Multi-Head Attention module that accepts concatenated QKV input. More...

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

Public Types
using	ComponentBase = Component<TDeviceType, TPrecision>
using	MR = typename DeviceTypeTraits<TDeviceType>::memory_resource
using	TensorType = Tensor<TPrecision, MR>

Public Member Functions
	MultiHeadAttention (const std::string &name, const MultiHeadAttentionConfig &config, std::optional< DeviceId > device_id=std::nullopt)
	Construct MultiHeadAttention component.
TensorType &	backward (const TensorType &input, const TensorType &output_grad)
	Run backward pass and return component-owned input-gradient tensor.
TensorType &	decode (const TensorType &input, dim_t position)
	Inference-only single-token decode pass.
TensorType &	forward (const TensorType &input)
	Standard forward pass.
const MultiHeadAttentionConfig &	getConfig () const noexcept
DeviceId	getDeviceId () const override
	Get the compute device id associated with this component.
std::vector< ITensor * >	getGradients () const override
	Return non-owning pointers to parameter gradient tensors.
MemoryStats	getMemoryStats () const override
	Return the current memory allocation breakdown for this component.
int64_t	getModelDim () const noexcept
int64_t	getNumHeads () const noexcept
std::vector< ITensor * >	getParameters () const override
	Return non-owning pointers to parameter tensors.
const ComponentType	getType () const override
	Get the component type identifier.
dim_t	parameterCount () const override
	Return number of trainable parameters.
void	save_ (ModelArchive &, SerializationMode) const override
bool	supportsKVCache () const noexcept
	Returns true when the underlying operation implements both IPositionalUnaryOp and IKVCacheLifecycle.
void	synchronize () override
std::string	toString () const override
	Produce a short, human-readable description of the component.
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).
const std::string	getName () const
	Get the component's name identifier.
virtual std::vector< std::string >	getParameterNames () const
	List all available parameter names for this component.
virtual MemoryStats	getRequiredMemory (const BuildContext &context) const
	Report what build( context ) would allocate, without allocating it.
TrainingMode	getTrainingMode () const noexcept
	The current runtime behavioral mode of this Component.
virtual bool	isBuilt () const final
	Returns true if build() has completed successfully.
virtual void	load_ (ModelArchive &archive, SerializationMode mode)
	Restore this component's parameters from its archive scope.
virtual void	loadParameter (const std::string &, const Serialization::ITensorBlob &)
	Load a parameter from serialized tensor data.
virtual void	requireSerializableParameters () const
	Verify this component can serialize whatever parameters it owns.
virtual void	saveFlatTensors (Serialization::SafeTensorsWriter &writer, const std::string &prefix, Serialization::TensorSavePass pass) const
	Drive this component's tensors through one pass of a flat safetensors save.
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	onBuilding (const BuildContext &build_config) override
	Hook invoked by build() to allocate component buffers.
void	onExecutionContextSet () override
	Lifecycle hook: Called immediately after ExecutionContext is set.
void	onTrainingModeChanging (TrainingMode training_mode) override
	Hook called before TrainingMode transitions.
Protected Member Functions inherited from Mila::Dnn::Component< TDeviceType, TPrecision >
IExecutionContext *	getExecutionContext () const
	Get the shared execution context.
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.
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
	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::MultiHeadAttention< TDeviceType, TPrecision >

Multi-Head Attention module that accepts concatenated QKV input.

The module requires a single input tensor in model-layout containing concatenated Q, K and V along the feature axis: input shape == [B, T, 3 * embedding_dim]

The backend compute implementation (registered as "MultiHeadAttentionOp") must accept the concatenated QKV input and produce an output of shape: output shape == [B, T, embedding_dim]

KV-cache inference is an optional backend capability. After build(), supportsKVCache() indicates whether the underlying operation implements both IPositionalUnaryOp (prefill/decode dispatch) and IKVCacheLifecycle (cache init/reset). Both pointers are resolved once at build time.

The KV cache lifecycle (initializeKVCache / resetKVCache) is intended to be driven exclusively by the owning transformer's generate() method. forward() is the sole entry point for prefill; decode() handles autoregressive single-token generation.

