Crate eredu_core 

Backend-neutral language-model contracts and orchestration.

This crate deliberately contains no tensor runtime. Backends own tensors, streams, executable models, caches, and completion primitives; core owns validation, lifecycle state, scheduling, and portable schemas.

Re-exports

pub use artifact::gguf_u32_metadata_values;

pub use artifact::inspect_artifact;

pub use artifact::plan_model_preparation;

pub use artifact::resolve_gguf_companions;

pub use artifact::validate_preparation_policy;

pub use artifact::ArtifactFormat;

pub use artifact::ArtifactInspection;

pub use artifact::GgufCompanionEncoding;

pub use artifact::GgufCompanionRequirement;

pub use artifact::GgufCompanionRole;

pub use artifact::LoadingProtocol;

pub use artifact::MaterializationRoute;

pub use artifact::ModelArtifact;

pub use artifact::ModelConfiguration;

pub use artifact::ModelConfigurationResolver;

pub use artifact::ModelPreparationPlan;

pub use artifact::PreparationPolicy;

pub use artifact::QuantizationRequest;

pub use artifact::ResidencyRequest;

pub use artifact::ResolvedModelConfiguration;

pub use artifact::ValidatedGguf;

pub use artifact::ValidatedGgufCompanion;

pub use attention::AttentionPolicy;

pub use attention::LayerSchedule;

pub use attention::LayerScheduleError;

pub use automatic::realize_execution_plan_drafting;

pub use automatic::realize_execution_plan_target;

pub use automatic::AllocatorTelemetry;

pub use automatic::AutomaticPlanRequest;

pub use automatic::AutomaticPlanner;

pub use automatic::AutomaticPlannerPolicy;

pub use automatic::AutomaticPlanningBackend;

pub use automatic::AutomaticPlanningError;

pub use automatic::BoundedResidencyRequirement;

pub use automatic::CandidateAdmission;

pub use automatic::DurationSeconds;

pub use automatic::ExecutionPlanBackendFactory;

pub use automatic::ExecutionPlanReport;

pub use automatic::ExecutionPlanTarget;

pub use automatic::ExecutionTelemetry;

pub use automatic::ExpertCacheTelemetry;

pub use automatic::ExternalDraftArtifact;

pub use automatic::HardwareBackendProfile;

pub use automatic::HardwareDeviceProfile;

pub use automatic::HardwareMemorySemantics;

pub use automatic::HardwareProfile;

pub use automatic::ModelResourceProfile;

pub use automatic::ObservationKind;

pub use automatic::Observed;

pub use automatic::PlanExplanation;

pub use automatic::PlanExplanationEntry;

pub use automatic::PlanExplanationLevel;

pub use automatic::RealizedDrafting;

pub use automatic::ResidencyTelemetry;

pub use automatic::SpeculativeDecodingTelemetry;

pub use automatic::TimingTelemetry;

pub use automatic::TokenizerCompatibilityError;

pub use automatic::TokenizerCompatibilityProof;

pub use automatic::TransferTelemetry;

pub use automatic::AUTOMATIC_SCHEMA_VERSION;

pub use backend::load_model;

pub use backend::prepare_inspected_model;

pub use backend::BackendDescriptor;

pub use backend::BackendError;

pub use backend::BackendProvider;

pub use backend::BackendSession;

pub use backend::CollectiveGroupDescriptor;

pub use backend::CollectiveGroupId;

pub use backend::CollectiveScope;

pub use backend::Completion;

pub use backend::ControlledTextGeneration;

pub use backend::ControlledTextGenerationError;

pub use backend::ControlledToken;

pub use backend::DeviceCapabilities;

pub use backend::DeviceDescriptor;

pub use backend::DistributedBackend;

pub use backend::DistributedCapabilities;

pub use backend::DistributedSession;

pub use backend::DistributedSessionDescriptor;

pub use backend::InspectableBackendSession;

pub use backend::ModelCapabilityBackend;

pub use backend::ModelLoadError;

pub use backend::ModelLoadingBackend;

pub use backend::ModelRuntime;

pub use backend::MultimodalPreparationBackend;

pub use backend::MultimodalPreparationFailure;

pub use backend::PreparedModel;

pub use backend::SessionCapabilities;

pub use backend::SessionCapabilityError;

pub use backend::SpeculativeTokenFilterController;

pub use backend::Submission;

pub use backend::TextGeneration;

pub use backend::TextGenerationBackend;

pub use backend::TextGenerationConfig;

pub use backend::TextSamplingStrategy;

pub use backend::TokenFilter;

pub use backend::TokenFilterController;

pub use backend::TokenFilterError;

pub use backend::TokenOutput;

pub use backend::ValueDescriptor;

pub use capability::apply_admission_policy;

pub use capability::estimate_runtime_state;

pub use capability::Admission;

pub use capability::AdmissionRejection;

pub use capability::AdmissionRequest;

pub use capability::AdmissionResult;

pub use capability::AvailableMemory;

