File size: 39,518 Bytes
bf64b40 61c5cf1 bf64b40 4275d16 bf64b40 4275d16 bf64b40 e6838c8 bf64b40 4275d16 bf64b40 4275d16 bf64b40 4275d16 bf64b40 4275d16 bf64b40 4275d16 bf64b40 4275d16 e6838c8 4275d16 4b1fd1d 4275d16 4b1fd1d 4275d16 bf64b40 e6838c8 bf64b40 4275d16 bf64b40 4275d16 bf64b40 61c5cf1 e6838c8 bf64b40 e6838c8 bf64b40 e6838c8 bf64b40 e6838c8 bf64b40 e6838c8 bf64b40 e6838c8 bf64b40 2de4a17 bf64b40 2de4a17 1acab08 2de4a17 1acab08 2de4a17 bf64b40 e6838c8 bf64b40 e6838c8 bf64b40 e6838c8 bf64b40 e6838c8 bf64b40 e6838c8 bf64b40 e6838c8 bf64b40 e6838c8 bf64b40 e6838c8 bf64b40 e6838c8 bf64b40 e6838c8 bf64b40 e6838c8 bf64b40 e6838c8 bf64b40 fa96858 ca9ef1a fa96858 ca9ef1a 2de4a17 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 |
import gc, os, sys, time, torch, logging, inspect, numpy as np, pandas as pd, importlib.util
from pathlib import Path
from threading import Lock
from collections import OrderedDict
from nltk.stem import WordNetLemmatizer
from typing import List, Dict, Any, Tuple, Optional
from sklearn.metrics.pairwise import cosine_similarity
from config import app_config
from dataset import TensorDataset
from utils.transformer_utils import get_sentence_transformer
from utils.smartHybridAttention import SmartHybridAttention, get_hybrid_attention_config
from transformers import AutoModelForCausalLM, AutoTokenizer
from service_registry import registry, MODEL, TOKENIZER, MODEL_MANAGER, COMMUNICATOR
logger = logging.getLogger(__name__)
try:
import psutil
PSUTIL_AVAILABLE = True
except ImportError:
logger.warning("psutil not available")
PSUTIL_AVAILABLE = False
class DummyProcess:
def __init__(self, pid=None): self.pid = pid or 1
def memory_info(self):
class MemInfo:
def __init__(self): self.rss = 1e6; self.vms = 1e6
return MemInfo()
def memory_percent(self): return 1.0
class DummyPsutil:
@staticmethod
def Process(pid=None): return DummyProcess(pid)
psutil = DummyPsutil()
def safe_get_config(config_obj, key, default=None):
if isinstance(config_obj, dict):
return config_obj.get(key, default)
elif hasattr(config_obj, key):
return getattr(config_obj, key, default)
return default
def safe_get_config_value(config_obj, key, default=None):
try:
if isinstance(config_obj, dict):
return config_obj.get(key, default)
elif hasattr(config_obj, key):
return getattr(config_obj, key, default)
elif isinstance(config_obj, (int, float, str, bool)):
return config_obj
return default
except:
return default
class DatasetManager:
def __init__(self):
self.datasets: Dict[str, Any] = {}
self.lock = Lock()
def load_dataset(self, path: str, specialization: str) -> Any:
with self.lock:
if specialization in self.datasets:
logger.info(f"Using cached dataset for {specialization}")
return self.datasets[specialization]
dataset = self._load_and_process_dataset(path, specialization)
self.datasets[specialization] = dataset
return dataset
def _load_and_process_dataset(self, path: str, specialization: str) -> TensorDataset:
if not os.path.exists(path):
raise FileNotFoundError(f"Dataset {path} not found.")
logger.info(f"Loading dataset: {specialization}")
data = pd.read_csv(path)
if "label" not in data.columns:
raise ValueError("Dataset must have a 'label' column.")
