File size: 14,259 Bytes
1d0ef1d |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 |
"""
Enhanced Tokenizer for Hugging Face Integration
Improved accuracy with comprehensive vocabulary and encoding
"""
import json
import re
from typing import List, Dict, Optional, Union
from transformers import PreTrainedTokenizer
import os
class IlluminatorTokenizer(PreTrainedTokenizer):
"""
Enhanced tokenizer for the Illuminator model with improved accuracy
Compatible with Hugging Face transformers
"""
vocab_files_names = {"vocab_file": "vocab.json", "merges_file": "merges.txt"}
def __init__(
self,
vocab_file=None,
merges_file=None,
errors="replace",
unk_token="<|unk|>",
bos_token="<|bos|>",
eos_token="<|eos|>",
pad_token="<|pad|>",
add_prefix_space=False,
**kwargs
):
super().__init__(
errors=errors,
unk_token=unk_token,
bos_token=bos_token,
eos_token=eos_token,
pad_token=pad_token,
add_prefix_space=add_prefix_space,
**kwargs
)
self.add_prefix_space = add_prefix_space
# Initialize enhanced vocabulary
if vocab_file and os.path.isfile(vocab_file):
with open(vocab_file, 'r', encoding='utf-8') as f:
self.encoder = json.load(f)
else:
self.encoder = self._build_enhanced_vocabulary()
self.decoder = {v: k for k, v in self.encoder.items()}
# Enhanced BPE merges for better subword handling
self.bpe_merges = []
if merges_file and os.path.isfile(merges_file):
with open(merges_file, 'r', encoding='utf-8') as f:
self.bpe_merges = [tuple(line.strip().split()) for line in f.readlines()[1:]]
else:
self.bpe_merges = self._build_enhanced_bpe_merges()
self.bpe_merges_dict = dict(self.bpe_merges)
self.cache = {}
def _build_enhanced_vocabulary(self) -> Dict[str, int]:
"""Build comprehensive vocabulary for maximum accuracy"""
vocab = {}
idx = 0
# Special tokens first
special_tokens = [
"<|pad|>", "<|unk|>", "<|bos|>", "<|eos|>",
"<|mask|>", "<|sep|>", "<|cls|>", "<|endoftext|>"
]
for token in special_tokens:
vocab[token] = idx
idx += 1
# Bytes for all possible byte values (0-255)
for i in range(256):
vocab[chr(i)] = idx
idx += 1
# Enhanced vocabulary for better accuracy
enhanced_words = self._get_enhanced_vocabulary_words()
for word in enhanced_words:
if word not in vocab:
vocab[word] = idx
idx += 1
# Common subwords and morphemes
subwords = self._get_subword_vocabulary()
for subword in subwords:
if subword not in vocab:
vocab[subword] = idx
idx += 1
# Technical terms for better domain coverage
technical_terms = self._get_technical_vocabulary()
for term in technical_terms:
if term not in vocab:
vocab[term] = idx
idx += 1
return vocab
def _get_enhanced_vocabulary_words(self) -> List[str]:
"""Get enhanced vocabulary for better accuracy"""
return [
# High-frequency words
"the", "be", "to", "of", "and", "a", "in", "that", "have", "i", "it", "for", "not", "on", "with", "he", "as", "you", "do", "at",
"this", "but", "his", "by", "from", "they", "we", "say", "her", "she", "or", "an", "will", "my", "one", "all", "would", "there", "their",
# AI/ML terms for domain accuracy
"artificial", "intelligence", "machine", "learning", "deep", "neural", "network", "algorithm", "model", "training", "data", "dataset",
"feature", "prediction", "classification", "regression", "supervised", "unsupervised", "reinforcement", "attention", "transformer",
"embedding", "gradient", "optimization", "backpropagation", "epoch", "batch", "loss", "accuracy", "validation", "testing",
# Programming terms
"python", "javascript", "java", "cpp", "function", "method", "class", "object", "variable", "parameter", "return", "loop",
"condition", "array", "list", "dictionary", "string", "integer", "boolean", "algorithm", "structure", "framework", "library",
