Spaces:
Running
Running
File size: 39,288 Bytes
6912ad8 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 |
from __future__ import annotations
import uuid
import numpy as np
from collections.abc import Callable
from enum import Enum
from itertools import islice
from operator import itemgetter
from typing import (
TYPE_CHECKING,
Any,
)
from langchain_core.documents import Document
from fastembed import TextEmbedding
from langchain_core.vectorstores import VectorStore
from qdrant_client import QdrantClient, models
if TYPE_CHECKING:
from collections.abc import Generator, Iterable, Sequence
from qdrant_sparse_embeddings import SparseEmbeddings
class QdrantVectorStoreError(Exception):
"""`QdrantVectorStore` related exceptions."""
class RetrievalMode(str, Enum):
"""Modes for retrieving vectors from Qdrant."""
DENSE = "dense"
SPARSE = "sparse"
HYBRID = "hybrid"
class QdrantVectorStore(VectorStore):
CONTENT_KEY: str = "page_content"
METADATA_KEY: str = "metadata"
VECTOR_NAME: str = ""
SPARSE_VECTOR_NAME: str = "test_collection"
def __init__(
self,
client: QdrantClient,
collection_name: str,
embedding: TextEmbedding | None = None,
retrieval_mode: RetrievalMode = RetrievalMode.DENSE,
vector_name: str = VECTOR_NAME,
content_payload_key: str = CONTENT_KEY,
metadata_payload_key: str = METADATA_KEY,
distance: models.Distance = models.Distance.COSINE,
sparse_embedding: SparseEmbeddings | None = None,
sparse_vector_name: str = SPARSE_VECTOR_NAME,
validate_embeddings: bool = True,
validate_collection_config: bool = True,
) -> None:
"""Initialize a new instance of `QdrantVectorStore`.
```python
qdrant = QdrantVectorStore(
client=client,
collection_name="my-collection",
embedding=OpenAIEmbeddings(),
retrieval_mode=RetrievalMode.HYBRID,
sparse_embedding=FastEmbedSparse(),
)
```
"""
if validate_embeddings:
self._validate_embeddings(retrieval_mode, embedding, sparse_embedding)
if validate_collection_config:
self._validate_collection_config(
client,
collection_name,
retrieval_mode,
vector_name,
sparse_vector_name,
distance,
embedding,
)
self._client = client
self.collection_name = collection_name
self._embeddings = embedding
self.retrieval_mode = retrieval_mode
self.vector_name = vector_name
self.content_payload_key = content_payload_key
self.metadata_payload_key = metadata_payload_key
self.distance = distance
self._sparse_embeddings = sparse_embedding
self.sparse_vector_name = sparse_vector_name
@property
def client(self) -> QdrantClient:
"""Get the Qdrant client instance that is being used.
Returns:
QdrantClient: An instance of `QdrantClient`.
"""
return self._client
@property
def embeddings(self) -> TextEmbedding | None:
"""Get the dense embeddings instance that is being used.
Returns:
Embeddings: An instance of `TextEmbedding`, or None for SPARSE mode.
"""
return self._embeddings
def _get_retriever_tags(self) -> list[str]:
"""Get tags for retriever.
Override the base class method to handle SPARSE mode where embeddings can be
None. In SPARSE mode, embeddings is None, so we don't include embeddings class
name in tags. In DENSE/HYBRID modes, embeddings is not None, so we include
embeddings class name.
"""
tags = [self.__class__.__name__]
# Handle different retrieval modes
if self.retrieval_mode == RetrievalMode.SPARSE:
# SPARSE mode: no dense embeddings, so no embeddings class name in tags
pass
# DENSE/HYBRID modes: include embeddings class name if available
elif self.embeddings is not None:
tags.append(self.embeddings.__class__.__name__)
return tags
def _require_embeddings(self, operation: str) -> TextEmbedding:
"""Require embeddings for operations that need them.
