Struct QuantizedWeightInfo 

pub struct QuantizedWeightInfo {
    pub name: String,
    pub bits: u8,
    pub scales: Vec<f32>,
    pub zero_points: Vec<i8>,
    pub original_length: usize,
    pub storage_bytes: usize,
}

Metadata about a quantized weight recovered from a QDQ-format model.

Fields

name: String

Original weight name (without _quantized suffix).

bits: u8

Quantization bit width (4 or 8).

scales: Vec<f32>

Quantization scales. len() == 1 for per-tensor quantization; len() == num_channels for per-channel.

zero_points: Vec<i8>

Quantization zero points. Same length as scales.

original_length: usize

Number of elements in the quantized tensor.

storage_bytes: usize

Actual on-disk byte count of the quantized initializer’s raw_data. For INT8 storage this equals original_length; for native INT4 (opset 21) it is ceil(original_length / 2).

Implementations

impl QuantizedWeightInfo

pub fn is_per_channel(&self) -> bool

true if the weight was quantized per-channel (more than one scale).

pub fn scale(&self) -> f32

Per-tensor convenience accessor: returns the first scale. Panics if empty.

For per-channel tensors, iterate over scales instead.

pub fn zero_point(&self) -> i8

Per-tensor convenience accessor: returns the first zero-point. Panics if empty.

For per-channel tensors, iterate over zero_points instead.

Trait Implementations

impl Clone for QuantizedWeightInfo

fn clone(&self) -> QuantizedWeightInfo

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for QuantizedWeightInfo

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

Formats the value using the given formatter.