hi. I only want to use “convolution” in libtorch. there are many .h files including what I don’t need, and it takes too much time to compile them all. Can I get some help using these? thank you.
I don’t think you can simply rip out a module from torch and build it.
In case you don’t need Autograd etc. you might want to check the actual conv kernel implementations for the CPU and CUDA, and just reuse them in your application.
Alternatively, check other repositories such as cutlass’ convolution implementation.