Confusion about different EVA architectures

[t0278611]

I read your EVA-CLIP Paper and I am a little bit confused about the timeline. It seems that you trained EVA-CLIP and EVA at the same time and EVA-02 is a successor. In the EVA-CLIP Paper you mention EVA-02-CLIP but nothing about which architecture you are using. When I look at the code, it does seem like EVA-02-CLIP is indeed using the EVA-02 architecture, so I am a little bit confused about how the models are related.

In Particular I would like to know which architecture you use for the gigantic 18B version. From your code and from your paper it would seem you used the vanilla ViT architecture. But I see in your code configs for a EVA02 style 16B version. Although I can't find anything in your paper. Are the experiments still ongoing?