File size: 9,942 Bytes
9abace2 | 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 | """
RAG Pipeline
Part of SOVEREIGN PYTHON LLM ENGINE
Complete Retrieval-Augmented Generation pipeline.
"""
from typing import AsyncIterator
from dataclasses import dataclass
from datetime import datetime
from ..models.entities import Message, MessageRole, RetrievalRequest, RetrievalResult
from ..core.protocols import Model, VectorStore
from ..tools.embeddings.encode import EmbeddingsTool
from .parallel import ParallelRetriever
from ..core.evidence import WORMLedger
@dataclass
class RAGConfig:
"""Configuration for RAG pipeline"""
top_k: int = 5
score_threshold: float = 0.7
use_reranking: bool = True
context_window: int = 8000 # Max tokens for context
log_to_worm: bool = True
class RAGPipeline:
"""
Complete RAG (Retrieval-Augmented Generation) pipeline.
Steps:
1. Embed query
2. Retrieve top-K documents
3. Optionally rerank
4. Format context
5. Generate answer with LLM
"""
def __init__(
self,
model: Model,
retriever: ParallelRetriever,
embeddings: EmbeddingsTool,
worm_ledger: WORMLedger | None = None,
config: RAGConfig | None = None
):
"""
Initialize RAG pipeline.
Args:
model: LLM for generation
retriever: Parallel retriever
embeddings: Embeddings tool
worm_ledger: Optional WORM ledger
config: Configuration
"""
self.model = model
self.retriever = retriever
self.embeddings = embeddings
self.worm_ledger = worm_ledger
self.config = config or RAGConfig()
async def query(
self,
question: str,
filters: dict | None = None,
stream: bool = False
) -> str | AsyncIterator[str]:
"""
Query the RAG system.
Args:
question: User question
filters: Optional metadata filters
stream: Whether to stream response
Returns:
Answer string (or stream of chunks)
"""
# 1. Retrieve relevant documents
retrieval_request = RetrievalRequest(
query=question,
limit=self.config.top_k,
score_threshold=self.config.score_threshold,
filters=filters
)
results = await self.retriever.retrieve(retrieval_request)
# 2. Format context
context = self._format_context(results)
# 3. Build prompt
prompt = self._build_prompt(question, context)
# 4. Generate answer
messages = [
{"role": "system", "content": "You are a helpful assistant that answers questions based on provided context."},
{"role": "user", "content": prompt}
]
if stream:
# Streaming not implemented in base Model protocol yet
# Return async generator wrapper
answer = await self.model.generate(
messages=messages,
temperature=0.3,
max_tokens=2048
)
# Log to WORM
if self.worm_ledger and self.config.log_to_worm:
await self._log_to_worm(question, results, answer)
async def stream_wrapper():
yield answer
return stream_wrapper()
else:
answer = await self.model.generate(
messages=messages,
temperature=0.3,
max_tokens=2048
)
# Log to WORM
if self.worm_ledger and self.config.log_to_worm:
await self._log_to_worm(question, results, answer)
return answer
async def query_with_citations(
self,
question: str,
filters: dict | None = None
) -> tuple[str, list[RetrievalResult]]:
"""
Query with source citations.
Args:
question: User question
filters: Optional filters
Returns:
(answer, source_documents) tuple
"""
# Retrieve
retrieval_request = RetrievalRequest(
query=question,
limit=self.config.top_k,
score_threshold=self.config.score_threshold,
filters=filters
)
results = await self.retriever.retrieve(retrieval_request)
# Format context
context = self._format_context(results)
# Build prompt with citation instruction
prompt = f"""Answer the question based on the provided context.
Cite sources using [Source N] notation.
Context:
{context}
Question: {question}
Answer:"""
# Generate
messages = [
{"role": "system", "content": "You are a helpful assistant that answers questions and cites sources."},
{"role": "user", "content": prompt}
]
answer = await self.model.generate(
messages=messages,
temperature=0.3,
max_tokens=2048
)
# Log to WORM
if self.worm_ledger and self.config.log_to_worm:
await self._log_to_worm(question, results, answer)
return answer, results
def _format_context(self, results: list[RetrievalResult]) -> str:
"""
Format retrieval results into context string.
Args:
results: Retrieved documents
Returns:
Formatted context
"""
if not results:
return "No relevant documents found."
context_parts = []
for i, result in enumerate(results, 1):
source_info = ""
if result.metadata:
source_info = f" ({result.metadata.get('source', 'Unknown')})"
context_parts.append(
f"[Source {i}]{source_info}\n{result.content}\n"
)
return "\n".join(context_parts)
def _build_prompt(self, question: str, context: str) -> str:
"""
Build RAG prompt.
Args:
question: User question
context: Retrieved context
Returns:
Formatted prompt
"""
return f"""Answer the question based on the following context.
If the context doesn't contain enough information, say so.
Context:
{context}
Question: {question}
Answer:"""
async def _log_to_worm(
self,
question: str,
results: list[RetrievalResult],
answer: str
) -> None:
"""Log RAG query to WORM ledger"""
if not self.worm_ledger:
return
await self.worm_ledger.append({
"event": "rag_query",
"question": question,
"retrieved_docs": len(results),
"top_score": results[0].score if results else None,
"answer_length": len(answer),
"timestamp": datetime.utcnow().isoformat()
})
class ConversationalRAG:
"""
RAG with conversation history.
Maintains context across multiple turns.
"""
def __init__(
self,
rag_pipeline: RAGPipeline,
max_history: int = 10
):
"""
Initialize conversational RAG.
Args:
rag_pipeline: Base RAG pipeline
max_history: Maximum conversation turns to keep
"""
self.rag = rag_pipeline
self.max_history = max_history
self.conversation_history: list[Message] = []
async def query(
self,
question: str,
filters: dict | None = None
) -> str:
"""
Query with conversation context.
Args:
question: User question
filters: Optional filters
Returns:
Answer
"""
# Add user message to history
self.conversation_history.append(
Message(
role=MessageRole.USER,
content=question,
created_at=datetime.utcnow()
)
)
# Retrieve using current question
retrieval_request = RetrievalRequest(
query=question,
limit=self.rag.config.top_k,
score_threshold=self.rag.config.score_threshold,
filters=filters
)
results = await self.rag.retriever.retrieve(retrieval_request)
# Format context
context = self.rag._format_context(results)
# Build messages with history
messages = [
{"role": "system", "content": "You are a helpful assistant. Use the provided context and conversation history to answer questions."}
]
# Add conversation history (limited)
history_messages = self.conversation_history[-(self.max_history * 2):]
for msg in history_messages:
messages.append({
"role": msg.role.value,
"content": msg.content
})
# Add current context
messages.append({
"role": "user",
"content": f"""Context:
{context}
Question: {question}
Answer:"""
})
# Generate
answer = await self.rag.model.generate(
messages=messages,
temperature=0.3,
max_tokens=2048
)
# Add assistant message to history
self.conversation_history.append(
Message(
role=MessageRole.ASSISTANT,
content=answer,
created_at=datetime.utcnow()
)
)
return answer
def clear_history(self) -> None:
"""Clear conversation history"""
self.conversation_history.clear()
|