Struct MemoryPlan 

pub struct MemoryPlan { /* private fields */ }

Implementations

impl MemoryPlan

pub fn startup_peak_bytes( &self, context_tokens: u64, active_sequences: u32, step_tokens: u64, ) -> Result<u64, VNextError>

Physical bytes needed by equal-length sequences and one execution wave.

This is startup planning evidence, never a reservation or an admission permit. Runtime admission still handles live cache retention, contention, and changing device availability. Each sequence is allocated separately; physical page padding, provider formulas, and proven workspace reuse are evaluated by the same plan helpers used to size runtime backing.

pub fn startup_workload_peak_bytes( &self, request_ceiling_tokens: u64, sequence_frontier_tokens: u64, active_sequences: u32, step_tokens: u64, ) -> Result<u64, VNextError>

Request-lifetime storage owns the complete requested input/output ceiling; sequence state owns only its committed frontier. Keep these dimensions separate when evaluating shallow concurrent decoding.

impl MemoryPlan

pub const fn device_capacity_bytes(&self) -> u64

pub const fn policy_capacity_bytes(&self) -> u64

pub const fn reserve_bytes(&self) -> u64

pub const fn usable_capacity_bytes(&self) -> u64

pub const fn maximum_active_sequences(&self) -> u32

pub const fn checkpoint_capacity(&self) -> Option<&CheckpointCapacityPolicy>

Aggregate retained-extent permission, without reserving bytes or implying that runtime cache admission has been enabled.

pub const fn capacity_bytes(&self) -> u64

pub const fn static_bytes(&self) -> u64

pub const fn minimum_request_bytes(&self) -> u64

pub const fn minimum_sequence_bytes(&self) -> u64

pub const fn minimum_step_bytes(&self) -> u64

pub const fn minimum_invocation_peak_bytes(&self) -> u64

pub const fn minimum_runnable_request_bytes(&self) -> u64

pub const fn invocation_liveness_mode(&self) -> InvocationLivenessMode

pub fn invocation_liveness(&self) -> &[InvocationResourceLiveness]

pub fn theoretical_ceiling_bytes(&self) -> u128

Conservative checked evidence across static allocations, live provider formula maxima, reusable execution workspace, and the protocol ceiling. It is never itself a reservation, admission target, or performance claim. Ordinary resource and reusable-workspace ceiling. Optional checkpoint growth is reported separately by checkpoint_capacity and pool bounds.

pub fn static_allocations(&self) -> &[ResourceAllocation]

pub fn dynamic_descriptors(&self) -> &[DynamicResourceDescriptor]

pub fn dynamic_pools(&self) -> &[DynamicBackingPoolSpec]

pub fn reusable_execution(&self) -> Option<&ReusableExecutionMemoryPlan>

pub fn static_buffer_requests(&self) -> Result<Vec<BufferRequest>, VNextError>

Trait Implementations

impl Clone for MemoryPlan

fn clone(&self) -> MemoryPlan

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for MemoryPlan

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for MemoryPlan

fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>

where D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Eq for MemoryPlan

impl PartialEq for MemoryPlan

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

Equality operator ==.

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

Inequality operator !=.

impl Serialize for MemoryPlan

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 MemoryPlan