the time of compilation caused by some third-party library download interruption. After recompiling, the file exists but is incomplete, failed to verify MD5. The solution is to delete the relevant third-party libraries in the .mslib cache path, and then recompile.


Q: What should I do when I have installed Python 3.7.5 and set environment variables accordingly, but still failed compiling MindSpore, with error message Python3 not found?

A: It’s probably due to the lack of shared libraries in current Python environment. Compiling MindSpore requires linking Python share libraries, hence you may need to compile and install Python 3.7.5 from source, using command ./configure --enable-shared.


Q: How to change installation directory of the third party libraries?

A: The third party library packages will be installed in build/mindspore/.mslib, and you can change the installation directory by setting the environment variable MSLIBS_CACHE_PATH, eg. export MSLIBS_CACHE_PATH = ~/.mslib.


Q: What are the directories to be cleaned if the previous compilation failed, so as to prevent the following compilation being affected by previous remains?

A: While compiling MindSpore, if:

  1. Failed while downloading or compiling third-party software. e.g. failed applying patch on icu4c, with error message Cmake Error at cmake/utils.cmake:301 (message): Failed patch:. In this case, go to build/mindspore/.mslib or the third-party software installation directory specified by the MSLIBS_CACHE_PATH environment variable and deletes the corresponding software from it.

  2. Failed in other stages of compilation, or if you wish to clean all previous build results, before recompilation, simply delete build directory.


Q: What should I do if it prompts that pthread cannot be found in CMakeError.txt after the compilation fails?

A: The real reason for the failure will be showed in the stdout log. CMakeError.txt has no reference value. Please look for the first error in the stdout log.


Q: After the compilation is successful, an error undefined reference to XXXX or undefined symbol XXXX occurs during runtime, what should I do?

A: The possible reasons are:

  1. If the problem occurs after the code is updated by git pull, please delete the build directory to exclude the impact of the previous build.

  2. If the problem occurs after modifying the code, you can use c++filt XXXX to view the meaning of the symbol. It may be caused by unimplemented functions, unimplemented virtual functions, and unlinked dependencies, etc.

  3. If the problem occurs on the Ascend platform, after excluding the above reasons, it is likely to be caused by a mismatch between the MindSpore version and the CANN version. For the version matching relationship, please refer to Installation Guide.


Q: A warning SetuptoolsDeprecationWarning: setup.py install is deprecated ... occurs when finishing compilation, what should I do?

A: Setuptools, a Python package which manages generation of Python whl packages, has deprecated direct calling of setup.py since 58.3.0, hence you may run into such warnings while using a Python environment with a newer version of setuptools installed. This does not affect the usage of MindSpore, and we will update our pacakaging methods in a future version. For more details, please refer to setuptools version history


Q: What should I do if the software version required by MindSpore is not the same with the Ubuntu default software version?

A: Currently, MindSpore only provides version matching relationships, which requires you to manually install and upgrade the companion software. (Note: MindSpore requires Python3.7.5 and gcc7.3, and the default version in Ubuntu 16.04 are Python3.5 and gcc5, whereas the default one in Ubuntu 18.04 are Python3.7.3 and gcc7.4)


Q: What should I do when the error messageNo module named 'mindpore.version is displayed when the use case is executed?

A: When there is such an error, it is possible to execute a use case in the path that created the same name as the MindSpore installation package, causing Python to preferentially find the current directory when importing the package, and the current directory does not version.py the file. The solution is to rename the directory or exit the one- or multi-level directory upwards.


Installing by Using Docker

Q: Whether MindSpore can support Nvidia GPU discrete graphics + Windows operating system pc?

A: At present, the support of MindSpore is the combination configuration of GPU +Linux and CPU + Windows, and the support of Windows + GPU is still under development.

