Text classification for low number of data

Question

I have a low amount of data for the text binary classification problem(1000 - 2000 samples).

I use Universal Sentence Encoding and SBERT for text embedding and then S