Links

sshyam

Hello, I was wondering what's the recommended way to list reference links (links to source material) along with a response while using chat_engine?

My knowledge base is basically a cloned github repo containing markdown files from my documentation and blog. What I ideally want to do is along with the response show the hyperlinks to the documents that were referred to build that response.

I couldn't find any guides or examples that does this.

15 comments

LLogan M

Each response has a source_nodes attribute. And each source node has metadata

So basically, you'll need to set some metadata on your input documents to properly retrieve it from the source nodes.

response.source_nodes[0].node.metadata

LLogan M

https://gpt-index.readthedocs.io/en/stable/core_modules/data_modules/documents_and_nodes/usage_documents.html#metadata

sshyam

thanks, so IIUC, the filename can be set as metadata, and then I can write the custom logic to convert the filename to the hyperlink I'm after

LLogan M

Yea exactly

sshyam

I seem to run into

Plain Text

AttributeError: 'Document' object has no attribute 'file_metadata'

sshyam

even after following the steps in https://gpt-index.readthedocs.io/en/stable/core_modules/data_modules/documents_and_nodes/usage_documents.html#metadata

LLogan M

Did you follow the same though?

Plain Text

from llama_index import SimpleDirectoryReader
filename_fn = lambda filename: {'file_name': filename}

# automatically sets the metadata of each document according to filename_fn
documents = SimpleDirectoryReader('./data', file_metadata=filename_fn).load_data()
print(documents[0].metadata['file_name'])

sshyam

yeah, let me try again in a simpler sample app..

LLogan M

Works for me locally 🤔

Plain Text

>>> from llama_index import SimpleDirectoryReader
>>> name_fn = lambda filename: {'file_name': filename}
>>> filename_fn = lambda filename: {'file_name': filename}
>>> documents = SimpleDirectoryReader('./docs/examples/data/paul_graham', file_metadata=filename_fn).load_data()
>>> print(documents[0].metadata['file_name'])
docs/examples/data/paul_graham/paul_graham_essay.txt
>>>

sshyam

Got it working, thanks !

sshyam

one last Q for the day 🙂

So in the response

Plain Text

 response = chat_engine.chat(prompt)

can it also return the metadata?

sshyam

I guess I'm asking during runtime how do I access the relevant filename for the question that was asked

LLogan M

Yea, exact same as I linked above

response.source_nodes[0].node.metadata

Although source nodes can be an empty list if the chat engine is an agent and didn't choose to use any tools

sshyam

Got it! And it works. Thanks for being awesome.

The support I got within minutes of hopping on here has been outstanding. Keep up the great work 🔥

LLogan M

:dotsCATJAM: :dotsHARDSTYLE:

Add a reply

Find answers from the community

Links