Struct GradientProcessor 

pub struct GradientProcessor<T: Float + Debug + Default + Clone + Sum + ScalarOperand + Send + Sync + 'static> { /* private fields */ }

Gradient processor for transformer optimizer

Implementations

impl<T: Float + Debug + Default + Clone + Sum + ScalarOperand + Send + Sync + 'static> GradientProcessor<T>

pub fn new(strategy: GradientProcessingStrategy) -> Self

Create new gradient processor

pub fn new_with_params( strategy: GradientProcessingStrategy, params: GradientProcessingParams<T>, ) -> Self

Create with custom parameters

pub fn process_gradients( &mut self, param_name: &str, gradients: &Array1<T>, ) -> Result<Array1<T>>

Process gradients according to the selected strategy.

param_name selects the per-parameter state slot; callers that only have a single parameter tensor may pass any stable identifier.

pub fn statistics(&self) -> &GradientStatistics<T>

Get aggregate gradient statistics

pub fn statistics_for(&self, param_name: &str) -> Option<&GradientStatistics<T>>

Get gradient statistics for a specific parameter

pub fn strategy(&self) -> GradientProcessingStrategy

Get the current processing strategy

pub fn parameters(&self) -> &GradientProcessingParams<T>

Get the current processing parameters

pub fn set_strategy(&mut self, strategy: GradientProcessingStrategy)

Update processing strategy

pub fn set_parameters(&mut self, params: GradientProcessingParams<T>)

Update processing parameters

pub fn reset(&mut self)

Reset processor state

Trait Implementations

impl<T: Clone + Float + Debug + Default + Clone + Sum + ScalarOperand + Send + Sync + 'static> Clone for GradientProcessor<T>

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<T: Debug + Float + Debug + Default + Clone + Sum + ScalarOperand + Send + Sync + 'static> Debug for GradientProcessor<T>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.