Tag: Test Time
All the articles with the tag "Test Time".
-
Test-time Correlation Alignment
本文提出测试时相关性对齐(TCA)范式,通过构建伪源域相关性并应用线性变换对齐测试数据特征,显著提升测试时适应(TTA)性能,同时保持高效性和源域知识。
-
Do We Truly Need So Many Samples? Multi-LLM Repeated Sampling Efficiently Scales Test-Time Compute
This paper introduces ModelSwitch, a multi-LLM repeated sampling strategy that leverages answer consistency to dynamically switch models, achieving superior performance and 34% sample efficiency over single-LLM self-consistency across diverse datasets.
-
Test-time regression: a unifying framework for designing sequence models with associative memory
本文提出一个基于测试时回归的统一框架,通过将关联回忆形式化为回归问题,推导出多种序列模型(如线性注意力、状态空间模型、softmax注意力),并通过合成实验验证其回归能力,同时提出高阶注意力泛化。
-
TTRL: Test-Time Reinforcement Learning
本文提出测试时强化学习(TTRL)方法,通过多数投票估计奖励,在无标签测试数据上训练大语言模型,实现模型自演化并显著提升推理任务性能。
-
A Survey on Test-Time Scaling in Large Language Models: What, How, Where, and How Well?
本文通过提出一个四维度分类框架(什么扩展、如何扩展、哪里扩展、扩展效果如何),系统综述了测试时扩展(TTS)在大型语言模型中的研究现状,为理解和应用推理阶段计算扩展提供了结构化视角和实践指导。