Quantum Backend Bench

PyPI: https://pypi.org/project/quantum-backend-bench/

Website: https://sidrichardsquantum.github.io/Quantum_Backend_Bench/

Compare local quantum SDKs and translate supported circuits and workflows between them through versioned neutral schemas. The package runs shared benchmark definitions across Cirq, PennyLane, Amazon Braket LocalSimulator, Qiskit Aer, CUDA-Q, and pyQuil QVM, then reports standardized runtime, structural, and distribution metrics. pytket is used for circuit analysis and compilation-style metrics, not as an execution backend.

See USAGE.md for a task-oriented guide to the CLI and Python API, ROADMAP.md for planned interop and translation work, and CHANGELOG.md for release notes. For research workflows and interpretation, see PROBLEM.md, THEORY.md, METHODOLOGY.md, RESULTS.md, SCHEMA.md, COMPATIBILITY.md, and LIMITATIONS.md.

Table of Contents

• Features
• Scope
• Backend Support
• Compatibility
• Release Artifacts
• Installation
• Two-Minute First Run
• GitHub Codespaces
• Quickstart
• Benchmark Suite
  • GHZ
  • QFT
  • Bernstein-Vazirani
  • Deutsch-Jozsa
  • Random Circuit
  • Quantum Volume Style
  • Grover
  • Hamiltonian Simulation
  • QAOA MaxCut
  • Noise Sensitivity
• Python API
• Result Bundles and DataFrames
• Tutorial Notebooks
• Project Layout
• Development
• SDK Utility Workflows
• Circuit Translation
• Roadmap
• Notes
• Author
• License

Features

• Unified BenchmarkSpec abstraction for reusable benchmark definitions
• Local-first execution backends with no cloud credentials required
• Built-in benchmarks for GHZ, Bernstein-Vazirani, Deutsch-Jozsa, QFT, random circuits, quantum-volume-style circuits, Grover search, Hamiltonian simulation, QAOA MaxCut, and noise sweeps
• Standardized metrics including depth, gate counts, runtime, success probability, and total variation distance
• CLI commands for discovery, backend capability reporting, single runs, backend comparison, presets, reports, SDK parity scorecards, semantic audits, round-trip translation audits, compilation audits, and noise sweeps
• Experiment manifests with environment capture and repeated runtime statistics
• Named benchmark suites for smoke, standard, and scaling runs
• Native circuit drawing through Cirq, PennyLane, Braket, and pytket renderers
• JSON/CSV export, summary rankings, and matplotlib plot generation
• Installable in GitHub Codespaces with Python 3.11+

Scope

The core package scope is local SDK comparison, translation, and verification:

• run the same small benchmark definitions across local simulator SDKs
• translate supported circuits, Pauli Hamiltonians, workflows, and result shapes through neutral schemas
• verify translations with exact, sampled, canonical, matrix, audit, and drift-check workflows
• report reproducible local runtime, structure, distribution, and environment metadata

Peripheral workflows are kept useful but deliberately secondary: plotting, tutorial notebooks, generated documentation assets, and reference-result bundles. These support reproducibility and review, but they are not the core API surface.

Non-goals for the default workflow are cloud job submission, QPU performance claims, provider billing workflows, arbitrary Python migration, and broad benchmark expansion that is not tied to SDK comparison or translation validation.

Backend Support

The core execution path is intentionally free, local, and credential-free. Backends that require cloud accounts, provider billing, remote queues, or private services should stay outside the default workflow unless they are added as clearly optional provider adapters. The default recommendation is to keep new-user benchmarking instantly accessible through local simulators.

Compatibility

See COMPATIBILITY.md for the supported Python versions, optional SDK extras, local runtime requirements, and CI coverage status for each integration.

Release Artifacts

The published wheel is intentionally lean and contains the importable package, bundled presets, metadata, entry point, and license. The source distribution is broader by design: it includes tests, examples, tutorial notebooks, docs source and generated docs assets, schemas, reference results, citation metadata, and the development container definition as research/reproducibility assets. See RELEASE_POLICY.md.

