Skip to content

Qdrant

QdrantVectorStore #

Bases: BasePydanticVectorStore

Qdrant Vector Store.

In this vector store, embeddings and docs are stored within a Qdrant collection.

During query time, the index uses Qdrant to query for the top k most similar nodes.

Parameters:

Name Type Description Default
collection_name str

(str): name of the Qdrant collection

required
client Optional[Any]

QdrantClient instance from qdrant-client package

None
aclient Optional[Any]

AsyncQdrantClient instance from qdrant-client package

None
url Optional[str]

url of the Qdrant instance

None
api_key Optional[str]

API key for authenticating with Qdrant

None
batch_size int

number of points to upload in a single request to Qdrant. Defaults to 64

64
parallel int

number of parallel processes to use during upload. Defaults to 1

1
max_retries int

maximum number of retries in case of a failure. Defaults to 3

3
client_kwargs Optional[dict]

additional kwargs for QdrantClient and AsyncQdrantClient

None
enable_hybrid bool

whether to enable hybrid search using dense and sparse vectors

False
sparse_doc_fn Optional[SparseEncoderCallable]

function to encode sparse vectors

None
sparse_query_fn Optional[SparseEncoderCallable]

function to encode sparse queries

None
hybrid_fusion_fn Optional[HybridFusionCallable]

function to fuse hybrid search results

None

Examples:

pip install llama-index-vector-stores-qdrant

import qdrant_client
from llama_index.vector_stores.qdrant import QdrantVectorStore

client = qdrant_client.QdrantClient()

vector_store = QdrantVectorStore(
    collection_name="example_collection", client=client
)
Source code in llama-index-integrations/vector_stores/llama-index-vector-stores-qdrant/llama_index/vector_stores/qdrant/base.py
 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
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
class QdrantVectorStore(BasePydanticVectorStore):
    """
    Qdrant Vector Store.

    In this vector store, embeddings and docs are stored within a
    Qdrant collection.

    During query time, the index uses Qdrant to query for the top
    k most similar nodes.

    Args:
        collection_name: (str): name of the Qdrant collection
        client (Optional[Any]): QdrantClient instance from `qdrant-client` package
        aclient (Optional[Any]): AsyncQdrantClient instance from `qdrant-client` package
        url (Optional[str]): url of the Qdrant instance
        api_key (Optional[str]): API key for authenticating with Qdrant
        batch_size (int): number of points to upload in a single request to Qdrant. Defaults to 64
        parallel (int): number of parallel processes to use during upload. Defaults to 1
        max_retries (int): maximum number of retries in case of a failure. Defaults to 3
        client_kwargs (Optional[dict]): additional kwargs for QdrantClient and AsyncQdrantClient
        enable_hybrid (bool): whether to enable hybrid search using dense and sparse vectors
        sparse_doc_fn (Optional[SparseEncoderCallable]): function to encode sparse vectors
        sparse_query_fn (Optional[SparseEncoderCallable]): function to encode sparse queries
        hybrid_fusion_fn (Optional[HybridFusionCallable]): function to fuse hybrid search results

    Examples:
        `pip install llama-index-vector-stores-qdrant`

        ```python
        import qdrant_client
        from llama_index.vector_stores.qdrant import QdrantVectorStore

        client = qdrant_client.QdrantClient()

        vector_store = QdrantVectorStore(
            collection_name="example_collection", client=client
        )
        ```
    """

    stores_text: bool = True
    flat_metadata: bool = False

    collection_name: str
    path: Optional[str]
    url: Optional[str]
    api_key: Optional[str]
    batch_size: int
    parallel: int
    max_retries: int
    client_kwargs: dict = Field(default_factory=dict)
    enable_hybrid: bool

    _client: Any = PrivateAttr()
    _aclient: Any = PrivateAttr()
    _collection_initialized: bool = PrivateAttr()
    _sparse_doc_fn: Optional[SparseEncoderCallable] = PrivateAttr()
    _sparse_query_fn: Optional[SparseEncoderCallable] = PrivateAttr()
    _hybrid_fusion_fn: Optional[HybridFusionCallable] = PrivateAttr()

    def __init__(
        self,
        collection_name: str,
        client: Optional[Any] = None,
        aclient: Optional[Any] = None,
        url: Optional[str] = None,
        api_key: Optional[str] = None,
        batch_size: int = 64,
        parallel: int = 1,
        max_retries: int = 3,
        client_kwargs: Optional[dict] = None,
        enable_hybrid: bool = False,
        sparse_doc_fn: Optional[SparseEncoderCallable] = None,
        sparse_query_fn: Optional[SparseEncoderCallable] = None,
        hybrid_fusion_fn: Optional[HybridFusionCallable] = None,
        **kwargs: Any,
    ) -> None:
        """Init params."""
        if (
            client is None
            and aclient is None
            and (url is None or api_key is None or collection_name is None)
        ):
            raise ValueError(
                "Must provide either a QdrantClient instance or a url and api_key."
            )

        if client is None and aclient is None:
            client_kwargs = client_kwargs or {}
            self._client = qdrant_client.QdrantClient(
                url=url, api_key=api_key, **client_kwargs
            )
            self._aclient = qdrant_client.AsyncQdrantClient(
                url=url, api_key=api_key, **client_kwargs
            )
        else:
            if client is not None and aclient is not None:
                logger.warning(
                    "Both client and aclient are provided. If using `:memory:` "
                    "mode, the data between clients is not synced."
                )

            self._client = client
            self._aclient = aclient

        if self._client is not None:
            self._collection_initialized = self._collection_exists(collection_name)
        else:
            #  need to do lazy init for async clients
            self._collection_initialized = False

