This tutorial is based on Hugging Face course.
You can see the YouTube video recorded for this tutorial (Persian)
distilbert-base-uncased is recommended, since it's faster than bert-base-uncased and offers a good performance. Also it was pretrained with the same corpus as BERT. This model is aimed at being fine-tuned for NLP tasks such as text classification, token classification, and question answering, for text generation you should go for models such as gpt2. More information about this model is available here.
sst2 from glue benchmark is used on this tutorial. It consists of sentences from movie reviews and human annotations of their sentiment. The task is to predict the sentiment of a given sentence. It uses the two-way (positive/negative) class split, with only sentence-level labels.