Pre-built Integration to external embedding and generative APIs

[eostis]

This is about APIs like OpenAI, Cohere, HuggingFace endPoints (and others). Obviously.

I could not find anything in the documentation or in issues, so I'm starting the subject.

The idea:

• Add some configuration for the fore-mentioned APIs, without too much sweating.

The target:

• Projects with little skills or time, but ready to pay for a quick and easy POC.
• A good example are WordPress and WooCommerce sites.

Pros & cons:

• Both APis and downloaded models have a cost: usage for the former, time and hosting for the later.
• For tiny projects, the API cost is negligible.
• For huge projects, the API cost is overwhelming.
• Some models are too big to even think of hosting them.

Features:

• Build vectors for documents and queries
• Transform query and results with a generative model (ask to fix typos, ask to generate recipes from ingredients ...)

[eostis]

Closing inactive issue.

[bratseth]

In development.

[zseymour]

@bratseth what is the status of this development?

[lesters]

@zseymour This is available now: https://docs.vespa.ai/en/llms-in-vespa.html

[zseymour]

@lesters Thank you! I did some searching, but are there any similar examples for hitting an API as the embeddings provider, too?