共形核取样（英文版）

英文标题：Conformal Nucleus Sampling中文摘要：本文通过使用 conformal prediction 方法对 $p$ 参数进行校准来研究 top-$p$ 采样在各种语言上下文环境下是否与其概率意义对齐，结果表明 OPT 模型存在过度自信，而校准与模型大小存在适度的反比关系。英文摘要：Language models generate text based on successively sampling the next word. Adecoding procedure based on nucleus (top-$p$) sampling chooses from thesmallest possible set of words whose cumulative probability exceeds theprobability $p$. In this work, we assess whether a top-$p$ set is indeedaligned with its probabilistic meaning in various ling