Spaces:
Running
Running
File size: 16,116 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 |
# DEPENDENCIES
import math
from typing import Any
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 config.models import TemperatureStrategy
from utils.error_handler import TemperatureControlError
# Setup Settings and Logging
settings = get_settings()
logger = get_logger(__name__)
class TemperatureController:
"""
Intelligent temperature control for LLM generation: Implements adaptive temperature strategies based on query type, complexity, and desired output characteristics
"""
def __init__(self, base_temperature: float = None, strategy: TemperatureStrategy = None):
"""
Initialize temperature controller
Arguments:
----------
base_temperature { float } : Base temperature value (default from settings)
strategy { str } : Temperature control strategy
"""
self.logger = logger
self.settings = get_settings()
self.base_temperature = base_temperature or self.settings.DEFAULT_TEMPERATURE
self.strategy = strategy or TemperatureStrategy.ADAPTIVE
# Validate base temperature
if not (0.0 <= self.base_temperature <= 1.0):
raise TemperatureControlError(f"Temperature must be between 0 and 1: {self.base_temperature}")
# Strategy configurations
self.strategy_configs = {TemperatureStrategy.FIXED : {"description" : "Fixed temperature for all queries", "range" : (0.0, 1.0)},
TemperatureStrategy.ADAPTIVE : {"description" : "Adapt temperature based on query complexity", "range" : (0.1, 0.8), "complexity_threshold" : 0.6},
TemperatureStrategy.CONFIDENCE : {"description" : "Adjust temperature based on retrieval confidence", "range" : (0.1, 0.9), "high_confidence_temp" : 0.1, "low_confidence_temp" : 0.7},
TemperatureStrategy.PROGRESSIVE : {"description" : "Progressively increase temperature for creative tasks", "range" : (0.1, 0.9), "creative_threshold" : 0.7}
}
self.logger.info(f"Initialized TemperatureController: base={self.base_temperature}, strategy={self.strategy}")
def get_temperature(self, query: str = "", context: str = "", retrieval_scores: Optional[list] = None, query_type: str = "qa") -> float:
"""
Get appropriate temperature for generation
Arguments:
----------
query { str } : User query
context { str } : Retrieved context
retrieval_scores { list } : Scores of retrieved chunks
query_type { str } : Type of query ('qa', 'creative', 'analytical', 'summary')
Returns:
--------
{ float } : Temperature value (0.0 - 1.0)
"""
if (self.strategy == TemperatureStrategy.FIXED):
return self._fixed_temperature()
elif (self.strategy == TemperatureStrategy.ADAPTIVE):
return self._adaptive_temperature(query = query,
context = context,
query_type = query_type,
)
elif (self.strategy == TemperatureStrategy.CONFIDENCE):
return self._confidence_based_temperature(retrieval_scores = retrieval_scores,
query_type = query_type,
)
elif (self.strategy == TemperatureStrategy.PROGRESSIVE):
return self._progressive_temperature(query_type = query_type,
query = query,
)
else:
self.logger.warning(f"Unknown strategy: {self.strategy}, using fixed")
return self.base_temperature
def _fixed_temperature(self) -> float:
"""
Fixed temperature strategy
"""
return self.base_temperature
def _adaptive_temperature(self, query: str, context: str, query_type: str) -> float:
"""
Adaptive temperature based on query complexity and type
"""
base_temp = self.base_temperature
# Adjust based on query type
type_adjustments = {"qa" : -0.2, # More deterministic for Q&A
"creative" : 0.3, # More creative for creative tasks
"analytical" : -0.1, # Slightly deterministic for analysis
"summary" : -0.15, # Deterministic for summarization
"comparison" : 0.1, # Slightly creative for comparisons
}
adjustment = type_adjustments.get(query_type, 0.0)
temp = base_temp + adjustment
# Adjust based on query complexity
complexity = self._calculate_query_complexity(query = query)
if (complexity > 0.7):
# High complexity
temp += 0.1
elif (complexity < 0.3):
# Low complexity
temp -= 0.1
# Adjust based on context quality
if context:
context_quality = self._calculate_context_quality(context = context)
# Poor context
if (context_quality < 0.5):
# More creative when context is poor
temp += 0.15
return self._clamp_temperature(temperature = temp)
def _confidence_based_temperature(self, retrieval_scores: Optional[list], query_type: str) -> float:
"""
Temperature based on retrieval confidence
"""
if not retrieval_scores:
self.logger.debug("No retrieval scores, using base temperature")
return self.base_temperature
# Calculate average confidence
avg_confidence = sum(retrieval_scores) / len(retrieval_scores)
config = self.strategy_configs[TemperatureStrategy.CONFIDENCE]
high_temp = config["high_confidence_temp"]
low_temp = config["low_confidence_temp"]
# High confidence -> low temperature (deterministic) & Low confidence -> high temperature (creative)
if (avg_confidence > 0.8):
temperature = high_temp
elif (avg_confidence < 0.3):
temperature = low_temp
else:
# Linear interpolation between high and low temps
normalized_confidence = (avg_confidence - 0.3) / (0.8 - 0.3)
temperature = high_temp + (low_temp - high_temp) * (1 - normalized_confidence)
# Adjust for query type
if (query_type == "creative"):
temperature = min(0.9, temperature + 0.2)
elif (query_type == "qa"):
temperature = max(0.1, temperature - 0.1)
return self._clamp_temperature(temperature = temperature)
def _progressive_temperature(self, query_type: str, query: str) -> float:
"""
Progressive temperature based on task requirements
"""
base_temp = self.base_temperature
# Task-based progression
if (query_type == "creative"):
# High creativity
return self._clamp_temperature(temperature = 0.8)
elif (query_type == "analytical"):
# Balanced
return self._clamp_temperature(temperature = 0.3)
elif (query_type == "qa"):
# For factual Q&A, use lower temperature
if self._is_factual_query(query):
return self._clamp_temperature(temperature = 0.1)
else:
return self._clamp_temperature(temperature = 0.4)
elif (query_type == "summary"):
# Deterministic summaries
return self._clamp_temperature(temperature = 0.2)
else:
return self._clamp_temperature(temperature = base_temp)
def _calculate_query_complexity(self, query: str) -> float:
"""
Simple, predictable complexity score
"""
if not query:
return 0.5
# Count words and questions
words = len(query.split())
has_why_how = any(word in query.lower() for word in ['why', 'how', 'explain'])
has_compare = any(word in query.lower() for word in ['compare', 'contrast', 'difference'])
# Simple rules
if has_compare:
# Complex
return 0.8
elif (has_why_how and( words > 15)):
return 0.7
elif words > 20:
return 0.6
else:
# Simple
return 0.3
def _calculate_context_quality(self, context: str) -> float:
"""
Calculate context quality (0.0 - 1.0)
"""
if not context:
return 0.0
factors = list()
# Length factor (adequate context)
words = len(context.split())
# Normalize
length_factor = min(words / 500, 1.0)
factors.append(length_factor)
# Diversity factor (multiple sources/citations)
citation_count = context.count('[')
diversity_factor = min(citation_count / 5, 1.0)
factors.append(diversity_factor)
# Coherence factor (simple measure)
sentence_count = context.count('.')
