Spaces:
Paused
Paused
File size: 11,584 Bytes
88bdcff f3ebc82 78caafb 88bdcff 78caafb 88bdcff f3ebc82 88bdcff 5f0db1e 88bdcff 5f0db1e 333c083 88bdcff 5f0db1e 88bdcff 5f0db1e 88bdcff 5f0db1e 88bdcff 5f0db1e 88bdcff 333c083 5f0db1e 88bdcff 5f0db1e 88bdcff 5f0db1e 88bdcff 78caafb 88bdcff 78caafb 88bdcff f3ebc82 88bdcff f3ebc82 88bdcff f3ebc82 88bdcff f3ebc82 88bdcff |
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 |
"""FDAM retriever with priority weighting and reranking.
Implements tiered retrieval:
1. Vector similarity search
2. Priority weighting (primary > reference-threshold > reference-narrative)
3. Optional reranking for production
"""
import logging
import time
from typing import Optional, TYPE_CHECKING
from dataclasses import dataclass
from config.settings import settings
# Type hints only - actual import deferred to __init__
if TYPE_CHECKING:
from .vectorstore import ChromaVectorStore
logger = logging.getLogger(__name__)
@dataclass
class RetrievalResult:
"""A single retrieval result with relevance score."""
chunk_id: str
text: str
source: str
category: str
section: str
priority: str
content_type: str
keywords: list[str]
similarity_score: float # 0-1, higher is better
weighted_score: float # After priority weighting
final_score: float # After reranking (if applied)
def to_dict(self) -> dict:
"""Convert to dictionary."""
return {
"chunk_id": self.chunk_id,
"text": self.text,
"source": self.source,
"category": self.category,
"section": self.section,
"priority": self.priority,
"content_type": self.content_type,
"keywords": self.keywords,
"similarity_score": self.similarity_score,
"weighted_score": self.weighted_score,
"final_score": self.final_score,
}
class MockReranker:
"""Mock reranker for local development.
Simply returns scores based on keyword overlap.
"""
def rerank(
self,
query: str,
documents: list[str],
) -> list[float]:
"""Score documents based on keyword overlap with query.
Args:
query: Query text
documents: List of document texts
Returns:
List of scores (0-1) for each document
"""
query_words = set(query.lower().split())
scores = []
for doc in documents:
doc_words = set(doc.lower().split())
# Jaccard-like overlap score
overlap = len(query_words & doc_words)
total = len(query_words | doc_words)
score = overlap / total if total > 0 else 0.0
scores.append(score)
return scores
class SharedReranker:
"""Reranker that uses the shared model from RealModelStack.
This avoids loading a duplicate reranker model - instead uses the
model already loaded by the pipeline at startup.
"""
def rerank(
self,
query: str,
documents: list[str],
) -> list[float]:
"""Score documents using the shared reranker model.
Args:
query: Query text
documents: List of document texts
Returns:
List of scores (0-1) for each document
"""
from models.loader import get_models
model_stack = get_models()
# Use the shared reranker model (always loaded at startup)
return model_stack.reranker.rerank(query, documents)
def get_reranker():
"""Get appropriate reranker based on settings.
For real models, uses SharedReranker which wraps the
model stack's reranker model (no duplicate loading).
"""
if settings.mock_models:
return MockReranker()
return SharedReranker()
class FDAMRetriever:
"""FDAM-specific retriever with priority weighting.
Priority weights:
- primary: 1.0 (FDAM methodology)
- reference-threshold: 0.9 (Threshold tables)
- reference-narrative: 0.8 (Supporting documentation)
"""
PRIORITY_WEIGHTS = {
"primary": 1.0,
"reference-threshold": 0.9,
"reference-narrative": 0.8,
}
def __init__(
self,
vectorstore: Optional["ChromaVectorStore"] = None,
reranker=None,
use_reranking: bool = True,
):
"""Initialize retriever.
Args:
vectorstore: ChromaDB vector store instance.
If None, creates default instance.
reranker: Reranker instance. If None, uses appropriate default.
use_reranking: Whether to apply reranking step.
"""
if vectorstore is None:
# Lazy import to avoid chromadb dependency at module load
from .vectorstore import ChromaVectorStore
vectorstore = ChromaVectorStore()
self.vectorstore = vectorstore
self.reranker = reranker if reranker is not None else get_reranker()
self.use_reranking = use_reranking
def retrieve(
self,
query: str,
top_k: int = 5,
category_filter: Optional[str] = None,
priority_filter: Optional[str] = None,
include_scores: bool = True,
) -> list[RetrievalResult]:
"""Retrieve relevant chunks for a query.
