Struct QuantizedWeights 

pub struct QuantizedWeights {
    pub name: String,
    pub weights_i8: Vec<i32>,
    pub shape: Vec<usize>,
    pub scale: f32,
    pub bias: Option<Vec<f32>>,
}

INT8 quantised weights for a single Linear layer.

Fields

name: String

Layer name.

weights_i8: Vec<i32>

Quantised weight values in row-major order (i8 serialised as i32 for serde compatibility).

shape: Vec<usize>

Weight tensor shape [out_features, in_features].

scale: f32

Per-tensor scale factor.

bias: Option<Vec<f32>>

Bias (kept in FP32).

Implementations

impl QuantizedWeights

pub fn from_f32( name: String, w: &[f32], shape: Vec<usize>, bias: Option<Vec<f32>>, scale: Option<f32>, ) -> Self

Quantise an FP32 weight matrix.

Arguments

• name – Layer name.
• w – FP32 weights in row-major layout.
• shape – [out_features, in_features].
• bias – Optional FP32 bias vector.
• scale – If None, computed from the max absolute weight.

pub fn dequantize(&self) -> Vec<f32>

Dequantise to FP32.

pub fn size_bytes(&self) -> usize

Memory used by the quantised weights in bytes.

pub fn original_size_bytes(&self) -> usize

Memory the original FP32 weights would have used.

pub fn compression_ratio(&self) -> f32

Compression ratio vs. FP32.

Trait Implementations

impl Clone for QuantizedWeights

fn clone(&self) -> QuantizedWeights

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for QuantizedWeights

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for QuantizedWeights

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for QuantizedWeights

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.