Struct TransformerModel 

pub struct TransformerModel {
    pub config: TransformerConfig,
    pub token_embedding: TokenEmbedding,
    pub encoder: TransformerEncoder,
    pub decoder: Option<TransformerDecoder>,
    /* private fields */
}

Complete transformer model for text processing

Fields

config: TransformerConfig

Model configuration

token_embedding: TokenEmbedding

Token embedding layer

encoder: TransformerEncoder

Transformer encoder

decoder: Option<TransformerDecoder>

Optional transformer decoder

Implementations

impl TransformerModel

pub fn new(config: TransformerConfig, vocabulary: Vec<String>) -> Result<Self>

Create new transformer model

pub fn encode_tokens(&self, tokens: &[String]) -> Result<Array2<f64>>

Encode text tokens to contextual embeddings

pub fn new_encoder_decoder( config: TransformerConfig, vocabulary: Vec<String>, ) -> Result<Self>

Create new encoder-decoder transformer model

pub fn encode_decode( &self, input_tokens: &[String], target_tokens: &[String], ) -> Result<Array2<f64>>

Perform encoder-decoder forward pass

pub fn generate( &self, input_tokens: &[String], max_length: usize, start_token: &str, ) -> Result<Vec<String>>

Generate text using the decoder (for generation tasks)

pub fn vocabulary(&self) -> (&HashMap<String, usize>, &HashMap<usize, String>)

Get vocabulary mapping

Trait Implementations

impl RegistrableModel for TransformerModel

Implementation of RegistrableModel for TransformerModel

fn serialize(&self) -> Result<SerializableModelData>

Serialize model to storable format

fn deserialize(data: &SerializableModelData) -> Result<Self>

Deserialize model from stored format

fn model_type(&self) -> ModelType

Get model type

fn get_config(&self) -> HashMap<String, String>

Get model configuration as string map