Spaces:
Running on Zero
Running on Zero
File size: 22,237 Bytes
0828c2c 6c64324 0828c2c 6c64324 0828c2c 6c64324 0828c2c | 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 | """RAG pipeline for retrieval-augmented generation."""
import logging
from collections.abc import Iterator
from typing import Any
from langchain_core.documents import Document
from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.retrievers import BaseRetriever
from .config_loader import get_config
from .llm_handler import get_llm_handler
from .main_document_loader import get_main_document_loader
from .response_enhancer import get_response_enhancer
# Import retrieval components
from .retrieval import RetrieverFactory
from .retrieval.strategies import BM25Strategy, BM25VectorStrategy, VectorStrategy # noqa: F401
logger = logging.getLogger(__name__)
class RAGPipeline:
"""RAG pipeline for question answering."""
def __init__(self, llm_handler=None, retrieval_strategy: str | None = None):
"""Initialize RAG pipeline.
Args:
llm_handler: Optional LLMHandler instance for custom provider/model
retrieval_strategy: Override retrieval strategy (uses config if None)
"""
self.config = get_config()
self.llm_handler = llm_handler or get_llm_handler()
self.response_enhancer = get_response_enhancer()
self.main_doc_loader = get_main_document_loader()
# Get LLM
self.llm = self.llm_handler.get_llm()
# Initialize retrieval strategy
self.retrieval_strategy_name = retrieval_strategy or self.config.get(
"retrieval.strategy", "vector"
)
self.retrieval_strategy = self._create_retrieval_strategy()
# Get retriever from strategy
self.retriever = self._get_retriever()
logger.info(f"Initialized RAG pipeline with '{self.retrieval_strategy_name}' strategy")
# Load main document if enabled
self.main_doc_content = ""
if self.config.get("main_document.enabled", False):
self.main_doc_content = self.main_doc_loader.load_main_document()
if self.main_doc_content:
token_count = self.main_doc_loader.count_tokens(self.main_doc_content)
logger.info(f"Main document loaded: {token_count} tokens available in context")
# Calculate and log token budget if main doc is loaded
if self.main_doc_content:
budget = self._calculate_context_budget()
logger.info(
f"Token budget - Main doc: {budget['main_doc_tokens']}, "
f"Available for retrieval: {budget['available_for_retrieval']}"
)
# Setup prompt template
self.prompt_template = self._create_prompt_template()
# Create QA chain using LCEL
self.qa_chain = self._create_qa_chain()
def _create_prompt_template(self) -> ChatPromptTemplate:
"""Create prompt template for RAG with main document and chat history support.
Returns:
ChatPromptTemplate instance
"""
system_prompt = self.llm_handler.get_system_prompt()
# Structure: System Prompt → Chat History → Main Doc → VectorDB Context → Question
template = f"""{system_prompt}
{{chat_history}}
{{main_document_section}}
{{context}}
Question: {{question}}
Answer: """
return ChatPromptTemplate.from_template(template)
def _format_chat_history(self, chat_history: list[dict] | None) -> str:
"""Format chat history for inclusion in prompt.
Args:
chat_history: List of message dicts with 'role' and 'content' keys
Returns:
Formatted chat history string
"""
if not chat_history:
return ""
formatted_lines = []
for msg in chat_history:
role = msg.get("role", "")
content = msg.get("content", "")
if role == "user":
formatted_lines.append(f"User: {content}")
elif role == "assistant":
formatted_lines.append(f"Assistant: {content}")
if formatted_lines:
return "=== PREVIOUS CONVERSATION ===\n" + "\n".join(formatted_lines) + "\n\n"
return ""
def _format_docs(self, docs: list[Document]) -> str:
"""Format documents into a single string.
Args:
docs: List of documents
Returns:
Formatted string
"""
if not docs:
return ""
return "\n\n".join(doc.page_content for doc in docs)
def _format_main_doc(self) -> str:
"""Format main document section for prompt.
Returns:
Formatted main document section (empty if not available)
"""
if not self.main_doc_content:
return ""
# Clear section header for main document
return f"""
=== ESSENTIAL PROFILE INFORMATION ===
(This information is always available and takes priority)
{self.main_doc_content}
=== ADDITIONAL CONTEXT FROM DOCUMENTS ===
"""
def _create_qa_chain(self):
"""Create retrieval QA chain using LCEL with main document and chat history.
