Struct rust_bert::bert::BertForMultipleChoice

pub struct BertForMultipleChoice { /* fields omitted */ }

BERT for multiple choices

Multiple choices model using a BERT base model and a linear classifier. Input should be in the form [CLS] Context [SEP] Possible choice [SEP]. The choice is made along the batch axis, assuming all elements of the batch are alternatives to be chosen from for a given context. It is made of the following blocks:

• bert: Base BertModel
• classifier: Linear layer for multiple choices

Implementations

impl BertForMultipleChoice

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

Build a new BertForMultipleChoice

Arguments

• p - Variable store path for the root of the BertForMultipleChoice model
• config - BertConfig object defining the model architecture

Example

use rust_bert::bert::{BertConfig, BertForMultipleChoice};
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 = BertConfig::from_file(config_path);
let bert = BertForMultipleChoice::new(&p.root() / "bert", &config);

pub fn forward_t(
    &self,
    input_ids: &Tensor,
    mask: Option<&Tensor>,
    token_type_ids: Option<&Tensor>,
    position_ids: Option<&Tensor>,
    train: bool
) -> BertSequenceClassificationOutput

Forward pass through the model

Arguments

• input_ids - Input tensor of shape (batch size, sequence_length).
• mask - Optional mask of shape (batch size, sequence_length). Masked position have value 0, non-masked value 1. If None set to 1
• 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.
• train - boolean flag to turn on/off the dropout layers in the model. Should be set to false for inference.

Returns

• BertSequenceClassificationOutput containing:
  • logits - Tensor of shape (1, batch size) containing the logits for each of the alternatives given
  • 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, sequence_length, hidden_size)

Example

let (num_choices, sequence_length) = (3, 128);
let input_tensor = Tensor::rand(&[num_choices, sequence_length], (Int64, device));
let mask = Tensor::zeros(&[num_choices, sequence_length], (Int64, device));
let token_type_ids = Tensor::zeros(&[num_choices, sequence_length], (Int64, device));
let position_ids = Tensor::arange(sequence_length, (Int64, device))
    .expand(&[num_choices, sequence_length], true);

let model_output = no_grad(|| {
    bert_model.forward_t(
        &input_tensor,
        Some(&mask),
        Some(&token_type_ids),
        Some(&position_ids),
        false,
    )
});