Skip to content

teff.rag.stores.qdrant

teff.rag.stores.qdrant

Qdrant vector store — requires qdrant-client.

Classes:

Name Description
QdrantVectorStore

Vector store backed by Qdrant.

QdrantVectorStore

Bases: VectorStore

Vector store backed by Qdrant.

Requires the qdrant-client package (install via teff[embedding]).

Source code in teff/rag/stores/qdrant.py
 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
class QdrantVectorStore(VectorStore):
    """Vector store backed by Qdrant.

    Requires the ``qdrant-client`` package (install via ``teff[embedding]``).
    """

    def __init__(
        self,
        host: str = "localhost",
        port: int = 6333,
        collection: str = "teff",
        client=None,
    ):
        if client is None:
            try:
                from qdrant_client import QdrantClient
            except ImportError as e:
                raise ImportError("install qdrant-client for QdrantVectorStore") from e
            client = QdrantClient(host=host, port=port)
        self._client = client
        self._collection = collection

    def _ensure_collection(self, dim: int) -> None:
        from qdrant_client.models import Distance, VectorParams

        collections = {c.name for c in self._client.get_collections().collections}
        if self._collection not in collections:
            self._client.create_collection(
                collection_name=self._collection,
                vectors_config=VectorParams(size=dim, distance=Distance.COSINE),
            )

    @staticmethod
    def _hash_id(vid: str) -> int:
        return int(hashlib.md5(vid.encode()).hexdigest()[:16], 16)

    async def add(self, vectors: list[tuple[str, list[float], dict]]) -> None:
        from qdrant_client.models import PointStruct

        if vectors:
            self._ensure_collection(len(vectors[0][1]))
        points = [
            PointStruct(
                id=self._hash_id(vid),
                vector=vec,
                payload={"doc_id": vid, **meta},
            )
            for vid, vec, meta in vectors
        ]
        self._client.upsert(collection_name=self._collection, points=points)

    async def search(
        self,
        query: list[float],
        k: int = 10,
        filter: dict | None = None,
        hybrid: bool = False,
        query_text: str | None = None,
    ) -> list[tuple[str, float, dict]]:
        results = self._client.query_points(
            collection_name=self._collection,
            query=query,
            limit=k,
            query_filter=_to_qdrant_filter(filter),
        )
        return [
            (r.payload.get("doc_id", str(r.id)), r.score, r.payload)  # type: ignore[misc, union-attr]
            for r in results.points
        ]

    async def delete(self, ids: list[str]) -> None:
        hashed = [self._hash_id(vid) for vid in ids]
        self._client.delete(collection_name=self._collection, points_selector=hashed)  # type: ignore[arg-type]

    async def count(self) -> int:
        return self._client.count(collection_name=self._collection).count

    async def entries(
        self, limit: int = 100, offset: int = 0
    ) -> list[tuple[str, dict]]:
        res = self._client.scroll(
            collection_name=self._collection,
            limit=limit,
            offset=offset,
            with_payload=True,
            with_vectors=False,
        )
        points, _ = res
        return [(p.payload.get("doc_id", str(p.id)), p.payload) for p in points]  # type: ignore[misc, union-attr]

    async def get(self, ids: list[str]) -> list[tuple[str, dict]]:
        hashed = [self._hash_id(vid) for vid in ids]
        res = self._client.retrieve(
            collection_name=self._collection,
            ids=hashed,
            with_payload=True,
            with_vectors=False,
        )
        return [(p.payload.get("doc_id", str(p.id)), p.payload) for p in res]  # type: ignore[misc, union-attr]

    async def update_metadata(self, id: str, metadata: dict) -> None:
        from qdrant_client import models

        self._client.set_payload(
            collection_name=self._collection,
            payload=metadata,
            points=models.FilterSelector(
                filter=models.Filter(
                    must=[
                        models.FieldCondition(
                            key="doc_id", match=models.MatchValue(value=id)
                        )
                    ]
                )
            ),
        )

    async def clear(self) -> None:
        from qdrant_client import models

        self._client.delete(
            collection_name=self._collection,
            points_selector=models.FilterSelector(filter=models.Filter(must=[])),
        )