AI14 hours agoMultimodal AI on Developer GPUs: Alibaba Releases Qwen2.5-Omni-3B with 50% Lower VRAM Usage and Nearly-7B Model Performance