About this document
Energy Management for Large Model Training by gliangdu is a document available to read on EtoBox.
The document presents a long-term energy management method for training large models, addressing the significant energy consumption and carbon emissions associated with such processes. It proposes a framework that utilizes multi-timescale forecasts and a linear aggregated energy model to optimize GPU server configurations and scheduling, ultimately reducing carbon emissions by 5.9% compared to existing methods. The approach aims to simplify the complexity of energy management by decomposing the scheduling p
- Author
- gliangdu
- Language
- EN