1#![warn(missing_docs)]
7
8pub mod automatic_support;
10pub mod backend;
12pub mod cache;
14pub mod communication;
16pub mod component;
18pub mod composite;
20pub mod decision;
22pub mod dense;
24pub mod draft;
26pub mod execution;
28pub mod execution_plan;
29pub mod expert;
30pub mod generation;
32pub mod input;
34pub mod inspection;
35pub mod layered;
37pub mod load_request;
39pub mod mechanism_synthesis;
41pub mod parallel;
43pub mod parameter;
45pub mod partition;
47pub mod partitioned_execution;
49pub mod placement;
50pub mod prefetch;
52pub mod realtime;
54pub mod realtime_executor;
56pub mod realtime_ingress;
58pub mod realtime_interpreter;
60pub mod realtime_mechanism_synthesis;
61pub mod realtime_model;
63pub mod realtime_payload;
65pub mod realtime_payload_state;
67pub mod realtime_selection;
69pub mod realtime_session;
71pub mod replicated_session;
73pub mod replicated_text;
75pub mod residency;
77pub mod speculative;
79pub mod speculative_selection;
81pub mod state;
83mod weight_residency;
84
85pub use automatic_support::{
86 residency_telemetry, selected_text_bounded_requirement, BoundedResidencySizingError,
87};
88pub use backend::{
89 BarrierBackend, BroadcastBackend, CollectiveBackend, CommunicationBackend, EvenGatherBackend,
90 FailureAgreementBackend, ParameterBackend, PointToPointBackend, RoleExactBoundaryValue,
91 SubmissionBackend, SumReductionBackend, TransferBackend, UnevenGatherBackend,
92 VariableAllToAllBackend,
93};
94pub use cache::{
95 finalize_prompt_cache_shard, hash_prompt_cache_shard_payload, inspect_prompt_cache,
96 prompt_cache_rank_path, resolve_prompt_cache_root, safe_prompt_cache_shard_path,
97 validate_prompt_cache_manifest, CacheBlockLifecycle, CacheBlockStorage,
98 CacheHostDemotionOperation, CacheHostPromotion, CacheIoAdmission, CacheIoCompletionDisposition,
99 CacheIoExecutionState, CacheIoExecutionStateError, CacheIoOperation, CacheIoOperationKey,
100 CacheIoOperationKind, CacheIoPreparation, CacheIoStartDisposition, CacheIoSubmission,
101 CacheIoSubmissionOutcome, CacheIoTicket, CacheIoWorker, CacheIoWorkerError,
102 CacheLayerResidencyReport, CacheLayerResidencyStats, CacheLifecycleError, CachePoolError,
103 CachePoolLimits, CachePoolMembership, CachePoolReport, CachePoolReservation, CachePoolResource,
104 CachePoolUsage, CacheResidencyConfigurationError, CacheResidencyPolicy, CacheResidencyPool,
105 CacheResidencyReport, CacheResidencyTelemetry, CacheStorageError, CacheStoragePhase,
106 LiveCacheBlockPublication, LiveCacheDiskPolicy, LiveCachePublicationError, MutableCacheTail,
107 PagedCacheOptions, PromptCachePersistenceError, PromptCachePublication,
108 ReversiblePromptCachePublication, CACHE_RESIDENCY_LAYER_REPORT_LIMIT,
109 MAX_PROMPT_CACHE_SHARD_HEADER_BYTES, PROMPT_CACHE_CURRENT_FILE,
110 PROMPT_CACHE_GENERATIONS_DIRECTORY,
111};
112pub use communication::validate_communication_manifest_consensus;
113pub use communication::{
114 prepare_communication_realization, project_all_communication_manifests,
115 project_communication_manifest, validate_compatible_communication_manifests,
116 BoundaryDimensionContract, BoundaryFramingProtocol, BoundaryRoleContract,
117 CommunicationCapabilities, CommunicationCapabilityError, CommunicationCompletionCapabilities,
118 CommunicationCompletionPolicy, CommunicationGroupDescriptor, CommunicationGroupRequirements,
119 CommunicationManifest, CommunicationManifestConsensusError, CommunicationManifestError,
120 CommunicationOperation, CommunicationOperationRequirement, CommunicationPeerCounts,
121 CommunicationRealizationError, CommunicationRouteDescriptor, CommunicationRouteId,
122 CommunicationTensorLimits, CommunicationTopologyCapabilities, PreparedCommunicationRealization,
123 RoleExactBoundaryContract, TopologyCommunicationPlan,
124};
125pub use component::{
126 ComponentDomain, ComponentGraph, ComponentGraphError, ComponentKind, ComponentResidencyClass,
