Can anyone reproduce the result of ViT?

[ZhiyuanChen]

I have tried the converted state_dict and the provided model, but I can only get 81.624 with ViT-L_16

[ZhiyuanChen]

Key is ToTensor of PyTorch normalize inputs to [1, 0], where flax implementation normalized it to [1, -1]

[rwightman]

That information is in the impl here and the validation scripts here use it to setup the pipeline properly.
https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/vision_transformer.py#L65