Researchers at the University of Michigan have developed a software tool called Perseus that significantly reduces the energy consumption of training large language models. This breakthrough study shows that by optimizing resource allocation during training, energy consumption can be reduced by 30% without affecting the training speed. This technology not only has important economic benefits, but also can make great contributions to the environmental protection cause. Its potential energy-saving effect is even enough to meet the electricity needs of millions of American households.
A new study by the University of Michigan found that a labor-saving method for training large language models can be completed within the same time, but energy consumption can be reduced by 30%. This approach can save enough energy to power 1.1 million U.S. homes by 2026.
Researchers have developed a software tool called Perseus that identifies critical paths, a series of subtasks that take the longest time to complete. Perseus then slows down processor speeds on non-critical paths so that they all do the job at the same time, eliminating unnecessary power consumption.
The team tested Perseus by training GPT-3, three other large language models and a computer vision model. The results show that Perseus can reduce the energy consumption of AI training while maintaining the same training speed.
Researchers say this labor-saving approach is of great significance to the fair use of artificial intelligence. If a country does not have enough power to run large models, they may need to use remote services or can only run smaller, less accurate models. This gap may further exacerbate the gap between different communities.
This study shows that by optimizing AI training methods, energy consumption can be reduced while maintaining the same training speed. This is of great significance for saving energy and reducing carbon footprint.
This study provides an effective solution to the energy efficiency problem in the field of artificial intelligence, paves the way for more sustainable development of artificial intelligence, and provides important guarantees for the fairness and inclusiveness of artificial intelligence in the future. Perseus has broad application prospects and is expected to promote the advancement of artificial intelligence technology and wider application.