Skip to main content

shim

Backend-agnostisc shims.

Classes​

BackendTensorShim​

class BackendTensorShim():

A shim for handling tensors of a particular type.

An abstract class representing a shim/bridge for tensor handling in a particular backend.

Ancestors​

Subclasses​

Static methods​

clamp_params​

def clamp_params(    p: _TensorLike, prime_q: int, precision: int, num_workers: int,) ‑> _TensorLike:

Clamps the parameter of a given tensor.

Constrains the parameters for secure sharing to be within the required range for secure sharing. Used only when steps_between_parameter_updates is 1.

Arguments

  • p: The tensor to be constrained.
  • prime_q: The prime use for secret aggregation.
  • precision: The precision used for secret aggregation.
  • num_workers: The number of workers taking part in the secure aggregation.

Returns The clamped parameters.

is_tensor​

def is_tensor(p: Any) ‑> bool:

Checks if the argument is a tensor.

Arguments

  • p: The argument to check.

Returns True if the supplied argument is a tensor according to this model's backend, False otherwise.

to_list​

def to_list(p: np.ndarray | _TensorLike) ‑> list[float]:

Converts the supplied tensor or numpy array to a list and returns it.

Arguments

  • p: The tensor or numpy array to convert to a list.

Returns A list.

to_numpy​

def to_numpy(t: _TensorLike | list[float]) ‑> np.ndarray:

Converts a tensor into a numpy array and returns it.

Arguments

  • t: The tensor or list to convert.

Returns A numpy array.

to_tensor​

def to_tensor(p: Any, **kwargs: Any) ‑> _TensorLike:

Converts the supplied argument to a tensor (if possible) and returns it.

Arguments

  • p: The argument to convert to a tensor.
  • **kwargs: Additional keyword arguments to pass to the tensor constructor.

Returns A tensor.