Estimating the Effects of Sample Training Orders for Large Language Models without Retraining

Hao Yang, Haoxuan Li, Mengyue Yang +2

The order of training samples plays a crucial role in large language models (LLMs), significantly impacting both their external performance and internal learning dynamics. Traditio…