Why adding a model to vllm only possible by branching from this repo?

[tsvisab]

I am not sure about the process of adding a model to VLLM, i saw the process is forking this repo, adding my model entire code to one of the folders here and implementing forward and load_weight, this seems like we are effectively implementing an LLM base class, does it make sense for you to allow supporting new models by deriving from some base class and having VLLM as a dependency?