Cudnn_status_internal_error

when I use nn.DataParallel, this CUDNN_STATUS_INTERNAL_ERROR would occur. And I’ve tried to use ‘rm -rf ~/.nv’, however it doesn’t work. Does anyone meet this problem?