Missing kthvalue on GPU

I am trying to use torch.kthvalue function. While the code runs perfectly on CPU, I get the following error on GPU.

RuntimeError: Type FloatTensor doesn't implement stateless method kthvalue

Is this a missing PyTorch feature? If so, are there any workarounds?

Edit: I currently perform the kthvalue operation in CPU and move the results to GPU.

kthvalue is a missing feature. There are 3 or 4 functions still missing on the GPU.
topk is available on the GPU for FloatTensor and currently being implemented for other tensor types. Maybe you can get torch.topk and get the last value? That might be faster