mirror of
https://github.com/qdrant/qdrant.git
synced 2026-07-26 12:41:04 -05:00
539606e5d4
* feat: integrate bm25 into edge * fix: linter * fix: imports * fix: qdrant-edge build errors * fix: spell check * feat: integrate inference * fix: api missmatch * chore: remove inference * fix: coderabbit comments * fix: compiler error * chore: remove wrapper type * feat: add some temp tests for legacy check * add example --------- Co-authored-by: Andrey Vasnetsov <andrey@vasnetsov.com>
44 lines
1.3 KiB
Python
44 lines
1.3 KiB
Python
"""Native BM25 sparse search: embed text manually with `Bm25`, then upsert and query
|
|
the produced sparse vectors through the regular update/query path.
|
|
"""
|
|
|
|
import os, shutil
|
|
from pathlib import Path
|
|
|
|
from qdrant_edge import (
|
|
EdgeShard, EdgeConfig, EdgeSparseVectorParams, Modifier,
|
|
Bm25, Bm25Config,
|
|
Point, Query, QueryRequest, UpdateOperation,
|
|
)
|
|
|
|
|
|
DATA_DIR = Path(__file__).parent.parent.parent / "data"
|
|
TMP_DIR = DATA_DIR / "tmp"
|
|
|
|
path = TMP_DIR / "qdrant_edge_bm25"
|
|
shutil.rmtree(path, ignore_errors=True)
|
|
os.makedirs(path)
|
|
|
|
config = EdgeConfig(
|
|
sparse_vectors={"text": EdgeSparseVectorParams(modifier=Modifier.Idf)},
|
|
)
|
|
shard = EdgeShard.create(str(path), config)
|
|
|
|
bm25 = Bm25(Bm25Config(language="english"))
|
|
|
|
shard.update(UpdateOperation.upsert_points([
|
|
Point(1, {"text": bm25.embed_document("the quick brown fox")}, {"doc": "1"}),
|
|
Point(2, {"text": bm25.embed_document("a lazy dog sleeps")}, {"doc": "2"}),
|
|
Point(3, {"text": bm25.embed_document("foxes are clever")}, {"doc": "3"}),
|
|
]))
|
|
shard.optimize()
|
|
print(f"Info: {shard.info()}")
|
|
|
|
q = bm25.embed_query("clever fox")
|
|
results = shard.query(QueryRequest(
|
|
limit=3, query=Query.Nearest(q, using="text"), with_payload=True,
|
|
))
|
|
print(f"Got {len(results)} results")
|
|
for r in results:
|
|
print(f" id={r.id} score={r.score:.4f} payload={r.payload}")
|