Spaces:
Sleeping
Sleeping
File size: 22,235 Bytes
1d10b0a d3be8f6 1d10b0a d3be8f6 1d10b0a d3be8f6 1d10b0a d3be8f6 1d10b0a d3be8f6 1d10b0a d3be8f6 1d10b0a |
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 |
"""Embedding models for document vectorization."""
from typing import List, Optional
import torch
from sentence_transformers import SentenceTransformer
from transformers import AutoTokenizer, AutoModel
import numpy as np
from tqdm import tqdm
import os
class EmbeddingModel:
"""Base class for embedding models."""
def __init__(self, model_name: str, device: Optional[str] = None):
"""Initialize embedding model.
Args:
model_name: Name/path of the model
device: Device to run model on (cuda/cpu)
"""
self.model_name = model_name
self.device = device or ("cuda" if torch.cuda.is_available() else "cpu")
self.model = None
self.tokenizer = None
def load_model(self):
"""Load the embedding model."""
raise NotImplementedError
def embed_documents(self, texts: List[str], batch_size: int = 32) -> np.ndarray:
"""Embed a list of documents.
Args:
texts: List of texts to embed
batch_size: Batch size for processing
Returns:
Numpy array of embeddings
"""
raise NotImplementedError
def embed_query(self, query: str) -> np.ndarray:
"""Embed a single query.
Args:
query: Query text
Returns:
Numpy array of embedding
"""
return self.embed_documents([query])[0]
class SentenceTransformerEmbedding(EmbeddingModel):
"""Sentence Transformer based embedding model."""
def load_model(self):
"""Load sentence transformer model."""
print(f"Loading SentenceTransformer model: {self.model_name}")
try:
self.model = SentenceTransformer(self.model_name, device=self.device)
print(f"Model loaded successfully on {self.device}")
except Exception as e:
print(f"Error loading model {self.model_name}: {str(e)}")
print("Falling back to default model...")
self.model = SentenceTransformer('all-MiniLM-L6-v2', device=self.device)
def embed_documents(self, texts: List[str], batch_size: int = 32) -> np.ndarray:
"""Embed documents using sentence transformer."""
if self.model is None:
self.load_model()
embeddings = []
for i in tqdm(range(0, len(texts), batch_size), desc="Embedding documents"):
batch = texts[i:i + batch_size]
batch_embeddings = self.model.encode(
batch,
convert_to_numpy=True,
show_progress_bar=False,
batch_size=batch_size
)
embeddings.append(batch_embeddings)
return np.vstack(embeddings) if embeddings else np.array([])
class BioMedicalEmbedding(EmbeddingModel):
"""Bio-medical BERT based embedding model."""
def load_model(self):
"""Load bio-medical BERT model."""
print(f"Loading Bio-Medical model: {self.model_name}")
try:
self.tokenizer = AutoTokenizer.from_pretrained(self.model_name)
self.model = AutoModel.from_pretrained(self.model_name).to(self.device)
self.model.eval()
print(f"Model loaded successfully on {self.device}")
except Exception as e:
print(f"Error loading model {self.model_name}: {str(e)}")
print("Falling back to default model...")
self.tokenizer = AutoTokenizer.from_pretrained('bert-base-uncased')
self.model = AutoModel.from_pretrained('bert-base-uncased').to(self.device)
self.model.eval()
def mean_pooling(self, model_output, attention_mask):
"""Apply mean pooling to get sentence embeddings."""
token_embeddings = model_output[0]
input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(
input_mask_expanded.sum(1), min=1e-9
)
def embed_documents(self, texts: List[str], batch_size: int = 32) -> np.ndarray:
"""Embed documents using bio-medical BERT."""
if self.model is None:
self.load_model()
embeddings = []
with torch.no_grad():
for i in tqdm(range(0, len(texts), batch_size), desc="Embedding documents"):
batch = texts[i:i + batch_size]
# Tokenize
encoded_input = self.tokenizer(
batch,
padding=True,
truncation=True,
max_length=512,
return_tensors='pt'
).to(self.device)
# Get embeddings
model_output = self.model(**encoded_input)
# Apply mean pooling
batch_embeddings = self.mean_pooling(
model_output,
encoded_input['attention_mask']
)
# Normalize
batch_embeddings = torch.nn.functional.normalize(batch_embeddings, p=2, dim=1)
embeddings.append(batch_embeddings.cpu().numpy())
return np.vstack(embeddings) if embeddings else np.array([])
class GeminiEmbedding(EmbeddingModel):
"""Gemini embedding model using Google AI API."""
