Model client configuration from JSON

[ekzhu]

What feature would you like to be added?

We need to support model client configuration from a file, just like in the autogen-agentchat 0.2 version. However, this time it needs to have a strict and documented schema per model client. Additionally, it needs to be extendable in terms of supporting new model clients. As far as extendable, we need to explore ways that new clients can dynamically register themselves as something that can be instantiated via the configuration processor.

Allow users to configure a model client from a configuration file.

Requirements:

• ability to dynamically register/load the plugin without needing to change your apps code
• The custom model MUST provide a schema for the available config for it so it can be documented and validated
• Access models from the config file via a key
• A user should be able to use the same config for both a Dotnet and Python application
• Document how to target and use a community provided client

Dynamic import:

• Specify import in configuration "python_provider": "gemini_agnext_client.GeminiClient",

Example json:

{
    "gpt4o": {
        "provider": {
            "python": "autogen_ext.models.AzureOpenAIChatCompletionClient",
        },
        "config": {
            "api_version": "2024-02-15-preview",
            "azure_endpoint": "https://aif-complex-tasks-west-us-3.openai.azure.com/",
            "azure_deployment": "gpt4o-1",
            "model": "gpt-4o-2024-05-13",
            "temperature": 0.1
        }
    }
}

Why is this needed?

Users are used to being able to configure model clients via a configuration file (JSON) from AutoGen.