Enum LayerCachePolicy 

pub enum LayerCachePolicy {
    NoState,
    KeyValue {
        attention: AttentionPolicy,
        num_key_value_heads: NonZeroU32,
        head_dim: NonZeroU32,
    },
    KeyOnly {
        attention: AttentionPolicy,
        num_key_heads: NonZeroU32,
        head_dim: NonZeroU32,
    },
    CompressedLatentRotary {
        attention: AttentionPolicy,
        latent_dim: NonZeroU32,
        rotary_dim: NonZeroU32,
    },
    FixedState {
        tensors: Vec<StateTensorPolicy>,
    },
    KeyValueWithFixedState {
        attention: AttentionPolicy,
        num_key_value_heads: NonZeroU32,
        head_dim: NonZeroU32,
        tensors: Vec<StateTensorPolicy>,
    },
    KeyOnlyWithFixedState {
        attention: AttentionPolicy,
        num_key_heads: NonZeroU32,
        head_dim: NonZeroU32,
        tensors: Vec<StateTensorPolicy>,
    },
}

Exact state kind, attention policy, and tensor geometry for one decoder layer.

Variants

NoState

This layer contributes no independently persisted state.

KeyValue

Ordinary attention keys and values.

Fields

attention: AttentionPolicy

Exact full or sliding attention range.

num_key_value_heads: NonZeroU32

Rank-local key/value head count.

head_dim: NonZeroU32

Per-head key/value dimension.

KeyOnly

Attention history whose value payload is intentionally empty.

Fields

attention: AttentionPolicy

Exact full or sliding attention range.

num_key_heads: NonZeroU32

Rank-local key head count.

head_dim: NonZeroU32

Per-head key dimension.

CompressedLatentRotary

Compressed latent state plus rotary keys.

Fields

attention: AttentionPolicy

Exact full or sliding attention range.

latent_dim: NonZeroU32

Compressed latent width.

rotary_dim: NonZeroU32

Rotary-key width.

FixedState

Fixed-size recurrent or convolution state without attention.

Fields

tensors: Vec<StateTensorPolicy>

Ordered tensors required to resume this layer.

KeyValueWithFixedState

Ordinary attention plus fixed-size state.

Fields

attention: AttentionPolicy

Exact full or sliding attention range.

num_key_value_heads: NonZeroU32

Rank-local key/value head count.

head_dim: NonZeroU32

Per-head key/value dimension.

tensors: Vec<StateTensorPolicy>

Ordered additional tensors.

KeyOnlyWithFixedState

Key-only attention plus fixed-size state.

Fields

attention: AttentionPolicy

Exact full or sliding attention range.

num_key_heads: NonZeroU32

Rank-local key head count.

head_dim: NonZeroU32

Per-head key dimension.

tensors: Vec<StateTensorPolicy>

Ordered additional tensors.

Implementations

impl LayerCachePolicy

pub const fn attention_residency_class(&self) -> Option<StateResidencyClass>

Returns the residency behavior of this layer’s attention payload.

pub fn key_value( attention: AttentionPolicy, num_key_value_heads: i32, head_dim: i32, ) -> Result<Self, CachePolicyError>

Constructs validated ordinary key/value state geometry.

pub fn key_only( attention: AttentionPolicy, num_key_heads: i32, head_dim: i32, ) -> Result<Self, CachePolicyError>

Constructs validated key-only state geometry.

pub fn compressed_latent_rotary( attention: AttentionPolicy, latent_dim: i32, rotary_dim: i32, ) -> Result<Self, CachePolicyError>

Constructs validated compressed-latent state geometry.

pub fn fixed_only( tensors: Vec<StateTensorPolicy>, ) -> Result<Self, CachePolicyError>

Constructs a validated fixed-state-only policy.

pub fn key_value_with_fixed_state( attention: AttentionPolicy, num_key_value_heads: i32, head_dim: i32, tensors: Vec<StateTensorPolicy>, ) -> Result<Self, CachePolicyError>

Constructs validated key/value plus fixed-state geometry.

pub fn key_only_with_fixed_state( attention: AttentionPolicy, num_key_heads: i32, head_dim: i32, tensors: Vec<StateTensorPolicy>, ) -> Result<Self, CachePolicyError>

Constructs validated key-only plus fixed-state geometry.

pub const fn attention(&self) -> Option<AttentionPolicy>

Returns the exact attention policy when present.

pub fn fixed_state(&self) -> &[StateTensorPolicy]

Returns ordered non-attention tensor policies.

pub fn components(&self) -> Vec<StateComponentPolicy>

Expands this layer policy into ordered, stably named semantic components shared by runtime residency and prompt persistence.

pub fn validate(&self) -> Result<(), CachePolicyError>

Validates dimensions and fixed-state invariants after deserialization.

Trait Implementations

impl Clone for LayerCachePolicy

fn clone(&self) -> LayerCachePolicy

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for LayerCachePolicy

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for LayerCachePolicy

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Eq for LayerCachePolicy

impl Hash for LayerCachePolicy

fn hash<__H: Hasher>(&self, state: &mut __H)

Feeds this value into the given Hasher.

fn hash_slice<H>(data: &[Self], state: &mut H)

where H: Hasher, Self: Sized,

Feeds a slice of this type into the given Hasher.

impl PartialEq for LayerCachePolicy

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

Equality operator ==.

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

Inequality operator !=.

impl Serialize for LayerCachePolicy

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl StructuralPartialEq for LayerCachePolicy