Installing Protobuf of the Python Version

If the training script depends on the Python version of Protobuf to store data in the serialized structure (for example, the serialization interfaces of TensorFlow), you need to install Protobuf Python.

Check that the system contains the dynamic library /usr/local/python3.7.5/lib/python3.7/site-packages/google/protobuf/pyext/_message.cpython-37m-<arch>-linux-gnu.so. If not, perform the following steps to install it. <arch> indicates the system architecture type.

/usr/local/python3.7.5/lib/python3.7/site-packages is the path for installing third-party libraries using pip. You can run the pip3 -V command to query the path.

If the system displays /usr/local/python3.7.5/lib/python3.7/site-packages/pip, the path for installing third-party libraries using pip is /usr/local/python3.7.5/lib/python3.7/site-packages.
Run the following command to uninstall Protobuf:
```
pip3 uninstall protobuf
```
Download the Protobuf software package.
Download protobuf-python-3.11.3.tar.gz (or an alternative version that is compatible with TensorFlow installed in the current environment) from https://github.com/protocolbuffers/protobuf/releases/download/v3.11.3/protobuf-python-3.11.3.tar.gz, upload the package to any directory on the Linux server as the root user, and decompress the package.
```
tar zxvf protobuf-python-3.11.3.tar.gz
```
Install Protobuf as the root user.
Go to the Protobuf software package directory.
1. Install the dependencies of Protobuf.
  If the OS is Ubuntu, run the following command:
  apt-get install autoconf automake libtool curl make g++ unzip libffi-dev -y
  If the OS is CentOS or BC-Linux, run the following command:
  yum install autoconf automake libtool curl make gcc-c++ unzip libffi-devel -y
2. Grant the execute permission on the autogen.sh script and execute the script.
```
chmod +x autogen.sh
./autogen.sh
```
3. Configure the installation path. The default installation path is /usr/local.
```
./configure
```
  To specify the installation path, run the following command:
```
./configure --prefix=/protobuf
```
  /protobuf indicates the installation path specified by the user.
4. Run the following commands to install Protobuf:
```
make -j15        # Check the number of CPUs by running grep -w processor /proc/cpuinfo|wc -l. In this example, the number is 15. You can set the parameters as required.
make install
```
5. Refresh the shared libraries.
```
ldconfig
```
  After Protobuf is installed, the google/protobuf folder is generated in the include directory in 4.c. This folder stores the header files related to Protobuf. Generate the protoc executable file in the bin directory in 4.c. The protoc executable file is used to compile the *.proto file and generate the C++ header file and implementation file of Protobuf.
6. Check whether the installation is complete.
```
ln -s /protobuf/bin/protoc /usr/bin/protoc
protoc --version
```
  /protobuf is the installation path configured in 4.c. If the installation path is not configured, run the protoc --version command to check whether the installation is successful.
Install the runtime library of Protobuf Python.
1. Go to the python subdirectory in the Protobuf software package directory and compile the Python runtime library.
```
python3 setup.py build --cpp_implementation
```
  Build a runtime library of the binary version. If the runtime library of the binary version cannot be generated by running the python3 setup.py build command, the processing of data in the serialized structure is slow.
2. Install the dynamic library.
```
cd .. && make install
```
  Go to the python subdirectory and install the Python runtime library.
```
python3 setup.py install --cpp_implementation
```
3. Check whether the installation is successful.
  Check whether the system contains the dynamic library /usr/local/python3.7.5/lib/python3.7/site-packages/protobuf-3.11.3-py3.7-linux-aarch64.egg/google/protobuf/pyext/_message.cpython-37m-<arch>-linux-gnu.so. <arch> indicates the system architecture type.
  
  /usr/local/python3.7.5/lib/python3.7/site-packages is the path for installing third-party libraries using pip. You can run the pip3 -V command to query the path.
  
  If the system displays /usr/local/python3.7.5/lib/python3.7/site-packages/pip, the path for installing third-party libraries using pip is /usr/local/python3.7.5/lib/python3.7/site-packages.
4. If you have specified the installation path in 4.c, add the following environment variable settings to the run script:
```
export LD_LIBRARY_PATH=/protobuf/lib:${LD_LIBRARY_PATH}
```
  /protobuf is the installation path configured in 4.c.
5. Create a soft link.
  If you have specified the installation path, you need to establish a soft link. Otherwise, an error will be reported when TensorFlow is imported. The command is as follows:
```
ln -s /protobuf/lib/libprotobuf.so.22.0.3 /usr/lib/libprotobuf.so.22
```
  /protobuf is the installation path configured in 4.c.

Parent topic: Common Operations