# Science terms
"physics", "chemistry", "biology", "mathematics", "quantum", "relativity", "evolution", "genetics", "climate", "environment",
"energy", "force", "matter", "atom", "molecule", "cell", "organism", "ecosystem", "theory", "experiment", "research",
# Technology terms
"computer", "software", "hardware", "internet", "network", "database", "security", "encryption", "server", "client",
"protocol", "application", "system", "platform", "technology", "digital", "electronic", "innovation", "development",
# Common prefixes and suffixes
"un", "re", "in", "dis", "en", "non", "over", "mis", "sub", "pre", "inter", "fore", "de", "trans", "super", "semi", "anti",
"ing", "ed", "er", "est", "ly", "tion", "sion", "ness", "ment", "ful", "less", "able", "ible", "ous", "ious", "ive",
]
def _get_subword_vocabulary(self) -> List[str]:
"""Get subword vocabulary for better tokenization"""
return [
# Common letter combinations
"th", "he", "in", "er", "an", "re", "ed", "nd", "on", "en", "at", "ou", "it", "is", "or", "ti", "as", "te", "et", "ng",
"of", "al", "de", "se", "le", "sa", "si", "ar", "ve", "ra", "ld", "ur", "ly", "ta", "ri", "ne", "me", "nt", "ty", "ic",
# Programming patterns
"def", "class", "import", "from", "return", "if", "else", "elif", "for", "while", "try", "except", "with", "lambda",
"self", "init", "len", "str", "int", "float", "bool", "list", "dict", "set", "tuple", "range", "print", "input",
# Technical patterns
"http", "https", "www", "com", "org", "net", "api", "json", "xml", "html", "css", "sql", "url", "uri", "uuid",
"config", "setup", "install", "version", "update", "upgrade", "debug", "error", "warning", "info", "log",
]
def _get_technical_vocabulary(self) -> List[str]:
"""Get technical vocabulary for domain expertise"""
return [
# AI/ML frameworks and tools
"pytorch", "tensorflow", "keras", "scikit", "pandas", "numpy", "matplotlib", "jupyter", "colab", "huggingface",
"openai", "anthropic", "deepmind", "nvidia", "cuda", "gpu", "cpu", "ram", "memory", "storage",
# Cloud and infrastructure
"aws", "azure", "gcp", "docker", "kubernetes", "linux", "ubuntu", "centos", "debian", "windows",
"server", "cluster", "container", "virtual", "machine", "instance", "deployment", "scaling",
# Programming languages and frameworks
"react", "angular", "vue", "nodejs", "express", "django", "flask", "fastapi", "spring", "laravel",
"mongodb", "postgresql", "mysql", "redis", "elasticsearch", "kafka", "rabbitmq", "nginx",
# Version control and development
"git", "github", "gitlab", "bitbucket", "branch", "commit", "merge", "pull", "push", "clone",
"repository", "fork", "issue", "release", "tag", "workflow", "pipeline", "cicd", "devops",
]
def _build_enhanced_bpe_merges(self) -> List[tuple]:
"""Build enhanced BPE merges for better subword tokenization"""
return [
# Common English patterns
("t", "h"), ("h", "e"), ("i", "n"), ("e", "r"), ("a", "n"), ("r", "e"), ("e", "d"), ("n", "d"),
("o", "n"), ("e", "n"), ("a", "t"), ("o", "u"), ("i", "t"), ("i", "s"), ("o", "r"), ("t", "i"),
("a", "s"), ("t", "e"), ("e", "t"), ("n", "g"), ("o", "f"), ("a", "l"), ("d", "e"), ("s", "e"),
# Programming patterns
("d", "ef"), ("cl", "ass"), ("im", "port"), ("fr", "om"), ("ret", "urn"), ("sel", "f"),
("in", "it"), ("le", "n"), ("st", "r"), ("in", "t"), ("pr", "int"), ("ran", "ge"),
# Technical patterns
("ht", "tp"), ("ww", "w"), ("co", "m"), ("or", "g"), ("ne", "t"), ("ap", "i"),
("js", "on"), ("ht", "ml"), ("cs", "s"), ("sq", "l"), ("ur", "l"), ("uu", "id"),
# AI/ML patterns
("ne", "ural"), ("net", "work"), ("mod", "el"), ("tra", "in"), ("dat", "a"), ("acc", "uracy"),
("los", "s"), ("gra", "dient"), ("opt", "im"), ("bat", "ch"), ("epo", "ch"), ("val", "id"),
]
def get_vocab(self) -> Dict[str, int]:
"""Return the vocabulary dictionary"""
