Chatgpt plugin

ChatGPTRetrievalPluginClient

Bases: BasePydanticVectorStore

聊天GPT检索插件客户端。

在此客户端中，我们利用了ChatGPT定义的端点。

Parameters:

Name	Type	Description	Default
endpoint_url	str	ChatGPT检索插件的URL。	required
bearer_token	Optional[str]	ChatGPT检索插件的Bearer令牌。	None
retries	Optional[Retry]	ChatGPT检索插件的重试对象。	None
batch_size	int	ChatGPT检索插件的批处理大小。	100

Source code in llama_index/vector_stores/chatgpt_plugin/base.py

class ChatGPTRetrievalPluginClient(BasePydanticVectorStore):
    """聊天GPT检索插件客户端。

在此客户端中，我们利用了ChatGPT定义的端点。

Args:
    endpoint_url (str): ChatGPT检索插件的URL。
    bearer_token (Optional[str]): ChatGPT检索插件的Bearer令牌。
    retries (Optional[Retry]): ChatGPT检索插件的重试对象。
    batch_size (int): ChatGPT检索插件的批处理大小。"""

    stores_text: bool = True
    is_embedding_query: bool = False

    _endpoint_url: str = PrivateAttr()
    _bearer_token: Optional[str] = PrivateAttr()
    _retries: Optional[Retry] = PrivateAttr()
    _batch_size: int = PrivateAttr()
    _s: requests.Session = PrivateAttr()

    def __init__(
        self,
        endpoint_url: str,
        bearer_token: Optional[str] = None,
        retries: Optional[Retry] = None,
        batch_size: int = 100,
        **kwargs: Any,
    ) -> None:
        """初始化参数。"""
        super().__init__()

        self._endpoint_url = endpoint_url
        self._bearer_token = bearer_token or os.getenv("BEARER_TOKEN")
        self._retries = retries
        self._batch_size = batch_size

        self._s = requests.Session()
        self._s.mount("http://", HTTPAdapter(max_retries=self._retries))

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

    @property
    def client(self) -> None:
        """获取客户端。"""
        return

    def add(
        self,
        nodes: List[BaseNode],
        **add_kwargs: Any,
    ) -> List[str]:
        """将节点添加到索引。"""
        headers = {"Authorization": f"Bearer {self._bearer_token}"}

        docs_to_upload = convert_docs_to_json(nodes)
        iterable_docs = get_tqdm_iterable(
            range(0, len(docs_to_upload), self._batch_size),
            show_progress=True,
            desc="Uploading documents",
        )
        for i in iterable_docs:
            i_end = min(i + self._batch_size, len(docs_to_upload))
            self._s.post(
                f"{self._endpoint_url}/upsert",
                headers=headers,
                json={"documents": docs_to_upload[i:i_end]},
            )

        return [result.node_id for result in nodes]

    def delete(self, ref_doc_id: str, **delete_kwargs: Any) -> None:
        """使用ref_doc_id删除节点。

Args:
    ref_doc_id（str）：要删除的文档的doc_id。
"""
        headers = {"Authorization": f"Bearer {self._bearer_token}"}
        self._s.post(
            f"{self._endpoint_url}/delete",
            headers=headers,
            json={"ids": [ref_doc_id]},
        )

    def query(
        self,
        query: VectorStoreQuery,
        **kwargs: Any,
    ) -> VectorStoreQueryResult:
        """获取响应的节点。"""
        if query.filters is not None:
            raise ValueError("Metadata filters not implemented for ChatGPT Plugin yet.")

        if query.query_str is None:
            raise ValueError("query_str must be provided")
        headers = {"Authorization": f"Bearer {self._bearer_token}"}
        # TODO: add metadata filter
        queries = [{"query": query.query_str, "top_k": query.similarity_top_k}]
        res = requests.post(
            f"{self._endpoint_url}/query", headers=headers, json={"queries": queries}
        )

        nodes = []
        similarities = []
        ids = []
        for query_result in res.json()["results"]:
            for result in query_result["results"]:
                result_id = result["id"]
                result_txt = result["text"]
                result_score = result["score"]
                result_ref_doc_id = result["source_id"]
                node = TextNode(
                    id_=result_id,
                    text=result_txt,
                    relationships={
                        NodeRelationship.SOURCE: RelatedNodeInfo(
                            node_id=result_ref_doc_id
                        )
                    },
                )
                nodes.append(node)
                similarities.append(result_score)
                ids.append(result_id)

            # NOTE: there should only be one query
            break

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

client property

client: None

获取客户端。

add

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

将节点添加到索引。

Source code in llama_index/vector_stores/chatgpt_plugin/base.py

def add(
    self,
    nodes: List[BaseNode],
    **add_kwargs: Any,
) -> List[str]:
    """将节点添加到索引。"""
    headers = {"Authorization": f"Bearer {self._bearer_token}"}

    docs_to_upload = convert_docs_to_json(nodes)
    iterable_docs = get_tqdm_iterable(
        range(0, len(docs_to_upload), self._batch_size),
        show_progress=True,
        desc="Uploading documents",
    )
    for i in iterable_docs:
        i_end = min(i + self._batch_size, len(docs_to_upload))
        self._s.post(
            f"{self._endpoint_url}/upsert",
            headers=headers,
            json={"documents": docs_to_upload[i:i_end]},
        )

    return [result.node_id for result in nodes]

delete

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

使用ref_doc_id删除节点。

Source code in llama_index/vector_stores/chatgpt_plugin/base.py

    def delete(self, ref_doc_id: str, **delete_kwargs: Any) -> None:
        """使用ref_doc_id删除节点。

Args:
    ref_doc_id（str）：要删除的文档的doc_id。
"""
        headers = {"Authorization": f"Bearer {self._bearer_token}"}
        self._s.post(
            f"{self._endpoint_url}/delete",
            headers=headers,
            json={"ids": [ref_doc_id]},
        )

query

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

获取响应的节点。

Source code in llama_index/vector_stores/chatgpt_plugin/base.py

def query(
    self,
    query: VectorStoreQuery,
    **kwargs: Any,
) -> VectorStoreQueryResult:
    """获取响应的节点。"""
    if query.filters is not None:
        raise ValueError("Metadata filters not implemented for ChatGPT Plugin yet.")

    if query.query_str is None:
        raise ValueError("query_str must be provided")
    headers = {"Authorization": f"Bearer {self._bearer_token}"}
    # TODO: add metadata filter
    queries = [{"query": query.query_str, "top_k": query.similarity_top_k}]
    res = requests.post(
        f"{self._endpoint_url}/query", headers=headers, json={"queries": queries}
    )

    nodes = []
    similarities = []
    ids = []
    for query_result in res.json()["results"]:
        for result in query_result["results"]:
            result_id = result["id"]
            result_txt = result["text"]
            result_score = result["score"]
            result_ref_doc_id = result["source_id"]
            node = TextNode(
                id_=result_id,
                text=result_txt,
                relationships={
                    NodeRelationship.SOURCE: RelatedNodeInfo(
                        node_id=result_ref_doc_id
                    )
                },
            )
            nodes.append(node)
            similarities.append(result_score)
            ids.append(result_id)

        # NOTE: there should only be one query
        break

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