Struct MultiQueryAttentionConfig 

pub struct MultiQueryAttentionConfig {
    pub num_heads: usize,
    pub num_kv_heads: usize,
    pub head_dim: usize,
    pub dropout_prob: f64,
    pub causal: bool,
}

Configuration for Multi-Query Attention

Fields

num_heads: usize

Number of query heads

num_kv_heads: usize

Number of KV heads (1 = pure MQA, num_heads = standard MHA)

head_dim: usize

Per-head dimension

dropout_prob: f64

Dropout probability

causal: bool

Whether to apply causal masking

Implementations

impl MultiQueryAttentionConfig

pub fn new(num_heads: usize, head_dim: usize) -> Self

Create a pure MQA config (1 KV head)

pub fn with_num_kv_heads(self, n: usize) -> Self

Set number of KV heads (1 = MQA, num_heads = MHA)

pub fn with_causal(self, causal: bool) -> Self

Enable or disable causal masking

pub fn with_dropout(self, prob: f64) -> Self

Set dropout probability

Trait Implementations

impl Clone for MultiQueryAttentionConfig

fn clone(&self) -> MultiQueryAttentionConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for MultiQueryAttentionConfig

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

Formats the value using the given formatter.

impl Default for MultiQueryAttentionConfig

fn default() -> Self

Returns the “default value” for a type.