Hi, I am currently testing the performance of torch.topk() on cpu and I see that the the runtime gets much slower when k becomes huge. Is torch.topk() using quickselection as subroutine or priority queue?
Hi, I am currently testing the performance of torch.topk() on cpu and I see that the the runtime gets much slower when k becomes huge. Is torch.topk() using quickselection as subroutine or priority queue?