beir-qdrant

BEIR是一个包含各种IR任务的异构基准。本项目将BEIR与Qdrant（一个向量搜索引擎）集成。

安装

pip install beir-qdrant

快速示例

以下示例演示了如何使用BEIR与Qdrant密集搜索。示例使用SciFact数据集和Sentence Transformers中的all-MiniLM-L6-v2模型生成密集嵌入。

from beir import util
from beir.datasets.data_loader import GenericDataLoader
from beir.retrieval.evaluation import EvaluateRetrieval
from qdrant_client import QdrantClient

from beir_qdrant.retrieval.models.fastembed import DenseFastEmbedModelAdapter
from beir_qdrant.retrieval.search.dense import DenseQdrantSearch

# Download and load the dataset
dataset = "scifact"
url = "https://public.ukp.informatik.tu-darmstadt.de/thakur/BEIR/datasets/{}.zip".format(dataset)
data_path = util.download_and_unzip(url, "datasets")
corpus, queries, qrels = GenericDataLoader(data_folder=data_path).load(split="test")

# Connect to Qdrant running on localhost
qdrant_client = QdrantClient("https://:6333")

# Create the retriever and evaluate it on the test set using
# one of the sentence-transformers models available in FastEmbed
model = DenseQdrantSearch(
    qdrant_client,
    model=DenseFastEmbedModelAdapter(
        model_name="sentence-transformers/all-MiniLM-L6-v2"
    ),
    collection_name="scifact-all-MiniLM-L6-v2",
    initialize=True,
)
retriever = EvaluateRetrieval(model)
results = retriever.retrieve(corpus, queries)

ndcg, _map, recall, precision = retriever.evaluate(qrels, results, retriever.k_values)

上述示例演示了如何使用密集嵌入，但更改搜索模式与更改模型实现一样简单。

支持的模式

Qdrant支持不同的搜索模式，包括

• 密集搜索：beir_qdrant.retrieval.search.dense.DenseQdrantSearch
• 稀疏搜索：beir_qdrant.retrieval.search.sparse.SparseQdrantSearch
• 多向量搜索：beir_qdrant.retrieval.search.multi_vector.MultiVectorQdrantSearch
• 与RRF的混合搜索：beir_qdrant.retrieval.search.hybrid.RRFHybridQdrantSearch