Struct MixedPrecisionPolicy 

pub struct MixedPrecisionPolicy {
    pub layer_bits: Vec<u32>,
    pub layer_names: Vec<String>,
    pub target_avg_bits: f32,
}

Per-layer bit-width assignment produced by the greedy sensitivity policy.

Fields

layer_bits: Vec<u32>

Bit-width assigned to each layer (same order as the input sensitivity list).

layer_names: Vec<String>

Layer names (mirrors the order of layer_bits).

target_avg_bits: f32

Target average bits-per-parameter (budget constraint).

Implementations

impl MixedPrecisionPolicy

pub fn from_sensitivity( sensitivities: &[LayerSensitivity], target_avg_bits: f32, ) -> QuantResult<Self>

Compute a mixed-precision policy from layer sensitivity profiles.

Parameters

• sensitivities — per-layer LayerSensitivity from SensitivityAnalyzer.
• target_avg_bits — target average bit-width (e.g., 4.0 for ~4-bit).

Errors

• QuantError::EmptyInput — sensitivities is empty.
• QuantError::InfeasibleCompressionTarget — target cannot be met even at max bits.

pub fn effective_average_bits(&self) -> f32

Effective average bits per parameter across all layers.

pub fn bits_for_layer(&self, name: &str) -> Option<u32>

Return the bit-width assigned to a layer by name.

Returns None if the name is not found.

pub fn n_layers(&self) -> usize

Number of layers in the policy.

Trait Implementations

impl Clone for MixedPrecisionPolicy

fn clone(&self) -> MixedPrecisionPolicy

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for MixedPrecisionPolicy

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

Formats the value using the given formatter.