Error using running model from torch script

I have a python3.6 virtualenv created, install opencv-python 3.4.5.20, torch 1.1.0 and torch vision 0.3.0 from https://download.pytorch.org/whl/cu90/torch_stable.html

My goal is to have torchscript save a model that i can run in libtorch 1.3.0, created by checking out the github repo to v1.3.0 and following first suggestions of the instructions here https://github.com/pytorch/pytorch/blob/master/docs/libtorch.rst)

I used torchscript to save a model and when i run it in c++, i got the following error.

terminate called after throwing an instance of ‘std::runtime_error’
what(): cuDNN error: CUDNN_STATUS_NOT_INITIALIZED
The above operation failed in interpreter, with the following stack trace:
at code/posenet_libtorch.py:111:11
_98 = self.offset
_99 = _98.weight
_100 = _98.bias
_101 = self.displacement_fwd
_102 = _101.weight
_103 = _101.bias
_104 = self.displacement_bwd
_105 = _104.weight
_106 = _104.bias
input0 = torch._convolution(input, _2, _3, [2, 2], [1, 1], [1, 1], False, [0, 0], 1, False, False, True)
~~~~~~~~~~~~~~~~~~ <— HERE
input1 = torch.hardtanh(input0, 0., 6.)
input2 = torch._convolution(input1, _6, _7, [1, 1], [1, 1], [1, 1], False, [0, 0], 32, False, False, True)
input3 = torch.hardtanh(input2, 0., 6.)
input4 = torch._convolution(input3, _9, _10, [1, 1], [0, 0], [1, 1], False, [0, 0], 1, False, False, True)
input5 = torch.hardtanh(input4, 0., 6.)
input6 = torch._convolution(input5, _13, _14, [2, 2], [1, 1], [1, 1], False, [0, 0], 64, False, False, True)
input7 = torch.hardtanh(input6, 0., 6.)
input8 = torch._convolution(input7, _16, _17, [1, 1], [0, 0], [1, 1], False, [0, 0], 1, False, False, True)
input9 = torch.hardtanh(input8, 0., 6.)