zilver 0.3.0

Open quantum simulation network for Apple Silicon

Zilver

Open quantum simulation network for Apple Silicon.

Zilver turns your Apple Silicon Mac into a quantum simulation node.

Built on MLX. Statevector, density matrix, and tensor network backends.

---

Join the network

Two commands to become a node operator:

pip install zilver[network]
zilver-node start --registry https://registry.siriusquantum.com:7701 --backends sv,dm

Your node auto-detects chip, RAM, and qubit ceilings. It registers, starts serving simulation jobs, and sends heartbeats every 30 seconds. That is it.

Early node operators will receive priority allocation of Sirius Quantum network rewards when the incentive layer launches. The simulation network is live. Rewards are retroactive to genesis nodes.

---

Hardware

Every Apple Silicon chip is supported. Qubit ceilings scale with unified memory.

Chip	RAM	SV qubits	DM qubits	TN qubits
M1	8 GB	28	14	50
M1 / M2	16 GB	30	15	50
M1 Pro / M2 Pro	32 GB	31	15	50
M1 Max / M2 Max	64 GB	32	16	50
M1 Ultra / M2 Ultra	128 GB	33	16	50
M3 / M4	16–24 GB	30–31	15	50
M3 Max / M4 Max	64–128 GB	32–33	16	50

SV = statevector (exact). DM = density matrix (noise-aware). TN = tensor network / MPS (sparse circuits, scales to 50+ qubits regardless of RAM).

---

What Zilver does

Local simulation

MLX-native simulator with full mx.vmap and mx.compile support. One Metal dispatch evaluates an entire loss landscape 400 parameter points in a single GPU kernel instead of 400 sequential circuit runs.

from zilver.circuit import hardware_efficient
from zilver.landscape import LossLandscape

circuit = hardware_efficient(n_qubits=6, depth=3)
result = LossLandscape(circuit, sweep_params=(0, 1), resolution=20).compute()

print(f"Plateau coverage:   {result.plateau_coverage():.1%}")
print(f"Trainability score: {result.trainability_score():.3f}")
print(f"Wall time:          {result.wall_time_seconds:.3f}s")

Distributed network

Submit jobs to any registered node. The coordinator finds an eligible node, dispatches the job, and returns a verified result with a cryptographic proof.

from zilver.client import NetworkCoordinator
from zilver.node import SimJob

coord = NetworkCoordinator("https://registry.siriusquantum.com:7701")

job = SimJob(
    circuit_ops=[{"type": "ry", "qubits": [0], "param_idx": 0}],
    n_qubits=4, n_params=1, params=[1.57], backend="sv",
)
result = coord.submit(job)
print(result.expectation)
print(result.verify(job))   # True cryptographic proof check

Gradient computation

Parameter-shift gradients, fully batched over parameter vectors. Plug directly into any VQA optimizer.

from zilver.gradients import param_shift_gradient

f = circuit.compile(observable="sum_z")
grads = param_shift_gradient(f, params)

---

Install

Simulator only

pip install zilver

Full network node

pip install zilver[network]

Requirements: Apple Silicon Mac, macOS 13 Ventura or later, Python 3.10+.

---

Gate library

Gate	Type	Params
H, X, Y, Z, S, T	Fixed single-qubit
RX, RY, RZ	Rotation	1
U3	Universal single-qubit	3 (theta, phi, lambda)
CNOT, CZ, SWAP, iSWAP	Fixed two-qubit
RZZ, RXX	Ising coupling	1
CRZ	Controlled rotation	1

All parameterized gates are MLX-native compatible with mx.vmap and mx.compile.

---

Development

git clone https://github.com/Sirius-Quantum/zilver
cd zilver
pip install -e ".[dev,network]"
pytest tests/   # 540 tests

---

License

Apache 2.0. See LICENSE.

---

Manifesto

We believe quantum compute should be open, distributed, and owned by the people who run it.

Read the Sirius Quantum Manifesto