return self.encoder.copy()
@property
def vocab_size(self) -> int:
"""Return the size of vocabulary"""
return len(self.encoder)
def _tokenize(self, text: str) -> List[str]:
"""Tokenize text using enhanced BPE"""
if not text:
return []
# Normalize text
text = self._normalize_text(text)
# Split into words
words = re.findall(r'\S+|\s+', text)
tokens = []
for word in words:
if word.isspace():
continue
# Apply BPE to each word
word_tokens = self._bpe_encode(word)
tokens.extend(word_tokens)
return tokens
def _normalize_text(self, text: str) -> str:
"""Normalize text for better tokenization"""
# Handle Unicode normalization
import unicodedata
text = unicodedata.normalize('NFKD', text)
# Handle common programming patterns
text = re.sub(r'([a-z])([A-Z])', r'\1 \2', text) # camelCase -> camel Case
text = re.sub(r'([a-zA-Z])(\d)', r'\1 \2', text) # word123 -> word 123
text = re.sub(r'(\d)([a-zA-Z])', r'\1 \2', text) # 123word -> 123 word
return text
def _bpe_encode(self, word: str) -> List[str]:
"""Apply BPE encoding to a word"""
if word in self.cache:
return self.cache[word]
# Convert to list of characters
word_chars = list(word)
if len(word_chars) == 1:
return word_chars
# Apply BPE merges
while len(word_chars) > 1:
pairs = self._get_pairs(word_chars)
if not pairs:
break
# Find the best pair to merge
best_pair = min(pairs, key=lambda x: self.bpe_merges_dict.get(x, float('inf')))
if best_pair not in self.bpe_merges_dict:
break
# Merge the best pair
new_word_chars = []
i = 0
while i < len(word_chars):
if (i < len(word_chars) - 1 and
word_chars[i] == best_pair[0] and
word_chars[i + 1] == best_pair[1]):
new_word_chars.append(best_pair[0] + best_pair[1])
i += 2
else:
new_word_chars.append(word_chars[i])
i += 1
word_chars = new_word_chars
self.cache[word] = word_chars
return word_chars
def _get_pairs(self, word_chars: List[str]) -> set:
"""Get all adjacent pairs in the word"""
pairs = set()
for i in range(len(word_chars) - 1):
pairs.add((word_chars[i], word_chars[i + 1]))
return pairs
def _convert_token_to_id(self, token: str) -> int:
"""Convert token to ID"""
return self.encoder.get(token, self.encoder.get(self.unk_token))
def _convert_id_to_token(self, index: int) -> str:
"""Convert ID to token"""
return self.decoder.get(index, self.unk_token)
def convert_tokens_to_string(self, tokens: List[str]) -> str:
"""Convert tokens back to string"""
text = ''.join(tokens)
# Clean up the text
text = text.replace('</w>', ' ')
text = re.sub(r' +', ' ', text)
return text.strip()
def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None) -> tuple:
"""Save vocabulary files"""
if not os.path.isdir(save_directory):
os.makedirs(save_directory)
vocab_file = os.path.join(
save_directory,
(filename_prefix + "-" if filename_prefix else "") + "vocab.json"
)
merges_file = os.path.join(
save_directory,
(filename_prefix + "-" if filename_prefix else "") + "merges.txt"
)
# Save vocabulary
with open(vocab_file, 'w', encoding='utf-8') as f:
json.dump(self.encoder, f, indent=2, sort_keys=True, ensure_ascii=False)
# Save merges
with open(merges_file, 'w', encoding='utf-8') as f:
f.write('#version: 0.2\n')
for merge in self.bpe_merges:
f.write(f'{merge[0]} {merge[1]}\n')
return vocab_file, merges_file
def build_inputs_with_special_tokens(self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None) -> List[int]:
"""Build model inputs by adding special tokens"""
bos = [self.bos_token_id] if self.bos_token_id is not None else []
eos = [self.eos_token_id] if self.eos_token_id is not None else []
if token_ids_1 is None:
return bos + token_ids_0 + eos
sep = [self.sep_token_id] if hasattr(self, 'sep_token_id') and self.sep_token_id is not None else []
return bos + token_ids_0 + sep + token_ids_1 + eos
|