Jetson TX2 configures common libraries such as tensorflow and pytoch

Before that PC Ubuntu Or raspberry pie , The method is the same , So this blog will Simple record Next process , For detailed methods and ideas, please refer to the previous series of blogs of bloggers （ Although the hardware platform is different , But the method is generally the same . The mainstream hardware platforms include raspberry pie ,NVIDIA Jetson,Google Of Coral Dev Board etc. . Teaching a man to fish is better than giving him a fish , Our focus is on learning methods , Maintaining the status quo . From the configuration of common libraries in the previous series of blogs , Inference of deep learning , How to cross compile has done some detailed experiments and explanations , These ideas and habits can be applied to the new embedded platform ）

 

https://blog.csdn.net/jiugeshao/category_11391557.html?spm=1001.2014.3001.5482

https://blog.csdn.net/jiugeshao/category_11447160.html?spm=1001.2014.3001.5482

  about Jetson TX2 The initial environment configuration of can be seen in the previous three blogs

NVIDIA Jetson TX2 brief introduction _ green bamboo snake lvye The blog of -CSDN Blog _nvidiatx2

NVIDIA Jetson Official website data sorting _ green bamboo snake lvye The blog of -CSDN Blog

NVIDIA Jetson TX2 install JetPack_ green bamboo snake lvye The blog of -CSDN Blog

One . Configure the specified python edition

Choose here python3.7, Refer to previous blogs of bloggers for methods

Raspberry pie 4B Multiple versions on python Switch （ One ）_ green bamboo snake lvye The blog of -CSDN Blog _ Raspberry pie python Version change

notes ：

(1) The deletion of soft links can be referred to Ubuntu Basic knowledge points under （ One ）_ green bamboo snake lvye The blog of -CSDN Blog _ubuntu Basic knowledge of

(2) here pip3 The link target of should be switched to the following （ Combine your own path ）

  After the completion of , The input terminal python3 and pip3 -V

  You can see that you have switched to python3.7 了 , Not the system's own python3.6. 

notes ：

If you find that clicking on the terminal on the desktop does not respond ,terminal Cannot be opened , You can open any folder , Right click in the blank to open the terminal . Edit the file in the following path

  The top is changed to the current system default python3.6, The original is python3.（ Because when you build a soft link ,python3 Has been linked to python3.7 Version of the , So this side pierced the window , Tell the system directly , What you need is python3.6）. After the completion of , Click on the terminal on the desktop to open .

Two . To configure CUDNN

The blogger reinstalled it in front JetPack

NVIDIA Jetson TX2 Reshipment system _ green bamboo snake lvye The blog of -CSDN Blog

It depends on the default cuda and cudnn The system can view the following command statements

 cd To the following directory

/usr/src/cudnn_samples_v8/mnistCUDNN

perform sudo make, Execute after completion ./mnistCUDNN, Show that the test passed .

Similar to previous blogs of bloggers , We need to cudnn Some header files and lib Copy library to cuda Let's go down

TensorRT Installation and verification _ green bamboo snake lvye The blog of -CSDN Blog _ verification tensorrt

For the present TX2 Configuration of , Execute the following command statement

sudo cp /usr/lib/aarch64-linux-gnu/libcudnn* /usr/local/cuda-10.2/lib64

3、 ... and . install Tensorflow

Mainly refer to the methods in bloggers' previous blogs

Raspberry pie 4B install Tensorflow2.4.0_ green bamboo snake lvye The blog of -CSDN Blog _ Raspberry pie installation tensorflow

  Select the following version here

  After downloading, use the following command to install

pip install tensorflow-2.4.0-cp37-none-linux_aarch64.whl --user

  If the following error message appears during installation ：

  You can manually compile the source code hdf5, Available from website Download HDF5 download , Bloggers choose hdf5-1.10.6

  After the completion of , stay hdf5-1.10.6 Create a new folder install Folder , Then execute the following command line statement to compile

./configure --prefix=/home/sxhlvye/Downloads/hdf5-1.10.6/install
sudo make -j4
sudo make install

After the completion of , stay install You can see the installed Library in the folder

  notes ： Don't forget it. ~/.bashrc Configure environment variables in , as follows ：

