Selfhosted GPU tests restricted to GPU backends only

[BrunoLiegiBastonLiegi]

Same as qiboteam/qibojit#206. A --gpu-only option is added to pytest to run tests on GPU backends only.

Checklist:

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

[github-actions]

Run on QPU sim completed!

You can download the coverage report as an artifact, from the workflow summary page:
https://github.com/qiboteam/qibo/actions/runs/13199560283

[BrunoLiegiBastonLiegi]

Ok now the gpu tests take ~6 minutes to complete. Should we include even the pytorch and tensorflow (I don't think we have gpu support for jax yet) gpu tests?

[andrea-pasquale]

Ok now the gpu tests take ~6 minutes to complete. Should we include even the pytorch and tensorflow (I don't think we have gpu support for jax yet) gpu tests?

If you want feel free to do so, however if we see that tests are not passing for PyTorch and tensorflow we can address them in another PR.

[github-actions]

Run on QPU sim completed!

You can download the coverage report as an artifact, from the workflow summary page:
https://github.com/qiboteam/qibo/actions/runs/13199807444

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.61%. Comparing base (5108c14) to head (d10aac8).
Report is 6 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #1579   +/-   ##
=======================================
  Coverage   99.61%   99.61%           
=======================================
  Files          76       76           
  Lines       11449    11449           
=======================================
  Hits        11405    11405           
  Misses         44       44           

Flag	Coverage Δ
unittests	99.61% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[BrunoLiegiBastonLiegi]

The thing is that right now I am not even sure how well is integrated the use of a device in the torch and tf backend. For example:

from qibo import set_backend
from qibo.models import QFT

set_backend(backend="qiboml", platform="pytorch")
c = QFT(5)
r = c()
r.probabilities().device
# device(type='cpu')
r.state().device
# device(type='cpu')

and the same happens for tensorflow.
Hence it's not worth it enabling them now as they would probably just run on CPU anyway. A revision of the device attribute has to be done for both of them first.

[andrea-pasquale]

The thing is that right now I am not even sure how well is integrated the use of a device in the torch and tf backend. For example:

from qibo import set_backend
from qibo.models import QFT

set_backend(backend="qiboml", platform="pytorch")
c = QFT(5)
r = c()
r.probabilities().device
# device(type='cpu')
r.state().device
# device(type='cpu')

and the same happens for tensorflow. Hence it's not worth it enabling them now as they would probably just run on CPU anyway. A revision of the device attribute has to be done for both of them first.

At this point I agree we can skip it, just open an issue so that we don't forget.