Args:
operation: Description of the operation requiring embeddings.
Returns:
The embeddings instance.
Raises:
ValueError: If embeddings are None and required for the operation.
"""
if self.embeddings is None:
msg = f"Embeddings are required for {operation}"
raise ValueError(msg)
return self.embeddings
@property
def sparse_embeddings(self) -> SparseEmbeddings:
"""Get the sparse embeddings instance that is being used.
Raises:
ValueError: If sparse embeddings are `None`.
Returns:
SparseEmbeddings: An instance of `SparseEmbeddings`.
"""
if self._sparse_embeddings is None:
msg = (
"Sparse embeddings are `None`. "
"Please set using the `sparse_embedding` parameter."
)
raise ValueError(msg)
return self._sparse_embeddings
@classmethod
def from_texts(
cls: type[QdrantVectorStore],
texts: list[str],
embedding: TextEmbedding | None = None,
metadatas: list[dict] | None = None,
ids: Sequence[str | int] | None = None,
collection_name: str | None = None,
location: str | None = None,
url: str | None = None,
port: int | None = 6333,
grpc_port: int = 6334,
prefer_grpc: bool = False,
https: bool | None = None,
api_key: str | None = None,
prefix: str | None = None,
timeout: int | None = None,
host: str | None = None,
path: str | None = None,
distance: models.Distance = models.Distance.COSINE,
content_payload_key: str = CONTENT_KEY,
metadata_payload_key: str = METADATA_KEY,
vector_name: str = VECTOR_NAME,
retrieval_mode: RetrievalMode = RetrievalMode.DENSE,
sparse_embedding: SparseEmbeddings | None = None,
sparse_vector_name: str = SPARSE_VECTOR_NAME,
collection_create_options: dict[str, Any] | None = None,
vector_params: dict[str, Any] | None = None,
sparse_vector_params: dict[str, Any] | None = None,
batch_size: int = 64,
force_recreate: bool = False,
validate_embeddings: bool = True,
validate_collection_config: bool = True,
**kwargs: Any,
) -> QdrantVectorStore:
"""
Construct an instance of `QdrantVectorStore` from a list of texts.
"""
if sparse_vector_params is None:
sparse_vector_params = {}
if vector_params is None:
vector_params = {}
if collection_create_options is None:
collection_create_options = {}
client_options = {
"location": location,
"url": url,
"port": port,
"grpc_port": grpc_port,
"prefer_grpc": prefer_grpc,
"https": https,
"api_key": api_key,
"prefix": prefix,
"timeout": timeout,
"host": host,
"path": path,
**kwargs,
}
qdrant = cls.construct_instance(
embedding,
retrieval_mode,
sparse_embedding,
client_options,
collection_name,
distance,
content_payload_key,
metadata_payload_key,
vector_name,
sparse_vector_name,
force_recreate,
collection_create_options,
vector_params,
sparse_vector_params,
validate_embeddings,
validate_collection_config,
)
qdrant.add_texts(texts, metadatas, ids, batch_size)
return qdrant
def add_documents(
self,
documents: Sequence[Document],
ids: Sequence[str | int] | None = None,
batch_size: int = 64,
**kwargs: Any,
) -> list[str | int]:
texts = [doc.page_content for doc in documents]
metadatas = [doc.metadata if doc.metadata is not None else {} for doc in documents]
return self.add_texts(
texts=texts,
metadatas=metadatas,
ids=ids,
batch_size=batch_size,
**kwargs,
)
@classmethod
def from_documents(
cls,
documents: list[Document],
embedding: TextEmbedding,
**kwargs: Any,
):
"""Return `VectorStore` initialized from documents and embeddings.
Args:
documents: List of `Document` objects to add to the `VectorStore`.
embedding: Embedding function to use.
**kwargs: Additional keyword arguments.
Returns:
`VectorStore` initialized from documents and embeddings.
