Struct ResidencyReport 

pub struct ResidencyReport {
    pub lines: Vec<ResidencyLine>,
    pub required_bytes: u64,
    pub budget_bytes: u64,
    pub usable_bytes: u64,
    pub assumptions: ResidencyAssumptions,
    pub weights_bytes: u64,
    pub kv_shape: KvShape,
}

Fields

lines: Vec<ResidencyLine>

required_bytes: u64

budget_bytes: u64

The ceiling this plan was priced against: a probed crate::device_budget::DeviceBudget::usable_bytes, or a hypothetical machine’s size for what-if planning.

usable_bytes: u64

budget_bytes minus the headroom fraction.

assumptions: ResidencyAssumptions

weights_bytes: u64

Weight bytes the plan charged (dense plus whatever the expert line resolved to), kept separately from the KV term so Self::kv_budget can rebuild the inequality exactly.

kv_shape: KvShape

The KV geometry this checkpoint runs on.

Implementations

impl ResidencyReport

pub fn from_gguf( path: impl AsRef<Path>, assumptions: ResidencyAssumptions, budget_bytes: u64, ) -> Result<Self, LoadError>

Computes the plan for the checkpoint at path (any shard of a split set, or a single file) against budget_bytes (pass a probed crate::device_budget::DeviceBudget::usable_bytes, or a hypothetical machine’s size for what-if planning).

pub fn fits(&self) -> bool

pub fn kv_budget(&self) -> KvBudget

The same plan expressed as the priced inequality, so --ctx auto and the server’s admission check reuse this report’s terms instead of recomputing their own.

The headroom fraction becomes the activation_headroom_bytes term rather than shrinking the budget, which is what makes kv_budget().check(context_tokens).is_ok() agree with Self::fits exactly.

pub fn auto_context(&self, cap: usize) -> ContextFit

Largest context that fits this plan, capped at cap (the model’s own trained context length).

pub fn check_strict(&self) -> Result<(), String>

Strict mode: an Err with the full plan text when the plan overcommits the usable budget. Callers refuse to load on Err.

Trait Implementations

impl Clone for ResidencyReport

fn clone(&self) -> ResidencyReport

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ResidencyReport

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

Formats the value using the given formatter.

impl Display for ResidencyReport

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

Formats the value using the given formatter.