Struct rust_bert::longformer::LongformerForSequenceClassification

pub struct LongformerForSequenceClassification { /* fields omitted */ }

Longformer for sequence classification

Base Longformer model with a classifier head to perform sentence or document-level classification It is made of the following blocks:

• longformer: Base Longformer
• classifier: Longformer classification head

Implementations

impl LongformerForSequenceClassification

pub fn new<'p, P>(
    p: P,
    config: &LongformerConfig
) -> LongformerForSequenceClassification where
    P: Borrow<Path<'p>>, 

Build a new LongformerForSequenceClassification

Arguments

• p - Variable store path for the root of the Longformer model
• config - LongformerConfig object defining the model architecture

Example

use rust_bert::longformer::{LongformerConfig, LongformerForSequenceClassification};
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 = LongformerConfig::from_file(config_path);
let longformer_model = LongformerForSequenceClassification::new(&p.root(), &config);

pub fn forward_t(
    &self,
    input_ids: Option<&Tensor>,
    attention_mask: Option<&Tensor>,
    global_attention_mask: Option<&Tensor>,
    token_type_ids: Option<&Tensor>,
    position_ids: Option<&Tensor>,
    input_embeds: Option<&Tensor>,
    train: bool
) -> Result<LongformerSequenceClassificationOutput, RustBertError>

Forward pass through the model

Arguments

• input_ids - Optional input tensor of shape (batch size, sequence_length). This or input_embeds must be provided.
• attention_mask - Optional attention mask of shape (batch size, sequence_length). Positions with a mask with value 0 will be masked.
• global_attention_mask - Optional attention mask of shape (batch size, sequence_length). Positions with a mask with value 1 will attend all other positions in the sequence.
• token_type_ids - Optional segment id of shape (batch size, sequence_length). Convention is value of 0 for the first sentence (incl. SEP) and 1 for the second sentence. If None set to 0.
• position_ids - Optional position ids of shape (batch size, sequence_length). If None, will be incremented from 0.
• input_embeds - Optional pre-computed input embeddings of shape (batch size, sequence_length, hidden_size). If None, input ids must be provided (see input_ids)
• train - boolean flag to turn on/off the dropout layers in the model. Should be set to false for inference.

Returns

• LongformerSequenceClassificationOutput containing:
  • logits - Tensor of shape (batch size, sequence_length, num_classes)
  • all_hidden_states - Option<Vec<Tensor>> of length num_hidden_layers with shape (batch size, sequence_length, hidden_size)
  • all_attentions - Option<Vec<Tensor>> of length num_hidden_layers with shape (batch size, num_heads, sequence_length, * attention_window_size*, x + attention_window_size + 1) where x is the number of tokens with global attention
  • all_global_attentions - Option<Vec<Tensor>> of length num_hidden_layers with shape (batch size, num_heads, sequence_length, attention_window_size, x) where x is the number of tokens with global attention

Example

use rust_bert::longformer::{LongformerConfig, LongformerForSequenceClassification};
let longformer_model = LongformerForSequenceClassification::new(&vs.root(), &config);
let (batch_size, sequence_length, target_sequence_length) = (64, 128, 32);
let input_tensor = Tensor::rand(&[batch_size, sequence_length], (Int64, device));
let attention_mask = Tensor::ones(&[batch_size, sequence_length], (Int64, device));
let global_attention_mask = Tensor::zeros(&[batch_size, sequence_length], (Int64, device));
let target_tensor = Tensor::ones(&[batch_size, sequence_length], (Int64, device));

let model_output = no_grad(|| {
    longformer_model
        .forward_t(
            Some(&input_tensor),
            Some(&attention_mask),
            Some(&global_attention_mask),
            None,
            None,
            None,
            false,
        )
        .unwrap()
});