Enum MoeSelectionOptions 

pub enum MoeSelectionOptions {
    Softmax {
        top_k: usize,
        renormalize: bool,
    },
    SigmoidGrouped {
        top_k: usize,
        groups: usize,
        selected_groups: usize,
        group_top_two: bool,
        renormalize: bool,
        scale: f32,
    },
}

Original discrete expert selection. Correction bias affects selection only.

Variants

Softmax

Original full softmax/top-k, with lower expert IDs resolving exact ties.

Fields

top_k: usize

Actual selected experts per token.

renormalize: bool

Original inference selection-weight normalization; training weights are explicit separately.

SigmoidGrouped

Original group-limited sigmoid selection with optional FP32 correction bias.

Fields

top_k: usize

Actual selected experts per token.

groups: usize

Actual equal expert-group count.

selected_groups: usize

Actual selected group count.

group_top_two: bool

Use the sum of the two highest corrected scores rather than the group maximum.

renormalize: bool

Original inference selected-weight normalization.

scale: f32

Original finite positive inference weight multiplier.

Trait Implementations

impl Clone for MoeSelectionOptions

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for MoeSelectionOptions

impl Debug for MoeSelectionOptions

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

Formats the value using the given formatter.

impl PartialEq for MoeSelectionOptions

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

Equality operator ==.

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

Inequality operator !=.

impl StructuralPartialEq for MoeSelectionOptions