> purposely tried to avoid adding any niche network modifications that would help it overfit to in-1k. All three of the modifications are applicable to other ne

> purposely tried to avoid adding any niche network modifications that would help it overfit to in-1k. All three of the modifications are applicable to other networks and datasets. You introduce a ridiculous number of degrees of freedom from your architectural modifications and the vast majority of them don’t seem to be some standard way of doing things. If these are generally applicable to most datasets, then show an eval on a new dataset against ViT that you haven’t looked at while making your arch modifications. The way this is described per the blog is just not sufficient for the sweeping titular claim without more robust evaluation without overfitting/p-hacking