Try Inference with SALMONN's Q former conda activate videollama_llx python demo_audio.py --gpu-id "your gpu id"