Checkpoint for ViT-L/14 Pretrained from Scratch (with SwiGLU)

[dfan]

In Table 17 of the paper, it is mentioned that SwiGLU is used as the FFN layer for ViT-L/14 when training from scratch. In the model card, only the ViT-G model trained from scratch with SwiGLU is available (the ViT-L uses MLP FFN and seems to be the distilled version). Is it possible to add the checkpoint for the ViT-L/14 trained from scratch?

I.e. matching this config: https://github.com/facebookresearch/dinov2/blob/main/dinov2/configs/train/vitl14.yaml