qiboteam / qiboml

Quantum Machine Learning using Qibo
Apache License 2.0
11 stars 3 forks source link

Move Pytorch backend #30

Closed Simone-Bordoni closed 1 week ago

Simone-Bordoni commented 4 months ago

Move pytorch backend from qibo to qiboml

Simone-Bordoni commented 4 months ago

We should wait for the new release of qibo in order to merge

MatteoRobbiati commented 2 months ago

We should wait for the new release of qibo in order to merge

Do you think now we can update this and merge @Simone-Bordoni? :)

MatteoRobbiati commented 2 months ago

I think now the tests should pass. The backend migrated here is updated with Qibo and a couple of errors have been fixed. Have a look whenever you can @BrunoLiegiBastonLiegi and imho we can merge it.

BrunoLiegiBastonLiegi commented 2 months ago

Honestly, I am not sure whether it is worth rushing this by friday. As I see it, there are still two relevant matters open:

MatteoRobbiati commented 2 months ago

Honestly, I am not sure whether it is worth rushing this by friday. As I see it, there are still two relevant matters open:

  • what to do with the tests? Do we leave them in qibo, thus creating the same ongoing problems we have with qibojit as well, or do we try to solve this once and for all, maybe splitting the tests in the different repos as we were discussing earlier in the meeting?
  • The PyTorchBackend is not really working for what it's more important (the gradients) Fix pytorch gradients qibo#1450, thus we would be moving something buggy anyway

I am fine in re-discussing the testing strategy and so post-poning this to next releases. Both your comments are correct and I agree. Anyway, it is useful if we merge this here, no? (without the need of rushing for the migration)

BrunoLiegiBastonLiegi commented 2 months ago

Yes we can, I mean qiboml is quite a work in progress in any case right now and even the main is not really tested, but I am not sure what this buys us. We will need to open a new PR for the PyTorchBackend in any case as soon the gradients problem gets fixed.

MatteoRobbiati commented 2 months ago

Yes we can, I mean qiboml is quite a work in progress in any case right now and even the main is not really tested, but I am not sure what this buys us. We will need to open a new PR for the PyTorchBackend in any case as soon the gradients problem gets fixed.

Ok, fine. We can keep this open until that moment.