How do name a connector so you have have 2 upstream connector to the same downstream connector?

[ringofhealth]

I have 2 retrievers,
BM25 and Embedding retrievers. I would like to connect both of them each to a sentence_window_retriever, then with a document joiner so we can select the best results.

in my mind it goes

BM25 -> Sentence--------> Document Joiner(the same instance)
Embedding -> Sentence --> Document Joiner(same Instance)

given this code

        rag_pipeline = Pipeline()
        rag_pipeline.add_component("text_embedder", text_embedder)
        rag_pipeline.add_component("embedding_retriever", embedding_retriever)
        rag_pipeline.add_component("bm25_retriever", bm25_retriever)
        rag_pipeline.add_component(
            "sentence_window__retriever",
            SentenceWindowRetriever(document_store=document_store, window_size=1),
        )

        rag_pipeline.add_component("document_joiner", document_joiner)
        rag_pipeline.add_component("prompt_builder", prompt_builder)
        rag_pipeline.add_component("generator", generator)

        # Connect components
        rag_pipeline.connect(
            "text_embedder.embedding", "embedding_retriever.query_embedding"
        )

        rag_pipeline.connect(
            "bm25_retriever.documents", "sentence_window__retriever.retrieved_documents"
        )
        rag_pipeline.connect(
            "embedding_retriever.documents",
            "sentence_window__retriever.retrieved_documents",
        )
        rag_pipeline.connect(
            "sentence_window__retriever.context_documents",
            "document_joiner.documents",
        )


        rag_pipeline.connect("document_joiner.documents", "prompt_builder.documents")
        rag_pipeline.connect("prompt_builder.prompt", "generator.messages")

2025-03-09 08:31:34,550 - __main__ - ERROR - Failed to create RAG pipeline: Cannot connect 'embedding_retriever.documents' with 'sentence_window__retriever.retrieved_documents': sentence_window__retriever.retrieved_documents is already connected to ['bm25_retriever'].

An error occurred: Cannot connect 'embedding_retriever.documents' with 'sentence_window__retriever.retrieved_documents': sentence_window__retriever.retrieved_documents is already connected to ['bm25_retriever'].

i mean this error makes sense.. but how do I actually do what I want to do?

[ringofhealth]

If I just use BM25 with the sentence window, then the graph looks like this..

        rag_pipeline = Pipeline()
        rag_pipeline.add_component("text_embedder", text_embedder)
        rag_pipeline.add_component("embedding_retriever", embedding_retriever)
        rag_pipeline.add_component("bm25_retriever", bm25_retriever)
        rag_pipeline.add_component(
            "sentence_window__retriever",
            SentenceWindowRetriever(document_store=document_store, window_size=2),
        )

        rag_pipeline.add_component("document_joiner", document_joiner)
        rag_pipeline.add_component("prompt_builder", prompt_builder)
        rag_pipeline.add_component("generator", generator)

        # Connect components
        rag_pipeline.connect(
            "text_embedder.embedding", "embedding_retriever.query_embedding"
        )

        rag_pipeline.connect(
            "embedding_retriever.documents", "document_joiner.documents"
        )
        rag_pipeline.connect(
            "bm25_retriever.documents", "sentence_window__retriever.retrieved_documents"
        )
        rag_pipeline.connect(
            "sentence_window__retriever.context_documents",
            "document_joiner.documents",
        )

        rag_pipeline.connect("bm25_retriever.documents", "document_joiner.documents")

        rag_pipeline.connect("document_joiner.documents", "prompt_builder.documents")
        rag_pipeline.connect("prompt_builder.prompt", "generator.messages")

1. i am not sure if the sentence_window_retriever result is the union of itself + BM25, or does it ONLY get get the windowed context. if this is the case, then it feels a bit odd because wouldnt it need the BM25 retrieved document + the context window retrieved) for greater relevance?

2. why does the sentence_window_retrieve have an extra context_windows that is by itself and connected to the end node?(is this suppose to happen?)

[julian-risch]

Hi @ringofhealth you could create a second instance of SentenceWindowRetriever by running

rag_pipeline.add_component(
    "sentence_window_retriever_2",
    SentenceWindowRetriever(document_store=document_store, window_size=1),
)

which you can then use in your pipeline. As an alternative you could use a DocumentJoiner right after the EmbeddingRetriever and the BM25Retriever and then a single SentenceWindowRetriever right after the DocumentJoiner.