Gemini openai Network Problem

[quantumalchemy]

Bug Report

Using the new Gemini openai compatible api url:
in settings OpenAI API connections
https://generativelanguage.googleapis.com/v1beta
Get--> Network Problem

using v0.5.3
anyone have a solution ?
works ok with curl , python requests ..

[mxmerce]

Same issue as well.

[sgnlwagfjalgnvds]

Same issue.

[j-arnott]

Same here

[sevenjay]

google provides a method to compatibility with openai, like https://generativelanguage.googleapis.com/v1beta/openai
but they are stupid enough, only curl and python api can be compatibility.

from openai import OpenAI

client = OpenAI(
    api_key="OPENAI_API_KEY",
    base_url="https://api.openai.com/v1"
    # real use https://api.openai.com/v1/chat/completions
)
client = OpenAI(
    api_key="GEMINI_API_KEY",
    base_url="https://generativelanguage.googleapis.com/v1beta/openai"
    # real use https://generativelanguage.googleapis.com/v1beta/openai/chat/completions
)

you can see the different, there is impossible for "v1" in the url like : https://generativelanguage.googleapis.com/v1beta/openai/v1/chat/completions

[turbosnail11]

When you use that URL and your key you will get an error - but as long as you add a model it will work.

[sgnlwagfjalgnvds]

thanks a lot for your help. however, mine still does not work with a 500: Open WebUI: Server Connection Error.

[turbosnail11]

Correct, it will throw an error but it should still work if you have a valid key and set it up exactly like I did.

Also, you to ensure you have access, you can try running this from the command line - just remember to insert your API key.

curl -X POST "https://generativelanguage.googleapis.com/v1beta/openai/chat/completions"
-H "Content-Type: application/json"
-H "Authorization: Bearer $GOOGLE_API_KEY"
-d '{
"model": "gemini-1.5-flash",
"messages": [
{
"role": "user",
"content": "What is the capital of the USA?"
}
],
"stream": false
}'

[spergware]

When you use that URL and your key you will get an error - but as long as you add a model it will work.

I did as you shown but after I manually add the model(s) and save still nothing shows, and when I go back to the API settings those models are indeed not there, no matter how many times I add them. I set up everything correctly I swear
EDIT: Tried for the Nth time right after this comment and for some reason the models finally shown up BUT when I try to chat with them I get error 404, it's all so tiresome

[Mongram-GTI]

I did as you shown but after I manually add the model(s) and save still nothing shows, and when I go back to the API settings those models are indeed not there, no matter how many times I add them. I set up everything correctly I swear EDIT: Tried for the Nth time right after this comment and for some reason the models finally shown up BUT when I try to chat with them I get error 404, it's all so tiresome

Were you able to figure this out? Having the same 404 error.

[spergware]

I did as you shown but after I manually add the model(s) and save still nothing shows, and when I go back to the API settings those models are indeed not there, no matter how many times I add them. I set up everything correctly I swear EDIT: Tried for the Nth time right after this comment and for some reason the models finally shown up BUT when I try to chat with them I get error 404, it's all so tiresome

Were you able to figure this out? Having the same 404 error.

Yeah it's the url, you actually have to use https://generativelanguage.googleapis.com/v1beta as is, don't forget to add the model IDs manually

[paulo05100]

The problem comes from Google's compression of the answer. Data is received by the backend, which sends it back uncompressed to the frontend. But the backend sends the same header back to the frontend with r.headers['Content-Encoding']. You need to del r.headers['Content-Encoding'].

[CGangel]

Just do as this, then you can use gemini api
（You must enter a Model ID and click on “Add”, otherwise, you will get network error）

[paulo05100]

Yes since 0.5.11 with direct_chat_completion it's ok, without code modification :-)

[hypo-thesis]

I am still stuck, cannot add gemini

[suurt8ll]

Hi everyone,

Following the discussion here about connecting Gemini models to Open WebUI and the network/connection issues, I wanted to share something I'm working on.

Myself and some collaborators are developing an Open WebUI pipe (gemini_manifold.py) specifically designed to work with Gemini models and leverage more of its unique features beyond just the basic OpenAI compatibility layer.

It's hosted on my GitHub within the plugins/pipes directory (https://github.com/suurt8ll/open_webui_functions/), and it gets updated quite often as we develop it.

Feel free to check it out!