I have converted a model to scipt module using torch.jit.scipt
. However, I find there is some difference between the output of the original model and the output of the scripted model like:
model.eval()
scripted_model = torch.jit.script(model)
# there is some difference between output and scripted_output
output = model(inputs)
scripted_output = scripted_model(inputs)
Each element has an average difference of 1e-6
Is this phenomenon normal?