Llm Config

This configures RuVLLM for local LLM inference with two different adaptation strategies. MicroLoRA is your traditional fine-tuning approach for task-specific work, taking minutes to train and saving adapter weights. SONA is the interesting bit: sub-millisecond neural adaptation for real-time feedback loops, though it only persists during your session. You get status checks, config generation, and separate workflows for creating and adapting each type. If you're running local inference and need something between base model performance and full retraining, the dual approach gives you options depending on whether you need durability or speed.