BERTweet is a pre-trained transformer-based language model developed by VinAI Research. It adapts the RoBERTa pre-training procedure specifically for English tweets, trained on a corpus of 850M English Tweets (including 5M COVID-19 related). The model supports downstream tasks including sentiment analysis, irony detection, named entity recognition, part-of-speech tagging, and text classification. It can be used via both the Hugging Face transformers library and fairseq.