Hi, is there a way or a library to perform only the inference part of XGBoost algorithm in CUDA C (I’m using Visual Studio)?
I downloaded the xgboost.dll library that supports GPU but it includes also the training part as I understood.
Hi @john455
We have Forest Inference Library as part of cuML here: API Reference — cuml 23.06.00 documentation. Yes, it is python API, but that’s built on top of the underlying C++ implementation in cuML.
It is possible to build a libcuml.so
that includes only FIL, and you should get better performance than with the native framework inference.
However this channel talks about issues related to cudnn, and Rapids should be the right channel to get better assistance.
Thank you.
So, no way to do it in CUDA C?