koba-jon / lightgrad

Lightest Gradient Calculation Framework based on Define-by-Run for Deep Learning in C++

LightGrad

LightGrad is Lightest Gradient Calculation Framework in C++.
This framework is a learning material created for us to understand Define-by-Run such as PyTorch.

(1) Clone

$ git clone https://github.com/koba-jon/lightgrad.git
$ cd lightgrad

(2) Create Library

$ cd cmake
$ mkdir build
$ cd build
$ cmake ..
$ make install
$ cd ../..

This operation created the directory "<this_repository_name>/lightgrad".

(3) Execute

$ cd example

Set Path

$ vi CMakeLists.txt

Please change the 5th line of "CMakeLists.txt" according to the path of the directory "<this_repository_name>/lightgrad".
The following is the default value for "example".

4: project(Example CXX)
5: list(APPEND CMAKE_PREFIX_PATH ${CMAKE_CURRENT_SOURCE_DIR}/../lightgrad)

Build

For release (default):

$ mkdir build
$ cd build
$ cmake -DCMAKE_BUILD_TYPE=Release ..
$ make
$ cd ..

For debug:

$ mkdir build
$ cd build
$ cmake -DCMAKE_BUILD_TYPE=Debug ..
$ make
$ cd ..

Run

$ ./Example