https://www.modelscope.cn/models/facebook/dinov3-vitb16-pretrain-lvd1689m run batch inference on single ARC gpu, FP16 or even FP8