Enum PackedExpertPayload 

pub enum PackedExpertPayload<B>
where
    B: Backend,
{
    Nf4(Nf4ExpertPayload<B>),
    Awq(AwqExpertPayload<B>),
    Nf4Window {
        payload: Nf4ExpertPayload<B>,
        element_offset: usize,
    },
}

Source-selected original packed expert format, not a quantizer or dense surrogate.

Variants

Nf4(Nf4ExpertPayload<B>)

Original RUDA high-nibble-first flat-block NF4.

Awq(AwqExpertPayload<B>)

Original AWQ per-input-group permuted eight-code I32 words.

Nf4Window

Actual owned NF4 bytes beginning at the first original scale-block boundary. The first owned logical coefficient can start within that retained block.

Fields

payload: Nf4ExpertPayload<B>

element_offset: usize

Implementations

impl<B> PackedExpertPayload<B>

where B: Backend,

pub fn expert_range(&self) -> (usize, usize)

Actual resident expert count and explicitly declared first global expert ID.

Trait Implementations

impl<B> Clone for PackedExpertPayload<B>

where B: Clone + Backend,

fn clone(&self) -> PackedExpertPayload<B>

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<B> Debug for PackedExpertPayload<B>

where B: Debug + Backend,

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

Formats the value using the given formatter.