Qwen2.5 VL 3B Instruct

Qwen2.5-VL open-source series model, with improved instruction following, mathematics, problem-solving, and coding capabilities, enhanced ability to recognize all things, supports diverse formats for direct and precise location of visual elements, supports understanding of up to one hour of video and second-level event moment positioning, can understand time sequence and speed, based on parsing and positioning capabilities supports manipulation of OS or Mobile agents, strong key information extraction capability and Json format output capability, this version is a 3B version suitable for mobile device use.