A fast multimodal LLM for real-time voice - View it on GitHub