arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2009.07243cs.CLcs.AIcs.LG
收藏

开放式语言生成抽样算法的系统刻画

A Systematic Characterization of Sampling Algorithms for Open-ended Language Generation

Massachusetts Institute of Technology(麻省理工学院);New York University(纽约大学)

Moin Nadeem, Tianxing He, Kyunghyun Cho, James Glass

摘要:

本文研究了自回归语言模型中被广泛采用的祖先抽样算法,这在文献中没有被广泛研究。我们使用质量多样性(Q-D)权衡调查三个流行的采样算法(top-k,核和回火采样)。我们专注于开放式语言生成的任务。我们首先表明,现有的采样算法具有类似的性能。在仔细检查了不同采样算法定义的变换后,我们确定了它们之间共享的三个关键属性:熵减少,顺序保持和斜率保持。为了验证所确定的属性的重要性,我们设计了两组新的采样算法:一组中,每个算法满足所有三个属性,和一组中,每个算法违反至少一个属性。我们将它们的性能与现有的采样算法进行比较,发现违反所确定的属性可能会导致急剧的性能下降,如Q-D权衡所测量的。另一方面,我们发现,满足这些属性的采样算法集与现有的采样算法表现相当。我们的数据和代码可在https://github.com/moinnadeem/characterizing-sampling-algorithms上获得

英文摘要:

This work studies the widely adopted ancestral sampling algorithms for auto-regressive language models, which is not widely studied in the literature. We use the quality-diversity (Q-D) trade-off to investigate three popular sampling algorithms (top-k, nucleus and tempered sampling). We focus on the task of open-ended language generation. We first show that the existing sampling algorithms have similar performance. After carefully inspecting the transformations defined by different sampling algorithms, we identify three key properties that are shared among them: entropy reduction, order preservation, and slope preservation. To validate the importance of the identified properties, we design two sets of new sampling algorithms: one set in which each algorithm satisfies all three properties, and one set in which each algorithm violates at least one of the properties. We compare their performance with existing sampling algorithms, and find that violating the identified properties could lead to drastic performance degradation, as measured by the Q-D trade-off. On the other hand, we find that the set of sampling algorithms that satisfies these properties performs on par with the existing sampling algorithms. Our data and code are available at https://github.com/moinnadeem/characterizing-sampling-algorithms

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。

↑