Struct MovingAvgObserver 

pub struct MovingAvgObserver {
    pub min_val: f32,
    pub max_val: f32,
    pub momentum: f32,
    pub bits: u32,
    pub symmetric: bool,
    /* private fields */
}

Tracks an exponential moving average of per-batch min/max statistics.

Update rule:

min_val ← momentum × min_val + (1 − momentum) × batch_min
max_val ← momentum × max_val + (1 − momentum) × batch_max

Fields

min_val: f32

Running EMA minimum.

max_val: f32

Running EMA maximum.

momentum: f32

EMA momentum (fraction of old statistics to retain, typically 0.9–0.99).

bits: u32

Quantization bit-width.

symmetric: bool

Symmetric vs asymmetric quantization.

Implementations

impl MovingAvgObserver

pub fn new(bits: u32, symmetric: bool, momentum: f32) -> Self

Create a new MovingAvgObserver.

Panics

Panics if bits is 0 or > 16 or if momentum is not in (0, 1).

Trait Implementations

impl Clone for MovingAvgObserver

fn clone(&self) -> MovingAvgObserver

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for MovingAvgObserver

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

Formats the value using the given formatter.

impl Observer for MovingAvgObserver

fn observe(&mut self, data: &[f32])

Observe a batch of values.

fn compute_params(&self) -> QuantResult<(f32, i32)>

Compute (scale, zero_point) from accumulated statistics.

fn reset(&mut self)

Reset all accumulated statistics.

fn is_calibrated(&self) -> bool

Whether any data has been observed.