Skip to contents

This is an implementation of multi-headed attention as described in the paper "Attention is all you Need" Vaswani et al., 2017. If query, key, value are the same, then this is self-attention. Each timestep in query attends to the corresponding sequence in key, and returns a fixed-width vector.

This layer first projects query, key and value. These are (effectively) a list of tensors of length num_attention_heads, where the corresponding shapes are (batch_size, <query dimensions>, key_dim), (batch_size, <key/value dimensions>, key_dim), (batch_size, <key/value dimensions>, value_dim).

Then, the query and key tensors are dot-producted and scaled. These are softmaxed to obtain attention probabilities. The value tensors are then interpolated by these probabilities, then concatenated back to a single tensor.

Finally, the result tensor with the last dimension as value_dim can take a linear projection and return.

Usage

layer_multi_head_attention(
  inputs,
  num_heads,
  key_dim,
  value_dim = NULL,
  dropout = 0,
  use_bias = TRUE,
  output_shape = NULL,
  attention_axes = NULL,
  kernel_initializer = "glorot_uniform",
  bias_initializer = "zeros",
  kernel_regularizer = NULL,
  bias_regularizer = NULL,
  activity_regularizer = NULL,
  kernel_constraint = NULL,
  bias_constraint = NULL,
  seed = NULL,
  ...
)

Arguments

inputs

see description

num_heads

Number of attention heads.

key_dim

Size of each attention head for query and key.

value_dim

Size of each attention head for value.

dropout

Dropout probability.

use_bias

Boolean, whether the dense layers use bias vectors/matrices.

output_shape

The expected shape of an output tensor, besides the batch and sequence dims. If not specified, projects back to the query feature dim (the query input's last dimension).

attention_axes

axes over which the attention is applied. NULL means attention over all axes, but batch, heads, and features.

kernel_initializer

Initializer for dense layer kernels.

bias_initializer

Initializer for dense layer biases.

kernel_regularizer

Regularizer for dense layer kernels.

bias_regularizer

Regularizer for dense layer biases.

activity_regularizer

Regularizer for dense layer activity.

kernel_constraint

Constraint for dense layer kernels.

bias_constraint

Constraint for dense layer kernels.

seed

Optional integer to seed the dropout layer.

...

For forward/backward compatability.

Value

The return value depends on the value provided for the first argument. If object is:

  • a keras_model_sequential(), then the layer is added to the sequential model (which is modified in place). To enable piping, the sequential model is also returned, invisibly.

  • a keras_input(), then the output tensor from calling layer(input) is returned.

  • NULL or missing, then a Layer instance is returned.

Call Arguments

  • query: Query tensor of shape (B, T, dim), where B is the batch size, T is the target sequence length, and dim is the feature dimension.

  • value: Value tensor of shape (B, S, dim), where B is the batch size, S is the source sequence length, and dim is the feature dimension.

  • key: Optional key tensor of shape (B, S, dim). If not given, will use value for both key and value, which is the most common case.

  • attention_mask: a boolean mask of shape (B, T, S), that prevents attention to certain positions. The boolean mask specifies which query elements can attend to which key elements, 1 indicates attention and 0 indicates no attention. Broadcasting can happen for the missing batch dimensions and the head dimension.

  • return_attention_scores: A boolean to indicate whether the output should be (attention_output, attention_scores) if TRUE, or attention_output if FALSE. Defaults to FALSE.

  • training: Python boolean indicating whether the layer should behave in training mode (adding dropout) or in inference mode (no dropout). Will go with either using the training mode of the parent layer/model, or FALSE (inference) if there is no parent layer.

  • use_causal_mask: A boolean to indicate whether to apply a causal mask to prevent tokens from attending to future tokens (e.g., used in a decoder Transformer).

Call return

  • attention_output: The result of the computation, of shape (B, T, E), where T is for target sequence shapes and E is the query input last dimension if output_shape is NULL. Otherwise, the multi-head outputs are projected to the shape specified by output_shape.

  • attention_scores: (Optional) multi-head attention coefficients over attention axes.

Properties

A MultiHeadAttention Layer instance has the following additional read-only properties:

  • attention_axes

  • dropout

  • key_dense

  • key_dim

  • num_heads

  • output_dense

  • output_shape

  • query_dense

  • use_bias

  • value_dense

  • value_dim

See also

Other attention layers:
layer_additive_attention()
layer_attention()
layer_group_query_attention()

Other layers:
Layer()
layer_activation()
layer_activation_elu()
layer_activation_leaky_relu()
layer_activation_parametric_relu()
layer_activation_relu()
layer_activation_softmax()
layer_activity_regularization()
layer_add()
layer_additive_attention()
layer_alpha_dropout()
layer_attention()
layer_average()
layer_average_pooling_1d()
layer_average_pooling_2d()
layer_average_pooling_3d()
layer_batch_normalization()
layer_bidirectional()
layer_category_encoding()
layer_center_crop()
layer_concatenate()
layer_conv_1d()
layer_conv_1d_transpose()
layer_conv_2d()
layer_conv_2d_transpose()
layer_conv_3d()
layer_conv_3d_transpose()
layer_conv_lstm_1d()
layer_conv_lstm_2d()
layer_conv_lstm_3d()
layer_cropping_1d()
layer_cropping_2d()
layer_cropping_3d()
layer_dense()
layer_depthwise_conv_1d()
layer_depthwise_conv_2d()
layer_discretization()
layer_dot()
layer_dropout()
layer_einsum_dense()
layer_embedding()
layer_feature_space()
layer_flatten()
layer_flax_module_wrapper()
layer_gaussian_dropout()
layer_gaussian_noise()
layer_global_average_pooling_1d()
layer_global_average_pooling_2d()
layer_global_average_pooling_3d()
layer_global_max_pooling_1d()
layer_global_max_pooling_2d()
layer_global_max_pooling_3d()
layer_group_normalization()
layer_group_query_attention()
layer_gru()
layer_hashed_crossing()
layer_hashing()
layer_identity()
layer_integer_lookup()
layer_jax_model_wrapper()
layer_lambda()
layer_layer_normalization()
layer_lstm()
layer_masking()
layer_max_pooling_1d()
layer_max_pooling_2d()
layer_max_pooling_3d()
layer_maximum()
layer_mel_spectrogram()
layer_minimum()
layer_multiply()
layer_normalization()
layer_permute()
layer_random_brightness()
layer_random_contrast()
layer_random_crop()
layer_random_flip()
layer_random_rotation()
layer_random_translation()
layer_random_zoom()
layer_repeat_vector()
layer_rescaling()
layer_reshape()
layer_resizing()
layer_rnn()
layer_separable_conv_1d()
layer_separable_conv_2d()
layer_simple_rnn()
layer_spatial_dropout_1d()
layer_spatial_dropout_2d()
layer_spatial_dropout_3d()
layer_spectral_normalization()
layer_string_lookup()
layer_subtract()
layer_text_vectorization()
layer_tfsm()
layer_time_distributed()
layer_torch_module_wrapper()
layer_unit_normalization()
layer_upsampling_1d()
layer_upsampling_2d()
layer_upsampling_3d()
layer_zero_padding_1d()
layer_zero_padding_2d()
layer_zero_padding_3d()
rnn_cell_gru()
rnn_cell_lstm()
rnn_cell_simple()
rnn_cells_stack()