QAISS keeps every provider behind an adapter. The Lab produces a Workload Passport; any adapter can run it and emit a new passport. Results then fall into two classes. Simulators that follow the sampling contract are checked bit-for-bit. Anything sampled by a provider's own service, including every quantum processor, can only be checked statistically.
Three execution classes
| Class | What actually runs | Result check | Today |
|---|---|---|---|
| Classical CPU simulation | Exact state vector or tensor network on an ordinary processor | Bit-for-bit | Browser, QAISS connector, Origin pyqpanda3 CPUQVM |
| GPU-accelerated classical simulation | The same mathematics on an NVIDIA GPU (cuStateVec, cuTensorNet). Faster, still classical | Bit-for-bit (to fp64 rounding) | Adapters written; never executed (no NVIDIA GPU) |
| Quantum hardware | A provider QPU, e.g. Origin Wukong, via its cloud | Statistical (TVD vs the ideal distribution) | Adapter written; never executed (no API key or access) |
GPU simulation is never quantum hardware execution, and nothing here demonstrates quantum advantage.
Interchange formats
- Workload Passport v0.1: the QAISS record (circuit, seed, shots, noise, backend, hashes).
- OpenQASM 2.0: exported by the Lab and the connector (pyqpanda3 reads it with convert_qasm_string_to_qprog).
- OriginIR: Origin Quantum's text format, exported by the Lab and the connector (convert_originir_string_to_qprog). The QPanda3 adapter loads exactly this text.
- Bit order: QAISS count keys write qubit 0 first; pyqpanda3 keys write c[0] last. The connector reverses keys when it reads Origin results.
Platform components
| Component | Purpose | Where | Status |
|---|---|---|---|
| Circuit studio | Build or import circuits, see the diagram, Bloch spheres and phases | /lab | Live (≤ 8 qubits × 20 steps in the visual builder) |
| Simulation engine | Exact fp64 state vector with seeded sampling | /lab, connector | Live |
| GPU simulation backend | cuStateVec / cuTensorNet adapters, code translator, memory planner | /gpu, connector | Untested on GPU |
| Hardware execution adapter | Origin Quantum Cloud through the documented pyqpanda3.qcloud API | connector | Untested: needs QPANDA_QCLOUD_API_KEY |
| Hybrid orchestrator | Choose a backend per passport; same record format everywhere | /lab#elsewhere, CLI --backend | Manual (CLI); no hosted scheduler |
| Results explorer | Histogram vs exact, TVD, output hashes, verifier | /lab | Live |
| Benchmark engine | One protocol for the browser and every connector backend (time, memory, accuracy vs reference); records compare at /benchmarks | /benchmarks, CLI bench | Live (browser, CPU, QPanda3); GPU runs pending |
| Experiment history | Session history plus up to 20 saved passports; import OpenQASM / OriginIR | /lab | Live (this browser only; no accounts or server storage) |
| AI assistant | Explain concepts, draft circuits | — | Not built |
Running a passport elsewhere
cd connectors/cuquantum
python -m qaiss_cuquantum run passport.json --backend qpanda3 # Origin pyqpanda3 CPUQVM
python -m qaiss_cuquantum export passport.json --format originir # or qasm
export QPANDA_QCLOUD_API_KEY=… # your own key; never commit it
python -m qaiss_cuquantum run passport.json --backend origin-cloud --origin-backend full_amplitudeOrigin Quantum Cloud access needs your own account and API key. QAISS holds no key, runs nothing on Origin hardware and has no partnership with Origin Quantum or NVIDIA.