arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.02597cs.LGcs.AI
收藏

ContextEvolve: 多智能体上下文压缩用于系统代码优化

ContextEvolve: Multi-Agent Context Compression for Systems Code Optimization

Tsinghua University, Beijing, China(清华大学);Zhongguancun Academy, Beijing, China(中关村学院);Massachusetts Institute of Technology, Cambridge MA, USA(麻省理工学院)

Hongyuan Su, Yu Zheng, Yong Li

AI总结:

ContextEvolve通过多智能体框架实现系统代码优化,利用上下文压缩提升搜索效率,实验表明其在ADRS基准测试中性能提升显著。

摘要:

大型语言模型正在通过自动化发现计算机系统中性能关键的算法来改变系统研究。尽管LLM生成的代码看似合理,但要产生符合系统严格正确性和性能要求的解决方案,仍需要迭代优化。测试时强化学习提供了高效的搜索效率,但需要参数更新,在仅API访问的情况下不可行,而现有的无训练进化方法则面临上下文利用效率低和搜索方向不明确的问题。我们引入了ContextEvolve,一种多智能体框架,通过将优化上下文分解为三个正交维度,在严格参数盲约束下实现RL级别的搜索效率:一个摘要智能体通过代码到语言抽象来压缩语义状态,一个导航智能体通过轨迹分析来提炼优化方向,一个采样智能体通过优先级示例检索来整理经验分布。这种协同作用形成了与RL映射到状态表示、策略梯度和经验回放的功能同构,使在文本潜在空间中实现原理化的优化成为可能。在ADRS基准测试中,ContextEvolve在性能上优于现有最先进的基线方法33.3%,同时将token消耗减少了29.0%。我们的工作代码已发布在https://anonymous.4open.science/r/ContextEvolve-ACC。

英文摘要:

Large language models are transforming systems research by automating the discovery of performance-critical algorithms for computer systems. Despite plausible codes generated by LLMs, producing solutions that meet the stringent correctness and performance requirements of systems demands iterative optimization. Test-time reinforcement learning offers high search efficiency but requires parameter updates infeasible under API-only access, while existing training-free evolutionary methods suffer from inefficient context utilization and undirected search. We introduce ContextEvolve, a multi-agent framework that achieves RL-level search efficiency under strict parameter-blind constraints by decomposing optimization context into three orthogonal dimensions: a Summarizer Agent condenses semantic state via code-to-language abstraction, a Navigator Agent distills optimization direction from trajectory analysis, and a Sampler Agent curates experience distribution through prioritized exemplar retrieval. This orchestration forms a functional isomorphism with RL-mapping to state representation, policy gradient, and experience replay-enabling principled optimization in a textual latent space. On the ADRS benchmark, ContextEvolve outperforms state-of-the-art baselines by 33.3% while reducing token consumption by 29.0%. Codes for our work are released at https://anonymous.4open.science/r/ContextEvolve-ACC

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。

↑