Mila::Dnn::Optimizers::AdamWConfig Class Reference

Configuration for AdamW optimizer. More...

Inheritance diagram for Mila::Dnn::Optimizers::AdamWConfig:

Collaboration diagram for Mila::Dnn::Optimizers::AdamWConfig:

Public Member Functions
void	fromMetadata (const SerializationMetadata &meta) override
	Populate configuration from provided metadata.
float	getBeta1 () const noexcept
float	getBeta2 () const noexcept
float	getEpsilon () const noexcept
float	getLearningRate () const noexcept
float	getWeightDecay () const noexcept
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)	withBeta1 (this Self &&self, float beta1) noexcept
template<typename Self>
decltype(auto)	withBeta2 (this Self &&self, float beta2) noexcept
template<typename Self>
decltype(auto)	withEpsilon (this Self &&self, float epsilon) noexcept
template<typename Self>
decltype(auto)	withLearningRate (this Self &&self, float learning_rate) noexcept
template<typename Self>
decltype(auto)	withWeightDecay (this Self &&self, float weight_decay) noexcept
Public Member Functions inherited from Mila::Dnn::ComponentConfig
virtual	~ComponentConfig ()=default
	Virtual destructor for polymorphic base.

Private Attributes
float	beta1_ { 0.9f }
float	beta2_ { 0.999f }
float	epsilon_ { 1e-8f }
float	learning_rate_ { 0.001f }
float	weight_decay_ { 0.01f }

Detailed Description

Configuration for AdamW optimizer.

Encapsulates hyperparameters for AdamW and provides fluent setters, validation and conversion to/from the framework's metadata abstraction.

Member Function Documentation

fromMetadata()

void Mila::Dnn::Optimizers::AdamWConfig::fromMetadata ( const SerializationMetadata & meta )

inlineoverridevirtual

Populate configuration from provided metadata.

Missing keys are ignored; type-safe try-get helpers are used so incorrect/missing entries don't throw but simply leave defaults intact.

Implements Mila::Dnn::ComponentConfig.

Here is the call graph for this function:

getBeta1()

float Mila::Dnn::Optimizers::AdamWConfig::getBeta1 ( ) const

inlinenoexcept

getBeta2()

float Mila::Dnn::Optimizers::AdamWConfig::getBeta2 ( ) const

inlinenoexcept

getEpsilon()

float Mila::Dnn::Optimizers::AdamWConfig::getEpsilon ( ) const

inlinenoexcept

getLearningRate()

float Mila::Dnn::Optimizers::AdamWConfig::getLearningRate ( ) const

inlinenoexcept

getWeightDecay()

float Mila::Dnn::Optimizers::AdamWConfig::getWeightDecay ( ) const

inlinenoexcept

toMetadata()

SerializationMetadata Mila::Dnn::Optimizers::AdamWConfig::toMetadata ( ) const

inlineoverridevirtual

Convert configuration into SerializationMetadata.

Keys produced:

• "learning_rate" : double
• "beta1" : double
• "beta2" : double
• "epsilon" : double
• "weight_decay" : double

Implements Mila::Dnn::ComponentConfig.

Here is the call graph for this function:

toString()

std::string Mila::Dnn::Optimizers::AdamWConfig::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::Optimizers::AdamWConfig::validate ( ) const

inlineoverridevirtual

Validate configuration parameters.

Throws std::invalid_argument on invalid parameters.

Implements Mila::Dnn::ComponentConfig.

Here is the caller graph for this function:

withBeta1()

template<typename Self>

decltype(auto) Mila::Dnn::Optimizers::AdamWConfig::withBeta1 ( this Self && self, float beta1 )

inlinenoexcept

withBeta2()

template<typename Self>

decltype(auto) Mila::Dnn::Optimizers::AdamWConfig::withBeta2 ( this Self && self, float beta2 )

inlinenoexcept

withEpsilon()

template<typename Self>

decltype(auto) Mila::Dnn::Optimizers::AdamWConfig::withEpsilon ( this Self && self, float epsilon )

inlinenoexcept

withLearningRate()

template<typename Self>

decltype(auto) Mila::Dnn::Optimizers::AdamWConfig::withLearningRate ( this Self && self, float learning_rate )

inlinenoexcept

withWeightDecay()

template<typename Self>

decltype(auto) Mila::Dnn::Optimizers::AdamWConfig::withWeightDecay ( this Self && self, float weight_decay )

inlinenoexcept

Member Data Documentation

beta1_

float Mila::Dnn::Optimizers::AdamWConfig::beta1_ { 0.9f }

private

beta2_

float Mila::Dnn::Optimizers::AdamWConfig::beta2_ { 0.999f }

private

epsilon_

float Mila::Dnn::Optimizers::AdamWConfig::epsilon_ { 1e-8f }

private

learning_rate_

float Mila::Dnn::Optimizers::AdamWConfig::learning_rate_ { 0.001f }

private

weight_decay_

float Mila::Dnn::Optimizers::AdamWConfig::weight_decay_ { 0.01f }

private

---

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

• /__w/Mila/Mila/Mila/Src/Dnn/Optimizers/AdamWConfig.ixx