def load_model(self):
"""Load Gemini embedding model."""
print(f"Initializing Gemini embedding model: {self.model_name}")
try:
import google.generativeai as genai
api_key = os.getenv("GEMINI_API_KEY")
if not api_key:
raise ValueError("GEMINI_API_KEY environment variable not set")
genai.configure(api_key=api_key)
self.model = genai
print(f"Gemini model initialized successfully")
except Exception as e:
print(f"Error loading Gemini model: {str(e)}")
print("Falling back to default model...")
self.model = SentenceTransformer('all-MiniLM-L6-v2', device=self.device)
def embed_documents(self, texts: List[str], batch_size: int = 32) -> np.ndarray:
"""Embed documents using Gemini API."""
if self.model is None:
self.load_model()
embeddings = []
# Gemini API has rate limits, process with delays
for i in tqdm(range(0, len(texts), batch_size), desc="Embedding documents"):
batch = texts[i:i + batch_size]
for text in batch:
try:
if hasattr(self.model, 'embed_content'):
result = self.model.embed_content(
model="models/embedding-001",
content=text,
task_type="retrieval_document"
)
embeddings.append(result['embedding'])
else:
# Fallback if Gemini not available
from sentence_transformers import SentenceTransformer
fallback_model = SentenceTransformer('all-MiniLM-L6-v2')
emb = fallback_model.encode([text])[0]
embeddings.append(emb)
except Exception as e:
print(f"Error embedding text: {str(e)}")
# Use zero vector as fallback
embeddings.append(np.zeros(768))
return np.array(embeddings)
class FinancialEmbedding(EmbeddingModel):
"""Financial domain BERT based embedding model."""
def load_model(self):
"""Load financial BERT model."""
print(f"Loading Financial domain model: {self.model_name}")
try:
self.tokenizer = AutoTokenizer.from_pretrained(self.model_name)
self.model = AutoModel.from_pretrained(self.model_name).to(self.device)
self.model.eval()
print(f"Model loaded successfully on {self.device}")
except Exception as e:
print(f"Error loading model {self.model_name}: {str(e)}")
print("Falling back to default model...")
self.tokenizer = AutoTokenizer.from_pretrained('bert-base-uncased')
self.model = AutoModel.from_pretrained('bert-base-uncased').to(self.device)
self.model.eval()
def mean_pooling(self, model_output, attention_mask):
"""Apply mean pooling to get sentence embeddings."""
token_embeddings = model_output[0]
input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(
input_mask_expanded.sum(1), min=1e-9
)
def embed_documents(self, texts: List[str], batch_size: int = 32) -> np.ndarray:
"""Embed documents using financial BERT."""
if self.model is None:
self.load_model()
embeddings = []
with torch.no_grad():
for i in tqdm(range(0, len(texts), batch_size), desc="Embedding financial documents"):
batch = texts[i:i + batch_size]
# Tokenize
encoded_input = self.tokenizer(
batch,
padding=True,
truncation=True,
max_length=512,
return_tensors='pt'
).to(self.device)
# Get embeddings
model_output = self.model(**encoded_input)
# Apply mean pooling
batch_embeddings = self.mean_pooling(
model_output,
encoded_input['attention_mask']
)
# Normalize
batch_embeddings = torch.nn.functional.normalize(batch_embeddings, p=2, dim=1)
embeddings.append(batch_embeddings.cpu().numpy())
return np.vstack(embeddings) if embeddings else np.array([])
class LawEmbedding(EmbeddingModel):
"""Legal domain BERT based embedding model."""
def load_model(self):
"""Load legal BERT model."""
print(f"Loading Legal domain model: {self.model_name}")
try:
self.tokenizer = AutoTokenizer.from_pretrained(self.model_name)
self.model = AutoModel.from_pretrained(self.model_name).to(self.device)
self.model.eval()
print(f"Model loaded successfully on {self.device}")
except Exception as e:
print(f"Error loading model {self.model_name}: {str(e)}")
print("Falling back to default model...")
