Release Notes

Contents

Release Notes#

v0.2.0#

Release date: 2026-09-11

  • Added features

    • Unified circuit API: fq.Circuit with the n_qubits spelling, semantic qubit keywords and compatible legacy aliases.

    • FlagQuantum IR: one versioned, serializable and validated representation shared by compilation, execution and deployment.

    • Runtime planning: fq.plan and Circuit.runtime_plan explain the selected representation, execution policy and blockers before anything runs.

    • A single execution entry point: fq.run returns fq.ExecutionResult with a stable value, state, samples, plan, accuracy, metrics, provenance, runtime and compatibility surface, plus fail-closed handling of unknown keywords.

    • PyTorch-native training: fq.Module returns autograd tensors, fq.train runs a caller-owned optimizer loop and returns fq.TrainingResult, and module checkpoints support save and restore.

    • Three simulation representations behind one program: statevector, matrix product state and tensor network, with optional JAX kernels behind the same PyTorch interface.

    • Compilation: target-independent optimization with compiler.optimize, target-aware compilation with explicit coupling maps and recorded routing, and compiler plugins discovered through the extension registry.

    • Measurement and noise: Pauli observables with expectation, probabilities, samples and counts outputs; one backend-neutral NoiseModel driving exact density-matrix evolution, batched statevector trajectories and MPS quantum trajectories.

    • Distributed execution: sharded statevector forward, backward, training and optimizer state, and rank-owned distributed MPS training.

    • Hardware execution paths: explicit compiler and target selection for remote submission, ordered logical-to-physical mapping, qubit-wise-commuting grouping for Hamiltonian measurement, and sealed deployment packages.

    • Ecosystem and research surfaces: interoperability adapters for Qiskit, PennyLane, Cirq, Braket and CUDA-Q; QPU digital twins; a repetition-code memory experiment; circuit drawers; and an algorithms package.

    • Stable error categories in flagquantum.errors with compatibility to the corresponding Python built-in exceptions.

  • Removed / replaced

    • The pre-release device-oriented API — DistributedQuantumDevice, GeneralEncoder, the invertible unitary mode, DTensor interchange helpers, and device-oriented gates and measurement — is not part of the v0.2.0 product and is removed without a compatibility layer.

v0.1.0#

Release date: 2026-06-24

  • Initial release of FlagQuantum as a distributed quantum statevector simulator built on PyTorch.

  • Distributed statevector simulation using DTensor for multi-GPU execution, with automatic resharding during gate operations.

  • Pauli, Clifford, rotation and controlled gates with parameterized support, and custom gate registration.

  • Angle, amplitude and basis encoding, plus a general user-defined encoder.

  • Invertible backpropagation for memory-efficient gradient computation, and measurement post-selection with a depolarizing noise model.

  • Text and Matplotlib circuit drawing, and an OpenQASM 2.0/3.0 exporter.

The repository development version additionally provides QPU digital-twin studies, detached remote job submission with receipts, dynamic circuits and further simulation representations. Install from source to use them.

Contents