Skip to main content

TensorFlow Quantum

What is it​

TensorFlow Quantum (TFQ) is an open-source library from Google for building hybrid quantum-classical machine learning models. It bridges two of Google's projects: quantum circuits are described with Cirq (see its page), and those circuits are wrapped as Keras layers so they train inside ordinary TensorFlow models. Its focus is research on quantum neural networks and variational QML, especially over batches of parameterized circuits.

TFQ's central idea is the quantum data layer: circuits become tensors, parameterized circuits become differentiable Keras layers, and you optimize them with standard TensorFlow optimizers and gradients.

Maintenance status​

Be realistic here: TensorFlow Quantum is largely in maintenance mode. Development has slowed considerably, and it is pinned to specific (older) versions of TensorFlow and Cirq, which can make installation finicky. It remains useful for reproducing TFQ-based papers and for teams already standardized on the Keras ecosystem, but for new QML projects most practitioners now reach for PennyLane, which is more actively developed and framework-agnostic. Choose TFQ deliberately, not by default.

Install​

pip install tensorflow-quantum

Because TFQ requires compatible, often older, versions of tensorflow and cirq, install it in a fresh virtual environment and consult the version matrix in the official docs before pinning your dependencies.

Hello Quantum: a parameterized quantum circuit (PQC) layer​

A minimal hybrid model: a parameterized Cirq circuit wrapped as a trainable Keras PQC layer.

import cirq
import sympy
import tensorflow as tf
import tensorflow_quantum as tfq

# A single qubit with a trainable rotation angle (symbol)
qubit = cirq.GridQubit(0, 0)
theta = sympy.Symbol("theta")

model_circuit = cirq.Circuit(cirq.rx(theta)(qubit))
readout = cirq.Z(qubit)   # measure <Z> as the layer output

# Wrap the parameterized circuit as a trainable Keras layer
quantum_layer = tfq.layers.PQC(model_circuit, readout)

# Feed an (empty) input circuit; PQC outputs the expectation value
input_circuit = tfq.convert_to_tensor([cirq.Circuit()])
print(quantum_layer(input_circuit))
# A trainable tensor: <Z> for the current value of theta

The PQC layer exposes theta as a trainable weight, so you can drop it into a tf.keras.Model, define a loss, and train it with model.fit exactly like a classical layer.

Strengths & use cases​

  • Keras-native QML. Quantum circuits behave as standard Keras layers, so they compose with classical neural networks.
  • Batched circuit execution. Designed to evaluate and differentiate large batches of parameterized circuits efficiently.
  • Research reproducibility. The reference implementation for a body of Google QML research.

Ecosystem​

  • Cirq — the circuit layer underneath TFQ.
  • TensorFlow / Keras — the classical ML stack TFQ plugs into.
  • tfq.layers — PQC, ControlledPQC, Expectation, and Sample layers.
  • For alternatives with broader, more current support, compare against PennyLane.

Resources​

To experiment with hybrid models, see the labs; to plan a QML learning path, see the roadmaps.