用预训练模型和字体数据微调训练个人字体,推理的字体总体效果可以,但出现部分崩字(结构变形、多笔、少笔)。 求解决办法,有尝 <img width="742" height="397" alt="Image" src="https://github.com/user-attachments/assets/5dcbef57-621f-4b5f-9c39-f61d75a46a2d" />