Trait Quantization

pub trait Quantization<E: Float + Send + Sync, Q: PrimInt + Send + Sync> {
    // Required methods
    fn range() -> (Q, Q);
    fn new(alpha: E, beta: E) -> Self;
    fn quantize(&self, values: &[E]) -> Vec<Q>;
    fn quantize_one(&self, value: E) -> Q;
    fn dequantize(&self, values: &[Q]) -> Vec<E>;
    fn dequantize_one(&self, value: Q) -> E;
}

Quantization scheme to convert elements of a higher precision data type E to a lower precision data type Q and vice-versa.

Required Methods

fn range() -> (Q, Q)

Returns the quantization range [a, b].

fn new(alpha: E, beta: E) -> Self

Create a new quantization scheme for an input range [alpha, beta].

fn quantize(&self, values: &[E]) -> Vec<Q>

Convert the values to a lower precision data type.

fn quantize_one(&self, value: E) -> Q

Convert a single value to a lower precision data type.

fn dequantize(&self, values: &[Q]) -> Vec<E>

Convert the values back to a higher precision data type.

fn dequantize_one(&self, value: Q) -> E

Convert a single value back to a higher precision data type.

Dyn Compatibility

This trait is not dyn compatible.

In older versions of Rust, dyn compatibility was called "object safety", so this trait is not object safe.

Implementors

impl<E: Float + Send + Sync, Q: PrimInt + Signed + Send + Sync> Quantization<E, Q> for SymmetricQuantization<E, Q>