Returns:
LCEL chain
"""
# Create the RAG chain with main document positioned BEFORE retrieval context
# Chat history will be injected via the invoke/stream methods
# The chain expects a dict with 'question' and optionally 'chat_history'
chain = (
{
"main_document_section": lambda _: self._format_main_doc(),
"context": (lambda x: x.get("question", x) if isinstance(x, dict) else x)
| self.retriever
| self._format_docs,
"question": lambda x: x.get("question", x) if isinstance(x, dict) else x,
"chat_history": lambda x: self._format_chat_history(x.get("chat_history", []))
if isinstance(x, dict)
else "",
}
| self.prompt_template
| self.llm
| StrOutputParser()
)
return chain
def query(self, question: str, chat_history: list[dict] | None = None) -> dict[str, Any]:
"""Query the RAG pipeline.
Args:
question: User question
chat_history: Optional list of previous messages with 'role' and 'content' keys
Returns:
Dictionary with 'result' and optionally 'source_documents'
"""
logger.info(f"Processing query: {question}")
if chat_history:
logger.debug(f"Including {len(chat_history)} previous messages in context")
try:
# Prepare input with chat history
chain_input = {
"question": question,
"chat_history": chat_history or [],
}
# Get the answer from the chain
answer = self.qa_chain.invoke(chain_input)
# Enhance the response for better tone and professionalism
if self.config.get("rag.enhance_responses", True):
original_answer = answer
answer = self.response_enhancer.enhance_with_context(answer, question)
if answer != original_answer:
logger.debug("Response enhanced for better tone")
# Retrieve source documents separately if needed
source_documents = []
if self.config.get("rag.include_sources", True):
source_documents = self.retriever.invoke(question)
logger.debug(f"Retrieved {len(source_documents)} source documents")
return {
"result": answer,
"source_documents": source_documents,
}
except Exception as e:
logger.error(f"Error processing query: {e}")
return {
"result": "I encountered a technical issue. Please try rephrasing your question or reach out directly to discuss further.",
"source_documents": [],
}
def get_answer(self, question: str, chat_history: list[dict] | None = None) -> str:
"""Get answer to a question (simplified interface).
Args:
question: User question
chat_history: Optional list of previous messages with 'role' and 'content' keys
Returns:
Answer string
"""
response = self.query(question, chat_history=chat_history)
return response.get("result", "I couldn't generate an answer.")
def get_answer_with_sources(
self, question: str, chat_history: list[dict] | None = None
) -> tuple[str, list[Document]]:
"""Get answer with source documents.
Args:
question: User question
chat_history: Optional list of previous messages with 'role' and 'content' keys
Returns:
Tuple of (answer, source_documents)
"""
response = self.query(question, chat_history=chat_history)
answer = response.get("result", "I couldn't generate an answer.")
sources = response.get("source_documents", [])
return answer, sources
def stream_query(self, question: str, chat_history: list[dict] | None = None) -> Iterator[str]:
"""Stream the response token by token.
Args:
question: User question
chat_history: Optional list of previous messages with 'role' and 'content' keys
Yields:
Response chunks as they are generated
"""
logger.info(f"Streaming query: {question}")
if chat_history:
logger.debug(f"Including {len(chat_history)} previous messages in context")
try:
# Prepare input with chat history
chain_input = {
"question": question,
"chat_history": chat_history or [],
}
# Stream the response using LCEL's stream method
yield from self.qa_chain.stream(chain_input)
except Exception as e:
logger.error(f"Error streaming query: {e}")
yield "I encountered a technical issue. Please try rephrasing your question."
def stream_answer(self, question: str, chat_history: list[dict] | None = None) -> Iterator[str]:
"""Stream answer to a question (simplified interface).
Args:
question: User question
chat_history: Optional list of previous messages with 'role' and 'content' keys
Yields:
Answer chunks as they are generated
"""
yield from self.stream_query(question, chat_history=chat_history)
def get_source_documents(
self,
question: str,
chat_history: list[dict] | None = None, # noqa: ARG002
) -> list[Document]:
"""Get source documents for a question (can run in parallel with streaming).
Args:
question: User question
chat_history: Optional list of previous messages (not used for retrieval, kept for API consistency)
Returns:
List of source documents
"""
if self.config.get("rag.include_sources", True):
return self.retriever.invoke(question)
return []
def format_sources(self, sources: list[Document]) -> str:
"""Format source documents for display.
