Different accuracy

Hello,
I am training the same BERT model from huggingface on the same dataset for text classification using google colab…but the accuracy and f1 are different in each time I repeat the process. Why and how can I fix that?