Skip to main content

QuantizationParameters

Type Alias QuantizationParameters 

Source
pub type QuantizationParameters = QParams<Tensor<1>>;
Expand description

The tensor quantization parameters.

Aliased Type§

pub struct QuantizationParameters {
    pub scales: Tensor<1>,
    pub global: Option<Tensor<1>>,
}

Fields§

§scales: Tensor<1>

The scaling factor.

§global: Option<Tensor<1>>

The per-tensor scale scales are relative to, for a two-level scheme.