Spaces:
Sleeping
Sleeping
File size: 23,363 Bytes
8205d6b |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 |
# DEPENDENCIES
import os
import gc
import json
import torch
import spacy
import threading
import subprocess
from typing import Any
from typing import Dict
from typing import Union
from pathlib import Path
from loguru import logger
from typing import Optional
from datetime import datetime
from transformers import pipeline
from collections import OrderedDict
from config.settings import settings
from transformers import GPT2Tokenizer
from transformers import AutoTokenizer
from transformers import GPT2LMHeadModel
from config.model_config import ModelType
from config.model_config import ModelConfig
from transformers import AutoModelForMaskedLM
from config.model_config import MODEL_REGISTRY
from config.model_config import get_model_config
from config.model_config import get_required_models
from sentence_transformers import SentenceTransformer
from transformers import AutoModelForSequenceClassification
class ModelCache:
"""
LRU cache for models with size limit
"""
def __init__(self, max_size: int = 5):
self.max_size = max_size
self.cache : OrderedDict = OrderedDict()
self.lock = threading.Lock()
def get(self, key: str) -> Optional[Any]:
"""
Get model from cache
"""
with self.lock:
if key in self.cache:
# Move to end (most recently used)
self.cache.move_to_end(key)
logger.debug(f"Cache hit for model: {key}")
return self.cache[key]
logger.debug(f"Cache miss for model: {key}")
return None
def put(self, key: str, model: Any):
"""
Add model to cache
"""
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
removed_key = next(iter(self.cache))
removed_model = self.cache.pop(removed_key)
# Clean up memory
if hasattr(removed_model, 'to'):
removed_model.to('cpu')
del removed_model
if torch.cuda.is_available():
torch.cuda.empty_cache()
logger.info(f"Evicted model from cache: {removed_key}")
self.cache[key] = model
logger.info(f"Added model to cache: {key}")
def clear(self):
"""
Clear all cached models
"""
with self.lock:
for model in self.cache.values():
if hasattr(model, 'to'):
model.to('cpu')
del model
self.cache.clear()
if torch.cuda.is_available():
torch.cuda.empty_cache()
logger.info("Cleared model cache")
def size(self) -> int:
"""
Get current cache size
"""
return len(self.cache)
class ModelManager:
"""
Central model management system
"""
def __init__(self):
self.cache = ModelCache(max_size = settings.MAX_CACHED_MODELS)
self.device = torch.device(settings.DEVICE if torch.cuda.is_available() else "cpu")
self.cache_dir = settings.MODEL_CACHE_DIR
self.cache_dir.mkdir(parents = True,
exist_ok = True,
)
# Model metadata tracking
self.metadata_file = self.cache_dir / "model_metadata.json"
self.metadata = self._load_metadata()
logger.info(f"ModelManager initialized with device: {self.device}")
logger.info(f"Model cache directory: {self.cache_dir}")
def _load_metadata(self) -> Dict:
"""
Load model metadata from disk
"""
if self.metadata_file.exists():
try:
with open(self.metadata_file, 'r') as f:
return json.load(f)
except Exception as e:
logger.warning(f"Failed to load metadata: {repr(e)}")
return {}
def _save_metadata(self):
"""
Save model metadata to disk
"""
try:
with open(self.metadata_file, 'w') as f:
json.dump(obj = self.metadata,
fp = f,
indent = 4,
)
except Exception as e:
logger.error(f"Failed to save metadata: {repr(e)}")
def _update_metadata(self, model_name: str, model_config: ModelConfig):
"""
Update metadata for a model
"""
self.metadata[model_name] = {"model_id" : model_config.model_id,
"model_type" : model_config.model_type.value,
"downloaded_at" : datetime.now().isoformat(),
"size_mb" : model_config.size_mb,
"last_used" : datetime.now().isoformat(),
}
self._save_metadata()
def is_model_downloaded(self, model_name: str) -> bool:
"""
Check if model is already downloaded
"""
model_config = get_model_config(model_name = model_name)
if not model_config:
return False