If you want to run on a GPU+Windows environment, you can try to use WSL+docker, the operation idea:

  1. For installing Ubuntu18.04 in WSL mode, refer to https://docs.microsoft.com/en-us/windows/wsl/install-win10.

  2. For installing Nvidia drivers that support WSL and deploying in environments where containers run on WSL, refer to https://docs.nvidia.com/cuda/wsl-user-guide/index.html.

Since CUDA on WSL is still a preview feature, pay attention to the description of the Windows version requirements in the reference link, and the version is not enough to be upgraded.

  1. Referring to https://gitee.com/mindspore/mindspore#docker镜像, take MindSpore-GPU images. For example, take the MindSpore1.0.0 version container, and execute docker pull mindspore/mindspore-gpu:1.0.0 to execute the container in WSL Ubuntu18.04:

docker run -it --runtime=nvidia mindspore/mindspore-gpu:1.0.0 /bin/bash

The detailed steps can refer to the practice provided by the community Zhang Xiaobai teaches you to install the GPU driver (CUDA and cuDNN). Thanks to the community member Zhang Hui for sharing.


Uninstall

Q: How to uninstall MindSpore?

A: First of all, please confirm the full name of MindSpore, for example, for the gpu version, and you can execute the command pip uninstall mindspore-gpu to uninstall.


Environment Variables

Q: Some frequently-used environment settings need to be reset in the newly started terminal window, which is easy to be forgotten. What should I do?

A: You can write the frequently-used environment settings to ~/.bash_profile or ~/.bashrc so that the settings can take effect immediately when you start a new terminal window.


Q: How to set environment variable DEVICE_ID when using GPU version of MindSpore?

A: Normally, GPU version of MindSpore doesn’t need to set DEVICE_ID. MindSpore automatically chooses visible GPU devices according to the cuda environment variable CUDA_VISIBLE_DEVICES. After setting CUDA_VISIBLE_DEVICES, DEVICE_ID refers to the ordinal of the GPU device:

  • After export CUDA_VISIBLE_DEVICES=1,3,5, DEVICE_ID should be exported as 0, 1 or 2. If 3 is exported, MindSpore will fail to execute because of the invalid device ordinal.


Q: What should I do when error /usr/bin/ld: warning: libxxx.so, needed by libmindspore.so, not found prompts during application compiling?

A: Find the directory where the missing dynamic library file is located, add the path to the environment variable LD_LIBRARY_PATH, and refer to Inference Using the MindIR Model on Ascend 310 AI Processors#Building Inference Code for environment variable settings.


Q: What should I do when error ModuleNotFoundError: No module named 'te' prompts during application running?

A: First confirm whether the system environment is installed correctly and whether the whl packages such as te and topi are installed correctly. If there are multiple Python versions in the user environment, such as Conda virtual environment, you need to execute ldd name_of_your_executable_app to confirm whether the application link libpython3.so is consistent with the current Python directory, if not, you need to adjust the order of the environment variable LD_LIBRARY_PATH. For example:

export LD_LIBRARY_PATH=`python -c "import distutils.sysconfig as sysconfig; print(sysconfig.get_config_var('LIBDIR'))"`:$LD_LIBRARY_PATH

Add the library directory of the program corresponding to the current python command to the top of LD_LIBRARY_PATH.


Q: What should I do when error error while loading shared libraries: libpython3.so: cannot open shared object file: No such file or directory prompts during application running?

A: This error usually occurs in an environment where multiple Python versions are installed. First, confirm whether the lib directory of Python is in the environment variable LD_LIBRARY_PATH, this command can set it:

export LD_LIBRARY_PATH=`python -c "import distutils.sysconfig as sysconfig ; print(sysconfig.get_config_var('LIBDIR'))"`:$LD_LIBRARY_PATH

Then, if Python 3.7.6 or lower does not contain this dynamic library in the Conda virtual environment, you can run the command to upgrade the Python version in Conda, such as: conda install python=3.7.11.


Q: Ascend AI processor software package and other prerequisites have been installed, but executing MindSpore failed with error message Cannot open shared objectfile: No such file or directory, what should I do?

