Mila::Dnn::MLPConfig Class Reference

Configuration class for the Multi-Layer Perceptron (MLP) block. More...

Inheritance diagram for Mila::Dnn::MLPConfig:

Collaboration diagram for Mila::Dnn::MLPConfig:

Public Member Functions
	MLPConfig (dim_t input_features, dim_t hidden_size)
	Construct MLP configuration.
void	fromMetadata (const SerializationMetadata &meta) override
	Populate configuration from provided metadata.
ActivationType	getActivationType () const noexcept
	Get the configured activation function type.
dim_t	getHiddenSize () const noexcept
	Get the hidden layer size.
dim_t	getInputFeatures () const noexcept
	Get the configured number of input (and output) features.
bool	hasBias () const noexcept
	Query whether linear layers include bias terms.
SerializationMetadata	toMetadata () const override
	Convert configuration into SerializationMetadata.
std::string	toString () const override
	Produce a short, human-readable summary of the configuration.
void	validate () const override
	Validate configuration parameters.
template<typename Self>
decltype(auto)	withActivation (this Self &&self, ActivationType activation)
	Set the activation function type (C++23 fluent style).
template<typename Self>
Self &&	withBias (this Self &&self, bool has_bias)
	Configure whether the linear layers use bias.
Public Member Functions inherited from Mila::Dnn::ComponentConfig
virtual	~ComponentConfig ()=default
	Virtual destructor for polymorphic base.

Private Attributes
ActivationType	activation_type_ { ActivationType::Gelu }
bool	has_bias_ { true }
dim_t	hidden_size_ { 0 }
dim_t	input_features_ { 0 }

Detailed Description

Configuration class for the Multi-Layer Perceptron (MLP) block.

MLPConfig specifies the architectural parameters for an MLP block:

• Input and hidden feature dimensions
• Activation function type
• Optional bias and layer normalization

The MLP block structure is: Input -> Linear(in_features, hidden_size) -> [LayerNorm] -> Activation -> Linear(hidden_size, in_features) -> Output

Constructor & Destructor Documentation

MLPConfig()

Mila::Dnn::MLPConfig::MLPConfig ( dim_t input_features, dim_t hidden_size )

inline

Construct MLP configuration.

Parameters

input_features	Number of input (and output) features (must be > 0).
hidden_size	Size of the intermediate hidden layer (must be > 0).

Member Function Documentation

fromMetadata()

void Mila::Dnn::MLPConfig::fromMetadata ( const SerializationMetadata & meta )

inlineoverridevirtual

Populate configuration from provided metadata.

Missing keys are ignored. Type-safe try-get helpers are used to avoid throwing on absent fields and to preserve forward/backward compatibility.

Parameters

meta	Metadata to read configuration values from.

Implements Mila::Dnn::ComponentConfig.

Here is the call graph for this function:

getActivationType()

ActivationType Mila::Dnn::MLPConfig::getActivationType ( ) const

inlinenoexcept

Get the configured activation function type.

Returns

ActivationType selected for the MLP.

getHiddenSize()

dim_t Mila::Dnn::MLPConfig::getHiddenSize ( ) const

inlinenoexcept

Get the hidden layer size.

Returns

Hidden layer dimension.

getInputFeatures()

dim_t Mila::Dnn::MLPConfig::getInputFeatures ( ) const

inlinenoexcept

Get the configured number of input (and output) features.

Returns

Number of input features.

hasBias()

bool Mila::Dnn::MLPConfig::hasBias ( ) const

inlinenoexcept

Query whether linear layers include bias terms.

Returns

True if bias is enabled.

toMetadata()

SerializationMetadata Mila::Dnn::MLPConfig::toMetadata ( ) const

inlineoverridevirtual

Convert configuration into SerializationMetadata.

Produces keys:

• "precision" : integer (underlying value of ComputePrecision::Policy)
• "input_features" : integer
• "hidden_size" : integer
• "has_bias" : boolean
• "activation" : integer (ActivationType)

Returns

SerializationMetadata Metadata representing this configuration.

Implements Mila::Dnn::ComponentConfig.

Here is the call graph for this function:

toString()

std::string Mila::Dnn::MLPConfig::toString ( ) const

inlineoverridevirtual

Produce a short, human-readable summary of the configuration.

Implementations should return a compact, single-line description suitable for logging and debugging.

Returns

std::string Human-readable summary of the configuration.

Implements Mila::Dnn::ComponentConfig.

validate()

void Mila::Dnn::MLPConfig::validate ( ) const

inlineoverridevirtual

Validate configuration parameters.

Exceptions

std::invalid_argument

If validation fails (e.g. zero-sized dimensions).

Implements Mila::Dnn::ComponentConfig.

withActivation()

template<typename Self>

decltype(auto) Mila::Dnn::MLPConfig::withActivation ( this Self && self, ActivationType activation )

inline

Set the activation function type (C++23 fluent style).

withBias()

template<typename Self>

Self && Mila::Dnn::MLPConfig::withBias ( this Self && self, bool has_bias )

inline

Configure whether the linear layers use bias.

Parameters

has_bias

True to include bias terms in linear layers, false to omit.

Returns

Self forwarding reference for method chaining.

Member Data Documentation

activation_type_

ActivationType Mila::Dnn::MLPConfig::activation_type_ { ActivationType::Gelu }

private

has_bias_

bool Mila::Dnn::MLPConfig::has_bias_ { true }

private

hidden_size_

dim_t Mila::Dnn::MLPConfig::hidden_size_ { 0 }

private

input_features_

dim_t Mila::Dnn::MLPConfig::input_features_ { 0 }

private

---

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

• /__w/Mila/Mila/Mila/Src/Dnn/Components/FFN/MLP.Config.ixx