# Check if model exists in cache directory
model_path = self.cache_dir / model_config.model_id.replace("/", "_")
return model_path.exists() and model_name in self.metadata
def load_model(self, model_name: str, force_download: bool = False) -> Any:
"""
Load a model by name
Arguments:
----------
model_name { str } : Name from MODEL_REGISTRY
force_download { bool } : Force re-download even if cached
Returns:
--------
{ Any } : Model instance
"""
# Check cache first
if not force_download:
cached = self.cache.get(key = model_name)
if cached is not None:
return cached
# Get model configuration
model_config = get_model_config(model_name = model_name)
if not model_config:
raise ValueError(f"Unknown model: {model_name}")
logger.info(f"Loading model: {model_name} ({model_config.model_id})")
try:
# Load based on model type
if (model_config.model_type == ModelType.SENTENCE_TRANSFORMER):
model = self._load_sentence_transformer(config = model_config)
elif (model_config.model_type == ModelType.GPT):
model = self._load_gpt_model(config = model_config)
elif (model_config.model_type == ModelType.CLASSIFIER):
model = self._load_classifier(config = model_config)
elif (model_config.model_type == ModelType.SEQUENCE_CLASSIFICATION):
model = self._load_sequence_classifier(config = model_config)
elif (model_config.model_type == ModelType.TRANSFORMER):
model = self._load_transformer(config = model_config)
elif (model_config.model_type == ModelType.RULE_BASED):
# Check if it's a spaCy model
if model_config.additional_params.get("is_spacy_model", False):
model = self._load_spacy_model(config = model_config)
else:
raise ValueError(f"Unknown rule-based model type: {model_name}")
else:
raise ValueError(f"Unsupported model type: {model_config.model_type}")
# Update metadata
self._update_metadata(model_name = model_name,
model_config = model_config,
)
# Cache the model
if model_config.cache_model:
self.cache.put(key = model_name,
model = model,
)
logger.success(f"Successfully loaded model: {model_name}")
return model
except Exception as e:
logger.error(f"Failed to load model {model_name}: {repr(e)}")
raise
def load_tokenizer(self, model_name: str) -> Any:
"""
Load tokenizer for a model
Arguments:
----------
model_name { str } : Name from MODEL_REGISTRY
Returns:
--------
{ Any } : Tokenizer instance
"""
model_config = get_model_config(model_name = model_name)
if not model_config:
raise ValueError(f"Unknown model: {model_name}")
logger.info(f"Loading tokenizer for: {model_name}")
try:
if (model_config.model_type in [ModelType.GPT, ModelType.CLASSIFIER, ModelType.SEQUENCE_CLASSIFICATION, ModelType.TRANSFORMER]):
tokenizer = AutoTokenizer.from_pretrained(pretrained_model_name_or_path = model_config.model_id,
cache_dir = str(self.cache_dir),
)
logger.success(f"Successfully loaded tokenizer for: {model_name}")
return tokenizer
else:
raise ValueError(f"Model type {model_config.model_type} doesn't require a separate tokenizer")
except Exception as e:
logger.error(f"Failed to load tokenizer for {model_name}: {repr(e)}")
raise
def _load_sentence_transformer(self, config: ModelConfig) -> SentenceTransformer:
"""
Load SentenceTransformer model
"""
model = SentenceTransformer(model_name_or_path = config.model_id,
cache_folder = str(self.cache_dir),
device = str(self.device),
)
return model
def _load_gpt_model(self, config: ModelConfig) -> tuple:
"""
Load GPT-style model with tokenizer
"""
model = GPT2LMHeadModel.from_pretrained(pretrained_model_name_or_path = config.model_id,
cache_dir = str(self.cache_dir),
)
tokenizer = GPT2Tokenizer.from_pretrained(pretrained_model_name_or_path = config.model_id,
cache_dir = str(self.cache_dir),
)
# Move to device
model = model.to(self.device)
model.eval()
# Apply quantization if enabled
if (settings.USE_QUANTIZATION and config.quantizable):
model = self._quantize_model(model = model)
return (model, tokenizer)
def _load_classifier(self, config: ModelConfig) -> Any:
"""
Load classification model (for zero-shot, etc.)
