Upstash

UpstashVectorStore

基类: BasePydanticVectorStore

Upstash 向量存储。

示例

pip install llama-index-vector-stores-upstash

from llama_index.vector_stores.upstash import UpstashVectorStore

# Create Upstash vector store
upstash_vector_store = UpstashVectorStore(
    url="your_upstash_vector_url",
    token="your_upstash_vector_token",
)

源代码位于 llama-index-integrations/vector_stores/llama-index-vector-stores-upstash/llama_index/vector_stores/upstash/base.py

class UpstashVectorStore(BasePydanticVectorStore):
    """
    Upstash Vector Store.

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

        ```python
        from llama_index.vector_stores.upstash import UpstashVectorStore

        # Create Upstash vector store
        upstash_vector_store = UpstashVectorStore(
            url="your_upstash_vector_url",
            token="your_upstash_vector_token",
        )
        ```

    """

    stores_text: bool = True
    flat_metadata: bool = False

    namespace: str = ""

    batch_size: int
    _index: Index = PrivateAttr()

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

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

    def __init__(
        self,
        url: str,
        token: str,
        batch_size: int = DEFAULT_BATCH_SIZE,
        # Upstash uses ("") as the default namespace, if not provided
        namespace: str = "",
    ) -> None:
        """
        Create a UpstashVectorStore. The index can be created using the Upstash console.

        Args:
            url (String): URL of the Upstash Vector instance, found in the Upstash console.
            token (String): Token for the Upstash Vector Index, found in the Upstash console.
            batch_size (Optional[int]): Batch size for adding nodes to the vector store.

        Raises:
            ImportError: If the upstash-vector python package is not installed.

        """
        super().__init__(batch_size=batch_size, namespace=namespace)
        self._index = Index(url=url, token=token)

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

        Args:
            nodes: List of nodes to add to the vector store.
            add_kwargs: Additional arguments to pass to the add method.

        Returns:
            List of ids of the added nodes.

        """
        ids = []
        vectors = []
        for node_batch in iter_batch(nodes, self.batch_size):
            for node in node_batch:
                metadata_dict = node_to_metadata_dict(node)
                ids.append(node.node_id)
                vectors.append((node.node_id, node.embedding, metadata_dict))

            self.client.upsert(vectors=vectors, namespace=self.namespace)

        return ids

    def delete(self, ref_doc_id: str, **delete_kwargs: Any) -> None:
        """
        Delete node from the vector store.

        Args:
            ref_doc_id: Reference doc id of the node to delete.
            delete_kwargs: Additional arguments to pass to the delete method.

        """
        raise NotImplementedError(
            "Delete is not currently supported, but will be in the future."
        )

    def query(self, query: VectorStoreQuery, **kwargs: Any) -> VectorStoreQueryResult:
        """
        Query the vector store.

        Args:
            query: Query to run against the vector store.
            kwargs: Additional arguments to pass to the query method.

        Returns:
            Query result.

        """
        if query.mode != VectorStoreQueryMode.DEFAULT:
            raise ValueError(f"Query mode {query.mode} not supported")

        # if query.filters:
        #     raise ValueError("Metadata filtering not supported")

        res = self.client.query(
            vector=query.query_embedding,
            top_k=query.similarity_top_k,
            include_vectors=True,
            include_metadata=True,
            filter=_to_upstash_filters(query.filters),
            namespace=self.namespace,
        )

        top_k_nodes = []
        top_k_ids = []
        top_k_scores = []
        for vector in res:
            node = metadata_dict_to_node(vector.metadata)
            node.embedding = vector.vector
            top_k_nodes.append(node)
            top_k_ids.append(vector.id)
            top_k_scores.append(vector.score)

        return VectorStoreQueryResult(
            nodes=top_k_nodes, similarities=top_k_scores, ids=top_k_ids
        )

客户端 property

client: Any

返回 Upstash 客户端。

添加

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

将节点添加到向量存储。

参数

名称	类型	描述	默认值
nodes	List[BaseNode]	要添加到向量存储的节点列表。	必需
add_kwargs	Any	要传递给 add 方法的额外参数。	{}

返回值

类型	描述
List[str]	添加的节点的 ID 列表。

源代码位于 llama-index-integrations/vector_stores/llama-index-vector-stores-upstash/llama_index/vector_stores/upstash/base.py

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

    Args:
        nodes: List of nodes to add to the vector store.
        add_kwargs: Additional arguments to pass to the add method.

    Returns:
        List of ids of the added nodes.

    """
    ids = []
    vectors = []
    for node_batch in iter_batch(nodes, self.batch_size):
        for node in node_batch:
            metadata_dict = node_to_metadata_dict(node)
            ids.append(node.node_id)
            vectors.append((node.node_id, node.embedding, metadata_dict))

        self.client.upsert(vectors=vectors, namespace=self.namespace)

    return ids

删除

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

从向量存储中删除节点。

参数

名称	类型	描述	默认值
ref_doc_id	str	要删除节点的引用文档 ID。	必需
delete_kwargs	Any	要传递给 delete 方法的额外参数。	{}

源代码位于 llama-index-integrations/vector_stores/llama-index-vector-stores-upstash/llama_index/vector_stores/upstash/base.py

def delete(self, ref_doc_id: str, **delete_kwargs: Any) -> None:
    """
    Delete node from the vector store.

    Args:
        ref_doc_id: Reference doc id of the node to delete.
        delete_kwargs: Additional arguments to pass to the delete method.

    """
    raise NotImplementedError(
        "Delete is not currently supported, but will be in the future."
    )

查询

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

查询向量存储。

参数

名称	类型	描述	默认值
查询	VectorStoreQuery	对向量存储运行的查询。	必需
kwargs	Any	要传递给 query 方法的额外参数。	{}

返回值

类型	描述
VectorStoreQueryResult	查询结果。

源代码位于 llama-index-integrations/vector_stores/llama-index-vector-stores-upstash/llama_index/vector_stores/upstash/base.py

def query(self, query: VectorStoreQuery, **kwargs: Any) -> VectorStoreQueryResult:
    """
    Query the vector store.

    Args:
        query: Query to run against the vector store.
        kwargs: Additional arguments to pass to the query method.

    Returns:
        Query result.

    """
    if query.mode != VectorStoreQueryMode.DEFAULT:
        raise ValueError(f"Query mode {query.mode} not supported")

    # if query.filters:
    #     raise ValueError("Metadata filtering not supported")

    res = self.client.query(
        vector=query.query_embedding,
        top_k=query.similarity_top_k,
        include_vectors=True,
        include_metadata=True,
        filter=_to_upstash_filters(query.filters),
        namespace=self.namespace,
    )

    top_k_nodes = []
    top_k_ids = []
    top_k_scores = []
    for vector in res:
        node = metadata_dict_to_node(vector.metadata)
        node.embedding = vector.vector
        top_k_nodes.append(node)
        top_k_ids.append(vector.id)
        top_k_scores.append(vector.score)

    return VectorStoreQueryResult(
        nodes=top_k_nodes, similarities=top_k_scores, ids=top_k_ids
    )