#3312 | Large Pulse backend mock

[IceKhan13]

Summary

Create a large (100Q) Pulse backend mock and/or create a util method or a script to generate reasonable Pulse-enabled backend

Issue #3312

Details and comments

The pulse module of Qiskit has yet to be benchmarked. The purpose of benchmarking is to provide a deterministic way to assess performance of software over time. There is room for improvement in the performance of pulse operations; building benchmarks will allow us to quantify progress and justify code refactors.

[lcapelluto]

It's exciting to see this getting started! Thank you @IceKhan13 :)

[taalexander]

This is coming along nicely @IceKhan13. I think you are on the right track with the FakeBackendBuilder. What I would like to see is an arbitrary FakeBackendBuilder that you can pass a number of qubits too, a grid size and potentially just examples for single-qubit properties and frequencies. We could then provide a FakeOpenPulse100Q or FakeOpenPulse1000Q which upon calling would generate the fake backend. This would remove the need for storing configuration and properties json.

For example

def fake_backend_generator(n_qubits, name, coupling_map=None, version='1.1', pulse=True, basis_gates=['id', 'u1', 'u2', 'u3', 'cx'], 1q_error_rate=0.001, 2q_error_rate=0.01, T1=100, T2=200) -> FakeBackend:
    ...

and then we would have

FakeOpenPulse100Q = fake_backend_generator(100, ...)
FakeOpenPulse1000Q = fake_backend_generator(1000, ...)

Note the backend generation shouldn't actually happen until the mock backend is instantiated, ie., calling FakeOpenPulse100Q() should cause the backend to be generated. These generated backends could then be tested by the tests described here #3892.

[taalexander]

This is coming along nicely, a couple of comments here and then we should figure out how to test this.

[IceKhan13]

Team, I need some input from you 😄
I'm trying to write tests for generated by builder backends (create circuit and run it on the backend.)

It goes as usual:

desired_vector = [1 / math.sqrt(2), 0, 0, 1 / math.sqrt(2)]
qr = QuantumRegister(2, "qr")
qc = QuantumCircuit(qr)
qc.initialize(desired_vector, [qr[0], qr[1]])

then I build a backend and try to transpile cirquit, create schedule and assemble qobj:

FakeOpenPulse10Q = FakeBackendBuilder("Tashkent", n_qubits=10).build()
backend = FakeOpenPulse10Q()

experiments = transpile(qc, backend=backend)
experiments = schedule(circuits=experiments, backend=backend)
qobj = assemble(experiments, backend=backend)

and everything works but to finish logical sequence for execute I need to run backend.run(qobj).
And here execution fails due to pickle error:

_pickle.PicklingError: Can't pickle <class 'abc.FakeOpenPulse10Q'>: attribute lookup FakeOpenPulse10Q on abc failed

Same thing happens when I try to use FakeOpenPulse2Q (existing) backend. But in case of existing backend it failed due to inaability to pickle local object:

AttributeError: Can't pickle local object 'QobjToInstructionConverter.convert_frame_change.<locals>.gen_fc_sched'

This can be solved (I assume) by using __setstate__ and __getstate__ to manually pickle-unpickle objects.

Anyhow both of this errors occurs in run function, where in FakeBackend we are using FakeJob which uses futures.ProcessPoolExecutor(). And after 15 seconds of googling I noticed that dynamic classes has issue with pickling (which is happening while using process pools).

I think, I can create other FakeJob which will not use pools and use it in generated fake backends run method. 🤔 Maybe it's 2am and I need to sleep a little to came up with better solution 😄

Any suggestions how to overcome this?)

