Question model with 1D tensor as an input generates wrong result

I am trying to use a model from Pytorch in Unity. I successfully converted it to onnx (the onnx results are almost the same as pytorch model's when I test it in python).

This is how I define the onnx model during the export:

Code (Boo):

1.  
2. dummy_input = torch.randn(1,636).to(device)
3. torch.onnx.export(model,                     # model being run
4.                  dummy_input,                # model input (or a tuple for multiple inputs)
5.                   "./onnx/model_A.onnx",     # where to save the model (can be a file or file-like object)
6.                   export_params=True,        # store the trained parameter weights inside the model file
7.                   opset_version=10,          # the ONNX version to export the model to
8.                   do_constant_folding=True,  # whether to execute constant folding for optimization
9.                   input_names = ['input'],   # the model's input names
10.                   output_names = ['output'], # the model's output names
11.                   dynamic_axes={'input' : {0 : 'batch_size', 1: 'vertices'},    # variable length axes
12.                                 'output' : {0 : 'batch_size',1: 'blendshape_values'}})
13.  

this is how I define input tensor inside unity:

Code (CSharp):

1.  
2. var inputTensorA = new Tensor(1, 636, input); # i tried this
3. var inputTensorB = new Tensor( new int[]{1,1,636,1,1,1,1,1}, input,"",true); # i also tried this
4.  

I tried putting the 636 in different dimensions as well. I tried all the combinations that came to my mind, however the model's output is always wrong. The model itself is very simple - just a few fully connected layers. What am I doing wrong? How to debug it?

 

please try rounding in python.
unity can not calculate precisely float 64.