"""
texts = [d.page_content for d in documents]
metadatas = [d.metadata for d in documents]
if "ids" not in kwargs:
ids = [doc.metadata.get("chunk_id") for doc in documents]
# If there's at least one valid ID, we'll assume that IDs
# should be used.
if any(ids):
kwargs["ids"] = ids
return cls.from_texts(texts, embedding, metadatas=metadatas, **kwargs)
@classmethod
def from_existing_collection(
cls: type[QdrantVectorStore],
collection_name: str,
embedding: TextEmbedding | None = None,
retrieval_mode: RetrievalMode = RetrievalMode.DENSE,
location: str | None = None,
url: str | None = None,
port: int | None = 6333,
grpc_port: int = 6334,
prefer_grpc: bool = False,
https: bool | None = None,
api_key: str | None = None,
prefix: str | None = None,
timeout: int | None = None,
host: str | None = None,
path: str | None = None,
distance: models.Distance = models.Distance.COSINE,
content_payload_key: str = CONTENT_KEY,
metadata_payload_key: str = METADATA_KEY,
vector_name: str = VECTOR_NAME,
sparse_vector_name: str = SPARSE_VECTOR_NAME,
sparse_embedding: SparseEmbeddings | None = None,
validate_embeddings: bool = True,
validate_collection_config: bool = True,
**kwargs: Any,
) -> QdrantVectorStore:
"""Construct `QdrantVectorStore` from existing collection without adding data.
Returns:
QdrantVectorStore: A new instance of `QdrantVectorStore`.
"""
client = QdrantClient(
location=location,
url=url,
port=port,
grpc_port=grpc_port,
prefer_grpc=prefer_grpc,
https=https,
api_key=api_key,
prefix=prefix,
timeout=timeout,
host=host,
path=path,
**kwargs,
)
return cls(
client=client,
collection_name=collection_name,
embedding=embedding,
retrieval_mode=retrieval_mode,
content_payload_key=content_payload_key,
metadata_payload_key=metadata_payload_key,
distance=distance,
vector_name=vector_name,
sparse_embedding=sparse_embedding,
sparse_vector_name=sparse_vector_name,
validate_embeddings=validate_embeddings,
validate_collection_config=validate_collection_config,
)
def add_texts( # type: ignore[override]
self,
texts: Iterable[str],
metadatas: list[dict] | None = None,
ids: Sequence[str | int] | None = None,
batch_size: int = 64,
**kwargs: Any,
) -> list[str | int]:
"""Add texts with embeddings to the `VectorStore`.
Returns:
List of ids from adding the texts into the `VectorStore`.
"""
added_ids = []
for batch_ids, points in self._generate_batches(
texts, metadatas, ids, batch_size
):
self.client.upsert(
collection_name=self.collection_name, points=points, **kwargs
)
added_ids.extend(batch_ids)
return added_ids
def similarity_search(
self,
query: str,
k: int = 4,
filter: models.Filter | None = None,
search_params: models.SearchParams | None = None,
offset: int = 0,
score_threshold: float | None = None,
consistency: models.ReadConsistency | None = None,
hybrid_fusion: models.FusionQuery | None = None,
**kwargs: Any,
) -> list[Document]:
"""Return docs most similar to query.
Returns:
List of `Document` objects most similar to the query.
"""
results = self.similarity_search_with_score(
query,
k,
filter=filter,
search_params=search_params,
offset=offset,
score_threshold=score_threshold,
consistency=consistency,
hybrid_fusion=hybrid_fusion,
**kwargs,
)
return list(map(itemgetter(0), results))
def similarity_search_with_score(
self,
query: str,
k: int = 4,
filter: models.Filter | None = None,
search_params: models.SearchParams | None = None,
offset: int = 0,
score_threshold: float | None = None,
consistency: models.ReadConsistency | None = None,
hybrid_fusion: models.FusionQuery | None = None,
**kwargs: Any,
) -> list[tuple[Document, float]]:
"""Return docs most similar to query.
