arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2306.01286cs.CLcs.AI
收藏

KL散度引导的温度采样

KL-Divergence Guided Temperature Sampling

Google Research(谷歌研究院)

Chung-Ching Chang, David Reitter, Renat Aksitov, Yun-Hsuan Sung

AI总结:

本文提出基于KL散度动态调整解码温度的方法,以缓解多样性与归因之间的权衡,并在对话问答和摘要任务中优于top-k和top-p采样。

摘要:

温度采样是使大语言模型预测多样化的常规方法。随着温度升高,预测变得多样,但也容易产生幻觉——生成看似合理但不符合事实的 token。缓解幻觉的一种常见方法是提供来源/依据文档,并训练模型生成与所提供来源绑定且可归因的预测。多样性与归因之间似乎存在权衡。为了缓解这种权衡,我们提出放宽在解码步骤中保持固定温度的约束,并提出一种机制,通过 KL 散度根据其与来源的相关性来引导动态温度。我们的实验证实了这种权衡,并表明我们的采样算法在对话式问答和摘要任务中优于传统的 top-k 和 top-p 算法。

英文摘要:

Temperature sampling is a conventional approach to diversify large language model predictions. As temperature increases, the prediction becomes diverse but also vulnerable to hallucinations -- generating tokens that are sensible but not factual. One common approach to mitigate hallucinations is to provide source/grounding documents and the model is trained to produce predictions that bind to and are attributable to the provided source. It appears that there is a trade-off between diversity and attribution. To mitigate any such trade-off, we propose to relax the constraint of having a fixed temperature over decoding steps, and a mechanism to guide the dynamic temperature according to its relevance to the source through KL-divergence. Our experiments justifies the trade-off, and shows that our sampling algorithm outperforms the conventional top-k and top-p algorithms in conversational question-answering and summarization tasks.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