What are the requirements when building the Windows version of libtorch program running on h100? It was built with libtorch 2.1.1 and cuda cudnn v11.8, but execution failed on h100.
In this case, is there no libtorch function that obtains any message? For example, PyTorch's get_device_name function can obtain even a few messages, but there is none in libtorch.
Does libtorch not support h100?