127 ComponentSpec,
128};
129pub use composite::{
130 select_composite_realization, select_processor_execution, CompositeSelectionError,
131 MediaPrimitiveCapabilities, ModalityProcessorRequirements, ProcessorExecutionRequirements,
132 ProcessorPrimitive, ProcessorSelectionError, ProcessorSelectionRequest,
133 SelectedCompositeRealization, SelectedProcessorExecution,
134};
135pub use decision::{
136 FullyForcedTailDecision, PredictionDirective, SequentialDecision, SequentialDecisionBoundary,
137 SequentialDecisionDiagnostic, SequentialDecisionDriver, SequentialDecisionError,
138 SequentialDecisionMode, SequentialDecisionPlan, SequentialDecisionPlanError,
139 SequentialDecisionSource, SequentialDecisionTraversal, SequentialSamplingState,
140};
141pub use dense::{
142 DenseCacheMetrics, DenseDiskStreamReport, DenseExecutionGroupReport, DensePassCounterSnapshot,
143 DensePassReport, DenseStreamTelemetry, DenseStreamTelemetryError, DenseTierResidencyReport,
144};
145pub use draft::{execute_draft_group, DraftGroupExecutionError, DraftStateTransaction};
146pub use execution::{
147 ExecutionGraph, ExecutionGraphError, ExecutionGroupId, ExecutionGroupSchedule,
148 ExecutionGroupSpec, ExecutionScheduleError, ExecutionUnitAddress, ExecutionUnitLayout,
149 ExecutionUnitLayoutError, ReadyGroupState,
150};
151pub use execution_plan::{
152 execution_plan_quantization, ExecutionPlanLoadError, ResidencyDiagnostics,
153};
154pub use expert::{
155 combine_routed_expert_tensor_parallel, combine_tensor_parallel_expert_outputs,
156 plan_addressable_bank_bindings, reduce_routed_expert_tensor_parallel,
157 reduce_tensor_parallel_expert_output, selected_addressable_parameter_bytes,
158 AddressableBankBindingPlan, AddressableBankDistribution, AddressableBankMember,
159 AddressableBankMemberError, AddressableBankMemberPlacement, AddressableBankParameter,
160 AddressableBindingTransform, AddressableExpertRouteProvider, AddressableExpertRouteRequest,
161 AddressableGatedProductBank, AddressableGroupedBank, ExpertRouteCombination,
162 ExpertRouteExchange, ExpertRouteTensorMovement, IndexedMovement, ObservedExpertProvider,
163 ObservedExpertProviderError, ParameterBankAcquisition, ResidentExpertProvider,
164 RoutedExpertProvider, RoutedExpertRequest, RoutedExpertTensorParallelOutput,
165 RoutedObservationPoint, TensorParallelRoutedExpertProvider,
166};
167pub use generation::{
168 CausalModel, ConstrainedSampler, DefaultSampler, GenerationSampler, MirostatV2Sampler,
169 PenaltyConfig, Sampler, SamplingBackend, SamplingConfigurationError, SpeculativeSampler,
170 TokenDomain,
171};
172pub use input::{
173 PreparedInputCacheIdentity, PreparedInputCacheIdentityError, PreparedInputInspector,
174 PreparedInputPart, PreparedInputPayload, PreparedModelInput,
175};
176pub use inspection::{
177 observe_and_intervene, observe_model_logits, ActivationObserver, NoopObserver,
178 RoutingObservation, TargetStateCapture, TargetStateCaptureError, TargetStateTap,
179};
180pub use layered::{
181 ArchitectureGroupKind, ArchitectureGroupPlacement, ArchitectureGroupTransport,
182 ArchitectureMergeDestination, ArchitectureParallelSubgroup, ArchitectureParameters,
183 CompositeLayeredTraversalHook, LayeredArchitecture, LayeredForwardState, LayeredPartitionInput,
184 LayeredPartitionOutput, LayeredPipelineSchedule, LayeredPipelineScheduleError,
185 LayeredTraversalHook, LayeredTraversalPoint, LayeredUnitAction, LayerwiseAcquireError,
186 LayerwisePolicy, LayerwisePolicyForward, LayerwiseRuntime, LayerwiseRuntimeError,
187 ParallelLayeredArchitecture, ParallelRoutedLayeredArchitecture, PartitionedLayeredArchitecture,
188 ResidentRuntime, ResidentUnitWindow, ResidentUnitWindowError, RoutedLayeredArchitecture,
189 StaticParameterVisitor, StaticParameterVisitorMut,
190};
191pub use load_request::{