Returns:
List of documents most similar to the query text and distance for each.
"""
query_options = {
"collection_name": self.collection_name,
"query_filter": filter,
"search_params": search_params,
"limit": k,
"offset": offset,
"with_payload": True,
"with_vectors": False,
"score_threshold": score_threshold,
"consistency": consistency,
**kwargs,
}
if self.retrieval_mode == RetrievalMode.DENSE:
embeddings = self._require_embeddings("DENSE mode")
query_dense_embedding = list(embeddings.embed(query))[0]
results = self.client.query_points(
query=query_dense_embedding,
using=self.vector_name,
**query_options,
).points
elif self.retrieval_mode == RetrievalMode.SPARSE:
query_sparse_embedding = self.sparse_embeddings.embed_query(query)
results = self.client.query_points(
query=models.SparseVector(
indices=query_sparse_embedding.indices,
values=query_sparse_embedding.values,
),
using=self.sparse_vector_name,
**query_options,
).points
elif self.retrieval_mode == RetrievalMode.HYBRID:
embeddings = self._require_embeddings("HYBRID mode")
query_dense_embedding = list(embeddings.embed(query))[0]
query_sparse_embedding = self.sparse_embeddings.embed_query(query)
results = self.client.query_points(
prefetch=[
models.Prefetch(
using=self.vector_name,
query=query_dense_embedding,
filter=filter,
limit=k,
params=search_params,
),
models.Prefetch(
using=self.sparse_vector_name,
query=models.SparseVector(
indices=query_sparse_embedding.indices,
values=query_sparse_embedding.values,
),
filter=filter,
limit=k,
params=search_params,
),
],
query=hybrid_fusion or models.FusionQuery(fusion=models.Fusion.RRF),
**query_options,
).points
else:
msg = f"Invalid retrieval mode. {self.retrieval_mode}."
raise ValueError(msg)
return [
(
self._document_from_point(
result,
self.collection_name,
self.content_payload_key,
self.metadata_payload_key,
),
result.score,
)
for result in results
]
def max_marginal_relevance_search(
self,
query: str,
k: int = 4,
fetch_k: int = 20,
lambda_mult: float = 0.5,
filter: models.Filter | None = None,
search_params: models.SearchParams | None = None,
score_threshold: float | None = None,
consistency: models.ReadConsistency | None = None,
**kwargs: Any,
) -> list[Document]:
"""Return docs selected using the maximal marginal relevance with dense vectors.
Maximal marginal relevance optimizes for similarity to query AND diversity
among selected documents.
Returns:
List of `Document` objects selected by maximal marginal relevance.
"""
self._validate_collection_for_dense(
self.client,
self.collection_name,
self.vector_name,
self.distance,
self.embeddings,
)
embeddings = self._require_embeddings("max_marginal_relevance_search")
query_embedding = list(embeddings.embed(query))[0]
return self.max_marginal_relevance_search_by_vector(
query_embedding,
k=k,
fetch_k=fetch_k,
lambda_mult=lambda_mult,
filter=filter,
search_params=search_params,
score_threshold=score_threshold,
consistency=consistency,
**kwargs,
)
def max_marginal_relevance_search_by_vector(
self,
embedding: list[float],
k: int = 4,
fetch_k: int = 20,
lambda_mult: float = 0.5,
filter: models.Filter | None = None, # noqa: A002
search_params: models.SearchParams | None = None,
score_threshold: float | None = None,
consistency: models.ReadConsistency | None = None,
**kwargs: Any,
) -> list[Document]:
"""Return docs selected using the maximal marginal relevance with dense vectors.
Maximal marginal relevance optimizes for similarity to query AND diversity
among selected documents.
Returns:
List of `Document` objects selected by maximal marginal relevance.
