Feature Description

Allow vector stores in llama index to skip embedding generation, leaving it to the underlying vector database to handle.

This allows users to make use of vector stores that provide an inherent embedding generation capability. Example, weaviate embeddings: https://weaviate.io/developers/wcs/embeddings

Reason

I noticed this while trying to use LlamaIndex with a Weaviate collection where 'Weaviate embeddings' were already enabled. In this situation, Weaviate only expects the objects (text inputs maybe) and will internally generate and attach embeddings using one of the models available on the platform, hosted by them.

But from what I see, the llamaindex vector stores will use an embedding model and generate embeddings by default, meaning even though embeddings on weaviate are enabled, they get overridden by the ones generated with llama index.

The embedding functionality in weaviate is convenient because it means the user has one less api key to worry about, and neither do they have to host an embedding model themselves.

Value of Feature

No response