Skip to content

Latest commit

 

History

History
95 lines (65 loc) · 2.63 KB

build.md

File metadata and controls

95 lines (65 loc) · 2.63 KB
Raw

Build Python Package from Source

Windows and macOS

For build instructions for Windows and macOS, go to:

Linux

This document demonstrates how to build NNabla Python package from source on Ubuntu 20.04 LTS and Python 3.10. It also works for many other linux distributions by replacing the installation commands that use apt in this example by commands for a package manager on your system (e.g. yum).

Setup

The following command install some system and dev tools.

sudo apt install -y cmake curl g++ make git unzip

Python dependencies are installed by:

sudo apt install python3-dev python3-pip python3-setuptools
sudo pip install --upgrade pip  # bump to latest version
sudo pip install --upgrade setuptools  # bump to latest version

You can also use another distribution (e.g. Pyenv, Miniconda).

Google Protocol Buffer compiler is also required to create NNabla's neural network format serializer/desrializer in Python or C++.

curl -L https://github.com/google/protobuf/releases/download/v3.20.3/protoc-3.20.3-linux-x86_64.zip -o /tmp/protoc-3.20.3-linux-x86_64.zip
sudo unzip -d /usr/local /tmp/protoc-3.20.3-linux-x86_64.zip && sudo chmod 755 /usr/local/bin/protoc

Build and installation

Get source code from Github.

git clone https://github.com/sony/nnabla

Install python requirements.

cd nnabla
sudo pip install -r python/setup_requirements.txt
sudo pip install -r python/requirements.txt

Build the C++ core libraries and NNabla Python package.

mkdir build
cd build
cmake ../
make

If you want to install nnabla for python 3.x, you may need to add -DPYTHON_COMMAND_NAME=python3.x to cmake. Without it, the installation may fail. So replace cmake ../. with

cmake ../ -DPYTHON_COMMAND_NAME=python3.10  # if you use python 3.10

Be careful if you have multiple python versions.

Finally, you get the NNabla Python package installer as a wheel file found at ./dist. You can install nnabla using the wheel file using pip command.

cd dist
sudo pip uninstall -y nnabla
sudo pip install nnabla-<package version>-<package-arch>.whl # a name depends on a version and an environment

Verify installation by unit testing

For unit testing, some additional requirements should be installed.

cd ../..
sudo apt install -y liblzma-dev libsndfile1
sudo pip install -r python/test_requirements.txt

Then run(on nnabla directory):

py.test python/test