Support Open WebUI as a Unified LLM Provider

[JBGruber]

We only need to make a few changes to support Open WebUI. I already made the bare minimum changes in https://github.com/JBGruber/rollama/tree/open-webui. The authentication can already be attached as described in #7.

The main advatages are:

• user authentication is possible, which allows for a central Ollama server, e.g., in a research lab. Everyone who has been granted access can then query models through their own account
• we can compare the models available through Ollama to proprietary ones

This is the current syntax, which could use some improvement, but works:

options(rollama_headers = list(Authorization = "Bearer sk-6b82a5443e7145d4a779922592701ee3"))
query(q = "Why is the sky blue?", 
            model = "gpt-4o", 
            server = "http://localhost:3000", 
            endpoint = "/api/chat/completions")
#>                      
#>  ── Answer from gpt-4o-2024-08-06 ───────────────────────────────────────────────────────────────
#>  The sky appears blue due to a phenomenon called Rayleigh scattering. When sunlight enters Earth's
#>  atmosphere, it is made up of many different colors, each with different wavelengths. Blue light has a
#>  shorter wavelength and is scattered in all directions by the gases and particles in the atmosphere more
#>  than other colors with longer wavelengths, like red or yellow.
#>  
#>  As sunlight passes through the atmosphere, the blue light is scattered more widely across the sky and is
#>  more likely to be seen from any given point. This widespread scattering of blue light makes the sky
#>  appear predominantly blue to our eyes during the day. Near sunrise and sunset, the sky can take on red
#>  or orange hues because the sun is lower on the horizon, and its light passes through more of the
#>  atmosphere, scattering shorter wavelengths out of the direct line of sight and allowing longer
#>  wavelengths like red and orange to dominate.

[JBGruber]

Authorization works now after setting the header:

options(
  rollama_server = "https://llm.societal-analytics.nl/ollama/",
  rollama_headers = list(
    # the token is stored in an .Renviron file
    Authorization = sprintf("Bearer %s", Sys.getenv("SAL"))
  )
)

But there are several todos left:

• So far, only the ollama endpoint works because we check the model before making a request. We should explore if this is neccesary or alternativly, if we can check the open webui endpoint if the Ollama endpoint does not exist

rollama/R/lib.R

Line 54 in 6b2099d

check_model_installed(model, server = server)

• The output parsing is also specific to Ollama. User can choose "httr2_response" of course, but that is not ideal