Attributes

59 public types in this namespace, organized by kind.

Enums (5)

Type Summary
ModelDimensionRole Architectural roles that participate in cross-parameter dimension constraints.
RecipeProvenance How a declared hyperparameter was arrived at in the paper that states it.
TensorLayoutDirection Whether a TensorLayoutAttribute describes an input or an output.
TensorPortDirection Identifies whether a declared tensor port receives or produces a tensor.
TrainingPhase Which stage of training a declared paper recipe describes.

Attributes (54)

Type Summary
ActivationCategoryAttribute Specifies the architectural category of an activation function.
ActivationPropertyAttribute Declares mathematical properties of an activation function for automatic test generation and cataloging.
ActivationTaskAttribute Specifies a task where this activation function is commonly used.
ArchitectureFromPaperAttribute Declares that a model’s layers come from a factory built for another paper’s architecture, and names that paper.
AutoParametersAttribute Marks a layer that has migrated to generated parameter plumbing.
BufferAttribute Marks a tensor field as persistent NON-TRAINABLE state: saved with the model, never touched by the optimizer.
ComponentDependencyAttribute Declares a dependency that a component requires from another component or interface.
ComponentTypeAttribute Specifies the type of an AI pipeline component (Tier 2 metadata).
CustomSerializationFormatAttribute Marks a type whose public checkpoint format has semantics that cannot be represented by the ordinary generated state envelope.
DimensionDivisibilityAttribute Declares that one integer model dimension must be evenly divisible by another integer configuration property.
DistinctImplementationOfAttribute Declares that a model and another model citing the same paper under the same name are different networks.
ElementWiseShapeAttribute Declares that a layer changes VALUES but never SHAPE: whatever tensor goes in, a tensor of exactly the same dimensions comes out, at any rank.
ExternalStateAttribute Marks numeric state owned by an external runtime.
FittedParameterAttribute Marks mathematical model parameters that become available through Fit rather than a gradient step.
FrozenParameterAttribute Marks mathematical parameters that participate in the parameter surface but are excluded from optimizer updates.
GenerateInputContractAttribute Marks the one method that contains a layer’s unique forward logic.
InfraTypeAttribute Specifies the type of infrastructure component (Tier 3 metadata).
LayerCategoryAttribute Specifies the architectural category of a neural network layer.
LayerPropertyAttribute Declares architectural properties of a neural network layer for cataloging and test generation.
LayerStateAttribute Marks a constructor parameter as state that must survive a serialization round-trip, so the layer can be rebuilt exactly rather than guessed at from its shape.
LayerTaskAttribute Specifies a task that a neural network layer performs.
LossCategoryAttribute Specifies the category of a loss function.
LossPropertyAttribute Declares mathematical properties of a loss function for automatic test generation and cataloging.
LossTaskAttribute Specifies a task this loss function is designed for.
ModelCategoryAttribute Specifies the algorithm family or category that a model belongs to.
ModelComplexityAttribute Specifies the computational complexity and resource requirements of a model.
ModelDimensionRoleAttribute Identifies the architectural role of an integer constructor parameter.
ModelDomainAttribute Specifies the application domain(s) that a model is designed for.
ModelInputAttribute Specifies the expected input and output types for a model.
ModelInputShapeConstraintAttribute Declares minimum fixture geometry from model configuration.
ModelMetadataExemptAttribute Marks a class that implements IFullModel as exempt from model metadata validation diagnostics.
ModelTaskAttribute Specifies the task(s) that a model is designed to perform.
NotScaledAttribute Marks an options size that test-scale construction must leave at its value, because the number is fixed by data the model reads rather than chosen for capacity: a tokenizer’s vocabulary, a phoneme inventory.
OptionBehaviourExemptAttribute Declares that an options property legitimately cannot change a model’s observable behaviour at test scale, and records why, so the generated option-sensitivity test does not report it.
PaperDefaultAttribute Declares the value a model’s research paper specifies for one options property, with the place in the paper it comes from, so a generated test can check that the options constructor really ships it.
PaperOptimizerAttribute Declares the training recipe a model’s research paper specifies — the optimizer, its hyperparameters, the learning-rate schedule and gradient clipping — so that a model given no optimizer trains the way its paper says rather than at an arbi…
ParameterAliasAttribute Marks a field or property as another view of parameter storage owned by a different member.
PipelineStageAttribute Specifies which stage(s) of an AI pipeline a component operates in.
PreprocessesInputAttribute Declares that the model transforms its input before the layer stack sees it, so the tensor reaching Layers[0] is NOT the tensor the caller passed to Predict.
RankRoutedInputDomainAttribute Declares a rank-routed model input.
ResearchPaperAttribute Specifies the academic paper(s) that introduced or describe a model, component, or algorithm.
ScalarKernelAttribute Marks a member whose element-by-element tensor access is deliberate, exempting it from the ADNELEM001/ADNELEM002 element-wise gate.
ScratchAttribute Marks a tensor field as transient working state: not trained, not counted, not saved.
ShapeContractRequiresPropertyOverrideAttribute Marks a symbolic shape-contract implementation that is available only when a concrete model overrides a named virtual property supplying model-specific shape metadata.
ShapeContractUnavailableAttribute Marks an IShapeContract.OutputAxesFor implementation that deliberately provides no symbolic output-shape law.
SharedLayerReferenceAttribute Marks a layer field that refers to a layer another part of the network owns, such as an output head that shares the token embedding’s table.
StackInputLayoutAttribute Declares the tensor layout produced by a model’s preprocessing step and consumed by Layers[0].
SubLayerInputAttribute Declares the input shape a composite layer feeds to one of its sub-layers, so the shape can be generated rather than hand-written.
TaskVariantOfAttribute Declares that a model is the same network as another model of the same paper, adapted to a different task.
TensorInputAttribute Refines the generated contract for a tensor method parameter.
TensorLayoutAttribute Declares one tensor layout a layer or model accepts as input, or produces as output, as an ordered list of axis ROLES.
TensorPortAttribute Declares a layer tensor port once so its runtime metadata, value-domain validation and generated diagnostics cannot drift into separate hand-written implementations.
TrainableParameterAttribute Marks a Tensor field as a trainable parameter that should be registered with the gradient tape training system.
ValidateSequentialLayerDomainsAttribute Compatibility marker retained for existing layer factories.