pub use capability::CacheStateStrategy;

pub use capability::CapabilityError;

pub use capability::EstimationCompleteness;

pub use capability::InputModalities;

pub use capability::InputTokenCount;

pub use capability::ModelCapabilities;

pub use capability::PhysicalMemorySemantics;

pub use capability::RuntimeStateEstimate;

pub use capability::SlidingWindowLayerCount;

pub use capability::StateMemoryAssumptions;

pub use capability::StateMemoryLayout;

pub use capability::StaticMemoryReport;

pub use execution::BackendId;

pub use execution::DevicePlan;

pub use execution::DraftPlacementPlan;

pub use execution::DraftingPlan;

pub use execution::ExecutionPlan;

pub use execution::ExecutionPlanError;

pub use execution::ExpertCachePlan;

pub use execution::ResidencyPlan;

pub use execution::WeightTransformationPlan;

pub use execution::DEFAULT_MAX_CACHED_SHARDS;

pub use execution::EXECUTION_PLAN_SCHEMA_VERSION;

pub use generation::resolve_generation_config;

pub use generation::resolve_optimistic_reuse;

pub use generation::CheckpointGenerationConfig;

pub use generation::FinishReason;

pub use generation::GenerationCancellationToken;

pub use generation::GenerationConfigOverrides;

pub use generation::GenerationError;

pub use generation::GenerationSequence;

pub use generation::OptimisticReuseDecision;

pub use generation::ResolvedGenerationConfig;

pub use generation::SemanticEvent;

pub use generation::SpeculativeCancellationDisposition;

pub use generation::SpeculativeCommitPlan;

pub use generation::SpeculativeConfig;

pub use generation::SpeculativeRequestId;

pub use generation::SpeculativeRequestLifecycle;

pub use generation::SpeculativeRequestStatus;

pub use generation::SpeculativeRound;

pub use generation::SpeculativeSchedulerOptions;

pub use generation::SpeculativeTail;

pub use generation::TokenCommit;

pub use generation::TokenTerminalSignals;

pub use input::InputExtent;

pub use input::InputMetadataKey;

pub use input::InputModality;

pub use input::InputPartDescriptor;

pub use input::InputPayloadKind;

pub use input::InputTensorIdentity;

pub use input::PreparedInputError;

pub use input::PreparedInputIdentity;

pub use inspection::ArtifactModality;

pub use inspection::ArtifactTensorEncoding;

pub use inspection::InspectionIssue;

pub use inspection::InspectionIssueCode;

pub use inspection::InspectionReadiness;

pub use inspection::InspectionRequirement;

pub use inspection::InspectionSeverity;

pub use inspection::ModelInspectionReport;

pub use media::Audio;

pub use media::Media;

pub use media::MediaBinding;

pub use media::MediaRequestError;

pub use media::MultimodalRequest;

pub use media::MultimodalSegment;

pub use media::RgbImage;

pub use media::TokenizedMultimodalRequest;

pub use media::TokenizedMultimodalSegment;

pub use media::Video;

pub use media::VideoSampling;

pub use observation::InspectedOutput;

pub use observation::ObservationError;

pub use observation::ObservationRequest;

pub use observation::ObservationSelector;

pub use observation::ObservationSet;

pub use observation::ObservationValue;

pub use observation::TensorObservation;

pub use observation::TensorObservationData;

pub use observation::MODEL_LOGITS_OBSERVATION_PATH;

pub use realtime::load_realtime_model;

pub use realtime::load_realtime_model_with_options;

pub use realtime::RealtimeBackend;

pub use realtime::RealtimeCompletedStep;

pub use realtime::RealtimeConfigError;

pub use realtime::RealtimeDecisionDiagnostics;

pub use realtime::RealtimeError;

pub use realtime::RealtimeFrameConvention;

pub use realtime::RealtimeFrameForcing;

pub use realtime::RealtimeFrameScheduleState;

pub use realtime::RealtimeFrameSlot;

pub use realtime::RealtimeFrameTransition;

pub use realtime::RealtimeInputFrame;

pub use realtime::RealtimeModel;

pub use realtime::RealtimeModelLoadingBackend;

pub use realtime::RealtimeOutputFrame;

pub use realtime::RealtimeSampling;

pub use realtime::RealtimeScheduleError;

pub use realtime::RealtimeScheduler;

pub use realtime::RealtimeSession;

pub use realtime::RealtimeSlotCoordinate;

pub use realtime::RealtimeSlotOccupancy;

pub use realtime::RealtimeSpeechConfig;

pub use realtime::RealtimeTargetDecision;

pub use realtime::RealtimeTargetSource;

pub use realtime::RealtimeTemporalSource;

pub use realtime::MAX_REALTIME_FRAME_DELAY;

pub use residency::BackgroundPrefetchReport;

pub use residency::PrefetchAdmission;

pub use residency::PrefetchCompletion;

pub use residency::PrefetchDemandObservation;

pub use residency::PrefetchDemandResolution;

