1#![warn(missing_docs)]
7
8pub mod backend;
10pub mod cache;
12pub mod component;
14pub mod decision;
16pub mod dense;
18pub mod draft;
20pub mod execution;
22pub mod expert;
23pub mod generation;
25pub mod input;
27pub mod inspection;
28pub mod layered;
30pub mod parallel;
32pub mod parameter;
34pub mod partition;
36pub mod prefetch;
38pub mod realtime;
40pub mod replicated_text;
42pub mod residency;
44pub mod speculative;
46pub mod state;
48mod weight_residency;
49
50pub use backend::{CollectiveBackend, ParameterBackend, SubmissionBackend, TransferBackend};
51pub use cache::{
52 finalize_prompt_cache_shard, hash_prompt_cache_shard_payload, inspect_prompt_cache,
53 resolve_prompt_cache_root, safe_prompt_cache_shard_path, validate_prompt_cache_manifest,
54 CacheBlockLifecycle, CacheBlockStorage, CacheHostDemotionOperation, CacheHostPromotion,
55 CacheIoAdmission, CacheIoCompletionDisposition, CacheIoExecutionState,
56 CacheIoExecutionStateError, CacheIoOperation, CacheIoOperationKey, CacheIoOperationKind,
57 CacheIoPreparation, CacheIoStartDisposition, CacheIoSubmission, CacheIoSubmissionOutcome,
58 CacheIoTicket, CacheIoWorker, CacheIoWorkerError, CacheLayerResidencyReport,
59 CacheLayerResidencyStats, CacheLifecycleError, CachePoolError, CachePoolLimits,
60 CachePoolMembership, CachePoolReport, CachePoolReservation, CachePoolResource, CachePoolUsage,
61 CacheResidencyConfigurationError, CacheResidencyPolicy, CacheResidencyPool,
62 CacheResidencyReport, CacheResidencyTelemetry, CacheStorageError, CacheStoragePhase,
63 LiveCacheBlockPublication, LiveCacheDiskPolicy, LiveCachePublicationError, MutableCacheTail,
64 PagedCacheOptions, PromptCachePersistenceError, PromptCachePublication,
65 CACHE_RESIDENCY_LAYER_REPORT_LIMIT, MAX_PROMPT_CACHE_SHARD_HEADER_BYTES,
66 PROMPT_CACHE_CURRENT_FILE, PROMPT_CACHE_GENERATIONS_DIRECTORY,
67};
68pub use component::{
69 ComponentDomain, ComponentGraph, ComponentGraphError, ComponentKind, ComponentResidencyClass,
70 ComponentSpec,
71};
72pub use decision::{
73 FullyForcedTailDecision, PredictionDirective, SequentialDecision, SequentialDecisionBoundary,
74 SequentialDecisionDiagnostic, SequentialDecisionDriver, SequentialDecisionError,
75 SequentialDecisionMode, SequentialDecisionPlan, SequentialDecisionPlanError,
76 SequentialDecisionSource, SequentialDecisionTraversal, SequentialSamplingState,
77};
78pub use dense::{
79 DenseCacheMetrics, DenseDiskStreamReport, DenseExecutionGroupReport, DensePassCounterSnapshot,
80 DensePassReport, DenseStreamTelemetry, DenseStreamTelemetryError, DenseTierResidencyReport,
81};
82pub use draft::{execute_draft_group, DraftGroupExecutionError, DraftStateTransaction};
83pub use execution::{
84 ExecutionGraph, ExecutionGraphError, ExecutionGroupId, ExecutionGroupSchedule,
85 ExecutionGroupSpec, ExecutionScheduleError, ExecutionUnitAddress, ExecutionUnitLayout,
86 ExecutionUnitLayoutError, ReadyGroupState,
87};
88pub use expert::{
89 combine_routed_expert_tensor_parallel, combine_tensor_parallel_expert_outputs,
90 reduce_routed_expert_tensor_parallel, reduce_tensor_parallel_expert_output,
91 AddressableGatedProductBank, ObservedExpertProvider, ObservedExpertProviderError,
92 ResidentExpertProvider, RoutedExpertProvider, RoutedExpertRequest,
93 RoutedExpertTensorParallelOutput, RoutedObservationPoint, TensorParallelRoutedExpertProvider,
94};
95pub use generation::{
96 CausalModel, ConstrainedSampler, DefaultSampler, GenerationSampler, MirostatV2Sampler,
97 PenaltyConfig, Sampler, SamplingBackend, SamplingConfigurationError, SpeculativeSampler,
98 TokenDomain,
99};
100pub use input::{PreparedInputPart, PreparedInputPayload, PreparedModelInput};
101pub use inspection::{
102 observe_and_intervene, observe_model_logits, ActivationObserver, NoopObserver,
103 RoutingObservation, TargetStateCapture, TargetStateCaptureError, TargetStateTap,
104};
105pub use layered::{
106 ArchitectureGroupKind, ArchitectureGroupPlacement, ArchitectureGroupTransport,
107 ArchitectureMergeDestination, ArchitectureParallelSubgroup, ArchitectureParameters,
108 CompositeLayeredTraversalHook, LayeredArchitecture, LayeredForwardState, LayeredPartitionInput,
109 LayeredPartitionOutput, LayeredPipelineSchedule, LayeredPipelineScheduleError,