192 DraftingLoadRequest, NormalizedLoadRequest, NormalizedLoadRequestError, ParallelLoadRequest,
193 ValidatedModelLoadRequest,
194};
195pub use mechanism_synthesis::{
196 synthesize_replicated_text_capabilities, BackendMechanismFacts, ReplicatedTextMechanismSupport,
197 StateLifecycleCapabilities,
198};
199pub use parallel::{
200 aligned_partition_units, expand_linear_format_parameter_groups, module_parameter_group,
201 partitioned_module_parameter_group, partitioned_projection_group, projection_parameter_group,
202 segmented_projection_group, LocalModelLayout, LocalTensorLayout, MemberSharding,
203 ParallelModelInfo, ParallelPlanError, ParameterGroupSpec, ParameterMemberSpec, ParameterRole,
204 ProjectionSharding, ShardingPolicy, TensorPlacement,
205};
206pub use parameter::{
207 bind_materialized_unit, bind_materialized_unit_excluding, bindings_from_recipe_set,
208 build_exact_replicated_text_bindings, build_module_binding_plan, materialize_bindings,
209 materialize_selected_bindings, preflight_bindings, select_bindings, BindingPlan,
210 BindingPlanError, MaterializedUnit, ModuleBindingPlan, ModuleBindingPlanError,
211 ParameterBindingTarget, ParameterOrchestrationError, PlannedBinding, RecipeBindingError,
212 SelectedBindingPlan,
213};
214pub use partition::{
215 validate_boundary_tensor_count, ArchitectureBoundary, ArchitectureBoundaryError,
216 ArchitectureBoundaryValue, ArchitectureParameterDescription, ArchitectureParameterError,
217 ArchitecturePartition, ArchitecturePartitionError, BoundaryTensorDimension,
218 BoundaryTensorDtype, BoundaryTensorSpec, BoundaryWireSchema, LayeredPartitionBeginError,
219 LayeredPartitionDriver, LayeredPartitionError, NoAuxiliaryBoundary, NoAuxiliaryBoundarySchema,
220 OwnedParameterGroupSpec, ParameterGroupOwner, PartitionGroup, PartitionOwnership,
221 PartitionState, PipelineActivationDtype, PipelineWireContract, ResolvedBoundaryTensorSpec,
222 ResolvedBoundaryWireSchema,
223};
224pub use partitioned_execution::{
225 CommunicationTensorMetadata, DistributedExecutionPhase, LayerwiseTraversalPartitionExecutor,
226 LayerwiseTraversalRuntime, NoBoundaryTransport, NoCommitAgreement, NoOutputPublisher,
227 OpaqueBoundaryTransport, OpaqueCommitAgreement, OpaqueFailureAgreement, OpaqueOutputPublisher,
228 PartitionBoundaryRoute, PartitionBoundaryTransport, PartitionCommitAgreement,
229 PartitionCommunication, PartitionCommunicationAuthority, PartitionExecutionError,
230 PartitionOutputAuthority, PartitionOutputPublication, PartitionOutputPublisher,
231 PartitionedExecutionPlan, PartitionedGroupExecutor, PartitionedTextExecution,
232 PartitionedTextRuntime, PartitionedTraversalError, PartitionedTraversalResult,
233 RealizedCommunicationGroup, RealizedCommunicationRoute,
234};
235pub use placement::{
236 place_addressable_member_bindings, place_weight_bindings, placement_selection,
237 BindingPlacementError, PlacementPlan, PlacementPlanError, PlacementRank,
238 ResolvedTensorPlacement, TensorSlice,
239};
240pub use prefetch::{BackgroundPrefetchWorker, BackgroundPrefetchWorkerError};
241pub use realtime::{
242 RealtimeCompletionAttachmentError, RealtimeFrameExecutionError, RealtimeFrameTransition,
243 RealtimeGenerationBranch, RealtimeGenerationState, RealtimeGenerationTransactionError,
244};
245pub use realtime_executor::{
246 execute_realtime_frame, PreparedRealtimeFrameExecutor, PrepublicationRealtimeFrame,
247 RealtimeCompletionCreationError, RealtimeDecisionExecution, RealtimeFrameCompletionMechanism,
248 RealtimeFrameCoordinatorError, RealtimeFrameHostObserver, RealtimeHostOutputUnavailable,
249 RealtimePrepublicationError, SubmittedRealtimeFrame,
250};
251pub use realtime_ingress::{
252 MaterializedRealtimeInput, RealtimeHostTokenMaterializer, RealtimeIngressContract,
253 RealtimeIngressError, RealtimePayloadKind, RealtimeTokenKind, ValidatedRealtimeInput,
254};
255pub use realtime_interpreter::{
