CrystalNet: A Deep Learning Library for Minimalists.
Crystalnet is a novel deep learning library written in modern C and C++. Its core library has less than 1K lines of code. Though small, it is able to achieve high-performance comparable to giant engines such as Google TenserFlow.
Crystalnet is very easy to learn and extend. It is a perfect choice for deep learning students and hackers. Students can easily learn the essential pieces of a modern deep learning system in hours. Hackers can quickly craft advanced system features based on Crystalnet in days.
Crystalnet is under active development. We are going to release more details once it is ready.
Getting start
Crystalnet relies on modern C and C++ features to write concise yet performant code. Its build process requires Cmake (>=3.9) and a contemporary C++ compiler (fully support C++17).
$ make
$ ./utils/download-mnist.sh
$ ./build/$(uname)/bin/mnist_slp_c