OpenAI Spec Expansion For Multi-Modal Models?

[norman-kong]

🚀 Feature

Expansion of the OpenAI Spec to include text-to-image and other multi-modal models.

Motivation

As a developer, I'd like to host several text-to-image (and other multi-modal models) on my own hardware, and access them via an API that conforms to the OpenAI API standard.

That is, I'd like to easily deploy my model and access it through an API that conforms to the OpenAI standard (as LitServe offers) but with text-to-image and other multi-modal models.

Pitch

See Feature/Motivation.

Alternatives

I've seen a couple of libraries that sort of offer this functionality (like LocalAI), but I want to stick with the simplicity of LitServe.

Additional context

Is this feature planned in the short term? Or not at all?

Or should I be writing a "spec"? I can't seem to find much documentation on that. Any insight would be appreciated!