Add run_options option and support for processing level 1 data to BackendSamplerV2

[wshanks]

This change adds support to the BackendSamplerV2 class so that it will accept a run_options dict option and pass its content through to the underlying BackendV2's run() method.

Additionally, to support meas_level=1, the results processing code checks for a meas_level option inside of run_options and handles meas_level=1 data appropriately when it is set, rather than always assuming the returned data is level 2.

Finally, the behavior of BackendSamplerV2 was changed so that it does clear the QuantumCircuit.metadata of its input circuits before passing them on to backend.run like it did previously.

[qiskit-bot]

One or more of the following people are relevant to this code:

• @Qiskit/terra-core
• @ajavadia
• @levbishop
• @t-imamichi

[wshanks]

These changes were needed to get the qiskit-experiments tests to pass in this PR that moves all circuit execution from Backend.run to SamplerV2.run, making use of BackendSamplerV2 as the translation layer. We can add a release note and a test if needed but I wanted to see if these changes would be acceptable or not first. There are a couple other related issues/questions I had:

• Right now, BackendSamplerV2 accepts an options dict and unpacks it into its Options class. These options are then passed through to Backend.run (except shots that is passed through only if it not overridden). This mechanism means that support for any additional arguments also requires changing the Options class again. Could all the options just be passed through besides shots?

• Secondary request to the previous one -- I can see why keeping to the Options format is more standard so it might not be acceptable to accept and pass through all options. If so, could we change the __init__ to filter the keys in options and only pass the ones through to Options that have defined fields and warn about the ones that do not? This would make it a little easier to add new options in the future without breaking support for previous versions of Qiskit since an option not in the older Options definition will cause Options(**options) to fail and Options being a non-public class it is not supported to peak at its fields and filter before passing the options (edit: one could create a dummy BackendSamplerV2 and check its options fields and use them to filter the options for the real BackendSamplerV2 without accessing a private class).

• Caveat to the previous points -- maybe we won't need to add support for any more backend run options.

• For the qiskit-experiments tests, I also needed to delete this line:

  qiskit/qiskit/primitives/backend_estimator.py

  Line 63 in 9a1d8d3

  circ.metadata = {}

  Is there any way to work around that issue? Can we revisit if the metadata still needs to be blanked out like that or could we pass through some kind of keep_metadata option? The qiskit-experiments test backends use the metadata to determine what results to return for the circuits. We can look into how to refactor the experiments tests to get the metadata a different way if not.

[coveralls]

Pull Request Test Coverage Report for Build 11599154700

Warning: This coverage report may be inaccurate.

This pull request's base commit is no longer the HEAD commit of its target branch. This means it includes changes from outside the original pull request, including, potentially, unrelated coverage changes.

• For more information on this, see Tracking coverage changes with pull request builds.
• To avoid this issue with future PRs, see these Recommended CI Configurations.
• For a quick fix, rebase this PR at GitHub. Your next report should be accurate.

Details

• 24 of 25 (96.0%) changed or added relevant lines in 2 files are covered.
• 54 unchanged lines in 5 files lost coverage.
• Overall coverage increased (+0.07%) to 88.758%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
qiskit/primitives/backend_sampler_v2.py	22	23	95.65%

Files with Coverage Reduction	New Missed Lines	%
crates/qasm2/src/lex.rs	5	92.23%
qiskit/result/mitigation/utils.py	8	89.74%
qiskit/result/mitigation/correlated_readout_mitigator.py	10	89.69%
qiskit/result/mitigation/local_readout_mitigator.py	13	90.15%
crates/accelerate/src/commutation_checker.rs	18	97.09%

Totals
Change from base Build 11597322858:	0.07%
Covered Lines:	75643
Relevant Lines:	85224

---

💛 - Coveralls

[wshanks]

Qiskit/qiskit-ibm-runtime#1990 tries to make use of this change to translate between the IBM Quantum SamplerV2 options and common BackendV2.run options.

[t-imamichi]

Is it possible to add some tests using GenericBackendV2 or something? It would be nice to have some tests at least.

[wshanks]

Is it possible to add some tests using GenericBackendV2 or something? It would be nice to have some tests at least.

There is not much support built in for testing with meas_level=1 results. I am investigating what is feasible to add.

[wshanks]

I added a simple test and also addressed the other points I had commented on in the way that works best for my case, but we can revisit them if those changes are a problem.

[raynelfss]

Is this still on track for 1.3? If so, who is interested in reviewing it? @Qiskit/qiskit-primitives

[wshanks]

On my end, I think it is ready for review, so it would be great if it could still make 1.3.

[t-imamichi]

Thank you for adding tests. It looks good overall. You don't have to add a test with a real device, but I would like to confirm that it works fine with real devices. Have you tested it yet?

[wshanks]

Thank you for adding tests. It looks good overall. You don't have to add a test with a real device, but I would like to confirm that it works fine with real devices. Have you tested it yet?

My main interest is in using BackendSamplerV2 with test backends in qiskit-experiments because we haven't had time to convert all the tests to use test samplers (maybe just using BackendSamplerV2 with test backends is good enough) because support for running with backend.run on a real device is going away very soon. I was still able to do it though. Here is a quick test using the readout discriminator experiment from qiskit-experiments:

[t-imamichi]

Thank you for sharing a result with a real device.

[t-imamichi]

LGTM