Struct ObservedExpertProvider 

pub struct ObservedExpertProvider<'a, P, O: ?Sized, E> { /* private fields */ }

Decorates a routed provider with normalized routing observation.

The decorator sees the exact request and output of canonical provider execution. It therefore adds observation without reimplementing a model family’s block, routing, shape, or residency lifecycle. Tensor-parallel requests are delegated without an event because their provider result may still require an architecture-owned reduction before it is observable.

Implementations

impl<'a, P, O: ?Sized, E> ObservedExpertProvider<'a, P, O, E>

pub fn new( provider: &'a mut P, observer: &'a mut O, point: RoutedObservationPoint, ) -> Self

Wraps provider for one canonical routed module invocation.

Trait Implementations

impl<B, P, O, E> RoutedExpertProvider<B> for ObservedExpertProvider<'_, P, O, E>

where B: GroupedNeuralBackend, P: RoutedExpertProvider<B>, O: ActivationObserver<B::Tensor, E> + ?Sized,

type Error = ObservedExpertProviderError<<P as RoutedExpertProvider<B>>::Error, E>

Provider-specific acquisition or execution failure.

fn routing_control( &mut self, token_rows: u64, ) -> Result<Option<GroupSelectionControl>, Self::Error>

Optional pre-dispatch control supplied by an instrumented provider adapter.

fn routing_applied( &mut self, original: Option<RoutingDecision<'_, B::Tensor>>, effective: RoutingDecision<'_, B::Tensor>, ) -> Result<(), Self::Error>

Consumes bounded decision evidence before dispatch. Shared experts are not included.

fn routing_failed(&mut self, message: &str)

Attributes selector failure without claiming cache rollback.

fn forward_grouped( &mut self, resident_bank: &mut B::GatedProductGroups, request: RoutedExpertRequest<'_, B::Tensor>, context: &<B::Tensor as Tensor>::Context, ) -> Result<B::Tensor, Self::Error>

Executes one typed route batch while retaining its acquired resources.

fn forward_relu2_routed( &mut self, resident_bank: &mut B::Relu2Groups, request: RoutedExpertRequest<'_, B::Tensor>, context: &<B::Tensor as Tensor>::Context, ) -> Result<B::Tensor, Self::Error>

Executes one ReLU-squared route batch through the same residency boundary.

fn forward_compact_grouped( &mut self, resident_bank: &mut B::GatedProductGroups, request: RoutedExpertRequest<'_, B::Tensor>, context: &<B::Tensor as Tensor>::Context, ) -> Result<B::Tensor, Self::Error>

Executes destination-local rows that were already expanded to one owner-local expert per row by the neutral expert exchange.

impl<B, P, O, E> TensorParallelRoutedExpertProvider<B> for ObservedExpertProvider<'_, P, O, E>

where B: GroupedNeuralBackend, P: TensorParallelRoutedExpertProvider<B>, O: ActivationObserver<B::Tensor, E> + ?Sized,

fn forward_grouped_tensor_parallel( &mut self, resident_bank: &mut B::GatedProductGroups, request: RoutedExpertRequest<'_, B::Tensor>, partitions: usize, context: &<B::Tensor as Tensor>::Context, ) -> Result<RoutedExpertTensorParallelOutput<B::Tensor>, Self::Error>

Executes a rank-local gated-product contribution.

fn forward_relu2_routed_tensor_parallel( &mut self, resident_bank: &mut B::Relu2Groups, request: RoutedExpertRequest<'_, B::Tensor>, partitions: usize, context: &<B::Tensor as Tensor>::Context, ) -> Result<RoutedExpertTensorParallelOutput<B::Tensor>, Self::Error>

Executes a rank-local ReLU-squared contribution.

fn forward_compact_grouped_tensor_parallel( &mut self, resident_bank: &mut B::GatedProductGroups, request: RoutedExpertRequest<'_, B::Tensor>, partitions: usize, context: &<B::Tensor as Tensor>::Context, ) -> Result<RoutedExpertTensorParallelOutput<B::Tensor>, Self::Error>

Executes destination-local, one-expert-per-row contributions while preserving the backend’s TP reduction and post-bias structure.