The models available in this repository are highly satisfactory, but is there a possibility to explore the CoAtNet model in some benchmark dataset? (e.g: ImageNet [1k, 21k], COCO, ...)