Installing MindSpore in GPU by Source Code

• Installing MindSpore in GPU by Source Code
  • System Environment Information Confirmation
  • Downloading Source Code from Code Repository
  • Compiling MindSpore
  • Installing MindSpore
  • Installation Verification
  • Installing MindInsight
  • Installing MindArmour
  • Installing MindSpore Hub

This document describes how to quickly install MindSpore by source code in a Linux system with a GPU environment.

System Environment Information Confirmation

• Confirm that Ubuntu 18.04 is installed with 64-bit operating system.

• Confirm that GCC 7.3.0 is installed.

• Confirm that gmp 6.1.2 is installed.

• Confirm that Python 3.7.5 is installed.

• Confirm that CMake 3.18.3 or later is installed.

  • After installing, add the path of cmake to the environment variable PATH.

• Confirm that patch 2.5 or later is installed.

  • After installing, add the path of patch to the environment variable PATH.

• Confirm that Autoconf 2.69 or later is installed. (Default versions of these tools built in their systems are supported.)

• Confirm that Libtool 2.4.6-29.fc30 or later is installed. (Default versions of these tools built in their systems are supported.)

• Confirm that Automake 1.15.1 or later is installed.(Default versions of these tools built in their systems are supported.)

• Confirm that cuDNN 7.6 or later is installed.

• Confirm that Flex 2.5.35 or later is installed.

• Confirm that wheel 0.32.0 or later is installed.

• Confirm that OpenSSL 1.1.1 or later is installed.

  • ensure that OpenSSL is installed and set system variable export OPENSSL_ROOT_DIR="OpenSSL installation directory".

• Confirm that CUDA 10.1 is installed as default configuration.

  • If CUDA is installed in a non-default path, after installing CUDA, environment variable PATH(e.g. export PATH=/usr/local/cuda-${version}/bin:$PATH) and LD_LIBRARY_PATH(e.g. export LD_LIBRARY_PATH=/usr/local/cuda-${version}/lib64:$LD_LIBRARY_PATH) need to be set. Please refer to CUDA installation guide for detailed post installation actions.

• Confirm that OpenMPI 3.1.5 is installed. (optional, required for single-node/multi-GPU and multi-node/multi-GPU training)

• Confirm that NCCL 2.7.6-1 is installed. (optional, required for single-node/multi-GPU and multi-node/multi-GPU training)

• Confirm that the git tool is installed.
  If not, use the following command to install it:

  apt-get install git
  

Downloading Source Code from Code Repository

git clone https://gitee.com/mindspore/mindspore.git -b r1.0

Compiling MindSpore

Run the following command in the root directory of the source code to compile MindSpore:

bash build.sh -e gpu

Of which,

• In the build.sh script, the default number of compilation threads is 8. If the compiler performance is poor, compilation errors may occur. You can add -j{Number of threads} in to script to reduce the number of threads. For example, bash build.sh -e ascend -j4.

Installing MindSpore

chmod +x build/package/mindspore_gpu-{version}-cp37-cp37m-linux_{arch}.whl
pip install build/package/mindspore_gpu-{version}-cp37-cp37m-linux_{arch}.whl -i https://pypi.tuna.tsinghua.edu.cn/simple

Of which,

• When the network is connected, dependency items are automatically downloaded during .whl package installation. (For details about other dependency items, see requirements.txt). In other cases, you need to manually install dependency items.
• {version} denotes the version of MindSpore. For example, when you are downloading MindSpore 1.0.1, {version} should be 1.0.1.
• {arch} denotes the system architecture. For example, the Linux system you are using is x86 architecture 64-bit, {arch} should be x86_64. If the system is ARM architecture 64-bit, then it should be aarch64.

Installation Verification

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

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

• The outputs should be the same as:

[[[ 2.  2.  2.  2.],
    [ 2.  2.  2.  2.],
    [ 2.  2.  2.  2.]],

    [[ 2.  2.  2.  2.],
    [ 2.  2.  2.  2.],
    [ 2.  2.  2.  2.]],

    [[ 2.  2.  2.  2.],
    [ 2.  2.  2.  2.],
    [ 2.  2.  2.  2.]]]

It means MindSpore has been installed successfully.

Version Update

Using the following command if you need update MindSpore version.

• Update online

  pip install --upgrade mindspore-gpu
  

• Update after source code compilation

  After successfully executing the compile script build.sh in the root path of the source code, find the whl package in path build/package, use the following command to update your version.

  pip install --upgrade mindspore_gpu-{version}-cp37-cp37m-linux_{arch}.whl
  

Installing MindInsight

If you need to analyze information such as model scalars, graphs, computation graphs and model traceback, you can install MindInsight.

For more details, please refer to MindInsight.

Installing MindArmour

If you need to conduct AI model security research or enhance the security of the model in you applications, you can install MindArmour.

For more details, please refer to MindArmour.

Installing MindSpore Hub

If you need to access and experience MindSpore pre-trained models quickly, you can install MindSpore Hub.

For more details, please refer to MindSpore Hub.