low-rank adaptation

Cyberbullying detection requires models to distinguish targeted harassment from general hostility and other forms of abusive language. Although large language models can perform this task through prompting, supervised adaptation generally achieves …