CUDA implementation of the softmax operation for neural networks. More...

Inheritance diagram for Mila::Dnn::Compute::CudaSoftmaxOp< TInput, TOutput >:

[legend]

Collaboration diagram for Mila::Dnn::Compute::CudaSoftmaxOp< TInput, TOutput >:

[legend]

Public Types
using	MR = typename CudaDevice::MR

using	UnaryOperationBase = UnaryOperation< DeviceType::Cuda, TInput, TOutput >

Public Types inherited from Mila::Dnn::Compute::UnaryOperation< TDeviceType, TInput, TOutput >
using	MR = std::conditional_t< TDeviceType==DeviceType::Cuda, CudaMemoryResource, HostMemoryResource >
	Memory resource type based on device type.

Public Member Functions
	CudaSoftmaxOp (const SoftmaxConfig &config)
	Constructs a new CUDA Softmax operation with the default device context.

	CudaSoftmaxOp (std::shared_ptr< DeviceContext > context, const SoftmaxConfig &config)
	Constructs a new CUDA Softmax operation with a specific device context.

void	backward (const Tensor< TInput, MR > &input, const Tensor< TInput, MR > &output, const Tensor< TInput, MR > &output_gradient, const std::vector< std::shared_ptr< Tensor< TInput, MR > > > &parameters, std::vector< std::shared_ptr< Tensor< TInput, MR > > > &parameter_gradients, Tensor< TInput, MR > &input_gradient, const OperationAttributes &properties, const std::vector< std::shared_ptr< Tensor< TInput, MR > > > &output_state) const
	Performs the backward pass of the softmax operation.

void	forward (const Tensor< TInput, MR > &input, const std::vector< std::shared_ptr< Tensor< TInput, MR > > > &parameters, const OperationAttributes &properties, Tensor< TOutput, MR > &output, std::vector< std::shared_ptr< Tensor< TOutput, MR > > > &output_state) const override
	Performs the forward pass of the softmax operation on CUDA.

std::string	getName () const override
	Gets the name of this operation.

Public Member Functions inherited from Mila::Dnn::Compute::UnaryOperation< TDeviceType, TInput, TOutput >
	UnaryOperation (OperationType operation_type)
	Constructs a UnaryOperation with the specified operation type.

	UnaryOperation (OperationType operation_type, std::shared_ptr< DeviceContext > context)
	Constructs a UnaryOperation with the specified operation type and device context.

virtual	~UnaryOperation ()=default
	Virtual destructor for proper cleanup of derived classes.

virtual void	backward (const Tensor< TInput, MR > &grad, const std::vector< std::shared_ptr< Tensor< TOutput, MR > > > &parameters, std::vector< std::shared_ptr< Tensor< TOutput, MR > > > &output_grads) const
	Executes the backward pass of a unary operation.

virtual void	backward (const Tensor< TInput, MR > &input, const Tensor< TOutput, MR > &output_grad, const std::vector< std::shared_ptr< Tensor< TOutput, MR > > > &parameters, std::vector< std::shared_ptr< Tensor< TOutput, MR > > > &parameter_grads, Tensor< TInput, MR > &input_grad, const OperationAttributes &properties, const std::vector< std::shared_ptr< Tensor< TOutput, MR > > > &output_state) const
	Executes the comprehensive backward pass of a unary operation.

virtual void	forward (const Tensor< TInput, MR > &input, const std::vector< std::shared_ptr< Tensor< TOutput, MR > > > &parameters, const OperationAttributes &properties, Tensor< TOutput, MR > &output, std::vector< std::shared_ptr< Tensor< TOutput, MR > > > &output_state) const =0
	Executes the forward pass of a unary operation.

Public Member Functions inherited from Mila::Dnn::Compute::OperationBase< TDeviceType, TInput1, TInput2, TOutput >
	OperationBase (OperationType operation_type, std::shared_ptr< DeviceContext > context)
	Constructs an OperationBase object with a specific device context and compute precision.

virtual	~OperationBase ()=default
	Virtual destructor for the OperationBase class.

std::shared_ptr< DeviceContext >	getDeviceContext () const
	Gets the device context associated with this operation.

DeviceType	getDeviceType () const
	Gets the device type for this operation.

OperationType	getOperationType () const
	Gets the operation type enumeration value.

Private Attributes
SoftmaxConfig	config_
	Configuration for the softmax operation, including axis and other parameters.

Detailed Description

template<typename TInput = float, typename TOutput = TInput>
requires ValidFloatTensorTypes<TInput, TOutput>
class Mila::Dnn::Compute::CudaSoftmaxOp< TInput, TOutput >

CUDA implementation of the softmax operation for neural networks.

This class provides a CUDA-based implementation of the softmax operation, which converts a vector of real numbers into a probability distribution. The softmax function is commonly used in classification tasks as the final activation function of a neural network, and in attention mechanisms within transformer architectures.

The implementation is optimized for NVIDIA GPUs using CUDA for high-performance computation, especially for large vocabulary sizes typical in language models.

Template Parameters

TInput	The data type of the input tensor elements.
TDataType	The data type of the output tensor elements (defaults to the input type).

