Enum Fp8Format 

pub enum Fp8Format {
    E4M3,
    E5M2,
}

8-bit floating-point formats modelled by fake_quant_fp8.

Variants

E4M3

4 exponent bits, 3 mantissa bits, bias 7, max-normal 448 (no infinities).

E5M2

5 exponent bits, 2 mantissa bits, bias 15, max-normal 57344 (IEEE-like).

Implementations

impl Fp8Format

pub fn mantissa_bits(self) -> i32

Number of explicit mantissa bits.

pub fn exponent_bias(self) -> i32

Exponent bias.

pub fn max_normal(self) -> f64

Largest finite representable magnitude (the documented max-normal).

pub fn min_normal(self) -> f64

Smallest normal magnitude, 2^(1 - bias).

pub fn min_subnormal(self) -> f64

Smallest (subnormal) positive magnitude, 2^(1 - bias - mantissa_bits).

Trait Implementations

impl Clone for Fp8Format

fn clone(&self) -> Fp8Format

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for Fp8Format

impl Debug for Fp8Format

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

Formats the value using the given formatter.

impl Eq for Fp8Format

impl PartialEq for Fp8Format

fn eq(&self, other: &Fp8Format) -> bool

Equality operator ==.

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

Inequality operator !=.

impl StructuralPartialEq for Fp8Format