Spaces:
Running
Running
File size: 10,437 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 |
# DEPENDENCIES
import os
import gc
import torch
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 EmbeddingError
from sentence_transformers import SentenceTransformer
# Setup Settings and Logging
settings = get_settings()
logger = get_logger(__name__)
class EmbeddingModelLoader:
"""
Manages loading and caching of embedding models: Supports multiple models with efficient resource management
"""
def __init__(self):
self.logger = logger
self._loaded_model = None
self._model_name = None
self._device = None
# Model cache for multiple models
self._model_cache = dict()
@handle_errors(error_type = EmbeddingError, log_error = True, reraise = True)
def load_model(self, model_name: Optional[str] = None, device: Optional[str] = None, force_reload: bool = False) -> SentenceTransformer:
"""
Load embedding model with caching and device optimization
Arguments:
----------
model_name { str } : Name of model to load (default from settings)
device { str } : Device to load on ('cpu', 'cuda', 'mps', 'auto')
force_reload { bool } : Force reload even if model is cached
Returns:
--------
{ SentenceTransformer } : Loaded model instance
Raises:
-------
EmbeddingError : If model loading fails
"""
model_name = model_name or settings.EMBEDDING_MODEL
device = self._resolve_device(device)
# Check cache first
cache_key = f"{model_name}_{device}"
if ((not force_reload) and (cache_key in self._model_cache)):
self.logger.debug(f"Using cached model: {cache_key}")
self._loaded_model = self._model_cache[cache_key]
self._model_name = model_name
self._device = device
return self._loaded_model
try:
self.logger.info(f"Loading embedding model: {model_name} on device: {device}")
# Load model with optimized settings
model = SentenceTransformer(model_name,
device = device,
cache_folder = os.path.expanduser("~/.cache/sentence_transformers"),
)
# Model-specific optimizations
model = self._optimize_model(model = model,
device = device,
)
# Cache the model
self._model_cache[cache_key] = model
self._loaded_model = model
self._model_name = model_name
self._device = device
# Log model info
self._log_model_info(model = model,
device = device,
)
self.logger.info(f"Successfully loaded model: {model_name}")
return model
except Exception as e:
self.logger.error(f"Failed to load model {model_name}: {repr(e)}")
raise EmbeddingError(f"Model loading failed: {repr(e)}")
def _resolve_device(self, device: Optional[str] = None) -> str:
"""
Resolve the best available device
Arguments:
----------
device { str } : Requested device
Returns:
--------
{ str } : Actual device to use
"""
if (device and (device != "auto")):
return device
# Auto device selection
if (settings.EMBEDDING_DEVICE != "auto"):
return settings.EMBEDDING_DEVICE
# Automatic detection
if torch.cuda.is_available():
return "cuda"
elif torch.backends.mps.is_available():
return "mps"
else:
return "cpu"
def _optimize_model(self, model: SentenceTransformer, device: str) -> SentenceTransformer:
"""
Apply optimizations to the model
Arguments:
----------
model { SentenceTransformer } : Model to optimize
device { str } : Device model is on
Returns:
--------
{ SentenceTransformer } : Optimized model
"""
# Enable eval mode for inference
model.eval()
# GPU optimizations
if (device == "cuda"):
# Use half precision for GPU if supported
try:
model = model.half()
self.logger.debug("Enabled half precision for GPU")
except Exception as e:
self.logger.warning(f"Could not enable half precision: {repr(e)}")
# Disable gradient computation
for param in model.parameters():
param.requires_grad = False
return model
def _log_model_info(self, model: SentenceTransformer, device: str):
"""
Log detailed model information
Arguments:
----------
model { SentenceTransformer } : Model to log info for
device { str } : Device model is on
"""
try:
# Get model architecture info
if hasattr(model, '_modules'):
modules = list(model._modules.keys())
else:
modules = ["unknown"]
# Get embedding dimension
if hasattr(model, 'get_sentence_embedding_dimension'):
dimension = model.get_sentence_embedding_dimension()
else:
dimension = "unknown"
# Count parameters
total_params = sum(p.numel() for p in model.parameters())
self.logger.info(f"Model Info: {len(modules)} modules, dimension={dimension}, parameters={total_params:,}, device={device}")
except Exception as e:
self.logger.debug(f"Could not get detailed model info: {repr(e)}")
def get_loaded_model(self) -> Optional[SentenceTransformer]:
"""
Get currently loaded model
Returns:
--------
{ SentenceTransformer } : Currently loaded model or None
"""
return self._loaded_model
def get_model_info(self) -> dict:
"""
Get information about loaded model
Returns:
--------
{ dict } : Model information dictionary
"""
if self._loaded_model is None:
return {"loaded": False}
info = {"loaded" : True,
"model_name" : self._model_name,
"device" : self._device,
"cache_size" : len(self._model_cache),
}
try:
if hasattr(self._loaded_model, 'get_sentence_embedding_dimension'):
info["embedding_dimension"] = self._loaded_model.get_sentence_embedding_dimension()
info["model_class"] = type(self._loaded_model).__name__
except Exception as e:
self.logger.warning(f"Could not get detailed model info: {e}")
return info
def clear_cache(self, model_name: Optional[str] = None):
"""
Clear model cache
Arguments:
----------
model_name { str } : Specific model to clear (None = all)
"""
if model_name:
# Clear specific model from all devices
keys_to_remove = [k for k in self._model_cache.keys() if k.startswith(model_name)]
for key in keys_to_remove:
del self._model_cache[key]
self.logger.info(f"Cleared cache for model: {model_name}")
else:
# Clear all cache
cache_size = len(self._model_cache)
self._model_cache.clear()
self.logger.info(f"Cleared all model cache ({cache_size} models)")
def unload_model(self):
"""
Unload current model and free memory
"""
if self._loaded_model:
model_name = self._model_name
# Clear from cache
if self._model_name and self._device:
cache_key = f"{self._model_name}_{self._device}"
self._model_cache.pop(cache_key, None)
# Clear references
self._loaded_model = None
self._model_name = None
self._device = None
# Force garbage collection
gc.collect()
if torch.cuda.is_available():
torch.cuda.empty_cache()
self.logger.info(f"Unloaded model: {model_name}")
# Global model loader instance
_model_loader = None
def get_model_loader() -> EmbeddingModelLoader:
"""
Get global model loader instance (singleton)
Returns:
--------
{ EmbeddingModelLoader } : Model loader instance
"""
global _model_loader
if _model_loader is None:
_model_loader = EmbeddingModelLoader()
return _model_loader
def load_embedding_model(model_name: Optional[str] = None, device: Optional[str] = None) -> SentenceTransformer:
"""
Convenience function to load embedding model
Arguments:
----------
model_name { str } : Model name
device { str } : Device
Returns:
--------
{ SentenceTransformer } : Loaded model
"""
loader = get_model_loader()
return loader.load_model(model_name, device) |