Enum AttentionProjection 

#[repr(u8)]
pub enum AttentionProjection {
    Full = 0,
    SharedKV = 1,
}

Attention projection configuration. Controls whether K and V projections share weights (Q-K=V tying).

Variants

Full = 0

Standard Q, K, V (3 projections, full KV cache)

SharedKV = 1

Q-K=V: K and V share projection (2 projections, K-only cache). 50% KV cache reduction, ~3% perplexity cost. Post-hoc weight merging: W_kv = (W_k + W_v) / 2.

Trait Implementations

impl Clone for AttentionProjection

fn clone(&self) -> AttentionProjection

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for AttentionProjection

impl Debug for AttentionProjection

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

Formats the value using the given formatter.

impl Default for AttentionProjection

fn default() -> AttentionProjection

Returns the “default value” for a type.

impl Eq for AttentionProjection

impl PartialEq for AttentionProjection

fn eq(&self, other: &AttentionProjection) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for AttentionProjection