features = data.drop("label", axis=1).values
labels = data["label"].values
features_tensor = torch.tensor(features, dtype=torch.float32)
labels_tensor = torch.tensor(labels, dtype=torch.long)
return TensorDataset(features_tensor, labels_tensor)
def get_status(self) -> Dict[str, Any]:
return {"loaded_datasets": list(self.datasets.keys()), "cache_size": len(self.datasets)}
def clear_cache(self):
with self.lock:
self.datasets.clear()
class ModelManager:
HF_ALIAS = {
"EvolphTech/Wildnerve-tlm01_Hybrid_Model": "model_Combn.Wildnerve_tlm01_Hybrid_Model",
"model_Custm": "EvolphTech/Wildnerve-tlm01_Hybrid_Model",
"model_Custm.py": "EvolphTech/Wildnerve-tlm01_Hybrid_Model",
}
def __init__(self, tokenizer=None, max_active_models=5, model_idle_threshold=600):
self.models = {}
self.lock = Lock()
self.model_pool = OrderedDict()
self.max_active_models = max_active_models if isinstance(max_active_models, int) and max_active_models > 0 else 2
self.model_idle_threshold = model_idle_threshold if isinstance(model_idle_threshold, int) and model_idle_threshold > 0 else 600
self.tokenizer = tokenizer
# Remove hardcoded specializations and use config values
# First try SPECIALIZATIONS directly from config
if hasattr(app_config, 'SPECIALIZATIONS') and app_config.SPECIALIZATIONS:
self.specializations = app_config.SPECIALIZATIONS
# Then try keys from DATASET_PATHS
elif hasattr(app_config, 'DATASET_PATHS') and isinstance(app_config.DATASET_PATHS, dict):
self.specializations = list(app_config.DATASET_PATHS.keys())
# Fallback to minimal set
else:
self.specializations = ["general", "programming", "mathematics"]
logger.info(f"Using {len(self.specializations)} specializations from config")
self._performance_metrics = {}
attention_config = get_hybrid_attention_config()
self.smart_attention = SmartHybridAttention(
dim=attention_config["DIM"],
num_heads=attention_config["NUM_HEADS"],
window_size=attention_config["WINDOW_SIZE"],
use_sliding=attention_config["USE_SLIDING"],
use_global=attention_config["USE_GLOBAL"],
use_hierarchical=attention_config["USE_HIERARCHICAL"],
global_token_ratio=attention_config["GLOBAL_TOKEN_RATIO"],
memory_tokens=attention_config["MEMORY_TOKENS"]
)
self.dataset_manager = DatasetManager()
transformer_config = safe_get_config(app_config, "TRANSFORMER_CONFIG", {})
model_name = safe_get_config(transformer_config, "MODEL_NAME", "Wildnerve-tlm01-0.05Bx12")
self.embedding_model = get_sentence_transformer(model_name)
self.similarity_threshold = safe_get_config(app_config, "SIMILARITY_THRESHOLD", 0.85)
self.top_k = safe_get_config(app_config, "TOP_K", 3)
self.prompt_analyzer = None
self.selected_models = self._get_selected_models()
try:
self._load_models()
except Exception:
logger.critical("Startup model loading failed, aborting ModelManager init", exc_info=True)
raise
logger.info(f"ModelManager initialized with {len(self.specializations)} specializations")
def _get_selected_models(self) -> List[str]:
model_files = safe_get_config(app_config, "SELECTED_MODEL", ["model_Custm.py"])
return model_files if model_files else ["model_Custm.py"]
def _import_model_class(self, model_key: str):
"""Robust import of model classes or HF hub repos; raises on failure."""
