Spaces:
Configuration error
Configuration error
File size: 7,765 Bytes
ba92f0e |
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 |
"""
vectors/embedding_validator.py - Embedding validation and security
"""
import numpy as np
from typing import Dict, List, Optional, Any, Tuple
from dataclasses import dataclass
from datetime import datetime
import hashlib
from ..core.logger import SecurityLogger
from ..core.exceptions import ValidationError
@dataclass
class EmbeddingMetadata:
"""Metadata for embeddings"""
dimension: int
model: str
timestamp: datetime
source: str
checksum: str
@dataclass
class ValidationResult:
"""Result of embedding validation"""
is_valid: bool
errors: List[str]
normalized_embedding: Optional[np.ndarray]
metadata: Dict[str, Any]
class EmbeddingValidator:
"""Validates and secures embeddings"""
def __init__(self, security_logger: Optional[SecurityLogger] = None):
self.security_logger = security_logger
self.known_models = {
"openai-ada-002": 1536,
"openai-text-embedding-ada-002": 1536,
"huggingface-bert-base": 768,
"huggingface-mpnet-base": 768
}
self.max_dimension = 2048
self.min_dimension = 64
def validate_embedding(self,
embedding: np.ndarray,
metadata: Optional[Dict[str, Any]] = None) -> ValidationResult:
"""Validate an embedding vector"""
try:
errors = []
# Check dimensions
if embedding.ndim != 1:
errors.append("Embedding must be a 1D vector")
if len(embedding) > self.max_dimension:
errors.append(f"Embedding dimension exceeds maximum {self.max_dimension}")
if len(embedding) < self.min_dimension:
errors.append(f"Embedding dimension below minimum {self.min_dimension}")
# Check for NaN or Inf values
if np.any(np.isnan(embedding)) or np.any(np.isinf(embedding)):
errors.append("Embedding contains NaN or Inf values")
# Validate against known models
if metadata and 'model' in metadata:
if metadata['model'] in self.known_models:
expected_dim = self.known_models[metadata['model']]
if len(embedding) != expected_dim:
errors.append(
f"Dimension mismatch for model {metadata['model']}: "
f"expected {expected_dim}, got {len(embedding)}"
)
# Normalize embedding
normalized = None
if not errors:
normalized = self._normalize_embedding(embedding)
# Calculate checksum
checksum = self._calculate_checksum(normalized)
# Create metadata
embedding_metadata = EmbeddingMetadata(
dimension=len(embedding),
model=metadata.get('model', 'unknown') if metadata else 'unknown',
timestamp=datetime.utcnow(),
source=metadata.get('source', 'unknown') if metadata else 'unknown',
checksum=checksum
)
result = ValidationResult(
is_valid=len(errors) == 0,
errors=errors,
normalized_embedding=normalized,
metadata=vars(embedding_metadata) if not errors else {}
)
if errors and self.security_logger:
self.security_logger.log_security_event(
"embedding_validation_failure",
errors=errors,
metadata=metadata
)
return result
except Exception as e:
if self.security_logger:
self.security_logger.log_security_event(
"embedding_validation_error",
error=str(e)
)
raise ValidationError(f"Embedding validation failed: {str(e)}")
def _normalize_embedding(self, embedding: np.ndarray) -> np.ndarray:
"""Normalize embedding vector"""
norm = np.linalg.norm(embedding)
if norm > 0:
return embedding / norm
return embedding
def _calculate_checksum(self, embedding: np.ndarray) -> str:
"""Calculate checksum for embedding"""
return hashlib.sha256(embedding.tobytes()).hexdigest()
def check_similarity(self,
embedding1: np.ndarray,
embedding2: np.ndarray) -> float:
"""Check similarity between two embeddings"""
try:
# Validate both embeddings
result1 = self.validate_embedding(embedding1)
result2 = self.validate_embedding(embedding2)
if not result1.is_valid or not result2.is_valid:
raise ValidationError("Invalid embeddings for similarity check")
# Calculate cosine similarity
return float(np.dot(
result1.normalized_embedding,
result2.normalized_embedding
))
except Exception as e:
if self.security_logger:
self.security_logger.log_security_event(
"similarity_check_error",
error=str(e)
)
raise ValidationError(f"Similarity check failed: {str(e)}")
def detect_anomalies(self,
embeddings: List[np.ndarray],
threshold: float = 0.8) -> List[int]:
"""Detect anomalous embeddings in a set"""
try:
anomalies = []
# Validate all embeddings
valid_embeddings = []
for i, emb in enumerate(embeddings):
result = self.validate_embedding(emb)
if result.is_valid:
valid_embeddings.append(result.normalized_embedding)
else:
anomalies.append(i)
if not valid_embeddings:
return list(range(len(embeddings)))
# Calculate mean embedding
mean_embedding = np.mean(valid_embeddings, axis=0)
mean_embedding = self._normalize_embedding(mean_embedding)
# Check similarities
for i, emb in enumerate(valid_embeddings):
similarity = float(np.dot(emb, mean_embedding))
if similarity < threshold:
anomalies.append(i)
if anomalies and self.security_logger:
self.security_logger.log_security_event(
"anomalous_embeddings_detected",
count=len(anomalies),
total_embeddings=len(embeddings)
)
return anomalies
except Exception as e:
if self.security_logger:
self.security_logger.log_security_event(
"anomaly_detection_error",
error=str(e)
)
raise ValidationError(f"Anomaly detection failed: {str(e)}")
def add_known_model(self, model_name: str, dimension: int):
"""Add a known model to the validator"""
self.known_models[model_name] = dimension
def verify_metadata(self, metadata: Dict[str, Any]) -> bool:
"""Verify embedding metadata"""
required_fields = {'model', 'dimension', 'timestamp'}
return all(field in metadata for field in required_fields) |