Struct NormalizedLoadRequest 

pub struct NormalizedLoadRequest { /* private fields */ }

One normalized backend-neutral request for cold model preparation.

Parallel rank, wire, and invocation fields are represented atomically. A concrete backend may pair this value with a separate native device or resource token, but must not add native identity to this request.

Implementations

impl NormalizedLoadRequest

pub fn from_execution_plan( plan: &ExecutionPlan, diagnostics: ResidencyDiagnostics, parallel: Option<ParallelLoadRequest>, ) -> Result<Self, ExecutionPlanLoadError>

Derives all portable load policy from a plan and optional complete rank request.

Distributed plans supply their exact rank, wire, invocation limits and completion policy explicitly because an execution plan does not contain those values. Native device identity and realized resources never enter this conversion.

impl NormalizedLoadRequest

pub const fn with_prompt_cache_persistence(self, required: bool) -> Self

Requires persisted prompt-prefix import/export independently of state residency.

pub const fn prompt_cache_persistence(&self) -> bool

Returns explicit persisted prompt-prefix import/export intent.

pub const fn with_max_cached_shards(self, maximum: NonZeroUsize) -> Self

Selects the exact reader-cache limit, including fully resident execution.

pub const fn max_cached_shards(&self) -> usize

Returns the source reader-cache bound retained by cold selection.

pub fn with_quantization(quantization: QuantizationRequest) -> Self

Creates a request that quantizes eligible dense weights on load.

pub fn with_parallel_execution( self, parallel: ParallelLoadRequest, ) -> Result<Self, NormalizedLoadRequestError>

Attaches one complete portable parallel-execution request.

pub const fn with_communication_completion_policy( self, policy: CommunicationCompletionPolicy, ) -> Self

Selects the bounded completion policy used by communication or local realtime work.

pub const fn set_communication_completion_policy( &mut self, policy: CommunicationCompletionPolicy, )

Replaces the bounded completion policy without changing another request field.

pub fn with_weight_residency(self, residency: WeightResidency) -> Self

Selects fully resident or bounded checkpoint-weight execution.

pub fn with_state_residency(self, residency: CacheResidencyPolicy) -> Self

Selects mutable-state residency and paging controls.

pub const fn with_required_session_capabilities( self, capabilities: SessionCapabilities, ) -> Self

Requires capabilities from the exact inspected and realized session.

pub const fn with_drafting(self, drafting: DraftingLoadRequest) -> Self

Selects portable drafting intent.

pub fn with_drafting_plan( self, plan: &DraftingPlan, ) -> Result<Self, NormalizedLoadRequestError>

Applies a portable execution plan’s drafting mode before payload selection.

pub const fn quantization(&self) -> Option<QuantizationRequest>

Returns the requested dense-weight transformation.

pub const fn parallel_execution(&self) -> Option<ParallelLoadRequest>

Returns the complete portable parallel request.

pub const fn parallel_topology(&self) -> Option<ParallelRankTopology>

Returns the selected portable rank, when parallel execution was attached.

pub const fn pipeline_wire_contract(&self) -> Option<PipelineWireContract>

Returns the activation wire contract for parallel execution.

pub const fn has_parallel_execution(&self) -> bool

Reports whether a portable parallel execution request is attached.

pub fn partitioned_invocation_limits(&self) -> Option<(i32, i32)>

Returns invocation limits already validated by parallel construction.

pub fn communication_completion_policy( &self, ) -> Result<Option<CommunicationCompletionPolicy>, NormalizedLoadRequestError>

Returns bounded communication completion for model preparation.

pub fn realtime_completion_policy( &self, ) -> Result<CommunicationCompletionPolicy, NormalizedLoadRequestError>

Returns completion policy for realtime work, including local async evaluation.

pub const fn weight_residency(&self) -> WeightResidency

Returns selected immutable-weight residency.

pub const fn state_residency(&self) -> &CacheResidencyPolicy

Returns selected mutable-state residency.

pub const fn required_session_capabilities(&self) -> SessionCapabilities

Returns capabilities required from the realized session.

pub const fn drafting(&self) -> DraftingLoadRequest

Returns the selected pre-payload drafting intent.

pub fn weight_quantization( &self, ) -> Result<Option<WeightQuantization>, NormalizedLoadRequestError>

Converts the requested transformation to the checkpoint lowering vocabulary.

pub fn validate_model_preparation( &self, ) -> Result<ValidatedModelLoadRequest<'_>, NormalizedLoadRequestError>

Validates the complete ordinary model-preparation request.

pub fn preparation_policy( &self, ) -> Result<PreparationPolicy, NormalizedLoadRequestError>

Converts this normalized request into core’s portable preparation policy.

Trait Implementations

impl Clone for NormalizedLoadRequest

fn clone(&self) -> NormalizedLoadRequest

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for NormalizedLoadRequest

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

Formats the value using the given formatter.

impl Default for NormalizedLoadRequest

fn default() -> NormalizedLoadRequest

Returns the “default value” for a type.

impl Eq for NormalizedLoadRequest

impl PartialEq for NormalizedLoadRequest

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

Equality operator ==.

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

Inequality operator !=.

impl StructuralPartialEq for NormalizedLoadRequest