Trained ModernBERT (base and large) for detection hallucinations in LLM responses. The models are trained as token classifications.