HPCA 2023 Supermarq Tutorial using cirq-superstaq
You can run this notebook on Colab or Binder. The Colab requires sign in while Binder does not.
[1]:
try:
import cirq_superstaq as css
except ImportError:
print("Installing cirq-superstaq...")
%pip install --quiet 'cirq-superstaq[examples]'
print("Installed cirq-superstaq.")
print("You may need to restart the kernel to import newly installed packages.")
import cirq_superstaq as css
[2]:
try:
import supermarq
except ImportError:
print("Installing supermarq...")
%pip install --quiet supermarq
print("Installed supermarq.")
print("You may need to restart the kernel to import newly installed packages.")
import supermarq
Basics:
1. Service creation
[3]:
# Provide your api key to `css.Service()` using the `api_key=` argument if
# the SUPERSTAQ_API_KEY environment variable is not set.
# Submit cirq circuits via `cirq-superstaq`
service = css.Service()
print(service.get_balance())
20 credits
[4]:
# See which targets are available
service.get_targets(available=True)
[4]:
[Target(target='aqt_keysight_qpu', supports_submit=False, supports_submit_qubo=False, supports_compile=True, available=True, retired=False, accessible=True),
Target(target='aqt_zurich_qpu', supports_submit=False, supports_submit_qubo=False, supports_compile=True, available=True, retired=False, accessible=True),
Target(target='aws_dm1_simulator', supports_submit=True, supports_submit_qubo=False, supports_compile=True, available=True, retired=False, accessible=True),
Target(target='aws_sv1_simulator', supports_submit=True, supports_submit_qubo=False, supports_compile=True, available=True, retired=False, accessible=True),
Target(target='aws_tn1_simulator', supports_submit=True, supports_submit_qubo=False, supports_compile=True, available=True, retired=False, accessible=True),
Target(target='cq_sqale_qpu', supports_submit=True, supports_submit_qubo=False, supports_compile=True, available=True, retired=False, accessible=True),
Target(target='cq_sqale_simulator', supports_submit=True, supports_submit_qubo=False, supports_compile=True, available=True, retired=False, accessible=True),
Target(target='eeroq_wonderlake_qpu', supports_submit=False, supports_submit_qubo=False, supports_compile=True, available=True, retired=False, accessible=True),
Target(target='ibmq_brisbane_qpu', supports_submit=True, supports_submit_qubo=False, supports_compile=True, available=True, retired=False, accessible=True),
Target(target='ibmq_fez_qpu', supports_submit=True, supports_submit_qubo=False, supports_compile=True, available=True, retired=False, accessible=True),
Target(target='ibmq_kingston_qpu', supports_submit=True, supports_submit_qubo=False, supports_compile=True, available=True, retired=False, accessible=True),
Target(target='ibmq_marrakesh_qpu', supports_submit=True, supports_submit_qubo=False, supports_compile=True, available=True, retired=False, accessible=True),
Target(target='ibmq_sherbrooke_qpu', supports_submit=True, supports_submit_qubo=False, supports_compile=True, available=True, retired=False, accessible=True),
Target(target='ibmq_torino_qpu', supports_submit=True, supports_submit_qubo=False, supports_compile=True, available=True, retired=False, accessible=True),
Target(target='ionq_forte-1_qpu', supports_submit=True, supports_submit_qubo=False, supports_compile=True, available=True, retired=False, accessible=True),
Target(target='ionq_ion_simulator', supports_submit=True, supports_submit_qubo=False, supports_compile=True, available=True, retired=False, accessible=True),
Target(target='qscout_peregrine_qpu', supports_submit=False, supports_submit_qubo=False, supports_compile=True, available=True, retired=False, accessible=True),
Target(target='qtm_h1-1_qpu', supports_submit=True, supports_submit_qubo=False, supports_compile=True, available=True, retired=False, accessible=True),
Target(target='qtm_h1-1e_simulator', supports_submit=True, supports_submit_qubo=False, supports_compile=True, available=True, retired=False, accessible=True),
Target(target='qtm_h2-1_qpu', supports_submit=True, supports_submit_qubo=False, supports_compile=True, available=True, retired=False, accessible=True),
Target(target='rigetti_ankaa-3_qpu', supports_submit=True, supports_submit_qubo=False, supports_compile=True, available=True, retired=False, accessible=True),
Target(target='ss_unconstrained_simulator', supports_submit=True, supports_submit_qubo=True, supports_compile=True, available=True, retired=False, accessible=True)]
2. Benchmark instantiation
NOTE: after executing a benchmark circuit, the score should always be evaluated using the same Benchmark object that was used to generate the circuit.
All of the benchmarks can be found in supermarq/benchmarks/.
Think about other circuits/benchmarks you would like to implement and run!
