Question about '_rebuild_tensor_v2'

1

Hi,

I have a model trained using the newest master branch of pytorch (0.4.0a0+18a76f5).
Now I want to use old version of pytorch (‘0.4.0’) to load the model.

It reports an error: ‘module’ object has no attribute ‘_rebuild_tensor_v2’

I noticed that ‘_rebuild_tensor_v2’ was added recently.
Is there any method to use ‘0.4.0’ to load a model trained by ‘0.4.0a0+18a76f5’?

bowen

2

I have the same problem, I develop and train using latest PyTorch from source, save a model, and want to deploy it on a robot which is using the stable PyTorch 0.3.1 from the repo.

It seems that this is intended behaviour, though. It is really annoying, actually.

3

Thanks to Python’s incredible flexibility, we can actually monkey-patch our way to making this work. This is very ugly, but gets the job done, add it at the top of your file, after import torch:

# Monkey-patch because I trained with a newer version.
# This can be removed once PyTorch 0.4.x is out.
# See https://discuss.pytorch.org/t/question-about-rebuild-tensor-v2/14560
import torch._utils
try:
    torch._utils._rebuild_tensor_v2
except AttributeError:
    def _rebuild_tensor_v2(storage, storage_offset, size, stride, requires_grad, backward_hooks):
        tensor = torch._utils._rebuild_tensor(storage, storage_offset, size, stride)
        tensor.requires_grad = requires_grad
        tensor._backward_hooks = backward_hooks
        return tensor
    torch._utils._rebuild_tensor_v2 = _rebuild_tensor_v2
38 Likes
4

Thanks man! This worked for me!

5

It works for me! Thanks!

6

That works for me, thanks!

7

how to use this? do you rebuild it from source? like you edit from environment or put it somewhere in the code for loading?
Sorry for newbie question here

8

just put it before loading checkpoint

9

Thank you so much!! These few lines saved my afternoon!

10

This worked for me! Thanks a lot!

11

This worked for me! Thank you very much!