Requirements

This section includes information about the hardware platforms and software requirements for FlagQuantum.

Software requirements

• Python 3.10, 3.11 or 3.12

• PyTorch 2.5 or higher, below 2.14

The released package depends on PyTorch only. Everything else is optional.

Supported hardware platforms

Platform	How it is reached
CPU	Default local statevector, MPS and tensor-network execution
NVIDIA GPU	CUDA device selected through fq.ExecutionOptions(device="cuda:0")
FlagOS-supported accelerators	The logical flagos:0 device, reached through Torch-FL

FlagQuantum does not detect or dispatch a specific domestic accelerator by device name. Torch-FL owns vendor detection, runtime activation and the compatibility route, and exposes it through the flagos contract.

Optional dependency groups

Group	Adds
dev	pytest, coverage, xdist, ruff, black, mypy, build, pre-commit
jax	JAX-backed kernels
cuda	Triton
qiskit	Qiskit and Aer interoperability
pennylane	PennyLane interoperability
cirq	Cirq interoperability
braket	Amazon Braket interoperability
cudaq	CUDA-Q kernel export
quafu	Quafu submission support
azure	Azure Quantum submission support
viz	Matplotlib circuit drawing
examples	datasets and transformers for the example workflows
all	Every optional dependency above

Interoperability adapters are optional control-plane boundaries: importing flagquantum never imports Qiskit, PennyLane, Cirq or another external framework.

Support boundaries to read before choosing a target

The maturity of every capability, and what each target has actually been executed with, is published in the capability reference. An implemented API is not automatically production support.