Struct CrossAttentionConfig 

pub struct CrossAttentionConfig {
    pub d_model: usize,
    pub d_context: usize,
    pub n_heads: usize,
    pub n_heads_kv: usize,
    pub d_head: usize,
    pub dropout: f64,
    pub min_float: f64,
    pub quiet_softmax: bool,
}

Configuration to create a CrossAttention layer using the init function.

Fields

d_model: usize

Dimension of the Query (e.g., Decoder state).

d_context: usize

Dimension of the Context (e.g., Encoder audio embeddings).

n_heads: usize

Number of heads for the Query.

n_heads_kv: usize

Number of heads for Key/Value (Set to 1 for MQA, set to n_heads for MHA).

d_head: usize

Dimension of a single head.

dropout: f64

Dropout rate.

min_float: f64

Masking value. Use -1.0e4 for f16/bf16 safety.

quiet_softmax: bool

Use quiet_softmax to allow zero-attention (good for sparse/quantized models).

Implementations

impl CrossAttentionConfig

pub fn new( d_model: usize, d_context: usize, n_heads: usize, n_heads_kv: usize, d_head: usize, ) -> CrossAttentionConfig

Create a new instance of the config.

Arguments

Required Arguments

d_model

Dimension of the Query (e.g., Decoder state).

d_context

Dimension of the Context (e.g., Encoder audio embeddings).

n_heads

Number of heads for the Query.

n_heads_kv

Number of heads for Key/Value (Set to 1 for MQA, set to n_heads for MHA).

d_head

Dimension of a single head.

Default Arguments

dropout

Dropout rate.

• Defaults to 0.1

min_float

Masking value. Use -1.0e4 for f16/bf16 safety.

• Defaults to -1.0e4

quiet_softmax

Use quiet_softmax to allow zero-attention (good for sparse/quantized models).

• Defaults to false

impl CrossAttentionConfig

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

Sets the value for the field dropout.

Dropout rate.

• Defaults to 0.1

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

Sets the value for the field min_float.

Masking value. Use -1.0e4 for f16/bf16 safety.

• Defaults to -1.0e4

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

Sets the value for the field quiet_softmax.

Use quiet_softmax to allow zero-attention (good for sparse/quantized models).

• Defaults to false

impl CrossAttentionConfig

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

where B: Backend,

Initializes a new cross-attention module.

Arguments

• device - The device on which to initialize the module.

Returns

A new CrossAttention module.

Trait Implementations

impl Clone for CrossAttentionConfig

fn clone(&self) -> CrossAttentionConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Config for CrossAttentionConfig

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 Debug for CrossAttentionConfig

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for CrossAttentionConfig

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

where D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Display for CrossAttentionConfig

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

Formats the value using the given formatter.

impl Serialize for CrossAttentionConfig

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.