adevolder / QMapDataset

Public

1

Dataset
Discussions

About dataset version

QMapDataset

QMapDataset is a dataset generator designed to help build machine learning models for qubit mapping. In many current quantum computers, qubits are not fully connected — for example, in superconducting quantum processors. In a quantum circuit, a two-qubit gate may need to be applied between qubits that are not directly connected in the hardware. To handle this, a series of SWAP gates (which effectively permute qubits) must be inserted. However, this increases the circuit depth, computation time, and overall execution cost. The qubit-mapping problem aims to find an optimal assignment between the physical qubits of the hardware and the logical qubits in the quantum circuit to minimize the number of required SWAP gates.

Dataset Description

QMapDataset samples a representative set of quantum circuits, including benchmark algorithms (Grover, QFT, Shor, etc.) and random circuits with varying numbers of qubits and depths. Additionally, data augmentation is applied via qubit permutations to further enrich the dataset. This release contains 4,001 samples (2,000 for the IBM Eagle-3 architecture and 2,001 for the IBM Heron-1 architecture).
Need more data?
You can generate it using the QMapDataset Python code available at: https://github.com/rscadrien/QMapDataset

File structure

Each sample includes three files:\
  • circuit.json contains information about the quantum circuit, including metadata on the sampling procedure (type of algorithm, number of permutations, etc.), the number of logical qubits, the circuit depth, and counts of single-qubit and two-qubit gates. \
  • hardware.json contains information about the hardware, including processor type, number of physical qubits, basis gates, coupling map, and qubit properties for each qubit (T1, T2, frequency, anharmonicity, readout error, probability of measuring 0/1 when 1/0 was prepared), as well as single- and two-qubit gate error rates. \
  • mapping.json contains the optimized qubit mapping computed using the IBM transpiler at its highest optimization level. This file serves as the label for the machine-learning model.

Goal

This dataset is intended to support the development of improved machine-learning models for qubit mapping, contributing to more efficient compilation and execution of quantum algorithms on current hardware.