Changelog¶
Changelog¶
All notable changes to qb-compiler, the open-source quantum circuit compiler by QubitBoost, will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
[Unreleased]¶
Added¶
Qiskit SDK 2.x compatibility:
qiskitdependency widened to>=1.0,<3.0.CI now runs the test suite against both Qiskit 1.4 and Qiskit 2.0 in matrix.
QBCalibrationLayoutPlugin— properqiskit.transpiler.layoutstage plugin. Invoke viagenerate_preset_pass_manager(layout_method="qb_calibration")with theQB_CALIBRATION_PATHenv var set. Plugin is now discoverable through Qiskit’s entry-point system.
Changed¶
qb_transpile()now injectsQBCalibrationLayoutinto the pass manager’spre_layoutstage instead oflayout. On Qiskit 2.x the previous approach triggeredApplyLayoutKeyErrorand silently fell back to stockqiskit.transpile, bypassing calibration-aware layout. The custom pipeline is now the primary code path on both Qiskit versions.QBTranspilerPluginentry-point group corrected from the non-existentqiskit.transpiler.stagetoqiskit.transpiler.layout, now pointing atQBCalibrationLayoutPlugin. The plugin was previously undiscoverable via Qiskit’s loader.
Deprecated¶
QBTranspilerPlugin.get_pass_manager(calibration_data=...)— emits aDeprecationWarningand will be removed in 0.4.0. Migrate togenerate_preset_pass_manager(layout_method="qb_calibration")withQB_CALIBRATION_PATHset, or callqb_transpile()directly.
Fixed¶
ci.ymlworkflow now triggers onmasteras well asmain(the repo’s default branch ismaster; the workflow had been dormant).Removed the phantom
[qiskit]optional-dependency extra from CI install commands (it did not exist inpyproject.tomland was silently ignored).
[0.1.0] - 2026-03-13¶
Added¶
Core IR: QBCircuit, QBDag, QBGate, QBMeasure, QBBarrier
Qiskit and OpenQASM 2.0 converters
CalibrationMapper: VF2-based calibration-weighted qubit placement
NoiseAwareRouter: Dijkstra shortest-error-path SWAP routing
NoiseAwareScheduler: ALAP scheduling with T1/T2 urgency scoring
GateDecomposition: Native basis decomposition (IBM ECR/RZ/SX/X, Rigetti CZ/RX/RZ, IonQ MS/GPI/GPI2, IQM CZ/PRX)
ErrorBudgetEstimator: Pre-execution fidelity prediction
T1 asymmetry awareness: readout-scaled penalty for qubits with high |1> decay
Temporal correlation detection: Pearson correlation across calibration snapshots
Calibration subsystem: StaticCalibrationProvider, CachedCalibrationProvider, BackendProperties
Noise modelling: EmpiricalNoiseModel, FidelityEstimator
Backend support: IBM Heron, Rigetti Ankaa, IonQ Aria/Forte, IQM Garnet/Emerald
Cost estimation with vendor pricing
Qiskit transpiler plugin: QBCalibrationLayout, qb_transpile(), QBPassManager
CLI:
qbc compile,qbc info,qbc calibration showGate cancellation and commutation analysis optimisation passes
Depth and gate count analysis passes
ML Phase 2: XGBoost layout predictor (AUC=0.94, 454KB, +5.4% fidelity on GHZ-8)
ML Phase 3: GNN layout predictor (dual-graph GCN, 42KB, +6.5% fidelity on QAOA-8)
ML Phase 4: RL SWAP router (PPO actor-critic, 190KB, calibration-aware routing)
ML training infrastructure: data generator, feature extraction, model training scripts
461 tests covering all passes, IR, calibration, backends, ML pipeline
CI/CD: GitHub Actions for lint, typecheck, test matrix (Python 3.10-3.12)
10 example scripts demonstrating key features
Comprehensive benchmark suite comparing all ML phases