        # setup hybrid search if enabled
        if enable_hybrid:
            self._sparse_doc_fn = sparse_doc_fn or self.get_default_sparse_doc_encoder(
                collection_name
            )
            self._sparse_query_fn = (
                sparse_query_fn
                or self.get_default_sparse_query_encoder(collection_name)
            )
            self._hybrid_fusion_fn = hybrid_fusion_fn or cast(
                HybridFusionCallable, relative_score_fusion
            )

        super().__init__(
            collection_name=collection_name,
            url=url,
            api_key=api_key,
            batch_size=batch_size,
            parallel=parallel,
            max_retries=max_retries,
            client_kwargs=client_kwargs or {},
            enable_hybrid=enable_hybrid,
        )

    @classmethod
    def class_name(cls) -> str:
        return "QdrantVectorStore"

    def set_query_functions(
        self,
        sparse_doc_fn: Optional[SparseEncoderCallable] = None,
        sparse_query_fn: Optional[SparseEncoderCallable] = None,
        hybrid_fusion_fn: Optional[HybridFusionCallable] = None,
    ):
        self._sparse_doc_fn = sparse_doc_fn
        self._sparse_query_fn = sparse_query_fn
        self._hybrid_fusion_fn = hybrid_fusion_fn

    def _build_points(self, nodes: List[BaseNode]) -> Tuple[List[Any], List[str]]:
        ids = []
        points = []
        for node_batch in iter_batch(nodes, self.batch_size):
            node_ids = []
            vectors: List[Any] = []
            sparse_vectors: List[List[float]] = []
            sparse_indices: List[List[int]] = []
            payloads = []

            if self.enable_hybrid and self._sparse_doc_fn is not None:
                sparse_indices, sparse_vectors = self._sparse_doc_fn(
                    [
                        node.get_content(metadata_mode=MetadataMode.EMBED)
                        for node in node_batch
                    ],
                )

            for i, node in enumerate(node_batch):
                assert isinstance(node, BaseNode)
                node_ids.append(node.node_id)

                if self.enable_hybrid:
                    if (
                        len(sparse_vectors) > 0
                        and len(sparse_indices) > 0
                        and len(sparse_vectors) == len(sparse_indices)
                    ):
                        vectors.append(
                            {
                                # Dynamically switch between the old and new sparse vector name
                                self.sparse_vector_name: rest.SparseVector(
                                    indices=sparse_indices[i],
                                    values=sparse_vectors[i],
                                ),
                                DENSE_VECTOR_NAME: node.get_embedding(),
                            }
                        )
                    else:
                        vectors.append(
                            {
                                DENSE_VECTOR_NAME: node.get_embedding(),
                            }
                        )
                else:
                    vectors.append(node.get_embedding())

                metadata = node_to_metadata_dict(
                    node, remove_text=False, flat_metadata=self.flat_metadata
                )

                payloads.append(metadata)

            points.extend(
                [
                    rest.PointStruct(id=node_id, payload=payload, vector=vector)
                    for node_id, payload, vector in zip(node_ids, payloads, vectors)
                ]
            )

            ids.extend(node_ids)

        return points, ids

    def add(self, nodes: List[BaseNode], **add_kwargs: Any) -> List[str]:
        """
        Add nodes to index.

        Args:
            nodes: List[BaseNode]: list of nodes with embeddings

        """
        if len(nodes) > 0 and not self._collection_initialized:
            self._create_collection(
                collection_name=self.collection_name,
                vector_size=len(nodes[0].get_embedding()),
            )

        points, ids = self._build_points(nodes)

        self._client.upload_points(
            collection_name=self.collection_name,
            points=points,
            batch_size=self.batch_size,
            parallel=self.parallel,
            max_retries=self.max_retries,
            wait=True,
        )

        return ids

    async def async_add(self, nodes: List[BaseNode], **kwargs: Any) -> List[str]:
        """
        Asynchronous method to add nodes to Qdrant index.

        Args:
            nodes: List[BaseNode]: List of nodes with embeddings.

        Returns:
            List of node IDs that were added to the index.

        Raises:
            ValueError: If trying to using async methods without aclient
        """
        collection_initialized = await self._acollection_exists(self.collection_name)

        if len(nodes) > 0 and not collection_initialized:
            await self._acreate_collection(
                collection_name=self.collection_name,
                vector_size=len(nodes[0].get_embedding()),
            )

        points, ids = self._build_points(nodes)

        await self._aclient.upload_points(
            collection_name=self.collection_name,
            points=points,
            batch_size=self.batch_size,
            parallel=self.parallel,
            max_retries=self.max_retries,
            wait=True,
        )

        return ids

    def delete(self, ref_doc_id: str, **delete_kwargs: Any) -> None:
        """
        Delete nodes using with ref_doc_id.

        Args:
            ref_doc_id (str): The doc_id of the document to delete.