Member Typedef Documentation

◆ MR

template<typename TInput = float, typename TOutput = TInput>

using Mila::Dnn::Compute::CudaSoftmaxOp< TInput, TOutput >::MR = typename CudaDevice::MR

◆ UnaryOperationBase

template<typename TInput = float, typename TOutput = TInput>

using Mila::Dnn::Compute::CudaSoftmaxOp< TInput, TOutput >::UnaryOperationBase = UnaryOperation<DeviceType::Cuda, TInput, TOutput>

Constructor & Destructor Documentation

◆ CudaSoftmaxOp() [1/2]

template<typename TInput = float, typename TOutput = TInput>

Mila::Dnn::Compute::CudaSoftmaxOp< TInput, TOutput >::CudaSoftmaxOp ( const SoftmaxConfig & config )

inline

Constructs a new CUDA Softmax operation with the default device context.

Initializes the operation with a CUDA device context (defaults to CUDA:0).

◆ CudaSoftmaxOp() [2/2]

template<typename TInput = float, typename TOutput = TInput>

Mila::Dnn::Compute::CudaSoftmaxOp< TInput, TOutput >::CudaSoftmaxOp	(	std::shared_ptr< DeviceContext >	context,
		const SoftmaxConfig &	config
	)

inline

Constructs a new CUDA Softmax operation with a specific device context.

Parameters

context The device context to use for this operation.

Exceptions

std::runtime_error If the context is not for a CUDA device.

Member Function Documentation

◆ backward()

template<typename TInput = float, typename TOutput = TInput>

void Mila::Dnn::Compute::CudaSoftmaxOp< TInput, TOutput >::backward	(	const Tensor< TInput, MR > &	input,
		const Tensor< TInput, MR > &	output,
		const Tensor< TInput, MR > &	output_gradient,
		const std::vector< std::shared_ptr< Tensor< TInput, MR > > > &	parameters,
		std::vector< std::shared_ptr< Tensor< TInput, MR > > > &	parameter_gradients,
		Tensor< TInput, MR > &	input_gradient,
		const OperationAttributes &	properties,
		const std::vector< std::shared_ptr< Tensor< TInput, MR > > > &	output_state
	)		const

inline

Performs the backward pass of the softmax operation.

Computes gradients with respect to inputs based on the output gradient. For softmax: dL/dx_i = ?_j (dL/dy_j * (y_i * (?_ij - y_j))) where ?_ij is the Kronecker delta.

Parameters

input	Input tensor from the forward pass.
output	Output tensor from the forward pass (softmax probabilities).
output_gradient	Gradient of the loss with respect to the output.
parameters	Parameters tensor from forward pass (not used).
parameter_gradients	Gradients for parameters (not used).
input_gradient	Gradient of the loss with respect to the input.
properties	Additional attributes for the operation.
output_state	Cache tensors from forward pass.

Here is the call graph for this function:

◆ forward()

template<typename TInput = float, typename TOutput = TInput>

void Mila::Dnn::Compute::CudaSoftmaxOp< TInput, TOutput >::forward	(	const Tensor< TInput, MR > &	input,
		const std::vector< std::shared_ptr< Tensor< TInput, MR > > > &	parameters,
		const OperationAttributes &	properties,
		Tensor< TOutput, MR > &	output,
		std::vector< std::shared_ptr< Tensor< TOutput, MR > > > &	output_state
	)		const

inlineoverride

Performs the forward pass of the softmax operation on CUDA.

Converts input logits into a probability distribution by taking the exponential of each element and normalizing by their sum. The computation is performed on the GPU using CUDA kernels for optimal performance.

The implementation includes numerical stability improvements by subtracting the maximum value before applying the exponential function.

Parameters

input	Input tensor containing logits of shape [B, TDataType, V], where B is batch size, TDataType is sequence length, and V is vocabulary size.
parameters	Additional parameters (not used in this operation).
properties	Additional attributes for the operation.
output	Output tensor of shape [B, TDataType, V] to store the resulting probability distribution.
output_state	Cache for intermediate results (not used in this operation).

Here is the call graph for this function:

◆ getName()

template<typename TInput = float, typename TOutput = TInput>

std::string Mila::Dnn::Compute::CudaSoftmaxOp< TInput, TOutput >::getName ( ) const

inlineoverridevirtual

Gets the name of this operation.

Returns: std::string The name of the operation ("Cuda::SoftmaxOp").

Implements Mila::Dnn::Compute::OperationBase< TDeviceType, TInput1, TInput2, TOutput >.

Member Data Documentation

◆ config_

template<typename TInput = float, typename TOutput = TInput>

SoftmaxConfig Mila::Dnn::Compute::CudaSoftmaxOp< TInput, TOutput >::config_

private

Configuration for the softmax operation, including axis and other parameters.

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

/home/runner/work/Mila/Mila/Mila/Src/Dnn/Compute/Operations/Cuda/CudaSoftmaxOp.ixx

Public Types

Public Member Functions

Private Attributes

Detailed Description

Member Typedef Documentation

◆ MR

◆ UnaryOperationBase

Constructor & Destructor Documentation

◆ CudaSoftmaxOp() [1/2]

◆ CudaSoftmaxOp() [2/2]

Member Function Documentation

◆ backward()

◆ forward()

◆ getName()

Member Data Documentation

◆ config_