tf.keras.layers.EinsumDense

A layer that uses einsum as the backing computation.

Inherits From: Layer, Operation

tf.keras.layers.EinsumDense(
    equation,
    output_shape,
    activation=None,
    bias_axes=None,
    kernel_initializer='glorot_uniform',
    bias_initializer='zeros',
    kernel_regularizer=None,
    bias_regularizer=None,
    kernel_constraint=None,
    bias_constraint=None,
    lora_rank=None,
    lora_alpha=None,
    gptq_unpacked_column_size=None,
    quantization_config=None,
    **kwargs
)

This layer can perform einsum calculations of arbitrary dimensionality.

equation An equation describing the einsum to perform. This equation must be a valid einsum string of the form ab,bc->ac, ...ab,bc->...ac, or ab...,bc->ac... where 'ab', 'bc', and 'ac' can be any valid einsum axis expression sequence.
output_shape The expected shape of the output tensor (excluding the batch dimension and any dimensions represented by ellipses). You can specify None for any dimension that is unknown or can be inferred from the input shape.
activation Activation function to use. If you don't specify anything, no activation is applied (that is, a "linear" activation: a(x) = x).
bias_axes A string containing the output dimension(s) to apply a bias to. Each character in the bias_axes string should correspond to a character in the output portion of the equation string.
kernel_initializer Initializer for the kernel weights matrix.
bias_initializer Initializer for the bias vector.
kernel_regularizer Regularizer function applied to the kernel weights matrix.
bias_regularizer Regularizer function applied to the bias vector.
kernel_constraint Constraint function applied to the kernel weights matrix.
bias_constraint Constraint function applied to the bias vector.
lora_rank Optional integer. If set, the layer's forward pass will implement LoRA (Low-Rank Adaptation) with the provided rank. LoRA sets the layer's kernel to non-trainable and replaces it with a delta over the original kernel, obtained via multiplying two lower-rank trainable matrices (the factorization happens on the last dimension). This can be useful to reduce the computation cost of fine-tuning large dense layers. You can also enable LoRA on an existing EinsumDense layer by calling layer.enable_lora(rank). lora_alpha: Optional integer. If set, this parameter scales the low-rank adaptation delta (computed as the product of two lower-rank trainable matrices) during the forward pass. The delta is scaled by lora_alpha / lora_rank, allowing you to fine-tune the strength of the LoRA adjustment independently of lora_rank.
**kwargs Base layer keyword arguments, such as name and dtype.

Examples:

Biased dense layer with einsums

This example shows how to instantiate a standard Keras dense layer using einsum operations. This example is equivalent to keras.layers.Dense(64, use_bias=True).

layer = keras.layers.EinsumDense("ab,bc->ac",
                                      output_shape=64,
                                      bias_axes="c")
input_tensor = keras.Input(shape=[32])
output_tensor = layer(input_tensor)
output_tensor.shape
(None, 64)

Applying a dense layer to a sequence

This example shows how to instantiate a layer that applies the same dense operation to every element in a sequence. Here, the output_shape has two values (since there are two non-batch dimensions in the output); the first dimension in the output_shape is None, because the sequence dimension b has an unknown shape.

layer = keras.layers.EinsumDense("abc,cd->abd",
                                      output_shape=(None, 64),
                                      bias_axes="d")
input_tensor = keras.Input(shape=[32, 128])
output_tensor = layer(input_tensor)
output_tensor.shape
(None, 32, 64)

Applying a dense layer to a sequence using ellipses

This example shows how to instantiate a layer that applies the same dense operation to every element in a sequence, but uses the ellipsis notation instead of specifying the batch and sequence dimensions.

Because we are using ellipsis notation and have specified only one axis, the output_shape arg is a single value. When instantiated in this way, the layer can handle any number of sequence dimensions - including the case where no sequence dimension exists.

layer = keras.layers.EinsumDense("...x,xy->...y",
                                      output_shape=64,
                                      bias_axes="y")
input_tensor = keras.Input(shape=[32, 128])
output_tensor = layer(input_tensor)
output_tensor.shape
(None, 32, 64)

input Retrieves the input tensor(s) of a symbolic operation.

Only returns the tensor(s) corresponding to the first time the operation was called.

kernel

output Retrieves the output tensor(s) of a layer.

Only returns the tensor(s) corresponding to the first time the operation was called.

variable_serialization_spec Returns a dict mapping quantization modes to variable names in order.

This spec is used by save_own_variables and load_own_variables to determine the correct ordering of variables during serialization for each quantization mode. None means no quantization.

Methods

enable_lora

View source

enable_lora(
    rank,
    lora_alpha=None,
    a_initializer='he_uniform',
    b_initializer='zeros'
)

from_config

View source

@classmethod
from_config(
    config
)

Creates an operation from its config.

This method is the reverse of get_config, capable of instantiating the same operation from the config dictionary.

if "dtype" in config and isinstance(config["dtype"], dict):
    policy = dtype_policies.deserialize(config["dtype"])

Args
config A Python dictionary, typically the output of get_config.

Returns
An operation instance.