Routes each token to its top_k experts. More...

Inheritance diagram for Mila::Dnn::Router< TDeviceType, TPrecision >:

Classes
struct	Routing
	One call's result. Both views are router-owned and valid until the next forward(). More...

Public Types
using	CompositeComponentBase = CompositeComponent<TDeviceType, TPrecision>
using	IndexTensorType = Tensor<TensorDataType::INT32, MR>
using	LinearType = Linear<TDeviceType, TPrecision>
using	MR = typename DeviceTypeTraits<TDeviceType>::memory_resource
using	RmsNormType = RmsNorm<TDeviceType, TPrecision>
using	TensorType = Tensor<TPrecision, MR>
Public Types inherited from Mila::Dnn::CompositeComponent< TDeviceType, TPrecision >
using	ComponentBase = Component<TDeviceType, TPrecision>
using	ComponentPtr = std::shared_ptr<Component<TDeviceType, TPrecision>>

Public Member Functions
	Router (const std::string &name, const RouterConfig &config, std::optional< DeviceId > device_id=std::nullopt)
Routing	forward (const TensorType &input)
	Select experts for every token of `input`.
MemoryStats	getMemoryStats () const override
	Return the current memory allocation breakdown for this component.
std::vector< ObservableStage >	getObservableStages () const override
	Stages this component publishes, and the passes each is published on.
std::vector< std::string >	getParameterNames () const override
	List all available parameter names for this component.
MemoryStats	getRequiredMemory (const BuildContext &context) const override
	What onBuilding() would allocate for this context, without allocating.
const ComponentType	getType () const override
	Get the component type identifier.
void	loadParameter (const std::string &name, const ITensorBlob &blob) override
	Load a parameter from serialized tensor data.
dim_t	parameterCount () const override
	The projection and the two owned vectors; the norm's derived weight is not counted.
void	saveFlatTensors (Serialization::SafeTensorsWriter &writer, const std::string &prefix, Serialization::TensorSavePass pass) const override
	The checkpoint's vocabulary: the projection, then the two owned vectors.
std::string	toString () const override
	Generate a human-readable description.
Public Member Functions inherited from Mila::Dnn::CompositeComponent< TDeviceType, TPrecision >
	CompositeComponent (CompositeComponent &&) noexcept=default
	CompositeComponent (const CompositeComponent &)=delete
	CompositeComponent (const std::string &name)
	Construct composite component with name.
CompositeComponent &	addComponent (ComponentPtr component)
	Add a pre-constructed child component (chainable).
size_t	childCount () const noexcept
	Get the number of direct children.
void	clearComponents ()
	Clear all child components.
std::vector< std::string >	componentPaths () const
	Every component path in this subtree, in traversal order.
ComponentPtr	findComponent (const std::string &path) const
	Resolve a dot-separated component path within this composite.
ComponentPtr	getComponent (const std::string &name) const
	Retrieve a direct child component by name.
const std::vector< ComponentPtr > &	getComponents () const
	Get all child components in insertion order.
DeviceId	getDeviceId () const override
	Get the compute device for this composite.
std::vector< ITensor * >	getGradients () const override
	Get all parameter gradients from all children.
std::vector< ITensor * >	getParameters () const override
	Get all parameters from all children.
bool	hasChildren () const noexcept
	Check if this composite has any children.
bool	hasComponent (const std::string &name) const
	Check if a named child component exists.
size_t	observe (std::string_view pattern, Compute::ComputePassMask passes, Compute::ActivationObserver sink)
	Attach sink to every component in this subtree whose path matches.
CompositeComponent &	operator= (CompositeComponent &&) noexcept=default
CompositeComponent &	operator= (const CompositeComponent &)=delete
bool	removeComponent (const std::string &name)
	Remove a child component by name.
void	stopObserving ()
	Detach every observer in this subtree and clear the shared sink.
void	synchronize () override
	Synchronize all child components.
ComponentPtr	tryFindComponent (const std::string &path) const
	Try to resolve a dot-separated component path within this composite.
Public Member Functions inherited from Mila::Dnn::Component< TDeviceType, TPrecision >
	Component (const std::string &name)
	Construct component with required name identifier.
virtual void	build (const BuildContext &context) final
	Build the component with the provided BuildContext (canonical overload).
IExecutionContext *	getExecutionContext () const
	Get the shared execution context.
const std::string	getName () const
	Get the component's name identifier.
Compute::ComputePassMask	getObservedPasses () const noexcept
virtual std::vector< const ITensor * >	getOutputs () const
	Return non-owning pointers to the tensors this component produces.
RuntimeMode	getRuntimeMode () const noexcept
TrainingMode	getTrainingMode () const noexcept
	The current runtime behavioral mode of this Component.
virtual bool	isBuilt () const final
	Returns true if build() has completed successfully.
bool	isInferenceMode () const noexcept
bool	isTrainingMode () const noexcept
void	setObservedPasses (Compute::ComputePassMask passes) noexcept
	Select which passes this component publishes on.
void	setTrainingMode (TrainingMode mode)
	Set the runtime behavioral mode for this Component.
virtual void	zeroGradients ()
	Clear all model-owned gradients for this component.

