# Sirius Quantum — Full Reference

Sirius Quantum Solutions builds the quantum data layer for Physical AI. We
publish open, quantum-native machine-learning datasets and build ReLab, the
quantum data relabelling engine that produces them.

What makes the datasets different: labels come from exact quantum computation —
full-configuration-interaction (FCI) energies, quantum-kernel matrices, exact
many-body ground states — rather than classical approximations such as DFT.
That gives models a noise-free target: every bit of model error is the model's,
not the label's.

All datasets are hosted on Hugging Face under https://huggingface.co/SiriusQuantum and load with the
standard `datasets` library. Licenses are stated on each dataset card.

## Datasets

### SIRIUS-14k

- Category: VQA · Barren Plateaus
- URL: https://huggingface.co/datasets/SiriusQuantum/sirius-14k

13,800 labeled optimization trajectories across 4 VQA circuit architectures. The first public labeled dataset for barren plateau research, with gradient variance profiles, convergence diagnostics, and 24-field multi-label trainability annotations.

Load it:

```python
from datasets import load_dataset
ds = load_dataset("SiriusQuantum/sirius-14k")
```

### QM7b Quantum Relabeled

- Category: Molecular Chemistry
- URL: https://huggingface.co/datasets/SiriusQuantum/qm7b-quantum-relabeled

Quantum-relabeled subset of QM7b (7,211-molecule benchmark): 400 seven-heavy-atom molecules (300 train / 100 test) with a precomputed 7-qubit Heisenberg quantum-kernel matrix, quantum-native labels, and 1-RDM observables. Drop-in for scikit-learn precomputed-kernel pipelines, no quantum hardware required.

Load it:

```python
from datasets import load_dataset
ds = load_dataset("SiriusQuantum/qm7b-quantum-relabeled")
```

### SQMolecular95k

- Category: Quantum Chemistry · Delta-Learning
- URL: https://huggingface.co/datasets/SiriusQuantum/SQMolecular95k

The 95k scale-up of SQMolecular: 94,376 geometries of exact FCI correlation energies, each paired in-file with its MP2 baseline. A harder cross-scaffold transfer benchmark that tests whether the quantum representation carries to unseen chemistry, not just unseen conformers.

Load it:

```python
from datasets import load_dataset
ds = load_dataset("SiriusQuantum/SQMolecular95k")
```

### BBBP Quantum Relabeled

- Category: Drug Discovery · Blood-Brain Barrier
- URL: https://huggingface.co/datasets/SiriusQuantum/bbbp-quantum-relabeled

Quantum-relabeled MoleculeNet BBBP benchmark for CNS drug penetration. 85 compounds encoded as a 25-qubit graph-Hamiltonian circuit, shipped as a precomputed pairwise quantum-fidelity kernel for drop-in SVM classification.

Load it:

```python
from datasets import load_dataset
ds = load_dataset("SiriusQuantum/bbbp-quantum-relabeled")
```

### Quantum Finance Risk Benchmark

- Category: Quantitative Finance
- URL: https://huggingface.co/datasets/SiriusQuantum/quantum-finance-risk-benchmark

1,000 correlated-asset market regimes encoded as Ising-Hamiltonian quantum states with systemic portfolio-risk labels. Quantum features hold 0.69 test error at 16 assets where the classical kernel degrades to chance — a widening sample-efficiency gap.

Load it:

```python
from datasets import load_dataset
ds = load_dataset("SiriusQuantum/quantum-finance-risk-benchmark")
```

### SQMolecular

- Category: Quantum Chemistry · Delta-Learning
- URL: https://huggingface.co/datasets/SiriusQuantum/SQMolecular
- License: cc-by-4.0

10,038 exact FCI correlation energies across 717 organic molecules, 14 thermal geometries each, paired in-file with matched MP2 baselines. A noise-free delta-learning target — every bit of model error is the model's, not the label's. Produced with the ReLab engine.

Load it:

```python
from datasets import load_dataset
ds = load_dataset("SiriusQuantum/SQMolecular")
```

### QM9 Quantum Relabeled

- Category: Molecular Chemistry
- URL: https://huggingface.co/datasets/SiriusQuantum/qm9-quantum-relabeled
- License: cc-by-4.0

Quantum-relabeled subset of QM9, the canonical 133,885-molecule small-molecule quantum-chemistry benchmark: 400 nine-heavy-atom molecules (300 train / 100 test) re-labeled with quantum-native targets alongside the original seventeen quantum-mechanical properties.

Load it:

```python
from datasets import load_dataset
ds = load_dataset("SiriusQuantum/qm9-quantum-relabeled")
```

### Quantum Ground States · 320 Qubits

- Category: Quantum Many-Body · Hardware-Validated
- URL: https://huggingface.co/datasets/SiriusQuantum/quantum-ground-states-320-qubits
- License: cc-by-4.0

750 exact ground states of disordered transverse-field Ising chains from 20 to 320 qubits, pairing local 1-RDM measurements with physical observables. A benchmark for learning properties of quantum states at a scale no state-vector simulator can reach.

Load it:

```python
from datasets import load_dataset
ds = load_dataset("SiriusQuantum/quantum-ground-states-320-qubits")
```

## ReLab Engine

ReLab is our quantum data relabelling engine: it takes an existing classical
dataset and replaces approximate labels with exact quantum-computed ones, or
augments samples with quantum-native features (kernels, 1-RDMs, observables).
Every public dataset above was produced or relabeled with it. Early access:
info@siriusquantum.com (subject "ReLab Early Access").

## Zilver

Distributed quantum simulation on Apple Silicon. Node waitlist:
https://siriusquantum.com/zilver

## Links

- Website: https://siriusquantum.com
- Hugging Face: https://huggingface.co/SiriusQuantum
- GitHub: https://github.com/Sirius-Quantum
- X / Twitter: https://x.com/SiriusQuantum
- LinkedIn: https://www.linkedin.com/company/sirius-quantum-solutions-ltd-sqs/
- Contact: info@siriusquantum.com