https://docs.unsloth.ai/models/tutorials-how-to-fine-tune-an...
Note that llama.cpp doesn't try to be production-grade engine, more focused on local usage.