Installation

Install from PyPI:

python -m pip install quantum-backend-bench

Install execution backends as needed:

python -m pip install "quantum-backend-bench[cirq]"
python -m pip install "quantum-backend-bench[pennylane]"
python -m pip install "quantum-backend-bench[braket]"
python -m pip install "quantum-backend-bench[qiskit]"
python -m pip install "quantum-backend-bench[cudaq]"
python -m pip install "quantum-backend-bench[pyquil]"
python -m pip install "quantum-backend-bench[yaml]"
python -m pip install "quantum-backend-bench[all]"
python -m pip install "quantum-backend-bench[full]"

The default package and .[dev] workflow are self-contained Python installs. The all extra is the practical Python-only comparison stack and intentionally excludes CUDA-Q and pyQuil. Use full only when you explicitly want every Python SDK extra, including heavy or external-runtime-backed adapters.

Install from a local checkout:

python -m pip install --upgrade pip
python -m pip install -e .

For development tools:

python -m pip install -e .[dev]
# CI uses reviewed major-version bands:
python -m pip install -c constraints/ci.txt -e .[dev]

For the practical local test matrix:

python -m pip install -e ".[all,dev]"

For the exhaustive optional SDK matrix:

python -m pip install -e ".[full,dev]"

The pyQuil execution test also requires local qvm and quilc executables on PATH. Those are external Rigetti runtime tools and are not installed by pip extras.

Two-Minute First Run

For the shortest public path, install the Cirq extra and run the smoke workflow:

python -m pip install "quantum-backend-bench[cirq]"
quantum-bench run ghz --backend cirq --n-qubits 3 --shots 128 --summary
quantum-bench suite smoke --backends cirq --shots 128 --summary

Expected output is a compact table with runtime, depth, gate counts, and benchmark quality metrics, followed by summary rankings. A saved reference bundle for this workflow is available under examples/reference_results/cirq_smoke_2026-06-03/.

GitHub Codespaces

The repository includes a Codespaces-ready .devcontainer/devcontainer.json using a Python 3.11 base image. On container creation it installs the package in editable mode with development dependencies.

Quickstart

List available benchmarks, suites, and local integrations:

quantum-bench list
quantum-bench info
quantum-bench doctor
quantum-bench recommend --use-case research
quantum-bench compatibility
quantum-bench sdk-parity
quantum-bench semantic-audit --backends cirq qiskit_aer
quantum-bench roundtrip-audit --include-hamiltonian --include-workflow
quantum-bench compile-audit --backends cirq qiskit_aer
quantum-bench noise-audit
quantum-bench validate

Run a single benchmark:

quantum-bench run ghz --backend cirq --n-qubits 5
quantum-bench run ghz --backend cirq --n-qubits 5 --repeats 5

Compare a benchmark across all execution backends and print summary rankings:

quantum-bench compare qft --backends cirq pennylane braket_local qiskit_aer --n-qubits 5 --summary

Run a random circuit:

quantum-bench run random-circuit --backend braket_local --n-qubits 4 --depth 10 --seed 42

Run Grover:

quantum-bench run grover --backend pennylane --n-qubits 3 --marked-state 101

Run Bernstein-Vazirani:

quantum-bench run bernstein-vazirani --backend cirq --n-qubits 4 --secret-string 101

Run Deutsch-Jozsa:

quantum-bench run deutsch-jozsa --backend cirq --n-qubits 4 --oracle-type balanced --bitmask 101

Run Hamiltonian simulation:

quantum-bench run hamiltonian-sim --backend cirq --n-qubits 4 --time 1.0 --trotter-steps 2

Run QAOA MaxCut:

quantum-bench run qaoa-maxcut --backend cirq --n-qubits 4 --graph ring --gamma 0.8 --beta 0.4

