How to Autostart tiny-Qwen2_5_VLForConditionalGeneration on AMD/Nvidia GPU with Native FP4 Full Method