Rapid Quantum Prototyping (RQP)

QML (VQC/QSVM) RQP Pro

Advanced quantum classification with workbook-driven configuration, async orchestration, mandatory classical baselines, live backend logs, runtime estimation, and consolidated local reporting.

Best used on a desktop or laptop screen. The quick interface is tuned for loading a workbook, loading data, adjusting the most important QML parameters, and then letting the backend carry the rest of the run.

Status

A tighter operator view of access, runtime, and the currently active async job.

Ready for the next step.-

Backend health

Access

Pending

Mode

VQC

Current job

Waiting for prepared data

Current stage

Ready

Job status

idle

Time

Initial ETA: -

Elapsed: -

Current ETA: -

Worker profile

small

Memory

in use: -

free: -

max: -

Input Workflow

Controls, checks, and artifact downloads.

1. Access

Use the normal access-key workflow here. A blank key still lets you activate the public demo for local testing.

QML access key

Check access first so the page knows whether you are running under an authenticated QML profile or the public demo profile.

2. Config Load

Load a workbook once. The page inspects it automatically and uses it as the current configuration snapshot.

Workbook

No file loaded

Select the workbook you want this run to trust. Re-select it after editing on disk so the browser sends the updated bytes.

3. Data Load

Load a dataset file. Preparation runs automatically and refreshes the runtime estimate.

Dataset Upload

No file loaded

Choose a dataset file to stage it in cloud storage. Bundled sample datasets can still be resolved from the workbook reference.

Fast Assessment

Lightweight dataset triage for signal strength, classical separability, nonlinear lift, and QML suitability.

Not assessed

Model Setup

Choose worker, simulator, QML method, scaling, and VQC/QSVM parameters before deeper analysis.

4. Execution Profile

Choose the simulator backend and the cloud-side execution profile. Worker profile is validated against both your access tier and the current qubit/runtime plan.

Worker profileSimulator backend

Selected worker: Small

5. QML Parameters

These controls are now live planning and execution overrides. Inspect the workbook first, then tune locally.

Quantum methodFeature mapAnsatzEntanglementMeasurementReadout

Ansatz repsQubits

Feature-map repeats

Scale circuit inputsClip scaled inputsInput scaling method

Robust percentile95%

707580859095

Quantum reduction

These overrides flow into preparation, planning, and async execution. If qubits or quantum reduction change after preparation, the client re-prepares the data before the run so the circuit width and feature matrix stay matched.

PCA compresses variance, mutual information keeps label-informative features, and none uses the selected feature view directly.

Measurement and readout affect the VQC expectation-value head; QSVM hides them because it uses state-fidelity kernel matrices instead.

Optimizer Parameters

Lean tuning controls for the parameters that most strongly affect runtime and convergence.

Optimizer

Adam uses exact gradients on the simulator. SPSA estimates the gradient from two circuit evaluations per iteration, which makes it tolerant to shot noise and the realistic choice for hardware-style runs.

Initialization

Identity block starts the ansatz near the identity so initial gradients stay usable on wide circuits (barren-plateau mitigation).

Head refit frequency

Every N iterations the classical readout head is refit exactly (convex solve) instead of gradient steps. 0 disables the refit.

IterationsLearning rate

RepeatsBatch size

Early stoppingBalance train split onlyOptimize decision threshold (binary only)

Class balancing undersamples the training split only. Threshold optimization tunes one binary validation threshold for QML and classical baselines; multiclass runs keep argmax prediction.

Patience checksValidate every

Validation sample rows

Validation metricClass weights

Validation is sampled during VQC training for speed. The backend still computes the full validation metrics at the end of the run.

6. Estimated Run Time

Auto-refreshed after data preparation. This is the right place to sanity-check whether a heavier QML run still fits an interactive workflow.

Nominal runtime

Selected worker

small

Validation cadence

Simulator backend

pennylane_default_qubit

The estimate appears after plan generation. For now, this page updates it automatically after prepare.

Deep Assessment

Run after model setup so the proxy kernel uses the current scaling and method choices.

