attentionLayer

Dot-product attention layer

Since R2024a

Description

A dot-product attention layer focuses on parts of the input using weighted multiplication operations.

Creation

Syntax

layer = attentionLayer(numHeads)

layer = attentionLayer(numHeads,Name=Value)

Description

layer = attentionLayer(numHeads) creates a dot-product attention layer and sets the NumHeads property.

layer = attentionLayer(numHeads,Name=Value) sets additional options using one or more name-value arguments. For example, attentionLayer(3,DropoutProbability=0.1) creates an attention layer with 3 heads and sets the DropoutProbability property to 0.1.

Input Arguments

numHeads — Number of attention heads

positive integer

Number of attention heads, specified as a positive integer.

Each head performs a separate linear transformation of the input and computes attention weights independently. The layer uses these attention weights to compute a weighted sum of the input representations, generating a context vector. Increasing the number of heads lets the model capture different types of dependencies and attend to different parts of the input simultaneously. Reducing the number of heads can lower the computational cost of the layer.

This argument sets the NumHeads property.

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64

Name-Value Arguments

Specify optional pairs of arguments as Name1=Value1,...,NameN=ValueN, where Name is the argument name and Value is the corresponding value. Name-value arguments must appear after other arguments, but the order of the pairs does not matter.

Example: attentionLayer(3,DropoutProbability=0.1) creates an attention layer with 3 heads and sets the DropoutProbability property to 0.1.

NumQueryGroups — Number of query groups

"num-heads" (default) | positive integer

Since R2026a

Number of query groups (equivalent to the number of key-value heads), specified as one of these values:

• "num-heads" — Use the numHeads argument value.

• Positive integer — Use the specified number of query groups. This value must divide the numHeads argument value.

The value of NumQueryGroups specifies the type of attention operation:

• For multihead attention, set NumQueryGroups to numHeads.

• For multiquery attention (MQA), set NumQueryGroups to 1.

• For grouped-query attention (GQA), set NumQueryGroups to a positive integer between 1 and numHeads.

Using multiquery attention and grouped-query attention in conjunction with the KeyState and ValueState properties can reduce memory and computation time for large inputs.

When the number of query groups is greater than 1, the operation creates groups of query channels-per-head, and applies the attention operation within each group.

For example, for six heads with three query groups, the operation splits the query channels into the heads (h₁, …, h₆) and then creates the groups of heads g₁=(h₁,h₂), g₂=(h₃,h₄), and g₃=(h₅,h₆). The operation also splits the key and value channels into the heads g₁, g₂, and g₃.

When the number of query groups matches the number of heads, the groups have one head each and is equivalent to multihead attention. When the number of query groups is 1, then all the heads are in the same group and is equivalent to multiquery attention.

This argument sets the NumQueryGroups property.

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64 | char | string

Scale — Multiplicative factor for scaling dot product of queries and keys

"auto" | numeric scalar

Multiplicative factor for scaling dot product of queries and keys, specified as one of these values:

• "auto" — Multiply the dot product by 1/sqrt(D), where D is the number of channels of the keys divided by NumHeads.

• Numeric scalar — Multiply the dot product by the specified scalar.

This argument sets the Scale property.

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64 | char | string | cell

HasPaddingMaskInput — Flag indicating whether layer has mask input

0 (false) (default) | 1 (true)

Flag indicating whether the layer has an input that represents the padding mask, specified as 0 (false) or 1 (true).

If the HasPaddingMaskInput property is 0 (false), then the layer has three inputs with the names "query", "key", and "value", which correspond to the input queries, keys, and values, respectively. In this case, the layer treats all elements as data.

If the HasPaddingMaskInput property is 1 (true), then the layer has an additional input with the name "mask", which corresponds to the padding mask. In this case, the padding mask is an array of ones and zeros. The layer uses or ignores elements of the queries, keys, and values when the corresponding element in the mask is one or zero, respectively.