Protected Member Functions
void	load_ (ModelArchive &archive, SerializationMode mode) override
	Restore children from their nested scopes, mirroring save_().
void	onBuilding (const BuildContext &context) override
	Hook invoked by build() to allocate component buffers.
void	onExecutionContextSet () override
	Hook invoked after ExecutionContext is set.
void	onTrainingModeChanging (TrainingMode training_mode) override
	Hook invoked when training mode is about to change.
void	save_ (ModelArchive &archive, SerializationMode mode) const override
Protected Member Functions inherited from Mila::Dnn::CompositeComponent< TDeviceType, TPrecision >
template<typename TComponent>
std::shared_ptr< TComponent >	getComponentAs (const std::string &name) const
	Retrieve a typed child component by name.
virtual void	optimize ()
	Virtual hook for graph optimization after construction.
void	requireSerializableParameters () const override
	No-op override: a composite names no parameters of its own.
Protected Member Functions inherited from Mila::Dnn::Component< TDeviceType, TPrecision >
bool	hasExecutionContext () const noexcept
	Check if execution context has been set.
template<TensorDataType TParameterPrecision, typename TMemoryResource>
void	loadParameterFromBlob (const std::string &param_name, const Serialization::ITensorBlob &blob, Tensor< TParameterPrecision, TMemoryResource > &target, const shape_t &expected_shape)
	Load a tensor blob into a parameter tensor with validation.
void	publish (Compute::ComputePass pass, std::string_view stage, const ITensor &value) const
	Publish one activation to the observer, if this pass is being observed.
template<TensorDataType TParameterPrecision, typename TMemoryResource>
void	saveParameterToArchive (ModelArchive &archive, const std::string &parameter_name, const Tensor< TParameterPrecision, TMemoryResource > &parameter) const
	Write one parameter tensor into the archive under "tensors/<name>".
template<TensorDataType TParameterPrecision, typename TMemoryResource>
void	saveParameterToWriter (Serialization::SafeTensorsWriter &writer, const std::string &flat_name, const Tensor< TParameterPrecision, TMemoryResource > &parameter, Serialization::TensorSavePass pass) const
	Drive one parameter through one pass of a flat safetensors save.
void	setExecutionContext (IExecutionContext *context)
	Set the execution context for this component.

Additional Inherited Members
Static Public Member Functions inherited from Mila::Dnn::Component< TDeviceType, TPrecision >
static constexpr DeviceType	getDeviceType ()
	Compile-time device type for this component instance.
static constexpr TensorDataType	getPrecision () noexcept
	Compile-time tensor precision for this component instance.
Protected Types inherited from Mila::Dnn::Component< TDeviceType, TPrecision >
using	HostStagingMemoryResource = typename DeviceTypeTraits<TDeviceType>::host_staging_memory_resource
	Host memory a device-resident parameter stages through.
Protected Attributes inherited from Mila::Dnn::Component< TDeviceType, TPrecision >
BuildContext	build_context_ { shape_t{ 1 }, RuntimeMode::Training }
	The BuildContext stored at build time.

Detailed Description

template<DeviceType TDeviceType, TensorDataType TPrecision>
requires PrecisionSupportedOnDevice<TPrecision, TDeviceType>
class Mila::Dnn::Router< TDeviceType, TPrecision >

Routes each token to its top_k experts.

The unscaled RMS norm and the learned scale are expressed as one RmsNorm child whose weight is scale * hidden_size^-0.5, derived whenever scale is initialized or loaded. That weight is therefore not a parameter of its own: the flat vocabulary is the checkpoint's – proj.weight, scale, per_expert_scale – and the derived weight is never written to it.

Inference-only. The projection stays unquantized: the router is a precision holdout (MixtureOfExperts.md section 7.5).

Member Function Documentation

◆ forward()

template<DeviceType TDeviceType, TensorDataType TPrecision>

Routing Mila::Dnn::Router< TDeviceType, TPrecision >::forward ( const TensorType & input )

inline

Select experts for every token of input.

Parameters

input [..., hidden_size], the residual stream the router reads unnormalized.

Returns: Combine weights [..., top_k] and expert indices [..., top_k]. Order within a row carries no meaning: the combine is a sum over the selected experts.

