cannot find cuDNN

[pvandwalle]

This is really amazing but so far it doesnt find cuDNN, which is installed in C:\Program Files\NVDIA\CUDNN\v8.x
Hence the project cannot be created at all.

" you choose to create a torch project with cuda but cuDNN isn't installed. "

whereas pytroch is able to use cudNN

[adlane98]

Try copying:

• all .h files in your CUDNN\v8.x\include folder to $CUDA_PATH\include
• all .lib files in your CUDNN\v8.x\lib folder to $CUDA_PATH\lib\x64

[mszhanyi]

https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html#install-windows

[NinjaDoggy]

I ran into the same issue, and I'm pretty sure cuDNN was installed properly because I was able to run a cmake project with libtorch.

Code:
#include <torch/torch.h>
#include

int main() {
torch::Tensor tensor = torch::rand({ 2, 3 });
std::cout << "cudnn is available: " << torch::cuda::cudnn_is_available() << "\n";
if (torch::cuda::is_available()) {
std::cout << "CUDA is available! Training on GPU\n";
auto tensor_cuda = tensor.cuda();
std::cout << tensor_cuda << "\n";
} else {
std::cout << "CUDA is not available! Training on CPU\n";
std::cout << tensor << "\n";
}
}

Result:
cudnn is available: 1
CUDA is available! Training on GPU
0.6249 0.2578 0.3856
0.5703 0.2497 0.4719
[ CUDAFloatType{2,3} ]

Copying the files like @adlane98 suggested fixed it!

[mszhanyi]

@NinjaDoggy , Thank your feedback. did you set CUDNN_PATH in your machine.?
So far, the extension didn't support CUDNN_PATH, I'll add the support later.
Now, the workaround is to copy cudnn files as #11 (comment)