pub use residency::PrefetchExecutionState;

pub use residency::PrefetchStateError;

pub use residency::PrefetchWork;

pub use speculative::cancel_pending_verification;

pub use speculative::propose_block;

pub use speculative::resolve_commit_and_publish;

pub use speculative::resolve_optimistic_branch;

pub use speculative::resolve_round;

pub use speculative::submit_verification_transaction;

pub use speculative::CompletedSpeculativeRequest;

pub use speculative::CompletedSpeculativeSchedule;

pub use speculative::PendingSpeculativeVerification;

pub use speculative::PreparedSpeculativeLane;

pub use speculative::ProposalDecision;

pub use speculative::PublishedSpeculativeResult;

pub use speculative::PublishedSpeculativeVerification;

pub use speculative::ResolvedSpeculativeRound;

pub use speculative::SamplingPlacement;

pub use speculative::SpeculativeAction;

pub use speculative::SpeculativeCallbackPublisher;

pub use speculative::SpeculativeCandidate;

pub use speculative::SpeculativeCapability;

pub use speculative::SpeculativeCommit;

pub use speculative::SpeculativeConstraint;

pub use speculative::SpeculativeContinuation;

pub use speculative::SpeculativeDraft;

pub use speculative::SpeculativeDraftBlock;

pub use speculative::SpeculativeDraftSource;

pub use speculative::SpeculativeDriverError;

pub use speculative::SpeculativeExecutionTopology;

pub use speculative::SpeculativeExecutor;

pub use speculative::SpeculativeGenerationBackend;

pub use speculative::SpeculativeGenerationBatchOutput;

pub use speculative::SpeculativeGenerationBatchRequest;

pub use speculative::SpeculativeGenerationLane;

pub use speculative::SpeculativeGenerationOutput;

pub use speculative::SpeculativeGenerationVisitor;

pub use speculative::SpeculativeOptimisticBranch;

pub use speculative::SpeculativeOutputError;

pub use speculative::SpeculativeOutputRuntime;

pub use speculative::SpeculativePrefill;

pub use speculative::SpeculativeProposal;

pub use speculative::SpeculativePublicationStatus;

pub use speculative::SpeculativePublisher;

pub use speculative::SpeculativeRandomness;

pub use speculative::SpeculativeRequest;

pub use speculative::SpeculativeRequestTable;

pub use speculative::SpeculativeSampling;

pub use speculative::SpeculativeSchedule;

pub use speculative::SpeculativeSchedulerStats;

pub use speculative::SpeculativeSemanticConstraint;

pub use speculative::SpeculativeSemanticState;

pub use speculative::SpeculativeStats;

pub use speculative::SpeculativeTelemetry;

pub use topology::balanced_contiguous_range;

pub use topology::ParallelAxis;

pub use topology::ParallelCoordinates;

pub use topology::ParallelRankTopology;

pub use topology::ParallelTopology;

pub use topology::SubgroupMembership;

pub use topology::TopologyError;

pub use topology::TopologyPreflightReport;

Modules

artifact

Portable artifact inspection and model-preparation planning. Backend-neutral model artifact inspection and preparation planning.

attention

Validated decoder attention schedules. Architecture-neutral decoder layer schedules and attention geometry.

automatic

Backend-neutral automatic execution planning. Backend-neutral automatic execution planning.

backend

High-level execution-backend contract. High-level contract implemented once per execution backend.

cache

Aggregate ownership and admission for backend-managed live caches. Portable cache/state policy and prompt-cache schemas.

capability

Portable model capabilities, runtime-state accounting, and admission policy. Portable model capabilities, runtime-state accounting, and admission policy.

checkpoint

Neutral checkpoint tensor descriptions and validation. Neutral checkpoint tensor catalog contracts.

consensus

Backend-neutral distributed scheduler consensus. Backend-neutral distributed scheduler consensus.

execution

Portable execution plans, capabilities, and telemetry. Portable execution-plan and telemetry schemas.

generation

Backend-independent generation lifecycle and output events. Portable generation configuration, lifecycle, and speculative bookkeeping.

input

Portable identity for ordered, prepared model input. Portable identity for ordered, prepared model input.

inspection

Portable model-artifact inspection results. Portable model-artifact inspection results.

media

Portable decoded-media requests and backend preparation inputs. Portable decoded-media requests and chat-placeholder composition.

observation

Portable, explicitly requested execution observations. Portable, explicitly requested execution observations.

realtime

Backend-generic realtime token-session execution and scheduling. Backend-generic realtime token-session execution and scheduling.

residency

Weight-residency ownership, capacity, and resource planning. Backend-neutral policy, ownership, capacity, accounting, and telemetry for weight residency.

scheduler

Transactional fair work scheduler. Fair transactional scheduling independent of event and tensor runtimes.

speculative

High-level speculative execution contracts and orchestration. High-level contracts and orchestration for speculative execution backends.

topology

Parallel topology and placement planning. Pure parallel topology coordinates and membership validation.