Struct MultiHeadAttentionConfig

pub struct MultiHeadAttentionConfig {
    pub d_model: usize,
    pub n_heads: usize,
    pub dropout: f64,
    pub min_float: f64,
    pub quiet_softmax: bool,
    pub initializer: Initializer,
}

Configuration to create a Multi Head Attention layer using the init function.

Fields

d_model: usize

The size of each linear layer.

n_heads: usize

The number of heads.

dropout: f64

The dropout rate. Default: 0.1

min_float: f64

The minimum value a float can take. Default: -1.0e4 This is used to mask attention scores before calculating attention weights. A value too low might result in NaN.

quiet_softmax: bool

Use “quiet softmax” instead of regular softmax.

• Usage may improve performance by allowing attention heads to deposit no information (if the sequence contains no information relevant to that head).
• Usage may reduce the entropy of weights in the model, enhancing quantization and compression.

Reference: https://www.evanmiller.org/attention-is-off-by-one.html

initializer: Initializer

The type of function used to initialize neural network parameters

Implementations

impl MultiHeadAttentionConfig

pub fn new(d_model: usize, n_heads: usize) -> MultiHeadAttentionConfig

Create a new instance of the config.

impl MultiHeadAttentionConfig

pub fn with_dropout(self, dropout: f64) -> MultiHeadAttentionConfig

The dropout rate. Default: 0.1

pub fn with_min_float(self, min_float: f64) -> MultiHeadAttentionConfig

The minimum value a float can take. Default: -1.0e4

pub fn with_quiet_softmax(self, quiet_softmax: bool) -> MultiHeadAttentionConfig

Use “quiet softmax” instead of regular softmax.

pub fn with_initializer( self, initializer: Initializer, ) -> MultiHeadAttentionConfig

The type of function used to initialize neural network parameters

impl MultiHeadAttentionConfig

pub fn init<B>(&self, device: &<B as Backend>::Device) -> MultiHeadAttention<B>

where B: Backend,

Initialize a new multihead attention module.

Trait Implementations

impl Clone for MultiHeadAttentionConfig

fn clone(&self) -> MultiHeadAttentionConfig

Returns a copy of the value. Read more

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source. Read more

impl Config for MultiHeadAttentionConfig

fn save<P>(&self, file: P) -> Result<(), Error>

where P: AsRef<Path>,

Saves the configuration to a file.

fn load<P>(file: P) -> Result<Self, ConfigError>

where P: AsRef<Path>,

Loads the configuration from a file.

fn load_binary(data: &[u8]) -> Result<Self, ConfigError>

Loads the configuration from a binary buffer.

impl<'de> Deserialize<'de> for MultiHeadAttentionConfig

fn deserialize<D>( deserializer: D, ) -> Result<MultiHeadAttentionConfig, <D as Deserializer<'de>>::Error>

where D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Display for MultiHeadAttentionConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter. Read more

impl Serialize for MultiHeadAttentionConfig

fn serialize<S>( &self, serializer: S, ) -> Result<<S as Serializer>::Ok, <S as Serializer>::Error>

where S: Serializer,

Serialize this value into the given Serde serializer.