The dimension labels of the padding mask must match the dimension labels of the keys, ignoring any "C" (channel) and "U" (unspecified) dimensions. (since R2026a).

Before R2026a: The format of the padding mask must match that of the input keys.

The size of the "S" (spatial) or "T" (time) dimension of the padding mask must match the sum of the corresponding dimension in the keys and values and the size of the second dimension of the KeyState and ValueState properties. The size of the "B" (batch) dimension of the padding mask must match the size of the corresponding dimensions in the keys and values.

The padding mask can have any number of channels. The software uses only the values in the first channel to indicate padding values.

This argument sets the HasPaddingMaskInput property.

HasScoresOutput — Flag indicating whether layer has scores output

0 (false) (default) | 1 (true)

Flag indicating whether the layer has an output that represents the scores (also known as the attention weights), specified as 0 (false) or 1 (true).

If the HasScoresOutput property is 0 (false), then the layer has one output with the name "out", which corresponds to the output data.

If the HasScoresOutput property is 1 (true), then the layer has two inputs with the names "out" and "scores", which correspond to the output data and the attention scores, respectively.

This argument sets the HasScoresOutput property.

AttentionMask — Attention mask

"none" (default) | "causal" | numeric array | logical array

Attention mask indicating which elements to include when applying the attention operation, specified as one of these values:

• "none" — Do not prevent attention to elements with respect to their positions. If AttentionMask is "none", then the software prevents attention using only the padding mask.

• "causal" — Prevent elements in position m in the "S" (spatial) or "T" (time) dimension of the input queries from providing attention to the elements in positions n, where n is greater than m in the corresponding dimension of the input keys and values. Use this option for autoregressive models.

• Logical or numeric array — Prevent attention to elements of the input keys and values when the corresponding element in the specified array is 0. The specified array must be an N_k-by-N_q matrix or a N_k-by-N_q-by-numObservations array, N_k is the size of the "S" (spatial) or "T" (time) dimension of the input keys, N_q is the size of the corresponding dimension of the input queries, and numObservations is the size of the "B" dimension in the input queries.

This argument sets the AttentionMask property.

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64 | logical | char | string

DropoutProbability — Dropout probability for attention scores

0 (default) | scalar in the range [0, 1)

Probability of dropping out attention scores, specified as a scalar in the range [0, 1).

During training, the software randomly sets values in the attention scores to zero using the specified probability. These dropouts can encourage the model to learn more robust and generalizable representations by preventing it from relying too heavily on specific dependencies.

This argument sets the DropoutProbability property.

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64

KeyState — Key state

[] (default) | numeric array

Since R2026a

Key state, specified as a c-by-x-by-b numeric array or [], where c is the size of the "C" (channel) dimension of the key input, b is the size of the "B" (batch) dimension of the input, and x can be any value.

When you update the state of a network containing a AttentionLayer object with the AttentionMask property set to "causal", the KeyState property contains keys used in previous forward passes of the network. If AttentionMask is "none", then the KeyState property must be empty.

This argument sets the KeyState property.

After you set this property manually, calls to the resetState function set the value state to this value.

Data Types: single | double

ValueState — Value state

[] (default) | numeric array

Since R2026a

Value state, specified as a c-by-x-by-b numeric array or [], where c is the size of the "C" (channel) dimension of the value input, b is the size of the "B" (batch) dimension of the input, and x can be any value.

When you update the state of a network containing a AttentionLayer object with the AttentionMask property set to "causal", the ValueState property contains values used in previous forward passes of the network. If AttentionMask is "none", then the ValueState property must be empty.

This argument sets the ValueState property.

After you set this property manually, calls to the resetState function set the value state to this value.

Data Types: single | double

Name — Layer name

"" (default) | character vector | string scalar

Layer name, specified as a character vector or a string scalar. For Layer array input, the trainnet and dlnetwork functions automatically assign names to unnamed layers.

Data Types: char | string

Properties

Attention

NumHeads — Number of attention heads

positive integer

This property is read-only after object creation. To set this property, use the corresponding positional input argument when you create the AttentionLayer object.