        """
        self._client.delete(
            collection_name=self.collection_name,
            points_selector=rest.Filter(
                must=[
                    rest.FieldCondition(
                        key="doc_id", match=rest.MatchValue(value=ref_doc_id)
                    )
                ]
            ),
        )

    async def adelete(self, ref_doc_id: str, **delete_kwargs: Any) -> None:
        """
        Asynchronous method to delete nodes using with ref_doc_id.

        Args:
            ref_doc_id (str): The doc_id of the document to delete.

        """
        await self._aclient.delete(
            collection_name=self.collection_name,
            points_selector=rest.Filter(
                must=[
                    rest.FieldCondition(
                        key="doc_id", match=rest.MatchValue(value=ref_doc_id)
                    )
                ]
            ),
        )

    @property
    def client(self) -> Any:
        """Return the Qdrant client."""
        return self._client

    def _create_collection(self, collection_name: str, vector_size: int) -> None:
        """Create a Qdrant collection."""
        from qdrant_client.http import models as rest
        from qdrant_client.http.exceptions import UnexpectedResponse

        try:
            if self.enable_hybrid:
                self._client.create_collection(
                    collection_name=collection_name,
                    vectors_config={
                        DENSE_VECTOR_NAME: rest.VectorParams(
                            size=vector_size,
                            distance=rest.Distance.COSINE,
                        )
                    },
                    # Newly created collection will have the new sparse vector name
                    sparse_vectors_config={
                        SPARSE_VECTOR_NAME: rest.SparseVectorParams(
                            index=rest.SparseIndexParams()
                        )
                    },
                )
            else:
                self._client.create_collection(
                    collection_name=collection_name,
                    vectors_config=rest.VectorParams(
                        size=vector_size,
                        distance=rest.Distance.COSINE,
                    ),
                )
        except (RpcError, ValueError, UnexpectedResponse) as exc:
            if "already exists" not in str(exc):
                raise exc  # noqa: TRY201
            logger.warning(
                "Collection %s already exists, skipping collection creation.",
                collection_name,
            )
        self._collection_initialized = True

    async def _acreate_collection(self, collection_name: str, vector_size: int) -> None:
        """Asynchronous method to create a Qdrant collection."""
        from qdrant_client.http import models as rest
        from qdrant_client.http.exceptions import UnexpectedResponse

        try:
            if self.enable_hybrid:
                await self._aclient.create_collection(
                    collection_name=collection_name,
                    vectors_config={
                        DENSE_VECTOR_NAME: rest.VectorParams(
                            size=vector_size,
                            distance=rest.Distance.COSINE,
                        )
                    },
                    sparse_vectors_config={
                        SPARSE_VECTOR_NAME: rest.SparseVectorParams(
                            index=rest.SparseIndexParams()
                        )
                    },
                )
            else:
                await self._aclient.create_collection(
                    collection_name=collection_name,
                    vectors_config=rest.VectorParams(
                        size=vector_size,
                        distance=rest.Distance.COSINE,
                    ),
                )
        except (RpcError, ValueError, UnexpectedResponse) as exc:
            if "already exists" not in str(exc):
                raise exc  # noqa: TRY201
            logger.warning(
                "Collection %s already exists, skipping collection creation.",
                collection_name,
            )
        self._collection_initialized = True

    def _collection_exists(self, collection_name: str) -> bool:
        """Check if a collection exists."""
        try:
            self._client.get_collection(collection_name)
        except (RpcError, UnexpectedResponse, ValueError):
            return False
        return True

    async def _acollection_exists(self, collection_name: str) -> bool:
        """Asynchronous method to check if a collection exists."""
        try:
            await self._aclient.get_collection(collection_name)
        except (RpcError, UnexpectedResponse, ValueError):
            return False
        return True

    def query(
        self,
        query: VectorStoreQuery,
        **kwargs: Any,
    ) -> VectorStoreQueryResult:
        """
        Query index for top k most similar nodes.

        Args:
            query (VectorStoreQuery): query
        """
        query_embedding = cast(List[float], query.query_embedding)
        #  NOTE: users can pass in qdrant_filters (nested/complicated filters) to override the default MetadataFilters
        qdrant_filters = kwargs.get("qdrant_filters")
        if qdrant_filters is not None:
            query_filter = qdrant_filters
        else:
            query_filter = cast(Filter, self._build_query_filter(query))

        if query.mode == VectorStoreQueryMode.HYBRID and not self.enable_hybrid:
            raise ValueError(
                "Hybrid search is not enabled. Please build the query with "
                "`enable_hybrid=True` in the constructor."
            )
        elif (
            query.mode == VectorStoreQueryMode.HYBRID
            and self.enable_hybrid
            and self._sparse_query_fn is not None
            and query.query_str is not None
        ):
            sparse_indices, sparse_embedding = self._sparse_query_fn(
                [query.query_str],
            )
            sparse_top_k = query.sparse_top_k or query.similarity_top_k

            sparse_response = self._client.search_batch(
                collection_name=self.collection_name,
                requests=[
                    rest.SearchRequest(
                        vector=rest.NamedVector(
                            name=DENSE_VECTOR_NAME,
                            vector=query_embedding,
                        ),
                        limit=query.similarity_top_k,
                        filter=query_filter,
                        with_payload=True,
                    ),
                    rest.SearchRequest(
                        vector=rest.NamedSparseVector(
                            # Dynamically switch between the old and new sparse vector name
                            name=self.sparse_vector_name,
                            vector=rest.SparseVector(
                                indices=sparse_indices[0],
                                values=sparse_embedding[0],
                            ),
                        ),
                        limit=sparse_top_k,
                        filter=query_filter,
                        with_payload=True,
                    ),
                ],
            )

