Minimal difference while converting .pt model to .onnx, using torch.export.onnx(), what is the reason?

Hi Everyone,

What could be the possible reason for the slight difference in model weights, while exporting .pt model to .onnx,
I havn’t looked into the implementation of torch.export() function but if anyone could give a heads up on what is happening, it would be great :slightly_smiling_face:

Ref : https://discuss.pytorch.org/t/conv-weights-changed-after-exporting-from-pytorch-pt-model-to-onnx-
model/197384