Number of attention heads, specified as a positive integer.

Each head performs a separate linear transformation of the input and computes attention weights independently. The layer uses these attention weights to compute a weighted sum of the input representations, generating a context vector. Increasing the number of heads lets the model capture different types of dependencies and attend to different parts of the input simultaneously. Reducing the number of heads can lower the computational cost of the layer.

Data Types: double

NumQueryGroups — Number of query groups

positive integer

Since R2026a

This property is read-only after object creation. To set this property, use the corresponding name-value argument when you create the AttentionLayer object.

Number of query groups (equivalent to the number of key-value heads), specified as one of these values:

• "num-heads" — Use the numHeads argument value.

• Positive integer — Use the specified number of query groups. This value must divide the numHeads argument value.

The value of NumQueryGroups specifies the type of attention operation:

• For multihead attention, set NumQueryGroups to numHeads.

• For multiquery attention (MQA), set NumQueryGroups to 1.

• For grouped-query attention (GQA), set NumQueryGroups to a positive integer between 1 and numHeads.

Using multiquery attention and grouped-query attention in conjunction with the KeyState and ValueState properties can reduce memory and computation time for large inputs.

When the number of query groups is greater than 1, the operation creates groups of query channels-per-head, and applies the attention operation within each group.

For example, for six heads with three query groups, the operation splits the query channels into the heads (h₁, …, h₆) and then creates the groups of heads g₁=(h₁,h₂), g₂=(h₃,h₄), and g₃=(h₅,h₆). The operation also splits the key and value channels into the heads g₁, g₂, and g₃.

When the number of query groups matches the number of heads, the groups have one head each and is equivalent to multihead attention. When the number of query groups is 1, then all the heads are in the same group and is equivalent to multiquery attention.

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64 | char | string

Scale — Multiplicative factor for scaling dot product of queries and keys

"auto" | numeric scalar

This property is read-only after object creation. To set this property, use the corresponding name-value argument when you create the AttentionLayer object.

Multiplicative factor for scaling dot product of queries and keys, specified as one of these values:

• "auto" — Multiply the dot product by 1/sqrt(D), where D is the number of channels of the keys divided by NumHeads.

• Numeric scalar — Multiply the dot product by the specified scalar.

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64 | char | string | cell

HasPaddingMaskInput — Flag indicating whether layer has mask input

0 (false) (default) | 1 (true)

This property is read-only after object creation. To set this property, use the corresponding name-value argument when you create the AttentionLayer object.

Flag indicating whether the layer has an input that represents the padding mask, specified as 0 (false) or 1 (true).

If the HasPaddingMaskInput property is 0 (false), then the layer has three inputs with the names "query", "key", and "value", which correspond to the input queries, keys, and values, respectively. In this case, the layer treats all elements as data.

If the HasPaddingMaskInput property is 1 (true), then the layer has an additional input with the name "mask", which corresponds to the padding mask. In this case, the padding mask is an array of ones and zeros. The layer uses or ignores elements of the queries, keys, and values when the corresponding element in the mask is one or zero, respectively.

The dimension labels of the padding mask must match the dimension labels of the keys, ignoring any "C" (channel) and "U" (unspecified) dimensions. (since R2026a).

Before R2026a: The format of the padding mask must match that of the input keys.

The size of the "S" (spatial) or "T" (time) dimension of the padding mask must match the sum of the corresponding dimension in the keys and values and the size of the second dimension of the KeyState and ValueState properties. The size of the "B" (batch) dimension of the padding mask must match the size of the corresponding dimensions in the keys and values.

The padding mask can have any number of channels. The software uses only the values in the first channel to indicate padding values.

HasScoresOutput — Flag indicating whether layer has scores output

0 (false) (default) | 1 (true)

This property is read-only after object creation. To set this property, use the corresponding name-value argument when you create the AttentionLayer object.

Flag indicating whether the layer has an output that represents the scores (also known as the attention weights), specified as 0 (false) or 1 (true).

