In the LLM world there are many open source solutions to find tuning, maybe the best one being from Meta: https://github.com/pytorch/torchtune

In terms of inference and interface (since you mentioned comfy) there are many truly open source options such as vLLM (though there isn't a single really performant open source solution for inference yet).