メインコンテンツまでスキップ

Sampling on Qiskit's real quantum computers

Here we introduce the Qiskit backends. For IBM devices, we provide 2 sampling backends:

  • QiskitSamplingBackend : For the IBMQ backends
  • QiskitRuntimeSamplingBackend: For the devices provided by the QiskitRuntimeService

which we provide further details on how to use them.

Prerequisite​

This section requires topics described in the previous sections (Sampler, Sampling estimation and Sampling Backend), so you need to read them before this section.

QURI Parts modules used in this tutorial: quri-parts-circuit, quri-parts-core and quri-parts-qiskit. You can install them as follows:

!pip install "quri-parts[qiskit]"

QiskitSamplingBackend and Sampler​

For using an IBMQ backend real device, you would need to activate your account first, then you may pick a device you prefer listed by the all_device variable defined below:

from qiskit import IBMQ

provider = IBMQ.enable_account("Please input your token")
all_devices = provider.backends()
print(all_devices)

Then, a QiskitSamplingBackend may be created by passing in the device

from quri_parts.qiskit.backend import QiskitSamplingBackend
backend = provider.get_backend("ibmq_qasm_simulator")
sampling_backend = QiskitSamplingBackend(backend=backend)

With the sampling backend we just created, we can run the exact codes as in the Sampling Backend and Sampler section of the Sampling Backend tutorial.

QiskitRuntimeSamplingBackend and Sampler​

Qiskit Runtime is a new service provided by IBM to perform experiments on real devices. In this section, we demonstrate how to submit jobs to Qiskit Runtime via QURI Parts. We first prepare some circuit

from quri_parts.circuit import QuantumCircuit

# construct quri-parts circuit
qp_circuit1 = QuantumCircuit(2)
qp_circuit1.add_H_gate(0)
qp_circuit1.add_H_gate(1)
qp_circuit1.add_RX_gate(0, 0.23)
qp_circuit1.add_RY_gate(1, -0.99)

qp_circuit2 = QuantumCircuit(2)
qp_circuit2.add_H_gate(0)
qp_circuit2.add_H_gate(1)
qp_circuit2.add_RX_gate(0, 1.23)
qp_circuit2.add_RY_gate(1, 4.56)

qp_circuit3 = QuantumCircuit(2)
qp_circuit3.add_H_gate(0)
qp_circuit3.add_H_gate(1)
qp_circuit3.add_RX_gate(0, 0.998)
qp_circuit3.add_RY_gate(1, 1.928)

For using the Qiskit Runtime service, you can create a QiskitRuntimeSamplingBackend by passing in a qiskit.providers.backend and a qiskit_ibm_runtime.QiskitRuntimeService object. To see the list of all the devices supported by the Qiskit Runtime Service, you may run:

from qiskit_ibm_runtime import QiskitRuntimeService

service = QiskitRuntimeService()

# Note: all available devices can be obtained by:
print(service.backends())
#output
    [<IBMBackend('ibmq_qasm_simulator')>, <IBMBackend('ibmq_lima')>, <IBMBackend('simulator_statevector')>, <IBMBackend('ibm_nairobi')>, <IBMBackend('ibm_lagos')>, <IBMBackend('ibm_perth')>, <IBMBackend('ibmq_jakarta')>, <IBMBackend('ibmq_quito')>, <IBMBackend('ibmq_belem')>, <IBMBackend('simulator_extended_stabilizer')>, <IBMBackend('simulator_mps')>, <IBMBackend('simulator_stabilizer')>, <IBMBackend('ibmq_manila')>]

We are now ready to create a qiskit runtime sampling backend

from quri_parts.qiskit.backend import QiskitRuntimeSamplingBackend

backend = service.backend("ibmq_qasm_simulator")
qiskit_runtime_sampling_backend = QiskitRuntimeSamplingBackend(
    backend=backend,
    service=service,
)

A sampler can be created as usual

from quri_parts.core.sampling import create_sampler_from_sampling_backend

sampler = create_sampler_from_sampling_backend(qiskit_runtime_sampling_backend)

sampling_cnt_1 = sampler(qp_circuit1, 1000)
sampling_cnt_2 = sampler(qp_circuit2, 2000)
sampling_cnt_3 = sampler(qp_circuit3, 3000)

Grouping multiple sampling jobs into a Session​

In Qiskit Runtime Service, they provide Session objects that allows us to group jobs together. In the above example, a Session is created every time the sampler is called. In QURI Parts, we can group jobs into a single Session with the QiskitRuntimeSamplingBackend as well.

