Up to 30% of the power used to train AI is wasted: A software tool could help fix that
A less wasteful way to train large language models, such as the GPT series, finishes in the same amount of time for up to 30% less energy, according to a new study from the University of Michigan.
Nov 7, 2024
0
11