arXiv · cs.AI · Ilija Subasic, Andrew Rabinovich, Zhao Chen · 11 小时前模型评分37
SQM:面向 大语言模型(LLM,Large Language Model)-as-a-Judge 校准的流式分位数匹配方法
Quantiles, Not Points: Streaming Quantile Matching for Ordinal-Aware LLM-as-a-Judge Calibration
导读
研究者提出流式分位数匹配(SQM,Streaming Quantile Matching),通过计算大语言模型(LLM,Large Language Model)与人类评分分布之间的最优传输映射来校准 LLM-as-a-Judge 打分,而非逐点映射。
分类置信度
中 · 待人工复核
- 只有单一媒体来源,需要人工复核。
来源级别:媒体与个人
事件类型:研究发布
证据摘录
- “We present streaming quantile matching (SQM), a version of a classical method not previously bench marked for LLMaJ calibration, which is built on the two positions by computing the optimal-transport map between the LLM and human score distributions.”
来源:arXiv · cs.AI · arxiv.org