一项名为词汇覆盖率得分（WCS）的新指标已被引入，用于评估大型语言模型（LLM）的标准采样过滤器如何无意中减少语言多样性。WCS量化了诸如Top-p和Top-k之类的采样方法对上下文适宜的、低频人类词汇的修剪。研究表明，这些默认的采样参数可以充当审查机制，导致文本同质化，并抹平独特的人类表达。 AI
影响 这项研究提供了一个诊断工具，用于优化LLM输出，以平衡连贯性与词汇丰富度，可能产生更多样化、不那么同质化的生成文本。
排序理由 该集群包含一篇详细介绍新指标和研究结果的学术论文。
在 Hugging Face Daily Papers 阅读 →
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
arXiv:2605.27268v1 Announce Type: cross Abstract: Modern Large Language Models (LLMs) are often criticized for producing repetitive and homogeneous text, despite possessing vast latent vocabularies. While previous research has focused on model knowledge and training data, we inve…
Modern Large Language Models (LLMs) are often criticized for producing repetitive and homogeneous text, despite possessing vast latent vocabularies. While previous research has focused on model knowledge and training data, we investigate the role of decoding mechanics in suppress…
Standard sampling filters in large language models unintentionally suppress linguistic diversity by pruning contextually appropriate vocabulary, creating a homogenized output despite vast latent vocabularies.