Args:
sources: List of source documents
Returns:
Formatted source string
"""
if not sources:
return ""
source_max_length = self.config.get("rag.source_max_length", 280)
formatted_sources = []
for i, doc in enumerate(sources, 1):
source_name = doc.metadata.get("source", "Unknown")
# Prefer basename for cleaner UI
try:
from pathlib import Path
source_name = Path(str(source_name)).name or source_name
except Exception: # noqa: BLE001
pass
page = doc.metadata.get("page", "")
page_info = f" (Page {page + 1})" if page != "" else ""
content = doc.page_content[:source_max_length]
if len(doc.page_content) > source_max_length:
content += "..."
formatted_sources.append(f"{i}. **{source_name}{page_info}**\n {content}")
return "\n\n".join(formatted_sources)
def format_retrieval_panel(self, question: str, sources: list[Document] | None = None) -> str:
"""Build a markdown panel describing the retrieval query and hits.
Args:
question: User query used for retrieval
sources: Optional pre-fetched docs; retrieved again if omitted
Returns:
Markdown string for UI panels
"""
if sources is None:
sources = self.get_source_documents(question)
strategy = self.retrieval_strategy_name
lines = [
"### 🔍 Retrieval debug",
f"**Query:** `{question}`",
f"**Strategy:** `{strategy}`",
f"**Hits:** {len(sources)}",
"",
]
if not sources:
lines.append("_No documents retrieved._")
return "\n".join(lines)
source_max_length = self.config.get("rag.source_max_length", 280)
for i, doc in enumerate(sources, 1):
raw_source = doc.metadata.get("source", "Unknown")
try:
from pathlib import Path
source_name = Path(str(raw_source)).name or str(raw_source)
except Exception: # noqa: BLE001
source_name = str(raw_source)
page = doc.metadata.get("page", "")
page_info = f", page {int(page) + 1}" if page != "" else ""
preview = doc.page_content[:source_max_length].replace("\n", " ")
if len(doc.page_content) > source_max_length:
preview += "..."
lines.append(f"**[{i}]** `{source_name}`{page_info}")
lines.append(f"> {preview}")
lines.append("")
return "\n".join(lines)
def get_answer_with_retrieval(
self, question: str, chat_history: list[dict] | None = None
) -> dict[str, Any]:
"""Return answer plus retrieval panel markdown and source docs.
Args:
question: User question
chat_history: Optional prior turns
Returns:
Dict with keys: answer, retrieval_panel, sources, source_documents
"""
sources = self.get_source_documents(question, chat_history=chat_history)
answer = self.get_answer(question, chat_history=chat_history)
return {
"answer": answer,
"retrieval_panel": self.format_retrieval_panel(question, sources=sources),
"sources": self.format_sources(sources),
"source_documents": sources,
}
def get_main_document_info(self) -> dict[str, Any]:
"""Get information about loaded main document.
Returns:
Dictionary with main document metadata
"""
if not self.main_doc_content:
return {
"enabled": self.config.get("main_document.enabled", False),
"loaded": False,
}
return {
"enabled": True,
"loaded": True,
"tokens": self.main_doc_loader.count_tokens(self.main_doc_content),
"path": str(self.main_doc_loader.path),
"size_bytes": len(self.main_doc_content.encode("utf-8")),
}
def reload_main_document(self) -> bool:
"""Reload main document (useful for runtime updates).
Returns:
True if reload successful, False otherwise
"""
try:
self.main_doc_loader.invalidate_cache()
self.main_doc_content = self.main_doc_loader.load_main_document()
logger.info("Main document reloaded successfully")
# Recalculate budget if main doc is loaded
if self.main_doc_content:
budget = self._calculate_context_budget()
logger.info(
f"Token budget after reload - Main doc: {budget['main_doc_tokens']}, "
f"Available for retrieval: {budget['available_for_retrieval']}"
)
return True
except Exception as e:
logger.error(f"Error reloading main document: {e}")
return False
def _calculate_context_budget(self) -> dict[str, int]:
"""Calculate and log token budget distribution.
Note: Chat history tokens are handled dynamically per query and are not
included in this static budget calculation. The app.py truncation logic
ensures chat history fits within the configured limits.
