Core dump when using PyTorch built from sources and setting cudnn.benchmark = True

I installed the 0.2+5254846 (master branch) just now but it seems that I still cannot use cudnn.benchmark = True. Am I misunderstanding something?

Thanks for your quick reply!