Replace convert_to_target with the Qiskit version

[t-imamichi]

Summary

This PR copies qiskit.providers.backend_compat.convert_to_target to utils.backend_converter.convert_to_target with some modifications for mypy and lint. Particularly, I added reset to required to ensure reset is included in target.

Fixes #1590

Details and comments

The script #1590 works fine as follows

from qiskit_ibm_runtime import QiskitRuntimeService

service = QiskitRuntimeService(channel="ibm_quantum")
backend = service.get_backend("ibm_sherbrooke")
print('ibm_sherbrooke')
print(type(backend.target["measure"][(0,)]._calibration))  # qiskit.pulse.calibration_entries.ScheduleDef
print(len(backend.target["measure"][(0,)].calibration.instructions))  # 381 -- this is not correct
print(len(backend.target["measure"][(1,)].calibration.instructions))  # 129
print()

from qiskit_ibm_runtime.fake_provider import FakeSherbrooke

backend = FakeSherbrooke()
print('fake_sherbrooke')
print(type(backend.target["measure"][(0,)]._calibration))  # qiskit.pulse.calibration_entries.PulseQobjDef
print(len(backend.target["measure"][(0,)].calibration.instructions))  # 129
print(len(backend.target["measure"][(1,)].calibration.instructions))  # 129

main

ibm_sherbrooke
<class 'qiskit.pulse.calibration_entries.ScheduleDef'>
381
129

fake_sherbrooke
<class 'qiskit.pulse.calibration_entries.PulseQobjDef'>
129
129

this PR

ibm_sherbrooke
<class 'qiskit.pulse.calibration_entries.PulseQobjDef'>
129
129

fake_sherbrooke
<class 'qiskit.pulse.calibration_entries.PulseQobjDef'>
129
129

[coveralls]

Pull Request Test Coverage Report for Build 8974814830

Details

• 98 of 117 (83.76%) changed or added relevant lines in 3 files are covered.
• 19 unchanged lines in 2 files lost coverage.
• Overall coverage decreased (-0.2%) to 83.48%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
qiskit_ibm_runtime/utils/backend_converter.py	95	114	83.33%

Files with Coverage Reduction	New Missed Lines	%
qiskit_ibm_runtime/ibm_qubit_properties.py	9	0.0%
qiskit_ibm_runtime/utils/backend_converter.py	10	77.04%

Totals
Change from base Build 8974407177:	-0.2%
Covered Lines:	6276
Relevant Lines:	7518

---

💛 - Coveralls

[jyu00]

@t-imamichi we recently had an issue that the convert_to_target from Qiskit wasn't picking up reset:

from qiskit_ibm_runtime.utils.backend_converter import convert_to_target as qr_c2t
from qiskit.providers.backend_compat import convert_to_target as qk_c2t

qr_target = qr_c2t(backend.configuration(), backend.properties(), backend.defaults())
qk_target = qk_c2t(backend.configuration(), backend.properties(), backend.defaults())

# then qr_target.operation_names gives:
dict_keys(['id', 'rz', 'sx', 'x', 'cx', 'reset', 'measure', 'switch_case', 'if_else', 'for_loop', 'delay'])

# but qk_target.operation_names gives:
dict_keys(['id', 'rz', 'switch_case', 'if_else', 'cx', 'x', 'measure', 'for_loop', 'delay', 'sx'])

Presumably there was a reason for qiskit-ibm-runtime to have its own convert_to_target, so I'd also consult with @mtreinish before removing it.

[mtreinish]

We can look at fixing the issue in Qiskit's converter function; is there an issue open about it already? I assume reset is being put in a field in the configuration that's not being looked at for the reset instruction by the qiskit version, or there are conflicting fields and qiskit is looking at one that's missing the reset instruction.

Besides that the other consideration is that we're planning to deprecate the BackendV1 interface in 1.2.x which will include removing the BackendConfiguration, BackendProperties, PulseDefaults, and by extension this will include dropping convert_to_target because there is nothing to convert from. The removal won't be until Qiskit 2.0, but it might be better to just leave this in place since we'll need to add it back before too long.

[t-imamichi]

I copied Qiskit convert_to_target to qiskit-ibm-runtime with some modifications.
Particularly, I added reset to required to ensure reset is included in target.

qiskit-ibm-runtime/qiskit_ibm_runtime/utils/backend_converter.py

Line 63 in 50bfff0

required = ["measure", "delay", "reset"]

[jyu00]

We can look at fixing the issue in Qiskit's converter function; is there an issue open about it already?

@mtreinish I'm guessing no if you're not aware. Somehow I thought you were added to the thread where this was discussed. Is it worth opening an issue in Qiskit at this point if it's going to be deprecated in 1.2?

Besides that the other consideration is that we're planning to deprecate the BackendV1 interface in 1.2.x which will include removing the BackendConfiguration, BackendProperties, PulseDefaults

Sounds like we'll need to copy those over to qiskit-ibm-runtime, since that's still how we get data from the server.

[jyu00]

I copied Qiskit convert_to_target to qiskit-ibm-runtime with some modifications.

@t-imamichi can you also add more tests? Like maybe copy the ones on faulty qubits from Qiskit (we already have one on control flow). They can just be unit tests using fake backends.

[mtreinish]

We can look at fixing the issue in Qiskit's converter function; is there an issue open about it already?

@mtreinish I'm guessing no if you're not aware. Somehow I thought you were added to the thread where this was discussed. Is it worth opening an issue in Qiskit at this point if it's going to be deprecated in 1.2?

I vaguely remember the issue, but I don't remember the specifics. But yeah we still support the converter function for > 1 year so it's good to ensure we get this fixed.

Besides that the other consideration is that we're planning to deprecate the BackendV1 interface in 1.2.x which will include removing the BackendConfiguration, BackendProperties, PulseDefaults

Sounds like we'll need to copy those over to qiskit-ibm-runtime, since that's still how we get data from the server.

Yeah, I would think so. There's no rush until ~August 2025 (that's when the 1.x series goes EoL) but planning for that and taking ownership of the container objects for the api payloads in the medium term is probably prudent.

[t-imamichi]

I copied unit tests on faulty qubits from https://github.com/Qiskit/qiskit/blob/main/test/python/providers/test_faulty_backend.py

[t-imamichi]

I noticed some error occurs due to random seed issue. Need to merge #1616

[t-imamichi]

Puala suggested replacing convert_to_target for fake backends too t-imamichi#2. Do you anyone have any suggestion or comment? If no, I will merge it.

[kt474]

Puala suggested replacing convert_to_target for fake backends too t-imamichi#2. Do you anyone have any suggestion or comment? If no, I will merge it.

No objections, after that, this PR LGTM

[wshanks]

This is labeled as new feature. Can it not be backported to 0.23.1 if there will be one? Besides fixing #1570 and #1590, it also fixes another issue I am running into (where a backend defines pulse calibrations for a qubit pair that does not have an entry in the gate properties; the old code in 0.23.0 raises a KeyError while the version here skips the missing keys without an exception).