Simulate quantum computations on classical hardware using PyTorch. It supports statevector simulation and pulse simulation on GPUs. It can scale up to the simulation of 30+ qubits with multiple GPUs.
Researchers on quantum algorithm design, parameterized quantum circuit training, quantum optimal control, quantum machine learning, quantum neural networks.
Dynamic computation graph, automatic gradient computation, fast GPU support, batch model tersorized processing.
- v0.1.8 Available!
- Check the dev branch for new latest features on quantum layers and quantum algorithms.
- Join our Slack for real time support!
- Welcome to contribute! Please contact us or post in the Github Issues if you want to have new examples implemented by TorchQuantum or any other questions.
- Qmlsys website goes online: qmlsys.mit.edu and torchquantum.org
- Easy construction and simulation of quantum circuits in PyTorch
- Dynamic computation graph for easy debugging
- Gradient support via autograd
- Batch mode inference and training on CPU/GPU
- Easy deployment on real quantum devices such as IBMQ
- Easy hybrid classical-quantum model construction
- (coming soon) pulse-level simulation
git clone https://github.com/mit-han-lab/torchquantum.git
cd torchquantum
pip install --editable .import torchquantum as tq
import torchquantum.functional as tqf
qdev = tq.QuantumDevice(n_wires=2, bsz=5, device="cpu", record_op=True) # use device='cuda' for GPU
# use qdev.op
qdev.h(wires=0)
qdev.cnot(wires=[0, 1])
# use tqf
tqf.h(qdev, wires=1)
tqf.x(qdev, wires=1)
# use tq.Operator
op = tq.RX(has_params=True, trainable=True, init_params=0.5)
op(qdev, wires=0)
# print the current state (dynamic computation graph supported)
print(qdev)
# obtain the qasm string
from torchquantum.plugin import op_history2qasm
print(op_history2qasm(qdev.n_wires, qdev.op_history))
# measure the state on z basis
print(tq.measure(qdev, n_shots=1024))
# obtain the expval on a observable by stochastic sampling (doable on simulator and real quantum hardware)
from torchquantum.measurement import expval_joint_sampling
expval_sampling = expval_joint_sampling(qdev, 'ZX', n_shots=1024)
print(expval_sampling)
# obtain the expval on a observable by analytical computation (only doable on classical simulator)
from torchquantum.measurement import expval_joint_analytical
expval = expval_joint_analytical(qdev, 'ZX')
print(expval)
# obtain gradients of expval w.r.t. trainable parameters
expval[0].backward()
print(op.params.grad)
# Apply gates to qdev with tq.QuantumModule
ops = [
{'name': 'hadamard', 'wires': 0},
{'name': 'cnot', 'wires': [0, 1]},
{'name': 'rx', 'wires': 0, 'params': 0.5, 'trainable': True},
{'name': 'u3', 'wires': 0, 'params': [0.1, 0.2, 0.3], 'trainable': True},
{'name': 'h', 'wires': 1, 'inverse': True}
]
qmodule = tq.QuantumModule.from_op_history(ops)
qmodule(qdev)We also prepare many example and tutorials using TorchQuantum.
For beginning level, you may check QNN for MNIST, Quantum Convolution (Quanvolution) and Quantum Kernel Method, and Quantum Regression.
For intermediate level, you may check Amplitude Encoding for MNIST, Clifford gate QNN, Save and Load QNN models, PauliSum Operation, How to convert tq to Qiskit.
For expert, you may check Parameter Shift on-chip Training, VQA Gradient Pruning, VQE, VQA for State Prepration, QAOA (Quantum Approximate Optimization Algorithm).
Construct parameterized quantum circuit models as simple as constructing a normal pytorch model.
