multi_dispatch(argnum=None, tensor_list=None)[source]

Decorater to dispatch arguments handled by the interface.

This helps simplify definitions of new functions inside PennyLane. We can decorate the function, indicating the arguments that are tensors handled by the interface:

>>> @qml.math.multi_dispatch(argnum=[0, 1])
... def some_function(tensor1, tensor2, option, like):
...     # the interface string is stored in `like`.
...     ...
  • argnum (list[int]) – A list of integers indicating indicating the indices to dispatch (i.e., the arguments that are tensors handled by an interface). If None, dispatch over all arguments.

  • tensor_lists (list[int]) – a list of integers indicating which indices in argnum are expected to be lists of tensors. If an argument marked as tensor list is not a tuple or list, it is treated as if it was not marked as tensor list. If None, this option is ignored.


A wrapped version of the function, which will automatically attempt to dispatch to the correct autodifferentiation framework for the requested arguments. Note that the like argument will be optional, but can be provided if an explicit override is needed.

Return type


See also



This decorator makes the interface argument “like” optional as it utilizes the utility function _multi_dispatch to automatically detect the appropriate interface based on the tensor types.


We can redefine external functions to be suitable for PennyLane. Here, we redefine Autoray’s stack function.

>>> stack = multi_dispatch(argnum=0, tensor_list=0)(autoray.numpy.stack)

We can also use the multi_dispatch decorator to dispatch arguments of more more elaborate custom functions. Here is an example of a custom_function that computes \(c \\sum_i (v_i)^T v_i\), where \(v_i\) are vectors in values and \(c\) is a fixed coefficient. Note how argnum=0 only points to the first argument values, how tensor_list=0 indicates that said first argument is a list of vectors, and that coefficient is not dispatched.

>>> @math.multi_dispatch(argnum=0, tensor_list=0)
>>> def custom_function(values, like, coefficient=10):
>>>     # values is a list of vectors
>>>     # like can force the interface (optional)
>>>     if like == "tensorflow":
>>>         # add interface-specific handling if necessary
>>>     return coefficient * np.sum([math.dot(v,v) for v in values])

We can then run

>>> values = [np.array([1, 2, 3]) for _ in range(5)]
>>> custom_function(values)