Struct GradientBuffer 

pub struct GradientBuffer {
    pub data: Vec<f64>,
    pub shape: Vec<usize>,
    pub accumulated_count: usize,
}

A single gradient buffer for one parameter.

Fields

data: Vec<f64>

Accumulated gradient values (flattened).

shape: Vec<usize>

Shape of the gradient tensor.

accumulated_count: usize

Number of micro-batches accumulated so far.

Implementations

impl GradientBuffer

pub fn new(shape: Vec<usize>) -> Self

Create a new gradient buffer initialized to zeros.

pub fn accumulate(&mut self, grad: &[f64]) -> Result<(), AccumulationError>

Add a micro-batch gradient to the buffer.

pub fn get(&self, normalize: bool) -> Vec<f64>

Get the accumulated gradient, optionally normalized by the count.

pub fn l2_norm(&self) -> f64

Compute the L2 norm of the accumulated gradient.

pub fn reset(&mut self)

Reset the buffer to zeros.

Trait Implementations

impl Clone for GradientBuffer

fn clone(&self) -> GradientBuffer

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for GradientBuffer

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

Formats the value using the given formatter.