Returns:
Dictionary with token budget breakdown
"""
# Model-specific context windows
model_name = self.config.get("llm.model", "llama3.2:3b")
context_windows = {
"llama3.2": 8192,
"llama3.1": 128000,
"phi3": 4096,
"gemma2": 8192,
}
# Extract base model name (before colon)
base_model = model_name.split(":")[0] if ":" in model_name else model_name
model_context_window = context_windows.get(base_model, 8192)
max_output_tokens = self.config.get("llm.max_tokens", 512)
main_doc_tokens = self.main_doc_loader.count_tokens(self.main_doc_content)
# Reserve space: Main Doc + Output + Safety Buffer + Chat History (estimated)
buffer_tokens = 500
# Estimate chat history tokens (max configured limit)
max_history_tokens = self.config.get("chat.max_history_tokens", 2000)
estimated_history_tokens = (
max_history_tokens if self.config.get("chat.enable_history", True) else 0
)
available_for_retrieval = (
model_context_window
- main_doc_tokens
- max_output_tokens
- buffer_tokens
- estimated_history_tokens
)
budget = {
"model_context_window": model_context_window,
"main_doc_tokens": main_doc_tokens,
"max_output_tokens": max_output_tokens,
"buffer_tokens": buffer_tokens,
"estimated_chat_history_tokens": estimated_history_tokens,
"available_for_retrieval": max(0, available_for_retrieval),
"total_input_budget": model_context_window - max_output_tokens,
}
logger.debug(f"Token budget: {budget}")
# Warning if main doc is too large
usage_percent = (
(main_doc_tokens / model_context_window) * 100 if model_context_window > 0 else 0
)
if usage_percent > 50:
logger.warning(
f"Main document uses {usage_percent:.1f}% of context window. "
f"Consider summarizing or reducing size."
)
# Warning if estimated total usage is high
total_estimated = main_doc_tokens + estimated_history_tokens + buffer_tokens
total_percent = (
(total_estimated / model_context_window) * 100 if model_context_window > 0 else 0
)
if total_percent > 70:
logger.warning(
f"Estimated context usage ({total_percent:.1f}%) is high. "
f"Consider reducing main document size or chat history limits."
)
return budget
def _create_retrieval_strategy(self):
"""Create the retrieval strategy based on configuration.
Returns:
BaseRetrieverStrategy instance
"""
# Build config dict for the strategy
config_dict = {
"retrieval": {
"strategy": self.retrieval_strategy_name,
"final_k": self.config.get("retrieval.final_k", 4),
"vector": {
"search_type": self.config.get("retrieval.vector.search_type", "similarity"),
"k": self.config.get("retrieval.vector.k", 10),
"search_kwargs": self.config.get("retrieval.vector.search_kwargs", {}),
},
"bm25": {
"k": self.config.get("retrieval.bm25.k", 10),
"persist_path": self.config.get("retrieval.bm25.persist_path", "./bm25_index"),
"tokenizer": self.config.get("retrieval.bm25.tokenizer", "simple"),
},
"fusion": {
"algorithm": self.config.get("retrieval.fusion.algorithm", "rrf"),
"rrf_k": self.config.get("retrieval.fusion.rrf_k", 60),
"weights": self.config.get(
"retrieval.fusion.weights", {"vector": 0.7, "bm25": 0.3}
),
},
}
}
strategy = RetrieverFactory.create(self.retrieval_strategy_name, config_dict)
# Load existing index
if not strategy.load_index():
logger.warning(
f"Could not load index for '{self.retrieval_strategy_name}' strategy. "
f"Run 'python -m src.build_vectorstore --strategy {self.retrieval_strategy_name}' first."
)
return strategy
def _get_retriever(self) -> BaseRetriever:
"""Get the LangChain retriever from the strategy.
Returns:
BaseRetriever instance
"""
final_k = self.config.get("retrieval.final_k", 4)
return self.retrieval_strategy.as_retriever(k=final_k)
def get_retrieval_info(self) -> dict[str, Any]:
"""Get information about the current retrieval strategy.
Returns:
Dictionary with retrieval strategy information
"""
return {
"strategy": self.retrieval_strategy_name,
"stats": self.retrieval_strategy.get_index_stats(),
}
def get_rag_pipeline(retrieval_strategy: str | None = None) -> RAGPipeline:
"""Get RAG pipeline instance.
Args:
retrieval_strategy: Override retrieval strategy (uses config if None)
Returns:
RAGPipeline instance
"""
return RAGPipeline(retrieval_strategy=retrieval_strategy)
|