Struct Quantization 

pub struct Quantization { /* private fields */ }

Quantization parameters for an integer tensor.

Covers all four modes the edgefirst.json spec defines:

Mode	scale.len()	zero_point	axis
Per-tensor symmetric	1	None	None
Per-tensor asymmetric	1	Some(len == 1)	None
Per-channel symmetric	>1	None	Some(c)
Per-channel asymmetric	>1	Some(len == scale.len())	Some(c)

The quantized storage type is carried on the parent Tensor<T>; this struct does not duplicate it. Construct via the four named constructors (the only public entry points); direct field mutation is not allowed so invalid combinations cannot be represented.

Dequantization formula:

  real_value = scale[c] × (quantized_value[c] - zero_point[c])

where c is the channel index (always 0 for per-tensor).

Implementations

impl Quantization

pub fn per_tensor_symmetric(scale: f32) -> Self

Per-tensor symmetric (zero_point = 0).

pub fn per_tensor(scale: f32, zero_point: i32) -> Self

Per-tensor asymmetric — the most common runtime shape.

pub fn per_channel_symmetric(scales: Vec<f32>, axis: usize) -> Result<Self>

Per-channel symmetric. Errors on empty scales.

pub fn per_channel( scales: Vec<f32>, zero_points: Vec<i32>, axis: usize, ) -> Result<Self>

Per-channel asymmetric. Errors on length mismatch between scales and zero_points, or empty arrays.

pub fn mode(&self) -> QuantMode<'_>

Borrow-based dispatch view. Match once at kernel entry.

pub fn is_per_tensor(&self) -> bool

Returns true for per-tensor quantization (scale.len() == 1).

pub fn is_per_channel(&self) -> bool

Returns true for per-channel quantization (scale.len() > 1).

pub fn is_symmetric(&self) -> bool

Returns true for symmetric quantization (no zero-point, or zero-point vector of all zeros).

pub fn scale(&self) -> &[f32]

Borrow the scale array. Length 1 for per-tensor; num_channels for per-channel.

pub fn zero_point(&self) -> Option<&[i32]>

Borrow the zero-point array. None for symmetric.

pub fn axis(&self) -> Option<usize>

Channel axis for per-channel quantization. None for per-tensor.

Trait Implementations

impl Clone for Quantization

fn clone(&self) -> Quantization

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Quantization

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for Quantization

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl From<(f32, i32)> for Quantization

fn from((scale, zero_point): (f32, i32)) -> Self

Convenience construction from a (scale, zero_point) tuple. Matches the legacy QuantTuple / Quantization::new calling convention so existing (0.1, -128).into() sites keep working.

impl PartialEq for Quantization

fn eq(&self, other: &Quantization) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Serialize for Quantization

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

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl StructuralPartialEq for Quantization