◆ getMemoryStats()

template<DeviceType TDeviceType, TensorDataType TPrecision>

MemoryStats Mila::Dnn::Router< TDeviceType, TPrecision >::getMemoryStats ( ) const

inlineoverridevirtual

Return the current memory allocation breakdown for this component.

Reflects allocations at the moment of the call. The returned stats naturally track the component lifecycle:

After construction – nothing; construction allocates none After build( Inference ) – parameters + T=1 state buffers After build( Training ) – parameters + T=full state buffers After setTrainingMode( Train ) – parameters + state + gradients

For CompositeComponent and Network, the returned stats are the recursive aggregate of all child components.

May be called at any time – no lifecycle preconditions.

Returns: MemoryStats reflecting current allocations.

Implements Mila::Dnn::Component< TDeviceType, TPrecision >.

◆ getObservableStages()

template<DeviceType TDeviceType, TensorDataType TPrecision>

std::vector< ObservableStage > Mila::Dnn::Router< TDeviceType, TPrecision >::getObservableStages ( ) const

inlineoverridevirtual

Stages this component publishes, and the passes each is published on.

Lets a caller tell an uninstrumented component from a clean one before running anything. The pass qualifier is load-bearing rather than decorative: the stage set genuinely differs by pass, since a prefill can produce intermediates its decode never materializes.

Returns: Vector of stages; empty for components that publish nothing.

Reimplemented from Mila::Dnn::Component< TDeviceType, TPrecision >.

◆ getParameterNames()

template<DeviceType TDeviceType, TensorDataType TPrecision>

std::vector< std::string > Mila::Dnn::Router< TDeviceType, TPrecision >::getParameterNames ( ) const

inlineoverridevirtual

List all available parameter names for this component.

Returns an empty vector by default. Leaf components with parameters should override to return their canonical parameter name list in the same stable order used by save_() and loadParameter().

Reimplemented from Mila::Dnn::Component< TDeviceType, TPrecision >.

◆ getRequiredMemory()

template<DeviceType TDeviceType, TensorDataType TPrecision>

MemoryStats Mila::Dnn::Router< TDeviceType, TPrecision >::getRequiredMemory ( const BuildContext & context ) const

inlineoverridevirtual

What onBuilding() would allocate for this context, without allocating.

Children by name, sized as onBuilding() builds them: both read the input shape.

Reimplemented from Mila::Dnn::Component< TDeviceType, TPrecision >.

◆ getType()

template<DeviceType TDeviceType, TensorDataType TPrecision>

const ComponentType Mila::Dnn::Router< TDeviceType, TPrecision >::getType ( ) const

inlineoverridevirtual

Get the component type identifier.

Used for serialization and runtime type identification.

Returns: Component type enum value.

Implements Mila::Dnn::Component< TDeviceType, TPrecision >.

◆ load_()

template<DeviceType TDeviceType, TensorDataType TPrecision>

void Mila::Dnn::Router< TDeviceType, TPrecision >::load_	(	ModelArchive &	archive,
		SerializationMode	mode )

inlineoverrideprotectedvirtual

Restore children from their nested scopes, mirroring save_().

Walks the same children in the same order under the same scopes, so a child's blobs are read from exactly the paths they were written to. Restores into the live graph – the composite and its children must already be constructed and built.

Parameters

archive	Archive to read from, scoped to this composite.
mode	Serialization mode (passed to children).

Exceptions

std::runtime_error if the composite is not built.

Reimplemented from Mila::Dnn::CompositeComponent< TDeviceType, TPrecision >.

◆ loadParameter()

template<DeviceType TDeviceType, TensorDataType TPrecision>

void Mila::Dnn::Router< TDeviceType, TPrecision >::loadParameter	(	const std::string &	name,
		const ITensorBlob &	blob )

inlineoverridevirtual

Load a parameter from serialized tensor data.

Loads raw tensor bytes directly into an existing parameter tensor, handling precision conversion and device upload as needed.

The component validates that the blob's shape matches the parameter's expected shape, then delegates to the backend to perform:

Precision conversion (blob dtype -> parameter dtype)
Device upload (CPU bytes -> target device)

Takes the parameter name used to locate the target tensor, and a blob holding serialized tensor metadata and raw bytes.

Exceptions

std::runtime_error	if component has no parameters to load.
std::runtime_error	if blob shape doesn't match parameter shape.

Reimplemented from Mila::Dnn::Component< TDeviceType, TPrecision >.

◆ onBuilding()

template<DeviceType TDeviceType, TensorDataType TPrecision>

void Mila::Dnn::Router< TDeviceType, TPrecision >::onBuilding ( const BuildContext & context )

inlineoverrideprotectedvirtual

Hook invoked by build() to allocate component buffers.

