Full step-by step description can be found at:
Links for model/tokenizer downloads: bert model, bert config, bert vocabulary for tokenizer, distilbert model, distilbert config, distilbert vocabulary for tokenizer
pashadude/troll_detection_nlp
revising reddit troll hunting with some new nlp techniques (DistilBERT, Multi-Sample Dropout, OOF, Group K-fold on subreddit and some fancy pre-processing)
Jupyter Notebook