Quick implementation/experimentation with OpenNMT's CTranslate2 library, primarily meant to just serve as an example of the core setup that is required.
- CUDA Developer Toolkit v11.0
- NVIDIA cuDNN for CUDA v11.0
- ICU Libraries
- A transformer model; for this project, the CTranslate quickstart model is recommended.
Ensure you have extracted cuDNN within your CUDA Developer Toolkit installation and that such libraries are within the system PATH, then:
git clone --recursive "https://github.com/NM-20/CTranslate2Test.git".
The converted model should be placed within the TranslateTest project's "res" directory.
Compile the libraries within externals, ensuring that CTranslate2 is compiled with the following settings:
-G "Visual Studio 16 2019" -DOPENMP_RUNTIME=COMP -DWITH_CUDA=ON -DWITH_CUDNN=ON -DWITH_MKL=OFF
From there, you may proceed with building the TranslateTest project in VS 2022.