tf.keras.Variable

Represents a backend-agnostic variable in Keras.

tf.keras.Variable(
    initializer,
    shape=None,
    dtype=None,
    trainable=True,
    autocast=True,
    aggregation='none',
    synchronization='auto',
    name=None,
    layout=None,
    **kwargs
)

A Variable acts as a container for state. It holds a tensor value and can be updated. With the JAX backend, variables are used to implement "functionalization", the pattern of lifting stateful operations out of a piece of computation to turn it into a stateless function.

initializer Initial value or callable for initialization. If a callable is used, it should take the arguments shape and dtype.
shape Optional. Tuple for the variable's shape. Required if initializer is a callable.
dtype Optional. Data type of the variable. Defaults to the global float dtype type ("float32" if never configured).
trainable Optional. Boolean indicating if variable is trainable. Defaults to True.
autocast Optional. Boolean indicating whether the variable supports autocasting. If True, the layer may first convert the variable to the compute data type when accessed. Defaults to True.
aggregation Optional string, one of None, "none", "mean", "sum" or "only_first_replica" specifying how a distributed variable will be aggregated. This serves as a semantic annotation, to be taken into account by downstream backends or users. Defaults to "none".
name Optional. A unique name for the variable. Automatically generated if not set.
layout Optional. Sharding layout for the variable. Can be a keras.distribution.TensorLayout or a backend-specific layout.
kwargs Additional backend-specific keyword arguments.

Examples:

Initializing a Variable with a NumPy array:

import numpy as np
import keras
initial_array = np.ones((3, 3))
variable_from_array = keras.Variable(initializer=initial_array)

Using a Keras initializer to create a Variable:

from keras.src.initializers import Ones
variable_from_initializer = keras.Variable(
    initializer=Ones(), shape=(3, 3), dtype="float32"
)

Updating the value of a Variable:

new_value = np.zeros((3, 3), dtype="float32")
variable_from_array.assign(new_value)

Marking a Variable as non-trainable:

non_trainable_variable = keras.Variable(
    initializer=np.ones((3, 3), dtype="float32"), trainable=False
)

shape The shape of the variable (tuple of integers).
ndim The number of dimensions of the variable (integer).
dtype The data type of the variable (string).
trainable Whether the variable is trainable (boolean).
autocast Whether the variable supports autocasting (boolean).
aggregation How a distributed variable will be aggregated (string).
value The current value of the variable (NumPy array or tensor).
name The name of the variable (string).
path The path of the variable within the Keras model or layer (string).
constraint

handle

overwrite_with_gradient Whether this variable should be overwritten by the gradient.

This property is designed for a special case where we want to overwrite the variable directly with its computed gradient. For example, in float8 training, new scale and amax_history are computed as gradients, and we want to overwrite them directly instead of following the typical procedure such as gradient descent with a learning rate, gradient clipping and weight decaying.

regularizer

synchronization The strategy for synchronizing this variable.

Methods

assign

View source

assign(
    value
)

assign_add

View source

assign_add(
    value
)

assign_sub

View source

assign_sub(
    value
)

numpy

View source

numpy()

__abs__

View source

__abs__()

__add__

View source

__add__(
    other
)

__and__

View source

__and__(
    other
)

__array__

View source

__array__(
    dtype=None
)

__bool__

View source

__bool__()

__eq__

View source

__eq__(
    other
)

Return self==value.

__floordiv__

View source

__floordiv__(
    other
)

__ge__

View source

__ge__(
    other
)

Return self>=value.

__getitem__

View source

__getitem__(
    idx
)

__gt__

View source

__gt__(
    other
)

Return self>value.

__invert__

View source

__invert__()

__le__

View source

__le__(
    other
)

Return self<=value.

__lt__

View source

__lt__(
    other
)

Return self<value.

__matmul__

View source

__matmul__(
    other
)

__mod__

View source

__mod__(
    other
)

__mul__

View source

__mul__(
    other
)

__ne__

View source

__ne__(
    other
)

Return self!=value.

__neg__

View source

__neg__()

__or__

View source

__or__(
    other
)

Return self|value.

__pos__

View source

__pos__()

__pow__

View source

__pow__(
    other
)

__radd__

View source

__radd__(
    other
)

__rand__

View source

__rand__(
    other
)

__rfloordiv__

View source

__rfloordiv__(
    other
)

__rmatmul__

View source

__rmatmul__(
    other
)

__rmod__

View source

__rmod__(
    other
)

__rmul__

View source

__rmul__(
    other
)

__ror__

View source

__ror__(
    other
)

Return value|self.

__rpow__

View source

__rpow__(
    other
)

__rsub__

View source

__rsub__(
    other
)

__rtruediv__

View source

__rtruediv__(
    other
)

__rxor__

View source

__rxor__(
    other
)

__sub__

View source

__sub__(
    other
)

__truediv__

View source

__truediv__(
    other
)

__xor__

View source

__xor__(
    other
)