项目作者: kyamagu

项目描述 :
Unofficial faiss wheel builder
高级语言: Python
项目地址: git://github.com/kyamagu/faiss-wheels.git
创建时间: 2019-06-27T02:29:15Z
项目社区:https://github.com/kyamagu/faiss-wheels

开源协议:MIT License

下载


faiss-wheels

Build
PyPI

faiss python wheel packages.

Overview

This repository provides scripts to build wheel packages for the
faiss library.

  • Builds CPU-only version with cibuildwheel.
  • Bundles OpenBLAS in Linux/Windows
  • Uses Accelerate framework in macOS

There is also a source package to customize the build process.

Note
GPU binary package is discontinued as of 1.7.3 release. Build a source package to support GPU features.

Install

Install the CPU-only binary package by:

  1. pip install faiss-cpu

Note that the package name is faiss-cpu.

Supporting GPU or customized build configuration

The PyPI binary package does not support GPU.
To support GPU methods or use faiss with different build configuration, build a source package.
For building the source package, swig 3.0.12 or later needs to be available.
Also, there should be all the required prerequisites for building faiss itself, such as nvcc and CUDA toolkit.

Building faiss

The source package assumes faiss is already built and installed in the system.
If not done so elsewhere, build and install the faiss library first.
The following example builds and installs faiss with GPU support and avx512 instruction set.

  1. git clone https://github.com/facebookresearch/faiss.git
  2. cd faiss
  3. cmake . -B build -DFAISS_ENABLE_GPU=ON -DFAISS_ENABLE_PYTHON=OFF -DFAISS_OPT_LEVEL=avx512
  4. cmake --build build --config Release -j
  5. cmake --install build install
  6. cd ..

See the official
faiss installation instruction
for more on how to build and install faiss.

Building and installing a source package

Once faiss is built and installed, build the source package.
The following builds and installs the faiss-cpu source package with GPU and AVX512.

  1. export FAISS_ENABLE_GPU=ON FAISS_OPT_LEVEL=avx512
  2. pip install --no-binary :all: faiss-cpu

There are a few environment variables that specifies build-time options.

  • FAISS_INSTALL_PREFIX: Specifies the install location of faiss library, default to /usr/local.
  • FAISS_OPT_LEVEL: Faiss SIMD optimization, one of generic, avx2, avx512. Note that AVX option is only available in x86_64 arch.
  • FAISS_ENABLE_GPU: Setting this variable to ON builds GPU wrappers. Set this variable if faiss is built with GPU support.
  • CUDA_HOME: Specifies CUDA install location for building GPU wrappers, default to /usr/local/cuda.

Note that you can build a custom wheel package without installing it. The resulting package can be installed in the other python environment as long as the ABI is the same. Otherwise, use auditwheel or similar tools to package the binary dependency after building a wheel.

  1. export FAISS_ENABLE_GPU=ON FAISS_OPT_LEVEL=avx512
  2. pip wheel --no-binary :all: faiss-cpu

Development

This repository is intended to support PyPI distribution for the official faiss library.
The repository contains the CI workflow based on cibuildwheel.
Feel free to make a pull request to fix packaging problems.

Other relevant resources: