toklen 0.1.0

A single-threaded, lightweight, and fast token counter.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169

use std::collections::HashMap;

use serde::Deserialize;
use serde_json::Value;
use strum::EnumDiscriminants;

use crate::bpe;
use crate::byte_level;
use crate::split as split_pt;

/// An entry in the `added_tokens` array of `tokenizer.json`.
#[derive(Clone, Debug, Deserialize)]
pub struct AddedTokenConfig {
    /// Token ID.
    pub id: u32,
    /// Literal text content that triggers this token.
    pub content: String,
    /// Whether the token should only match as a whole word.
    #[serde(default)]
    pub single_word: bool,
    /// Whether to strip whitespace on the left when matching.
    #[serde(default)]
    pub lstrip: bool,
    /// Whether to strip whitespace on the right when matching.
    #[serde(default)]
    pub rstrip: bool,
    /// Whether the content should be matched against normalized text.
    #[serde(default)]
    pub normalized: bool,
    /// Whether this is a "special" token.
    #[serde(default)]
    pub special: bool,
}

/// Parsed `tokenizer.json`.
#[derive(Debug, Deserialize)]
pub struct TokenizerJson {
    #[serde(default)]
    pub added_tokens: Vec<AddedTokenConfig>,
    pub normalizer: Option<NormalizerConfig>,
    pub pre_tokenizer: Option<PreTokenizerConfig>,
    pub model: ModelConfig,
    #[serde(flatten)]
    pub extra: HashMap<String, Value>,
}

/// Normalizer pipeline step from `tokenizer.json`.
#[derive(Clone, Debug, Deserialize, EnumDiscriminants)]
#[strum_discriminants(name(NormalizerKind), derive(strum::Display, Hash, Ord, PartialOrd))]
#[serde(tag = "type")]
pub enum NormalizerConfig {
    Sequence {
        #[serde(default)]
        normalizers: Vec<NormalizerConfig>,
    },
    #[serde(rename = "NFC")]
    #[strum_discriminants(strum(to_string = "NFC"))]
    Nfc,
    #[serde(rename = "NFD")]
    #[strum_discriminants(strum(to_string = "NFD"))]
    Nfd,
    #[serde(rename = "NFKC")]
    #[strum_discriminants(strum(to_string = "NFKC"))]
    Nfkc,
    #[serde(rename = "NFKD")]
    #[strum_discriminants(strum(to_string = "NFKD"))]
    Nfkd,
    Lowercase,
    StripAccents,
    Strip {
        #[serde(default)]
        strip_left: bool,
        #[serde(default)]
        strip_right: bool,
    },
    Prepend {
        #[serde(default)]
        prepend: String,
    },
    Replace {
        pattern: Value,
        #[serde(default)]
        content: String,
    },
    BertNormalizer {
        #[serde(default)]
        clean_text: bool,
        #[serde(default)]
        handle_chinese_chars: bool,
        strip_accents: Option<bool>,
        #[serde(default)]
        lowercase: bool,
    },
    ByteLevel,
    #[serde(untagged)]
    Other(Value),
}

/// Pre-tokenizer pipeline step from `tokenizer.json`.
#[derive(Clone, Debug, Deserialize, EnumDiscriminants)]
#[strum_discriminants(name(PreTokenizerKind), derive(strum::Display, Hash, Ord, PartialOrd))]
#[serde(tag = "type")]
pub enum PreTokenizerConfig {
    Sequence {
        #[serde(default)]
        pretokenizers: Vec<PreTokenizerConfig>,
    },
    ByteLevel(byte_level::ByteLevel),
    Whitespace,
    WhitespaceSplit,
    Split(split_pt::Split),
    Punctuation {
        #[serde(default)]
        behavior: String,
    },
    Metaspace {
        #[serde(default = "default_meta_replacement")]
        replacement: char,
        prepend_scheme: Option<String>,
        add_prefix_space: Option<bool>,
    },
    Digits {
        #[serde(default)]
        individual_digits: bool,
    },
    BertPreTokenizer,
    UnicodeScripts,
    #[serde(untagged)]
    Other(Value),
}

#[inline(always)]
const fn default_meta_replacement() -> char {
    '\u{2581}' // ▁
}

/// Tokenization model from `tokenizer.json`.
#[derive(Clone, Debug, Deserialize, EnumDiscriminants)]
#[strum_discriminants(name(ModelKind), derive(strum::Display, Hash, Ord, PartialOrd))]
#[serde(tag = "type")]
pub enum ModelConfig {
    #[serde(rename = "BPE")]
    #[strum_discriminants(strum(to_string = "BPE"))]
    Bpe(Box<bpe::Bpe>),
    WordPiece {
        #[serde(default)]
        vocab: HashMap<String, u32>,
        #[serde(default)]
        unk_token: String,
        #[serde(default)]
        continuing_subword_prefix: String,
        max_input_chars_per_word: Option<u64>,
    },
    WordLevel {
        #[serde(default)]
        vocab: HashMap<String, u32>,
        #[serde(default)]
        unk_token: String,
    },
    Unigram {
        #[serde(default)]
        vocab: Vec<(String, f64)>,
        unk_id: Option<u32>,
        #[serde(default)]
        byte_fallback: bool,
    },
    #[serde(untagged)]
    Other(Value),
}