Struct SmoothQuantConfig 

pub struct SmoothQuantConfig {
    pub alpha: f32,
    pub epsilon: f32,
}

SmoothQuant configuration: redistribute quantization difficulty from activations to weights.

Smoothing factor: s_j = max(|A_j|)^α / max(|W_j|)^(1-α) Then: à = A / s, W̃ = W * s

Fields

alpha: f32

Balance factor in [0, 1]. Typically 0.5.

epsilon: f32

Floor for scale values to avoid division by zero.

Implementations

impl SmoothQuantConfig

pub fn new(alpha: f32) -> Self

Create a new config with the given alpha (must be in [0, 1]).

pub fn default_alpha() -> Self

Default config with alpha = 0.5.

Trait Implementations

impl Clone for SmoothQuantConfig

fn clone(&self) -> SmoothQuantConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for SmoothQuantConfig

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

Formats the value using the given formatter.