Spaces:
Running
Running
File size: 13,887 Bytes
0a4529c |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 |
# DEPENDENCIES
import tiktoken
from typing import List
from typing import Dict
from typing import Optional
from config.settings import get_settings
from config.logging_config import get_logger
from utils.error_handler import handle_errors
from utils.error_handler import TokenManagementError
# Setup Settings and Logging
settings = get_settings()
logger = get_logger(__name__)
class TokenManager:
"""
Token management for LLM context windows: Handles token counting, context window management, and optimization for different LLM providers (Ollama, OpenAI)
"""
def __init__(self, model_name: str = None):
"""
Initialize token manager
Arguments:
----------
model_name { str } : Model name for tokenizer selection
"""
self.logger = logger
self.settings = get_settings()
self.model_name = model_name or self.settings.OLLAMA_MODEL
self.encoding = None
self.context_window = self._get_context_window()
self._initialize_tokenizer()
def _initialize_tokenizer(self):
"""
Initialize appropriate tokenizer based on model
"""
try:
# Determine tokenizer based on model
if self.model_name.startswith(('gpt-3.5', 'gpt-4')):
# OpenAI models
self.encoding = tiktoken.encoding_for_model(self.model_name)
self.logger.debug(f"Initialized tiktoken for {self.model_name}")
else:
# Default for Ollama/local models
self.encoding = tiktoken.get_encoding("cl100k_base")
self.logger.debug(f"Using cl100k_base tokenizer for local model {self.model_name}")
except Exception as e:
self.logger.warning(f"Failed to initialize specific tokenizer: {repr(e)}, using approximation")
self.encoding = None
def _get_context_window(self) -> int:
"""
Get context window size based on model
Returns:
--------
{ int } : Context window size in tokens
"""
model_contexts = {"gpt-3.5-turbo" : 4096,
"mistral:7b" : 8192,
}
# Find matching model
for model_pattern, context_size in model_contexts.items():
if model_pattern in self.model_name.lower():
return context_size
# Default context window
default_context = self.settings.CONTEXT_WINDOW
self.logger.info(f"Using default context window {default_context} for model {self.model_name}")
return default_context
def count_tokens(self, text: str) -> int:
"""
Count tokens in text
Arguments:
----------
text { str } : Input text
Returns:
--------
{ int } : Number of tokens
"""
if not text:
return 0
if self.encoding is not None:
try:
return len(self.encoding.encode(text))
except Exception as e:
self.logger.warning(f"Tokenizer failed, using approximation: {repr(e)}")
# Fallback approximation
return self._approximate_token_count(text = text)
def _approximate_token_count(self, text: str) -> int:
"""
Approximate token count when tokenizer is unavailable
Arguments:
----------
text { str } : Input text
Returns:
--------
{ int } : Approximate token count
"""
if not text:
return 0
# Use word-based approximation (more reliable than char-based)
words = text.split()
# English text averages ~1.3 tokens per word : (accounting for punctuation and subword tokenization)
estimated_tokens = int(len(words) * 1.3)
# Add 5% buffer for punctuation and special tokens
estimated_tokens = int(estimated_tokens * 1.05)
return estimated_tokens
def count_message_tokens(self, messages: List[Dict]) -> int:
"""
Count tokens in chat messages
Arguments:
----------
messages { list } : List of message dictionaries
Returns:
--------
{ int } : Total tokens in messages
"""
if not messages:
return 0
total_tokens = 0
for message in messages:
# Count content tokens
content = message.get('content', '')
total_tokens += self.count_tokens(text = content)
# Count role tokens (approximate)
role = message.get('role', '')
total_tokens += self.count_tokens(text = role)
# Add overhead for message structure: Approximate overhead per message
total_tokens += 5
return total_tokens
def fits_in_context(self, prompt: str, max_completion_tokens: int = 1000) -> bool:
"""
Check if prompt fits in context window with room for completion
Arguments:
----------
prompt { str } : Prompt text
max_completion_tokens { int } : Tokens to reserve for completion
Returns:
--------
{ bool } : True if prompt fits
"""
prompt_tokens = self.count_tokens(text = prompt)
total_required = prompt_tokens + max_completion_tokens
return (total_required <= self.context_window)
def truncate_to_fit(self, text: str, max_tokens: int, strategy: str = "end") -> str:
"""
Truncate text to fit within token limit
Arguments:
----------
text { str } : Text to truncate
max_tokens { int } : Maximum tokens allowed
strategy { str } : Truncation strategy ('end', 'start', 'middle')
Returns:
--------
{ str } : Truncated text
"""
current_tokens = self.count_tokens(text = text)
if (current_tokens <= max_tokens):
return text
if (strategy == "end"):
return self._truncate_from_end(text = text,
max_tokens = max_tokens,
)
elif (strategy == "start"):
return self._truncate_from_start(text = text,