REVIEW: initializeKVCache() and resetKVCache() are currently public. When TransformerBase<> is introduced as the common base for GptTransformer, LlamaTransformer, MistralTransformer etc., revisit whether these should become private with 'friend class TransformerBase<TDeviceType, TPrecision>' to enforce that only the generate() orchestration path may manage the KV cache lifecycle.

Constructor & Destructor Documentation

◆ MultiHeadAttention()

template<DeviceType TDeviceType, TensorDataType TPrecision>

Mila::Dnn::MultiHeadAttention< TDeviceType, TPrecision >::MultiHeadAttention	(	const std::string &	name,
		const MultiHeadAttentionConfig &	config,
		std::optional< DeviceId >	device_id = std::nullopt )

inlineexplicit

Construct MultiHeadAttention component.

Parameters

name	Component name identifier (mandatory).
config	MultiHeadAttention configuration.
device_id	Optional DeviceId to create owned ExecutionContext (standalone mode).

Member Function Documentation

◆ backward()

template<DeviceType TDeviceType, TensorDataType TPrecision>

TensorType & Mila::Dnn::MultiHeadAttention< TDeviceType, TPrecision >::backward	(	const TensorType &	input,
		const TensorType &	output_grad )

inline

Run backward pass and return component-owned input-gradient tensor.

Parameters

input	Concatenated QKV input tensor used in forward.
output_grad	Gradient w.r.t. the module output.

Returns: Reference to component-owned TensorType containing the input gradient.

◆ decode()

template<DeviceType TDeviceType, TensorDataType TPrecision>

TensorType & Mila::Dnn::MultiHeadAttention< TDeviceType, TPrecision >::decode	(	const TensorType &	input,
		dim_t	position )

inline

Inference-only single-token decode pass.

When the backend implements IPositionalUnaryOp and the cache has been populated by a prior forward() call, uses the fast O(n) KV cache path. When the backend does not support positional dispatch (CpuMultiHeadAttentionOp), falls back to forward(). The caller never needs to know which path was taken.

Precondition: forward() must have been called at least once to populate the KV cache before decode() is called.

Parameters

input	Single-token QKV input [B, 1, 3 * embedding_dim].
position	Current sequence position (0-based).

Returns: Reference to component-owned single-token output tensor.

◆ forward()

template<DeviceType TDeviceType, TensorDataType TPrecision>

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

inline

Standard forward pass.

Always available regardless of backend. When the backend supports KV caching, the first forward() call initializes and populates the cache (prefill with position_offset=0). When called again after decode() steps, it automatically resets the cache and begins a new prefill session – no explicit session management required by callers.

Parameters

input Concatenated QKV input [B, T, 3 * embedding_dim].

Returns: Reference to component-owned output tensor.

◆ getDeviceId()

template<DeviceType TDeviceType, TensorDataType TPrecision>

DeviceId Mila::Dnn::MultiHeadAttention< TDeviceType, TPrecision >::getDeviceId ( ) const

inlineoverridevirtual

Get the compute device id associated with this component.

Must return the device on which parameters and operations execute.

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

◆ getGradients()

template<DeviceType TDeviceType, TensorDataType TPrecision>

std::vector< ITensor * > Mila::Dnn::MultiHeadAttention< TDeviceType, TPrecision >::getGradients ( ) const

inlineoverridevirtual

Return non-owning pointers to parameter gradient tensors.

Gradient buffers are allocated only when the component is built in training mode, so a component built for inference returns an empty vector. Stateless components return empty in either mode. This is the accessor counterpart to getParameters() and does not throw on mode.

Returns: Vector of gradient pointers; empty when built for inference.

Exceptions

std::runtime_error if called before the component has been built.

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

◆ getMemoryStats()

template<DeviceType TDeviceType, TensorDataType TPrecision>

MemoryStats Mila::Dnn::MultiHeadAttention< 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 >.

◆ getParameters()

template<DeviceType TDeviceType, TensorDataType TPrecision>

std::vector< ITensor * > Mila::Dnn::MultiHeadAttention< TDeviceType, TPrecision >::getParameters ( ) const

inlineoverridevirtual

Return non-owning pointers to parameter tensors.

The returned tensor pointers remain valid for the lifetime of the component. Order should be canonical (weights before biases).

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

◆ getType()

template<DeviceType TDeviceType, TensorDataType TPrecision>

const ComponentType Mila::Dnn::MultiHeadAttention< 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 >.

