Python: ADR for new AI connector abstract methods

[TaoChenOSU]

Motivation and Context

This ADR documents a proposal to introduce four new abstract methods in ChatCompletionClientBase and TextCompletionClientBase to optimize code and further streamline the process of adding new AI connectors.

Description

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the SK Contribution Guidelines and the pre-submission formatting script raises no violations
• All unit tests pass, and I have added new tests where possible
• I didn't break anyone 😄

[nmoeller]

@TaoChenOSU i really like this ! With this we could potentially running FunctionCall in ChatCompletionClientBase Class using your new abstract Methods, this would remove a lot of duplicate Code in the Connectors !

[eavanvalkenburg]

This is great, Tao, thanks for writing this up, I like the thinking, I've used a very similar setup for the new memory connectors, where the classes implement a _inner_* method, while the get, upsert, etc methods are all default implemented in the base class, indeed makes it much simpler to implement a new one! One thing to add is a way to indicate which model can and can't do function calling, since the function call behavior is part of the base prompt execution settings and that is therefore not a way to distinguish!

[TaoChenOSU]

This is great, Tao, thanks for writing this up, I like the thinking, I've used a very similar setup for the new memory connectors, where the classes implement a _inner_* method, while the get, upsert, etc methods are all default implemented in the base class, indeed makes it much simpler to implement a new one! One thing to add is a way to indicate which model can and can't do function calling, since the function call behavior is part of the base prompt execution settings and that is therefore not a way to distinguish!

That's great! I think a ClassVar will be sufficient to indicate whether a connector will support function calling. I will add that.