Struct LearnablePosEmbed 

pub struct LearnablePosEmbed {
    pub table: Vec<f32>,
    pub n_positions: usize,
    pub embed_dim: usize,
}

Learnable position embedding table: [n_positions, embed_dim].

Row i is the position embedding for the i-th token (index 0 is conventionally the CLS token position).

Fields

table: Vec<f32>

Flat [n_positions × embed_dim] parameter table.

n_positions: usize

Number of positions (including CLS if present).

embed_dim: usize

Embedding dimension.

Implementations

impl LearnablePosEmbed

pub fn new( n_positions: usize, embed_dim: usize, rng: &mut LcgRng, ) -> VisionResult<Self>

Create a learnable position embedding with small Gaussian init.

pub fn position_embedding(&self, i: usize) -> VisionResult<&[f32]>

Return the embedding for position i as a slice of length embed_dim.

Trait Implementations

impl Clone for LearnablePosEmbed

fn clone(&self) -> LearnablePosEmbed

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for LearnablePosEmbed

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

Formats the value using the given formatter.