Receives the stored BuildContext. Implementations must use config.allocationSeqLen() when sizing output buffers – this is the single call that makes Inference and Training allocate the correct buffer sizes automatically without per-component logic.

// Example -- Linear component:
shape_t out_shape =
{
    config.batchSize(),
    config.allocationSeqLen(),   // 1 for Inference, T for Training
    config_.getOutputFeatures()
};
output_ = std::make_unique<TensorType>( device, out_shape,
    this->getName() + ".output" );

The default implementation forwards to the legacy onBuilding( const shape_t& ) overload for backwards compatibility. New components should override this overload directly.

Note: Do not call build() or onBuilding() from within this hook.; Implementations should either succeed fully or leave no partial state, as a failed build() may be retried.

Takes the build-time configuration; use its allocationSeqLen() to obtain the correct output buffer sequence dimension.

Reimplemented from Mila::Dnn::Component< TDeviceType, TPrecision >.

◆ onExecutionContextSet()

template<DeviceType TDeviceType, TensorDataType TPrecision>

void Mila::Dnn::Router< TDeviceType, TPrecision >::onExecutionContextSet ( )

inlineoverrideprotectedvirtual

Hook invoked after ExecutionContext is set.

Propagates the execution context to all child components that don't already have one. This enables the pattern where composites define their architecture graph in the constructor (context-independent) and context is bound later when available.

Called by Component::setExecutionContext() after the context is registered. Automatically invoked for both standalone mode (component creates own context) and shared mode (parent provides context).

Override this in derived classes if additional context-dependent initialization is required beyond context propagation to children.

Reimplemented from Mila::Dnn::CompositeComponent< TDeviceType, TPrecision >.

◆ onTrainingModeChanging()

template<DeviceType TDeviceType, TensorDataType TPrecision>

void Mila::Dnn::Router< TDeviceType, TPrecision >::onTrainingModeChanging ( TrainingMode training_mode )

inlineoverrideprotectedvirtual

Hook invoked when training mode is about to change.

Propagates the new mode to all child components. The hook runs with the Component's training mutex held; it MUST NOT call setTrainingMode().

Parameters

training_mode New training mode (Normal or Eval)

Reimplemented from Mila::Dnn::CompositeComponent< TDeviceType, TPrecision >.

◆ parameterCount()

template<DeviceType TDeviceType, TensorDataType TPrecision>

dim_t Mila::Dnn::Router< TDeviceType, TPrecision >::parameterCount ( ) const

inlineoverridevirtual

The projection and the two owned vectors; the norm's derived weight is not counted.

Reimplemented from Mila::Dnn::CompositeComponent< TDeviceType, TPrecision >.

◆ save_()

template<DeviceType TDeviceType, TensorDataType TPrecision>

void Mila::Dnn::Router< TDeviceType, TPrecision >::save_	(	ModelArchive &	archive,
		SerializationMode	mode ) const

inlineoverrideprotectedvirtual

Reimplemented from Mila::Dnn::CompositeComponent< TDeviceType, TPrecision >.

◆ saveFlatTensors()

template<DeviceType TDeviceType, TensorDataType TPrecision>

void Mila::Dnn::Router< TDeviceType, TPrecision >::saveFlatTensors	(	Serialization::SafeTensorsWriter &	writer,
		const std::string &	prefix,
		Serialization::TensorSavePass	pass ) const

inlineoverridevirtual

The checkpoint's vocabulary: the projection, then the two owned vectors.

Deliberately not the base recursion, which would also write the norm child's derived weight – a tensor no checkpoint carries and every load would have to ignore.

Reimplemented from Mila::Dnn::CompositeComponent< TDeviceType, TPrecision >.

◆ toString()

template<DeviceType TDeviceType, TensorDataType TPrecision>

std::string Mila::Dnn::Router< TDeviceType, TPrecision >::toString ( ) const

inlineoverridevirtual

Generate a human-readable description.

Returns: String representation showing children

Reimplemented from Mila::Dnn::CompositeComponent< TDeviceType, TPrecision >.

The documentation for this class was generated from the following file:

Mila/Src/Dnn/Components/MixtureOfExperts/Router.ixx

Classes

Public Types

Public Member Functions

Protected Member Functions

Additional Inherited Members

Detailed Description

Member Function Documentation

◆ forward()

◆ getMemoryStats()

◆ getObservableStages()

◆ getParameterNames()

◆ getRequiredMemory()

◆ getType()

◆ load_()

◆ loadParameter()

◆ onBuilding()

◆ onExecutionContextSet()

◆ onTrainingModeChanging()

◆ parameterCount()

◆ save_()

◆ saveFlatTensors()

◆ toString()