Run a noise sweep:

quantum-bench noise-sweep ghz --backend cirq --n-qubits 5

Run free local SDK audits:

quantum-bench sdk-parity --save-json artifacts/sdk_parity.json
quantum-bench semantic-audit --backends cirq qiskit_aer pennylane
quantum-bench roundtrip-audit --targets cirq qiskit_aer pennylane braket_local --include-hamiltonian --include-workflow
quantum-bench compile-audit --backends cirq qiskit_aer
quantum-bench noise-audit
quantum-bench noise-audit --run --backends cirq qiskit_aer --noise-types depolarizing amplitude_damping readout_error

Run a quantum-volume-style circuit:

quantum-bench run quantum-volume --backend cirq --n-qubits 4 --depth 4 --seed 42

Draw a circuit with a native SDK renderer:

quantum-bench draw ghz --backend cirq --n-qubits 5
quantum-bench draw qft --backend pennylane --n-qubits 5 --save-path artifacts/qft_pennylane.png
quantum-bench draw ghz --backend tket --n-qubits 5 --save-path artifacts/ghz_tket.txt

Save JSON and plots:

quantum-bench compare ghz --backends cirq pennylane braket_local --n-qubits 5 --save-json artifacts/ghz.json --save-plot artifacts/ghz.png

Save distribution, heatmap, noise-quality, and suite plots:

quantum-bench run grover --backend cirq --n-qubits 3 --marked-state 101 --save-distribution artifacts/grover_distribution.png
quantum-bench compare ghz --backends cirq pennylane --n-qubits 4 --save-heatmap artifacts/ghz_heatmap.png
quantum-bench noise-sweep ghz --backend cirq --n-qubits 4 --save-quality-plot artifacts/noise_quality.png
quantum-bench suite smoke --backends cirq --save-suite-plot artifacts/smoke_runtime.png

Save CSV:

quantum-bench compare ghz --backends cirq pennylane --n-qubits 5 --save-csv artifacts/ghz.csv

Compare saved results and fail on regressions:

quantum-bench diff artifacts/baseline.json artifacts/current.json --relative-threshold 0.05 --fail-on-regression
quantum-bench bundle artifacts/current.json --output artifacts/current_bundle
quantum-bench diff artifacts/baseline.csv artifacts/current.csv --metric runtime_seconds

Generate a Markdown report:

quantum-bench report artifacts/current.json --output artifacts/current_report.md

Use packaged comparison presets:

quantum-bench preset list
quantum-bench preset show algorithmic --save-json artifacts/algorithmic_manifest.json
quantum-bench preset run runtime --backends cirq pennylane qiskit_aer --save-report artifacts/runtime_report.md

Run a named suite:

quantum-bench suite smoke --backends cirq --summary
quantum-bench suite standard --backends cirq pennylane braket_local --save-csv artifacts/standard.csv
quantum-bench suite standard --list-cases --save-json artifacts/standard_manifest.json

Run a reproducible experiment manifest:

quantum-bench experiment run examples/manifests/runtime_scaling.json

For more complete workflows, result interpretation, and Python examples, see USAGE.md.

Benchmark Suite

GHZ

Generates GHZ states for configurable qubit counts. Ideal output is concentrated on 00...0 and 11...1.

QFT

Implements the Quantum Fourier Transform for structural and runtime comparisons.

Bernstein-Vazirani

Recovers a hidden bitstring with one oracle query. The final qubit is used as the oracle work qubit, so --secret-string must have length --n-qubits - 1.

Deutsch-Jozsa

Runs constant or linear balanced oracle cases. Balanced cases use --bitmask; constant cases use --oracle-type constant --constant-value 0|1.

Random Circuit

Builds reproducible random circuits using a fixed gate set and explicit seed control.

Quantum Volume Style

Builds reproducible shuffled-pair random layers inspired by quantum volume workloads. This is a portable workload, not a formal quantum volume certification routine.

