Semi-supervised finetuning doesn't improve accuracy of finetuned models.

[opencomvis]

Thank you for your work,
I have followed the instructions mention here and in the code for Supervised Finetuning and Semi-supervised Finetuning the ViT architectures. The Supervised Finetuning process has finished successfully, however, when doing the Semi-supervised Finetuning, the results doesn't improved over the supervised finetuned models, which made me confused, wither I am doing something wrong or there is an issue. Your kind help will be appreciated.