If the HasScoresOutput property is 0 (false), then the layer has one output with the name "out", which corresponds to the output data.

If the HasScoresOutput property is 1 (true), then the layer has two inputs with the names "out" and "scores", which correspond to the output data and the attention scores, respectively.

AttentionMask — Attention mask

"none" (default) | "causal" | numeric array | logical array

Attention mask indicating which elements to include when applying the attention operation, specified as one of these values:

• "none" — Do not prevent attention to elements with respect to their positions. If AttentionMask is "none", then the software prevents attention using only the padding mask.

• "causal" — Prevent elements in position m in the "S" (spatial) or "T" (time) dimension of the input queries from providing attention to the elements in positions n, where n is greater than m in the corresponding dimension of the input keys and values. Use this option for autoregressive models.

• Logical or numeric array — Prevent attention to elements of the input keys and values when the corresponding element in the specified array is 0. The specified array must be an N_k-by-N_q matrix or a N_k-by-N_q-by-numObservations array, N_k is the size of the "S" (spatial) or "T" (time) dimension of the input keys, N_q is the size of the corresponding dimension of the input queries, and numObservations is the size of the "B" dimension in the input queries.

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64 | logical | char | string

DropoutProbability — Dropout probability for attention scores

0 (default) | scalar in the range [0, 1)

Probability of dropping out attention scores, specified as a scalar in the range [0, 1).

During training, the software randomly sets values in the attention scores to zero using the specified probability. These dropouts can encourage the model to learn more robust and generalizable representations by preventing it from relying too heavily on specific dependencies.

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64

State

KeyState — Key state

[] (default) | numeric array

Since R2026a

Key state, specified as a c-by-x-by-b numeric array or [], where c is the size of the "C" (channel) dimension of the key input, b is the size of the "B" (batch) dimension of the input, and x can be any value.

When you update the state of a network containing a AttentionLayer object with the AttentionMask property set to "causal", the KeyState property contains keys used in previous forward passes of the network. If AttentionMask is "none", then the KeyState property must be empty.

After you set this property manually, calls to the resetState function set the value state to this value.

Data Types: single | double

ValueState — Value state

[] (default) | numeric array

Since R2026a

Value state, specified as a c-by-x-by-b numeric array or [], where c is the size of the "C" (channel) dimension of the value input, b is the size of the "B" (batch) dimension of the input, and x can be any value.

When you update the state of a network containing a AttentionLayer object with the AttentionMask property set to "causal", the ValueState property contains values used in previous forward passes of the network. If AttentionMask is "none", then the ValueState property must be empty.

After you set this property manually, calls to the resetState function set the value state to this value.

Data Types: single | double

Layer

Name — Layer name

'' (default) | character vector

Layer name, specified as a character vector. For Layer array input, the trainnet and dlnetwork functions automatically assign names to unnamed layers.

Data Types: char

NumInputs — Number of inputs

3 (default) | 4

Number of inputs to the layer, returned as 3 or 4.

If the HasPaddingMaskInput property is 0 (false), then the layer has three inputs with the names "query", "key", and "value", which correspond to the input queries, keys, and values, respectively. In this case, the layer treats all elements as data.

If the HasPaddingMaskInput property is 1 (true), then the layer has an additional input with the name "mask", which corresponds to the padding mask. In this case, the padding mask is an array of ones and zeros. The layer uses or ignores elements of the queries, keys, and values when the corresponding element in the mask is one or zero, respectively.

Data Types: double

InputNames — Input names

["query" "key" "value"] (default) | ["query" "key" "value" "mask"]

Input names of the layer, returned as a cell array of character vectors.

If the HasPaddingMaskInput property is 0 (false), then the layer has three inputs with the names "query", "key", and "value", which correspond to the input queries, keys, and values, respectively. In this case, the layer treats all elements as data.

