224 lines
9.7 KiB
Python
224 lines
9.7 KiB
Python
"""Production answer path: retrieval -> model selection -> guard -> rendered text."""
|
|
|
|
from __future__ import annotations
|
|
|
|
import threading
|
|
from pathlib import Path
|
|
|
|
from tools.rag.common import DB_FILE
|
|
from tools.rag.rag_utils import build_rag_context
|
|
from tools.rag.llm_context import compact_rag_payload
|
|
from tools.graphrag.graph_search import search_graph
|
|
from tools.answers.graph_context import graph_context
|
|
from tools.answers.checked_generation import generate_checked_answer
|
|
from tools.answers.verified_rag import generate_verified_rag
|
|
from tools.answers.rag_evidence import expand_comparison_sources
|
|
from tools.answers.rag_preparation import prepare_named_content
|
|
from tools.sync.index_state import index_snapshot
|
|
from tools.answers.model_client import complete_with_configured_model
|
|
from tools.answers.answer_errors import public_failure, clarification_result
|
|
|
|
_capacity = threading.BoundedSemaphore(2)
|
|
|
|
|
|
def retrieval_failure(engine: str) -> dict:
|
|
return public_failure(
|
|
{
|
|
"reason": "retrieval_unavailable",
|
|
"engine": engine,
|
|
"index_version": None,
|
|
"complete": False,
|
|
"truncated": False,
|
|
"truncation_reasons": [],
|
|
}
|
|
)
|
|
|
|
|
|
def name_failure(retrieved: dict, engine: str) -> dict | None:
|
|
resolution = retrieved.get("name_resolution") or {}
|
|
if not resolution or resolution.get("status") == "ok":
|
|
return None
|
|
return {
|
|
"status": resolution["status"],
|
|
"reason": resolution["reason"],
|
|
"engine": engine,
|
|
"accepted_for_delivery": True,
|
|
"answer": resolution["message"],
|
|
"claims": [],
|
|
"source_urls": [],
|
|
"name_resolution": resolution,
|
|
"index_version": retrieved.get("index_version"),
|
|
"complete": False,
|
|
"truncated": False,
|
|
"truncation_reasons": [],
|
|
}
|
|
|
|
|
|
def with_name_resolution(result: dict, retrieved: dict) -> dict:
|
|
resolution = retrieved.get("name_resolution") or {}
|
|
if result.get("accepted_for_delivery") and resolution.get("notices"):
|
|
result["answer"] = "\n".join(resolution["notices"]) + "\n\n" + result["answer"]
|
|
if resolution:
|
|
result["name_resolution"] = resolution
|
|
return result
|
|
|
|
|
|
def deliver_answer(
|
|
question: str,
|
|
*,
|
|
engine: str = "rag",
|
|
limit: int = 5,
|
|
complete=None,
|
|
db_file: Path = DB_FILE,
|
|
include_diagnostics: bool = False,
|
|
) -> dict:
|
|
from tools.answers.reference_links import checked_repository_links
|
|
|
|
result = _deliver_answer(
|
|
question,
|
|
engine=engine,
|
|
limit=limit,
|
|
complete=complete,
|
|
db_file=db_file,
|
|
include_diagnostics=include_diagnostics,
|
|
)
|
|
return checked_repository_links(result)
|
|
|
|
|
|
def _deliver_answer(
|
|
question: str,
|
|
*,
|
|
engine: str = "rag",
|
|
limit: int = 5,
|
|
complete=None,
|
|
db_file: Path = DB_FILE,
|
|
include_diagnostics: bool = False,
|
|
) -> dict:
|
|
if not _capacity.acquire(blocking=False):
|
|
return public_failure({"reason": "answer_capacity_exceeded", "engine": engine})
|
|
try:
|
|
model = complete or complete_with_configured_model
|
|
from tools.answers.question_plan import (build_question_plan, planning_catalog, QuestionPlan, PlanMeaning,
|
|
PlanRejected, PlanModelUnavailable)
|
|
from tools.knowledge.query import deliver_content, ContentModelUnavailable, ContentAnswerRejected
|
|
try:
|
|
with index_snapshot(db_file.parent, db_file=db_file) as planned_version:
|
|
catalog = planning_catalog(db_file)
|
|
plan_model = complete or (lambda messages: complete_with_configured_model(
|
|
messages, output_schema=QuestionPlan.model_json_schema()))
|
|
verify_model = complete or (lambda messages: complete_with_configured_model(
|
|
messages, output_schema=PlanMeaning.model_json_schema()))
|
|
plan = build_question_plan(question, catalog, complete=plan_model,verify_complete=verify_model,index_version=planned_version)
|
|
except PlanModelUnavailable:
|
|
return public_failure({'reason':'model_unavailable','engine':engine})
|
|
except PlanRejected:
|
|
return public_failure({'reason':'answer_rejected','engine':engine})
|
|
except Exception:
|
|
return retrieval_failure(engine)
|
|
if plan.clarification:
|
|