max_tokens = max_tokens,
)
elif (strategy == "middle"):
return self._truncate_from_middle(text = text,
max_tokens = max_tokens,
)
else:
self.logger.warning(f"Unknown truncation strategy: {strategy}, using 'end'")
return self._truncate_from_end(text = text,
max_tokens = max_tokens,
)
def _truncate_from_end(self, text: str, max_tokens: int) -> str:
"""
Truncate from the end of the text
"""
if self.encoding is not None:
tokens = self.encoding.encode(text)
truncated_tokens = tokens[:max_tokens]
return self.encoding.decode(truncated_tokens)
# Approximate truncation
words = text.split()
# Conservative estimate
target_words = int(max_tokens * 0.75)
truncated_words = words[:target_words]
return " ".join(truncated_words)
def _truncate_from_start(self, text: str, max_tokens: int) -> str:
"""
Truncate from the start of the text
"""
if self.encoding is not None:
tokens = self.encoding.encode(text)
# Take from end
truncated_tokens = tokens[-max_tokens:]
return self.encoding.decode(truncated_tokens)
# Approximate truncation
words = text.split()
target_words = int(max_tokens * 0.75)
# Take from end
truncated_words = words[-target_words:]
return " ".join(truncated_words)
def _truncate_from_middle(self, text: str, max_tokens: int) -> str:
"""
Truncate from the middle of the text
"""
if self.encoding is not None:
tokens = self.encoding.encode(text)
if (len(tokens) <= max_tokens):
return text
# Keep beginning and end, remove middle
keep_start = max_tokens // 3
keep_end = max_tokens - keep_start
start_tokens = tokens[:keep_start]
end_tokens = tokens[-keep_end:]
return self.encoding.decode(start_tokens) + " [...] " + self.encoding.decode(end_tokens)
# Approximate truncation
words = text.split()
if (len(words) <= max_tokens):
return text
keep_start = max_tokens // 3
keep_end = max_tokens - keep_start
start_words = words[:keep_start]
end_words = words[-keep_end:]
return " ".join(start_words) + " [...] " + " ".join(end_words)
def calculate_max_completion_tokens(self, prompt: str, reserve_tokens: int = 100) -> int:
"""
Calculate maximum completion tokens given prompt length
Arguments:
----------
prompt { str } : Prompt text
reserve_tokens { int } : Tokens to reserve for safety
Returns:
--------
{ int } : Maximum completion tokens
"""
prompt_tokens = self.count_tokens(text = prompt)
available_tokens = self.context_window - prompt_tokens - reserve_tokens
return max(0, available_tokens)
def optimize_context_usage(self, context: str, prompt: str, max_completion_tokens: int = 1000) -> str:
"""
Optimize context to fit within context window
Arguments:
----------
context { str } : Context text
prompt { str } : Prompt template
max_completion_tokens { int } : Tokens needed for completion
Returns:
--------
{ str } : Optimized context
"""
total_prompt_tokens = self.count_tokens(text = prompt.format(context=""))
available_for_context = self.context_window - total_prompt_tokens - max_completion_tokens
if (available_for_context <= 0):
self.logger.warning("Prompt too large for context window")
return ""
context_tokens = self.count_tokens(text = context)
if (context_tokens <= available_for_context):
return context
# Truncate context to fit
optimized_context = self.truncate_to_fit(context, available_for_context, strategy="end")
reduction_pct = ((context_tokens - self.count_tokens(text = optimized_context)) / context_tokens) * 100
self.logger.info(f"Context reduced by {reduction_pct:.1f}% to fit context window")
return optimized_context
def get_token_stats(self, text: str) -> Dict:
"""
Get detailed token statistics
Arguments:
----------
text { str } : Input text
Returns:
--------
{ dict } : Token statistics
"""
tokens = self.count_tokens(text = text)
chars = len(text)
words = len(text.split())
return {"tokens" : tokens,
"characters" : chars,
"words" : words,
"chars_per_token" : chars / tokens if tokens > 0 else 0,
"tokens_per_word" : tokens / words if words > 0 else 0,
"context_window" : self.context_window,
"model" : self.model_name,
}
# Global token manager instance
_token_manager = None
def get_token_manager(model_name: str = None) -> TokenManager:
"""
Get global token manager instance
Arguments:
----------
model_name { str } : Model name for tokenizer selection
Returns:
--------
{ TokenManager } : TokenManager instance
"""
global _token_manager
if _token_manager is None or (model_name and _token_manager.model_name != model_name):
_token_manager = TokenManager(model_name)
return _token_manager
@handle_errors(error_type = TokenManagementError, log_error = True, reraise = False)
def count_tokens_safe(text: str, model_name: str = None) -> int:
"""
Safe token counting with error handling
Arguments:
----------
text { str } : Text to count tokens for
model_name { str } : Model name for tokenizer
Returns:
--------
{ int } : Token count (0 on error)
"""
try:
manager = get_token_manager(model_name = model_name)
return manager.count_tokens(text = text)
except Exception:
return 0 |