Submitted by KAI LIU 18 JavisGPT: A Unified Multi-modal LLM for Sounding-Video Comprehension and Generation JavisVerse 55 3