RAG - Provide Reference / Source link of Knowledgebase in llm Output

[Constey]

I've a ConfluenceLoader loading a Space into a ChromaDB that works well.
I have the content of each page in {text} and the url in the {source} collumn:

In the Prompt Template i've tried to pass {source} aswell as {text}, but it did not work. I've aswell found nothing in the docs. May someone can give me a hint please ? :-)

The result should be that in the response the url to the page will be linked where the {context} text has been found.
Thanks in advance

[dosubot[bot]]

Hello @Constey! I'm here to help you with bug fixes, answer questions, and guide you on becoming a contributor. Let's tackle this challenge together!

To include additional fields like {source} in your Langflow prompt template, you need to ensure that the prompt template is correctly set up to handle these fields. Here are some steps and best practices you can follow:

1. Validate the Prompt Template: Use the validate_prompt function to ensure that your template is valid and to extract input variables, including {source} and {text}.

2. Add New Variables: Use the add_new_variables_to_template function to add new input variables to your template. This function will ensure that each variable is represented correctly in the template.

3. Update Input Variables: Make sure the input_variables field in your template is updated with the current list of input variables, including {source}.

4. Process the Prompt Template: Utilize the process_prompt_template function to handle the integration of custom fields and ensure the template is processed correctly.

By following these steps, you should be able to include the {source} field in your prompt template and have it processed correctly in the LLM output ^[1][2].

To continue talking to Dosu, mention @dosu.

[Constey]

Ok I've tried aswell with a second "parse data" to fetch the {source} collumn of the chroma db. But it's still not working :-(

[Constey]

@dosu. any ideas how I can provide the references from the vector db (source) collumn into my message output?

[dosubot[bot]]

To provide the references from the vector database's {source} column into your message output, you can use the return_source_documents option in the RetrievalQAComponent. When this option is set to True, the source documents are included in the result. Here's a code snippet that shows how the references are appended to the result string:

if self.return_source_documents and len(source_docs):
    references_str = self.create_references_from_data(source_docs)
    result_str = f"{result_str}\n{references_str}"

This code checks if return_source_documents is enabled and if there are any source documents. It then creates a string of references using create_references_from_data(source_docs) and appends it to the result string ^[1].

Make sure that your workflow is configured to enable this option and that the Parse Data node is correctly set up to extract the {source} column.

To continue talking to Dosu, mention @dosu.

[Constey]

The Retrieval QA is marked as Legacy what is the new component to use? @dosu.

[dosubot[bot]]

The new component that replaces the Legacy Retrieval QA in Langflow is the RetrievalQAComponent ^[1].

To continue talking to Dosu, mention @dosu.