Add support on QuantumInstance/run_circuits for Backends that don't handle QasmQobj #6299

manoelmarques · 2021-04-25T13:54:11Z

Summary

Until a bigger QuantumInstance/run_circuits refactor happens, this handles backends that don't accept QasmQobj.

Details and comments

@mtreinish added to Quantum Instance the error mitigation logic to also support circuits.

mtreinish

Just took a quick pass through it to start, overall lgtm just a couple of questions inline. As for the unit tests I agree they're good to have, but they won't be backportable (which is something we need to do here) because we can't backport #6286 (it's a breaking change for users of the mock backends). But, I guess it'll be simple enough to strip the tests in the backport so that's not an issue.

mtreinish · 2021-04-26T15:00:46Z

qiskit/utils/quantum_instance.py

+                not is_aer_provider(self._backend) and \
+                not is_basicaer_provider(self._backend) and \
+                not is_ibmq_provider(self._backend):


I'm not sure how I feel about this, qobj support is dated and will be going away in ~<3months for the ibmq provider; for aer and basic aer it'll be around a bit longer, but not much. They're also all emitting warnings either a DeprecationWarning or a PendingDeprecationWarning which means anytime someone runs an algorithm on up-to-date versions of these backends we'll be emitting a warning for each job. But, I get that this is a workaround just to fix the bug for now. Maybe just expand on the comment here that we can't rely on Qobj for much longer even if the backends are from ibmq, aer, or basicaer.

I agree that the ideal would be to just not use Qobj anymore however many things will change possibly including how AQGD works and the whole error mitigation logic. Excluding error mitigation, currently, if I use Aer without Qobj, I get several unit tests failures while the same pass using BasicAer without QObj.

I expect this is because when running against aer it expects to be able to hack the qobj up to pass a paramterized circuit, and the circuit interface to aer is missing equivalent functionality (it was hacked in aer as a custom path for aqua and not really exposed as real interface). We should open an issue on aer to add an alternative for passing in parameterized circuits and a list of bound values as a kwarg to run without needing qobj.

We should open an issue on aer to add an alternative for passing in parameterized circuits and a list of bound values as a kwarg to run without needing qobj.

Actually it would be nice for for such a capability no matter the backend.

Well most backends don't have a method to natively run parameterized circuits over a parameter list like aer does, but in such cases the provider could just expand the circuits before sending them to the backend.

I think the place to define this would be in the next version of the backend interface (which I have a rfc/prototype up in #5885 )

qiskit/utils/quantum_instance.py

woodsp-ibm · 2021-04-26T15:36:31Z

I think its a question here of how much rework/fixing can be easily (sensibly) done as a bug fix. Mitigation logic is all written around qobj. qobj too is used with Aer and has dedicated logic for it to allow it to be reused across different parameterizations. #5717 was written up a while back with a view to re-doing the QI. With the runtime work it indeed needs this - the above patch, should allow newer backends to run albeit without noise mitigation for now.

mtreinish · 2021-04-26T21:16:31Z

I think its a question here of how much rework/fixing can be easily (sensibly) done as a bug fix. Mitigation logic is all written around qobj. qobj too is used with Aer and has dedicated logic for it to allow it to be reused across different parameterizations. #5717 was written up a while back with a view to re-doing the QI. With the runtime work it indeed needs this - the above patch, should allow newer backends to run albeit without noise mitigation for now.

Sure I agree with this, and as I said before this is a fine compromise, except I think we should have the measurement mitigation in place, it's honestly not hard to do with circuits (it's basically identical you just skip the assemble here: https://github.com/Qiskit/qiskit-terra/blob/main/qiskit/utils/measurement_error_mitigation.py#L148 and it's basically all the same). There are 2 reasons I think this, first everyone keeps saying the builtin measurement mitigation is the reason the quantum instance exists as a standalone thing, so not supporting that seems to still be a problem, and second there is no user facing indication that measurement mitigation isn't being run when you use a BackendV1 object, even if you set it to be enabled.

woodsp-ibm · 2021-04-27T14:26:53Z

There are 2 reasons I think this, first everyone keeps saying the builtin measurement mitigation is the reason the quantum instance exists as a standalone thing, so not supporting that seems to still be a problem, and second there is no user facing indication that measurement mitigation isn't being run when you use a BackendV1 object, even if you set it to be enabled.

