arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2501.16634cs.DCcs.AI
收藏

迈向资源高效的复合 AI 系统

Towards Resource-Efficient Compound AI Systems

MIT CSAIL(麻省理工学院计算机科学与人工智能实验室);Microsoft Azure(微软Azure)

Gohar Irfan Chaudhry, Esha Choukse, Íñigo Goiri, Rodrigo Fonseca, Adam Belay, Ricardo Bianchini

AI总结:

本文提出面向复合 AI 系统的声明式工作流模型与自适应运行时,并通过 Murakkab 原型实现编排和资源管理协同,最高获得 3.4 倍加速与 4.5 倍能效提升。

摘要:

复合 AI 系统集成了模型、检索器和外部工具等多个交互组件,已成为解决复杂 AI 任务的关键。然而,当前实现存在资源利用效率低下的问题,原因包括应用逻辑与执行细节之间的紧耦合、编排层与资源管理层之间的脱节,以及人们认为效率与质量相互排斥。我们提出一种资源高效复合 AI 系统的愿景,通过声明式工作流编程模型和用于动态调度、资源感知决策的自适应运行时系统来实现。将应用逻辑与底层细节解耦,可为运行时暴露调节手段,使其灵活配置执行环境和资源,同时不损害质量。促进工作流编排器与集群管理器之间的协作,可通过更好的调度和资源管理实现更高效率。我们正在构建一个名为 Murakkab 的原型系统以实现这一愿景。初步评估表明,该系统在工作流完成时间上最高可实现约 3.4 倍加速,同时能源效率提高约 4.5 倍,显示出优化资源和推进 AI 系统设计的前景。

英文摘要:

Compound AI Systems, integrating multiple interacting components like models, retrievers, and external tools, have emerged as essential for addressing complex AI tasks. However, current implementations suffer from inefficient resource utilization due to tight coupling between application logic and execution details, a disconnect between orchestration and resource management layers, and the perceived exclusiveness between efficiency and quality. We propose a vision for resource-efficient Compound AI Systems through a declarative workflow programming model and an adaptive runtime system for dynamic scheduling and resource-aware decision-making. Decoupling application logic from low-level details exposes levers for the runtime to flexibly configure the execution environment and resources, without compromising on quality. Enabling collaboration between the workflow orchestration and cluster manager enables higher efficiency through better scheduling and resource management. We are building a prototype system, called Murakkab, to realize this vision. Our preliminary evaluation demonstrates speedups up to $\sim 3.4\times$ in workflow completion times while delivering $\sim 4.5\times$ higher energy efficiency, showing promise in optimizing resources and advancing AI system design.

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。

↑