Provide support for gpt-4 turbo (gpt-4-1106-preview)

[gitthub89]

Duplicates

• I have searched the existing issues

Summary 💡

The option to set gpt-4-1106-preview in the .env should be allowed and thus set the gpt4 model of autogpt to this gpt-4-1106-preview (or any other available one).
Then combined with --gpt4only will let the user use only this model.

Examples 🌈

Motivation 🔦

Use the newest model available, and save money because it's cheaper as described here:
https://openai.com/blog/new-models-and-developer-products-announced-at-devday

[Bentlybro]

the new model was only just released a few hours ago, we already plan to add it some time today/tomorrow

[gitthub89]

I am not sure why the models must be hardcoded though, for every change openAI makes, you have to edit the code again seems like an overhead. also if someone wants to use one of the lesser used models it's impossible without changing the code. just something I thought about, but I also think others thought about it so there's probably a good reason I am not aware of.

[ph-ausseil]

A more sensible change is actually that AutoGPT uses ChatCompletion API & a new API (Assistant API) is available & it will need a new ModelProvider such as AssistantModelProvider to benefit from this new model. Assistant API from my understanding is should be optimized for function calling (Which has to be benchmarked against ChatCompletion API).

Also this new model could or would (depends of the way you want to implement it) implements new functions such as Thread.

[abhawan888]

Waiting on this as well.

[Pwuts]

Resolved in #6147

[sfxworks]

Can you put this in the stable branch please?

[jackchang1025]

Although it is supported, it still cannot be used in the docker environment I built. I found that the image did not follow the new code.

This is my docker-compose.yml file

gpt-4-1106-preview is set in my .env

Error when running

I checked the built image file and it doesn't seem to update the code

Can the complete dokefile be displayed?