Struct Bpe 

pub struct Bpe { /* private fields */ }

Byte-Pair Encoding model.

Example

use std::collections::HashMap;
use morpheme::models::Bpe;
use morpheme::Model;

let vocab: HashMap<String, u32> =
    [("a", 0), ("b", 1), ("c", 2), ("ab", 3), ("abc", 4)].map(|(t, i)| (t.to_string(), i)).into();
let merges = vec![("a".into(), "b".into()), ("ab".into(), "c".into())];
let bpe = Bpe::builder().vocab_and_merges(vocab, merges).build()?;

let tokens: Vec<String> = bpe.tokenize("abcab")?.into_iter().map(|t| t.value).collect();
assert_eq!(tokens, ["abc", "ab"]);

Implementations

impl Bpe

pub fn builder() -> BpeBuilder

Start building a model.

pub fn new(vocab: Vocab, merges: Merges) -> Result<Self>

Build from a vocabulary and merges with default options.

pub fn unk_token(&self) -> Option<&str>

The unknown token, if any.

pub fn continuing_subword_prefix(&self) -> Option<&str>

The continuing-subword prefix, if any.

pub fn end_of_word_suffix(&self) -> Option<&str>

The end-of-word suffix, if any.

pub fn dropout(&self) -> Option<f32>

BPE-dropout probability (None: deterministic).

pub fn set_dropout(&mut self, dropout: Option<f32>) -> Result<()>

Set the BPE-dropout probability at runtime, with the same validation as the builder: None or a value in [0, 1], where Some(0.0) is normalized to None (it never skips a merge). Anything else, including NaN, is a config error and leaves the model unchanged. The word cache is cleared, since cached results assume the previous merge behaviour.

pub fn fuse_unk(&self) -> bool

Whether consecutive unknown chars are fused into one unknown token.

pub fn byte_fallback(&self) -> bool

Whether unknown chars fall back to <0xNN> byte tokens.

pub fn ignore_merges(&self) -> bool

Whether words already in the vocabulary skip merging.

pub fn merges(&self) -> &[(String, String)]

Merges ordered by priority, as token strings.

pub fn clear_cache(&self)

Drop all cached words.

Trait Implementations

impl Clone for Bpe

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for Bpe

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for Bpe

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for Bpe

fn deserialize<D: Deserializer<'de>>(deserializer: D) -> Result<Self, D::Error>

Deserialize this value from the given Serde deserializer.

impl From<Bpe> for ModelWrapper

fn from(m: Bpe) -> Self

Converts to this type from the input type.

impl Model for Bpe

fn tokenize(&self, sequence: &str) -> Result<Vec<Token>>

Tokenize sequence. Token offsets are byte offsets into sequence.

fn token_to_id(&self, token: &str) -> Option<u32>

Id of token, if it is in the vocabulary.

fn id_to_token(&self, id: u32) -> Option<String>

Token for id, if it is in the vocabulary.

fn vocab(&self) -> HashMap<String, u32>

The full vocabulary.

fn vocab_size(&self) -> usize

Vocabulary size.

impl PartialEq for Bpe

fn eq(&self, other: &Self) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl Serialize for Bpe

fn serialize<S: Serializer>(&self, serializer: S) -> Result<S::Ok, S::Error>

Serialize this value into the given Serde serializer.