Branching out on model support

[pcfreak30]

@b4rtaz I am opening this issue because I would like to get info/understand high level what is required to support other model families.

Two that come to mind are the actual deepseek R1 and the Qwen families. But my question is more abstract as well. Is every effort having to be a reverse engineering of the model to run case by case or can the process of a "distributed" model be abstracted and generalized? Im aware of you having a HF script, but not yet sure on the details of what the conversions are doing.

I just found this project after spending a decent amount of my time doing research/testing around llamas GRPC via localai.

Kudos.