README.md

ddnet-nn

To compile you would need to install libtorch with cuda support.

Steps to build on Windows:

1. Install CUDA the same version that libtorch with cuda support is built for.
2. Install Nsight NVTX from CUDA 11.8.
3. Download libtorch with cuda support and add it to ddnet-libs
4. Call cmake so it will generate files to build
5. Link dependencies in Visual Studio for DDNet-Server:

   Libraries:

   D:\GitHub\ddnet-nn\ddnet-libs\libtorch-win-shared-with-deps-latest+cu\libtorch\lib\c10.lib
   D:\GitHub\ddnet-nn\ddnet-libs\libtorch-win-shared-with-deps-latest+cu\libtorch\lib\kineto.lib
   D:\GitHub\ddnet-nn\ddnet-libs\libtorch-win-shared-with-deps-latest+cu\libtorch\lib\caffe2_nvrtc.lib
   D:\GitHub\ddnet-nn\ddnet-libs\libtorch-win-shared-with-deps-latest+cu\libtorch\lib\c10_cuda.lib
   D:\GitHub\ddnet-nn\ddnet-libs\libtorch-win-shared-with-deps-latest+cu\libtorch\lib\torch.lib
   D:\GitHub\ddnet-nn\ddnet-libs\libtorch-win-shared-with-deps-latest+cu\libtorch\lib\torch_cuda.lib
   C:\Program Files\NVIDIA Corporation\NvToolsExt\lib\x64\nvToolsExt64_1.lib
   D:\GitHub\ddnet-nn\ddnet-libs\libtorch-win-shared-with-deps-latest+cu\libtorch\lib\torch_cpu.lib
   -INCLUDE:?warp_size@cuda@at@@YAHXZ
   C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4\lib\x64\cudart.lib

   Include directories

   C:/Program Files/NVIDIA Corporation/NvToolsExt
   C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.4/include
   D:\GitHub\ddnet-nn\ddnet-libs\libtorch-win-shared-with-deps-latest+cu\libtorch\include\torch\csrc\api\include
   D:\GitHub\ddnet-nn\ddnet-libs\libtorch-win-shared-with-deps-latest+cu\libtorch\include
6. Now try to compile DDNet-Server.
7. Copy DLL-s from libtorch folder to run it

Example of trained neural network - https://www.youtube.com/watch?v=LaiUJSzhEJc