miwieg

roberta with long text instances

8

I am using FLAIR for text classification ("TextClassifier") with RoBERTa. My dataset just contains about 2000 instances but the text instances themselves are fairly long, i.e. longer than 512 tokens....