"""
results = self.max_marginal_relevance_search_with_score_by_vector(
embedding,
k=k,
fetch_k=fetch_k,
lambda_mult=lambda_mult,
filter=filter,
search_params=search_params,
score_threshold=score_threshold,
consistency=consistency,
**kwargs,
)
return list(map(itemgetter(0), results))
def max_marginal_relevance_search_with_score_by_vector(
self,
embedding: list[float],
k: int = 4,
fetch_k: int = 20,
lambda_mult: float = 0.5,
filter: models.Filter | None = None, # noqa: A002
search_params: models.SearchParams | None = None,
score_threshold: float | None = None,
consistency: models.ReadConsistency | None = None,
**kwargs: Any,
) -> list[tuple[Document, float]]:
"""Return docs selected using the maximal marginal relevance.
Maximal marginal relevance optimizes for similarity to query AND diversity
among selected documents.
Returns:
List of `Document` objects selected by maximal marginal relevance and
distance for each.
"""
results = self.client.query_points(
collection_name=self.collection_name,
query=models.NearestQuery(
nearest=embedding,
mmr=models.Mmr(diversity=lambda_mult, candidates_limit=fetch_k),
),
query_filter=filter,
search_params=search_params,
limit=k,
with_payload=True,
with_vectors=True,
score_threshold=score_threshold,
consistency=consistency,
using=self.vector_name,
**kwargs,
).points
return [
(
self._document_from_point(
result,
self.collection_name,
self.content_payload_key,
self.metadata_payload_key,
),
result.score,
)
for result in results
]
def max_marginal_relevance_search_with_score(
self,
query: str,
k: int = 4,
fetch_k: int = 20,
lambda_mult: float = 0.5,
filter: models.Filter | None = None,
search_params: models.SearchParams | None = None,
score_threshold: float | None = None,
consistency: models.ReadConsistency | None = None,
**kwargs: Any,
) -> list[tuple[Document, float]]:
"""Return docs selected using the maximal marginal relevance with dense vectors.
Maximal marginal relevance optimizes for similarity to query AND diversity
among selected documents.
Returns:
List of `Document` objects selected by maximal marginal relevance.
"""
self._validate_collection_for_dense(
self.client,
self.collection_name,
self.vector_name,
self.distance,
self.embeddings,
)
embeddings = self._require_embeddings("max_marginal_relevance_search")
query_embedding = list(embeddings.embed(query))[0]
return self.max_marginal_relevance_search_with_score_by_vector(
query_embedding,
k=k,
fetch_k=fetch_k,
lambda_mult=lambda_mult,
filter=filter,
search_params=search_params,
score_threshold=score_threshold,
consistency=consistency,
**kwargs,
)
