English

Trusting Your Evidence: Hallucinate Less with Context-aware Decoding

Computation and Language 2023-05-25 v1

Abstract

Language models (LMs) often struggle to pay enough attention to the input context, and generate texts that are unfaithful or contain hallucinations. To mitigate this issue, we present context-aware decoding (CAD), which follows a contrastive output distribution that amplifies the difference between the output probabilities when a model is used with and without context. Our experiments show that CAD, without additional training, significantly improves the faithfulness of different LM families, including OPT, GPT, LLaMA and FLAN-T5 for summarization tasks (e.g., 14.3% gain for LLaMA in factuality metrics). Furthermore, CAD is particularly effective in overriding a model's prior knowledge when it contradicts the provided context, leading to substantial improvements in tasks where resolving the knowledge conflict is essential.

Keywords

Cite

@article{arxiv.2305.14739,
  title  = {Trusting Your Evidence: Hallucinate Less with Context-aware Decoding},
  author = {Weijia Shi and Xiaochuang Han and Mike Lewis and Yulia Tsvetkov and Luke Zettlemoyer and Scott Wen-tau Yih},
  journal= {arXiv preprint arXiv:2305.14739},
  year   = {2023}
}
R2 v1 2026-06-28T10:44:00.388Z