Richie
c9f859cac5
Move protected phrase detection into the retrieval gather so it runs concurrently with vector and BM25 candidates instead of sequentially before them. Make the search API accept real bool form fields for rerank/phrase_matching, gate phrase matching on both the request and config kill switch, and reflow log f-strings for readability.
124 lines
4.2 KiB
Python
124 lines
4.2 KiB
Python
"""Search routes for the EPUB search web UI."""
|
|
|
|
from __future__ import annotations
|
|
|
|
import logging
|
|
from dataclasses import replace
|
|
from time import perf_counter
|
|
from typing import TYPE_CHECKING, Annotated
|
|
|
|
from fastapi import APIRouter, Form, Request
|
|
from fastapi.responses import HTMLResponse
|
|
|
|
from python.ebook_search.answer import answer_query
|
|
from python.ebook_search.api.dependencies import ( # noqa: TC001 FastAPI resolves these annotated dependencies at runtime
|
|
AppConfig,
|
|
AppEngine,
|
|
AppHttpClient,
|
|
)
|
|
from python.ebook_search.api.web import templates
|
|
from python.ebook_search.guardrails import (
|
|
CitationReport,
|
|
is_confident,
|
|
retrieval_confidence,
|
|
validate_citations,
|
|
)
|
|
from python.ebook_search.search import SearchResponse, search_ebooks
|
|
from python.ebook_search.timing import runtime_step_from_start
|
|
|
|
if TYPE_CHECKING:
|
|
import httpx
|
|
|
|
from python.ebook_search.config import EbookSearchConfig
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
router = APIRouter()
|
|
|
|
|
|
async def build_answer(
|
|
client: httpx.AsyncClient,
|
|
query: str,
|
|
response: SearchResponse,
|
|
config: EbookSearchConfig,
|
|
) -> tuple[str, bool, CitationReport | None]:
|
|
"""Generate the answer for a search, returning ``(answer, low_confidence, citation_report)``."""
|
|
if not config.answer_enabled:
|
|
logger.info("ebook_answer_skipped_disabled")
|
|
return "Answer generation is disabled. Source chunks are shown below.", False, None
|
|
|
|
if not is_confident(response.results, config):
|
|
logger.info(
|
|
f"ebook_answer_low_confidence confidence={retrieval_confidence(response.results):.4f} "
|
|
f"{config.min_retrieval_confidence=:.4f}"
|
|
)
|
|
answer = (
|
|
"Retrieval confidence is low for this query, so answer generation was skipped. "
|
|
"Source chunks are shown below."
|
|
)
|
|
return answer, True, None
|
|
|
|
try:
|
|
answer = await answer_query(client, query, response.results, config)
|
|
except RuntimeError as error:
|
|
logger.warning(f"ebook_answer_request_failed_falling_back {error=}")
|
|
return "Answer generation failed. Source chunks are still shown below.", False, None
|
|
|
|
citation_report = None
|
|
if config.validate_citations_enabled and response.results:
|
|
citation_report = validate_citations(answer, len(response.results))
|
|
if citation_report.invalid or not citation_report.grounded:
|
|
logger.warning(f"ebook_answer_citation_issue {citation_report.invalid=} {citation_report.grounded=}")
|
|
return answer, False, citation_report
|
|
|
|
|
|
@router.post("/search", response_class=HTMLResponse)
|
|
async def search(
|
|
request: Request,
|
|
config: AppConfig,
|
|
engine: AppEngine,
|
|
client: AppHttpClient,
|
|
query: Annotated[str, Form()],
|
|
*,
|
|
rerank: Annotated[bool, Form()] = False,
|
|
phrase_matching: Annotated[bool, Form()] = False,
|
|
) -> HTMLResponse:
|
|
"""Run a search and render HTMX results."""
|
|
try:
|
|
response = await search_ebooks(
|
|
engine,
|
|
client,
|
|
query,
|
|
config,
|
|
rerank=rerank,
|
|
phrase_matching=phrase_matching,
|
|
)
|
|
except Exception as error:
|
|
logger.exception("ebook_search_request_failed")
|
|
return templates.TemplateResponse(request, "partials/error.html", {"message": str(error)}, status_code=500)
|
|
|
|
answer_start = perf_counter()
|
|
answer, low_confidence, citation_report = await build_answer(client, query, response, config)
|
|
answer_step_name = "Answer generation" if config.answer_enabled else "Answer skipped"
|
|
response = replace(
|
|
response,
|
|
timings=(*response.timings, runtime_step_from_start(answer_step_name, answer_start)),
|
|
)
|
|
|
|
for step in response.timings:
|
|
logger.info(f"ebook_search_timing {step.name=} {step.duration_ms=:.1f}")
|
|
logger.info(
|
|
f"ebook_search_request_complete results={len(response.results)} {response.rank_label=} "
|
|
f"{response.total_runtime_ms=:.1f}"
|
|
)
|
|
return templates.TemplateResponse(
|
|
request,
|
|
"partials/results.html",
|
|
{
|
|
"answer": answer,
|
|
"response": response,
|
|
"low_confidence": low_confidence,
|
|
"citation_report": citation_report,
|
|
},
|
|
)
|