Spaces:
Running
Running
File size: 11,869 Bytes
33fba47 |
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 |
import torch
import numpy as np
import gc
import threading
import langdetect
import logging
from collections import OrderedDict, Counter
from transformers import AutoTokenizer, AutoModelForSequenceClassification
from concurrent.futures import ThreadPoolExecutor
from functools import lru_cache, wraps
from contextlib import contextmanager
from typing import List, Dict, Optional, Tuple, Any, Callable
import re
from config import config
logger = logging.getLogger(__name__)
# Decorators and Context Managers
def handle_errors(default_return=None):
"""Centralized error handling decorator"""
def decorator(func: Callable) -> Callable:
@wraps(func)
def wrapper(*args, **kwargs):
try:
return func(*args, **kwargs)
except Exception as e:
logger.error(f"{func.__name__} failed: {e}")
return default_return if default_return is not None else f"Error: {str(e)}"
return wrapper
return decorator
@contextmanager
def memory_cleanup():
"""Context manager for memory cleanup"""
try:
yield
finally:
gc.collect()
if torch.cuda.is_available():
torch.cuda.empty_cache()
class ThemeContext:
"""Theme management context"""
def __init__(self, theme: str = 'default'):
self.theme = theme
self.colors = config.THEMES.get(theme, config.THEMES['default'])
class LRUModelCache:
"""LRU Cache for models with memory management"""
def __init__(self, max_size: int = 2):
self.max_size = max_size
self.cache = OrderedDict()
self.lock = threading.Lock()
def get(self, key):
with self.lock:
if key in self.cache:
# Move to end (most recently used)
self.cache.move_to_end(key)
return self.cache[key]
return None
def put(self, key, value):
with self.lock:
if key in self.cache:
self.cache.move_to_end(key)
else:
if len(self.cache) >= self.max_size:
# Remove least recently used
oldest_key = next(iter(self.cache))
old_model, old_tokenizer = self.cache.pop(oldest_key)
# Force cleanup
del old_model, old_tokenizer
gc.collect()
if torch.cuda.is_available():
torch.cuda.empty_cache()
self.cache[key] = value
def clear(self):
with self.lock:
for model, tokenizer in self.cache.values():
del model, tokenizer
self.cache.clear()
gc.collect()
if torch.cuda.is_available():
torch.cuda.empty_cache()
# Enhanced Model Manager with Optimized Memory Management
class ModelManager:
"""Optimized multi-language model manager with LRU cache and lazy loading"""
_instance = None
def __new__(cls):
if cls._instance is None:
cls._instance = super().__new__(cls)
cls._instance._initialized = False
return cls._instance
def __init__(self):
if not self._initialized:
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
self.model_cache = LRUModelCache(config.MODEL_CACHE_SIZE)
self.loading_lock = threading.Lock()
self._initialized = True
logger.info(f"ModelManager initialized on device: {self.device}")
def _load_model(self, model_name: str, cache_key: str):
"""Load model with memory optimization"""
try:
logger.info(f"Loading model: {model_name}")
# Load with memory optimization
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(
model_name,
torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
device_map="auto" if torch.cuda.is_available() else None
)
if not torch.cuda.is_available():
model.to(self.device)
# Set to eval mode to save memory
model.eval()
# Cache the model
self.model_cache.put(cache_key, (model, tokenizer))
logger.info(f"Model {model_name} loaded and cached successfully")
return model, tokenizer
except Exception as e:
logger.error(f"Failed to load model {model_name}: {e}")
raise
def get_model(self, language='en'):
"""Get model for specific language with lazy loading and caching"""
# Determine cache key and model name
if language == 'zh':
cache_key = 'zh'
model_name = config.MODELS['zh']
else:
cache_key = 'multilingual'
model_name = config.MODELS['multilingual']
# Try to get from cache first
cached_model = self.model_cache.get(cache_key)
if cached_model is not None:
return cached_model
# Load model if not in cache (with thread safety)
with self.loading_lock:
# Double-check pattern
cached_model = self.model_cache.get(cache_key)
if cached_model is not None:
return cached_model
return self._load_model(model_name, cache_key)
@staticmethod
def detect_language(text: str) -> str:
"""Detect text language"""
try:
