Skip to main content

Encoder

syntaxdot_transformers::models

Trait Encoder 

Source 
pub trait Encoder {
    // Required methods
    fn encode(
        &self,
        input: &Tensor,
        attention_mask: Option<&Tensor>,
        train: bool,
    ) -> Result<Vec<LayerOutput>, TransformerError>;
    fn n_layers(&self) -> i64;
}
Expand description

Encoder networks.

Required Methods§

Source

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

Apply the encoder.

Returns the output and attention per layer. The (optional) attention mask of shape [batch_size, time_steps] indicates which tokens should be included (true) and excluded (false) from attention. This can be used to mask inactive timesteps.

Source

fn n_layers(&self) -> i64

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

Dyn Compatibility§

This trait is dyn compatible.

In older versions of Rust, dyn compatibility was called "object safety".

Implementors§