3.5 and 4 support?

Yes the gpt-X models use a conversational transaction in its payload, the older models (davinci, etc) use the simpler payload.

One enhancement down the road will be to simplify this into a config form where a user can select from different models and the its on the implementation to decide which endpoint to call. What makes it a bit cumbersome is that the gpt-X models are typically (I think) assuming you are going to continue interacting with it for refinement. Sometimes you do, sometimes you just want one thing.

Yes, the models are definitely slightly different.

I'm not sure what to make of the continued interaction. Is it a way to increase the number of tokens used as the conversation gets progressively longer? I noticed that the chatgpt module keeps an array of the earlier posts in $message and then resubmits them. (This confused me for a bit because I assumed the interaction was stateless.)

Thx.

No, its just the way those APIs work. GPT models use the chats endpoint, text- ones use the completion endpoint. The compability for endpoints and models is listed here:

https://platform.openai.com/docs/models/model-endpoint-compatibility

The reason GPT-3.5+ is able to interact with users (ChatGPT) is because it can reference the submission history for context. The Completion endpoints do not.

https://platform.openai.com/docs/models/model-endpoint-compatibility

I think we are good here for now unless I am missing something.

FYI: 📌 Enable GPT models for CKEditor. Fixed

I think moving forward, completion endpoint will be phased out for simply using the chat GPT endpoints.

https://platform.openai.com/docs/deprecations/2023-07-06-gpt-and-embeddings