Find answers from the community

Updated 10 months ago

Hey guys, I have a question regarding

Hey guys, I have a question regarding document metadata - in all examples/tutorials metadata is presented as a list of simple {key}:{value} string pairs. But what about more complex cases, let's say:

value is a list of strings
value is a dict
value is a list of dicts

Are there ways to represent such metadata in LlamaIndex Documents to make sure it is understood/handled properly by the embedder and LLM?

4 comments

WWhiteFang_Jr

Currently Metadata is in the form of Dict only.

Did you try passing your cases like
key: lIst of strings
key: dict
key: list of dict
?

ddmitry.danilov

That sure is the first/simplest idea that comes to mind but I was wondering if lists/dicts represented as strings (specifically, all the punctuation chars such as { } [ ] , : ) will be handled properly by embedder/LLM

LLogan M

it will just be converted/casted as string

LLogan M

so if the LLM/embed model has seen code before, it should be fine

Add a reply