is there a standardised way to interact with LLMs

[mosajjal]

Hi.

has anyone come up with a unified, standard API to talk to all LLMs? (LLaMA, Bard, OpenAI etc)? if not, it would be helpful to start defining that as a community and build the compatibility layers for that. keen to hear your thoughts.

[iaalm]

OpenAI API is the de facto standard in this area. That's why I start this project to call custom LLM in OpenAI way. Any suggestion?

[mosajjal]

I agree that currently OpenAI is considered the standard. However, I don't think OpenAI is committed to keep the API specification intact for the foreseeable future.

My suggestion is to start a community-led group to codify a standard, and start building a compatibility shim between major models (OpenAI, LLaMA etc) according to said model. We can always start with OpenAI's reference and build on top of it.

[iaalm]

Agree, that would be very useful😎

[mosajjal]

cool :) so how should we start this? need to start a group of some sort and start exchanging ideas on what it would look like.

[iaalm]

Seriously I don't know. But maybe we can define an API spec and implement adaptors for popular models?

[mosajjal]

sounds good. I can help with Bard.