让大模型在测试时积累知识，无需改参数也能越用越强。
Test-Time Learning with an Evolving Library
- 用模型自身推理轨迹自动提取技能和反思，构建共享知识库
- 通过权重优化机制，让临时知识逐步变成通用能力
- 在数学、编程等任务上超越现有测试时学习方法
我们提出EvoLib，一种无需参数更新或外部监督的测试时学习框架，使大语言模型能够跨问题实例累积、复用并演化知识。该方法不调整模型参数，而是维护一个共享的知识抽象库，包括从模型自身推理轨迹中自动提取的模块化技能与反思性洞察。为支持持续改进，我们引入一种兼顾即时效用与长期价值的加权与整合机制，使简单、特定于实例的抽象随时间演变为更通用、可复用的形式。在数学推理、代码生成及多轮代理环境等挑战性基准上，EvoLib在无真值反馈的情况下，显著优于当前最先进的测试时扩展与学习方法。
原文摘要 · Abstract (English)
We introduce EvoLib, a test-time learning framework that enables large language models to accumulate, reuse, and evolve knowledge across problem instances without parameter updates or external supervision. Instead of adapting model parameters, our approach maintains a shared library of knowledge abstractions, including modular skills and reflective insights, automatically extracted from the model's own inference trajectories. To support continual improvement, we introduce a principled weighting and consolidation mechanism that jointly optimizes for immediate utility and long-term value. This allows simple, instance-specific abstractions to evolve into more general and reusable ones over time. Across challenging benchmarks in mathematical reasoning, code generation, and multi-turn agentic environments, EvoLib improves substantially over the top test-time scaling and learning methods without ground-truth feedback.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。