Enum CacheFormat 

pub enum CacheFormat {
    F16,
    Int8,
    Int8Hadamard,
}

How an attention block’s keys and values are stored.

Variants

F16

Each element binary16.

Int8

Each head row of a position int8 codes under one f16 scale (cache_row).

Int8Hadamard

Each head row rotated by the Walsh–Hadamard transform (hadamard) before its int8 codes, so a few large channels spread across the row and its one scale fits every element; the query rotated alike (a score is the rotated dot over the head’s length) and the weighted sum of values rotated back.

Implementations

impl CacheFormat

pub fn is_int8(self) -> bool

Whether the cache holds int8 codes, so attention can run on the array.

Trait Implementations

impl Clone for CacheFormat

fn clone(&self) -> CacheFormat

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for CacheFormat

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

Formats the value using the given formatter.

impl Hash for CacheFormat

fn hash<__H: Hasher>(&self, state: &mut __H)

Feeds this value into the given Hasher.

fn hash_slice<H>(data: &[Self], state: &mut H)

where H: Hasher, Self: Sized,

Feeds a slice of this type into the given Hasher.

impl PartialEq for CacheFormat

fn eq(&self, other: &CacheFormat) -> 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 Copy for CacheFormat

impl Eq for CacheFormat

impl StructuralPartialEq for CacheFormat