Struct rust_bert::reformer::ReformerModelWithLMHead

pub struct ReformerModelWithLMHead { /* fields omitted */ }

Reformer Model for text generation

Reformer model with a vocabulary decoding head It is made of the following blocks:

• reformer: ReformerModel Base Reformer model
• lm_head: ReformerLMHead projecting hidden states to the vocabulary dimension

Implementations

impl ReformerModelWithLMHead

pub fn new<'p, P>(
    p: P,
    config: &ReformerConfig
) -> Result<ReformerModelWithLMHead, RustBertError> where
    P: Borrow<Path<'p>>, 

Build a new ReformerModelWithLMHead

Arguments

• p - Variable store path for the root of the BART model
• config - ReformerConfig object defining the model architecture

Example

use rust_bert::reformer::{ReformerConfig, ReformerModelWithLMHead};
use rust_bert::Config;
use std::path::Path;
use tch::{nn, Device};

let config_path = Path::new("path/to/config.json");
let device = Device::Cpu;
let p = nn::VarStore::new(device);
let config = ReformerConfig::from_file(config_path);
let reformer_model: ReformerModelWithLMHead =
    ReformerModelWithLMHead::new(&p.root(), &config).unwrap();

pub fn forward_t(
    &self,
    input_ids: Option<&Tensor>,
    position_ids: Option<&Tensor>,
    input_embeds: Option<Tensor>,
    attention_mask: Option<&Tensor>,
    num_hashes: Option<i64>,
    old_layer_states: Option<Vec<Option<LayerState>>>,
    train: bool
) -> Result<ReformerLMModelOutput, RustBertError>

Forward pass through the model

Arguments

• input_ids - Optional input tensor of shape (batch size, sequence_length). Must be provided when no pre-computed embeddings are given.
• position_ids - Optional input tensor of shape (batch size, sequence_length). If not provided will be calculated on the fly starting from position 0.
• input_embeds - Optional input tensor of shape (batch size, sequence_length, embeddings_dim). Must be provided when no input ids are given.
• attention_mask - Optional attention mask of shape (batch size, sequence_length). Positions with a mask with value 0 will be masked.
• num_hashes - Optional specification of the number of hashes to use. If not provided will use the value provided in the model configuration.
• old_layer_states - Optional cached input (Option<Vec<Option<LayerState>>>) containing previous values for the cached states and buckets.
• train - boolean flag to turn on/off the dropout layers in the model. Should be set to false for inference.

Returns

• ReformerLMModelOutput containing:
  • logits - Tensor of shape (batch size, sequence_length, vocab_size) representing the logits for each vocabulary item
  • all_hidden_states - Option<Vec<Tensor>> of length n_layers with shape (batch size, sequence_length, hidden_size)
  • all_attentions - Option<Vec<Tensor>> of length n_layers with shape (batch size, sequence_length, hidden_size)
  • cache - Option<Vec<Option<LayerState>>> of length n_layer containing values for the states and buckets for future use.

Example

use rust_bert::reformer::{ReformerConfig, ReformerModelWithLMHead};
let (batch_size, sequence_length) = (64, 128);
let input_tensor = Tensor::rand(&[batch_size, sequence_length], (Int64, device));
let input_positions = Tensor::arange(sequence_length, (Kind::Int64, device)).unsqueeze(0).expand(&[batch_size, sequence_length], true);
let attention_mask = Tensor::ones(&[batch_size, sequence_length], (Int64, device));

let model_output = no_grad(|| {
    reformer_model.forward_t(
        Some(&input_tensor),    
        Some(&input_positions),
        None,
        Some(&attention_mask),
        Some(4),
        None,
        false,
    )
});

Trait Implementations

impl LMHeadModel for ReformerModelWithLMHead

fn forward_t(
    &self,
    input_ids: &Option<Tensor>,
    cache: Cache,
    attention_mask: &Option<Tensor>,
    _token_type_ids: &Option<Tensor>,
    _position_ids: &Option<Tensor>,
    _input_embeds: &Option<Tensor>,
    _encoder_outputs: Option<&Tensor>,
    _decoder_input_ids: &Option<Tensor>,
    train: bool
) -> Result<LMModelOutput, RustBertError>

Forward pass through the model. Example provided for GPT2.

impl LanguageGenerator<ReformerModelWithLMHead, ReformerVocab, ReformerTokenizer> for ReformerGenerator

fn generate<'a, S>(
    &self,
    prompt_texts: Option<S>,
    attention_mask: Option<Tensor>,
    min_length: impl Into<Option<i64>>,
    max_length: impl Into<Option<i64>>,
    decoder_start_token_id: impl Into<Option<i64>>
) -> Vec<String> where
    S: AsRef<[&'a str]>, 

Generate text based on a vector of promp texts.

fn generate_indices<'a, S>(
    &self,
    prompt_texts: Option<S>,
    attention_mask: Option<Tensor>,
    min_length: impl Into<Option<i64>>,
    max_length: impl Into<Option<i64>>,
    decoder_start_token_id: impl Into<Option<i64>>
) -> Vec<Vec<i64>> where
    S: AsRef<[&'a str]>, 

Generate token indices without decoding (useful for token-level operations before returning final text or as validation step during training).

fn generate_from_ids_and_past(
    &self,
    input_ids: Tensor,
    attention_mask: Option<Tensor>,
    min_length: impl Into<Option<i64>>,
    max_length: impl Into<Option<i64>>,
    decoder_start_token_id: impl Into<Option<i64>>
) -> Vec<Vec<i64>>