If the HasPaddingMaskInput property is 1 (true), then the layer has an additional input with the name "mask", which corresponds to the padding mask. In this case, the padding mask is an array of ones and zeros. The layer uses or ignores elements of the queries, keys, and values when the corresponding element in the mask is one or zero, respectively.

The AttentionLayer object stores this property as a cell array of character vectors.

NumOutputs — Number of outputs

Read-only: 1 (default) | 2

This property is read-only.

Number of outputs of the layer.

If the HasScoresOutput property is 0 (false), then the layer has one output with the name "out", which corresponds to the output data.

If the HasScoresOutput property is 1 (true), then the layer has two inputs with the names "out" and "scores", which correspond to the output data and the attention scores, respectively.

Data Types: double

OutputNames — Output names

Read-only: "out" (default) | ["out" "scores"]

This property is read-only.

Output names of the layer.

If the HasScoresOutput property is 0 (false), then the layer has one output with the name "out", which corresponds to the output data.

If the HasScoresOutput property is 1 (true), then the layer has two inputs with the names "out" and "scores", which correspond to the output data and the attention scores, respectively.

The AttentionLayer object stores this property as a cell array of character vectors.

Examples

Create Attention Layer

Create a dot-product attention layer with 10 heads.

layer = attentionLayer(10)

layer = 
  AttentionLayer with properties:

                   Name: ''
              NumInputs: 3
             InputNames: {'query'  'key'  'value'}
               NumHeads: 10
         NumQueryGroups: 10
                  Scale: 'auto'
          AttentionMask: 'none'
     DropoutProbability: 0
    HasPaddingMaskInput: 0
        HasScoresOutput: 0

   Learnable Parameters
    No properties.

   State Parameters
               KeyState: []
             ValueState: []

  Show all properties

Create Cross-Attention Neural Network

Create a simple neural network with cross-attention.

numChannels = 256;
numHeads = 8;

net = dlnetwork;

layers = [
    sequenceInputLayer(1,Name="query")
    fullyConnectedLayer(numChannels)
    attentionLayer(numHeads,Name="attention")
    fullyConnectedLayer(numChannels,Name="fc-out")];

net = addLayers(net,layers);

layers = [
    sequenceInputLayer(1, Name="key-value")
    fullyConnectedLayer(numChannels,Name="fc-key")];

net = addLayers(net,layers);
net = connectLayers(net,"fc-key","attention/key");

net = addLayers(net, fullyConnectedLayer(numChannels,Name="fc-value"));
net = connectLayers(net,"key-value","fc-value");
net = connectLayers(net,"fc-value","attention/value");

View the network in a plot.

figure
plot(net)

Algorithms

Dot-Product Attention

The attention operation focuses on parts of the input using weighted multiplication operations.

The single-head dot-product attention operation is given by

$$\text{attention}(Q,K,V)=\text{dropout}\left(\text{softmax}\left(\text{mask}\left(\lambda Q{K}^{\top },M\right)\right),p\right)V,$$

where:

• Q denotes the queries.

• K denotes the keys.

• V denotes the values.

• $$\lambda$$ denotes the scaling factor.

• M is a mask array of ones and zeros.

• p is the dropout probability.

The mask operation includes or excludes the values of the matrix multiplication by setting values of the input to $$-\infty$$ for zero-valued mask elements. The mask is the union of the padding and attention masks. The softmax function normalizes the value of the input data across the channel dimension such that it sums to one. The dropout operation sets elements to zero with probability p.

Multihead Attention

The multihead attention operation applies the attention operation across multiple heads. Each head uses its own learnable query, key, and value projection matrices.

The multihead attention operation for the queries Q, keys K, and values V is given by

$$\text{multiheadAttention}(Q,K,V)=\text{concatenate}({\text{head}}_1,\dots ,{\text{head}}_h)W^O,$$

where

• h is the number of heads.

• W^O is a learnable projection matrix for the output.

For the multihead attention operation, each learnable projection matrix for the queries, keys, and values are composed of concatenated matrices W_i, where i indexes over the heads.

