RAG - Show document metadata in the stream response.

[pironev]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

const retriever = await loadedVectorStore.asRetriever();
 const retrievalChain = await createRetrievalChain({
 documentsChain,
 retriever
 });
 const response = await retrievalChain.stream({ input: "Where is London?" }
  for await (const chunk of response) {
     if(chunk?.answer)
      {
        console.log(chunk.answer)
      }
 }

Description

If is generate a stream with retrievalChain as in example.
How can I show the metadata of the document found at the end of the response?

System Info

langchain v0.2
llama cpp

[dosubot[bot]]

To show the metadata of the document found at the end of the response when generating a stream with retrievalChain in LangChain v0.2, you need to ensure that the metadata is included in the Document objects created from the search results. Here's how you can achieve this:

1. Extract Metadata: Use the _getMetadata function to extract the metadata from the SearchResult and include it in the Document object.

2. Include Metadata in Stream: Modify your stream handling to include the metadata in the output.

Here's an example of how you can do this:

Step 1: Extract Metadata

Ensure you have a function to extract metadata from the search results:

export function _getMetadata<T extends ContentsOptions = { text: true }>(
  result: SearchResult<T>
): Record<string, unknown> {
  const newMetadata: Record<string, unknown> = { ...result };
  delete newMetadata.text;
  return newMetadata;
}

async _getRelevantDocuments(query: string): Promise<Document[]> {
  const res: SearchResponse<T> = await this.client.searchAndContents<T>(
    query,
    this.searchArgs
  );

  const documents: Document[] = [];
  for (const result of res.results) {
    let pageContent;
    if ("text" in result) {
      pageContent = result.text;
    } else if ("highlights" in result) {
      pageContent = result.highlights.join("\n\n");
    } else {
      pageContent = "No results found.";
    }
    documents.push(
      new Document({
        pageContent,
        metadata: _getMetadata<T>(result),
      })
    );
  }
  return documents;
}

Step 2: Include Metadata in Stream

Modify your stream handling to include the metadata in the output:

const retriever = await loadedVectorStore.asRetriever();
const retrievalChain = await createRetrievalChain({
  combineDocsChain,
  retriever,
  returnSourceDocuments: true, // Ensure source documents are returned
});

const response = await retrievalChain.stream({ input: "Where is London?" });

for await (const chunk of response) {
  if (chunk?.answer) {
    console.log(chunk.answer);
  }
  if (chunk?.sourceDocuments) {
    chunk.sourceDocuments.forEach((doc) => {
      console.log(doc.metadata); // Log the metadata of each document
    });
  }
}

This approach ensures that the metadata is included in the Document objects and is accessible in the stream response ^[1][2].

To continue talking to Dosu, mention @dosu.

[pironev]

The proposed solutions did not work for me but I saw that the first chunks are "input", "chat_history" and "context". It can extract the metadata from the context.
I don't know if it's the best solution. Anyway, it works.