Struct LearnedPositionalEmbedding 

pub struct LearnedPositionalEmbedding {
    pub max_positions: usize,
    pub embed_dim: usize,
    pub weight: WeightTensor,
}

Learned positional embedding table (GPT-2 style).

Weight shape: [max_positions × embed_dim].

Fields

max_positions: usize

Maximum number of positions.

embed_dim: usize

Embedding dimension.

weight: WeightTensor

Weight table.

Implementations

impl LearnedPositionalEmbedding

pub fn new(max_positions: usize, embed_dim: usize) -> LmResult<Self>

Construct with zero-initialised weights.

pub fn from_weight(weight: WeightTensor) -> LmResult<Self>

Construct from an existing weight tensor.

pub fn forward(&self, seq_len: usize, offset: usize) -> LmResult<Vec<f32>>

Return positional embeddings for positions [offset, offset + seq_len).

Returns flat buffer of shape [seq_len × embed_dim].

Trait Implementations

impl Clone for LearnedPositionalEmbedding

fn clone(&self) -> LearnedPositionalEmbedding

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for LearnedPositionalEmbedding

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

Formats the value using the given formatter.