Monitoring Decoding: Mitigating Hallucination via Evaluating the Factuality of Partial Response during Generation

Yurui Chang, Bochuan Cao, Lu Lin

While large language models have demonstrated exceptional performance across a wide range of tasks, they remain susceptible to hallucinations -- generating plausible yet factually…