Struct bert_tokenizer::FullTokenizer

pub struct FullTokenizer { /* private fields */ }

A FullTokenizer that runs basic tokenization and WordPiece tokenization.

Example

A full tokenizer can be built from a vocabulary as HashMap.

use bert_tokenizer::{FullTokenizer, Tokenizer, Vocab};

let mut vocab = Vocab::new();
vocab.insert("hello".to_string(), 0);
vocab.insert("world".to_string(), 1);
vocab.insert("!".to_string(), 2);
vocab.insert(",".to_string(), 3);
vocab.insert("##,".to_string(), 4);
vocab.insert("##!".to_string(), 5);
vocab.insert("##world".to_string(), 6);
vocab.insert("##hello".to_string(), 7);

let tokenizer = FullTokenizer::new().vocab(vocab).do_lower_case(true).build();
let tokens = tokenizer.tokenize("Hello, World!");
assert_eq!(tokens, vec!["hello", ",", "world", "!"]);

Or from a vocabulary file.

use bert_tokenizer::{FullTokenizer, Tokenizer};
let tokenizer = FullTokenizer::new().vocab_from_file("tests/cased_L-12_H-768_A-12/vocab.txt").build();
let tokens = tokenizer.tokenize("Hello, World!");
assert_eq!(tokens, vec!["Hello", ",", "World", "!"]);

You can also specify whether to do lower case.

use bert_tokenizer::{FullTokenizer, Tokenizer};
let tokenizer = FullTokenizer::new().vocab_from_file("tests/uncased_L-12_H-768_A-12/vocab.txt").do_lower_case(true).build();
let tokens = tokenizer.tokenize("Hello, World!");
assert_eq!(tokens, vec!["hello", ",", "world", "!"]);

Implementations

impl FullTokenizer

pub fn new() -> FullTokenizerBuilder

pub fn convert_tokens_to_ids(&self, tokens: &Vec<String>) -> Vec<u32>

Converts a sequence of tokens to a sequence of ids.

Example

use bert_tokenizer::{FullTokenizer, Tokenizer, Vocab};

let mut vocab = Vocab::new();
vocab.insert("hello".to_string(), 0);
vocab.insert("world".to_string(), 1);
vocab.insert("!".to_string(), 2);
vocab.insert("##!".to_string(), 3);
vocab.insert("##world".to_string(), 4);
vocab.insert("##hello".to_string(), 5);

let tokenizer = FullTokenizer::new().vocab(vocab).build();
let ids = tokenizer.convert_tokens_to_ids(&vec!["hello".to_string(), "!".to_string(), "world".to_string(), "##!".to_string()]);
assert_eq!(ids, vec![0, 2, 1, 3]);

Arguments

• tokens - A sequence of tokens.

Returns

A sequence of ids.

pub fn convert_ids_to_tokens(&self, ids: &Ids) -> Vec<String>

Converts a sequence of ids to a sequence of tokens.

Example

use bert_tokenizer::{FullTokenizer, Tokenizer, Vocab};

let mut vocab = Vocab::new();
vocab.insert("hello".to_string(), 0);
vocab.insert("world".to_string(), 1);
vocab.insert("!".to_string(), 2);
vocab.insert("##!".to_string(), 3);
vocab.insert("##world".to_string(), 4);
vocab.insert("##hello".to_string(), 5);

let tokenizer = FullTokenizer::new().vocab(vocab).build();
let tokens = tokenizer.convert_ids_to_tokens(&vec![0, 2, 4, 3]);
assert_eq!(tokens, vec!["hello", "!", "##world", "##!"]);

Arguments

• ids - A sequence of ids.

Returns

A sequence of tokens.

pub fn convert_tokens_to_string(&self, tokens: &[String]) -> String

Converts a sequence of subword tokens to a single text string.

Example

use bert_tokenizer::{FullTokenizer, Tokenizer, Vocab};

let mut vocab = Vocab::new();
vocab.insert("hello".to_string(), 0);
vocab.insert("world".to_string(), 1);
vocab.insert("!".to_string(), 2);
vocab.insert("##!".to_string(), 3);
vocab.insert("##world".to_string(), 4);
vocab.insert("##hello".to_string(), 5);

let tokenizer = FullTokenizer::new().vocab(vocab).build();
let text = "hello, world!";
let tokens = tokenizer.tokenize(text);
let text2 = tokenizer.convert_tokens_to_string(&tokens);
println!("Before: {} -> Tokens: {:?} -> After: {}", text, tokens, text2);

Arguments

• tokens - A sequence of tokens.

Returns

A single text string.

pub fn get_vocab_words(&self) -> Vec<String>

Get subword tokens from the vocabulary.

Example

use bert_tokenizer::{FullTokenizer, Tokenizer, Vocab};

let mut vocab = Vocab::new();
vocab.insert("hello".to_string(), 0);
vocab.insert("world".to_string(), 1);
vocab.insert("!".to_string(), 2);
vocab.insert("##!".to_string(), 3);
vocab.insert("##world".to_string(), 4);
vocab.insert("##hello".to_string(), 5);

let tokenizer = FullTokenizer::new().vocab(vocab).build();
let tokens = tokenizer.get_vocab_words();
assert_eq!(tokens, vec!["hello", "world", "!", "##!", "##world", "##hello"]);

Returns

A sequence of subword tokens.

Trait Implementations

impl Tokenizer for FullTokenizer

fn tokenize(&self, text: &str) -> Vec<String>

Tokenize by applying basic and wordpiece tokenization.

Example

use bert_tokenizer::{FullTokenizer, Tokenizer, Vocab};

let mut vocab = Vocab::new();
vocab.insert("hello".to_string(), 0);
vocab.insert("world".to_string(), 1);
vocab.insert("!".to_string(), 2);
vocab.insert("##!".to_string(), 3);
vocab.insert("##world".to_string(), 4);
vocab.insert("##hello".to_string(), 5);

let tokenizer = FullTokenizer::new().vocab(vocab).build();
let text = "hello, world!";
let tokens = tokenizer.tokenize(text);
println!("Text: {} -> Tokens: {:?}", text, tokens);