New method could increase LLM training efficiency

Researchers have developed a new method that could double the speed of large language model (LLM) training by leveraging idle computing time. This approach maintains the accuracy of the models during the training process.