self.tokenizer = AutoTokenizer.from_pretrained('bert-base-uncased')
self.model = AutoModel.from_pretrained('bert-base-uncased').to(self.device)
self.model.eval()
def mean_pooling(self, model_output, attention_mask):
"""Apply mean pooling to get sentence embeddings."""
token_embeddings = model_output[0]
input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(
input_mask_expanded.sum(1), min=1e-9
)
def embed_documents(self, texts: List[str], batch_size: int = 32) -> np.ndarray:
"""Embed documents using legal BERT."""
if self.model is None:
self.load_model()
embeddings = []
with torch.no_grad():
for i in tqdm(range(0, len(texts), batch_size), desc="Embedding legal documents"):
batch = texts[i:i + batch_size]
# Tokenize
encoded_input = self.tokenizer(
batch,
padding=True,
truncation=True,
max_length=512,
return_tensors='pt'
).to(self.device)
# Get embeddings
model_output = self.model(**encoded_input)
# Apply mean pooling
batch_embeddings = self.mean_pooling(
model_output,
encoded_input['attention_mask']
)
# Normalize
batch_embeddings = torch.nn.functional.normalize(batch_embeddings, p=2, dim=1)
embeddings.append(batch_embeddings.cpu().numpy())
return np.vstack(embeddings) if embeddings else np.array([])
class CustomerServiceEmbedding(EmbeddingModel):
"""Customer service domain specialized embedding model."""
def load_model(self):
"""Load customer service domain model."""
print(f"Loading Customer Service domain model: {self.model_name}")
try:
self.model = SentenceTransformer(self.model_name, device=self.device)
print(f"Model loaded successfully on {self.device}")
except Exception as e:
print(f"Error loading model {self.model_name}: {str(e)}")
print("Falling back to default model...")
self.model = SentenceTransformer('all-MiniLM-L6-v2', device=self.device)
def embed_documents(self, texts: List[str], batch_size: int = 32) -> np.ndarray:
"""Embed documents using customer service model."""
if self.model is None:
self.load_model()
embeddings = []
for i in tqdm(range(0, len(texts), batch_size), desc="Embedding customer service documents"):
batch = texts[i:i + batch_size]
batch_embeddings = self.model.encode(
batch,
convert_to_numpy=True,
show_progress_bar=False,
batch_size=batch_size
)
embeddings.append(batch_embeddings)
return np.vstack(embeddings) if embeddings else np.array([])
class EmbeddingFactory:
"""Factory for creating embedding model instances."""
# Map model names to their types
MODEL_TYPES = {
"sentence-transformers/all-mpnet-base-v2": "sentence-transformer", # Stable, well-supported
"emilyalsentzer/Bio_ClinicalBERT": "biomedical", # Clinical domain
"microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract": "biomedical", # Medical domain
"sentence-transformers/all-MiniLM-L6-v2": "sentence-transformer", # Fast, lightweight
"sentence-transformers/multilingual-MiniLM-L12-v2": "sentence-transformer", # Multilingual
"sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2": "sentence-transformer", # Paraphrase
"allenai/specter": "biomedical", # Academic paper embeddings
"ProsusAI/finbert": "financial", # Financial domain BERT
"gemini-embedding-001": "gemini", # Gemini API
"nlpaueb/legal-bert-base-uncased": "law", # Legal domain BERT
"sentence-transformers/all-mpnet-base-v2-legal": "law", # Legal domain specialized
"sentence-transformers/paraphrase-mpnet-base-v2-customer-service": "customer-service", # Customer service
"sentence-transformers/all-MiniLM-L6-v2-customer-service": "customer-service" # Customer service lightweight
}
@classmethod
def create_embedding_model(cls, model_name: str, device: Optional[str] = None) -> EmbeddingModel:
"""Create an embedding model instance.
Args:
model_name: Name of the embedding model
device: Device to run model on
Returns:
EmbeddingModel instance
"""
model_type = cls.MODEL_TYPES.get(model_name, "sentence-transformer")
if model_type == "gemini":
return GeminiEmbedding(model_name, device)
elif model_type == "biomedical":
return BioMedicalEmbedding(model_name, device)
elif model_type == "financial":
return FinancialEmbedding(model_name, device)
elif model_type == "law":
return LawEmbedding(model_name, device)
elif model_type == "customer-service":
return CustomerServiceEmbedding(model_name, device)
else:
return SentenceTransformerEmbedding(model_name, device)
@classmethod
def get_available_models(cls) -> List[str]:
"""Get list of available embedding models."""
return list(cls.MODEL_TYPES.keys())
@classmethod
def get_model_info(cls, model_name: str) -> dict:
"""Get information about a specific model.