            # sanity check
            assert len(sparse_response) == 2
            assert self._hybrid_fusion_fn is not None

            # flatten the response
            return self._hybrid_fusion_fn(
                self.parse_to_query_result(sparse_response[0]),
                self.parse_to_query_result(sparse_response[1]),
                # NOTE: only for hybrid search (0 for sparse search, 1 for dense search)
                alpha=query.alpha or 0.5,
                # NOTE: use hybrid_top_k if provided, otherwise use similarity_top_k
                top_k=query.hybrid_top_k or query.similarity_top_k,
            )
        elif (
            query.mode == VectorStoreQueryMode.SPARSE
            and self.enable_hybrid
            and self._sparse_query_fn is not None
            and query.query_str is not None
        ):
            sparse_indices, sparse_embedding = self._sparse_query_fn(
                [query.query_str],
            )
            sparse_top_k = query.sparse_top_k or query.similarity_top_k

            sparse_response = self._client.search_batch(
                collection_name=self.collection_name,
                requests=[
                    rest.SearchRequest(
                        vector=rest.NamedSparseVector(
                            # Dynamically switch between the old and new sparse vector name
                            name=self.sparse_vector_name,
                            vector=rest.SparseVector(
                                indices=sparse_indices[0],
                                values=sparse_embedding[0],
                            ),
                        ),
                        limit=sparse_top_k,
                        filter=query_filter,
                        with_payload=True,
                    ),
                ],
            )
            return self.parse_to_query_result(sparse_response[0])

        elif self.enable_hybrid:
            # search for dense vectors only
            response = self._client.search_batch(
                collection_name=self.collection_name,
                requests=[
                    rest.SearchRequest(
                        vector=rest.NamedVector(
                            name=DENSE_VECTOR_NAME,
                            vector=query_embedding,
                        ),
                        limit=query.similarity_top_k,
                        filter=query_filter,
                        with_payload=True,
                    ),
                ],
            )

            return self.parse_to_query_result(response[0])
        else:
            response = self._client.search(
                collection_name=self.collection_name,
                query_vector=query_embedding,
                limit=query.similarity_top_k,
                query_filter=query_filter,
            )
            return self.parse_to_query_result(response)

    async def aquery(
        self, query: VectorStoreQuery, **kwargs: Any
    ) -> VectorStoreQueryResult:
        """
        Asynchronous method to query index for top k most similar nodes.

        Args:
            query (VectorStoreQuery): query
        """
        query_embedding = cast(List[float], query.query_embedding)

        #  NOTE: users can pass in qdrant_filters (nested/complicated filters) to override the default MetadataFilters
        qdrant_filters = kwargs.get("qdrant_filters")
        if qdrant_filters is not None:
            query_filter = qdrant_filters
        else:
            # build metadata filters
            query_filter = cast(Filter, self._build_query_filter(query))

        if query.mode == VectorStoreQueryMode.HYBRID and not self.enable_hybrid:
            raise ValueError(
                "Hybrid search is not enabled. Please build the query with "
                "`enable_hybrid=True` in the constructor."
            )
        elif (
            query.mode == VectorStoreQueryMode.HYBRID
            and self.enable_hybrid
            and self._sparse_query_fn is not None
            and query.query_str is not None
        ):
            sparse_indices, sparse_embedding = self._sparse_query_fn(
                [query.query_str],
            )
            sparse_top_k = query.sparse_top_k or query.similarity_top_k

            sparse_response = await self._aclient.search_batch(
                collection_name=self.collection_name,
                requests=[
                    rest.SearchRequest(
                        vector=rest.NamedVector(
                            name=DENSE_VECTOR_NAME,
                            vector=query_embedding,
                        ),
                        limit=query.similarity_top_k,
                        filter=query_filter,
                        with_payload=True,
                    ),
                    rest.SearchRequest(
                        vector=rest.NamedSparseVector(
                            # Dynamically switch between the old and new sparse vector name
                            name=self.sparse_vector_name,
                            vector=rest.SparseVector(
                                indices=sparse_indices[0],
                                values=sparse_embedding[0],
                            ),
                        ),
                        limit=sparse_top_k,
                        filter=query_filter,
                        with_payload=True,
                    ),
                ],
            )

