flashlight is a fast, flexible machine learning library written entirely in C++ from the Facebook AI Research Speech team and the creators of Torch and Deep Speech. It uses the ArrayFire tensor library and features just-in-time compilation with modern C++. flashlight supports both CPU (still in active development) and GPU backends for maximum portability, and has an emphasis on efficiency and scale.
All documentation (including build/install instructions) can be found here
Experimental and in-progress project components are located in flashlight/contrib. Breaking changes may be made to APIs therein.
Contact: [email protected], [email protected], [email protected], [email protected], [email protected], [email protected], [email protected], [email protected]
flashlight is being very actively developed. See CONTRIBUTING for more on how to help out.
flashlight and relevant dependencies can be easily built and installed on Linux-based systems using the Conan package manager. Only the flashlight CUDA backend is supported at this time. After installing Conan:
conan remote add bincrafters https://api.bintray.com/conan/bincrafters/public-conan # for OpenMPI
conan remote add flashlight https://api.bintray.com/conan/flashlight/flashlight # for ArrayFire/flashlight
conan install flashlight/0.1@conan/stable -r=flashlight
if no pre-built version of flashlight or ArrayFire exists on your system, you can build everything from source with:
conan install flashlight 0.1@conan/stable -r=flashlight --build arrayfire --build flashlight
See the documentation for more on how to use Conan and flashlight with your own project.
Some of flashlight's code is derived from arrayfire-ml.
flashlight is under a BSD license. See LICENSE.