Grover

Implements a small search benchmark for 2 to 4 qubits and reports marked-state success probability.

Hamiltonian Simulation

Implements first-order Trotterized evolution for a simple Ising-style Hamiltonian:

H = sum_i Z_i Z_{i+1} + 0.5 * sum_i X_i

QAOA MaxCut

Builds a single-layer QAOA circuit for line or ring MaxCut instances and reports success probability as probability mass on optimal cut bitstrings.

Noise Sensitivity

Wraps a base benchmark and sweeps depolarizing noise levels. Noise behavior differs by backend and is reported in result metadata. Cirq, PennyLane, and Qiskit Aer inject depolarizing noise in this project; other adapters may report the request without applying noise.

Python API

from quantum_backend_bench.benchmarks.ghz import build_benchmark
from quantum_backend_bench import build_suite, run_benchmark, summarize_results

benchmark = build_benchmark(n_qubits=5)
results = run_benchmark(benchmark, ["cirq", "pennylane", "braket_local"], shots=1024)
suite_results = [
    result
    for benchmark in build_suite("smoke")
    for result in run_benchmark(benchmark, ["cirq"], shots=128)
]
summary = summarize_results(suite_results)

Result Bundles and DataFrames

Create a shareable result bundle from any saved JSON or CSV result file:

quantum-bench bundle artifacts/results.json --output artifacts/results_bundle

The bundle contains normalized JSON/CSV outputs, flattened records, a Markdown report, plots when matplotlib is available, environment metadata when present, and a README for interpretation. Python users can also flatten results directly:

from quantum_backend_bench import results_to_dataframe

frame = results_to_dataframe("artifacts/results.json")

Tutorial Notebooks

The notebooks/ directory contains succinct package-client tutorials. Each notebook starts with problem context, quantum-advantage scope, and parameter definitions, then uses readable tables, plots where useful, saved artifacts, and verification checks.

• 01_quickstart_cirq.ipynb: GHZ and smoke-suite workflow on the free local Cirq simulator.
• 02_compare_local_simulators.ipynb: installed local simulator comparison for GHZ and QFT.
• 03_hamiltonian_simulation_case_study.ipynb: small Ising-style Hamiltonian simulation scaling study.
• 04_compare_sdk_workflows.ipynb: compare one export, execution, artifact, and verification workflow across installed core SDK adapters.
• 05_circuit_translation_workflow.ipynb: all-target local SDK circuit translation with native diagram comparison, reports, runner output, and diagnostics.
• 06_observable_hamiltonian_translation_workflow.ipynb: Pauli observable and Hamiltonian translation across local SDK formats with canonical verification.
• 07_parameterized_workflow_translation.ipynb: parameterized workflow translation, canonical and semantic workflow verification, executable neutral results, result normalization, and Pauli grouping.
• 08_translation_migration_audit_workflow.ipynb: migration diagnostics, supported boundaries, and round-trip audit workflow.

Install notebook helpers with:

python -m pip install -e ".[cirq,plot,notebooks]"

Project Layout

quantum_backend_bench/
├── backends/
├── benchmarks/
├── core/
├── utils/
└── cli.py

Example scripts live in examples/, with a run order and expected outputs documented in examples/README.md. They include backend comparison, GHZ execution, oracle benchmarks, quantum-volume-style execution, suite export, plot generation, circuit diagram export, research manifest generation, repeated-runtime analysis, schema inspection, Markdown report generation, backend capability inspection, and a Cirq noise sweep demo. Treat examples, plots, notebooks, and generated Pages assets as research/reproducibility material around the core compare/translate/verify workflows. The plot gallery example uses compact workloads by default, but remains the heaviest basic example because it renders multiple plots and includes noisy simulation.