256 complete_realtime_frame, prepare_realtime_frame, CompletedRealtimeFrame, PreparedRealtimeFrame,
257 RealtimeFrameInterpretationError, RealtimeFrameTensorMechanisms,
258};
259pub use realtime_mechanism_synthesis::{
260 synthesize_realtime_capabilities, RealtimeMechanismFacts, RealtimeMechanismSupport,
261};
262pub use realtime_model::{
263 construct_realtime_model, preflight_realtime_materialization_tasks, realtime_task_binding_plan,
264 ConstructedRealtimeExecution, ConstructedRealtimeModel, PreparedRealtimeModelContract,
265 RealizedRealtimePolicy, RealizedRealtimeState, RealtimeArchitectureConstructionIdentity,
266 RealtimeArchitectureIdentity, RealtimeLayerwiseRuntime, RealtimeMaterializationComponent,
267 RealtimeMaterializationTask, RealtimeModelConstructionError,
268 RealtimeModelConstructionMechanisms, RealtimeModelContractError, RealtimeTaskBindingPlan,
269};
270pub use realtime_payload::{
271 RealtimePayloadContract, RealtimePayloadContractError, RealtimePayloadEnvelope,
272 RealtimePayloadGeneration, RealtimePayloadHistory, RealtimePayloadHistoryError,
273 RealtimePayloadOwnerIdentity,
274};
275pub use realtime_payload_state::{
276 RealtimePayloadBranch, RealtimePayloadState, RealtimePayloadStateTransactionError,
277};
278pub use realtime_selection::{
279 select_and_prepare_realtime_realization, select_realtime_realization,
280 ConstructedRealtimeResources, PreparedRealtimeRealization, RealtimeArchitectureProof,
281 RealtimeArchitectureRequirements, RealtimeContractError, RealtimeExecutionRequirements,
282 RealtimeIdentity, RealtimeMechanism, RealtimeMechanismCapabilities,
283 RealtimeMechanismRequirements, RealtimeObservationRequirements, RealtimePreparationError,
284 RealtimeSelectionError, RealtimeSelectionIssue, RealtimeSelectionRequest,
285 RealtimeTopologyPolicy, RealtimeWeightComponentRequirement, RealtimeWeightComponentRole,
286 RealtimeWeightLoweringRequirement, SelectedRealtimeRealization,
287 SelectedRealtimeStateComponentRealization, SelectedRealtimeStateRealization,
288};
289pub use realtime_session::{
290 RealtimeHistoryGeneration, RealtimeModelOwnerIdentity, RealtimeModelSessionIdentity,
291 RealtimeSamplingReplacementError, RealtimeSamplingUpdateError, RealtimeSessionBranch,
292 RealtimeSessionError, RealtimeSessionExecutionError, RealtimeSessionIncarnation,
293 RealtimeSessionResumeError, RealtimeSessionScheduler, RealtimeSessionState,
294 RealtimeSessionTransactionError, ReleasedRealtimeSession,
295};
296pub use replicated_session::{
297 construct_replicated_text_session, construct_replicated_text_session_with_execution,
298 construct_replicated_text_session_with_runtime, prepare_default_partitioned_runtime,
299 prepare_layered_text_contract, prepare_layered_text_contract_with_addressable_parameters,
300 prepare_partitioned_session_runtime, prepare_partitioned_session_runtime_with_exclusions,
301 prepare_replicated_text_contract, prepare_replicated_text_contract_with_addressable_parameters,
302 DirectReplicatedTextExecution, DistributedSessionCheckpoint, DistributedStateCheckpoint,
303 PartitionedRuntimeConstructionError, PartitionedSessionFactoryInput,
304 PartitionedSessionPreparationError, PartitionedUnitScope, PredictionTargetOperation,
305 PreparedPartitionedRuntimeComponents, PreparedPartitionedSessionRuntime,
306 PreparedReplicatedTextContract, ReplicatedRuntimeExecutionStrategy,
307 ReplicatedTextExecutionStrategy, ReplicatedTextRuntime, ReplicatedTextSession,
308 ReplicatedTextSessionCheckpoint, ReplicatedTextSessionError, ReplicatedTextSessionMechanisms,
309 ReplicatedTextSessionReport, RoutedReplicatedTextExecution, SessionStateRealization,
310 TransactionalPromptCacheMechanisms,
311};
312pub use replicated_text::{
313 group_replicated_text_transform_tasks, locally_materialized_replicated_text_outputs,
314 partition_selected_replicated_text_materialization_tasks,