"""
# For zero-shot classification models
pipe = pipeline("zero-shot-classification",
model = config.model_id,
device = 0 if self.device.type == "cuda" else -1,
model_kwargs = {"cache_dir": str(self.cache_dir)},
)
return pipe
def _load_sequence_classifier(self, config: ModelConfig) -> Any:
"""
Load sequence classification model (for domain classification)
"""
model = AutoModelForSequenceClassification.from_pretrained(pretrained_model_name_or_path = config.model_id,
cache_dir = str(self.cache_dir),
num_labels = config.additional_params.get('num_labels', 2),
)
# Move to device
model = model.to(self.device)
model.eval()
# Apply quantization if enabled
if (settings.USE_QUANTIZATION and config.quantizable):
model = self._quantize_model(model = model)
return model
def _load_transformer(self, config: ModelConfig) -> tuple:
"""
Load masking transformer model
"""
model = AutoModelForMaskedLM.from_pretrained(pretrained_model_name_or_path = config.model_id,
cache_dir = str(self.cache_dir),
)
tokenizer = AutoTokenizer.from_pretrained(pretrained_model_name_or_path = config.model_id,
cache_dir = str(self.cache_dir),
)
# Move to device
model = model.to(self.device)
model.eval()
# Apply quantization if enabled
if (settings.USE_QUANTIZATION and config.quantizable):
model = self._quantize_model(model)
return (model, tokenizer)
def _quantize_model(self, model):
"""
Apply INT8 quantization to model
"""
try:
if hasattr(torch.quantization, 'quantize_dynamic'):
quantized_model = torch.quantization.quantize_dynamic(model = model,
qconfig_spec = {torch.nn.Linear},
dtype = torch.qint8,
)
logger.info("Applied INT8 quantization to model")
return quantized_model
except Exception as e:
logger.warning(f"Quantization failed: {repr(e)}, using original model")
return model
def load_pipeline(self, model_name: str, task: str) -> pipeline:
"""
Load a Hugging Face pipeline
"""
model_config = get_model_config(model_name = model_name)
if not model_config:
raise ValueError(f"Unknown model: {model_name}")
logger.info(f"Loading pipeline: {task} with {model_name}")
pipe = pipeline(task = task,
model = model_config.model_id,
device = 0 if self.device.type == "cuda" else -1,
model_kwargs = {"cache_dir": str(self.cache_dir)},
)
return pipe
def _load_spacy_model(self, config: ModelConfig):
"""
Load spaCy model
"""
try:
model = spacy.load(config.model_id)
logger.info(f"Loaded spaCy model: {config.model_id}")
return model
except OSError:
# Model not downloaded, install it
logger.info(f"Downloading spaCy model: {config.model_id}")
subprocess.run(["python", "-m", "spacy", "download", config.model_id], check = True)
model = spacy.load(config.model_id)
return model
def download_model(self, model_name: str) -> bool:
"""
Download model without loading it into memory
Arguments:
----------
model_name { str } : Name from MODEL_REGISTRY
Returns:
--------
{ bool } : True if successful, False otherwise
"""
model_config = get_model_config(model_name)
if not model_config:
logger.error(f"Unknown model: {model_name}")
return False
if self.is_model_downloaded(model_name):
logger.info(f"Model already downloaded: {model_name}")
return True
logger.info(f"Downloading model: {model_name} ({model_config.model_id})")
try:
if model_config.model_type == ModelType.SENTENCE_TRANSFORMER:
SentenceTransformer(model_name_or_path = model_config.model_id,
cache_folder = str(self.cache_dir),
)
elif (model_config.model_type == ModelType.GPT):
GPT2LMHeadModel.from_pretrained(pretrained_model_name_or_path = model_config.model_id,
cache_dir = str(self.cache_dir),
)
GPT2Tokenizer.from_pretrained(pretrained_model_name_or_path = model_config.model_id,
cache_dir = str(self.cache_dir),
