Fine-tuning without loading position embedding from pre-trained model

[Blosslzy]

Hi!
When fine-tuning in my own data set, can I just load the parameters of the encoder block from the pre-trained ViT-B? Thanks!