FYI QI existed before measurement mitigation. It was done to hold the settings etc for an algorithm running over multiple circuit executions and have a common place for such function so it was not done in each algorithm. Anyway thats somewhat irrelevant to the current situation. An external indication could simply be added to log a warning that measure mitigation is ignored, in the absence of it being supported for V1 backend for this fix.

Co-authored-by: Matthew Treinish <[email protected]>

mtreinish

Overall LGTM, I found one small issue in the run_circuits code but other than that I think it's good to go. One question (but I don't think it's worth blocking over), do we have any test coverage for the v1 interface with error mitigation? The tests added didn't seem to set it, but I wasn't sure.

qiskit/utils/run_circuits.py

Add check for providers that don't support it. Co-authored-by: Matthew Treinish <[email protected]>

…backend

mtreinish

LGTM, thanks for the quick update.

…andle QasmQobj (#6299) * Support BackendV1 * Add error mitigation logic. Co-authored-by: Matthew Treinish <[email protected]> * Fix run_circuits calls * add unit tests * Update qiskit/utils/run_circuits.py Add check for providers that don't support it. Co-authored-by: Matthew Treinish <[email protected]> * Add error mitigation test Co-authored-by: Matthew Treinish <[email protected]> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com> (cherry picked from commit a0ed7ff)

2 of the tests added in #6299 depend on the fake backends in qiskit.test.mock being strict BackendV1 backends (that reject qobj) and also having other aspects of the interface implemented (mainly run kwargs). However the change which updated the fake backends to use BackendV1 (#6286) is not backportable as it's an API change, so these tests are runnable in stable/0.17. This commit just removes these tests because even if we could get them running without #6286 we wouldn't be actually testing the v1 interface (which is their purpose).

…andle QasmQobj (backport #6299) (#6343) * Add support on QuantumInstance/run_circuits for Backends that don't handle QasmQobj (#6299) * Support BackendV1 * Add error mitigation logic. Co-authored-by: Matthew Treinish <[email protected]> * Fix run_circuits calls * add unit tests * Update qiskit/utils/run_circuits.py Add check for providers that don't support it. Co-authored-by: Matthew Treinish <[email protected]> * Add error mitigation test Co-authored-by: Matthew Treinish <[email protected]> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com> (cherry picked from commit a0ed7ff) * Remove tests that depend on BackendV1 fake backends 2 of the tests added in #6299 depend on the fake backends in qiskit.test.mock being strict BackendV1 backends (that reject qobj) and also having other aspects of the interface implemented (mainly run kwargs). However the change which updated the fake backends to use BackendV1 (#6286) is not backportable as it's an API change, so these tests are runnable in stable/0.17. This commit just removes these tests because even if we could get them running without #6286 we wouldn't be actually testing the v1 interface (which is their purpose). * Fix lint Co-authored-by: Manoel Marques <[email protected]> Co-authored-by: Matthew Treinish <[email protected]> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>

In Qiskit#6299 support was fixed for strict BackendV1 backends that only take QuantumCircuit objects (instead of qobj) for the input. That was fixed by adding a parallel path when running with a new backend. However that parallel path wasn't identical and was missing the support the qobj path had for splitting an algorithm run into multiple jobs if the backend if the number of circuits was greater than the max_experiments set in the backend. This would result on some providers' backends, such as ionq and aqt, both of which have max_experiments set to 1. This commit fixes this issue by splitting the circuits list into smaller sublists when the len(circuits) > max_experiments (or the old env var, which we should change the name of at some point).