The head operation is given by

$${\text{head}}_i=\text{attention}\left(Q{W}_i^Q,K{W}_i^K,V{W}_i^V\right),$$

where:

• i indexes over the heads.

• W^Q is a learnable projection matrix for the queries.

• W^K is a learnable projection matrix for the keys.

• W^V is a learnable projection matrix for the values.

Multiquery Attention

The multiquery attention (MQA) operation applies the attention operation across multiple heads. Each head uses its own learnable query projection matrix. The operation uses the same learnable key and value projection matrices across all heads.

The multiquery attention operation for the queries Q, keys K, and values V is given by

$$\text{multiqueryAttention}(Q,K,V)=\text{concatenate}({\text{head}}_1,\dots ,{\text{head}}_h)W^O,$$

where

• h is the number of heads.

• W^O is a learnable projection matrix for the output.

For the multiquery attention operation, only the learnable projection matrix for the queries is composed of concatenated matrices W_i, where i indexes over the heads.

For multiquery attention, the head operation is given by

$${\text{head}}_i=\text{attention}\left(Q{W}_i^Q,K{W}^K,V{W}^V\right),$$

where:

• i indexes over the heads.

• W^Q is a learnable projection matrix for the queries.

• W^K is a learnable projection matrix for the keys.

• W^V is a learnable projection matrix for the values.

Grouped-Query Attention

The grouped query attention (GQA) operation applies the attention operation across several heads. The operation partitions the heads into groups that use the same learnable query projection matrix. The operation uses the same learnable key and value projection matrices for each group of query heads.

The grouped-query attention operation for the queries Q, keys K, and values V is given by

$$\text{groupedQueryAttention}(Q,K,V)=\text{concatenate}({\text{head}}_1,\dots ,{\text{head}}_h)W^O,$$

where

• h is the number of heads.

• W^O is a learnable projection matrix for the output.

For the grouped query attention operation:

• The learnable projection matrix for the queries is composed of concatenated matrices W_i, where i indexes over the heads.

• The learnable projection matrices for the keys and values are composed of concatenated matrices W_j, where j indexes over the groups.

When the number of query groups is greater than 1, the operation creates groups of query channels-per-head, and applies the attention operation within each group.

For example, for six heads with three query groups, the operation splits the query channels into the heads (h₁, …, h₆) and then creates the groups of heads g₁=(h₁,h₂), g₂=(h₃,h₄), and g₃=(h₅,h₆). The operation also splits the key and value channels into the heads g₁, g₂, and g₃.

When the number of query groups matches the number of heads, the groups have one head each and is equivalent to multihead attention. When the number of query groups is 1, then all the heads are in the same group and is equivalent to multiquery attention.

The head operation is given by

$${\text{head}}_i=\text{attention}\left(Q{W}_i^Q,K{W}_{g(i)}^K,V{W}_{g(i)}^V\right),$$

where:

• i indexes over the heads.

• g(i) is the group number of head i.

• W^Q is a learnable projection matrix for the queries.

• W^K is a learnable projection matrix for the keys.

• W^V is a learnable projection matrix for the values.

Layer Input and Output Formats

Most layers in a layer array or layer graph pass data to subsequent layers as formatted dlarray objects. The format of a dlarray object is a string of characters in which each character describes the corresponding dimension of the data. The format consists of one or more of these characters:

• "S" — Spatial

• "C" — Channel

• "B" — Batch

• "T" — Time

• "U" — Unspecified

For example, you can describe 2-D image data that is represented as a 4-D array, where the first two dimensions correspond to the spatial dimensions of the images, the third dimension corresponds to the channels of the images, and the fourth dimension corresponds to the batch dimension, as having the format "SSCB" (spatial, spatial, channel, batch).