  After the completion of , You can continue to execute the following two installation statements ：

sudo apt-get install libhdf5-serial-dev
sudo apt-get install libhdf5-dev

After that, execute the following command statement again

pip install tensorflow-2.4.0-cp37-none-linux_aarch64.whl --user

This installation is successful , Here's the picture ：

  test ： as follows import The library does not report an error

Four . install pycharm

Refer to blog download and installation （ The community version is selected here , There is no need to activate ）Ubuntu20.04 C++ Simple compilation and QT and Pycharm Configuration of _ green bamboo snake lvye The blog of -CSDN Blog

In execution ./pycharm.sh When , There is an error ：No JRE found.Please make sure $PYCHARM_JDK,$JDK_HOME,or $JAVA_HHOME point to valid JRE installation

Need to configure jdk.

1. Use apt-cache search openjdk Look at the current TX2 Supported in the environment jdk edition

Configuration here jdk11

2. The following command line statement can be used to install

sudu apt-get install openjdk-11-jdk

  After installation , Executable statement view openjdk Installation path for

dpkg -L openjdk-11-jdk

  The following commands can be viewed jdk Version of

java -version

3. Re execution ./pycharm.sh, pycharm It can be loaded normally

  But see the error message ：Failed to load moudle "canberra-gtk-moudle"

  We'll solve it later .

5、 ... and . Update source

You can refer to the blog update source before the blogger

VMware Ubuntu20.04 Next mount disk 、 Update source and other common configurations _ green bamboo snake lvye The blog of -CSDN Blog _ubuntu vmware Mount the hard disk

  Here, select the above source , But remember to put the inside ubuntu Change it to ubuntu-ports

  After the completion of , Don't forget to execute the statement sudo apt-get update

  Updated completely , Try to install the following statement

 sudo apt-get install libcanberra-gtk-module

To solve the error reporting mentioned in the third part above . Open it again at this time pycharm, There will be no error messages in the background . stay pycharm To test the previous configuration tensorflow Information about

 pycharm You can refer to previous blogs of bloggers , As can be seen from the above figure , The second part contains tensorflow The version is not GPU Version of , So next, through the virtual environment , stay python3.7 Under the virtual environment configuration of version tensorflow-gpu

6、 ... and . To configure Tensorflow-GPU

python3 For the configuration of virtual environment, please refer to the blogger's previous blog

Raspberry pie 4B Multiple versions on python Switch （ Two ）_ green bamboo snake lvye The blog of -CSDN Blog _ Raspberry pie switch python2

https://files.pythonhosted.org/packages/9e/34/e86fc6a8f84329b49321a532b3c1fef103c67765df957fbb3852eea39d00/virtualenv-20.14.1-py2.py3-none-any.whl 

setup script , Mainly refer to the official website DOCS This column

NVIDIA Documentation Center | NVIDIA Developer

  After entering the point , You can see the installation instructions

Installing TensorFlow for Jetson Platform :: NVIDIA Deep Learning Frameworks Documentation

  Bloggers will not repeat , Just follow the document , It's still very detailed . Used by bloggers TX2 The configuration of is visible before the blog

NVIDIA Jetson TX2 Reshipment system _ green bamboo snake lvye The blog of -CSDN Blog

The link also contains examples of official tensorflow Some information

TensorFlow for Jetson Platform Release Notes :: NVIDIA Deep Learning Frameworks Documentation

  Bloggers install Tensorflow 2.7 edition ( Used by bloggers python3 Environment or installation Jetpack The system defaults to python3.6 Environmental Science , Not configured in the previous python3.7 Installation in environment tensorflow2.7).

