Closed filiptibell closed 5 months ago
It would be nice if we supported CUDA. However, I'm a Mac user, and testing with NVIDIA's GPUs is not easy for me...
It seems like bindgen_cuda has some logic for finding proper CUDA paths to link and it can probably be added as an optional dependency similar to intel-mkl-src
in #23 . It needs an attached gpu to build but seems like that is also pretty common. I can take a look at this in a week or so
CUDA_TOOLKIT_ROOT_DIR
environment variable is set.
Similar to #23 , would you be open to adding support for building for CUDA? I have not yet figured out how to do this (don't have much experience with cc/cxx/build scripts) but this would be great for deploying on infra such as HF Inference Endpoints.