key = model_key.rstrip(".py")
alias = self.HF_ALIAS.get(key)
try:
# 1) HF hub repo
if alias and "/" in alias:
logger.info(f"Loading HF model from repo '{alias}'")
model = AutoModelForCausalLM.from_pretrained(alias, use_auth_token=os.getenv("HF_TOKEN"))
tok = AutoTokenizer.from_pretrained(alias, use_auth_token=os.getenv("HF_TOKEN"))
model.tokenizer = tok
return model.__class__ # return the class (caller will instantiate)
# 2) Explicit module.Class mapping
if alias and "." in alias:
module_name, cls_name = alias.split(".", 1)
mod = importlib.import_module(module_name)
return getattr(mod, cls_name)
# 3) Local file fallback
file_path = os.path.join(os.path.dirname(__file__), f"{key}.py")
if os.path.isfile(file_path):
spec = importlib.util.spec_from_file_location(key, file_path)
mod = importlib.util.module_from_spec(spec)
spec.loader.exec_module(mod)
cls = getattr(mod, key, None) or getattr(mod, "Wildnerve_tlm01", None)
if cls:
return cls
raise ImportError(f"No class '{key}' or 'Wildnerve_tlm01' in {file_path}")
# 4) Standard Python import
mod = importlib.import_module(key)
cls = getattr(mod, key, None) or getattr(mod, "Wildnerve_tlm01", None)
if cls:
return cls
raise ImportError(f"No class '{key}' or 'Wildnerve_tlm01' in module '{key}'")
except Exception as e:
logger.error(f"Failed to import model class for '{model_key}': {e}", exc_info=True)
raise
def _initialize_model_for_specialization(self, spec: str, data_dir: str):
"""Instantiate a model for a given spec, with dataset-path handling and timeout warnings."""
# Resolve dataset path
ds_paths = safe_get_config_value(app_config, "DATASET_PATHS", {})
raw = ds_paths.get(spec)
if isinstance(raw, (list, tuple)):
dataset_path = raw[0]
else:
dataset_path = raw or os.path.join(data_dir, f"{spec}.csv")
# Ensure dataset exists (create minimal CSV if missing)
if not os.path.exists(dataset_path):
try:
with open(dataset_path, "w") as f:
f.write("text,label\nsample text,0\n")
logger.info(f"Created placeholder dataset for '{spec}'")
except Exception as e:
logger.warning(f"Could not create dataset for '{spec}': {e}")
# Import and instantiate model with GPT-2 parameters instead of BERT
model_cls = self._import_model_class(self.selected_models[0])
params = dict(
vocab_size=50257, # GPT-2 vocab size
specialization=spec,
dataset_path=dataset_path,
model_name=safe_get_config_value(app_config, "TRANSFORMER_CONFIG", {}).get("MODEL_NAME", "gpt2"),
embedding_dim=768, # Ensure 768-dimensional model
num_heads=12, # 12 heads for 768-dim
hidden_dim=768, # Ensure 768-dimensional model
num_layers=12, # More layers for larger model
output_size=50257, # GPT-2 vocab size
dropout=0.1,
max_seq_length=1024, # Increased for 768-dim model
pooling_mode=safe_get_config_value(app_config, "TRANSFORMER_CONFIG", {}).get("POOLING_MODE", "last"),
tokenizer=self.tokenizer
)
start = time.time()
try:
model = model_cls(**params)
except Exception as e:
logger.error(f"Error instantiating '{model_cls.__name__}' for '{spec}': {e}", exc_info=True)
raise
elapsed = time.time() - start
if elapsed > 30:
logger.warning(f"Model creation for '{spec}' took {elapsed:.1f}s (>30s)")
# Register
self.models[spec] = model
self.model_pool[spec] = None
self._performance_metrics[spec] = {
"inference_time": 0.0,
"memory_usage": 0.0,
"last_accessed": time.time(),
"num_inferences": 0
}
def _load_models(self):
"""Load initial specializations - now only preloading 'general' to save resources."""