Bloggers download it first , Install again , The download address is as follows

Index of /compute/redist/jp/v461/tensorflow

After downloading , Again pip Installation , that will do

verification ：

stay pycharm Test the following code ：

import tensorflow as tf
print(tf.test.is_gpu_available())
print(tf.config.list_physical_devices('GPU'))
print(tf.test.is_built_with_cuda())

The results are as follows ：

/usr/bin/python3 /home/sxhlvye/PycharmProjects/pythonProject/test1.py
WARNING:tensorflow:From /home/sxhlvye/PycharmProjects/pythonProject/test1.py:2: is_gpu_available (from tensorflow.python.framework.test_util) is deprecated and will be removed in a future version.
Instructions for updating:
Use `tf.config.list_physical_devices('GPU')` instead.
2022-06-25 07:03:13.240374: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1019] ARM64 does not support NUMA - returning NUMA node zero
2022-06-25 07:03:13.465336: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1019] ARM64 does not support NUMA - returning NUMA node zero
2022-06-25 07:03:13.465659: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1019] ARM64 does not support NUMA - returning NUMA node zero
2022-06-25 07:03:17.279189: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1019] ARM64 does not support NUMA - returning NUMA node zero
2022-06-25 07:03:17.279540: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1019] ARM64 does not support NUMA - returning NUMA node zero
2022-06-25 07:03:17.279799: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1019] ARM64 does not support NUMA - returning NUMA node zero
2022-06-25 07:03:17.280005: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1525] Created device /device:GPU:0 with 638 MB memory:  -> device: 0, name: NVIDIA Tegra X2, pci bus id: 0000:00:00.0, compute capability: 6.2
True
[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]
True
2022-06-25 07:03:17.291743: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1019] ARM64 does not support NUMA - returning NUMA node zero
2022-06-25 07:03:17.293825: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1019] ARM64 does not support NUMA - returning NUMA node zero
2022-06-25 07:03:17.295500: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1019] ARM64 does not support NUMA - returning NUMA node zero

Process finished with exit code 0

As you can see above, it has been successfully installed , Can run in GPU On .

This side also runs down the previous blog （ stay Raspberry pie Up operation ） The code in

Raspberry pie 4B install Tensorflow2.4.0_ green bamboo snake lvye The blog of -CSDN Blog _ Raspberry pie installation tensorflow

The operation results are as follows ,ct need 19s：

7、 ... and . To configure Pytorch

Refer to official documentation

Installing PyTorch for Jetson Platform :: NVIDIA Deep Learning Frameworks Documentation

Bloggers should download the installation package first , Again pip install install （ Before installation, pre configure the environment according to the official website ）

Index of /compute/redist/jp/v461/pytorch

After installation , The following tests , No report error ,torch Successful installation

Continue to install torchvision library , The following command line statement is sufficient ,torch、torchvision、torchaudio The corresponding relationship of can be viewed from the official website

https://pytorch.org/get-started/locally/

pip install torchvision==0.10.0 --default-timeout=100000
pip install torchaudio

  After the completion of , stay pycharm The following code tests are carried out in ：

import torch
import torchvision
from torchvision import transforms
import torchvision.models as models

print("over")
print(torch.cuda.is_available())
print(torch.__version__)
print(torch.backends.cudnn.version())

The operation results are as follows ： 

  You can see pytorch Configured successfully . Blog before running off the blogger （ stay Raspberry pie On ） Test code in

Raspberry pie 4B install pytorch_ green bamboo snake lvye The blog of -CSDN Blog _ Raspberry pie 4b install pytorch

  You can see it in TX2 Time to run , need 28s.

8、 ... and . To configure TensorRT

After reinstalling the system in front , default python3.6 There are already in the environment tensorrt The library of , Directly paste the acceleration effect picture here , Refer to previous blogs of bloggers for specific methods .

TensorRT Introduction to acceleration methods （python pytorch Model ）_ green bamboo snake lvye The blog of -CSDN Blog _tensorrt Speed up

need pip3 install pycuda, If the following errors occur during installation

In file included from src/cpp/cuda.cpp:4:0:
  src/cpp/cuda.hpp:14:10: fatal error: cuda.h: No such file or directory
   #include <cuda.h>
            ^~~~~~~~
  compilation terminated.
  error: command 'aarch64-linux-gnu-gcc' failed with exit status 1

You can add related variables to environment variables

export CPATH=$CPATH:/usr/local/cuda/include

In execution import torch when , If reported ：OSError: libomp.so No problem found , You can add the following environment variables

  stay TX2 The operation information on is as follows ,ct Time is greatly accelerated ：

Energy co., LTD. , Only record the configuration of the above common libraries , You can access its process and on other hardware platforms ubuntu There is little difference in the system environment .