# TO-DO
# def delete(
# self,
# ids: list[str | int] | None = None,
# **kwargs: Any,
# ) -> bool | None:
# """Delete documents by their ids.
# Args:
# ids: List of ids to delete.
# **kwargs: Other keyword arguments that subclasses might use.
# Returns:
# True if deletion is successful, `False` otherwise.
# """
# result = self.client.delete(
# collection_name=self.collection_name,
# points_selector=ids,
# )
# return result.status == models.UpdateStatus.COMPLETED
@classmethod
def construct_instance(
cls: type[QdrantVectorStore],
embedding: TextEmbedding | None = None,
retrieval_mode: RetrievalMode = RetrievalMode.DENSE,
sparse_embedding: SparseEmbeddings | None = None,
client_options: dict[str, Any] | None = None,
collection_name: str | None = None,
distance: models.Distance = models.Distance.COSINE,
content_payload_key: str = CONTENT_KEY,
metadata_payload_key: str = METADATA_KEY,
vector_name: str = VECTOR_NAME,
sparse_vector_name: str = SPARSE_VECTOR_NAME,
force_recreate: bool = False,
collection_create_options: dict[str, Any] | None = None,
vector_params: dict[str, Any] | None = None,
sparse_vector_params: dict[str, Any] | None = None,
validate_embeddings: bool = True,
validate_collection_config: bool = True,
) -> QdrantVectorStore:
if sparse_vector_params is None:
sparse_vector_params = {}
if vector_params is None:
vector_params = {}
if collection_create_options is None:
collection_create_options = {}
if client_options is None:
client_options = {}
if validate_embeddings:
cls._validate_embeddings(retrieval_mode, embedding, sparse_embedding)
collection_name = collection_name or uuid.uuid4().hex
client = QdrantClient(**client_options)
collection_exists = client.collection_exists(collection_name)
if collection_exists and force_recreate:
client.delete_collection(collection_name)
collection_exists = False
if collection_exists:
if validate_collection_config:
cls._validate_collection_config(
client,
collection_name,
retrieval_mode,
vector_name,
sparse_vector_name,
distance,
embedding,
)
else:
vectors_config, sparse_vectors_config = {}, {}
if retrieval_mode == RetrievalMode.DENSE:
partial_embeddings = list(embedding.embed(["dummy_text"]))
vector_params["size"] = len(partial_embeddings[0])
vector_params["distance"] = distance
vectors_config = {
vector_name: models.VectorParams(
**vector_params,
)
}
elif retrieval_mode == RetrievalMode.SPARSE:
sparse_vectors_config = {
sparse_vector_name: models.SparseVectorParams(
**sparse_vector_params
)
}
elif retrieval_mode == RetrievalMode.HYBRID:
partial_embeddings = list(embedding.embed(["dummy_text"]))
vector_params["size"] = len(partial_embeddings[0])
vector_params["distance"] = distance
vectors_config = {
vector_name: models.VectorParams(
**vector_params,
)
}
sparse_vectors_config = {
sparse_vector_name: models.SparseVectorParams(
**sparse_vector_params
)
}
collection_create_options["collection_name"] = collection_name
collection_create_options["vectors_config"] = vectors_config
collection_create_options["sparse_vectors_config"] = sparse_vectors_config
client.create_collection(**collection_create_options)
return cls(
client=client,
collection_name=collection_name,
embedding=embedding,
retrieval_mode=retrieval_mode,
content_payload_key=content_payload_key,
metadata_payload_key=metadata_payload_key,
distance=distance,
vector_name=vector_name,
sparse_embedding=sparse_embedding,
sparse_vector_name=sparse_vector_name,
validate_embeddings=False,
validate_collection_config=False,
)
@staticmethod
def _cosine_relevance_score_fn(distance: float) -> float:
"""Normalize the distance to a score on a scale `[0, 1]`."""
return (distance + 1.0) / 2.0
def _select_relevance_score_fn(self) -> Callable[[float], float]:
"""Your "correct" relevance function may differ depending on a few things.
Including:
- The distance / similarity metric used by the VectorStore
- The scale of your embeddings (OpenAI's are unit normed. Many others are not!)
- Embedding dimensionality
- etc.
"""
if self.distance == models.Distance.COSINE:
return self._cosine_relevance_score_fn
if self.distance == models.Distance.DOT:
return self._max_inner_product_relevance_score_fn
if self.distance == models.Distance.EUCLID:
return self._euclidean_relevance_score_fn
msg = "Unknown distance strategy, must be COSINE, DOT, or EUCLID."