if (sentence_count > 0):
avg_sentence_length = words / sentence_count
# Ideal ~20 words/sentence
coherence_factor = 1.0 - min(abs(avg_sentence_length - 20) / 50, 1.0)
factors.append(coherence_factor)
return sum(factors) / len(factors)
def _is_factual_query(self, query: str) -> bool:
"""
Check if query is factual (requires precise answers)
"""
factual_indicators = ['what is', 'who is', 'when did', 'where is', 'how many', 'how much', 'definition of', 'meaning of', 'calculate', 'number of']
query_lower = query.lower()
return any(indicator in query_lower for indicator in factual_indicators)
def _clamp_temperature(self, temperature: float) -> float:
"""
Clamp temperature to valid range
"""
strategy_config = self.strategy_configs.get(self.strategy, {})
temp_range = strategy_config.get("range", (0.0, 1.0))
clamped = max(temp_range[0], min(temperature, temp_range[1]))
# Round to 2 decimal places
clamped = round(clamped, 2)
return clamped
def get_temperature_parameters(self, temperature: float) -> Dict[str, Any]:
"""
Get additional parameters based on temperature
Arguments:
----------
temperature { float } : Temperature value
Returns:
--------
{ dict } : Additional generation parameters
"""
params = {"temperature" : temperature,
"top_p" : 0.9,
}
# Adjust top_p based on temperature
if (temperature < 0.3):
# Broader distribution for low temp
params["top_p"] = 0.95
elif (temperature > 0.7):
# Narrower distribution for high temp
params["top_p"] = 0.7
# Adjust presence_penalty based on temperature
if (temperature > 0.5):
# Encourage novelty for creative tasks
params["presence_penalty"] = 0.1
else:
params["presence_penalty"] = 0.0
return params
def explain_temperature_choice(self, query: str, context: str, retrieval_scores: list, query_type: str, final_temperature: float) -> Dict[str, Any]:
"""
Explain why a particular temperature was chosen
Arguments:
----------
query { str } : User query
context { str } : Retrieved context
retrieval_scores { list } : Retrieval scores
query_type { str } : Query type
final_temperature { float } : Chosen temperature
Returns:
--------
{ dict } : Explanation dictionary
"""
explanation = {"strategy" : self.strategy.value,
"final_temperature" : final_temperature,
"base_temperature" : self.base_temperature,
"factors" : {},
}
if (self.strategy == TemperatureStrategy.ADAPTIVE):
complexity = self._calculate_query_complexity(query = query)
context_quality = self._calculate_context_quality(context = context)
explanation["factors"] = {"query_complexity" : round(complexity, 3),
"context_quality" : round(context_quality, 3),
"query_type" : query_type,
}
elif (self.strategy == TemperatureStrategy.CONFIDENCE):
if retrieval_scores:
avg_confidence = sum(retrieval_scores) / len(retrieval_scores)
explanation["factors"] = {"average_retrieval_confidence" : round(avg_confidence, 3),
"query_type" : query_type,
}
elif (self.strategy == TemperatureStrategy.PROGRESSIVE):
is_factual = self._is_factual_query(query)
explanation["factors"] = {"query_type" : query_type,
"is_factual_query" : is_factual,
}
return explanation
# Global temperature controller instance
_temperature_controller = None
def get_temperature_controller() -> TemperatureController:
"""
Get global temperature controller instance (singleton)
Returns:
--------
{ TemperatureController } : TemperatureController instance
"""
global _temperature_controller
if _temperature_controller is None:
_temperature_controller = TemperatureController()
return _temperature_controller
@handle_errors(error_type=TemperatureControlError, log_error=True, reraise=False)
def get_adaptive_temperature(query: str = "", context: str = "", retrieval_scores: list = None, query_type: str = "qa") -> float:
"""
Convenience function for getting adaptive temperature
Arguments:
----------
query { str } : User query
context { str } : Retrieved context
retrieval_scores { list } : Retrieval scores
query_type { str } : Query type
Returns:
--------
{ float } : Temperature value
"""
controller = get_temperature_controller()
return controller.get_temperature(query = query,
context = context,
retrieval_scores = retrieval_scores,
query_type = query_type,
) |