* Respect max_experiments in QuantumInstance BackendV1 path In #6299 support was fixed for strict BackendV1 backends that only take QuantumCircuit objects (instead of qobj) for the input. That was fixed by adding a parallel path when running with a new backend. However that parallel path wasn't identical and was missing the support the qobj path had for splitting an algorithm run into multiple jobs if the backend if the number of circuits was greater than the max_experiments set in the backend. This would result on some providers' backends, such as ionq and aqt, both of which have max_experiments set to 1. This commit fixes this issue by splitting the circuits list into smaller sublists when the len(circuits) > max_experiments (or the old env var, which we should change the name of at some point). * Fix issues with results and split circuits path * Fix copy paste issue * Update qiskit/utils/run_circuits.py * Add release notes * Fix whitespace Co-authored-by: Kevin Krsulich <[email protected]> Co-authored-by: Manoel Marques <[email protected]>

* Respect max_experiments in QuantumInstance BackendV1 path In #6299 support was fixed for strict BackendV1 backends that only take QuantumCircuit objects (instead of qobj) for the input. That was fixed by adding a parallel path when running with a new backend. However that parallel path wasn't identical and was missing the support the qobj path had for splitting an algorithm run into multiple jobs if the backend if the number of circuits was greater than the max_experiments set in the backend. This would result on some providers' backends, such as ionq and aqt, both of which have max_experiments set to 1. This commit fixes this issue by splitting the circuits list into smaller sublists when the len(circuits) > max_experiments (or the old env var, which we should change the name of at some point). * Fix issues with results and split circuits path * Fix copy paste issue * Update qiskit/utils/run_circuits.py * Add release notes * Fix whitespace Co-authored-by: Kevin Krsulich <[email protected]> Co-authored-by: Manoel Marques <[email protected]> (cherry picked from commit 9d4bb91) # Conflicts: # qiskit/utils/run_circuits.py # test/python/algorithms/test_backendv1.py

…6391) (#6414) * Respect max_experiments in QuantumInstance BackendV1 path (#6391) * Respect max_experiments in QuantumInstance BackendV1 path In #6299 support was fixed for strict BackendV1 backends that only take QuantumCircuit objects (instead of qobj) for the input. That was fixed by adding a parallel path when running with a new backend. However that parallel path wasn't identical and was missing the support the qobj path had for splitting an algorithm run into multiple jobs if the backend if the number of circuits was greater than the max_experiments set in the backend. This would result on some providers' backends, such as ionq and aqt, both of which have max_experiments set to 1. This commit fixes this issue by splitting the circuits list into smaller sublists when the len(circuits) > max_experiments (or the old env var, which we should change the name of at some point). * Fix issues with results and split circuits path * Fix copy paste issue * Update qiskit/utils/run_circuits.py * Add release notes * Fix whitespace Co-authored-by: Kevin Krsulich <[email protected]> Co-authored-by: Manoel Marques <[email protected]> (cherry picked from commit 9d4bb91) # Conflicts: # qiskit/utils/run_circuits.py # test/python/algorithms/test_backendv1.py * Fix merge conflicts This commit fixes the merge conflicts from backporting #6391 to stable/0.17. The tests added in #6391 are just removed because they can't actually run without #6286 which isn't backportable. * Fix lint Co-authored-by: Matthew Treinish <[email protected]>

…andle QasmQobj (Qiskit#6299) * Support BackendV1 * Add error mitigation logic. Co-authored-by: Matthew Treinish <[email protected]> * Fix run_circuits calls * add unit tests * Update qiskit/utils/run_circuits.py Add check for providers that don't support it. Co-authored-by: Matthew Treinish <[email protected]> * Add error mitigation test Co-authored-by: Matthew Treinish <[email protected]> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>

* Respect max_experiments in QuantumInstance BackendV1 path In Qiskit#6299 support was fixed for strict BackendV1 backends that only take QuantumCircuit objects (instead of qobj) for the input. That was fixed by adding a parallel path when running with a new backend. However that parallel path wasn't identical and was missing the support the qobj path had for splitting an algorithm run into multiple jobs if the backend if the number of circuits was greater than the max_experiments set in the backend. This would result on some providers' backends, such as ionq and aqt, both of which have max_experiments set to 1. This commit fixes this issue by splitting the circuits list into smaller sublists when the len(circuits) > max_experiments (or the old env var, which we should change the name of at some point). * Fix issues with results and split circuits path * Fix copy paste issue * Update qiskit/utils/run_circuits.py * Add release notes * Fix whitespace Co-authored-by: Kevin Krsulich <[email protected]> Co-authored-by: Manoel Marques <[email protected]>