            # sanity check
            assert len(sparse_response) == 2
            assert self._hybrid_fusion_fn is not None

            # flatten the response
            return self._hybrid_fusion_fn(
                self.parse_to_query_result(sparse_response[0]),
                self.parse_to_query_result(sparse_response[1]),
                alpha=query.alpha or 0.5,
                # NOTE: use hybrid_top_k if provided, otherwise use similarity_top_k
                top_k=query.hybrid_top_k or query.similarity_top_k,
            )
        elif (
            query.mode == VectorStoreQueryMode.SPARSE
            and self.enable_hybrid
            and self._sparse_query_fn is not None
            and query.query_str is not None
        ):
            sparse_indices, sparse_embedding = self._sparse_query_fn(
                [query.query_str],
            )
            sparse_top_k = query.sparse_top_k or query.similarity_top_k

            sparse_response = await self._aclient.search_batch(
                collection_name=self.collection_name,
                requests=[
                    rest.SearchRequest(
                        vector=rest.NamedSparseVector(
                            # Dynamically switch between the old and new sparse vector name
                            name=self.sparse_vector_name,
                            vector=rest.SparseVector(
                                indices=sparse_indices[0],
                                values=sparse_embedding[0],
                            ),
                        ),
                        limit=sparse_top_k,
                        filter=query_filter,
                        with_payload=True,
                    ),
                ],
            )
            return self.parse_to_query_result(sparse_response[0])
        elif self.enable_hybrid:
            # search for dense vectors only
            response = await self._aclient.search_batch(
                collection_name=self.collection_name,
                requests=[
                    rest.SearchRequest(
                        vector=rest.NamedVector(
                            name=DENSE_VECTOR_NAME,
                            vector=query_embedding,
                        ),
                        limit=query.similarity_top_k,
                        filter=query_filter,
                        with_payload=True,
                    ),
                ],
            )

            return self.parse_to_query_result(response[0])
        else:
            response = await self._aclient.search(
                collection_name=self.collection_name,
                query_vector=query_embedding,
                limit=query.similarity_top_k,
                query_filter=query_filter,
            )

            return self.parse_to_query_result(response)

    def parse_to_query_result(self, response: List[Any]) -> VectorStoreQueryResult:
        """
        Convert vector store response to VectorStoreQueryResult.

        Args:
            response: List[Any]: List of results returned from the vector store.
        """
        nodes = []
        similarities = []
        ids = []

        for point in response:
            payload = cast(Payload, point.payload)
            try:
                node = metadata_dict_to_node(payload)
            except Exception:
                # NOTE: deprecated legacy logic for backward compatibility
                logger.debug("Failed to parse Node metadata, fallback to legacy logic.")
                metadata, node_info, relationships = legacy_metadata_dict_to_node(
                    payload
                )

                node = TextNode(
                    id_=str(point.id),
                    text=payload.get("text"),
                    metadata=metadata,
                    start_char_idx=node_info.get("start", None),
                    end_char_idx=node_info.get("end", None),
                    relationships=relationships,
                )
            nodes.append(node)
            similarities.append(point.score)
            ids.append(str(point.id))

        return VectorStoreQueryResult(nodes=nodes, similarities=similarities, ids=ids)

    def _build_query_filter(self, query: VectorStoreQuery) -> Optional[Any]:
        if not query.doc_ids and not query.query_str:
            return None

        must_conditions = []

        if query.doc_ids:
            must_conditions.append(
                FieldCondition(
                    key="doc_id",
                    match=MatchAny(any=query.doc_ids),
                )
            )

        # Point id is a “service” id, it is not stored in payload. There is ‘HasId’ condition to filter by point id
        # https://qdrant.tech/documentation/concepts/filtering/#has-id
        if query.node_ids:
            must_conditions.append(
                HasIdCondition(has_id=query.node_ids),
            )

        # Qdrant does not use the query.query_str property for the filtering. Full-text
        # filtering cannot handle longer queries and can effectively filter our all the
        # nodes. See: https://github.com/jerryjliu/llama_index/pull/1181

        if query.filters is None:
            return Filter(must=must_conditions)

        for subfilter in query.filters.filters:
            # only for exact match
            if not subfilter.operator or subfilter.operator == "==":
                if isinstance(subfilter.value, float):
                    must_conditions.append(
                        FieldCondition(
                            key=subfilter.key,
                            range=Range(
                                gte=subfilter.value,
                                lte=subfilter.value,
                            ),
                        )
                    )
                else:
                    must_conditions.append(
                        FieldCondition(
                            key=subfilter.key,
                            match=MatchValue(value=subfilter.value),
                        )
                    )
            elif subfilter.operator == "<":
                must_conditions.append(
                    FieldCondition(
                        key=subfilter.key,
                        range=Range(lt=subfilter.value),
                    )
                )
            elif subfilter.operator == ">":
                must_conditions.append(
                    FieldCondition(
                        key=subfilter.key,
                        range=Range(gt=subfilter.value),
                    )
                )
            elif subfilter.operator == ">=":
                must_conditions.append(
                    FieldCondition(
                        key=subfilter.key,
                        range=Range(gte=subfilter.value),
                    )
                )
            elif subfilter.operator == "<=":
                must_conditions.append(
                    FieldCondition(
                        key=subfilter.key,
                        range=Range(lte=subfilter.value),
                    )
                )
            elif subfilter.operator == "text_match":
                must_conditions.append(
                    FieldCondition(
                        key=subfilter.key,
                        match=MatchText(text=subfilter.value),
                    )
                )
            elif subfilter.operator == "!=":
                must_conditions.append(
                    FieldCondition(
                        key=subfilter.key,
                        match=MatchExcept(**{"except": [subfilter.value]}),
                    )
                )
            elif subfilter.operator == "in":
                # match any of the values
                # https://qdrant.tech/documentation/concepts/filtering/#match-any
                must_conditions.append(
                    FieldCondition(
                        key=subfilter.key,
                        match=MatchAny(any=str(subfilter.value).split(",")),
                    )
                )

        return Filter(must=must_conditions)

    def use_old_sparse_encoder(self, collection_name: str) -> bool:
        collection_info = self.client.get_collection(collection_name)
        return (
            self._collection_exists(collection_name)
            and collection_info.config.params.sparse_vectors is not None
            and SPARSE_VECTOR_NAME_OLD in collection_info.config.params.sparse_vectors
        )

