jit - What is the difference between a pytorch model and its executable version?

Question

I have converted my pytorch model and converted into an executable by using torch.jit.trace. I noticed that for same input, the intermediate outputs of the model and executable are different. Furthermore, I created two different executable models, one with single return value, and the other with multiple return values. Surprisingly, intermediate outputs are different for both executables too. What can be the reason behind this behavior?

question from:https://stackoverflow.com/questions/65880740/what-is-the-difference-between-a-pytorch-model-and-its-executable-version

Answer 1

Waitting for answers