Execution Session

This section explains how to execute a quantum program using the ExecutionSession class.

This class enables the execution of a quantum program with varying parameters and operations, eliminating the need to re-synthesize the model.

Initialization

ExecutionSession is initialized with a quantum program; the output of the synthesis operation. For example:

from classiq import (
    qfunc,
    create_model,
    synthesize,
)
from classiq.execution import ExecutionSession


@qfunc
def main():
    pass


qmod = create_model(main)
qprog = synthesize(qmod)

es = ExecutionSession(qprog)

Operations

ExecutionSession supports two types of operations: sample: Executes the quantum program with the specified parameters. estimate: Computes the expectation value of the specified Hamiltonian using the quantum program.

Each invocation mode can be done using one of the three options: single, batch, submit.

batch: Samples/estimate the quantum program multiple times. The number of samples/estimations is determined by the length of the parameters list.

submit: When using submit as a prefix - the job object is returned immediately, and the results should be polled. See ExecutionJob in the sdk reference.

For the sample method, the following variations are available:

es.sample(parameter: Optional[ExecutionParams])

es.batch_sample(parameter: List[ExecutionParams])

es.submit_sample(parameter: Optional[ExecutionParams])

es.submit_batch_sample(parameter: List[ExecutionParams])

estimate has the same variations:

es.estimate(hamiltonian: Union[str, PauliTerm], parameters: Optional[ExecutionParams])

es.batch_estimate(hamiltonian, parameters: List[ExecutionParams])

es.submit_estimate(hamiltonia, parameters: Optional[ExecutionParams])

es.submit_batch_estimate(hamiltonian, parameters: List[ExecutionParams])

Examples

sample()

A simple example on how to use sample and its variations

from classiq import qfunc, Output, QBit, CReal, allocate, create_model, synthesize, RX
from classiq.execution import ExecutionSession


@qfunc
def main(x: Output[QBit], t: CReal):
    allocate(1, x)
    RX(t, x)


qmod = create_model(main)
qprog = synthesize(qmod)

execution_session = ExecutionSession(qprog)

sample_result = execution_session.sample({"t": 0.5})
batch_sample_result = execution_session.batch_sample([{"t": 0.5}, {"t": 0.6}])
sample_job = execution_session.submit_sample({"t": 0.5})
batch_sample_job = execution_session.submit_batch_sample([{"t": 0.5}, {"t": 0.6}])

estimate()

An example that show how to use estimate and its variations:

from classiq import (
    qfunc,
    Output,
    QBit,
    CReal,
    allocate,
    create_model,
    synthesize,
    PauliTerm,
    Pauli,
    RX,
)
from classiq.execution import ExecutionSession


@qfunc
def main(x: Output[QBit], t: CReal):
    allocate(1, x)
    RX(t, x)


qmod = create_model(main)
qprog = synthesize(qmod)

execution_session = ExecutionSession(qprog)

hamiltonian = [
    PauliTerm(pauli=[Pauli.I], coefficient=1),
    PauliTerm(pauli=[Pauli.Z], coefficient=2),
]

estimate_result = execution_session.estimate(hamiltonian, {"t": 0.5})
batch_estimate_result = execution_session.batch_estimate(
    hamiltonian, [{"t": 0.5}, {"t": 0.6}]
)
estimate_job = execution_session.submit_estimate(hamiltonian, {"t": 0.5})
batch_estimate_job = execution_session.submit_batch_estimate(
    hamiltonian, [{"t": 0.5}, {"t": 0.6}]
)

Handling Long Jobs

When handling long-running jobs (jobs that are submitted to HW providers with potentially very long queues) it is advisable to retrieve and save the job ID for future reference:

...
session = ExecutionSession(qprog)
job = session.submit_sample()
job_ID = job.id

restored_job = ExecutionJob.from_id(job_ID)
results = restored_job.get_sample_result()

Alternatively, for scenarios requiring result polling, e.g., iterative hybrid algorithms, consider the following partial code example: This example runs a series of quantum estimation jobs, step by step. It automatically submits each job, collects the results, checks they meet some specified criteria (for example, if the cost does not exceed a certain threshold), and adjusts the parameters for the next job based on those results.

from classiq.execution import ExecutionSession, ExecutionJob


def get_data_from_file():
    # read data from file
    return data


def store_data_in_file(data):
    # save data in file
    return


session = ExecutionSession(qprog)

iterations_data = get_data_from_file() or []

for _ in range(max_iterations - len(iterations_data)):
    if iterations_data and "result" not in iterations_data[-1]:
        # continue the loop from the last iteration
        last_job = ExecutionJob.from_id(iterations_data[-1]["job_id"])
    else:
        last_job = session.submit_estimate(hamiltonian, params)
        iterations_data.append({"job_id": last_job.id, "params": params})
        store_data_in_file(iterations_data)

    result = last_job.get_estimate_result()
    iterations_data[-1]["result"] = result
    store_data_in_file(iterations_data)

    if result_is_good_enough(result):
        break

    params = compute_the_next_parameters(result)

@cfunc

Please note that algorithms utilizing @cfunc (or cscope in Qmod native) are not supported for long job execution.