[Feature]: Google Gemini support

[donbowman]

What problem does the new feature solve?

How may I use this with Google Gemini? I believe it supports both its own APi and also a compatibility API.
https://developers.googleblog.com/en/gemini-is-now-accessible-from-the-openai-library/

I tried

select ai.ollama_generate( 'gemini-1.5-flash', 'what is the weather in waterloo in june', host=>'https://generativelanguage.googleapis.com/v1beta/');

I looked at some proxies, but don't believe that is a good approach.

What does the feature do?

Make Gemini a peer to Ollama and OpenAI

Implementation challenges

No response

Are you going to work on this feature?

None

[alejandrodnm]

@donbowman we've just merged support for liteLLM. We released a new version of the vectorizer worker with support for it, but we haven't made a new extension release there are some pending things like docs:

https://github.com/timescale/pgai/pull/397/files

You could try building the extension from source. Then specify the model with gemini/text-embedding-004, something like:

    SELECT ai.create_vectorizer(
        'my_table'::regclass,
        embedding => ai.embedding_litellm(
          'gemini/text-embedding-004',
          768,
          api_key_name => 'GEMINI_API_KEY'
        ));

[donbowman]

excellent!
I am still working on the PR, but got bogged down in both it and other things. its working but one of the tests fails still.

I was also adding the e.g. query lookup.