raise ValueError(msg)
@classmethod
def _document_from_point(
cls,
scored_point: Any,
collection_name: str,
content_payload_key: str,
metadata_payload_key: str,
) -> Document:
metadata = scored_point.payload.get(metadata_payload_key) or {}
metadata["_id"] = scored_point.id
metadata["_collection_name"] = collection_name
return Document(
page_content=scored_point.payload.get(content_payload_key, ""),
metadata=metadata,
)
def _generate_batches(
self,
texts: Iterable[str],
metadatas: list[dict] | None = None,
ids: Sequence[str | int] | None = None,
batch_size: int = 64,
) -> Generator[tuple[list[str | int], list[models.PointStruct]], Any, None]:
texts_iterator = iter(texts)
metadatas_iterator = iter(metadatas or [])
ids_iterator = iter(ids or [uuid.uuid4().hex for _ in iter(texts)])
while batch_texts := list(islice(texts_iterator, batch_size)):
batch_metadatas = list(islice(metadatas_iterator, batch_size)) or None
batch_ids = list(islice(ids_iterator, batch_size))
points = [
models.PointStruct(
id=point_id,
vector=vector,
payload=payload,
)
for point_id, vector, payload in zip(
batch_ids,
self._build_vectors(batch_texts),
self._build_payloads(
batch_texts,
batch_metadatas,
self.content_payload_key,
self.metadata_payload_key,
),
strict=False,
)
]
yield batch_ids, points
@staticmethod
def _build_payloads(
texts: Iterable[str],
metadatas: list[dict] | None,
content_payload_key: str,
metadata_payload_key: str,
) -> list[dict]:
payloads = []
for i, text in enumerate(texts):
if text is None:
msg = (
"At least one of the texts is None. Please remove it before "
"calling .from_texts or .add_texts."
)
raise ValueError(msg)
metadata = metadatas[i] if metadatas is not None else None
payloads.append(
{
content_payload_key: text,
metadata_payload_key: metadata,
}
)
return payloads
def _build_vectors(
self,
texts: Iterable[str],
) -> list[models.VectorStruct]:
if self.retrieval_mode == RetrievalMode.DENSE:
embeddings = self._require_embeddings("DENSE mode")
batch_embeddings = list(embeddings.embed(list(texts)))
return [
{
self.vector_name: vector,
}
for vector in batch_embeddings
]
if self.retrieval_mode == RetrievalMode.SPARSE:
batch_sparse_embeddings = self.sparse_embeddings.embed_documents(
list(texts)
)
return [
{
self.sparse_vector_name: models.SparseVector(
values=vector.values, indices=vector.indices
)
}
for vector in batch_sparse_embeddings
]
if self.retrieval_mode == RetrievalMode.HYBRID:
embeddings = self._require_embeddings("HYBRID mode")
dense_embeddings = list(embeddings.embed(list(texts)))
sparse_embeddings = self.sparse_embeddings.embed_documents(list(texts))
if len(dense_embeddings) != len(sparse_embeddings):
msg = "Mismatched length between dense and sparse embeddings."
raise ValueError(msg)
return [
{
self.vector_name: dense_vector,
self.sparse_vector_name: models.SparseVector(
values=sparse_vector.values, indices=sparse_vector.indices
),
}
for dense_vector, sparse_vector in zip(
dense_embeddings, sparse_embeddings, strict=False
)
]
msg = f"Unknown retrieval mode. {self.retrieval_mode} to build vectors."
raise ValueError(msg)
@classmethod
def _validate_collection_config(
cls: type[QdrantVectorStore],
client: QdrantClient,
collection_name: str,
retrieval_mode: RetrievalMode,
vector_name: str,
sparse_vector_name: str,
distance: models.Distance,
embedding: TextEmbedding | None,
) -> None:
if retrieval_mode == RetrievalMode.DENSE:
cls._validate_collection_for_dense(
client, collection_name, vector_name, distance, embedding
)
elif retrieval_mode == RetrievalMode.SPARSE:
cls._validate_collection_for_sparse(
client, collection_name, sparse_vector_name
)
elif retrieval_mode == RetrievalMode.HYBRID:
cls._validate_collection_for_dense(
client, collection_name, vector_name, distance, embedding
)
cls._validate_collection_for_sparse(
client, collection_name, sparse_vector_name
)
@classmethod
def _validate_collection_for_dense(
cls: type[QdrantVectorStore],
client: QdrantClient,
collection_name: str,
vector_name: str,
distance: models.Distance,
dense_embeddings: TextEmbedding | list[float] | None,
) -> None:
collection_info = client.get_collection(collection_name=collection_name)
vector_config = collection_info.config.params.vectors
if isinstance(vector_config, dict):
# vector_config is a Dict[str, VectorParams]
if vector_name not in vector_config:
msg = (
f"Existing Qdrant collection {collection_name} does not "
f"contain dense vector named {vector_name}. "
"Did you mean one of the "
f"existing vectors: {', '.join(vector_config.keys())}? "
f"If you want to recreate the collection, set `force_recreate` "
f"parameter to `True`."