◆ onBuilding()

template<DeviceType TDeviceType, TensorDataType TPrecision>

void Mila::Dnn::MultiHeadAttention< TDeviceType, TPrecision >::onBuilding ( const BuildContext & build_config )

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::MultiHeadAttention< TDeviceType, TPrecision >::onExecutionContextSet ( )

inlineoverrideprotectedvirtual

Lifecycle hook: Called immediately after ExecutionContext is set.

Override this to perform initialization that requires a valid ExecutionContext. At the time this is called, getExecutionContext() is guaranteed to return a valid context.

Common uses:

Composite components: Create and configure child components.
Device resource allocation: Query device capabilities.

Default implementation does nothing.

Exceptions

Any	exception thrown will cause setExecutionContext() to fail and restore the component to a "context not set" state.

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

◆ onTrainingModeChanging()

template<DeviceType TDeviceType, TensorDataType TPrecision>

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

inlineoverrideprotectedvirtual

Hook called before TrainingMode transitions.

Called by setTrainingMode() after validation and lock acquisition, before the internal state is updated. Derived classes override to respond to the transition – e.g. zeroing gradient buffers on transition to Eval, or re-enabling dropout on transition to Training.

The default implementation is a no-op.

Takes the incoming TrainingMode.

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

◆ parameterCount()

template<DeviceType TDeviceType, TensorDataType TPrecision>

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

inlineoverridevirtual

Return number of trainable parameters.

For leaf components this is the element count of owned parameter tensors. CompositeComponent and Network implementations should return the recursive aggregate across all children.

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

◆ save_()

template<DeviceType TDeviceType, TensorDataType TPrecision>

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

inlineoverridevirtual

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

◆ supportsKVCache()

template<DeviceType TDeviceType, TensorDataType TPrecision>

bool Mila::Dnn::MultiHeadAttention< TDeviceType, TPrecision >::supportsKVCache ( ) const

inlinenoexcept

Returns true when the underlying operation implements both IPositionalUnaryOp and IKVCacheLifecycle.

Resolved once at build time. CPU backends return false; CUDA backends return true when CudaMultiHeadAttentionOp is in use. Safe to query before calling generate() to determine which forward path is available.

◆ synchronize()

template<DeviceType TDeviceType, TensorDataType TPrecision>

void Mila::Dnn::MultiHeadAttention< TDeviceType, TPrecision >::synchronize ( )

inlineoverridevirtual

         @brief Convenience accessor -- true if currently in Eval mode.

         Equivalent to getTrainingMode() == TrainingMode::Eval.
         Valid for both RuntimeMode::Inference and RuntimeMode::Training
         built components.

         @return true if in Eval mode.
        &zwj;/

bool isEvalMode() const noexcept { return getTrainingMode() == TrainingMode::Eval; }

    RuntimeMode getRuntimeMode() const noexcept
    {
        return build_context_.getRuntimeMode();
    }

    bool isInferenceMode() const noexcept
    {
        return build_context_.isInferenceMode();
    }

    bool isTrainingMode() const noexcept
    {
        return build_context_.isTrainingMode();
    }

====================================================================

Synchronization

    /**
       @brief Wait for outstanding device work submitted by this component.

       On CPU this may be a no-op. Use to ensure results are visible to
       the host or to measure synchronous timings.

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

◆ toString()

template<DeviceType TDeviceType, TensorDataType TPrecision>

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

inlineoverridevirtual

Produce a short, human-readable description of the component.

Implementations should keep output concise and avoid throwing.

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

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

Mila/Src/Dnn/Components/Attention/MHA/MultiHeadAttention.ixx

Public Types

Public Member Functions

Protected Member Functions

Additional Inherited Members

Detailed Description

Constructor & Destructor Documentation

◆ MultiHeadAttention()

Member Function Documentation

◆ backward()

◆ decode()

◆ forward()

◆ getDeviceId()

◆ getGradients()

◆ getMemoryStats()

◆ getParameters()

◆ getType()

◆ onBuilding()

◆ onExecutionContextSet()

◆ onTrainingModeChanging()

◆ parameterCount()

◆ save_()

◆ supportsKVCache()

◆ synchronize()

Synchronization

◆ toString()