Struct SpeculativeRequestTable 

pub struct SpeculativeRequestTable<'cache, E, S, C, P>
where
    E: SpeculativeExecutor,
    S: SpeculativeSampling<Logits = E::Logits, Error = E::Error>,
    C: SpeculativeConstraint,
    P: SpeculativePublisher<C>,
{ /* private fields */ }

Canonical table and action coordinator for speculative requests.

Implementations

impl<'cache, E, S, C, P> SpeculativeRequestTable<'cache, E, S, C, P>

where E: SpeculativeExecutor, S: SpeculativeSampling<Logits = E::Logits, Error = E::Error>, C: SpeculativeConstraint, P: SpeculativePublisher<C>,

pub fn new( options: SpeculativeSchedulerOptions, topology: SpeculativeExecutionTopology, ) -> Result<Self, GenerationError>

Creates an empty validated request table.

pub fn request( &self, id: SpeculativeRequestId, ) -> Option<&SpeculativeRequest<'cache, E, S, C, P>>

Returns one request by stable identity.

pub fn status( &self, id: SpeculativeRequestId, ) -> Option<SpeculativeRequestStatus>

Returns one request’s current status.

pub fn is_finished(&self) -> bool

Whether every request is terminal.

pub const fn options(&self) -> SpeculativeSchedulerOptions

Validated scheduler options.

pub fn submit<'context>( &mut self, executor: &mut E, cache: &'cache mut E::Cache, input: E::Input, config: SpeculativeConfig, runtime: SpeculativeOutputRuntime<S, C, P>, randomness: SpeculativeRandomness<S::RandomState, S::DraftRandomness>, component_timings_collected: bool, context: E::Context<'context>, ) -> Result<SpeculativeRequestId, SpeculativeDriverError<E::Error>>

where E: 'context, S: SpeculativeSampling<Logits = E::Logits, Error = E::Error, Context<'context> = E::Context<'context>> + 'context,

Prefills and inserts one request, or records its pre-existing terminal state.

pub fn cancel( &mut self, id: SpeculativeRequestId, ) -> Result<(), SpeculativeDriverError<E::Error>>

Requests cancellation without releasing an exact in-flight transaction.

pub fn step<'context>( &mut self, executor: &mut E, optimistic_execution_available: bool, context: E::Context<'context>, ) -> Result<bool, SpeculativeDriverError<E::Error>>

where E: 'context, S: SpeculativeSampling<Logits = E::Logits, Error = E::Error, Context<'context> = E::Context<'context>> + 'context,

Applies one fairly selected request action.

pub fn run<'context>( &mut self, executor: &mut E, optimistic_execution_available: bool, context: E::Context<'context>, ) -> Result<(), SpeculativeDriverError<E::Error>>

where E: 'context, S: SpeculativeSampling<Logits = E::Logits, Error = E::Error, Context<'context> = E::Context<'context>> + 'context,

Drives every request to a terminal state.

pub fn finish( self, ) -> Result<CompletedSpeculativeSchedule<S>, SpeculativeDriverError<E::Error>>

Consumes a terminal table and returns outputs in stable submission order.