Development

Run formatting and linting:

black quantum_backend_bench tests examples
ruff check quantum_backend_bench tests examples

Run the default core test suite:

pytest

Run optional SDK and generated-site documentation checks explicitly when needed:

pytest -m optional_sdk
pytest tests/test_docs_links.py -m "docs or not docs"

Build and inspect release artifacts:

python -m build
python -m twine check dist/*

Continuous integration is handled by .github/workflows/ci.yml, which runs formatting, linting, tests, documentation link validation, Cirq smoke regression checks, package builds, and distribution checks. Publishing is handled by .github/workflows/publish.yml when a version tag such as v0.2.8 is pushed. The workflow expects PyPI trusted publishing to be configured for this repository.

SDK Utility Workflows

The CLI includes SDK-facing workflows beyond local benchmark execution:

quantum-bench export ghz --n-qubits 3 --format openqasm
quantum-bench export ghz --n-qubits 3 --format openqasm3
quantum-bench export ghz --n-qubits 3 --format native --backend cirq
quantum-bench import-qasm artifacts/ghz.qasm3 --name imported_ghz
quantum-bench translate artifacts/ghz.qasm --from-format openqasm --to-format cirq --verify exact --output artifacts/ghz_cirq.py
quantum-bench translate examples/qiskit_circuit.py --from-format qiskit --to-format pennylane --verify exact --output artifacts/circuit_pennylane.py
quantum-bench translate-check examples/translation/qiskit_registers.py --from-format qiskit
quantum-bench translate-check examples/translation/qiskit_registers.py --from-format qiskit --json
quantum-bench translate-check examples/translation/qiskit_registers.py --from-format qiskit --to-format cirq --explain
quantum-bench translate-check examples/translation/qiskit_registers.py --from-format qiskit --to-format cirq --save-markdown artifacts/translation_check.md
quantum-bench translate examples/translation/ghz.qasm --from-format openqasm --to-format cirq --verify exact --save-report artifacts/translation_report.json
quantum-bench translate-all examples/translation/qiskit_registers.py --from-format qiskit --output-dir artifacts/qiskit_registers_all
quantum-bench translate-hamiltonian examples/translation/ising_hamiltonian.json --from-format pauli-json --to-format pennylane --output artifacts/ising_pennylane.py
quantum-bench translate-workflow examples/translation/parameterized_workflow.json --to-format qiskit_aer --verify semantic
quantum-bench translate-result examples/translation/qiskit_counts_result.json --from-format qiskit-counts-json
quantum-bench group-pauli-terms examples/translation/ising_hamiltonian.json --from-format pauli-json
quantum-bench translation-audit
quantum-bench translation-audit --from-format qiskit --to-format qiskit_aer --json
quantum-bench exact ghz --n-qubits 3 --top-k 4 --amplitudes --observable ZZI
quantum-bench run random-circuit --backend cirq --sweep n-qubits=2:5 --sweep depth=4,8
quantum-bench noise-sweep ghz --backend cirq --noise-type bit_flip
quantum-bench diagnose artifacts/ghz.json
quantum-bench recommend --needs-noise --no-external-runtime

translate and the schema-specific translation commands share one versioned neutral model. See Circuit Translation and Schema for supported syntax, diagnostics, verification, and report formats.

Workflow outputs now include schema-versioned runtime counts, probabilities, and Pauli expectations. Use translate-workflow --verify semantic for neutral distribution-TVD, expectation-error, and result-schema checks, or roundtrip-audit --include-workflow --workflow-verify semantic to record those metrics across targets.

Result tables and reports retain compile/transpile metadata where an SDK provides it. The consolidated SDK notebook compares the same workflow across Cirq, Qiskit Aer, PennyLane, and Braket LocalSimulator.

Notes

• The project targets standard pip environments with Python 3.11 or newer.
• No AWS account, cloud credentials, GPUs, or paid services are required.
• The internal circuit model is intentionally simple to keep backend translation maintainable.

Author

Sid Richards

• LinkedIn: sid-richards-21374b30b
• GitHub: SidRichardsQuantum

License

MIT. See LICENSE.