[5]:
# Create your benchmark using cirq circuits
ghz = supermarq.ghz.GHZ(5)
print(ghz.cirq_circuit())
0: ───H───@───────────────M───
│ │
1: ───────X───@───────────M───
│ │
2: ───────────X───@───────M───
│ │
3: ───────────────X───@───M───
│ │
4: ───────────────────X───M───
3. Circuit evaluation
The generated circuits can be evaluated on a backend using any compatible service: AWS Braket, IBM Qiskit, cirq-superstaq, qiskit-superstaq, etc. Here we use cirq-superstaq
[6]:
job_css = service.create_job(
ghz.cirq_circuit(), repetitions=1000, target="ss_unconstrained_simulator", method="dry-run"
)
4. Compute the score
NOTE: after executing a benchmark circuit, the score should always be evaluated using the same Benchmark object that was used to generate the circuit.
[7]:
job_css.status()
[7]:
'Done'
[8]:
counts = job_css.counts(0)
print(counts)
score_css = ghz.score(counts)
print(score_css)
{'11111': 519, '00000': 481}
0.9996388695848232
5. Visualize the results
The function in supermarq/plotting.plot_results produces a simple bar plot. Feel free to copy and paste that code here to generate more detailed figures.
[9]:
supermarq.plotting.plot_results([score_css], ["ghz_css"])
Quantum Program Profiling
The current hardware-agnostic application features can be found in supermarq/features.py. Quantum program profiling is an exciting area of research that is just getting started – what kinds of features would you find meaningful? Try implementing them!
Compute features
[10]:
ghz_circuit = supermarq.ghz.GHZ(10).cirq_circuit()
ghz_features = [
supermarq.features.compute_communication(ghz_circuit),
supermarq.features.compute_depth(ghz_circuit),
supermarq.features.compute_entanglement(ghz_circuit),
supermarq.features.compute_liveness(ghz_circuit),
supermarq.features.compute_measurement(ghz_circuit),
supermarq.features.compute_parallelism(ghz_circuit),
]
print(ghz_features)
[0.2, 1.0, 0.9, np.float64(0.2636363636363636), 0.0, 0]
Visualize the feature vector
[11]:
supermarq.plotting.plot_benchmark(
title="A single GHZ benchmark",
labels=["ghz10"],
features=[ghz_features],
spoke_labels=["PC", "CD", "Ent", "Liv", "Mea", "Par"],
)
Correlating performance <> features
To correlate the performance of a particular device with a certain application feature, it is helpful to evaluate several different benchmarks on that same device. This code example creates 4 different benchmarks, computes their feature vectors, evaluates them on a backend, and then measures the correlation between the performance seen and the application features.
Characterize the benchmarks
[12]:
benchmark_features = {}
benchmarks = [
(supermarq.ghz.GHZ(5), "ghz5"),
(supermarq.hamiltonian_simulation.HamiltonianSimulation(4), "hsim4"),
(supermarq.mermin_bell.MerminBell(3), "mb3"),
(supermarq.bit_code.BitCode(3, 3, [1, 0, 1]), "bitcode3"),
]
for benchmark, label in benchmarks:
benchmark_features[label] = [
supermarq.features.compute_communication(benchmark.cirq_circuit()),
supermarq.features.compute_depth(benchmark.cirq_circuit()),
supermarq.features.compute_entanglement(benchmark.cirq_circuit()),
supermarq.features.compute_liveness(benchmark.cirq_circuit()),
supermarq.features.compute_measurement(benchmark.cirq_circuit()),
supermarq.features.compute_parallelism(benchmark.cirq_circuit()),
]
print(benchmark_features)
{'ghz5': [0.4, 1.0, 0.8, np.float64(0.4666666666666667), 0.0, 0], 'hsim4': [0.5, 1.0, 0.2857142857142857, np.float64(0.5961538461538461), 0.0, 0.20512820512820515], 'mb3': [1.0, 1.0, 0.4375, np.float64(0.6666666666666666), 0.0, 0.11538461538461542], 'bitcode3': [0.4, 0.5, 0.8571428571428571, np.float64(0.6142857142857143), 0.46153846153846156, 0.0]}
Evaluate
[13]:
jobs = []
for benchmark, label in benchmarks:
job = service.create_job(
benchmark.cirq_circuit(), repetitions=1000, target="ss_unconstrained_simulator"
)
jobs.append((label, job, benchmark))
Wait until the benchmarks have successfully executed…
[14]:
device_scores = {}
for label, job, benchmark in jobs:
if job.status() == "Done":
counts = job.counts(0)
print(counts)
score = benchmark.score(counts)
print(score)
device_scores[label] = score
else:
print(label, "not done!")
{'00000': 501, '11111': 499}
0.999998999999
{'1000': 5, '1111': 530, '1011': 102, '1001': 18, '0111': 87, '1101': 83, '1100': 14, '1010': 17, '1110': 95, '0100': 5, '0110': 11, '0101': 20, '0001': 2, '0011': 10, '0010': 1}
0.9999466094067264
{'001': 1000}
1.0
{'11111110001': 1000}
1.0
Measure the correlation between device performance and application features
[15]:
supermarq.plotting.plot_correlations(
benchmark_features,
device_scores,
["PC", "CD", "Ent", "Liv", "Mea", "Par"],
device_name="ibmq_sim",
)