)
elif (model_config.model_type == ModelType.SEQUENCE_CLASSIFICATION):
AutoModelForSequenceClassification.from_pretrained(pretrained_model_name_or_path = model_config.model_id,
cache_dir = str(self.cache_dir),
)
AutoTokenizer.from_pretrained(pretrained_model_name_or_path = model_config.model_id,
cache_dir = str(self.cache_dir),
)
elif (model_config.model_type == ModelType.RULE_BASED):
if model_config.additional_params.get("is_spacy_model", False):
subprocess.run(["python", "-m", "spacy", "download", model_config.model_id], check = True)
else:
logger.warning(f"Cannot pre-download rule-based model: {model_name}")
# Mark as "downloaded"
return True
else:
# Generic transformer models
AutoModelForSequenceClassification.from_pretrained(pretrained_model_name_or_path = model_config.model_id,
cache_dir = str(self.cache_dir),
)
AutoTokenizer.from_pretrained(pretrained_model_name_or_path = model_config.model_id,
cache_dir = str(self.cache_dir),
)
self._update_metadata(model_name, model_config)
logger.success(f"Successfully downloaded: {model_name}")
return True
except Exception as e:
logger.error(f"Failed to download {model_name}: {repr(e)}")
return False
def download_all_required(self) -> Dict[str, bool]:
"""
Download all required models
Returns:
--------
{ dict } : Dict mapping model names to success status
"""
required_models = get_required_models()
results = dict()
logger.info(f"Downloading {len(required_models)} required models...")
for model_name in required_models:
results[model_name] = self.download_model(model_name = model_name)
success_count = sum(1 for v in results.values() if v)
logger.info(f"Downloaded {success_count}/{len(required_models)} required models")
return results
def get_model_info(self, model_name: str) -> Optional[Dict]:
"""
Get information about a model
"""
return self.metadata.get(model_name)
def list_downloaded_models(self) -> list:
"""
List all downloaded models
"""
return list(self.metadata.keys())
def clear_cache(self):
"""
Clear model cache
"""
self.cache.clear()
logger.info("Model cache cleared")
def unload_model(self, model_name: str):
"""
Unload a specific model from cache
"""
with self.cache.lock:
if model_name in self.cache.cache:
model = self.cache.cache.pop(model_name)
if hasattr(model, 'to'):
model.to('cpu')
del model
if torch.cuda.is_available():
torch.cuda.empty_cache()
logger.info(f"Unloaded model: {model_name}")
def get_memory_usage(self) -> Dict[str, Any]:
"""
Get current memory usage statistics
"""
stats = {"cached_models" : self.cache.size(),
"device" : str(self.device),
}
if torch.cuda.is_available():
stats.update({"gpu_allocated_mb" : torch.cuda.memory_allocated() / 1024**2,
"gpu_reserved_mb" : torch.cuda.memory_reserved() / 1024**2,
"gpu_max_allocated_mb" : torch.cuda.max_memory_allocated() / 1024**2,
})
return stats
def optimize_memory(self):
"""
Optimize memory usage
"""
logger.info("Optimizing memory...")
# Clear unused cached models
self.cache.clear()
# Force garbage collection
gc.collect()
# Clear CUDA cache if available
if torch.cuda.is_available():
torch.cuda.empty_cache()
logger.info("Memory optimization complete")
logger.info(f"Memory usage: {self.get_memory_usage()}")
# Singleton instance
_model_manager_instance : Optional[ModelManager] = None
_manager_lock = threading.Lock()
def get_model_manager() -> ModelManager:
"""
Get singleton ModelManager instance
"""
global _model_manager_instance
if _model_manager_instance is None:
with _manager_lock:
if _model_manager_instance is None:
_model_manager_instance = ModelManager()
return _model_manager_instance
# Export
__all__ = ["ModelManager",
"ModelCache",
"get_model_manager",
] |