feat: pull model list for openai-compatible endpoints

[cpacker]

Closes #592

Please describe the purpose of this pull request.

• During memgpt configure if using an openai endpoint, pull the list of models
• Also, allow users to choose a "enter yourself" custom option (useful for custom OpenAI compatible endpoints)

How to test

Run memgpt configure

OpenAI

Azure

vLLM

Have you tested this PR?

• Tested on OpenAI
• Tested on Azure OpenAI
• Tested on vLLM

Other comments

@sarahwooders

• I'm concerned for some users the total amount of models is information overload (see screenshot).
• Also, this is even after filtering for models only prefixed by gpt-
• Additionally, there are models in the list that are misleading to show (eg gpt-4-vision-preview shows up, but we don't support vision inputs at the moment)

[sarahwooders]

Could you also call the same function for the case where vllm is the endpoint type? e.g. with our endpoint you can do http://api.memgpt.ai/v1/models (since we require the OpenAI compatible vLLM endpoint).

Also might be nice to add Azure too, but not that important.

[cpacker]

Could you also call the same function for the case where vllm is the endpoint type? e.g. with our endpoint you can do http://api.memgpt.ai/v1/models (since we require the OpenAI compatible vLLM endpoint).

Also might be nice to add Azure too, but not that important.

Added vLLM support + tested against our vLLM API (see picture in OP)

[cpacker]

@sarahwooders added a new method to resolve the "too many models" problem:

Start by showing a curated list:

Allow the user to expand the list (if they expand the list, they get a warning):

Finally, they can still enter a manual model if the internet / GET request is broken and they know what they're doing:

[sarahwooders]

lgtm!