Add support for Gemini models

[NimJay]

Describe the feature

• This is a feature request to add support for Google's Gemini models.
• I'm mainly creating this GitHub issue to first check if the owners of this repository are open to having Gemini support. :)
  • (I personally would want to try out this repo using Google's Gemini models.)

Potential Solutions

• This changes would be similar to:
  • feat: Amazon Bedrock support (Claude models) #207
  • Commit adding Ollama support
• See Python Gemini client library example here.
• See source code for GenerativeModel class here

Related documentation

• cloud.google.com - Model API reference for Generative AI

[ofirpress]

Yup, this would be great to have, we welcome contributions on this.

[klieret]

@NimJay Could you check if we could add the new model via litellm? We're slowly starting to accumulate a lot of models/code in models.py and perhaps slowly migrating some of them to use litellm might allow us to simplify (also see #64). For the models that are used in the paper, we need to be extra careful to replicate the cost function etc., so let's perhaps test this for new models first.

Disclaimer: I haven't looked too closely at litellm, so feel free to let me know if this would make things more/equally complicated instead of simplifying things under one common interface.

[ofirpress]

We're going to add support for something like LiteLLM therefore this issue will be deprecated. Closing.