    @property
    def sparse_vector_name(self) -> str:
        return (
            SPARSE_VECTOR_NAME_OLD
            if self.use_old_sparse_encoder(self.collection_name)
            else SPARSE_VECTOR_NAME
        )

    def get_default_sparse_doc_encoder(
        self, collection_name: str
    ) -> SparseEncoderCallable:
        if self.use_old_sparse_encoder(collection_name):
            return default_sparse_encoder("naver/efficient-splade-VI-BT-large-doc")

        return fastembed_sparse_encoder(model_name="prithvida/Splade_PP_en_v1")

    def get_default_sparse_query_encoder(
        self, collection_name: str
    ) -> SparseEncoderCallable:
        if self.use_old_sparse_encoder(collection_name):
            return default_sparse_encoder("naver/efficient-splade-VI-BT-large-query")

        return fastembed_sparse_encoder(model_name="prithvida/Splade_PP_en_v1")

client property #

client: Any

Return the Qdrant client.

add #

add(nodes: List[BaseNode], **add_kwargs: Any) -> List[str]

Add nodes to index.

Parameters:

Name Type Description Default
nodes List[BaseNode]

List[BaseNode]: list of nodes with embeddings

required
Source code in llama-index-integrations/vector_stores/llama-index-vector-stores-qdrant/llama_index/vector_stores/qdrant/base.py
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
def add(self, nodes: List[BaseNode], **add_kwargs: Any) -> List[str]:
    """
    Add nodes to index.

    Args:
        nodes: List[BaseNode]: list of nodes with embeddings

    """
    if len(nodes) > 0 and not self._collection_initialized:
        self._create_collection(
            collection_name=self.collection_name,
            vector_size=len(nodes[0].get_embedding()),
        )

    points, ids = self._build_points(nodes)

    self._client.upload_points(
        collection_name=self.collection_name,
        points=points,
        batch_size=self.batch_size,
        parallel=self.parallel,
        max_retries=self.max_retries,
        wait=True,
    )

    return ids

async_add async #

async_add(nodes: List[BaseNode], **kwargs: Any) -> List[str]

Asynchronous method to add nodes to Qdrant index.

Parameters:

Name Type Description Default
nodes List[BaseNode]

List[BaseNode]: List of nodes with embeddings.

required

Returns:

Type Description
List[str]

List of node IDs that were added to the index.

Raises:

Type Description
ValueError

If trying to using async methods without aclient

Source code in llama-index-integrations/vector_stores/llama-index-vector-stores-qdrant/llama_index/vector_stores/qdrant/base.py
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
async def async_add(self, nodes: List[BaseNode], **kwargs: Any) -> List[str]:
    """
    Asynchronous method to add nodes to Qdrant index.

    Args:
        nodes: List[BaseNode]: List of nodes with embeddings.

    Returns:
        List of node IDs that were added to the index.

    Raises:
        ValueError: If trying to using async methods without aclient
    """
    collection_initialized = await self._acollection_exists(self.collection_name)

    if len(nodes) > 0 and not collection_initialized:
        await self._acreate_collection(
            collection_name=self.collection_name,
            vector_size=len(nodes[0].get_embedding()),
        )

    points, ids = self._build_points(nodes)

    await self._aclient.upload_points(
        collection_name=self.collection_name,
        points=points,
        batch_size=self.batch_size,
        parallel=self.parallel,
        max_retries=self.max_retries,
        wait=True,
    )

    return ids

delete #

delete(ref_doc_id: str, **delete_kwargs: Any) -> None

Delete nodes using with ref_doc_id.

Parameters:

Name Type Description Default
ref_doc_id str

The doc_id of the document to delete.

required
Source code in llama-index-integrations/vector_stores/llama-index-vector-stores-qdrant/llama_index/vector_stores/qdrant/base.py
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
def delete(self, ref_doc_id: str, **delete_kwargs: Any) -> None:
    """
    Delete nodes using with ref_doc_id.

    Args:
        ref_doc_id (str): The doc_id of the document to delete.

    """
    self._client.delete(
        collection_name=self.collection_name,
        points_selector=rest.Filter(
            must=[
                rest.FieldCondition(
                    key="doc_id", match=rest.MatchValue(value=ref_doc_id)
                )
            ]
        ),
    )

adelete async #

adelete(ref_doc_id: str, **delete_kwargs: Any) -> None

Asynchronous method to delete nodes using with ref_doc_id.

Parameters:

Name Type Description Default
ref_doc_id str

The doc_id of the document to delete.

required
Source code in llama-index-integrations/vector_stores/llama-index-vector-stores-qdrant/llama_index/vector_stores/qdrant/base.py
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
async def adelete(self, ref_doc_id: str, **delete_kwargs: Any) -> None:
    """
    Asynchronous method to delete nodes using with ref_doc_id.

    Args:
        ref_doc_id (str): The doc_id of the document to delete.

