future7/CogniDet
TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kArchitecture:Transformer0.0K Cold

CogniDet by future7 is an 8 billion parameter Llama3-backbone model designed for detecting both factual and cognitive hallucinations in Large Language Model outputs. It specializes in identifying unsupported inferences and evaluations, going beyond simple factual contradictions. Trained on the large-scale CogniBench-L dataset, CogniDet offers efficient, single-pass detection, making it suitable for robust LLM evaluation, particularly in English knowledge-grounded dialogues.

Loading preview...