Why pretrained weights are bigger than finetune weights?

[eeyrw]

I download:

But I find out that pretrained weights is about 3x larger comparing with finetune. What's the difference between them?
When I loading the pretrained weights a message from console saying:
reshape position embedding from 196 to 576.
Does it means that two weights correspond to different model configuration?

[LiJunnan1992]

1. pretrained weights contains additional parameters such as the momentum model used for contrastive learning.
2. yes the input resolution is different between pretrain & finetune

[eeyrw]

I see. Thank you.