Quantum task batching

Quantum task batching is available on every Amazon Braket device, except the local simulator. Batching is especially useful for quantum tasks you run on the on-demand simulators (TN1 or SV1) because they can process multiple quantum tasks in parallel. To help you set up various quantum tasks, Amazon Braket provides example notebooks.

Batching allows you to launch quantum tasks in parallel. For example, if you wish to make a calculation that requires 10 quantum tasks and the circuits in those quantum tasks are independent of each other, it is a good idea to use batching. That way, you don’t have to wait for one quantum task to be complete before another task begins.

The following example shows how to run a batch of quantum tasks:

circuits = [bell for _ in range(5)]
batch = device.run_batch(circuits, s3_folder, shots=100)
print(batch.results()[0].measurement_counts)  # The result of the first quantum task in the batch

For more information, see the Amazon Braket examples on GitHub or Quantum task batching, which has more specific information about batching.

About quantum task batching and costs

A few caveats to keep in mind regarding quantum task batching and billing costs:

Quantum task batching and PennyLane

Take advantage of batching when you’re using PennyLane on Amazon Braket by setting parallel = True when you instantiate an Amazon Braket device, as shown in the following example.

device = qml.device("braket.aws.qubit",device_arn="arn:aws:braket:::device/quantum-simulator/amazon/sv1",wires=wires,s3_destination_folder=s3_folder,parallel=True,)

For more information about batching with PennyLane, see Parallelized optimization of quantum circuits.

Task batching and parametrized circuits

When submitting a quantum task batch that contains parametrized circuits, you can either provide an inputs dictionary, which is used for all quantum tasks in the batch, or a list of input dictionaries, in which case the i-th dictionary is paired with the i-th task, as shown in the following example.

from braket.circuits import Circuit, FreeParameter, Observable
from braket.aws import AwsQuantumTaskBatch

# create the free parameters
alpha = FreeParameter('alpha')
beta = FreeParameter('beta')

# create two circuits
circ_a = Circuit().rx(0, alpha).ry(1, alpha).cnot(0,2).xx(0, 2, beta)
circ_a.variance(observable=Observable.Z(), target=0)

circ_b = Circuit().rx(0, alpha).rz(1, alpha).cnot(0,2).zz(0, 2, beta)
circ_b.expectation(observable=Observable.Z(), target=2)

# use the same inputs for both circuits in one batch

tasks = device.run_batch([circ_a, circ_b], inputs={'alpha': 0.1, 'beta':0.2})

# or provide each task its own set of inputs

inputs_list = [{'alpha': 0.3,'beta':0.1}, {'alpha': 0.1,'beta':0.4}]

tasks = device.run_batch([circ_a, circ_b], inputs=inputs_list)

You can also prepare a list of input dictionaries for a single parametric circuit and submit them as a quantum task batch. If there is N input dictionaries in the list, the batch contains N quantum task. The i-th quantum task corresponds to the circuit executed with i-th input dictionary.

from braket.circuits import Circuit, FreeParameter

# create a parametric circuit
circ = Circuit().rx(0, FreeParameter('alpha'))

# provide a list of inputs to execute with the circuit
inputs_list = [{'alpha': 0.1}, {'alpha': 0.2}, {'alpha': 0.3}]

tasks = device.run_batch(circ, inputs=inputs_list)