NVVP erroneously reporting single precision float values being used?

Hello, I am trying to profile a PyTorch application using nvvp using half precision tensors. However, nvvp appears to be reporting that I am using single precision float tensors, despite the Kernel function calls appearing as they are using half.
You can see in the attached image, the kernel calls (left column) are using half, but the actual metric on the right is counting the operations as single precision.
I believe that I am correctly casting both the ConvTranspose2D and the input tensor to .half() precision.