315 partitioned_replicated_text_materialization_tasks,
316 plan_local_replicated_text_materialization_tasks, plan_replicated_text_materialization_tasks,
317 replicated_text_materialization_tasks, select_replicated_text_realization,
318 selected_materialization_task_bytes, AddressableStorageCapabilities, AddressableStorageTiers,
319 BackendMechanismCapabilities, GroupedOperationRequirement, ParameterTransformConstraint,
320 ParameterTransformTarget, ReplicatedTextArchitecture, ReplicatedTextContractError,
321 ReplicatedTextMaterializationPartitionPlan, ReplicatedTextMaterializationTask,
322 ReplicatedTextOutputCompanion, ReplicatedTextOutputSelection, ReplicatedTextParameterOwner,
323 ReplicatedTextParameterPresence, ReplicatedTextParameterRequirement,
324 ReplicatedTextParameterRole, ReplicatedTextPhysicalSource, ReplicatedTextRequirements,
325 ReplicatedTextSelectionError, ReplicatedTextSelectionRequest, ReplicatedTextStateAccess,
326 ReplicatedTextTransformGroup, SelectedParameterRealization, SelectedReplicatedTextRealization,
327 SelectedStateComponentRealization, SelectedStateRealization, StateComponentMechanism,
328 StateComponentPlacement, StateMechanismCapabilities, StateStorageDtype,
329 WeightLoweringCapability, WeightLoweringDescriptor, WeightLoweringKind,
330 WeightResidencyMechanism,
331};
332pub use residency::{
333 DeviceLayerWindow, OffloadUnit, QuantizationCompanionBindings, ResidencyAcquisition,
334 ResidencyController, ResidencyControllerError, ResidencyDeclarationError, ResidencyLease,
335 ResidencyLeaseOwner, ResidencyLeaseStorage, ResidencyReport, ResidencyTransfer,
336 ResidencyTransferOwner, ResidencyWindowError, ResidencyWindowManager, ResidentLayerGroup,
337 ResidentLayerGroupReport, WeightBinding, WeightBindingPlan, WeightBindingSelectionError,
338 WeightMaterializationReport,
339};
340pub use speculative::{RunSpeculativeGeneration, SpeculativeScheduler};
341pub use speculative_selection::{
342 select_and_prepare_speculative_realization,
343 select_and_prepare_speculative_realization_observed, select_speculative_realization,
344 ConstructedSpeculativeResources, PreparedSpeculativeRealization, SelectedSpeculativeCompletion,
345 SelectedSpeculativePlacement, SelectedSpeculativeRealization, SelectedSpeculativeSampling,
346 SelectedSpeculativeState, SpeculativeArchitectureCompatibilityProof, SpeculativeCaptureEntry,
347 SpeculativeCaptureEnvelope, SpeculativeCaptureError, SpeculativeCaptureMetadata,
348 SpeculativeCaptureSchema, SpeculativeContractError, SpeculativeIdentity,
349 SpeculativeLaneIdentity, SpeculativeMechanism, SpeculativeMechanismCapabilities,
350 SpeculativeMechanismRequirements, SpeculativePlacementRequest, SpeculativePreparationError,
351 SpeculativeRealizationRequirements, SpeculativeSelectionError, SpeculativeSelectionRequest,
352 SpeculativeStateCacheIdentityIngredients, SpeculativeStrategyClass,
353 SpeculativeStrategyRequirements,
354};
355pub use state::{
356 realize_architecture_state, ArchitectureStateFactory, ArchitectureStatePartitionError,
357 ArchitectureStatePartitionPlan, ArchitectureStatePartitionRule, ArchitectureStatePlacement,
358 ArchitectureStateRealizationError, DeviceState, LayerRuntimeState, ModelStateIdentity,
359 ResettableRuntimeLayerState, ResettableRuntimeState, RuntimeLayerState, RuntimeState,
360 RuntimeStateComponents, StateError, StateLayout, StateSegmentId, StateSegmentLifetime,
361 StateSegmentSpec, DEFAULT_STATE_SEGMENT_ID,
362};
363pub use weight_residency::{
364 DenseDiskStreamLoadOptions, DenseTransferSchedule, DenseTransferScheduleError,
365 ExecutionResidency, ExpertPass, LayerWeightResidency, LayerwiseLoadOptions,
366 LayerwiseModelMetadata, OrdinaryWeightResidency, ParameterBankAccess, ParameterBankKey,
367 ParameterBankLoadOptions, ParameterBankResidency, StaticUnitBindings, WeightResidency,
368 WeightResidencyPolicyError, DENSE_TRANSFER_WINDOW,
369};