Struct syntaxdot_transformers::models::albert::AlbertEncoder

pub struct AlbertEncoder { /* private fields */ }

ALBERT encoder.

This encoder uses the BERT encoder with two modifications:

1. The embeddings are projected to fit the hidden layer size.
2. Weights are shared between layers.

Implementations

impl AlbertEncoder

pub fn new<'a>(
    vs: impl Borrow<PathExt<'a>>,
    config: &AlbertConfig
) -> Result<Self, TransformerError>

Trait Implementations

impl Debug for AlbertEncoder

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

Formats the value using the given formatter.

impl Encoder for AlbertEncoder

fn encode(
    &self,
    input: &Tensor,
    attention_mask: Option<&Tensor>,
    train: bool
) -> Result<Vec<LayerOutput>, TransformerError>

Apply the encoder.

fn n_layers(&self) -> i64

Get the number of layers that is returned by the encoder.