Struct Quantization 

pub struct Quantization {
    pub scale: Vec<f32>,
    pub zero_point: Option<Vec<i32>>,
    pub axis: Option<usize>,
    pub dtype: Option<DType>,
}

Quantization parameters for a quantized tensor.

Supports per-tensor (scalar scale) and per-channel (array scale) quantization. Symmetric quantization is indicated by an absent or all-zero zero_point.

Fields

scale: Vec<f32>

Scale factor(s). One element for per-tensor quantization, or one element per slice for per-channel quantization.

zero_point: Option<Vec<i32>>

Zero point offset(s). Omit or set to all-zero for symmetric quantization. For per-channel quantization, length must match scale length.

axis: Option<usize>

Tensor dimension index that scale/zero_point arrays correspond to. Required when per-channel; ignored otherwise.

dtype: Option<DType>

Quantized data type. Required on v2 metadata files; may be absent on programmatically-constructed configurations where the dtype is resolved at decode time from the actual tensor.

Implementations

impl Quantization

pub fn is_per_tensor(&self) -> bool

Returns true when per-tensor (scalar scale).

pub fn is_per_channel(&self) -> bool

Returns true when per-channel (array scale of length > 1).

pub fn is_symmetric(&self) -> bool

Returns true when all zero points are 0 (or absent).

pub fn zero_point_at(&self, channel: usize) -> i32

Returns the zero point for the given channel index, or 0 when the quantization is symmetric.

pub fn scale_at(&self, channel: usize) -> f32

Returns the scale for the given channel index.

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 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 TryFrom<&Quantization> for Quantization

Convert a schema-level Quantization (which also carries the quantized dtype) into the tensor-crate edgefirst_tensor::Quantization attached to a Tensor<T> at runtime. The dtype field is dropped — the tensor’s T supplies the storage element type directly.

Returns Err on any length-mismatch / axis-out-of-range condition; the schema’s Quantization::validate_shape has looser rules (accepts absent axis on per-channel) that the tensor crate does not. Callers should prefer the v2 parse path which normalizes.

type Error = Error

The type returned in the event of a conversion error.

fn try_from(q: &Quantization) -> Result<Self, Self::Error>

Performs the conversion.

impl StructuralPartialEq for Quantization