You can interact with these dlarray objects in automatic differentiation workflows, such as those for:

• developing a custom layer

• using a functionLayer object

• using the forward and predict functions with dlnetwork objects

This table shows the supported input formats of AttentionLayer objects and the corresponding output format. If the software passes the output of the layer to a custom layer that does not inherit from the nnet.layer.Formattable class, or to a FunctionLayer object with the Formattable property set to 0 (false), then the layer receives an unformatted dlarray object with dimensions ordered according to the formats in this table. The formats listed here are only a subset of the formats that the layer supports. The layer might support additional formats, such as formats with additional "S" (spatial) or "U" (unspecified) dimensions.

Query, Key, and Value Format	Output Format	Scores Output Format (When HasScoresOutput is 1 (true)
"CB" (channel, batch)	"CB" (channel, batch)	"UUUU" (unspecified, unspecified, unspecified, unspecified)
"SCB" (spatial, channel, batch)	"SCB" (spatial, channel, batch)	"UUUU" (unspecified, unspecified, unspecified, unspecified)
"CBT" (channel, batch, time)	"CBT" (channel, batch, time)	"UUUU" (unspecified, unspecified, unspecified, unspecified)
"SC" (spatial, channel)	"SC" (spatial, channel)	"UUU" (unspecified, unspecified, unspecified)
"CT" (channel, time)	"CT" (channel, time)	"UUU" (unspecified, unspecified, unspecified)
"BT" (batch, time)	"CBT" (channel, batch, time)	"UUUU" (unspecified, unspecified, unspecified, unspecified)
"SB" (spatial, batch)	"SCB" (spatial, channel, batch)	"UUUU" (unspecified, unspecified, unspecified, unspecified)

If HasMaskInput is 1 (true), then the mask must have the same format as the queries, keys, and values.

References

[1] Vaswani, Ashish, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, and Illia Polosukhin. "Attention is all you need." In Advances in Neural Information Processing Systems, Vol. 30. Curran Associates, Inc., 2017. https://papers.nips.cc/paper_files/paper/2017/hash/3f5ee243547dee91fbd053c1c4a845aa-Abstract.html.

Extended Capabilities

C/C++ Code Generation

Generate C and C++ code using MATLAB® Coder™.

Usage notes and limitations:

• Code generation is not supported when HasScoresOutput is set to true.

• Code generation does not support passing dlarray objects with "U" (unspecified) dimensions to this layer.

GPU Code Generation

Generate CUDA® code for NVIDIA® GPUs using GPU Coder™.

Refer to the usage notes and limitations in the C/C++ Code Generation section. The same usage notes and limitations apply to GPU code generation.

Version History

Introduced in R2024a

R2026a: Apply multiquery and grouped-query attention

The value of NumQueryGroups specifies the type of attention operation:

• For multihead attention, set NumQueryGroups to numHeads.

• For multiquery attention (MQA), set NumQueryGroups to 1.

• For grouped-query attention (GQA), set NumQueryGroups to a positive integer between 1 and numHeads.

Using multiquery attention and grouped-query attention in conjunction with the KeyState and ValueState properties can reduce memory and computation time for large inputs.

R2026a: Cache previous keys and values

You can now cache previous keys and values using the new KeyState and ValueState properties to accelerate making autoregressive predictions with transformer networks. Caching previous keys and values allows you to pass only one previous prediction as input, instead of an entire sequence. Previously, forecasting a future value required using all previous time steps as input.

To cache the keys and values and use them to make autoregressive predictions, set the state of a transformer network after each prediction. For example, this code uses a transformer network to forecast future values of a sequence using one previous time step as input.

% Reset the network state.
net = resetState(net);

% Get the first time step in the sequence.
sequence = initialValue;

% Predict values for the next 100 time steps.
for idx = 1:100
    % Make predicitions by passing the previous value as input.
    previousStep = sequence(end);
    [sequence(end+1),state] = predict(net,previousStep);

    % Update the network state.
    net.State = state;
end

R2026a: Padding mask does not require channel dimension

If the HasPaddingMaskInput property value is 1 (true), then the input data for the corresponding padding mask input does not require a channel dimension. For masks that do not specify a channel dimension, the operation assumes a singleton channel dimension.