Qwen3-VL-4B and 8B runs locally on NPU, GPU, and CPU with one SDK nexa.ai 3 points by alanzhuly 10 hours ago