File size: 6,718 Bytes
f24563f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
"""
Tokenizer for LLM model.
"""

import os
import json
import regex as re
from typing import Dict, List, Optional, Tuple, Union
import sentencepiece as spm
from abc import ABC, abstractmethod


class Tokenizer(ABC):
    """
    Abstract base class for tokenizers.
    """
    
    @abstractmethod
    def encode(self, text: str, add_bos: bool = True, add_eos: bool = False) -> List[int]:
        """
        Encode text to token IDs.
        
        Args:
            text: Input text
            add_bos: Whether to add beginning of sequence token
            add_eos: Whether to add end of sequence token
            
        Returns:
            List of token IDs
        """
        pass
    
    @abstractmethod
    def decode(self, token_ids: List[int], skip_special_tokens: bool = True) -> str:
        """
        Decode token IDs to text.
        
        Args:
            token_ids: List of token IDs
            skip_special_tokens: Whether to skip special tokens
            
        Returns:
            Decoded text
        """
        pass
    
    @property
    @abstractmethod
    def vocab_size(self) -> int:
        """
        Get vocabulary size.
        
        Returns:
            Vocabulary size
        """
        pass
    
    @property
    @abstractmethod
    def bos_token_id(self) -> int:
        """
        Get beginning of sequence token ID.
        
        Returns:
            Beginning of sequence token ID
        """
        pass
    
    @property
    @abstractmethod
    def eos_token_id(self) -> int:
        """
        Get end of sequence token ID.
        
        Returns:
            End of sequence token ID
        """
        pass
    
    @property
    @abstractmethod
    def pad_token_id(self) -> int:
        """
        Get padding token ID.
        
        Returns:
            Padding token ID
        """
        pass


class SentencePieceTokenizer(Tokenizer):
    """
    SentencePiece tokenizer.
    
    Attributes:
        model_path: Path to SentencePiece model
        bos_id: Beginning of sequence token ID
        eos_id: End of sequence token ID
        pad_id: Padding token ID
    """
    
    def __init__(
        self,
        model_path: str,
        bos_id: int = 1,
        eos_id: int = 2,
        pad_id: int = 0
    ):
        """
        Initialize tokenizer.
        
        Args:
            model_path: Path to SentencePiece model
            bos_id: Beginning of sequence token ID
            eos_id: End of sequence token ID
            pad_id: Padding token ID
        """
        self.model_path = model_path
        self._bos_id = bos_id
        self._eos_id = eos_id
        self._pad_id = pad_id
        
        # Load SentencePiece model
        self.sp_model = spm.SentencePieceProcessor()
        self.sp_model.Load(model_path)
    
    def encode(self, text: str, add_bos: bool = True, add_eos: bool = False) -> List[int]:
        """
        Encode text to token IDs.
        
        Args:
            text: Input text
            add_bos: Whether to add beginning of sequence token
            add_eos: Whether to add end of sequence token
            
        Returns:
            List of token IDs
        """
        # Encode text
        token_ids = self.sp_model.EncodeAsIds(text)
        
        # Add special tokens
        if add_bos:
            token_ids = [self.bos_token_id] + token_ids
        
        if add_eos:
            token_ids = token_ids + [self.eos_token_id]
        
        return token_ids
    
    def decode(self, token_ids: List[int], skip_special_tokens: bool = True) -> str:
        """
        Decode token IDs to text.
        
        Args:
            token_ids: List of token IDs
            skip_special_tokens: Whether to skip special tokens
            
        Returns:
            Decoded text
        """
        # Filter special tokens if requested
        if skip_special_tokens:
            token_ids = [
                token_id for token_id in token_ids
                if token_id not in [self.bos_token_id, self.eos_token_id, self.pad_token_id]
            ]
        
        # Decode token IDs
        text = self.sp_model.DecodeIds(token_ids)
        
        return text
    
    @property
    def vocab_size(self) -> int:
        """
        Get vocabulary size.
        
        Returns:
            Vocabulary size
        """
        return self.sp_model.GetPieceSize()
    
    @property
    def bos_token_id(self) -> int:
        """
        Get beginning of sequence token ID.
        
        Returns:
            Beginning of sequence token ID
        """
        return self._bos_id
    
    @property
    def eos_token_id(self) -> int:
        """
        Get end of sequence token ID.
        
        Returns:
            End of sequence token ID
        """
        return self._eos_id
    
    @property
    def pad_token_id(self) -> int:
        """
        Get padding token ID.
        
        Returns:
            Padding token ID
        """
        return self._pad_id


def train_sentencepiece_tokenizer(
    texts: List[str],
    vocab_size: int,
    model_prefix: str,
    character_coverage: float = 0.9995,
    model_type: str = "bpe",
    user_defined_symbols: Optional[List[str]] = None
) -> str:
    """
    Train SentencePiece tokenizer.
    
    Args:
        texts: List of training texts
        vocab_size: Vocabulary size
        model_prefix: Prefix for model files
        character_coverage: Character coverage
        model_type: Model type (bpe, unigram, char, word)
        user_defined_symbols: List of user-defined symbols
        
    Returns:
        Path to trained model
    """
    # Write training data to temporary file
    with open(f"{model_prefix}.txt", "w", encoding="utf-8") as f:
        for text in texts:
            f.write(text + "\n")
    
    # Set training arguments
    args = [
        f"--input={model_prefix}.txt",
        f"--model_prefix={model_prefix}",
        f"--vocab_size={vocab_size}",
        f"--character_coverage={character_coverage}",
        f"--model_type={model_type}",
        "--pad_id=0",
        "--bos_id=1",
        "--eos_id=2",
        "--unk_id=3",
        "--pad_piece=<pad>",
        "--bos_piece=<bos>",
        "--eos_piece=<eos>",
        "--unk_piece=<unk>",
        "--normalization_rule_name=nmt_nfkc_cf",
    ]
    
    # Add user-defined symbols if provided
    if user_defined_symbols:
        args.append(f"--user_defined_symbols={','.join(user_defined_symbols)}")
    
    # Train tokenizer
    spm.SentencePieceTrainer.Train(" ".join(args))
    
    # Remove temporary file
    os.remove(f"{model_prefix}.txt")
    
    return f"{model_prefix}.model"