How to configure Custom Embedding Provider

[NirnayK]

In Custom Embedding Provider For the function: async def search(self, text: str, max_results: int = 20) -> List[IndexItem]:

Can anyone explain what needs to be returned in an IndexItem
what should be returned in the text attribute and what should be returned in meta attribute.

[NirnayK]

Is there any code with a fully implemented Custom Embedding Provider ?

[drazvan]

@NirnayK : the default implementation is a good reference point: https://github.com/NVIDIA/NeMo-Guardrails/blob/main/nemoguardrails/embeddings/basic.py.

And to respond to your questions:

1. The text is the text of the chunk.
2. In the meta it's a dict with different types of info depending on the purpose of the index (e.g., body and title for the kb; intent for the user utterances).