# List of specializations to preload - can be expanded if desired
initial = ["general"] # Only preload the 'general' specialization for efficiency
# Uncomment the following line to preload more specializations
# initial = self.specializations[:2] # Load first 2 specializations
data_dir = os.environ.get("TLM_DATA_DIR", "/tmp/tlm_data")
os.makedirs(data_dir, exist_ok=True)
for spec in initial:
logger.info(f"Initializing model for specialization '{spec}'")
try:
self._initialize_model_for_specialization(spec, data_dir)
logger.info(f"Model for '{spec}' loaded successfully")
except Exception:
logger.error(f"Failed to load model for '{spec}'", exc_info=True)
raise
logger.info(f"{len(self.models)} models loaded at startup (of {len(self.specializations)} total)")
# Add debug info about available specializations
logger.debug(f"Available specializations: {', '.join(self.specializations)}")
def get_or_create_model(self, specialization: str) -> Any:
"""Get an existing model or create it on demand if not already loaded"""
with self.lock:
# Check if model already exists
model = self.get_model(specialization)
if (model):
logger.info(f"Using existing model for {specialization}")
return model
# Check if it's a valid specialization
if specialization not in self.all_specializations and specialization != "general":
logger.warning(f"Unknown specialization: {specialization}, using general")
specialization = "general"
# Create model if needed
logger.info(f"Lazily loading model for {specialization}")
# Remove least recently used model if needed
if len(self.models) >= self.max_active_models:
lru_specialization = next(iter(self.model_pool))
self.remove_model_instance(lru_specialization)
# Initialize the requested model
data_dir = os.environ.get("TLM_DATA_DIR", "/tmp/tlm_data")
try:
self._initialize_model_for_specialization(specialization, data_dir)
return self.models.get(specialization)
except Exception as e:
logger.error(f"Error initializing model: {e}")
# Fallback to general model
if specialization != "general" and "general" in self.models:
return self.models["general"]
# Last resort - create a minimal model
return self._create_minimal_model()
def _create_minimal_model(self):
"""Create a minimal fallback model for emergencies"""
try:
from model_Custm import Wildnerve_tlm01
model = Wildnerve_tlm01(
vocab_size=50257, # GPT-2 vocab size
specialization="minimal",
dataset_path=None,
model_name="gpt2", # Use GPT-2 instead of BERT
embedding_dim=768,
num_heads=12,
hidden_dim=768,
num_layers=2, # Reduced layers
output_size=50257, # Match GPT-2 vocab
dropout=0.1,
max_seq_length=128, # Reduced sequence length
pooling_mode="last", # GPT-2 uses last token
tokenizer=self.tokenizer
)
model._is_minimal = True # Mark as minimal model
return model
except Exception as e:
logger.error(f"Failed to create minimal model: {e}")
return None
def get_model(self, specialization: str) -> Any:
with self.lock:
model = self.models.get(specialization)
if model:
self.model_pool.move_to_end(specialization)
if specialization in self._performance_metrics:
self._performance_metrics[specialization]["last_accessed"] = time.time()
return model
def route_input(self, input_text: str) -> dict:
# Create embedding for input text
input_embedding = self.embedding_model.encode(input_text)
# Process input through SmartHybridAttention for enhanced understanding
if hasattr(self, 'smart_attention') and self.smart_attention:
try:
# Convert embedding to tensor format needed by attention
import torch
input_tensor = torch.tensor(input_embedding).unsqueeze(0).unsqueeze(0) # [1, 1, dim]
# Process through attention mechanism to extract key patterns
enhanced, _ = self.smart_attention( # FIXED: Properly unpack tuple
query=input_tensor,
key=input_tensor,
value=input_tensor
)
# Convert back to numpy for similarity calculations
if isinstance(enhanced, torch.Tensor):
enhanced_embedding = enhanced.squeeze().cpu().numpy()
# Use enhanced embedding for similarity calculation
input_embedding = enhanced_embedding