…andle QasmQobj (Qiskit/qiskit#6299) * Support BackendV1 * Add error mitigation logic. Co-authored-by: Matthew Treinish <[email protected]> * Fix run_circuits calls * add unit tests * Update qiskit/utils/run_circuits.py Add check for providers that don't support it. Co-authored-by: Matthew Treinish <[email protected]> * Add error mitigation test Co-authored-by: Matthew Treinish <[email protected]> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>

…kit#6391) * Respect max_experiments in QuantumInstance BackendV1 path In Qiskit/qiskit#6299 support was fixed for strict BackendV1 backends that only take QuantumCircuit objects (instead of qobj) for the input. That was fixed by adding a parallel path when running with a new backend. However that parallel path wasn't identical and was missing the support the qobj path had for splitting an algorithm run into multiple jobs if the backend if the number of circuits was greater than the max_experiments set in the backend. This would result on some providers' backends, such as ionq and aqt, both of which have max_experiments set to 1. This commit fixes this issue by splitting the circuits list into smaller sublists when the len(circuits) > max_experiments (or the old env var, which we should change the name of at some point). * Fix issues with results and split circuits path * Fix copy paste issue * Update qiskit/utils/run_circuits.py * Add release notes * Fix whitespace Co-authored-by: Kevin Krsulich <[email protected]> Co-authored-by: Manoel Marques <[email protected]>

…andle QasmQobj (Qiskit/qiskit#6299) * Support BackendV1 * Add error mitigation logic. Co-authored-by: Matthew Treinish <[email protected]> * Fix run_circuits calls * add unit tests * Update qiskit/utils/run_circuits.py Add check for providers that don't support it. Co-authored-by: Matthew Treinish <[email protected]> * Add error mitigation test Co-authored-by: Matthew Treinish <[email protected]> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>

…kit#6391) * Respect max_experiments in QuantumInstance BackendV1 path In Qiskit/qiskit#6299 support was fixed for strict BackendV1 backends that only take QuantumCircuit objects (instead of qobj) for the input. That was fixed by adding a parallel path when running with a new backend. However that parallel path wasn't identical and was missing the support the qobj path had for splitting an algorithm run into multiple jobs if the backend if the number of circuits was greater than the max_experiments set in the backend. This would result on some providers' backends, such as ionq and aqt, both of which have max_experiments set to 1. This commit fixes this issue by splitting the circuits list into smaller sublists when the len(circuits) > max_experiments (or the old env var, which we should change the name of at some point). * Fix issues with results and split circuits path * Fix copy paste issue * Update qiskit/utils/run_circuits.py * Add release notes * Fix whitespace Co-authored-by: Kevin Krsulich <[email protected]> Co-authored-by: Manoel Marques <[email protected]>

manoelmarques added the bug Something isn't working label Apr 25, 2021

manoelmarques requested review from mtreinish, woodsp-ibm and stefan-woerner April 25, 2021 13:54

manoelmarques requested a review from a team as a code owner April 25, 2021 13:54

manoelmarques self-assigned this Apr 25, 2021

mtreinish added the stable backport potential The bug might be minimal and/or import enough to be port to stable label Apr 26, 2021

mtreinish reviewed Apr 26, 2021

View reviewed changes

woodsp-ibm mentioned this pull request Apr 26, 2021

AQGD optimizer groups objective funtion calls by default #6305

Closed

manoelmarques force-pushed the backend branch 5 times, most recently from ac22589 to 4312086 Compare April 29, 2021 13:34

mtreinish added the Changelog: Bugfix Include in the "Fixed" section of the changelog label Apr 29, 2021

manoelmarques force-pushed the backend branch 4 times, most recently from 36e4dcc to 5964b72 Compare April 29, 2021 22:16

manoelmarques added 2 commits April 30, 2021 08:47

Support BackendV1

ff92faf

Add error mitigation logic.