7. Deep Assessment

Optional capped kernel due-diligence before committing to a full QML run.

Deep Assessment

Default cap is 500 stratified rows. Diagnostics are O(N²), so larger caps can take minutes.

Optional

Uses capped bootstrap stability probes, a label-permutation sanity check, linear/RBF SVM probes, an angle-fidelity quantum-kernel proxy, and kernel-target alignment. It is a richer triage signal, not a replacement for a real VQC/QSVM run.

Execute

Queue the selected QML run, then save reports and notebooks from the completed result.

8. Execute Run

Queue the remaining stages asynchronously. The backend will plan, run baselines, run the QML method, and generate the report in the background.

Config load triggers inspect automatically. Data load triggers prepare and planning automatically. Execute run handles the async orchestration from there.

Check license or demo first so the page can load the active QML limits.

9. Review Files

Export and reload controls for saved QML review snapshots. Reload stays available before a run so you can reopen earlier results at any time.

Run artifacts are staged in backend storage for about 24 hours. The JSON archive embeds the report, Excel, notebook, Python export, and small replay files so those can be downloaded again after reload.

Results View

Filter the result panes while keeping the input workflow on the left.

Client Log

Lean operator-side events from the browser workflow.

Client-side events will appear here as you check access, load files, queue runs, and reopen previous jobs.

Backend Log

Backend-stage events from async polling. This is the place to watch the actual QML job move across plan, baselines, QML, and report.

Backend log entries appear after a job has been queued or reloaded from the backend.

Configuration Summary

The current dataset, task framing, and effective model settings flowing through the backend.

Mandatory baselines

Mandatory Baseline Status

Workbook baseline status has not been parsed yet.

Warnings

Warnings are deduplicated across workbook inspection, preparation, async execution, and reporting so the important caveats stay visible.

No active warnings. This usually means the latest configuration, preparation, and async stages all agreed with each other.

Quantum Feature Scaling Preview

Live preview from the prepared train split. It updates when you change input scaling, percentile, or clipping controls.

Scaled feature histogram appears after data preparation.

Data Preparation Summary

Prepared shapes, feature counts, class balance, and preprocessing choices from the current job context.

Prepared job ID

Selected features

Quantum feature names

Prepared split distribution plot

No class distribution plot available yet.

Effective training distribution

Mirrors prepared train split while balancing is off

No class distribution available yet.

Effective training distribution plot

No class distribution plot available yet.

No preprocessing summary available yet.

Run Plan

Workload, runtime, memory, and circuit estimates automatically refreshed from the prepared dataset and current workbook settings.

Hardware Feasibility

No details available yet.

Recommendations

Baseline Results

Mandatory classical baselines now arrive through the async pipeline and remain the reference frame for any QML claim.

Benchmark methods

Each classical method is reported twice: once in strict parity mode against the QML feature and row budget, and once as a stronger practical classical reference.

Benchmark method details appear after the classical stage finishes.

Baseline comparison will appear after the async baseline stage completes.

QML Results

Validation and test metrics, circuit summary, training trace summary, and the baseline comparison preview all come together here.

Training History Summary

Training history becomes available after a QML run.

Circuit Summary

Features reduced usingPCA

Circuit Executions

No details available yet.

Counts are row-level QNode sample evaluations, not multiplied by shots or internal autodiff device executions.

Training Loss

Training loss plot appears after at least two optimization trace points.

Baseline Comparison Preview

Comparison preview appears after both baselines and the QML stage complete.

Memory Telemetry

Worker memory is sampled through the async job and mirrored back into the status surface, backend log, report, and this chart.

Configured memory

Memory used

Memory free

Memory limit

Max memory used

Memory used %

Cgroup used

Process tree RSS

Peak process RSS

Memory telemetry appears after at least two worker samples.

Final Report

Reporting stays local for now. We surface the run summary and final seven-method comparison without flooding the page with internal storage paths.

Job ID

Embedded archive files

Skipped archive files

Run summary becomes available after report generation.