value=clarification_result(plan.clarification,plan.reason,engine=engine,index_version=planned_version)
|
|
if plan.resolution.get('status')=='no_answer' and plan.reason==plan.resolution.get('reason'):value['status']='no_answer'
|
|
value['name_resolution']=plan.resolution
|
|
if include_diagnostics:value['question_plan']=plan.model_dump()
|
|
return value
|
|
question = plan.spec.question
|
|
try:
|
|
content_result = deliver_content(
|
|
question, engine=engine, limit=limit, db_file=db_file, complete=model, plan=plan
|
|
)
|
|
if content_result is not None:
|
|
if include_diagnostics:content_result["question_plan"]=plan.model_dump()
|
|
return content_result
|
|
except ContentModelUnavailable:
|
|
return public_failure({"reason": "model_unavailable", "engine": engine})
|
|
except ContentAnswerRejected:
|
|
return public_failure({"reason": "answer_rejected", "engine": engine})
|
|
except Exception:
|
|
return public_failure({"reason": "content_index_unavailable", "engine": engine})
|
|
if engine == "rag":
|
|
try:
|
|
with index_snapshot(db_file.parent, db_file=db_file):
|
|
retrieved = build_rag_context(db_file, question, limit=limit, question_plan=plan.spec)
|
|
if retrieved.get("status") == "error":
|
|
return retrieval_failure(engine)
|
|
if failure := name_failure(retrieved, engine):
|
|
return failure
|
|
question = retrieved.get("resolved_query") or question
|
|
if planned_version is not None and retrieved.get("index_version") != planned_version:
|
|
return retrieval_failure(engine)
|
|
retrieved["question_plan"] = plan.spec.model_dump()
|
|
retrieved["name_resolution"] = plan.resolution
|
|
retrieved = expand_comparison_sources(db_file, question, retrieved)
|
|
retrieved = prepare_named_content(db_file, question, retrieved)
|
|
retrieved = compact_rag_payload(retrieved)
|
|
except Exception:
|
|
return retrieval_failure(engine)
|
|
result = generate_verified_rag(question, retrieved, model)
|
|
context = retrieved
|
|
elif engine == "graphrag":
|
|
if plan.spec.intent != "metadata":
|
|
return clarification_result(
|
|
"Táto otázka vyžaduje textové podklady o zadaní, priebehu alebo výsledkoch práce. Použi /rag s rovnakou otázkou; graf samotný tieto výňatky neposkytuje.",
|
|
"text_evidence_required",
|
|
engine=engine,
|
|
)
|
|
try:
|
|
retrieved = search_graph(question, limit=limit, question_plan=plan.spec)
|
|
if retrieved.get("status") == "error":
|
|
return retrieval_failure(engine)
|
|
if failure := name_failure(retrieved, engine):
|
|
return failure
|
|
question = retrieved.get("resolved_query") or question
|
|
if planned_version is not None and retrieved.get("index_version") != planned_version:
|
|
return retrieval_failure(engine)
|
|
retrieved["name_resolution"] = plan.resolution
|
|
context = graph_context(retrieved)
|
|
except Exception:
|
|
return retrieval_failure(engine)
|
|
if context.get("clarification"):
|
|
return with_name_resolution(
|
|
clarification_result(
|
|
context["clarification"],
|
|
context.get("reason"),
|
|
engine=engine,
|
|
index_version=context.get("index_version"),
|
|
),
|
|
retrieved,
|
|
)
|
|
prompt = (
|
|
Path(__file__).resolve().parents[1] / "answers/graph_answer_prompt.txt"
|
|
).read_text()
|
|
generated = generate_checked_answer(question, context, prompt, model)
|
|
if generated.get("accepted_for_delivery"):
|
|
result = {
|
|
**generated["answer"],
|
|
"accepted_for_delivery": True,
|
|
"reason": None,
|
|
"index_version": context.get("index_version"),
|
|
"complete": context.get("complete"),
|
|
"truncated": context.get("truncated"),
|
|
"truncation_reasons": context.get("truncation_reasons"),
|
|
"attempts": generated["attempts"],
|
|
}
|
|
else:
|
|
reason = generated.get("reason", "answer_rejected")
|
|
result = public_failure(
|
|
{
|
|
"reason": reason,
|
|
"index_version": context.get("index_version"),
|
|
"attempts": generated["attempts"],
|
|
}
|
|
)
|
|
else:
|
|
raise ValueError("Unsupported answer engine")
|
|
if result.get("status") == "error" or result.get("accepted_for_delivery") is not True:
|
|
result = public_failure(result)
|
|
result = with_name_resolution(result, retrieved)
|
|
return {
|
|
**result,
|
|
"engine": engine,
|
|
**({"retrieval_context": context, "question_plan": plan.model_dump()} if include_diagnostics else {}),
|
|
}
|
|
finally:
|
|
_capacity.release()
|