Add support for Gemini models

[mnvsk97]

• [x] Models and providers should be configurable from models_config.yaml -> No extra work needed. Ref here
• [x] Add support to inference servers that are not OpenAI compatible. -> Not needed. Ref here
• [ ] Support all modes of authentication to communicate with Gemini models via OpenAI API

[AbhishekRP2002]

hi @mnvsk97 if this is not taken up yet by anyone, I would love to take this up and contribute.

• I think for the first point we have to update model_gateway.py and model_config.sample.yaml
• for the second one, can u please explain a little more ?

[AbhishekRP2002]

any updates on this ? @mnvsk97

cc: @chiragjn

[AbhishekRP2002]

Hi @mnvsk97 @chiragjn please lemme know if this enhancement is still in the roadmap. Would be super grateful if given the chance to contribute

[chiragjn]

Hey @AbhishekRP2002 , sorry I didn't get chance to reply earlier. I am wondering, now that Gemini has a openai compatible API (https://ai.google.dev/gemini-api/docs/openai), is this still needed?

---

I know accessing Gemini without API Key via some service account will need separate work

[AbhishekRP2002]

Hey @AbhishekRP2002 , sorry I didn't get chance to reply earlier. I am wondering, now that Gemini has a openai compatible API (https://ai.google.dev/gemini-api/docs/openai), is this still needed?

I know accessing Gemini without API Key via some service account will need separate work

yeah u are right actually, i think with Gemini models accessible using the OpenAI lib we don't need to make any changes. I thought in the direction of using gemini models either via Google AI Studio ( google.generativeai package ) or Vertex AI. Apologies. yeah for service account we need the GOOGLE_APPLICATIONS_CREDENTIALS with authentication via google auth.

[mnvsk97]

Hello @AbhishekRP2002, I've updated the issue description as per the conversation done so far. Would you still like to pick this up? Please let us know if you need any help in getting started.