    """
    await self._aclient.delete(
        collection_name=self.collection_name,
        points_selector=rest.Filter(
            must=[
                rest.FieldCondition(
                    key="doc_id", match=rest.MatchValue(value=ref_doc_id)
                )
            ]
        ),
    )

query #

query(query: VectorStoreQuery, **kwargs: Any) -> VectorStoreQueryResult

Query index for top k most similar nodes.

Parameters:

Name Type Description Default
query VectorStoreQuery

query

required
Source code in llama-index-integrations/vector_stores/llama-index-vector-stores-qdrant/llama_index/vector_stores/qdrant/base.py
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
def query(
    self,
    query: VectorStoreQuery,
    **kwargs: Any,
) -> VectorStoreQueryResult:
    """
    Query index for top k most similar nodes.

    Args:
        query (VectorStoreQuery): query
    """
    query_embedding = cast(List[float], query.query_embedding)
    #  NOTE: users can pass in qdrant_filters (nested/complicated filters) to override the default MetadataFilters
    qdrant_filters = kwargs.get("qdrant_filters")
    if qdrant_filters is not None:
        query_filter = qdrant_filters
    else:
        query_filter = cast(Filter, self._build_query_filter(query))

    if query.mode == VectorStoreQueryMode.HYBRID and not self.enable_hybrid:
        raise ValueError(
            "Hybrid search is not enabled. Please build the query with "
            "`enable_hybrid=True` in the constructor."
        )
    elif (
        query.mode == VectorStoreQueryMode.HYBRID
        and self.enable_hybrid
        and self._sparse_query_fn is not None
        and query.query_str is not None
    ):
        sparse_indices, sparse_embedding = self._sparse_query_fn(
            [query.query_str],
        )
        sparse_top_k = query.sparse_top_k or query.similarity_top_k

        sparse_response = self._client.search_batch(
            collection_name=self.collection_name,
            requests=[
                rest.SearchRequest(
                    vector=rest.NamedVector(
                        name=DENSE_VECTOR_NAME,
                        vector=query_embedding,
                    ),
                    limit=query.similarity_top_k,
                    filter=query_filter,
                    with_payload=True,
                ),
                rest.SearchRequest(
                    vector=rest.NamedSparseVector(
                        # Dynamically switch between the old and new sparse vector name
                        name=self.sparse_vector_name,
                        vector=rest.SparseVector(
                            indices=sparse_indices[0],
                            values=sparse_embedding[0],
                        ),
                    ),
                    limit=sparse_top_k,
                    filter=query_filter,
                    with_payload=True,
                ),
            ],
        )

        # sanity check
        assert len(sparse_response) == 2
        assert self._hybrid_fusion_fn is not None

        # flatten the response
        return self._hybrid_fusion_fn(
            self.parse_to_query_result(sparse_response[0]),
            self.parse_to_query_result(sparse_response[1]),
            # NOTE: only for hybrid search (0 for sparse search, 1 for dense search)
            alpha=query.alpha or 0.5,
            # NOTE: use hybrid_top_k if provided, otherwise use similarity_top_k
            top_k=query.hybrid_top_k or query.similarity_top_k,
        )
    elif (
        query.mode == VectorStoreQueryMode.SPARSE
        and self.enable_hybrid
        and self._sparse_query_fn is not None
        and query.query_str is not None
    ):
        sparse_indices, sparse_embedding = self._sparse_query_fn(
            [query.query_str],
        )
        sparse_top_k = query.sparse_top_k or query.similarity_top_k

        sparse_response = self._client.search_batch(
            collection_name=self.collection_name,
            requests=[
                rest.SearchRequest(
                    vector=rest.NamedSparseVector(
                        # Dynamically switch between the old and new sparse vector name
                        name=self.sparse_vector_name,
                        vector=rest.SparseVector(
                            indices=sparse_indices[0],
                            values=sparse_embedding[0],
                        ),
                    ),
                    limit=sparse_top_k,
                    filter=query_filter,
                    with_payload=True,
                ),
            ],
        )
        return self.parse_to_query_result(sparse_response[0])

    elif self.enable_hybrid:
        # search for dense vectors only
        response = self._client.search_batch(
            collection_name=self.collection_name,
            requests=[
                rest.SearchRequest(
                    vector=rest.NamedVector(
                        name=DENSE_VECTOR_NAME,
                        vector=query_embedding,
                    ),
                    limit=query.similarity_top_k,
                    filter=query_filter,
                    with_payload=True,
                ),
            ],
        )

        return self.parse_to_query_result(response[0])
    else:
        response = self._client.search(
            collection_name=self.collection_name,
            query_vector=query_embedding,
            limit=query.similarity_top_k,
            query_filter=query_filter,
        )
        return self.parse_to_query_result(response)

aquery async #

aquery(query: VectorStoreQuery, **kwargs: Any) -> VectorStoreQueryResult

Asynchronous method to query index for top k most similar nodes.

Parameters:

Name Type Description Default
query VectorStoreQuery

query

required
Source code in llama-index-integrations/vector_stores/llama-index-vector-stores-qdrant/llama_index/vector_stores/qdrant/base.py
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
async def aquery(
    self, query: VectorStoreQuery, **kwargs: Any
) -> VectorStoreQueryResult:
    """
    Asynchronous method to query index for top k most similar nodes.

