pytorch backend

[Simone-Bordoni]

Backend based on pytorch for optimised automatic differentiation.

Checklist:

• Reviewers confirm new code works as expected.
• Tests are passing.
• Coverage does not decrease.
• Documentation is updated.

[Simone-Bordoni]

I need a bit of help to finish this work.
For now i have created the pytorch backend starting from the tf backend and basically replacing all tf functions/tensors with torch functions/tensors.
Where can I add the tests of the code?
What else do i have to change to make everything work?
Where can i write the documentation of the backend?
I get a pylint error when using torch.linalg.eigvalsh, torch.linalg.eigh, torch.linalg.matrix_exp, torch.sparse.mm (not-callable error) for now i have disabled this error on pylint, however i can't find a similar error and I don't know if there is a way to solve it without disabling the pylint error.

[scarrazza]

Thanks @Simone-Bordoni for this. Could you please include pytorch in tests by updating this file and check coverage?

[renatomello]

I converted this PR to draft just because it's gonna take a little bit to finish it

[Simone-Bordoni]

I converted this PR to draft just because it's gonna take a little bit to finish it

I thought it would have been an easy thing but there are many parts of qibo that are not backend agnostic and struggle with pytorch.

[Simone-Bordoni]

It doesn’t change every time. It’s always the same dict on CI, but it’s a different duct when the test is run locally

Ok, now it makes sense. sorry if i changed it again, i will not touch it anymore

[Simone-Bordoni]

I have reverted the changes to the test with seed so now everything should pass. Thanks to @BrunoLiegiBastonLiegi I also managed to remove some more unuseful methods in the pytorch backend.

[BrunoLiegiBastonLiegi]

Thanks @Simone-Bordoni, please address these last minor things and open an issue about adding support to gradients in the pytorch backend. Then we can merge this I guess.