Serve pepekberbulu Q4_K_M GGUF live via llama-server (fast CPU inference)

Deploy VibeThinker-3B -> GGUF Q4_K_M conversion+upload job

initial commit