Args:
model_name: Name of the model
Returns:
Dictionary with model information
"""
info = {
"sentence-transformers/all-mpnet-base-v2": {
"description": "High-quality, general-purpose sentence embeddings (384d)",
"dimension": 768,
"type": "sentence-transformer",
"note": "Recommended for general use"
},
"emilyalsentzer/Bio_ClinicalBERT": {
"description": "Clinical BERT for biomedical and clinical text",
"dimension": 768,
"type": "biomedical"
},
"microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract": {
"description": "PubMedBERT for biomedical and medical text",
"dimension": 768,
"type": "biomedical"
},
"sentence-transformers/all-MiniLM-L6-v2": {
"description": "Fast, lightweight sentence embeddings",
"dimension": 384,
"type": "sentence-transformer",
"note": "Good for speed-sensitive applications"
},
"sentence-transformers/multilingual-MiniLM-L12-v2": {
"description": "Fast multilingual sentence embeddings",
"dimension": 384,
"type": "sentence-transformer",
"note": "Supports 50+ languages"
},
"sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2": {
"description": "Multilingual paraphrase embeddings",
"dimension": 384,
"type": "sentence-transformer",
"note": "Good for paraphrase detection"
},
"allenai/specter": {
"description": "Embeddings for academic papers and citations",
"dimension": 768,
"type": "biomedical",
"note": "Optimized for scientific literature"
},
"ProsusAI/finbert": {
"description": "BERT model fine-tuned for financial domain NLP tasks",
"dimension": 768,
"type": "financial",
"note": "Optimized for financial documents, reports, and SEC filings"
},
"gemini-embedding-001": {
"description": "Google Gemini embedding model via API",
"dimension": 768,
"type": "gemini",
"url": "https://ai.google.dev/gemini-api/docs/embeddings",
"note": "Requires GEMINI_API_KEY environment variable"
},
"nlpaueb/legal-bert-base-uncased": {
"description": "Legal BERT pre-trained on a large corpus of legal documents",
"dimension": 768,
"type": "law",
"note": "Optimized for contracts, statutes, and legal documents"
},
"sentence-transformers/all-mpnet-base-v2-legal": {
"description": "Sentence Transformer fine-tuned for legal domain",
"dimension": 768,
"type": "law",
"note": "High-quality embeddings for legal text similarity and retrieval"
},
"sentence-transformers/paraphrase-mpnet-base-v2-customer-service": {
"description": "Specialized embeddings for customer service queries and responses",
"dimension": 768,
"type": "customer-service",
"note": "Optimized for FAQs, support tickets, and customer interactions"
},
"sentence-transformers/all-MiniLM-L6-v2-customer-service": {
"description": "Lightweight customer service embeddings",
"dimension": 384,
"type": "customer-service",
"note": "Fast and efficient for real-time customer service applications"
}
}
return info.get(model_name, {"description": "Unknown model", "dimension": 768})
@classmethod
def get_embedding_dimension(cls, model_name: str) -> int:
"""Get embedding dimension for a model.
Args:
model_name: Name of the model
Returns:
Embedding dimension
"""
# Default dimensions (adjust based on actual models)
dimensions = {
"sentence-transformers/all-mpnet-base-v2": 768,
"emilyalsentzer/Bio_ClinicalBERT": 768,
"microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract": 768,
"sentence-transformers/all-MiniLM-L6-v2": 384,
"sentence-transformers/multilingual-MiniLM-L12-v2": 384,
"sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2": 384,
"allenai/specter": 768,
"ProsusAI/finbert": 768,
"gemini-embedding-001": 768,
"nlpaueb/legal-bert-base-uncased": 768,
"sentence-transformers/all-mpnet-base-v2-legal": 768,
"sentence-transformers/paraphrase-mpnet-base-v2-customer-service": 768,
"sentence-transformers/all-MiniLM-L6-v2-customer-service": 384
}
return dimensions.get(model_name, 768)
|