Args:
query: Query text
top_k: Number of results to return
category_filter: Optional category to filter by
priority_filter: Optional priority to filter by
include_scores: Whether to include score details
Returns:
List of RetrievalResult objects, sorted by final_score descending
"""
start_time = time.time()
logger.debug(f"RAG retrieve: query='{query[:50]}...' top_k={top_k}")
# Build metadata filter
where_filter = None
if category_filter or priority_filter:
where_filter = {}
if category_filter:
where_filter["category"] = category_filter
if priority_filter:
where_filter["priority"] = priority_filter
# Fetch more results than needed for reranking
fetch_k = top_k * 3 if self.use_reranking else top_k
# Query vector store
raw_results = self.vectorstore.query(
query_text=query,
n_results=fetch_k,
where=where_filter,
)
if not raw_results:
logger.debug("RAG retrieve: no results found")
return []
# Convert to RetrievalResult objects with priority weighting
results = []
for r in raw_results:
# Convert distance to similarity (cosine distance: 0 = identical)
similarity = 1.0 - r["distance"]
# Apply priority weight
priority = r["metadata"].get("priority", "reference-narrative")
weight = self.PRIORITY_WEIGHTS.get(priority, 0.8)
weighted_score = similarity * weight
# Parse keywords
keywords_str = r["metadata"].get("keywords", "")
keywords = keywords_str.split(",") if keywords_str else []
results.append(
RetrievalResult(
chunk_id=r["id"],
text=r["document"],
source=r["metadata"].get("source", "unknown"),
category=r["metadata"].get("category", "unknown"),
section=r["metadata"].get("section", "unknown"),
priority=priority,
content_type=r["metadata"].get("content_type", "narrative"),
keywords=keywords,
similarity_score=similarity,
weighted_score=weighted_score,
final_score=weighted_score, # Will be updated by reranking
)
)
# Apply reranking if enabled
if self.use_reranking and results:
logger.debug(f"Applying reranking to {len(results)} results")
documents = [r.text for r in results]
rerank_scores = self.reranker.rerank(query, documents)
# Combine weighted score with rerank score
# Final = 0.6 * weighted + 0.4 * rerank
for i, result in enumerate(results):
rerank_score = rerank_scores[i]
result.final_score = 0.6 * result.weighted_score + 0.4 * rerank_score
# Sort by final score (descending) and take top_k
results.sort(key=lambda x: x.final_score, reverse=True)
final_results = results[:top_k]
# Log retrieval summary
elapsed = time.time() - start_time
if final_results:
top_score = final_results[0].final_score
top_source = final_results[0].source
logger.debug(f"RAG retrieve: {len(final_results)} results in {elapsed:.3f}s, "
f"top_score={top_score:.3f}, top_source={top_source}")
else:
logger.debug(f"RAG retrieve: 0 results in {elapsed:.3f}s")
return final_results
def retrieve_for_context(
self,
query: str,
top_k: int = 5,
) -> str:
"""Retrieve and format chunks as context string for LLM.
Args:
query: Query text
top_k: Number of chunks to include
Returns:
Formatted context string with source citations
"""
results = self.retrieve(query, top_k=top_k)
if not results:
return "No relevant context found."
context_parts = []
for i, r in enumerate(results, 1):
context_parts.append(
f"[{i}] Source: {r.source} | Section: {r.section}\n{r.text}"
)
return "\n\n---\n\n".join(context_parts)
def retrieve_thresholds(
self,
material_type: str,
facility_type: str,
) -> list[RetrievalResult]:
"""Retrieve threshold values for a specific material and facility type.
Convenience method for threshold lookups.
Args:
material_type: Type of material (e.g., "lead", "soot", "char")
facility_type: Facility classification
Returns:
Relevant threshold results
"""
query = f"{material_type} threshold {facility_type} clearance criteria"
return self.retrieve(
query=query,
top_k=3,
category_filter="thresholds",
)
def retrieve_disposition(
self,
zone: str,
condition: str,
material_type: Optional[str] = None,
) -> list[RetrievalResult]:
"""Retrieve disposition guidance for zone/condition combination.
Convenience method for disposition lookups.
Args:
zone: Zone classification (burn-zone, near-field, far-field)
condition: Condition level (background, light, moderate, heavy, structural-damage)
material_type: Optional material type for specific guidance
Returns:
Relevant disposition results
"""
query = f"disposition {zone} {condition}"
if material_type:
query += f" {material_type}"
query += " cleaning recommendation"
return self.retrieve(
query=query,
top_k=5,
priority_filter="primary", # Prefer FDAM methodology
)
def retrieve_cleaning_method(
self,
surface_type: str,
condition: str,
) -> list[RetrievalResult]:
"""Retrieve cleaning method recommendations.
Args:
surface_type: Type of surface (e.g., "drywall", "concrete", "metal")
condition: Condition level
Returns:
Relevant cleaning method results
"""
query = f"cleaning method {surface_type} {condition} procedure hepa"
return self.retrieve(
query=query,
top_k=5,
)
|