logger.info("Using SmartHybridAttention for enhanced prompt routing")
except Exception as e:
logger.warning(f"Error using SmartHybridAttention: {e}")
# Continue with existing similarity calculation
similarities = {}
for spec in self.specializations:
model = self.get_model(spec)
if model and hasattr(model, "embedding"):
sim = cosine_similarity(input_embedding.reshape(1, -1), model.embedding.reshape(1, -1))[0][0]
similarities[spec] = sim
if similarities:
best_match = max(similarities.items(), key=lambda x: x[1])
return {"matched_specialization": best_match[0], "confidence": best_match[1], "all_scores": similarities}
return {"matched_specialization": self.specializations[0], "confidence": 0.0, "all_scores": similarities}
def get_model_for_prompt(self, prompt: str) -> Tuple[Any, str]:
try:
routing_result = self.route_input(prompt)
specialization = routing_result.get("matched_specialization", self.specializations[0])
model = self.get_or_create_model(specialization)
start_time = time.time()
def update_metrics():
if specialization in self._performance_metrics:
m = self._performance_metrics[specialization]
elapsed = time.time() - start_time
n = m.get("num_inferences", 0) + 1
m["inference_time"] = ((m.get("inference_time", 0) * (n-1)) + elapsed) / n
m["num_inferences"] = n
m["last_accessed"] = time.time()
if hasattr(model, "get_memory_usage"):
m["memory_usage"] = model.get_memory_usage()
update_metrics()
return model, specialization
except Exception as e:
logger.error(f"Error selecting model: {e}")
if self.models:
default_key = list(self.models.keys())[0]
return self.models[default_key], default_key
else:
logger.error("No models available for routing")
return None, "none"
def generate(self, prompt: str, **kwargs):
self.validate_input(prompt)
model, specialization = self.get_model_for_prompt(prompt)
start_time = time.time()
try:
result = model.generate(prompt=prompt, **kwargs)
elapsed = time.time() - start_time
if specialization in self._performance_metrics:
m = self._performance_metrics[specialization]
n = m.get("num_inferences", 0) + 1
m["inference_time"] = ((m.get("inference_time", 0) * (n-1)) + elapsed) / n
m["num_inferences"] = n
m["last_accessed"] = time.time()
return result
except Exception as e:
logger.error(f"Error generating with {specialization}: {e}")
default_spec = self.specializations[0]
default_model = self.get_or_create_model(default_spec)
return default_model.generate(prompt=prompt, **kwargs)
def generate_streaming(self, prompt: str, **kwargs):
self.validate_input(prompt)
model, specialization = self.get_model_for_prompt(prompt)
start_time = time.time()
try:
if hasattr(model, "generate_streaming"):
for token in model.generate_streaming(prompt=prompt, **kwargs):
yield token
else:
logger.info("Simulating streaming generation")
result = model.generate(prompt=prompt, **kwargs)
for word in result.split():
yield word + " "
elapsed = time.time() - start_time
if specialization in self._performance_metrics:
m = self._performance_metrics[specialization]
n = m.get("num_inferences", 0) + 1
m["inference_time"] = ((m.get("inference_time", 0) * (n-1)) + elapsed) / n
m["num_inferences"] = n
m["last_accessed"] = time.time()
except Exception as e:
logger.error(f"Error in streaming generation: {e}")
default_spec = self.specializations[0]
default_model = self.get_or_create_model(default_spec)
if hasattr(default_model, "generate_streaming"):
for token in default_model.generate_streaming(prompt=prompt, **kwargs):
yield token
else:
fallback_result = default_model.generate(prompt=prompt, **kwargs)
for word in fallback_result.split():
yield word + " "
def remove_model_instance(self, specialization: str) -> bool:
with self.lock:
if specialization in self.models:
del self.models[specialization]
self.model_pool.pop(specialization, None)
gc.collect()
if torch.cuda.is_available():
torch.cuda.empty_cache()
logger.info(f"Removed model for {specialization}")
return True
return False
def validate_input(self, input_text: str) -> bool:
if not input_text or len(input_text.strip()) == 0:
raise ValueError("Empty input text")
max_length = safe_get_config(app_config, "MAX_INPUT_LENGTH", safe_get_config(app_config, "MAX_SEQ_LENGTH", 128))
if len(input_text) > max_length:
raise ValueError(f"Input exceeds maximum length of {max_length}")
return True
def get_health_status(self) -> Dict[str, Any]:
with self.lock:
process = psutil.Process(os.getpid())
mem_info = process.memory_info()
return {
"active_models": len(self.models),
"memory_usage": {
"rss_mb": mem_info.rss / (1024 * 1024),
"vms_mb": mem_info.vms / (1024 * 1024),
"percent": process.memory_percent()
},
"model_performance": self._get_model_metrics(),
"dataset_status": self.dataset_manager.get_status(),
"cache_efficiency": len(self.model_pool) / max(1, self.max_active_models)
}
def _get_model_metrics(self) -> Dict[str, Dict[str, Any]]:
metrics = {}
for spec, model in self.models.items():
base = self._performance_metrics.get(spec, {})
mem_usage = 0
if hasattr(model, "get_memory_usage"):
mem_usage = model.get_memory_usage()
elif hasattr(model, "parameters"):
mem_usage = sum(p.numel() * p.element_size() for p in model.parameters()) / (1024 * 1024)
metrics[spec] = {
"inference_time": base.get("inference_time", 0),
"memory_usage_mb": mem_usage,
"last_accessed": base.get("last_accessed", 0),
"num_inferences": base.get("num_inferences", 0),
"model_type": model.__class__.__name__
}
return metrics
def get_available_models(self) -> Dict[str, Any]:
with self.lock:
return dict(self.models)
def shutdown(self):
try:
logger.info("Initiating shutdown")
for spec in list(self.models.keys()):
self.remove_model_instance(spec)
self.dataset_manager.clear_cache()
logger.info("Shutdown complete")
except Exception as e:
logger.error(f"Error during shutdown: {e}")
def manage_model_cache(self):
try:
current = time.time()
with self.lock:
while len(self.models) > self.max_active_models:
oldest = next(iter(self.model_pool))
self.remove_model_instance(oldest)
logger.info(f"Removed LRU model: {oldest}")
for spec, last in list(self.model_pool.items()):
m = self._performance_metrics.get(spec, {})
if m.get("last_accessed", 0) and (current - m["last_accessed"] > self.model_idle_threshold):
self.remove_model_instance(spec)
logger.info(f"Removed idle model: {spec}")
sorted_models = sorted(self.model_pool.items(), key=lambda x: self._performance_metrics.get(x[0], {}).get("last_accessed", 0), reverse=True)
self.model_pool = OrderedDict(sorted_models)
except Exception as e:
logger.error(f"Error in cache management: {e}")
def set_tokenizer(self, tokenizer):
self.tokenizer = tokenizer
with self.lock:
for name, model in self.models.items():
if hasattr(model, "set_tokenizer"):
try:
model.tokenizer = tokenizer
logger.debug(f"Updated tokenizer for {name}")
except Exception as ex:
logger.warning(f"Failed to set tokenizer for {name}: {ex}")
logger.info("Tokenizer updated for models")
return self
def initialize_models(self):
try:
logger.info("Initializing models from weights")
prompt_analyzer = registry.get("prompt_analyzer")
if not prompt_analyzer:
try:
from pathlib import Path
model_list_path = Path(__file__).parent / "model_List.py"
if model_list_path.exists():
spec = importlib.util.find_spec("model_List")
if spec:
model_list = importlib.util.module_from_spec(spec)
spec.loader.exec_module(model_list)
if hasattr(model_list, "PromptAnalyzer"):
prompt_analyzer = model_list.PromptAnalyzer()
registry.register("prompt_analyzer", prompt_analyzer)
logger.info("Imported PromptAnalyzer")
except Exception as e:
logger.error(f"Error importing PromptAnalyzer: {e}")
self.prompt_analyzer = prompt_analyzer
selected_models_list = prompt_analyzer.get_selected_models() if prompt_analyzer and hasattr(prompt_analyzer, "get_selected_models") else ["model_Custm.py"]
logger.info(f"Selected model types: {selected_models_list}")
# Use specializations from class property rather than hardcoding
selected_specializations = self.specializations[:5] # Only load the first 5
for spec in selected_specializations:
try:
model_name = selected_models_list[0].replace(".py", "")
from pathlib import Path
model_path = Path(__file__).parent / f"{model_name}.py"
if model_path.exists():
spec_obj = importlib.util.find_spec(model_name)
if spec_obj:
model_module = importlib.util.module_from_spec(spec_obj)
spec_obj.loader.exec_module(model_module)
if hasattr(model_module, "Wildnerve_tlm01"):
model_class = getattr(model_module, "Wildnerve_tlm01")
embedding_dim = 768
num_heads = 12 if embedding_dim % 12 == 0 else 1
model_instance = model_class(
vocab_size=50257, # GPT-2 vocab size
specialization=spec,
dataset_path=None,
model_name="gpt2", # Changed from bert-base-uncased
embedding_dim=embedding_dim,
num_heads=num_heads,
hidden_dim=768,
num_layers=2,
output_size=50257, # Match GPT-2 vocab
dropout=0.1,
max_seq_length=128,
pooling_mode="last" # GPT-2 uses last token
)
self.models[spec] = model_instance
logger.info(f"Created model for {spec}")
except Exception as e:
logger.error(f"Error creating model for {spec}: {e}")
if not self.models:
logger.error("No models created")
return False
try:
import os
attention_config_path = os.path.join(app_config.DATA_DIR, "attention_configuration.json")
from utils.attention_connector import get_attention_connector
attention_connector = get_attention_connector()
if hasattr(attention_connector, "config_path"):
attention_connector.config_path = attention_config_path
attention_connector._init_profile_selector()
logger.info(f"Initialized attention connector with config: {attention_config_path}")
except Exception as e:
logger.warning(f"Failed to initialize attention connector: {e}")
logger.info(f"Successfully initialized {len(self.models)} models")
return True
except Exception as e:
logger.error(f"Error initializing models: {e}", exc_info=True)
return False
def get_alternative_model_for_prompt(self, prompt: str, current_model=None) -> any:
try:
if self.prompt_analyzer and hasattr(self.prompt_analyzer, "choose_model"):
model_type = self.prompt_analyzer.choose_model(prompt)
if model_type:
# Creates an instance of the model chosen by prompt_analyzer
from model_Custm import Wildnerve_tlm01
alt_model = Wildnerve_tlm01(
vocab_size=50257, # GPT-2 vocab size
specialization="general",
dataset_path=None,
model_name="gpt2", # Changed from bert-base-uncased
embedding_dim=768,
num_heads=12,
hidden_dim=768,
num_layers=6,
output_size=50257, # Match GPT-2 vocab
dropout=0.1,
max_seq_length=512,
pooling_mode="last", # GPT-2 uses last token
tokenizer=self.tokenizer
)
if alt_model != current_model:
logger.info("Found alternative model via prompt_analyzer")
return alt_model
for name, model in self.get_available_models().items():
if model != current_model:
logger.info(f"Using alternative model: {name}")
return model
try:
from model_Custm import Wildnerve_tlm01
fallback_model = Wildnerve_tlm01(
vocab_size=50257, # GPT-2 vocab size
specialization="general",
model_name="gpt2", # Changed from bert-base-uncased
embedding_dim=768,
num_heads=12,
hidden_dim=768,
num_layers=6,
output_size=50257, # Match GPT-2 vocab
dropout=0.1,
max_seq_length=512,
pooling_mode="last", # GPT-2 uses last token
tokenizer=self.tokenizer
)
logger.info("Created fallback model")
return fallback_model
except Exception as e:
logger.error(f"Error creating fallback model: {e}")
return None
except Exception as e:
logger.error(f"Error getting alternative model: {e}")
return None
def prepare_model_input(self, text: str, model) -> dict:
device = next(model.parameters()).device
try:
tokenizer = getattr(model, "tokenizer", None)
if tokenizer:
inputs = tokenizer(
text,
return_tensors="pt",
padding=True,
truncation=True,
max_length=safe_get_config_value(app_config, "MAX_SEQ_LENGTH", 512)
)
input_ids = inputs["input_ids"].to(device)
return {"input_ids": input_ids, "max_length": safe_get_config_value(app_config, "MAX_SEQ_LENGTH", 512), "device": device, "temperature": getattr(self, "generation_config", {}).get("temperature", 0.7)}
else:
logger.warning("No tokenizer in model; using basic input")