    Args:
        query (VectorStoreQuery): query
    """
    query_embedding = cast(List[float], query.query_embedding)

    #  NOTE: users can pass in qdrant_filters (nested/complicated filters) to override the default MetadataFilters
    qdrant_filters = kwargs.get("qdrant_filters")
    if qdrant_filters is not None:
        query_filter = qdrant_filters
    else:
        # build metadata filters
        query_filter = cast(Filter, self._build_query_filter(query))

    if query.mode == VectorStoreQueryMode.HYBRID and not self.enable_hybrid:
        raise ValueError(
            "Hybrid search is not enabled. Please build the query with "
            "`enable_hybrid=True` in the constructor."
        )
    elif (
        query.mode == VectorStoreQueryMode.HYBRID
        and self.enable_hybrid
        and self._sparse_query_fn is not None
        and query.query_str is not None
    ):
        sparse_indices, sparse_embedding = self._sparse_query_fn(
            [query.query_str],
        )
        sparse_top_k = query.sparse_top_k or query.similarity_top_k

        sparse_response = await self._aclient.search_batch(
            collection_name=self.collection_name,
            requests=[
                rest.SearchRequest(
                    vector=rest.NamedVector(
                        name=DENSE_VECTOR_NAME,
                        vector=query_embedding,
                    ),
                    limit=query.similarity_top_k,
                    filter=query_filter,
                    with_payload=True,
                ),
                rest.SearchRequest(
                    vector=rest.NamedSparseVector(
                        # Dynamically switch between the old and new sparse vector name
                        name=self.sparse_vector_name,
                        vector=rest.SparseVector(
                            indices=sparse_indices[0],
                            values=sparse_embedding[0],
                        ),
                    ),
                    limit=sparse_top_k,
                    filter=query_filter,
                    with_payload=True,
                ),
            ],
        )

        # sanity check
        assert len(sparse_response) == 2
        assert self._hybrid_fusion_fn is not None

        # flatten the response
        return self._hybrid_fusion_fn(
            self.parse_to_query_result(sparse_response[0]),
            self.parse_to_query_result(sparse_response[1]),
            alpha=query.alpha or 0.5,
            # NOTE: use hybrid_top_k if provided, otherwise use similarity_top_k
            top_k=query.hybrid_top_k or query.similarity_top_k,
        )
    elif (
        query.mode == VectorStoreQueryMode.SPARSE
        and self.enable_hybrid
        and self._sparse_query_fn is not None
        and query.query_str is not None
    ):
        sparse_indices, sparse_embedding = self._sparse_query_fn(
            [query.query_str],
        )
        sparse_top_k = query.sparse_top_k or query.similarity_top_k

        sparse_response = await self._aclient.search_batch(
            collection_name=self.collection_name,
            requests=[
                rest.SearchRequest(
                    vector=rest.NamedSparseVector(
                        # Dynamically switch between the old and new sparse vector name
                        name=self.sparse_vector_name,
                        vector=rest.SparseVector(
                            indices=sparse_indices[0],
                            values=sparse_embedding[0],
                        ),
                    ),
                    limit=sparse_top_k,
                    filter=query_filter,
                    with_payload=True,
                ),
            ],
        )
        return self.parse_to_query_result(sparse_response[0])
    elif self.enable_hybrid:
        # search for dense vectors only
        response = await self._aclient.search_batch(
            collection_name=self.collection_name,
            requests=[
                rest.SearchRequest(
                    vector=rest.NamedVector(
                        name=DENSE_VECTOR_NAME,
                        vector=query_embedding,
                    ),
                    limit=query.similarity_top_k,
                    filter=query_filter,
                    with_payload=True,
                ),
            ],
        )

        return self.parse_to_query_result(response[0])
    else:
        response = await self._aclient.search(
            collection_name=self.collection_name,
            query_vector=query_embedding,
            limit=query.similarity_top_k,
            query_filter=query_filter,
        )

        return self.parse_to_query_result(response)

parse_to_query_result #

parse_to_query_result(response: List[Any]) -> VectorStoreQueryResult

Convert vector store response to VectorStoreQueryResult.

Parameters:

Name Type Description Default
response List[Any]

List[Any]: List of results returned from the vector store.

required
Source code in llama-index-integrations/vector_stores/llama-index-vector-stores-qdrant/llama_index/vector_stores/qdrant/base.py
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
def parse_to_query_result(self, response: List[Any]) -> VectorStoreQueryResult:
    """
    Convert vector store response to VectorStoreQueryResult.

    Args:
        response: List[Any]: List of results returned from the vector store.
    """
    nodes = []
    similarities = []
    ids = []

    for point in response:
        payload = cast(Payload, point.payload)
        try:
            node = metadata_dict_to_node(payload)
        except Exception:
            # NOTE: deprecated legacy logic for backward compatibility
            logger.debug("Failed to parse Node metadata, fallback to legacy logic.")
            metadata, node_info, relationships = legacy_metadata_dict_to_node(
                payload
            )

            node = TextNode(
                id_=str(point.id),
                text=payload.get("text"),
                metadata=metadata,
                start_char_idx=node_info.get("start", None),
                end_char_idx=node_info.get("end", None),
                relationships=relationships,
            )
        nodes.append(node)
        similarities.append(point.score)
        ids.append(str(point.id))

    return VectorStoreQueryResult(nodes=nodes, similarities=similarities, ids=ids)