Fusion retriever

HybridFusionRetrieverPack

Bases: BaseLlamaPack

混合融合检索包。

使用融合技术集成向量和bm25检索器。

Source code in llama_index/packs/fusion_retriever/hybrid_fusion/base.py

class HybridFusionRetrieverPack(BaseLlamaPack):
    """混合融合检索包。

    使用融合技术集成向量和bm25检索器。"""

    def __init__(
        self,
        nodes: List[TextNode] = None,
        chunk_size: int = 256,
        mode: str = "reciprocal_rerank",
        vector_similarity_top_k: int = 2,
        bm25_similarity_top_k: int = 2,
        fusion_similarity_top_k: int = 2,
        num_queries: int = 4,
        documents: List[Document] = None,
        cache_dir: str = None,
        **kwargs: Any,
    ) -> None:
        """初始化参数。"""
        service_context = ServiceContext.from_defaults(chunk_size=chunk_size)
        if cache_dir is not None and os.path.exists(cache_dir):
            # Load from cache
            from llama_index import StorageContext, load_index_from_storage

            # rebuild storage context
            storage_context = StorageContext.from_defaults(persist_dir=cache_dir)
            # load index
            index = load_index_from_storage(storage_context)
        elif documents is not None:
            index = VectorStoreIndex.from_documents(
                documents=documents, service_context=service_context
            )
        else:
            index = VectorStoreIndex(nodes, service_context=service_context)

        if cache_dir is not None and not os.path.exists(cache_dir):
            index.storage_context.persist(persist_dir=cache_dir)

        self.vector_retriever = index.as_retriever(
            similarity_top_k=vector_similarity_top_k
        )

        self.bm25_retriever = BM25Retriever.from_defaults(
            docstore=index.docstore, similarity_top_k=bm25_similarity_top_k
        )
        self.fusion_retriever = QueryFusionRetriever(
            [self.vector_retriever, self.bm25_retriever],
            similarity_top_k=fusion_similarity_top_k,
            num_queries=num_queries,  # set this to 1 to disable query generation
            mode=mode,
            use_async=True,
            verbose=True,
            # query_gen_prompt="...",  # we could override the query generation prompt here
        )

        self.query_engine = RetrieverQueryEngine.from_args(self.fusion_retriever)

    def get_modules(self) -> Dict[str, Any]:
        """获取模块。"""
        return {
            "vector_retriever": self.vector_retriever,
            "bm25_retriever": self.bm25_retriever,
            "fusion_retriever": self.fusion_retriever,
            "query_engine": self.query_engine,
        }

    def retrieve(self, query_str: str) -> Any:
        """检索。"""
        return self.fusion_retriever.retrieve(query_str)

    def run(self, *args: Any, **kwargs: Any) -> Any:
        """运行流水线。"""
        return self.query_engine.query(*args, **kwargs)

get_modules

get_modules() -> Dict[str, Any]

获取模块。

Source code in llama_index/packs/fusion_retriever/hybrid_fusion/base.py

def get_modules(self) -> Dict[str, Any]:
    """获取模块。"""
    return {
        "vector_retriever": self.vector_retriever,
        "bm25_retriever": self.bm25_retriever,
        "fusion_retriever": self.fusion_retriever,
        "query_engine": self.query_engine,
    }

retrieve

retrieve(query_str: str) -> Any

检索。

Source code in llama_index/packs/fusion_retriever/hybrid_fusion/base.py

def retrieve(self, query_str: str) -> Any:
    """检索。"""
    return self.fusion_retriever.retrieve(query_str)

run

run(*args: Any, **kwargs: Any) -> Any

运行流水线。

Source code in llama_index/packs/fusion_retriever/hybrid_fusion/base.py

def run(self, *args: Any, **kwargs: Any) -> Any:
    """运行流水线。"""
    return self.query_engine.query(*args, **kwargs)

QueryRewritingRetrieverPack

Bases: BaseLlamaPack

查询重写检索器包。

给定输入节点，构建一个向量索引。

然后将查询重写为多个查询，并重新对结果进行排名。

Source code in llama_index/packs/fusion_retriever/query_rewrite/base.py

class QueryRewritingRetrieverPack(BaseLlamaPack):
    """查询重写检索器包。

给定输入节点，构建一个向量索引。

然后将查询重写为多个查询，并重新对结果进行排名。
"""

    def __init__(
        self,
        nodes: List[TextNode] = None,
        chunk_size: int = 256,
        mode: str = "reciprocal_rerank",
        vector_similarity_top_k: int = 2,
        fusion_similarity_top_k: int = 2,
        num_queries: int = 4,
        **kwargs: Any,
    ) -> None:
        """初始化参数。"""
        service_context = ServiceContext.from_defaults(chunk_size=chunk_size)
        index = VectorStoreIndex(nodes, service_context=service_context)
        self.vector_retriever = index.as_retriever(
            similarity_top_k=vector_similarity_top_k
        )

        self.fusion_retriever = QueryFusionRetriever(
            [self.vector_retriever],
            similarity_top_k=fusion_similarity_top_k,
            num_queries=num_queries,  # set this to 1 to disable query generation
            mode=mode,
            use_async=True,
            verbose=True,
            # query_gen_prompt="...",  # we could override the query generation prompt here
        )

        self.query_engine = RetrieverQueryEngine.from_args(self.fusion_retriever)

    def get_modules(self) -> Dict[str, Any]:
        """获取模块。"""
        return {
            "vector_retriever": self.vector_retriever,
            "fusion_retriever": self.fusion_retriever,
            "query_engine": self.query_engine,
        }

    def retrieve(self, query_str: str) -> Any:
        """检索。"""
        return self.fusion_retriever.retrieve(query_str)

    def run(self, *args: Any, **kwargs: Any) -> Any:
        """运行流水线。"""
        return self.query_engine.query(*args, **kwargs)

get_modules

get_modules() -> Dict[str, Any]

获取模块。

Source code in llama_index/packs/fusion_retriever/query_rewrite/base.py

def get_modules(self) -> Dict[str, Any]:
    """获取模块。"""
    return {
        "vector_retriever": self.vector_retriever,
        "fusion_retriever": self.fusion_retriever,
        "query_engine": self.query_engine,
    }

retrieve

retrieve(query_str: str) -> Any

检索。

Source code in llama_index/packs/fusion_retriever/query_rewrite/base.py

def retrieve(self, query_str: str) -> Any:
    """检索。"""
    return self.fusion_retriever.retrieve(query_str)

run

run(*args: Any, **kwargs: Any) -> Any

运行流水线。

Source code in llama_index/packs/fusion_retriever/query_rewrite/base.py

def run(self, *args: Any, **kwargs: Any) -> Any:
    """运行流水线。"""
    return self.query_engine.query(*args, **kwargs)