6ec1700

Co-authored-by: Matthew Treinish <[email protected]>

manoelmarques force-pushed the backend branch from 5964b72 to 6ec1700 Compare April 30, 2021 12:47

manoelmarques added 2 commits April 30, 2021 12:35

Fix run_circuits calls

b124398

add unit tests

bf5c7db

manoelmarques changed the title ~~[WIP] Add support on QuantumInstance/run_circuits for Backends that don't handle QasmQobj~~ Add support on QuantumInstance/run_circuits for Backends that don't handle QasmQobj Apr 30, 2021

manoelmarques requested a review from mtreinish April 30, 2021 18:02

manoelmarques removed the bug Something isn't working label Apr 30, 2021

mtreinish reviewed Apr 30, 2021

View reviewed changes

qiskit/utils/run_circuits.py Outdated Show resolved Hide resolved

manoelmarques and others added 3 commits April 30, 2021 16:55

Update qiskit/utils/run_circuits.py

19a2b74

Add check for providers that don't support it. Co-authored-by: Matthew Treinish <[email protected]>

Add error mitigation test

09a8c84

Merge branch 'backend' of github.com:manoelmarques/qiskit-terra into …

cdd9c3c

…backend

manoelmarques requested a review from mtreinish April 30, 2021 21:04

mtreinish approved these changes May 3, 2021

View reviewed changes

mtreinish added the automerge label May 3, 2021

Merge branch 'main' into backend

de625ed

mergify bot merged commit a0ed7ff into Qiskit:main May 3, 2021

manoelmarques deleted the backend branch May 3, 2021 17:12

mergify bot mentioned this pull request May 3, 2021

Add support on QuantumInstance/run_circuits for Backends that don't handle QasmQobj (backport #6299) #6343

Merged

mtreinish mentioned this pull request May 10, 2021

Respect max_experiments in QuantumInstance BackendV1 path #6391

Merged

ElePT mentioned this pull request Aug 22, 2023

[from qiskit] AQGD optimizer groups objective funtion calls by default qiskit-community/qiskit-algorithms#50

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add support on QuantumInstance/run_circuits for Backends that don't handle QasmQobj #6299

Add support on QuantumInstance/run_circuits for Backends that don't handle QasmQobj #6299

manoelmarques commented Apr 25, 2021 •

edited

Loading

mtreinish left a comment

mtreinish Apr 26, 2021

manoelmarques Apr 26, 2021

mtreinish Apr 26, 2021

woodsp-ibm Apr 27, 2021

mtreinish Apr 28, 2021

woodsp-ibm commented Apr 26, 2021

mtreinish commented Apr 26, 2021

woodsp-ibm commented Apr 27, 2021

mtreinish left a comment

mtreinish left a comment

Add support on QuantumInstance/run_circuits for Backends that don't handle QasmQobj #6299

Add support on QuantumInstance/run_circuits for Backends that don't handle QasmQobj #6299

Conversation

manoelmarques commented Apr 25, 2021 • edited Loading

Summary

Details and comments

mtreinish left a comment

Choose a reason for hiding this comment

mtreinish Apr 26, 2021

Choose a reason for hiding this comment

manoelmarques Apr 26, 2021

Choose a reason for hiding this comment

mtreinish Apr 26, 2021

Choose a reason for hiding this comment

woodsp-ibm Apr 27, 2021

Choose a reason for hiding this comment

mtreinish Apr 28, 2021

Choose a reason for hiding this comment

woodsp-ibm commented Apr 26, 2021

mtreinish commented Apr 26, 2021

woodsp-ibm commented Apr 27, 2021

mtreinish left a comment

Choose a reason for hiding this comment

mtreinish left a comment

Choose a reason for hiding this comment

manoelmarques commented Apr 25, 2021 •

edited

Loading