)
raise QdrantVectorStoreError(msg)
# Get the VectorParams object for the specified vector_name
vector_config = vector_config[vector_name] # type: ignore[assignment, index]
# vector_config is an instance of VectorParams
# Case of a collection with single/unnamed vector.
elif vector_name != "":
msg = (
f"Existing Qdrant collection {collection_name} is built "
"with unnamed dense vector. "
f"If you want to reuse it, set `vector_name` to ''(empty string)."
f"If you want to recreate the collection, "
"set `force_recreate` to `True`."
)
raise QdrantVectorStoreError(msg)
if vector_config is None:
msg = "VectorParams is None"
raise ValueError(msg)
if isinstance(dense_embeddings, TextEmbedding):
embeddings = list(dense_embeddings.embed(["dummy_text"]))[0]
vector_size = len(embeddings)
elif isinstance(dense_embeddings, list):
vector_size = len(dense_embeddings)
else:
msg = "Invalid `embeddings` type."
raise TypeError(msg)
if vector_config.size != vector_size:
msg = (
f"Existing Qdrant collection is configured for dense vectors with "
f"{vector_config.size} dimensions. "
f"Selected embeddings are {vector_size}-dimensional. "
f"If you want to recreate the collection, set `force_recreate` "
f"parameter to `True`."
)
raise QdrantVectorStoreError(msg)
if vector_config.distance != distance:
msg = (
f"Existing Qdrant collection is configured for "
f"{vector_config.distance.name} similarity, but requested "
f"{distance.upper()}. Please set `distance` parameter to "
f"`{vector_config.distance.name}` if you want to reuse it. "
f"If you want to recreate the collection, set `force_recreate` "
f"parameter to `True`."
)
raise QdrantVectorStoreError(msg)
@classmethod
def _validate_collection_for_sparse(
cls: type[QdrantVectorStore],
client: QdrantClient,
collection_name: str,
sparse_vector_name: str,
) -> None:
collection_info = client.get_collection(collection_name=collection_name)
sparse_vector_config = collection_info.config.params.sparse_vectors
if (
sparse_vector_config is None
or sparse_vector_name not in sparse_vector_config
):
msg = (
f"Existing Qdrant collection {collection_name} does not "
f"contain sparse vectors named {sparse_vector_name}. "
f"If you want to recreate the collection, set `force_recreate` "
f"parameter to `True`."
)
raise QdrantVectorStoreError(msg)
@classmethod
def _validate_embeddings(
cls: type[QdrantVectorStore],
retrieval_mode: RetrievalMode,
embedding: TextEmbedding | None,
sparse_embedding: SparseEmbeddings | None,
) -> None:
if retrieval_mode == RetrievalMode.DENSE and embedding is None:
msg = "'embedding' cannot be None when retrieval mode is 'dense'"
raise ValueError(msg)
if retrieval_mode == RetrievalMode.SPARSE and sparse_embedding is None:
msg = "'sparse_embedding' cannot be None when retrieval mode is 'sparse'"
raise ValueError(msg)
if retrieval_mode == RetrievalMode.HYBRID and any(
[embedding is None, sparse_embedding is None]
):
msg = (
"Both 'embedding' and 'sparse_embedding' cannot be None "
"when retrieval mode is 'hybrid'"
)
raise ValueError(msg) |