University of Chinese Academy of Sciences Open-Sources Multimodal LLM LLaMA-OmniLLaMA-Omni outperforms traditional baseline models in speech and text processing while requiring less training data and compute resources.