Integration

Lol google is hard to setup auth for, for sure.

Bedrock is also up there. It was so bad before they had the converse api. But even now, trying to figure out the correct model name and if the model is available in your region is insane

I'm having troubles in integrating studio + vertex ai into an unified interface for my work

Horrible

Seems that each platform contains different SDKs. Google.generativeai and vertexai.generative_models

Also they don't follow the OpenAI API style (of course)

🙂 hehe they don't need to and why should they? Not that am a google fan but usually you would use one model stack. if you want another you setup a different railwork sort to say

that would keep your setup clean and the only thing you access in common is data

Yeah, they should not I know 😁

Why would they do that, right?

A lot of other providers follow OpenAI schema, but I understand their choices

The only problem is that cloud SDKs seems (imo) to be made to be intentionally hard to change to another provider. So we get stuck to them.

I mean.. it's easy to change from OpenAI to anthropic, groq etc

tbh I think openai shaked some feathers and not sure if google was on the ball for that. they do have very stable and well documented apis for anything else

I use GCP for the past 8 years and compared to other clouds it was the easist to integrate with

now ... from an ai perspective that is quite the oppposite and I hear you 🙂

Yeah... I've never seen anything more confusing than the

google.ai
google.generativeai
vertexai.generative_models
google.ai.generative_language

And some of them have classes with the same names and responsibilities, but doesn't accept objects from others. For example, google.generativeai's GenerativeModel class accepts different Tool's objects than vertexai.generative_models. GenerativeModel, which accepts Tool objects, for example.