Switching the Execution Backend

In qiboml the backend is the object in charge of, both, executing the quantum circuit in the first place and handling any side processing needed for training and evaluating your model (e.g. calculation of probabilities, sampling, etc…).

qiboml provides itself three different backends for computation that are compliant with the qibo standards: qiboml.backends.JaxBackend, qiboml.backends.PyTorchBackend and qiboml.backends.TensorflowBackend.

Any one of them can be set up by using the usual set_backend method from qibo:

from qibo import set_backend

# set jax
set_backend("qiboml", platform="jax")
# set torch
set_backend("qiboml", platform="pytorch")
# set tensorflow
set_backend("qiboml", platform="tensorflow")

this sets the backend globally, which means it will be automatically used everywhere it is needed.

Alternatively, you can also directly construct an instance of the backend you wish to use, remember, however, to explicitely set it in all the pieces that require a backend specification (e.g. the decoding layer):

from qiboml.backends import PyTorchBackend, TensorflowBackend
from qiboml.models.decoding import Probabilities

# construct a local istance of the backend
backend = PyTorchBackend()
# assign it to the decoding layer upon initialization
decoding = Probabilities(nqubits=3, backend=backend)
# you can even switch it after initialization,
# however, pay always attention to consistency!
tf_backend = TensorflowBackend()
decoding.set_backend(tf_backend)

Note that three backends mentioned above are just those that qiboml provides out of the box, but actually any qibo-compatible backend can be used. In particular, for istance, in case you wished to try out your model on a real quantum hardware chip, you could just set up qibolab for running and not worry about anything else.