110 LayeredTraversalHook, LayeredTraversalPoint, LayeredUnitAction, LayerwiseAcquireError,
111 LayerwisePolicy, LayerwiseRuntime, LayerwiseRuntimeError, ParallelLayeredArchitecture,
112 ParallelRoutedLayeredArchitecture, PartitionedLayeredArchitecture, ResidentRuntime,
113 ResidentUnitWindow, ResidentUnitWindowError, RoutedLayeredArchitecture, StaticParameterVisitor,
114 StaticParameterVisitorMut,
115};
116pub use parallel::{
117 aligned_partition_units, expand_linear_format_parameter_groups, module_parameter_group,
118 partitioned_module_parameter_group, partitioned_projection_group, projection_parameter_group,
119 segmented_projection_group, LocalModelLayout, LocalTensorLayout, MemberSharding,
120 ParallelModelInfo, ParallelPlanError, ParameterGroupSpec, ParameterMemberSpec, ParameterRole,
121 ProjectionSharding, ShardingPolicy, TensorPlacement,
122};
123pub use parameter::{
124 bind_materialized_unit, bindings_from_recipe_set, materialize_bindings, MaterializedUnit,
125 ParameterOrchestrationError, RecipeBindingError,
126};
127pub use partition::{
128 validate_boundary_tensor_count, ArchitectureBoundary, ArchitectureBoundaryError,
129 ArchitectureParameterDescription, ArchitectureParameterError, ArchitecturePartition,
130 ArchitecturePartitionError, BoundaryTensorDimension, BoundaryTensorDtype, BoundaryTensorSpec,
131 BoundaryWireSchema, LayeredPartitionDriver, LayeredPartitionError, NoAuxiliaryBoundary,
132 NoAuxiliaryBoundarySchema, OwnedParameterGroupSpec, ParameterGroupOwner, PartitionGroup,
133 PartitionOwnership, PartitionState, PipelineActivationDtype, PipelineWireContract,
134 ResolvedBoundaryTensorSpec, ResolvedBoundaryWireSchema,
135};
136pub use prefetch::{BackgroundPrefetchWorker, BackgroundPrefetchWorkerError};
137pub use realtime::{
138 RealtimeCompletionAttachmentError, RealtimeFrameExecutionError, RealtimeFrameTransition,
139 RealtimeGenerationBranch, RealtimeGenerationState, RealtimeGenerationTransactionError,
140};
141pub use replicated_text::{
142 select_replicated_text_realization, BackendMechanismCapabilities, GroupedOperationRequirement,
143 ParameterTransformConstraint, ParameterTransformTarget, ReplicatedTextArchitecture,
144 ReplicatedTextContractError, ReplicatedTextParameterOwner, ReplicatedTextParameterPresence,
145 ReplicatedTextParameterRequirement, ReplicatedTextParameterRole, ReplicatedTextPhysicalSource,
146 ReplicatedTextRequirements, ReplicatedTextSelectionError, ReplicatedTextSelectionRequest,
147 SelectedParameterRealization, SelectedReplicatedTextRealization, StateResidencyMechanism,
148 WeightLoweringCapability, WeightLoweringDescriptor, WeightLoweringKind,
149 WeightResidencyMechanism,
150};
151pub use residency::{
152 DeviceLayerWindow, OffloadUnit, ResidencyAcquisition, ResidencyController,
153 ResidencyControllerError, ResidencyDeclarationError, ResidencyLease, ResidencyLeaseOwner,
154 ResidencyLeaseStorage, ResidencyReport, ResidencyTransfer, ResidencyTransferOwner,
155 ResidencyWindowError, ResidencyWindowManager, ResidentLayerGroup, ResidentLayerGroupReport,
156 WeightBinding, WeightBindingPlan, WeightBindingSelectionError, WeightMaterializationReport,
157};
158pub use speculative::{RunSpeculativeGeneration, SpeculativeScheduler};
159pub use state::{
160 realize_architecture_state, ArchitectureStateFactory, ArchitectureStatePartitionError,
161 ArchitectureStatePartitionPlan, ArchitectureStatePartitionRule, ArchitectureStatePlacement,
162 ArchitectureStateRealizationError, DeviceState, LayerRuntimeState, ModelStateIdentity,
163 ResettableRuntimeLayerState, ResettableRuntimeState, RuntimeLayerState, RuntimeState,
164 RuntimeStateComponents, StateError, StateLayout, StateSegmentId, StateSegmentLifetime,
165 StateSegmentSpec, DEFAULT_STATE_SEGMENT_ID,
166};
167pub use weight_residency::{
168 DenseDiskStreamLoadOptions, DenseTransferSchedule, DenseTransferScheduleError,
169 ExecutionResidency, ExpertPass, LayerWeightResidency, LayerwiseLoadOptions,
170 LayerwiseModelMetadata, OrdinaryWeightResidency, ParameterBankKey, ParameterBankLoadOptions,
171 ParameterBankResidency, StaticUnitBindings, WeightResidency, WeightResidencyPolicyError,
172 DENSE_TRANSFER_WINDOW,
173};