return {"input_text": text, "max_length": safe_get_config_value(app_config, "MAX_SEQ_LENGTH", 512)}
except Exception as e:
logger.error(f"Error preparing model input: {e}")
return {"input_text": text}
def process_with_context(self, input_text: str, context: Optional[dict] = None) -> dict:
conversation_context = self.get_conversation_context(window_size=3)
contextualized_prompt = input_text
if (conversation_context):
max_seq_length = safe_get_config_value(app_config, "MAX_SEQ_LENGTH", 512)
max_seq_length = int(max_seq_length) if isinstance(max_seq_length, (int, float)) else 512
contextualized_prompt = f"Previous conversation:\n{conversation_context}\n\nCurrent question: {input_text}"
result = self.process_input(contextualized_prompt, context)
if isinstance(result, dict):
result["original_query"] = input_text
return result
def get_conversation_context(self, window_size: int = 3) -> str:
if not hasattr(self, "conversation_history"):
self.conversation_history = []
recent = self.conversation_history[-window_size*2:]
lines = []
for entry in recent:
prefix = "User: " if entry["role"]=="user" else "Assistant: "
lines.append(f"{prefix}{entry['content']}")
return "\n".join(lines)
# Factory methods for model manager creation
def create_model_manager(tokenizer=None) -> ModelManager:
try:
max_active_models = safe_get_config_value(app_config, "MAX_ACTIVE_MODELS", 2)
model_idle_threshold = safe_get_config_value(app_config, "MODEL_IDLE_THRESHOLD", 600)
manager = ModelManager(tokenizer=tokenizer, max_active_models=max_active_models, model_idle_threshold=model_idle_threshold)
if tokenizer:
manager.set_tokenizer(tokenizer)
elif registry.has(TOKENIZER):
manager.set_tokenizer(registry.get(TOKENIZER))
registry.register(MODEL_MANAGER, manager)
return manager
except Exception as e:
logger.error(f"Error creating ModelManager: {e}")
minimal_manager = ModelManager(tokenizer=tokenizer, max_active_models=1)
registry.register(MODEL_MANAGER, minimal_manager)
return minimal_manager
def create_model_manager_with_tokenizer(tokenizer):
try:
max_active_models = safe_get_config_value(app_config, "MAX_ACTIVE_MODELS", 2)
model_idle_threshold = safe_get_config_value(app_config, "MODEL_IDLE_THRESHOLD", 600)
manager = ModelManager(max_active_models=max_active_models, model_idle_threshold=model_idle_threshold)
manager.tokenizer = tokenizer
manager.initialize_models()
registry.register(MODEL_MANAGER, manager)
return manager
except Exception as e:
logger.error(f"Error creating ModelManager with tokenizer: {e}")
minimal_manager = ModelManager(max_active_models=1)
minimal_manager.tokenizer = tokenizer
registry.register(MODEL_MANAGER, minimal_manager)
return minimal_manager
if __name__ == "__main__":
tokenizer = registry.get(TOKENIZER)
if not tokenizer:
from utils.transformer_utils import get_tokenizer
tokenizer = get_tokenizer("bert-base-uncased")
registry.register(TOKENIZER, tokenizer)
model_manager = create_model_manager(tokenizer)
logger.info(f"Model Manager initialized with {len(model_manager.models)} models")
else:
model_manager = None
logger.info("ModelManager module imported; initialization deferred")
# Optional late registration - can be moved to a function to be called after imports
def register_models():
"""Register models after imports to avoid circular dependencies."""
import os
from service_registry import registry, MODEL, PRETRAINED_MODEL, TOKENIZER
from tokenizer import TokenizerWrapper
# Import here to avoid circular imports
try:
from model_Custm import Wildnerve_tlm01
from model_PrTr import Wildnerve_tlm01 as PretrainedModel
# Instantiate & register tokenizer
tok = TokenizerWrapper()
registry.register(TOKENIZER, tok, overwrite=True)
# Instantiate & register custom model
custom = Wildnerve_tlm01(tokenizer=tok)
registry.register(MODEL, custom, overwrite=True)
# Instantiate & register pretrained model
pre = PretrainedModel(tokenizer=tok)
registry.register(PRETRAINED_MODEL, pre, overwrite=True)
return True
except Exception as e:
logger.error(f"Failed to register models: {e}")
return False
|