Environment Setup

Download and install the CANN software package and framework package by referring to Environment Setup.

Install the dependencies listed in Table 1.

**Table 1** Dependency information
Dependency	Version
gcc	8.4 or later You can run the cc --version command to check the GCC version in use. If GCC 9.x is used, Bazel compilation will fail. Therefore, GCC 9.x is not recommended. If you have to use it, click here to rectify related faults.
g++	8.4 or later You can run the c++ --version command to check the GCC version in use. If GCC 9.x is used, Bazel compilation will fail. Therefore, GCC 9.x is not recommended. If you have to use it, click here to rectify related faults.
zip	Any version
unzip	Any version
libtool	Any version
automake	Any version
Python	3.7.5
TensorFlow	1.15.0
tensorflow-serving-api	1.15.0
future	Any version
bazel	0.24.1 or later
CMake	3.14.0 or later
swig	If the operating system architecture is aarch64, the software version must be 3.0.12 or later. If the operating system architecture is x86_64, the software version must be 4.0.1 or later.

The GCC and G++ versions must match. Otherwise, errors may occur during the compilation of TF Serving.
For details about how to use bazel for building, see Installing bazel 0.24.1 from Source Code.
For details about how to use CMake for building, see Installing CMake 3.14.0 from Source Code.
If the swig software package fails to be installed, rectify the fault by referring to What Should I Do If the swig Dependency Package Cannot Be Installed?.
The gcc, g++, zip, unzip, libtool, and automake software packages must be installed using apt or yum. The TensorFlow, tensorflow-serving-api, and future software packages must be installed using pip3.

Parent topic: Deploying TensorFlow Serving–based Online Inference Services