[FEATURE] Use local LLM deployment as Judge

[pascal-pfeiffer]

🚀 Feature

Additional to the hardcoded models, add support to use local models as judges for evaluation. Can be simplified to require the OpenAI API.
Should be basically an endpoint selection, probably the Azure hosted pipeline can be extended to cover this. If already working, add documentation on how this can be done.

Motivation

Local development and evals

[pascal-pfeiffer]

One way this is already supported in the current version:

1. Have an endpoint running that supports the OpenAI API format, specifically chat.completions.

2. Start LLM Studio with environment variable to point to that endpoint: OPENAI_API_BASE="http://111.111.111.111:8000/v1"

3. Validate correct usage in "Settings page". Note that "Use Azure" must be off, and the environment variable that was set above should be visible below. Changing it here has no effect! This is only for testing the correct setting of the environment variable.
   

4. Run an experiment with GPT metric and use the correct model name at your endpoint:
   

5. Calls to the LLM judge are now directed to your own LLM endpoint
   

[sherenem]

Documented in #756