LoRA 基础
常见 LoRA 训练错误
常见问题包括数据太少、重复图、标注不清、概念混杂和过拟合。
这篇文章属于 LoRA 学习参考库。
Common mistakes include tiny datasets, duplicated images, unclear captions, mixed concepts, overtraining, and testing only one prompt. A safer workflow is to train small, test broadly, and stop when the model is useful rather than extreme.