Kareus: Joint Reduction of Dynamic and Static Energy in Large Model Training

2026-01-25

AI Summary

This work addresses the limitation of existing large model training optimization methods, which typically focus on either dynamic or static energy consumption in isolation, thereby failing to achieve holistic energy efficiency. To overcome this, the paper introduces a novel partition-based multi-objective optimization framework that jointly models dynamic and static energy consumption for the first time. By co-optimizing fine-grained operator scheduling and frequency scaling, the approach achieves a superior trade-off between training time and energy usage. The method decomposes the otherwise intractable global problem into efficiently solvable local subproblems and employs an iterative optimization strategy to coordinate execution scheduling and frequency control. Experimental results demonstrate that, compared to state-of-the-art techniques, the proposed method reduces energy consumption by up to 28.3% under identical training time constraints, or shortens training time by up to 27.5% at the same energy cost.

Technology Category

• Machine Learning: Optimization
• Search and Optimization: Learning to Search
• Computer Vision: Learning & Optimization for CV

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Systems and Infrastructure for Web, Mobile and WoT: Energy management for devices in mobile Web and WoT environments

Abstract

The computing demand of AI is growing at an unprecedented rate, but energy supply is not keeping pace. As a result, energy has become an expensive, contended resource that requires explicit management and optimization. Although recent works have made significant progress in large model training optimization, they focus only on a single aspect of energy consumption: dynamic or static energy. We find that fine-grained kernel scheduling and frequency scaling jointly and interdependently impact both dynamic and static energy consumption. Based on this finding, we design Kareus, a training system that pushes the time--energy tradeoff frontier by optimizing both aspects. Kareus decomposes the intractable joint optimization problem into local, partition-based subproblems. It then uses a multi-pass multi-objective optimization algorithm to find execution schedules that push the time--energy tradeoff frontier. Compared to the state of the art, Kareus reduces training energy by up to 28.3% at the same training time, or reduces training time by up to 27.5% at the same energy consumption.

Problem

• dynamic energy
• static energy
• large model training
• energy optimization
• joint optimization

Innovation

• joint energy optimization
• dynamic and static energy
• kernel scheduling
• frequency scaling
• multi-objective optimization