Struct Data2VecConfig 

pub struct Data2VecConfig {
    pub mask_ratio: f32,
    pub momentum: f32,
    pub beta: f32,
    pub normalize_targets: bool,
    pub top_k_average: usize,
}

Hyper-parameters for the data2vec training objective.

Fields

mask_ratio: f32

Fraction of tokens to mask (default 0.65, paper canonical for vision).

momentum: f32

EMA coefficient for the teacher network (default 0.999).

beta: f32

Huber loss threshold β (default 2.0).

normalize_targets: bool

Whether to L2-normalise teacher representations per feature dimension across the batch/token axis before computing the loss (default true).

top_k_average: usize

Number of teacher layer outputs to average for target computation (default 1; set > 1 to enable a rolling exponential history buffer).

Trait Implementations

impl Clone for Data2VecConfig

fn clone(&self) -> Data2VecConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Data2VecConfig

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

Formats the value using the given formatter.

impl Default for Data2VecConfig

fn default() -> Self

Returns the “default value” for a type.