SUDOZOVCHIK/vk_hackathon
1
0
Fork 0
forked from zovos/vk_hackathon
Code Pull requests Activity

Revert to v1.0-working baseline (score 0.5094)

Browse source 
Improvements to RERANK_LIMIT, search_text, variants made score worse.
Reverting to investigate better approach.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
q 2026-04-18 19:37:38 +03:00
parent d5ed8c7764
commit f4b14edabf
2 changed files with 10 additions and 17 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
index/main.py
View file

@ -11,7 +11,7 @@ from pydantic import BaseModel
HOST = os.getenv("HOST", "0.0.0.0") HOST = os.getenv("HOST", "0.0.0.0")
PORT = int(os.getenv("PORT", "8000")) PORT = int(os.getenv("PORT", "8000"))
UVICORN_WORKERS = 4 UVICORN_WORKERS = 8
logging.basicConfig(level=os.getenv("LOG_LEVEL", "INFO")) logging.basicConfig(level=os.getenv("LOG_LEVEL", "INFO"))
logger = logging.getLogger("index-service") logger = logging.getLogger("index-service")

25
search/main.py
View file

@ -165,7 +165,7 @@ app = FastAPI(title="Search Service", version="0.1.0", lifespan=lifespan)
DENSE_PREFETCH_K = 80 DENSE_PREFETCH_K = 80
SPARSE_PREFETCH_K = 200 SPARSE_PREFETCH_K = 200
RETRIEVE_K = 150 RETRIEVE_K = 150
RERANK_LIMIT = 60 RERANK_LIMIT = 10
async def embed_dense(client: httpx.AsyncClient, text: str) -> list[float]: async def embed_dense(client: httpx.AsyncClient, text: str) -> list[float]:
@ -211,16 +211,15 @@ def embed_sparse_sync(text: str) -> SparseVector:
def build_dense_query(question: Question) -> str: def build_dense_query(question: Question) -> str:
if question.search_text:
return question.search_text.strip()
return question.text.strip() return question.text.strip()
def build_sparse_query(question: Question) -> str: def build_sparse_query(question: Question) -> str:
base = question.search_text.strip() if question.search_text else question.text.strip() parts = [question.text.strip()]
parts = [base]
if question.keywords: if question.keywords:
parts.extend(question.keywords) parts.extend(question.keywords)
if question.search_text:
parts = [question.search_text]
return " ".join(parts) return " ".join(parts)
@ -359,19 +358,13 @@ async def search(payload: SearchAPIRequest) -> SearchAPIResponse:
dense_vector, sparse_vector = await asyncio.gather(dense_task, sparse_task) dense_vector, sparse_vector = await asyncio.gather(dense_task, sparse_task)
dense_vectors = [dense_vector] dense_vectors = [dense_vector]
extra_texts: list[str] = [] if question.hyde and len(question.hyde) > 0:
for v in (question.variants or [])[:2]:
if v.strip():
extra_texts.append(v.strip())
for h in (question.hyde or [])[:2]:
if h.strip():
extra_texts.append(h.strip())
if extra_texts:
try: try:
extra_vectors = await embed_dense_batch(client, extra_texts) hyde_texts = question.hyde[:2]
dense_vectors.extend(extra_vectors) hyde_vectors = await embed_dense_batch(client, hyde_texts)
dense_vectors.extend(hyde_vectors)
except Exception as e: except Exception as e:
logger.warning(f"Extra dense embedding failed: {e}") logger.warning(f"HyDE embedding failed: {e}")
all_points = await qdrant_search(qdrant, dense_vectors, sparse_vector) all_points = await qdrant_search(qdrant, dense_vectors, sparse_vector)