Dnn.Components.Rope Module Reference

Exported Modules
module	Compute.DeviceTypeTraits
module	Dnn.TensorOps
module	Compute.OperationTraits
module	Dnn.TensorTypes
module	Compute.IPositionalPairedOp
module	Dnn.TensorDataType
module	Compute.PairedOperation
module	Compute.DeviceType
module	Dnn.ITensor
module	Serialization.Mode
module	Compute.ExecutionContext
module	Serialization.ModelArchive
module	Dnn.TensorDataTypeTraits
module	Compute.ExecutionContextFactory
module	Dnn.Components.RopeConfig
module	Dnn.Component
module	Compute.CpuMemoryResource
module	Dnn.ComponentType
module	Compute.Device
module	Dnn.Tensor
module	Compute.DeviceId

Classes
class	Mila::Dnn::Rope< TDeviceType, TPrecision >
	Device-templated RoPE component. More...

Typedefs
using	ComponentBase = Component<TDeviceType, TPrecision>
using	MR = typename DeviceTypeTraits<TDeviceType>::memory_resource
using	OpType = typename OperationTraits<OperationType::RopeOp, TDeviceType, TPrecision>::type
using	TensorType = Tensor<TPrecision, MR>

Functions
	Rope (const std::string &name, const RopeConfig &config, std::optional< DeviceId > device_id=std::nullopt)
	~Rope () override=default
std::pair< TensorType &, TensorType & >	backward (TensorType &grad_Q, TensorType &grad_K)
	Backpropagate gradients through RoPE.
void	createOperation ()
void	decode (TensorType &Q, TensorType &K, int position)
	Single-token decode with explicit position.
void	forward (TensorType &Q, TensorType &K)
	Apply rotary position embeddings to Q and K in-place.
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.
std::vector< ITensor * >	getParameters () const override
	Return non-owning pointers to parameter tensors.
const ComponentType	getType () const override
	Get the component type identifier.
void	onBuilding (const BuildContext &build_context) 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.
size_t	parameterCount () const override
	Return number of trainable parameters.
void	prefill (TensorType &Q, TensorType &K, int position_offset)
	Apply rotary position embeddings with an explicit position offset.
void	save_ (ModelArchive &archive, SerializationMode mode) const override
void	synchronize () override
	Wait for outstanding device work submitted by this component.
std::string	toString () const override
	Produce a short, human-readable description of the component.
void	validateBuildContext (const BuildContext &context) const
void	zeroGradients () override
	Clear all model-owned gradients for this component.

Variables
RopeConfig	config_
shape_t	k_shape_
std::shared_ptr< OpType >	operation_ { nullptr }
std::unique_ptr< IExecutionContext >	owned_exec_context_ { nullptr }
std::unique_ptr< TensorType >	owned_K_grad_ { nullptr }
std::unique_ptr< TensorType >	owned_Q_grad_ { nullptr }
IPositionalPairedOp *	positional_op_ { nullptr }
shape_t	q_shape_

Files
file	/__w/Mila/Mila/Mila/Src/Dnn/Components/Encodings/Rope/Rope.ixx
	Rotary positional embedding (RoPE) component.