Support Gemini (Google Generative AI) Platform

[enricoros]

Support the Google generative AI platform and the Gemini models, as documented here:

• docs: ..
• login: ...

Requirements:

• Add Gemini Vendor
• Add Gemini vector icon
• Implement Setup and (transport) Access
• Implement Transport
• Implement Model List mapping (tokens, name, description, etc)
• Implement Streaming endpoint
• Test
• Ship 🚀

[david-strejc]

Support the Google generative AI platform and the Gemini models, as documented here:

• docs: ..
• login: ...

Requirements:

• Add Gemini Vendor
• Add Gemini vector icon
• Implement Setup and (transport) Access
• Implement Transport
• Implement Model List mapping (tokens, name, description, etc)
• Implement Streaming endpoint
• Test
• Ship 🚀

When will be first version ready?

[enricoros]

@david-strejc This week, ideally tomorrow, at least with the basic features.

[david-strejc]

@david-strejc This week, ideally tomorrow, at least with the basic features.

Many MANY thanks. We are curious what we can do with Gemini.

[enricoros]

Support almost complete. Will finish tomorrow.

• list models; ✅
• immediate generation: ✅
• streaming generation: will be done next
• image/multi-part input: not ready yet, needs to change all LLM providers - prob preview late this week
• function calling: very experimental, nowhere near OpenAI

Note that Gemini models:

• does not support system prompts well (they do not have the concept of system prompts)
• ... more quirks? ...

[enricoros]

Code landed for a very good implementation of Gemini.

• includes streaming, non-streaming, list models
• includes changeable safety settings
• includes multipart text conversation building

Next up:

• ship this in Release 1.8.0 #283, quickly
• enable the images multi-part pipeline