detected = langdetect.detect(text)
language_mapping = {
'zh-cn': 'zh',
'zh-tw': 'zh'
}
detected = language_mapping.get(detected, detected)
return detected if detected in config.SUPPORTED_LANGUAGES else 'en'
except:
return 'en'
# Core Sentiment Analysis Engine with Performance Optimizations
class SentimentEngine:
"""Optimized multi-language sentiment analysis engine"""
def __init__(self):
self.model_manager = ModelManager()
self.executor = ThreadPoolExecutor(max_workers=4)
@handle_errors(default_return={'sentiment': 'Unknown', 'confidence': 0.0})
def analyze_single(self, text: str, language: str = 'auto', preprocessing_options: Dict = None) -> Dict:
"""Optimized single text analysis"""
if not text.strip():
raise ValueError("Empty text provided")
# Detect language
if language == 'auto':
detected_lang = self.model_manager.detect_language(text)
else:
detected_lang = language
# Get appropriate model
model, tokenizer = self.model_manager.get_model(detected_lang)
# Preprocessing
options = preprocessing_options or {}
processed_text = text
if options.get('clean_text', False) and not re.search(r'[\u4e00-\u9fff]', text):
from data_utils import TextProcessor
processed_text = TextProcessor.clean_text(
text,
options.get('remove_punctuation', True),
options.get('remove_numbers', False)
)
# Tokenize and analyze with memory optimization
inputs = tokenizer(processed_text, return_tensors="pt", padding=True,
truncation=True, max_length=config.MAX_TEXT_LENGTH).to(self.model_manager.device)
# Use no_grad for inference to save memory
with torch.no_grad():
outputs = model(**inputs)
probs = torch.nn.functional.softmax(outputs.logits, dim=-1).cpu().numpy()[0]
# Clear GPU cache after inference
if torch.cuda.is_available():
torch.cuda.empty_cache()
# Handle different model outputs
if len(probs) == 3: # negative, neutral, positive
sentiment_idx = np.argmax(probs)
sentiment_labels = ['Negative', 'Neutral', 'Positive']
sentiment = sentiment_labels[sentiment_idx]
confidence = float(probs[sentiment_idx])
result = {
'sentiment': sentiment,
'confidence': confidence,
'neg_prob': float(probs[0]),
'neu_prob': float(probs[1]),
'pos_prob': float(probs[2]),
'has_neutral': True
}
else: # negative, positive
pred = np.argmax(probs)
sentiment = "Positive" if pred == 1 else "Negative"
confidence = float(probs[pred])
result = {
'sentiment': sentiment,
'confidence': confidence,
'neg_prob': float(probs[0]),
'pos_prob': float(probs[1]),
'neu_prob': 0.0,
'has_neutral': False
}
# Add metadata
result.update({
'language': detected_lang,
'word_count': len(text.split()),
'char_count': len(text)
})
return result
def _analyze_text_batch(self, text: str, language: str, preprocessing_options: Dict, index: int) -> Dict:
"""Single text analysis for batch processing"""
try:
result = self.analyze_single(text, language, preprocessing_options)
result['batch_index'] = index
result['text'] = text[:100] + '...' if len(text) > 100 else text
result['full_text'] = text
return result
except Exception as e:
return {
'sentiment': 'Error',
'confidence': 0.0,
'error': str(e),
'batch_index': index,
'text': text[:100] + '...' if len(text) > 100 else text,
'full_text': text
}
@handle_errors(default_return=[])
def analyze_batch(self, texts: List[str], language: str = 'auto',
preprocessing_options: Dict = None, progress_callback=None) -> List[Dict]:
"""Optimized parallel batch processing"""
if len(texts) > config.BATCH_SIZE_LIMIT:
texts = texts[:config.BATCH_SIZE_LIMIT]
if not texts:
return []
# Pre-load model to avoid race conditions
self.model_manager.get_model(language if language != 'auto' else 'en')
# Use ThreadPoolExecutor for parallel processing
with ThreadPoolExecutor(max_workers=min(4, len(texts))) as executor:
futures = []
for i, text in enumerate(texts):
future = executor.submit(
self._analyze_text_batch,
text, language, preprocessing_options, i
)
futures.append(future)
results = []
for i, future in enumerate(futures):
if progress_callback:
progress_callback((i + 1) / len(futures))
try:
result = future.result(timeout=30) # 30 second timeout per text
results.append(result)
except Exception as e:
results.append({
'sentiment': 'Error',
'confidence': 0.0,
'error': f"Timeout or error: {str(e)}",
'batch_index': i,
'text': texts[i][:100] + '...' if len(texts[i]) > 100 else texts[i],
'full_text': texts[i]
})
return results |