P.S
Is is ok that I always mention your names in PR discussions @taalexander @lcapelluto ? It probably creates a lot of notifications in your accounts. Sorry for that. I can just leave a comments and wait 😄

[taalexander]

It's totally ok to ping me 😄. I was tried to run the sample code

import math

from qiskit import QuantumRegister, QuantumCircuit, transpile, schedule, assemble
from qiskit.test.mock.utils.fake_backend_builder import FakeBackendBuilder

desired_vector = [1 / math.sqrt(2), 0, 0, 1 / math.sqrt(2)]
qr = QuantumRegister(2, "qr")
qc = QuantumCircuit(qr)
qc.initialize(desired_vector, [qr[0], qr[1]])

FakeOpenPulse10Q = FakeBackendBuilder("Tashkent", n_qubits=10).build()
backend = FakeOpenPulse10Q()
experiments = transpile(qc, backend=backend)
experiments = schedule(circuits=experiments, backend=backend)
qobj = assemble(experiments, backend=backend)

but am getting the error

TranspilerError: "The 'layout' must be full (with ancilla)."

If I look at the coupling_map, it includes too many qubits.

>>> backend.configuration().coupling_map
[[0, 4], [1, 2], [2, 3], [2, 6], [3, 4], [4, 8], [5, 6], [6, 7], [6, 10], [7, 8], [8, 12], [9, 10]]

[taalexander]

The backend is missing a parameter rep_times I think.

[taalexander]

I can, however, recreate your error with

import math

from qiskit import QuantumRegister, QuantumCircuit, transpile, schedule, assemble
from qiskit.test.mock.utils.fake_backend_builder import FakeBackendBuilder
from qiskit.test.mock import FakeOpenPulse2Q

desired_vector = [1 / math.sqrt(2), 0, 0, 1 / math.sqrt(2)]
qr = QuantumRegister(2, "qr")
qc = QuantumCircuit(qr)
qc.initialize(desired_vector, [qr[0], qr[1]])

FakeOpenPulse10Q = FakeBackendBuilder("Tashkent", n_qubits=4, coupling_map=[[0, 1], [1,2], [2,3]]).build()
#backend = FakeOpenPulse2Q()
backend = FakeOpenPulse10Q()
experiments = transpile(qc, backend=backend)
experiments = schedule(circuits=experiments, backend=backend)
qobj = assemble(experiments, backend=backend, rep_time=100)

job = backend.run(qobj)
result = job.result()

Please try again after updating to master of qiskit which has changed this method.

[taalexander]

Did merging in master help?

[IceKhan13]

Fake job changed. So, no it's using Aer simulator and reads hamiltonian. Now I need to generate Hamiltonian correctly and continue with tests. But all of your comments above correct and it helped :)
I'll get back and ask you again if I'll not be able to run it.

[mtreinish]

LGTM, thanks for sticking with this and pushing it through to completion

[mtreinish]

There is partial overlap here with: https://github.com/Qiskit/qiskit-terra/blob/master/test/python/providers/test_fake_backends.py#L51 but this is different enough that it's probably fine. It might be a good follow up to expand this or the other to cover both use cases and deduplicate things a bit.

[jyu00]

@IceKhan13 Can the name be changed to ConfigurableFakeBackend? We may have real configurable backends in the future.

[MGLetsoalo]

Hello, we had a similar project to this one titled: open pulse mock backend for spin based qubits from qiskit camp Africa 2019, the mock qubits where associated with the 13-C nuclear spins coupled to NV centres in a diamond. Based on the computational ability of carbon atoms, it was my believe that these carbon atoms could be used as fake backends that will participate as reactive groups in drug design to combat the covid19 pandemic. If we treat the electron spin as a quantum bus, we can adjust the states that correspond to the syndrome measurements and thus the 13-C nuclear spins provide a platform to design the proper chemistry that will help mitigate the syndrome. There is a lot of research on the development of viral nanoparticles as a template for organising materials at the nano scales and I think that it is time to consolidate these efforts.

[taalexander]

I think it is fine to still have an example backend for an NV center system. Would it be possible to use @IceKhan13's work for this and choose the right parameters? Maybe some slight modifications would be required?

[MGLetsoalo]

Yes, with 100 qubits we can realise a 100*100 tabular qubit ordering, how do I become part of this discussion so that I may be able to present my attempts?

[taalexander]

If you open up a draft PR with your attempts we can comment directly on it 😄