A: There are 2 common reasons: an incorrect version of Ascend AI processor software package is installed, or the packages are installed in customized paths, yet the environment varialbes are not set accordingly.

  1. Go to directory where Ascend AI processor software package is installed, being /usr/local/Ascend by default, open version.info files which are located under directories of subpackages, and check if the version number matches the requirement of MindSpore. Please refer to Install for the detailed version required by MindSpore versions. If the version number does not match, please update the packages accordingly.

  2. Check if Ascend AI processor software package is installed in the default directory. MindSpore attempts to load packages from default directory /usr/local/Ascend, if the packages are installed in a customized directory, please follow the instructions from Configuring Environment Variables section of Install and set environment variables to match the changes. If other dependencies are installed in customized directies, then please set LD_LIBRARY_PATH environment variable accordingly.


Verifying the Installation

Q: Does MindSpore of the GPU version have requirements on the computing capability of devices?

A: Currently, MindSpore supports only devices with the computing capability version greater than 5.3.


Q: After MindSpore is installed on a CPU of a PC, an error message the pointer[session] is null is displayed during code verification. The specific code is as follows. How do I verify whether MindSpore is successfully installed?

import numpy as np
from mindspore import Tensor
import mindspore.ops as ops
import mindspore.context as context

context.set_context(device_target="Ascend")
x = Tensor(np.ones([1,3,3,4]).astype(np.float32))
y = Tensor(np.ones([1,3,3,4]).astype(np.float32))
print(ops.add(x,y))

A: To verify whether MindSpore is installed successfully, please follow the Installation Verification part under installation guide:

python -c "import mindspore;mindspore.run_check()"

The outputs should be the same as:

MindSpore version: __version__
The result of multiplication calculation is correct, MindSpore has been installed successfully!

It means MindSpore has been installed successfully.


Q: What should I do do when the errors prompts, such as sh:1:python:not found, No module named mindspore._extends.remote that the Python was linked to Python2.7 when the use case is executed on linux platform?

A: When you encounter such problem, it is most likely caused by Python environment. Use the following command to check whether the current Python environment meets the requirements of MindSpore.

  • Enter python in the terminal window and check the following version information entered into the Python interactive environment. If an error is reported directly, there is no Python soft connection; if you enter a non-Python 3.7 environment, the current Python environment is not the MindSpore runtime needs.

  • Execute the sudo ln -sf /usr/bin/python3.7.x /usr/bin/python command to create Python’s soft connection, and then check the execution.


Q: Here in script when we import other python lib before import mindspore, error raised like follows (/{your_path}/libgomp.so.1: cannot allocate memory in static TLS block), how can we solve it?

A: Above question is relatively common, and there are two feasible solutions, you can choose one of them:

  • Exchange the order of import, first import mindspore and then import other third party libraries.

  • Before executing the program, we can add environment variables first (export LD_PRELOAD=/{your_path}/libgomp.so.1), where {your_path} is the path mentioned in above error.


Q: What should I do if an error message ImportError: libgmpxx.so: cannot open shared object file: No such file or directory is displayed when running import mindspore in a script after the source code of mindspore and gmp are compiled and installed?

A: The reason is that we didn’t set --enable-cxx when installing gmp. The correct steps for installing gmp is (suppose that we have download gmp installation repository):

$cd gmp-6.1.2
$./configure --enable-cxx
$make
$make check
$sudo make install

Q: What should I do if an warning message UserWarning: The value of the smallest subnormal for <class 'numpy.float64'> type is zero. is displayed when running Mindspore?

A: The above issue occurs in environments with a newer version of numpy (>=1.22.0) version of ARM python 3.9 installed. We observed such warnings on ARM environment, with python 3.9 and numpy >=1.22.0 installed. The wharnings come from numpy and not from MindSpore. If the wharnings affect the normal debugging of the code, you can consider manually installing an earlier version of numpy (<=1.21.2) to circumvent it.