import torch.nn as nn
import torch.nn.functional as F
import torchquantum as tq
import torchquantum.functional as tqf
class QFCModel(nn.Module):
def __init__(self):
super().__init__()
self.n_wires = 4
self.measure = tq.MeasureAll(tq.PauliZ)
self.encoder_gates = [tqf.rx] * 4 + [tqf.ry] * 4 + \
[tqf.rz] * 4 + [tqf.rx] * 4
self.rx0 = tq.RX(has_params=True, trainable=True)
self.ry0 = tq.RY(has_params=True, trainable=True)
self.rz0 = tq.RZ(has_params=True, trainable=True)
self.crx0 = tq.CRX(has_params=True, trainable=True)
def forward(self, x):
bsz = x.shape[0]
# down-sample the image
x = F.avg_pool2d(x, 6).view(bsz, 16)
# create a quantum device to run the gates
qdev = tq.QuantumDevice(n_wires=self.n_wires, bsz=bsz, device=x.device)
# encode the classical image to quantum domain
for k, gate in enumerate(self.encoder_gates):
gate(qdev, wires=k % self.n_wires, params=x[:, k])
# add some trainable gates (need to instantiate ahead of time)
self.rx0(qdev, wires=0)
self.ry0(qdev, wires=1)
self.rz0(qdev, wires=3)
self.crx0(qdev, wires=[0, 2])
# add some more non-parameterized gates (add on-the-fly)
qdev.h(wires=3)
qdev.sx(wires=2)
qdev.cnot(wires=[3, 0])
qdev.qubitunitary(wires=[1, 2], params=[[1, 0, 0, 0],
[0, 1, 0, 0],
[0, 0, 0, 1j],
[0, 0, -1j, 0]])
# perform measurement to get expectations (back to classical domain)
x = self.measure(qdev).reshape(bsz, 2, 2)
# classification
x = x.sum(-1).squeeze()
x = F.log_softmax(x, dim=1)
return xTrain a quantum circuit to perform VQE task. Quito quantum computer as in simple_vqe.py script:
cd examples/vqe
python vqe.pyTrain a quantum circuit to perform MNIST classification task and deploy on the real IBM Quito quantum computer as in mnist_example.py script:
cd examples/mnist
python mnist.py| File | Description |
|---|---|
| devices.py | QuantumDevice class which stores the statevector |
| encoding.py | Encoding layers to encode classical values to quantum domain |
| functional.py | Quantum gate functions |
| operators.py | Quantum gate classes |
| layers.py | Layer templates such as RandomLayer |
| measure.py | Measurement of quantum states to get classical values |
| graph.py | Quantum gate graph used in static mode |
| super_layer.py | Layer templates for SuperCircuits |
| plugins/qiskit* | Convertors and processors for easy deployment on IBMQ |
| examples/ | More examples for training QML and VQE models |
torchquantum uses pre-commit hooks to ensure Python style consistency and prevent common mistakes in its codebase.
To enable it pre-commit hooks please reproduce:
pip install pre-commit
pre-commit install- [HPCA'22] Wang et al., "QuantumNAS: Noise-Adaptive Search for Robust Quantum Circuits"
- [DAC'22] Wang et al., "QuantumNAT: Quantum Noise-Aware Training with Noise Injection, Quantization and Normalization"
- [DAC'22] Wang et al., "QOC: Quantum On-Chip Training with Parameter Shift and Gradient Pruning"
- [QCE'22] Liang et al., "Variational Quantum Pulse Learning"
- [ICCAD'22] Hu et al., "Quantum Neural Network Compression"
- [ICCAD'22] Wang et al., "QuEst: Graph Transformer for Quantum Circuit Reliability Estimation"
- [ICML Workshop] Yun et al., "Slimmable Quantum Federated Learning"
- [IEEE ICDCS] Yun et al., "Quantum Multi-Agent Reinforcement Learning via Variational Quantum Circuit Design"
- [QCE'23] Zhan et al., "Quantum Sensor Network Algorithms for Transmitter Localization"
Manuscripts
- Yun et al., "Projection Valued Measure-based Quantum Machine Learning for Multi-Class Classification"
- Baek et al., "3D Scalable Quantum Convolutional Neural Networks for Point Cloud Data Processing in Classification Applications"
- Baek et al., "Scalable Quantum Convolutional Neural Networks"
- Yun et al., "Quantum Multi-Agent Meta Reinforcement Learning"
- 3.9 >= Python >= 3.7 (Python 3.10 may have the
concurrentpackage issue for Qiskit) - PyTorch >= 1.8.0
- configargparse >= 0.14
- GPU model training requires NVIDIA GPUs
TorchQuantum Forum
Hanrui Wang hanrui@mit.edu
Jiannan Cao, Jessica Ding, Jiai Gu, Song Han, Zhirui Hu, Zirui Li, Zhiding Liang, Pengyu Liu, Yilian Liu, Mohammadreza Tavasoli, Hanrui Wang, Zhepeng Wang, Zhuoyang Ye
@inproceedings{hanruiwang2022quantumnas,
title = {Quantumnas: Noise-adaptive search for robust quantum circuits},
author = {Wang, Hanrui and Ding, Yongshan and Gu, Jiaqi and Li, Zirui and Lin, Yujun and Pan, David Z and Chong, Frederic T and Han, Song},
booktitle = {The 28th IEEE International Symposium on High-Performance Computer Architecture (HPCA-28)},
year = {2022}
}