Struct PartitionedTextRuntime 

pub struct PartitionedTextRuntime<A, B, S, P, E, G, R, I, T, U, V>
where
    B: CommunicationBackend,
    G: Borrow<B::CommunicationGroup>,
    R: Borrow<B::CommunicationRoute>,
    I: CommunicationTensorMetadata<B>,
    T: PartitionBoundaryTransport<B, G, R, I>,
    U: PartitionOutputPublisher<B, G, R, I>,
    V: PartitionCommitAgreement<B, G, R, I>,
{ /* private fields */ }

Complete rank-local runtime installed behind the shared replicated session.

Implementations

impl<A, B, S, P, E, G, R, I, T, U, V> PartitionedTextRuntime<A, B, S, P, E, G, R, I, T, U, V>

where B: CommunicationBackend, G: Borrow<B::CommunicationGroup>, R: Borrow<B::CommunicationRoute>, I: CommunicationTensorMetadata<B>, T: PartitionBoundaryTransport<B, G, R, I>, U: PartitionOutputPublisher<B, G, R, I>, V: PartitionCommitAgreement<B, G, R, I>,

pub fn new( plan: PartitionedExecutionPlan, executor: E, communication: PartitionCommunication<B, G, R, I>, communication_executor: B::OwnedExecutor, boundary_transport: T, output_publisher: U, commit_agreement: V, residency: ExecutionResidency, bounded_policy: Option<P>, ) -> Result<Self, PartitionExecutionError>

Pairs one immutable plan with its rank-local executable and opaque resources.

impl<A, B, S, P, ExecutionPolicy, G, R, I, T, U, V> PartitionedTextRuntime<A, B, S, P, LayerwiseTraversalPartitionExecutor<A, B, S, ExecutionPolicy>, G, R, I, T, U, V>

where B: CommunicationBackend<Executor = <<B as NeuralBackend>::Tensor as Tensor>::Context>, S: RuntimeState<B>, A: ParallelLayeredArchitecture<B, S>, ExecutionPolicy: LayerwisePolicy<B, A::Unit>, G: Borrow<B::CommunicationGroup>, R: Borrow<B::CommunicationRoute>, I: CommunicationTensorMetadata<B>, T: PartitionBoundaryTransport<B, G, R, I>, U: PartitionOutputPublisher<B, G, R, I>, V: PartitionCommitAgreement<B, G, R, I>, B::ParallelContext: Sized, A::Error: Display, ExecutionPolicy::Error: Display,

pub fn forward_with_traversal_hook<'a, H>( &mut self, input: A::Input<'a>, state: &mut S, context: &<B::Tensor as Tensor>::Context, hook: &mut H, ) -> PartitionedTraversalResult<B::Tensor, A::ForwardContext, A::Error, ExecutionPolicy::Error>

where H: LayeredTraversalHook<B, A::ForwardContext, A::Error> + ?Sized,

Runs one architecture-owned traversal through the selected neutral partitioned base.

Every architecture group must be local. Point-to-point routes, output publication, and phase agreement use the ordinary partitioned session entry instead and are rejected here rather than silently bypassed.

pub const fn traversal_executor( &self, ) -> &LayerwiseTraversalPartitionExecutor<A, B, S, ExecutionPolicy>

Borrows the installed traversal executor for neutral residency reports.