with QiskitRuntimeSamplingBackend(backend=backend, service=service) as qiskit_runtime_sampling_backend:
	sampler = create_sampler_from_sampling_backend(qiskit_runtime_sampling_backend)
	sampling_cnt_1 = sampler(qp_circuit1, 1000)
	sampling_cnt_2 = sampler(qp_circuit2, 2000)
	sampling_cnt_3 = sampler(qp_circuit3, 3000)

Billable Time Tracker​

When executing jobs on real devices, it can be useful to keep track of the cost by tracking the billable time. In the QiskitRuntimeSamplingBackend, we provide the total_time_limit option that allows you to track the total billable time. If the total billable time exceeds the time limit, the backend will reject new job submissions and cancel all unfinished jobs. Let’s create a backend with run time limit of 100 seconds:

TIME_LIMIT = 100

backend = service.backend("ibmq_qasm_simulator")

sampling_backend = QiskitRuntimeSamplingBackend(
	backend=backend,
	service=service,
	total_time_limit=TIME_LIMIT
)

When the total time limit is set, a Tracker object is created along with the sampling backend. You may access it by the .tracker attribute.

tracker = sampling_backend.tracker

The total billable run time of jobs submitted by the backend is not tracked automatically. Instead, it is computed whenever any of the total_run_time, running_jobs, finished_jobs properties is accessed. As we have not submitted any jobs with the backend we just created, it should be 0 at the moment.

tracker.total_run_time

0.0

When we submit jobs with the backend, the jobs will be registered to the tracker for tracking. For example let’s submit 3 sampling jobs:

sampling_job_1 = sampling_backend.sample(qp_circuit1, 10)
sampling_job_2 = sampling_backend.sample(qp_circuit2, 20)
sampling_job_3 = sampling_backend.sample(qp_circuit3, 30)

These jobs will be stored inside the tracker and can be accessed by the .running_jobs property if they are still being executed by the real device. If any of the job is finished, you may access them with the .finished_jobs property.

print(tracker.running_jobs)
print(tracker.finished_jobs)
#output
    [<quri_parts.qiskit.backend.primitive.QiskitRuntimeSamplingJob object at 0x17c66f1f0>, <quri_parts.qiskit.backend.primitive.QiskitRuntimeSamplingJob object at 0x17b24fdf0>, <quri_parts.qiskit.backend.primitive.QiskitRuntimeSamplingJob object at 0x17b24a3d0>]
    []

Data Saving and Replaying​

As we might want to perform different analysis using the same data generated by real devices, having a way to save and retrieve past experiment data can be useful. In this section, we explain how to save and replay past experiment data generated by Qiskit devices.

The data saving feature can be activated by setting the save_data_while_sampling to True. Both QiskitSamplingBackend and QiskitRuntimeSamplingBackend support this feature. Let’s use the local Aer simulator as an example.

from quri_parts.qiskit.backend import QiskitSamplingBackend
from qiskit_aer import AerSimulator
from quri_parts.core.sampling import create_sampler_from_sampling_backend

sampling_backend = QiskitSamplingBackend(
	backend=AerSimulator(),
	save_data_while_sampling=True # activate data saving feature
)

sampler = create_sampler_from_sampling_backend(sampling_backend)

cnt1 = sampler(qp_circuit1, 100)
cnt2 = sampler(qp_circuit2, 200)
cnt3 = sampler(qp_circuit3, 300)

print(cnt1)
print(cnt2)
print(cnt3)
#output
    {2: 3, 3: 5, 0: 56, 1: 36}
    {3: 1, 2: 1, 0: 108, 1: 90}
    {1: 7, 0: 9, 2: 147, 3: 137}

After performing sampling job like the above, we may save the sampling data into a json file:

import json

with open('saved_sampling_job.json', 'w') as fp:
	json.dump(sampling_backend.jobs_json, fp)

The jobs_json property accessed above encodes all the past sampling jobs in the order they were submitted. Now, let’s load it back to the memory and replay with the QiskitSavedDataSamplingBackend.

from quri_parts.qiskit.backend import QiskitSavedDataSamplingBackend

with open('saved_sampling_job.json', 'r') as fp:
	saved_data = json.load(fp)

replay_backend = QiskitSavedDataSamplingBackend(
	backend=AerSimulator(),
	saved_data=saved_data
)

replay_sampler = create_sampler_from_sampling_backend(replay_backend)

replay_cnt1 = replay_sampler(qp_circuit1, 100)
replay_cnt2 = replay_sampler(qp_circuit2, 200)
replay_cnt3 = replay_sampler(qp_circuit3, 300)

print(replay_cnt1)
print(replay_cnt2)
print(replay_cnt3)
#output
    {2: 3, 3: 5, 0: 56, 1: 36}
    {3: 1, 2: 1, 0: 108, 1: 90}
    {1: 7, 0: 9, 2: 147, 3: 137}